This patch series adds bindings document for RPU subsystem found on Xilinx
ZynqMP platforms. It also adds device nodes and driver to enable RPU subsytem
in split mode and lockstep mode.
Xilinx ZynqMP platform contains Remote Processing Unit(RPU). RPU subsystem
contains two arm cortex r5f cores. RPU subsystem can be configured in
split mode, locsktep mode and single-cpu mode.
RPU subsystem also contains 4 Tightly Coupled Memory(TCM) banks. In lockstep
mode, all 4 banks are combined and total of 256KB memory is made available to
r5 core0. In split mode, both cores can access two TCM banks i.e. 128 KB.
RPU can also fetch data and execute instructions from DDR memory along with
TCM memory.
---
Changes in v3:
- Fix checkpatch script indentation warning
- Remove unused variable from xilinx remoteproc driver
- use C style comments, i.e /*...*/
- Remove redundant debug information which can be derived using /proc/device-tree
- Fix multilined comment format
- s/"final fot TCM"/"final for TCM"
- Function devm_kzalloc() does not return an code on error, just NULL.
Remove redundant error check for this function throughout the driver.
- Fix RPU mode configuration and add documentation accordingly
- Get rid of the indentations to match function documentation style with rest of the driver
- Fix memory leak by only using r5_rproc->priv and not replace it with new instance
- Use 'i' for the outer loop and 'j' for the inner one as per convention
- Remove redundant error and NULL checks throughout the driver
- Use devm_kcalloc() when more than one element is required
- Add memory-regions carveouts during driver probe instead of parse_fw call
This removes redundant copy of reserved_mem object in r5_core structure.
- Fix memory leak by using of_node_put()
- Fix indentation of tcm_mem_map function args
- Remove redundant init of variables
- Initialize tcm bank size variable for lockstep mode
- Replace u32 with phys_addr_t for variable stroing memory bank address
- Add documentation of TCM behavior in lockstep mode
- Use dev_get_drvdata instead of platform driver API
- Remove info level messages
- Fix checkpatch.pl warnings
- Add documentation for the Xilinx r5f platform to understand driver design
Changes in v2:
- Remove proprietary copyright footer from cover letter
Ben Levinsky (3):
firmware: xilinx: Add ZynqMP firmware ioctl enums for RPU
configuration.
firmware: xilinx: Add shutdown/wakeup APIs
firmware: xilinx: Add RPU configuration APIs
Tanmay Shah (3):
dt-bindings: remoteproc: Add Xilinx RPU subsystem bindings
arm64: dts: xilinx: zynqmp: Add RPU subsystem device node
drivers: remoteproc: Add Xilinx r5 remoteproc driver
.../bindings/remoteproc/xlnx,r5f-rproc.yaml | 139 +++
arch/arm64/boot/dts/xilinx/zynqmp.dtsi | 17 +
drivers/firmware/xilinx/zynqmp.c | 97 ++
drivers/remoteproc/Kconfig | 12 +
drivers/remoteproc/Makefile | 1 +
drivers/remoteproc/xlnx_r5_remoteproc.c | 973 ++++++++++++++++++
include/dt-bindings/power/xlnx-zynqmp-power.h | 6 +
include/linux/firmware/xlnx-zynqmp.h | 60 ++
8 files changed, 1305 insertions(+)
create mode 100644 Documentation/devicetree/bindings/remoteproc/xlnx,r5f-rproc.yaml
create mode 100644 drivers/remoteproc/xlnx_r5_remoteproc.c
base-commit: f5d4b6d998d403352d0143e0c158706412384121
--
2.25.1
RPU subsystem can be configured in cluster-mode or split mode.
Also each r5 core has separate power domains.
Signed-off-by: Tanmay Shah <[email protected]>
---
Changes in v3:
- Fix checkpatch.pl style warning
arch/arm64/boot/dts/xilinx/zynqmp.dtsi | 17 +++++++++++++++++
1 file changed, 17 insertions(+)
diff --git a/arch/arm64/boot/dts/xilinx/zynqmp.dtsi b/arch/arm64/boot/dts/xilinx/zynqmp.dtsi
index 74e66443e4ce..9ac649a7b52c 100644
--- a/arch/arm64/boot/dts/xilinx/zynqmp.dtsi
+++ b/arch/arm64/boot/dts/xilinx/zynqmp.dtsi
@@ -647,6 +647,23 @@ qspi: spi@ff0f0000 {
power-domains = <&zynqmp_firmware PD_QSPI>;
};
+ r5fss: r5fss@ff9a0000 {
+ compatible = "xlnx,zynqmp-r5fss";
+ xlnx,cluster-mode = <1>;
+
+ reg = <0x0 0Xff9a0000 0x0 0x228>;
+
+ r5f_core0: r5f-0 {
+ compatible = "xlnx,zynqmp-r5f";
+ power-domains = <&zynqmp_firmware PD_RPU_0>;
+ };
+
+ r5f_core1: r5f-1 {
+ compatible = "xlnx,zynqmp-r5f";
+ power-domains = <&zynqmp_firmware PD_RPU_1>;
+ };
+ };
+
psgtr: phy@fd400000 {
compatible = "xlnx,zynqmp-psgtr-v1.1";
status = "disabled";
--
2.25.1
This driver enables r5f dual core Real time Processing Unit subsystem
available on Xilinx Zynq Ultrascale MPSoC Platform. RPU subsystem
(cluster) can be configured in different modes e.g. split mode in which
two r5f cores work independent of each other and lock-step mode in which
both r5f cores execute same code clock-for-clock and notify if the
result is different.
The Xilinx r5 Remoteproc Driver boots the RPU cores via calls to the Xilinx
Platform Management Unit that handles the R5 configuration, memory access
and R5 lifecycle management. The interface to this manager is done in this
driver via zynqmp_pm_* function calls.
Signed-off-by: Ben Levinsky <[email protected]>
Signed-off-by: Tanmay Shah <[email protected]>
---
Changes in v3:
- Fix checkpatch script indentation warning
- Remove unused variable from xilinx remoteproc driver
- use C style comments, i.e /*...*/
- Remove redundant debug information which can be derived using /proc/device-tree
- Fix multilined comment format
- s/"final fot TCM"/"final for TCM"
- Function devm_kzalloc() does not return an code on error, just NULL.
Remove redundant error check for this function throughout the driver.
- Fix RPU mode configuration and add documentation accordingly
- Get rid of the indentations to match function documentation style with rest of the driver
- Fix memory leak by only using r5_rproc->priv and not replace it with new instance
- Use 'i' for the outer loop and 'j' for the inner one as per convention
- Remove redundant error and NULL checks throughout the driver
- Use devm_kcalloc() when more than one element is required
- Add memory-regions carveouts during driver probe instead of parse_fw call
This removes redundant copy of reserved_mem object in r5_core structure.
- Fix memory leak by using of_node_put()
- Fix indentation of tcm_mem_map function args
- Remove redundant init of variables
- Initialize tcm bank size variable for lockstep mode
- Replace u32 with phys_addr_t for variable stroing memory bank address
- Add documentation of TCM behavior in lockstep mode
- Use dev_get_drvdata instead of platform driver API
- Remove info level messages
- Fix checkpatch.pl warnings
- Add documentation for the Xilinx r5f platform to understand driver design
drivers/remoteproc/Kconfig | 12 +
drivers/remoteproc/Makefile | 1 +
drivers/remoteproc/xlnx_r5_remoteproc.c | 973 ++++++++++++++++++++++++
3 files changed, 986 insertions(+)
create mode 100644 drivers/remoteproc/xlnx_r5_remoteproc.c
diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
index 166019786653..5637a71c0677 100644
--- a/drivers/remoteproc/Kconfig
+++ b/drivers/remoteproc/Kconfig
@@ -352,6 +352,18 @@ config TI_K3_R5_REMOTEPROC
It's safe to say N here if you're not interested in utilizing
a slave processor.
+config XLNX_R5_REMOTEPROC
+ tristate "Xilinx R5 remoteproc support"
+ depends on PM && ARCH_ZYNQMP
+ depends on ZYNQMP_FIRMWARE
+ select RPMSG_VIRTIO
+ select ZYNQMP_IPI_MBOX
+ help
+ Say y or m here to support Xilinx R5 remote processors via the remote
+ processor framework.
+
+ It's safe to say N if not interested in using RPU r5f cores.
+
endif # REMOTEPROC
endmenu
diff --git a/drivers/remoteproc/Makefile b/drivers/remoteproc/Makefile
index 5478c7cb9e07..91314a9b43ce 100644
--- a/drivers/remoteproc/Makefile
+++ b/drivers/remoteproc/Makefile
@@ -38,3 +38,4 @@ obj-$(CONFIG_ST_SLIM_REMOTEPROC) += st_slim_rproc.o
obj-$(CONFIG_STM32_RPROC) += stm32_rproc.o
obj-$(CONFIG_TI_K3_DSP_REMOTEPROC) += ti_k3_dsp_remoteproc.o
obj-$(CONFIG_TI_K3_R5_REMOTEPROC) += ti_k3_r5_remoteproc.o
+obj-$(CONFIG_XLNX_R5_REMOTEPROC) += xlnx_r5_remoteproc.o
diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c
new file mode 100644
index 000000000000..ce2ac364b995
--- /dev/null
+++ b/drivers/remoteproc/xlnx_r5_remoteproc.c
@@ -0,0 +1,973 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * ZynqMP R5 Remote Processor driver
+ *
+ */
+
+#include <dt-bindings/power/xlnx-zynqmp-power.h>
+#include <linux/firmware/xlnx-zynqmp.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/mailbox_client.h>
+#include <linux/mailbox/zynqmp-ipi-message.h>
+#include <linux/module.h>
+#include <linux/of_address.h>
+#include <linux/of_platform.h>
+#include <linux/of_reserved_mem.h>
+#include <linux/platform_device.h>
+#include <linux/remoteproc.h>
+#include <linux/skbuff.h>
+#include <linux/sysfs.h>
+
+#include "remoteproc_internal.h"
+
+/* settings for RPU cluster mode */
+enum zynqmp_r5_cluster_mode {
+ SPLIT_MODE = 0, /* When cores run as separate processor */
+ LOCKSTEP_MODE = 1, /* cores execute same code in lockstep,clk-for-clk */
+ SINGLE_CPU_MODE = 2, /* core0 is held in reset and only core1 runs */
+};
+
+/**
+ * struct mem_bank_data - Memory Bank description
+ *
+ * @addr: Start address of memory bank
+ * @size: Size of Memory bank
+ * @pm_domain_id: Power-domains id of memory bank for firmware to turn on/off
+ * @bank_name: name of the bank for remoteproc framework
+ */
+struct mem_bank_data {
+ phys_addr_t addr;
+ size_t size;
+ u32 pm_domain_id;
+ char *bank_name;
+};
+
+static const struct mem_bank_data zynqmp_tcm_banks[] = {
+ {0xffe00000UL, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */
+ {0xffe20000UL, 0x10000UL, PD_R5_0_BTCM, "btcm0"},
+ {0xffe90000UL, 0x10000UL, PD_R5_1_ATCM, "atcm1"},
+ {0xffeb0000UL, 0x10000UL, PD_R5_1_BTCM, "btcm1"},
+};
+
+/**
+ * struct zynqmp_r5_core - ZynqMP R5 core structure
+ *
+ * @dev: device of RPU instance
+ * @np: device node of RPU instance
+ * @tcm_bank_count: number TCM banks accessible to this RPU
+ * @tcm_banks: array of each TCM bank data
+ * @rproc: rproc handle
+ * @pm_domain_id: RPU CPU power domain id
+ */
+struct zynqmp_r5_core {
+ struct device *dev;
+ struct device_node *np;
+ int tcm_bank_count;
+ struct mem_bank_data *tcm_banks;
+ struct rproc *rproc;
+ u32 pm_domain_id;
+};
+
+/**
+ * struct zynqmp_r5_cluster - ZynqMP R5 cluster structure
+ *
+ * @dev: r5f subsystem cluster device node
+ * @mode: cluster mode of type zynqmp_r5_cluster_mode
+ * @core_count: number of r5 cores used for this cluster mode
+ * @r5_cores: Array of pointers pointing to r5 core
+ */
+struct zynqmp_r5_cluster {
+ struct device *dev;
+ enum zynqmp_r5_cluster_mode mode;
+ int core_count;
+ struct zynqmp_r5_core **r5_cores;
+};
+
+/*
+ * zynqmp_r5_set_mode - set RPU operation mode
+ *
+ * set RPU operation mode
+ *
+ * Return: 0 for success, negative value for failure
+ */
+static int zynqmp_r5_set_mode(struct zynqmp_r5_core *r5_core,
+ enum zynqmp_r5_cluster_mode rpu_mode)
+{
+ enum rpu_tcm_comb tcm_mode;
+ int ret, reg_val;
+
+ /*
+ * SINGLE CPU mode is not supported in driver yet.
+ * For now we only set lockstep mode or split mode.
+ */
+ if (rpu_mode == LOCKSTEP_MODE) {
+ reg_val = 0;
+ } else if (rpu_mode == SPLIT_MODE) {
+ reg_val = 1;
+ } else if (rpu_mode == SINGLE_CPU_MODE) {
+ pr_err("RPU single cpu mode is not supported\n");
+ return -EINVAL;
+ }
+
+ ret = zynqmp_pm_set_rpu_mode(r5_core->pm_domain_id, reg_val);
+ if (ret < 0) {
+ pr_err("failed to set RPU mode\n");
+ return ret;
+ }
+
+ /*
+ * At this point RPU is assumed to be configured only in
+ * lockstep mode or split mode. Configure TCM accordingly.
+ */
+ tcm_mode = (rpu_mode == LOCKSTEP_MODE) ?
+ PM_RPU_TCM_COMB : PM_RPU_TCM_SPLIT;
+ ret = zynqmp_pm_set_tcm_config(r5_core->pm_domain_id, tcm_mode);
+ if (ret < 0)
+ pr_err("failed to configure TCM\n");
+
+ return ret;
+}
+
+/*
+ * zynqmp_r5_rproc_start
+ * @rproc: single R5 core's corresponding rproc instance
+ *
+ * Start R5 Core from designated boot address.
+ *
+ * return 0 on success, otherwise non-zero value on failure
+ */
+static int zynqmp_r5_rproc_start(struct rproc *rproc)
+{
+ struct zynqmp_r5_core *r5_core = rproc->priv;
+ enum rpu_boot_mem bootmem;
+ int ret;
+
+ /*
+ * The exception vector pointers (EVP) refer to the base-address of
+ * exception vectors (for reset, IRQ, FIQ, etc). The reset-vector
+ * starts at the base-address and subsequent vectors are on 4-byte
+ * boundaries.
+ *
+ * Exception vectors can start either from 0x0000_0000 (LOVEC) or
+ * from 0xFFFF_0000 (HIVEC) which is mapped in the OCM (On-Chip Memory)
+ *
+ * Usually firmware will put Exception vectors at LOVEC.
+ *
+ * It is not recommend that you change the exception vector.
+ * Changing the EVP to HIVEC will result in increased interrupt latency
+ * and jitter. Also, if the OCM is secured and the Cortex-R5F processor
+ * is non-secured, then the Cortex-R5F processor cannot access the
+ * HIVEC exception vectors in the OCM.
+ */
+ bootmem = (rproc->bootaddr >= 0xFFFC0000) ?
+ PM_RPU_BOOTMEM_HIVEC : PM_RPU_BOOTMEM_LOVEC;
+
+ dev_dbg(r5_core->dev, "RPU boot addr 0x%llx from %s.", rproc->bootaddr,
+ bootmem == PM_RPU_BOOTMEM_HIVEC ? "OCM" : "TCM");
+
+ ret = zynqmp_pm_request_wake(r5_core->pm_domain_id, 1,
+ bootmem, ZYNQMP_PM_REQUEST_ACK_NO);
+ if (ret)
+ pr_err("failed to start RPU = 0x%x\n", r5_core->pm_domain_id);
+ return ret;
+}
+
+/*
+ * zynqmp_r5_rproc_stop
+ * @rproc: single R5 core's corresponding rproc instance
+ *
+ * Power down R5 Core.
+ *
+ * return 0 on success, otherwise non-zero value on failure
+ */
+static int zynqmp_r5_rproc_stop(struct rproc *rproc)
+{
+ struct zynqmp_r5_core *r5_core = rproc->priv;
+ int ret;
+
+ ret = zynqmp_pm_force_pwrdwn(r5_core->pm_domain_id,
+ ZYNQMP_PM_REQUEST_ACK_BLOCKING);
+ if (ret)
+ pr_err("failed to stop remoteproc RPU %d\n", ret);
+
+ return ret;
+}
+
+/*
+ * zynqmp_r5_rproc_mem_map
+ * @rproc: single R5 core's corresponding rproc instance
+ * @mem: mem entry to map
+ *
+ * Callback to map va for memory-region's carveout.
+ *
+ * return 0 on success, otherwise non-zero value on failure
+ */
+static int zynqmp_r5_rproc_mem_map(struct rproc *rproc,
+ struct rproc_mem_entry *mem)
+{
+ void __iomem *va;
+
+ va = ioremap_wc(mem->dma, mem->len);
+ if (IS_ERR_OR_NULL(va))
+ return -ENOMEM;
+
+ mem->va = (void *)va;
+
+ return 0;
+}
+
+/*
+ * zynqmp_r5_rproc_mem_unmap
+ * @rproc: single R5 core's corresponding rproc instance
+ * @mem: mem entry to unmap
+ *
+ * Unmap memory-region carveout
+ *
+ * return 0 on success, otherwise non-zero value on failure
+ */
+static int zynqmp_r5_rproc_mem_unmap(struct rproc *rproc,
+ struct rproc_mem_entry *mem)
+{
+ iounmap((void __iomem *)mem->va);
+ return 0;
+}
+
+/*
+ * add_mem_regions
+ * @r5_core: single R5 core's corresponding zynqmp_r5_core type instance
+ * @rmem: reserved mem region parsed from dt node
+ *
+ * Construct rproc mem carveouts from carveout provided in
+ * memory-region property
+ *
+ * return 0 on success, otherwise non-zero value on failure
+ */
+static int add_mem_regions_carveout(struct zynqmp_r5_core *r5_core,
+ struct reserved_mem *rmem)
+{
+ struct device *dev;
+ struct rproc_mem_entry *mem;
+ struct rproc *rproc;
+
+ rproc = r5_core->rproc;
+ dev = r5_core->dev;
+
+ /* Register associated reserved memory regions */
+ mem = rproc_mem_entry_init(dev, NULL,
+ (dma_addr_t)rmem->base,
+ rmem->size, rmem->base,
+ zynqmp_r5_rproc_mem_map,
+ zynqmp_r5_rproc_mem_unmap,
+ rmem->name);
+ if (!mem)
+ return -ENOMEM;
+
+ rproc_add_carveout(rproc, mem);
+
+ return 0;
+}
+
+/*
+ * zynqmp_r5_rproc_mem_unmap
+ * @rproc: single R5 core's corresponding rproc instance
+ * @mem: mem entry to unmap
+ *
+ * Unmap TCM banks when powering down R5 core.
+ *
+ * return 0 on success, otherwise non-zero value on failure
+ */
+static int tcm_mem_unmap(struct rproc *rproc, struct rproc_mem_entry *mem)
+{
+ struct zynqmp_r5_core *r5_core;
+ int i;
+ u32 pm_domain_id;
+
+ r5_core = rproc->priv;
+ if (!r5_core) {
+ pr_err("r5 core is not available\n");
+ return -EINVAL;
+ }
+
+ iounmap((void __iomem *)mem->va);
+
+ for (i = 0; i < r5_core->tcm_bank_count; i++) {
+ pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
+ if (zynqmp_pm_release_node(pm_domain_id))
+ pr_warn("can't turn off TCM bank 0x%x", pm_domain_id);
+ }
+
+ return 0;
+}
+
+/*
+ * tcm_mem_map
+ * @rproc: single R5 core's corresponding rproc instance
+ * @mem: mem entry to initialize the va and da fields of
+ *
+ * Given TCM bank entry, this callback will set device address for R5
+ * running on TCM and also setup virtual address for TCM bank
+ * remoteproc carveout.
+ *
+ * return 0 on success, otherwise non-zero value on failure
+ */
+static int tcm_mem_map(struct rproc *rproc,
+ struct rproc_mem_entry *mem)
+{
+ void __iomem *va;
+
+ va = ioremap_wc(mem->dma, mem->len);
+ if (IS_ERR_OR_NULL(va))
+ return -ENOMEM;
+
+ /* Update memory entry va */
+ mem->va = (void *)va;
+
+ /* clear TCMs */
+ memset_io(va, 0, mem->len);
+
+ /*
+ * The R5s expect their TCM banks to be at address 0x0 and 0x2000,
+ * while on the Linux side they are at 0xffexxxxx.
+ *
+ * Zero out the high 12 bits of the address. This will give
+ * expected values for TCM Banks 0A and 0B (0x0 and 0x20000).
+ */
+ mem->da &= 0x000fffff;
+
+ /*
+ * TCM Banks 1A and 1B still have to be translated.
+ *
+ * Below handle these two banks' absolute addresses (0xffe90000 and
+ * 0xffeb0000) and convert to the expected relative addresses
+ * (0x0 and 0x20000).
+ */
+ if (mem->da == 0x90000 || mem->da == 0xB0000)
+ mem->da -= 0x90000;
+
+ /* if translated TCM bank address is not valid report error */
+ if (mem->da != 0x0 && mem->da != 0x20000) {
+ dev_err(&rproc->dev, "invalid TCM address: %x\n", mem->da);
+ return -EINVAL;
+ }
+ return 0;
+}
+
+static int add_tcm_carveout_split_mode(struct rproc *rproc)
+{
+ int i, num_banks, ret;
+ struct rproc_mem_entry *mem;
+ u32 pm_domain_id;
+ phys_addr_t bank_addr;
+ size_t bank_size;
+ char *bank_name;
+ struct device *dev;
+ struct zynqmp_r5_core *r5_core;
+
+ r5_core = (struct zynqmp_r5_core *)rproc->priv;
+
+ dev = r5_core->dev;
+
+ /* go through zynqmp banks for r5 node */
+ num_banks = r5_core->tcm_bank_count;
+ if (num_banks <= 0) {
+ dev_err(dev, "need to specify TCM banks\n");
+ return -EINVAL;
+ }
+
+ /*
+ * Power-on Each 64KB TCM,
+ * register its address space, map and unmap functions
+ * and add carveouts accordingly
+ */
+ for (i = 0; i < num_banks; i++) {
+ bank_addr = r5_core->tcm_banks[i].addr;
+ bank_name = r5_core->tcm_banks[i].bank_name;
+ bank_size = r5_core->tcm_banks[i].size;
+ pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
+
+ ret = zynqmp_pm_request_node(pm_domain_id,
+ ZYNQMP_PM_CAPABILITY_ACCESS, 0,
+ ZYNQMP_PM_REQUEST_ACK_BLOCKING);
+ if (ret < 0) {
+ dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
+ return ret;
+ }
+
+ dev_dbg(dev, "TCM carveout split mode %s addr=%llx, size=0x%lx",
+ bank_name, bank_addr, bank_size);
+
+ mem = rproc_mem_entry_init(dev, NULL, bank_addr,
+ bank_size, bank_addr,
+ tcm_mem_map, tcm_mem_unmap,
+ bank_name);
+ if (!mem) {
+ /* If failed, Turn off all TCM banks turned on before */
+ do {
+ pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
+ ret = zynqmp_pm_release_node(pm_domain_id);
+ if (ret)
+ dev_warn(dev,
+ "fail to release node: 0x%x, %x\n",
+ pm_domain_id, ret);
+ } while (i--);
+ return -ENOMEM;
+ }
+
+ rproc_add_carveout(rproc, mem);
+ }
+
+ return 0;
+}
+
+static int add_tcm_carveout_lockstep_mode(struct rproc *rproc)
+{
+ int i, num_banks, ret;
+ struct rproc_mem_entry *mem;
+ u32 pm_domain_id;
+ phys_addr_t bank_addr;
+ size_t bank_size = 0;
+ char *bank_name;
+ struct device *dev;
+ struct zynqmp_r5_core *r5_core;
+
+ r5_core = (struct zynqmp_r5_core *)rproc->priv;
+ dev = r5_core->dev;
+
+ /* Go through zynqmp banks for r5 node */
+ num_banks = r5_core->tcm_bank_count;
+ if (num_banks <= 0) {
+ dev_err(dev, "need to specify TCM banks\n");
+ return -EINVAL;
+ }
+
+ /*
+ * In lockstep mode, TCM is contiguous memory block
+ * However, each TCM block still needs to be enabled individually.
+ * So, Enable each TCM block individually, but add their size
+ * to create contiguous memory region.
+ */
+ bank_addr = r5_core->tcm_banks[0].addr;
+ bank_name = r5_core->tcm_banks[0].bank_name;
+
+ for (i = 0; i < num_banks; i++) {
+ bank_size += r5_core->tcm_banks[i].size;
+ pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
+
+ /* Turn on each TCM bank individually */
+ ret = zynqmp_pm_request_node(pm_domain_id,
+ ZYNQMP_PM_CAPABILITY_ACCESS, 0,
+ ZYNQMP_PM_REQUEST_ACK_BLOCKING);
+ if (ret < 0) {
+ dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
+ return ret;
+ }
+ }
+
+ dev_dbg(dev, "TCM add carveout lockstep mode %s addr=0x%llx, size=0x%lx",
+ bank_name, bank_addr, bank_size);
+
+ /* Register TCM address range, TCM map and unmap functions */
+ mem = rproc_mem_entry_init(dev, NULL, bank_addr,
+ bank_size, bank_addr,
+ tcm_mem_map, tcm_mem_unmap,
+ bank_name);
+ if (!mem) {
+ /* In case of error, turn off TCM banks */
+ for (i = 0; i < num_banks; i++) {
+ pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
+ ret = zynqmp_pm_release_node(pm_domain_id);
+ if (ret)
+ dev_warn(dev,
+ "fail to release node: 0x%x ret: %x\n",
+ pm_domain_id, ret);
+ }
+ return -ENOMEM;
+ }
+
+ /* If registration is success, add carveouts */
+ rproc_add_carveout(rproc, mem);
+
+ return 0;
+}
+
+/*
+ * add_tcm_banks()
+ * @rproc: single R5 core's corresponding rproc instance
+ *
+ * Given R5 node in remoteproc instance
+ * allocate remoteproc carveout for TCM memory
+ * needed for firmware to be loaded
+ *
+ * return 0 on success, otherwise non-zero value on failure
+ */
+static int add_tcm_banks(struct rproc *rproc)
+{
+ struct device *dev;
+ struct platform_device *parent_pdev;
+ struct zynqmp_r5_cluster *cluster;
+ struct zynqmp_r5_core *r5_core;
+
+ r5_core = (struct zynqmp_r5_core *)rproc->priv;
+ if (!r5_core)
+ return -EINVAL;
+
+ dev = r5_core->dev;
+
+ cluster = dev_get_drvdata(dev->parent);
+ if (!cluster) {
+ dev_err(&parent_pdev->dev, "Invalid driver data\n");
+ return -EINVAL;
+ }
+
+ /*
+ * In lockstep mode TCM banks are one contiguous memory region of 256Kb
+ * In split mode, each TCM bank is 64Kb and not contiguous.
+ * We add memory carveouts accordingly.
+ */
+ if (cluster->mode == SPLIT_MODE)
+ return add_tcm_carveout_split_mode(rproc);
+ else if (cluster->mode == LOCKSTEP_MODE)
+ return add_tcm_carveout_lockstep_mode(rproc);
+
+ dev_err(cluster->dev, "invalid cluster mode\n");
+ return -EINVAL;
+}
+
+/*
+ * zynqmp_r5_parse_fw()
+ * @rproc: single R5 core's corresponding rproc instance
+ * @fw: ptr to firmware to be loaded onto r5 core
+ *
+ * When loading firmware, ensure the necessary carveouts are in remoteproc
+ *
+ * return 0 on success, otherwise non-zero value on failure
+ */
+static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct firmware *fw)
+{
+ int ret;
+ struct zynqmp_r5_core *r5_core;
+ struct device *dev;
+
+ r5_core = rproc->priv;
+
+ dev = r5_core->dev;
+
+ ret = add_tcm_banks(rproc);
+ if (ret) {
+ dev_err(dev, "failed to get TCM banks, err %d\n", ret);
+ return ret;
+ }
+
+ ret = rproc_elf_load_rsc_table(rproc, fw);
+ if (ret == -EINVAL) {
+ /*
+ * resource table only required for IPC.
+ * if not present, this is not necessarily an error;
+ * for example, loading r5 hello world application
+ * so simply inform user and keep going.
+ */
+ dev_info(&rproc->dev, "no resource table found.\n");
+ ret = 0;
+ }
+ return ret;
+}
+
+static struct rproc_ops zynqmp_r5_rproc_ops = {
+ .start = zynqmp_r5_rproc_start,
+ .stop = zynqmp_r5_rproc_stop,
+ .load = rproc_elf_load_segments,
+ .parse_fw = zynqmp_r5_parse_fw,
+ .find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
+ .sanity_check = rproc_elf_sanity_check,
+ .get_boot_addr = rproc_elf_get_boot_addr,
+};
+
+/**
+ * zynqmp_r5_add_rproc_core()
+ * Probes ZynqMP R5 processor device node
+ * this is called for each individual R5 core
+ *
+ * @cdev: Device node of each r5 core
+ *
+ * Return: zynqmp_r5_core object for success, error pointer in case of error.
+ */
+static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev)
+{
+ int ret;
+ struct rproc *r5_rproc;
+ struct zynqmp_r5_core *r5_core;
+
+ /* Set up DMA mask */
+ ret = dma_set_coherent_mask(cdev, DMA_BIT_MASK(32));
+ if (ret)
+ return ERR_PTR(ret);
+
+ /* Allocate remoteproc instance */
+ r5_rproc = devm_rproc_alloc(cdev, dev_name(cdev),
+ &zynqmp_r5_rproc_ops,
+ NULL, sizeof(struct zynqmp_r5_core));
+ if (!r5_rproc) {
+ dev_err(cdev, "failed to allocate memory for rproc instance\n");
+ return ERR_PTR(-ENOMEM);
+ }
+
+ r5_rproc->auto_boot = false;
+ r5_core = (struct zynqmp_r5_core *)r5_rproc->priv;
+ r5_core->dev = cdev;
+
+ /* Add R5 remoteproc core */
+ ret = devm_rproc_add(cdev, r5_rproc);
+ if (ret) {
+ dev_err(cdev, "failed to add r5 remoteproc\n");
+ return ERR_PTR(ret);
+ }
+
+ r5_core->rproc = r5_rproc;
+ return r5_core;
+}
+
+/**
+ * zynqmp_r5_get_tcm_node()
+ * Ideally this function should parse tcm node and store information
+ * in r5_core instance. We will use hardcoded TCM information from
+ * driver for now in this function.
+ *
+ * @cluster: pointer to zynqmp_r5_cluster type object
+ *
+ * Return: 0 for success and error code for failure.
+ */
+static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster)
+{
+ int tcm_bank_count, tcm_node;
+ int i = 0, j;
+ struct zynqmp_r5_core *r5_core;
+ const struct mem_bank_data *tcm = zynqmp_tcm_banks;
+ struct device *dev = cluster->dev;
+
+ /*
+ * ToDo: Use predefined TCM address space values from driver until
+ * system-dt spec is not final for TCM
+ */
+ tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks);
+
+ /* count per core tcm banks */
+ tcm_bank_count = tcm_bank_count / cluster->core_count;
+
+ /* r5 core 0 will use all of TCM banks in lockstep mode.
+ * In split mode, r5 core0 will use 128k and r5 core1 will use another
+ * 128k. Assign TCM banks to each core accordingly
+ */
+ tcm_node = 0;
+ for (i = 0; i < cluster->core_count; i++) {
+ r5_core = cluster->r5_cores[i];
+ r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count,
+ sizeof(struct mem_bank_data),
+ GFP_KERNEL);
+ if (!r5_core->tcm_banks)
+ return -ENOMEM;
+
+ for (j = 0; j < tcm_bank_count; j++) {
+ /* Use pre-defined TCM reg values.
+ * Eventually this should be replaced by values
+ * parsed from dts.
+ */
+ r5_core->tcm_banks[j].addr = tcm[tcm_node].addr;
+ r5_core->tcm_banks[j].size = tcm[tcm_node].size;
+ r5_core->tcm_banks[j].pm_domain_id = tcm[tcm_node].pm_domain_id;
+ r5_core->tcm_banks[j].bank_name = tcm[tcm_node].bank_name;
+ tcm_node++;
+ }
+
+ r5_core->tcm_bank_count = tcm_bank_count;
+ }
+
+ return 0;
+}
+
+/**
+ * zynqmp_r5_get_mem_region_node()
+ * parse memory-region property from dt node and add
+ * memory region carveouts
+ *
+ * @r5_core: pointer to zynqmp_r5_core type object
+ *
+ * Return: 0 for success and error code for failure.
+ */
+static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core *r5_core)
+{
+ int res_mem_count, i, ret;
+ struct device *dev;
+ struct device_node *np, *rmem_np;
+ struct reserved_mem *rmem;
+
+ dev = r5_core->dev;
+
+ np = r5_core->np;
+
+ res_mem_count = of_property_count_elems_of_size(np, "memory-region",
+ sizeof(phandle));
+ if (res_mem_count <= 0) {
+ dev_warn(dev, "failed to get memory-region property %d\n",
+ res_mem_count);
+ return -EINVAL;
+ }
+
+ for (i = 0; i < res_mem_count; i++) {
+ rmem_np = of_parse_phandle(np, "memory-region", i);
+ if (!rmem_np)
+ return -EINVAL;
+
+ rmem = of_reserved_mem_lookup(rmem_np);
+ if (!rmem) {
+ of_node_put(rmem_np);
+ return -EINVAL;
+ }
+
+ ret = add_mem_regions_carveout(r5_core, rmem);
+ if (ret)
+ dev_warn(dev, "failed to get reserve mem regions %d\n",
+ ret);
+
+ of_node_put(rmem_np);
+ }
+
+ return 0;
+}
+
+/*
+ * zynqmp_r5_core_init()
+ * Create and initialize zynqmp_r5_core type object
+ *
+ * @cluster: pointer to zynqmp_r5_cluster type object
+ *
+ * Return: 0 for success and error code for failure.
+ */
+static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
+{
+ int ret, i;
+ struct zynqmp_r5_core *r5_core;
+ struct device *dev = cluster->dev;
+
+ ret = zynqmp_r5_get_tcm_node(cluster);
+ if (ret < 0) {
+ dev_err(dev, "can't get tcm node, err %d\n", ret);
+ return ret;
+ }
+
+ for (i = 0; i < cluster->core_count; i++) {
+ r5_core = cluster->r5_cores[i];
+
+ ret = zynqmp_r5_get_mem_region_node(r5_core);
+ if (ret)
+ dev_warn(dev, "memory-region prop failed %d\n", ret);
+
+ /* Initialize r5 cores with power-domains parsed from dts */
+ ret = of_property_read_u32_index(r5_core->np, "power-domains",
+ 1, &r5_core->pm_domain_id);
+ if (ret) {
+ dev_err(dev, "failed to get power-domains property\n");
+ return ret;
+ }
+
+ ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
+ if (ret) {
+ dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
+ cluster->mode, ret);
+ return ret;
+ }
+ }
+
+ return 0;
+}
+
+/*
+ * zynqmp_r5_cluster_init()
+ * Create and initialize zynqmp_r5_cluster type object
+ *
+ * @cluster: pointer to zynqmp_r5_cluster type object
+ *
+ * Return: 0 for success and error code for failure.
+ */
+static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
+{
+ struct device *dev = cluster->dev;
+ struct device_node *dev_node = dev_of_node(dev);
+ struct device_node *child;
+ struct platform_device *child_pdev;
+ int core_count = 0, ret, i;
+ enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
+ struct zynqmp_r5_core **r5_cores;
+
+ ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
+
+ /*
+ * on success returns 0, if not defined then returns -EINVAL,
+ * In that case, default is LOCKSTEP mode
+ */
+ if (ret != -EINVAL && ret != 0) {
+ dev_err(dev, "Invalid xlnx,cluster-mode property\n");
+ return -EINVAL;
+ }
+
+ /*
+ * For now driver only supports split mode and lockstep mode.
+ * fail driver probe if either of that is not set in dts
+ */
+ if (cluster_mode == SINGLE_CPU_MODE) {
+ dev_err(dev, "driver does not support single cpu mode\n");
+ return -EINVAL;
+ } else if ((cluster_mode != SPLIT_MODE &&
+ cluster_mode != LOCKSTEP_MODE)) {
+ dev_err(dev, "Invalid cluster mode\n");
+ return -EINVAL;
+ }
+
+ /*
+ * Number of cores is decided by number of child nodes of
+ * r5f subsystem node in dts. If Split mode is used in dts
+ * 2 child nodes are expected.
+ * In lockstep mode if two child nodes are available,
+ * only use first child node and consider it as core0
+ * and ignore core1 dt node.
+ */
+ core_count = of_get_available_child_count(dev_node);
+ if (core_count <= 0) {
+ dev_err(dev, "Invalid number of r5 cores %d", core_count);
+ return -EINVAL;
+ } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
+ dev_err(dev, "Invalid number of r5 cores for split mode\n");
+ return -EINVAL;
+ } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
+ dev_warn(dev, "Only r5 core0 will be used\n");
+ core_count = 1;
+ }
+
+ r5_cores = devm_kcalloc(dev, core_count,
+ sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
+ if (!r5_cores)
+ return -ENOMEM;
+
+ i = 0;
+ for_each_available_child_of_node(dev_node, child) {
+ child_pdev = of_find_device_by_node(child);
+ if (!child_pdev) {
+ of_node_put(child);
+ return -ENODEV;
+ }
+
+ /* create and add remoteproc instance of type struct rproc */
+ r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
+ r5_cores[i]->dev = &child_pdev->dev;
+ if (!r5_cores[i]->dev) {
+ dev_err(dev, "can't get device for r5 core %d\n", i);
+ of_node_put(child);
+ return -ENODEV;
+ }
+
+ r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
+ if (!r5_cores[i]->np) {
+ dev_err(dev, "can't get device node for r5 core %d\n", i);
+ of_node_put(child);
+ return -ENODEV;
+ }
+
+ i++;
+
+ /*
+ * If two child nodes are available in dts in lockstep mode,
+ * then ignore second child node.
+ */
+ if (i == core_count) {
+ of_node_put(child);
+ break;
+ }
+ of_node_put(child);
+ }
+
+ cluster->mode = cluster_mode;
+ cluster->core_count = core_count;
+ cluster->r5_cores = r5_cores;
+
+ ret = zynqmp_r5_core_init(cluster);
+ if (ret < 0) {
+ dev_err(dev, "failed to init r5 core err %d\n", ret);
+ return ret;
+ }
+
+ return 0;
+}
+
+static void zynqmp_r5_cluster_exit(void *data)
+{
+ struct platform_device *pdev = (struct platform_device *)data;
+
+ platform_set_drvdata(pdev, NULL);
+
+ pr_info("Exit r5f subsystem driver\n");
+}
+
+/*
+ * zynqmp_r5_remoteproc_probe()
+ *
+ * @pdev: domain platform device for R5 cluster
+ *
+ * called when driver is probed, for each R5 core specified in DT,
+ * setup as needed to do remoteproc-related operations
+ *
+ * Return: 0 for success, negative value for failure.
+ */
+static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
+{
+ int ret;
+ struct zynqmp_r5_cluster *cluster;
+ struct device *dev = &pdev->dev;
+
+ cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
+ if (!cluster)
+ return -ENOMEM;
+
+ cluster->dev = dev;
+
+ ret = devm_of_platform_populate(dev);
+ if (ret) {
+ dev_err(dev, "failed to populate platform dev %d\n", ret);
+ return ret;
+ }
+
+ /* wire in so each core can be cleaned up at driver remove */
+ platform_set_drvdata(pdev, cluster);
+
+ ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
+ if (ret)
+ return ret;
+
+ ret = zynqmp_r5_cluster_init(cluster);
+ if (ret) {
+ dev_err(dev, "Invalid r5f subsystem device tree\n");
+ return ret;
+ }
+
+ return 0;
+}
+
+/* Match table for OF platform binding */
+static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
+ { .compatible = "xlnx,zynqmp-r5fss", },
+ { /* end of list */ },
+};
+MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
+
+static struct platform_driver zynqmp_r5_remoteproc_driver = {
+ .probe = zynqmp_r5_remoteproc_probe,
+ .driver = {
+ .name = "zynqmp_r5_remoteproc",
+ .of_match_table = zynqmp_r5_remoteproc_match,
+ },
+};
+module_platform_driver(zynqmp_r5_remoteproc_driver);
+
+MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
+MODULE_AUTHOR("Xilinx Inc.");
+MODULE_LICENSE("GPL v2");
--
2.25.1
From: Ben Levinsky <[email protected]>
This patch adds APIs to access to configure RPU and its
processor-specific memory.
That is query the run-time mode of RPU as either split or lockstep as well
as API to set this mode. In addition add APIs to access configuration of
the RPUs' tightly coupled memory (TCM).
Signed-off-by: Ben Levinsky <[email protected]>
Signed-off-by: Tanmay Shah <[email protected]>
---
Changes in v3:
- Add missing function argument documentation
drivers/firmware/xilinx/zynqmp.c | 62 ++++++++++++++++++++++++++++
include/linux/firmware/xlnx-zynqmp.h | 18 ++++++++
2 files changed, 80 insertions(+)
diff --git a/drivers/firmware/xilinx/zynqmp.c b/drivers/firmware/xilinx/zynqmp.c
index 7928d52c8ade..0b28c0cf1a82 100644
--- a/drivers/firmware/xilinx/zynqmp.c
+++ b/drivers/firmware/xilinx/zynqmp.c
@@ -1057,6 +1057,68 @@ int zynqmp_pm_release_node(const u32 node)
}
EXPORT_SYMBOL_GPL(zynqmp_pm_release_node);
+/**
+ * zynqmp_pm_get_rpu_mode() - Get RPU mode
+ * @node_id: Node ID of the device
+ * @rpu_mode: return by reference value
+ * either split or lockstep
+ *
+ * Return: return 0 on success or error+reason.
+ * if success, then rpu_mode will be set
+ * to current rpu mode.
+ */
+int zynqmp_pm_get_rpu_mode(u32 node_id, enum rpu_oper_mode *rpu_mode)
+{
+ u32 ret_payload[PAYLOAD_ARG_CNT];
+ int ret;
+
+ ret = zynqmp_pm_invoke_fn(PM_IOCTL, node_id,
+ IOCTL_GET_RPU_OPER_MODE, 0, 0, ret_payload);
+
+ /* only set rpu_mode if no error */
+ if (ret == XST_PM_SUCCESS)
+ *rpu_mode = ret_payload[0];
+
+ return ret;
+}
+EXPORT_SYMBOL_GPL(zynqmp_pm_get_rpu_mode);
+
+/**
+ * zynqmp_pm_set_rpu_mode() - Set RPU mode
+ * @node_id: Node ID of the device
+ * @rpu_mode: Argument 1 to requested IOCTL call. either split or lockstep
+ *
+ * This function is used to set RPU mode to split or
+ * lockstep
+ *
+ * Return: Returns status, either success or error+reason
+ */
+int zynqmp_pm_set_rpu_mode(u32 node_id, enum rpu_oper_mode rpu_mode)
+{
+ return zynqmp_pm_invoke_fn(PM_IOCTL, node_id,
+ IOCTL_SET_RPU_OPER_MODE, (u32)rpu_mode,
+ 0, NULL);
+}
+EXPORT_SYMBOL_GPL(zynqmp_pm_set_rpu_mode);
+
+/**
+ * zynqmp_pm_set_tcm_config - configure TCM
+ * @node_id: Firmware specific TCM subsystem ID
+ * @tcm_mode: Argument 1 to requested IOCTL call
+ * either PM_RPU_TCM_COMB or PM_RPU_TCM_SPLIT
+ *
+ * This function is used to set RPU mode to split or combined
+ *
+ * Return: status: 0 for success, else failure
+ */
+int zynqmp_pm_set_tcm_config(u32 node_id, enum rpu_tcm_comb tcm_mode)
+{
+ return zynqmp_pm_invoke_fn(PM_IOCTL, node_id,
+ IOCTL_TCM_COMB_CONFIG, (u32)tcm_mode, 0,
+ NULL);
+}
+EXPORT_SYMBOL_GPL(zynqmp_pm_set_tcm_config);
+
/**
* zynqmp_pm_force_pwrdwn - PM call to request for another PU or subsystem to
* be powered down forcefully
diff --git a/include/linux/firmware/xlnx-zynqmp.h b/include/linux/firmware/xlnx-zynqmp.h
index 94ba5a40f6e2..047798405cb3 100644
--- a/include/linux/firmware/xlnx-zynqmp.h
+++ b/include/linux/firmware/xlnx-zynqmp.h
@@ -475,6 +475,9 @@ int zynqmp_pm_request_wake(const u32 node,
const bool set_addr,
const u64 address,
const enum zynqmp_pm_request_ack ack);
+int zynqmp_pm_get_rpu_mode(u32 node_id, enum rpu_oper_mode *rpu_mode);
+int zynqmp_pm_set_rpu_mode(u32 node_id, u32 arg1);
+int zynqmp_pm_set_tcm_config(u32 node_id, u32 arg1);
#else
static inline int zynqmp_pm_get_api_version(u32 *version)
{
@@ -731,6 +734,21 @@ static inline int zynqmp_pm_request_wake(const u32 node,
{
return -ENODEV;
}
+
+static inline int zynqmp_pm_get_rpu_mode(u32 node_id, enum rpu_oper_mode *rpu_mode)
+{
+ return -ENODEV;
+}
+
+static inline int zynqmp_pm_set_rpu_mode(u32 node_id, u32 arg1)
+{
+ return -ENODEV;
+}
+
+static inline int zynqmp_pm_set_tcm_config(u32 node_id, u32 arg1)
+{
+ return -ENODEV;
+}
#endif
#endif /* __FIRMWARE_ZYNQMP_H__ */
--
2.25.1
From: Ben Levinsky <[email protected]>
Add ZynqMP firmware ioctl enums for RPU configuration and TCM Nodes for
later use via request_node and release_node
Signed-off-by: Ben Levinsky <[email protected]>
Signed-off-by: Tanmay Shah <[email protected]>
---
Changes in v3:
- None
include/linux/firmware/xlnx-zynqmp.h | 19 +++++++++++++++++++
1 file changed, 19 insertions(+)
diff --git a/include/linux/firmware/xlnx-zynqmp.h b/include/linux/firmware/xlnx-zynqmp.h
index 907cb01890cf..0834b00ee2b1 100644
--- a/include/linux/firmware/xlnx-zynqmp.h
+++ b/include/linux/firmware/xlnx-zynqmp.h
@@ -128,6 +128,10 @@ enum pm_ret_status {
};
enum pm_ioctl_id {
+ IOCTL_GET_RPU_OPER_MODE = 0,
+ IOCTL_SET_RPU_OPER_MODE = 1,
+ IOCTL_RPU_BOOT_ADDR_CONFIG = 2,
+ IOCTL_TCM_COMB_CONFIG = 3,
IOCTL_SD_DLL_RESET = 6,
IOCTL_SET_SD_TAPDELAY = 7,
IOCTL_SET_PLL_FRAC_MODE = 8,
@@ -162,6 +166,21 @@ enum pm_query_id {
PM_QID_CLOCK_GET_MAX_DIVISOR = 13,
};
+enum rpu_oper_mode {
+ PM_RPU_MODE_LOCKSTEP = 0,
+ PM_RPU_MODE_SPLIT = 1,
+};
+
+enum rpu_boot_mem {
+ PM_RPU_BOOTMEM_LOVEC = 0,
+ PM_RPU_BOOTMEM_HIVEC = 1,
+};
+
+enum rpu_tcm_comb {
+ PM_RPU_TCM_SPLIT = 0,
+ PM_RPU_TCM_COMB = 1,
+};
+
enum zynqmp_pm_reset_action {
PM_RESET_ACTION_RELEASE = 0,
PM_RESET_ACTION_ASSERT = 1,
--
2.25.1
From: Ben Levinsky <[email protected]>
Add shutdown/wakeup a resource eemi operations to shutdown
or bringup a resource.
Note alignment of args matches convention of other fn's in this file.
The reason being that the long fn name results in aligned args that
otherwise go over 80 chars so shift right to avoid this
Signed-off-by: Ben Levinsky <[email protected]>
Signed-off-by: Tanmay Shah <[email protected]>
---
Changes in v3:
- None
drivers/firmware/xilinx/zynqmp.c | 35 ++++++++++++++++++++++++++++
include/linux/firmware/xlnx-zynqmp.h | 23 ++++++++++++++++++
2 files changed, 58 insertions(+)
diff --git a/drivers/firmware/xilinx/zynqmp.c b/drivers/firmware/xilinx/zynqmp.c
index 450c5f6a1cbf..7928d52c8ade 100644
--- a/drivers/firmware/xilinx/zynqmp.c
+++ b/drivers/firmware/xilinx/zynqmp.c
@@ -1057,6 +1057,41 @@ int zynqmp_pm_release_node(const u32 node)
}
EXPORT_SYMBOL_GPL(zynqmp_pm_release_node);
+/**
+ * zynqmp_pm_force_pwrdwn - PM call to request for another PU or subsystem to
+ * be powered down forcefully
+ * @node: Node ID of the targeted PU or subsystem
+ * @ack: Flag to specify whether acknowledge is requested
+ *
+ * Return: status, either success or error+reason
+ */
+int zynqmp_pm_force_pwrdwn(const u32 node,
+ const enum zynqmp_pm_request_ack ack)
+{
+ return zynqmp_pm_invoke_fn(PM_FORCE_POWERDOWN, node, ack, 0, 0, NULL);
+}
+EXPORT_SYMBOL_GPL(zynqmp_pm_force_pwrdwn);
+
+/**
+ * zynqmp_pm_request_wake - PM call to wake up selected master or subsystem
+ * @node: Node ID of the master or subsystem
+ * @set_addr: Specifies whether the address argument is relevant
+ * @address: Address from which to resume when woken up
+ * @ack: Flag to specify whether acknowledge requested
+ *
+ * Return: status, either success or error+reason
+ */
+int zynqmp_pm_request_wake(const u32 node,
+ const bool set_addr,
+ const u64 address,
+ const enum zynqmp_pm_request_ack ack)
+{
+ /* set_addr flag is encoded into 1st bit of address */
+ return zynqmp_pm_invoke_fn(PM_REQUEST_WAKEUP, node, address | set_addr,
+ address >> 32, ack, NULL);
+}
+EXPORT_SYMBOL_GPL(zynqmp_pm_request_wake);
+
/**
* zynqmp_pm_set_requirement() - PM call to set requirement for PM slaves
* @node: Node ID of the slave
diff --git a/include/linux/firmware/xlnx-zynqmp.h b/include/linux/firmware/xlnx-zynqmp.h
index 0834b00ee2b1..94ba5a40f6e2 100644
--- a/include/linux/firmware/xlnx-zynqmp.h
+++ b/include/linux/firmware/xlnx-zynqmp.h
@@ -12,6 +12,7 @@
#ifndef __FIRMWARE_ZYNQMP_H__
#define __FIRMWARE_ZYNQMP_H__
+#include <linux/types.h>
#include <linux/err.h>
@@ -81,6 +82,8 @@ enum pm_api_cb_id {
enum pm_api_id {
PM_GET_API_VERSION = 1,
PM_REGISTER_NOTIFIER = 5,
+ PM_FORCE_POWERDOWN = 8,
+ PM_REQUEST_WAKEUP = 10,
PM_SYSTEM_SHUTDOWN = 12,
PM_REQUEST_NODE = 13,
PM_RELEASE_NODE = 14,
@@ -466,6 +469,12 @@ int zynqmp_pm_load_pdi(const u32 src, const u64 address);
int zynqmp_pm_register_notifier(const u32 node, const u32 event,
const u32 wake, const u32 enable);
int zynqmp_pm_feature(const u32 api_id);
+int zynqmp_pm_force_pwrdwn(const u32 target,
+ const enum zynqmp_pm_request_ack ack);
+int zynqmp_pm_request_wake(const u32 node,
+ const bool set_addr,
+ const u64 address,
+ const enum zynqmp_pm_request_ack ack);
#else
static inline int zynqmp_pm_get_api_version(u32 *version)
{
@@ -666,6 +675,12 @@ static inline int zynqmp_pm_pinctrl_request(const u32 pin)
return -ENODEV;
}
+static inline int zynqmp_pm_force_pwrdwn(const u32 target,
+ const enum zynqmp_pm_request_ack ack)
+{
+ return -ENODEV;
+}
+
static inline int zynqmp_pm_pinctrl_release(const u32 pin)
{
return -ENODEV;
@@ -708,6 +723,14 @@ static inline int zynqmp_pm_feature(const u32 api_id)
{
return -ENODEV;
}
+
+static inline int zynqmp_pm_request_wake(const u32 node,
+ const bool set_addr,
+ const u64 address,
+ const enum zynqmp_pm_request_ack ack)
+{
+ return -ENODEV;
+}
#endif
#endif /* __FIRMWARE_ZYNQMP_H__ */
--
2.25.1
On Thu, Feb 10, 2022 at 03:28:24AM -0800, Tanmay Shah wrote:
> This driver enables r5f dual core Real time Processing Unit subsystem
> available on Xilinx Zynq Ultrascale MPSoC Platform. RPU subsystem
> (cluster) can be configured in different modes e.g. split mode in which
> two r5f cores work independent of each other and lock-step mode in which
> both r5f cores execute same code clock-for-clock and notify if the
> result is different.
>
> The Xilinx r5 Remoteproc Driver boots the RPU cores via calls to the Xilinx
> Platform Management Unit that handles the R5 configuration, memory access
> and R5 lifecycle management. The interface to this manager is done in this
> driver via zynqmp_pm_* function calls.
>
> Signed-off-by: Ben Levinsky <[email protected]>
> Signed-off-by: Tanmay Shah <[email protected]>
> ---
>
> Changes in v3:
> - Fix checkpatch script indentation warning
> - Remove unused variable from xilinx remoteproc driver
> - use C style comments, i.e /*...*/
> - Remove redundant debug information which can be derived using /proc/device-tree
> - Fix multilined comment format
> - s/"final fot TCM"/"final for TCM"
> - Function devm_kzalloc() does not return an code on error, just NULL.
> Remove redundant error check for this function throughout the driver.
> - Fix RPU mode configuration and add documentation accordingly
> - Get rid of the indentations to match function documentation style with rest of the driver
> - Fix memory leak by only using r5_rproc->priv and not replace it with new instance
> - Use 'i' for the outer loop and 'j' for the inner one as per convention
> - Remove redundant error and NULL checks throughout the driver
> - Use devm_kcalloc() when more than one element is required
> - Add memory-regions carveouts during driver probe instead of parse_fw call
> This removes redundant copy of reserved_mem object in r5_core structure.
> - Fix memory leak by using of_node_put()
> - Fix indentation of tcm_mem_map function args
> - Remove redundant init of variables
> - Initialize tcm bank size variable for lockstep mode
> - Replace u32 with phys_addr_t for variable stroing memory bank address
> - Add documentation of TCM behavior in lockstep mode
> - Use dev_get_drvdata instead of platform driver API
> - Remove info level messages
> - Fix checkpatch.pl warnings
> - Add documentation for the Xilinx r5f platform to understand driver design
>
> drivers/remoteproc/Kconfig | 12 +
> drivers/remoteproc/Makefile | 1 +
> drivers/remoteproc/xlnx_r5_remoteproc.c | 973 ++++++++++++++++++++++++
> 3 files changed, 986 insertions(+)
> create mode 100644 drivers/remoteproc/xlnx_r5_remoteproc.c
>
> diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
> index 166019786653..5637a71c0677 100644
> --- a/drivers/remoteproc/Kconfig
> +++ b/drivers/remoteproc/Kconfig
> @@ -352,6 +352,18 @@ config TI_K3_R5_REMOTEPROC
> It's safe to say N here if you're not interested in utilizing
> a slave processor.
>
> +config XLNX_R5_REMOTEPROC
> + tristate "Xilinx R5 remoteproc support"
> + depends on PM && ARCH_ZYNQMP
> + depends on ZYNQMP_FIRMWARE
> + select RPMSG_VIRTIO
> + select ZYNQMP_IPI_MBOX
> + help
> + Say y or m here to support Xilinx R5 remote processors via the remote
> + processor framework.
> +
> + It's safe to say N if not interested in using RPU r5f cores.
> +
> endif # REMOTEPROC
>
> endmenu
> diff --git a/drivers/remoteproc/Makefile b/drivers/remoteproc/Makefile
> index 5478c7cb9e07..91314a9b43ce 100644
> --- a/drivers/remoteproc/Makefile
> +++ b/drivers/remoteproc/Makefile
> @@ -38,3 +38,4 @@ obj-$(CONFIG_ST_SLIM_REMOTEPROC) += st_slim_rproc.o
> obj-$(CONFIG_STM32_RPROC) += stm32_rproc.o
> obj-$(CONFIG_TI_K3_DSP_REMOTEPROC) += ti_k3_dsp_remoteproc.o
> obj-$(CONFIG_TI_K3_R5_REMOTEPROC) += ti_k3_r5_remoteproc.o
> +obj-$(CONFIG_XLNX_R5_REMOTEPROC) += xlnx_r5_remoteproc.o
> diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c
> new file mode 100644
> index 000000000000..ce2ac364b995
> --- /dev/null
> +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c
> @@ -0,0 +1,973 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * ZynqMP R5 Remote Processor driver
> + *
> + */
> +
> +#include <dt-bindings/power/xlnx-zynqmp-power.h>
> +#include <linux/firmware/xlnx-zynqmp.h>
> +#include <linux/interrupt.h>
> +#include <linux/kernel.h>
> +#include <linux/list.h>
> +#include <linux/mailbox_client.h>
> +#include <linux/mailbox/zynqmp-ipi-message.h>
No mailboxes are used in this revision and as such the above are not needed.
> +#include <linux/module.h>
> +#include <linux/of_address.h>
> +#include <linux/of_platform.h>
> +#include <linux/of_reserved_mem.h>
> +#include <linux/platform_device.h>
> +#include <linux/remoteproc.h>
> +#include <linux/skbuff.h>
> +#include <linux/sysfs.h>
As far as I can tell the above two includes are also not needed. Moreover
dma-mapping.h should be included because of dma_set_coherent_mask() and
DMA_BIT_MASK().
> +
> +#include "remoteproc_internal.h"
> +
> +/* settings for RPU cluster mode */
> +enum zynqmp_r5_cluster_mode {
> + SPLIT_MODE = 0, /* When cores run as separate processor */
> + LOCKSTEP_MODE = 1, /* cores execute same code in lockstep,clk-for-clk */
> + SINGLE_CPU_MODE = 2, /* core0 is held in reset and only core1 runs */
> +};
> +
> +/**
> + * struct mem_bank_data - Memory Bank description
> + *
> + * @addr: Start address of memory bank
> + * @size: Size of Memory bank
> + * @pm_domain_id: Power-domains id of memory bank for firmware to turn on/off
> + * @bank_name: name of the bank for remoteproc framework
> + */
> +struct mem_bank_data {
> + phys_addr_t addr;
> + size_t size;
> + u32 pm_domain_id;
> + char *bank_name;
> +};
> +
> +static const struct mem_bank_data zynqmp_tcm_banks[] = {
> + {0xffe00000UL, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */
> + {0xffe20000UL, 0x10000UL, PD_R5_0_BTCM, "btcm0"},
> + {0xffe90000UL, 0x10000UL, PD_R5_1_ATCM, "atcm1"},
> + {0xffeb0000UL, 0x10000UL, PD_R5_1_BTCM, "btcm1"},
> +};
> +
> +/**
> + * struct zynqmp_r5_core - ZynqMP R5 core structure
> + *
> + * @dev: device of RPU instance
> + * @np: device node of RPU instance
> + * @tcm_bank_count: number TCM banks accessible to this RPU
> + * @tcm_banks: array of each TCM bank data
> + * @rproc: rproc handle
> + * @pm_domain_id: RPU CPU power domain id
> + */
> +struct zynqmp_r5_core {
> + struct device *dev;
> + struct device_node *np;
> + int tcm_bank_count;
> + struct mem_bank_data *tcm_banks;
> + struct rproc *rproc;
> + u32 pm_domain_id;
> +};
> +
> +/**
> + * struct zynqmp_r5_cluster - ZynqMP R5 cluster structure
> + *
> + * @dev: r5f subsystem cluster device node
> + * @mode: cluster mode of type zynqmp_r5_cluster_mode
> + * @core_count: number of r5 cores used for this cluster mode
> + * @r5_cores: Array of pointers pointing to r5 core
> + */
> +struct zynqmp_r5_cluster {
> + struct device *dev;
> + enum zynqmp_r5_cluster_mode mode;
> + int core_count;
> + struct zynqmp_r5_core **r5_cores;
> +};
> +
> +/*
> + * zynqmp_r5_set_mode - set RPU operation mode
> + *
> + * set RPU operation mode
> + *
> + * Return: 0 for success, negative value for failure
> + */
> +static int zynqmp_r5_set_mode(struct zynqmp_r5_core *r5_core,
> + enum zynqmp_r5_cluster_mode rpu_mode)
> +{
> + enum rpu_tcm_comb tcm_mode;
> + int ret, reg_val;
> +
> + /*
> + * SINGLE CPU mode is not supported in driver yet.
> + * For now we only set lockstep mode or split mode.
> + */
> + if (rpu_mode == LOCKSTEP_MODE) {
> + reg_val = 0;
> + } else if (rpu_mode == SPLIT_MODE) {
> + reg_val = 1;
> + } else if (rpu_mode == SINGLE_CPU_MODE) {
> + pr_err("RPU single cpu mode is not supported\n");
> + return -EINVAL;
> + }
> +
> + ret = zynqmp_pm_set_rpu_mode(r5_core->pm_domain_id, reg_val);
> + if (ret < 0) {
> + pr_err("failed to set RPU mode\n");
> + return ret;
> + }
> +
> + /*
> + * At this point RPU is assumed to be configured only in
> + * lockstep mode or split mode. Configure TCM accordingly.
> + */
> + tcm_mode = (rpu_mode == LOCKSTEP_MODE) ?
> + PM_RPU_TCM_COMB : PM_RPU_TCM_SPLIT;
> + ret = zynqmp_pm_set_tcm_config(r5_core->pm_domain_id, tcm_mode);
> + if (ret < 0)
> + pr_err("failed to configure TCM\n");
> +
> + return ret;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_start
> + * @rproc: single R5 core's corresponding rproc instance
> + *
> + * Start R5 Core from designated boot address.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_start(struct rproc *rproc)
> +{
> + struct zynqmp_r5_core *r5_core = rproc->priv;
> + enum rpu_boot_mem bootmem;
> + int ret;
> +
> + /*
> + * The exception vector pointers (EVP) refer to the base-address of
> + * exception vectors (for reset, IRQ, FIQ, etc). The reset-vector
> + * starts at the base-address and subsequent vectors are on 4-byte
> + * boundaries.
> + *
> + * Exception vectors can start either from 0x0000_0000 (LOVEC) or
> + * from 0xFFFF_0000 (HIVEC) which is mapped in the OCM (On-Chip Memory)
> + *
> + * Usually firmware will put Exception vectors at LOVEC.
> + *
> + * It is not recommend that you change the exception vector.
> + * Changing the EVP to HIVEC will result in increased interrupt latency
> + * and jitter. Also, if the OCM is secured and the Cortex-R5F processor
> + * is non-secured, then the Cortex-R5F processor cannot access the
> + * HIVEC exception vectors in the OCM.
> + */
> + bootmem = (rproc->bootaddr >= 0xFFFC0000) ?
> + PM_RPU_BOOTMEM_HIVEC : PM_RPU_BOOTMEM_LOVEC;
> +
> + dev_dbg(r5_core->dev, "RPU boot addr 0x%llx from %s.", rproc->bootaddr,
> + bootmem == PM_RPU_BOOTMEM_HIVEC ? "OCM" : "TCM");
> +
> + ret = zynqmp_pm_request_wake(r5_core->pm_domain_id, 1,
> + bootmem, ZYNQMP_PM_REQUEST_ACK_NO);
> + if (ret)
> + pr_err("failed to start RPU = 0x%x\n", r5_core->pm_domain_id);
> + return ret;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_stop
> + * @rproc: single R5 core's corresponding rproc instance
> + *
> + * Power down R5 Core.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_stop(struct rproc *rproc)
> +{
> + struct zynqmp_r5_core *r5_core = rproc->priv;
> + int ret;
> +
> + ret = zynqmp_pm_force_pwrdwn(r5_core->pm_domain_id,
> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
> + if (ret)
> + pr_err("failed to stop remoteproc RPU %d\n", ret);
> +
> + return ret;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_mem_map
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to map
> + *
> + * Callback to map va for memory-region's carveout.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_mem_map(struct rproc *rproc,
> + struct rproc_mem_entry *mem)
> +{
> + void __iomem *va;
> +
> + va = ioremap_wc(mem->dma, mem->len);
> + if (IS_ERR_OR_NULL(va))
> + return -ENOMEM;
> +
> + mem->va = (void *)va;
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_mem_unmap
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to unmap
> + *
> + * Unmap memory-region carveout
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_mem_unmap(struct rproc *rproc,
> + struct rproc_mem_entry *mem)
> +{
> + iounmap((void __iomem *)mem->va);
> + return 0;
> +}
> +
> +/*
> + * add_mem_regions
> + * @r5_core: single R5 core's corresponding zynqmp_r5_core type instance
> + * @rmem: reserved mem region parsed from dt node
> + *
> + * Construct rproc mem carveouts from carveout provided in
> + * memory-region property
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int add_mem_regions_carveout(struct zynqmp_r5_core *r5_core,
> + struct reserved_mem *rmem)
> +{
> + struct device *dev;
> + struct rproc_mem_entry *mem;
> + struct rproc *rproc;
> +
> + rproc = r5_core->rproc;
> + dev = r5_core->dev;
> +
> + /* Register associated reserved memory regions */
> + mem = rproc_mem_entry_init(dev, NULL,
> + (dma_addr_t)rmem->base,
> + rmem->size, rmem->base,
> + zynqmp_r5_rproc_mem_map,
> + zynqmp_r5_rproc_mem_unmap,
> + rmem->name);
> + if (!mem)
> + return -ENOMEM;
> +
> + rproc_add_carveout(rproc, mem);
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_mem_unmap
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to unmap
> + *
> + * Unmap TCM banks when powering down R5 core.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int tcm_mem_unmap(struct rproc *rproc, struct rproc_mem_entry *mem)
> +{
> + struct zynqmp_r5_core *r5_core;
> + int i;
> + u32 pm_domain_id;
> +
> + r5_core = rproc->priv;
> + if (!r5_core) {
> + pr_err("r5 core is not available\n");
> + return -EINVAL;
> + }
> +
> + iounmap((void __iomem *)mem->va);
> +
> + for (i = 0; i < r5_core->tcm_bank_count; i++) {
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> + if (zynqmp_pm_release_node(pm_domain_id))
> + pr_warn("can't turn off TCM bank 0x%x", pm_domain_id);
> + }
> +
> + return 0;
> +}
> +
> +/*
> + * tcm_mem_map
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to initialize the va and da fields of
> + *
> + * Given TCM bank entry, this callback will set device address for R5
> + * running on TCM and also setup virtual address for TCM bank
> + * remoteproc carveout.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int tcm_mem_map(struct rproc *rproc,
> + struct rproc_mem_entry *mem)
> +{
> + void __iomem *va;
> +
> + va = ioremap_wc(mem->dma, mem->len);
> + if (IS_ERR_OR_NULL(va))
> + return -ENOMEM;
> +
> + /* Update memory entry va */
> + mem->va = (void *)va;
> +
> + /* clear TCMs */
> + memset_io(va, 0, mem->len);
> +
> + /*
> + * The R5s expect their TCM banks to be at address 0x0 and 0x2000,
> + * while on the Linux side they are at 0xffexxxxx.
> + *
> + * Zero out the high 12 bits of the address. This will give
> + * expected values for TCM Banks 0A and 0B (0x0 and 0x20000).
> + */
> + mem->da &= 0x000fffff;
> +
> + /*
> + * TCM Banks 1A and 1B still have to be translated.
> + *
> + * Below handle these two banks' absolute addresses (0xffe90000 and
> + * 0xffeb0000) and convert to the expected relative addresses
> + * (0x0 and 0x20000).
> + */
> + if (mem->da == 0x90000 || mem->da == 0xB0000)
> + mem->da -= 0x90000;
> +
> + /* if translated TCM bank address is not valid report error */
> + if (mem->da != 0x0 && mem->da != 0x20000) {
> + dev_err(&rproc->dev, "invalid TCM address: %x\n", mem->da);
> + return -EINVAL;
> + }
> + return 0;
> +}
> +
> +static int add_tcm_carveout_split_mode(struct rproc *rproc)
> +{
> + int i, num_banks, ret;
> + struct rproc_mem_entry *mem;
> + u32 pm_domain_id;
> + phys_addr_t bank_addr;
> + size_t bank_size;
> + char *bank_name;
> + struct device *dev;
> + struct zynqmp_r5_core *r5_core;
> +
> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
> +
> + dev = r5_core->dev;
> +
> + /* go through zynqmp banks for r5 node */
> + num_banks = r5_core->tcm_bank_count;
> + if (num_banks <= 0) {
> + dev_err(dev, "need to specify TCM banks\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * Power-on Each 64KB TCM,
> + * register its address space, map and unmap functions
> + * and add carveouts accordingly
> + */
> + for (i = 0; i < num_banks; i++) {
> + bank_addr = r5_core->tcm_banks[i].addr;
> + bank_name = r5_core->tcm_banks[i].bank_name;
> + bank_size = r5_core->tcm_banks[i].size;
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> +
> + ret = zynqmp_pm_request_node(pm_domain_id,
> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
> + if (ret < 0) {
> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
> + return ret;
> + }
> +
> + dev_dbg(dev, "TCM carveout split mode %s addr=%llx, size=0x%lx",
> + bank_name, bank_addr, bank_size);
> +
> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
> + bank_size, bank_addr,
> + tcm_mem_map, tcm_mem_unmap,
> + bank_name);
> + if (!mem) {
> + /* If failed, Turn off all TCM banks turned on before */
> + do {
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> + ret = zynqmp_pm_release_node(pm_domain_id);
> + if (ret)
> + dev_warn(dev,
> + "fail to release node: 0x%x, %x\n",
> + pm_domain_id, ret);
> + } while (i--);
> + return -ENOMEM;
> + }
> +
> + rproc_add_carveout(rproc, mem);
> + }
> +
> + return 0;
> +}
> +
> +static int add_tcm_carveout_lockstep_mode(struct rproc *rproc)
> +{
> + int i, num_banks, ret;
> + struct rproc_mem_entry *mem;
> + u32 pm_domain_id;
> + phys_addr_t bank_addr;
> + size_t bank_size = 0;
> + char *bank_name;
> + struct device *dev;
> + struct zynqmp_r5_core *r5_core;
> +
> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
> + dev = r5_core->dev;
> +
> + /* Go through zynqmp banks for r5 node */
> + num_banks = r5_core->tcm_bank_count;
> + if (num_banks <= 0) {
> + dev_err(dev, "need to specify TCM banks\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * In lockstep mode, TCM is contiguous memory block
> + * However, each TCM block still needs to be enabled individually.
> + * So, Enable each TCM block individually, but add their size
> + * to create contiguous memory region.
> + */
> + bank_addr = r5_core->tcm_banks[0].addr;
> + bank_name = r5_core->tcm_banks[0].bank_name;
> +
> + for (i = 0; i < num_banks; i++) {
> + bank_size += r5_core->tcm_banks[i].size;
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> +
> + /* Turn on each TCM bank individually */
> + ret = zynqmp_pm_request_node(pm_domain_id,
> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
> + if (ret < 0) {
> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
> + return ret;
> + }
> + }
> +
> + dev_dbg(dev, "TCM add carveout lockstep mode %s addr=0x%llx, size=0x%lx",
> + bank_name, bank_addr, bank_size);
> +
> + /* Register TCM address range, TCM map and unmap functions */
> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
> + bank_size, bank_addr,
> + tcm_mem_map, tcm_mem_unmap,
> + bank_name);
> + if (!mem) {
> + /* In case of error, turn off TCM banks */
> + for (i = 0; i < num_banks; i++) {
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> + ret = zynqmp_pm_release_node(pm_domain_id);
> + if (ret)
> + dev_warn(dev,
> + "fail to release node: 0x%x ret: %x\n",
> + pm_domain_id, ret);
> + }
> + return -ENOMEM;
> + }
> +
> + /* If registration is success, add carveouts */
> + rproc_add_carveout(rproc, mem);
> +
> + return 0;
> +}
> +
> +/*
> + * add_tcm_banks()
> + * @rproc: single R5 core's corresponding rproc instance
> + *
> + * Given R5 node in remoteproc instance
> + * allocate remoteproc carveout for TCM memory
> + * needed for firmware to be loaded
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int add_tcm_banks(struct rproc *rproc)
> +{
> + struct device *dev;
> + struct platform_device *parent_pdev;
> + struct zynqmp_r5_cluster *cluster;
> + struct zynqmp_r5_core *r5_core;
> +
> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
> + if (!r5_core)
> + return -EINVAL;
> +
> + dev = r5_core->dev;
> +
> + cluster = dev_get_drvdata(dev->parent);
> + if (!cluster) {
> + dev_err(&parent_pdev->dev, "Invalid driver data\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * In lockstep mode TCM banks are one contiguous memory region of 256Kb
> + * In split mode, each TCM bank is 64Kb and not contiguous.
> + * We add memory carveouts accordingly.
> + */
> + if (cluster->mode == SPLIT_MODE)
> + return add_tcm_carveout_split_mode(rproc);
> + else if (cluster->mode == LOCKSTEP_MODE)
> + return add_tcm_carveout_lockstep_mode(rproc);
> +
> + dev_err(cluster->dev, "invalid cluster mode\n");
> + return -EINVAL;
> +}
> +
> +/*
> + * zynqmp_r5_parse_fw()
> + * @rproc: single R5 core's corresponding rproc instance
> + * @fw: ptr to firmware to be loaded onto r5 core
> + *
> + * When loading firmware, ensure the necessary carveouts are in remoteproc
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct firmware *fw)
> +{
> + int ret;
> + struct zynqmp_r5_core *r5_core;
> + struct device *dev;
> +
> + r5_core = rproc->priv;
> +
> + dev = r5_core->dev;
> +
> + ret = add_tcm_banks(rproc);
> + if (ret) {
> + dev_err(dev, "failed to get TCM banks, err %d\n", ret);
> + return ret;
> + }
> +
> + ret = rproc_elf_load_rsc_table(rproc, fw);
> + if (ret == -EINVAL) {
> + /*
> + * resource table only required for IPC.
> + * if not present, this is not necessarily an error;
> + * for example, loading r5 hello world application
> + * so simply inform user and keep going.
> + */
> + dev_info(&rproc->dev, "no resource table found.\n");
> + ret = 0;
> + }
> + return ret;
> +}
> +
> +static struct rproc_ops zynqmp_r5_rproc_ops = {
> + .start = zynqmp_r5_rproc_start,
> + .stop = zynqmp_r5_rproc_stop,
> + .load = rproc_elf_load_segments,
> + .parse_fw = zynqmp_r5_parse_fw,
> + .find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
> + .sanity_check = rproc_elf_sanity_check,
> + .get_boot_addr = rproc_elf_get_boot_addr,
> +};
> +
> +/**
> + * zynqmp_r5_add_rproc_core()
> + * Probes ZynqMP R5 processor device node
> + * this is called for each individual R5 core
> + *
> + * @cdev: Device node of each r5 core
> + *
> + * Return: zynqmp_r5_core object for success, error pointer in case of error.
> + */
> +static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev)
> +{
> + int ret;
> + struct rproc *r5_rproc;
> + struct zynqmp_r5_core *r5_core;
> +
> + /* Set up DMA mask */
> + ret = dma_set_coherent_mask(cdev, DMA_BIT_MASK(32));
> + if (ret)
> + return ERR_PTR(ret);
> +
> + /* Allocate remoteproc instance */
> + r5_rproc = devm_rproc_alloc(cdev, dev_name(cdev),
> + &zynqmp_r5_rproc_ops,
> + NULL, sizeof(struct zynqmp_r5_core));
> + if (!r5_rproc) {
> + dev_err(cdev, "failed to allocate memory for rproc instance\n");
> + return ERR_PTR(-ENOMEM);
> + }
> +
> + r5_rproc->auto_boot = false;
> + r5_core = (struct zynqmp_r5_core *)r5_rproc->priv;
> + r5_core->dev = cdev;
> +
> + /* Add R5 remoteproc core */
> + ret = devm_rproc_add(cdev, r5_rproc);
> + if (ret) {
> + dev_err(cdev, "failed to add r5 remoteproc\n");
> + return ERR_PTR(ret);
> + }
> +
> + r5_core->rproc = r5_rproc;
> + return r5_core;
> +}
> +
> +/**
> + * zynqmp_r5_get_tcm_node()
> + * Ideally this function should parse tcm node and store information
> + * in r5_core instance. We will use hardcoded TCM information from
> + * driver for now in this function.
> + *
> + * @cluster: pointer to zynqmp_r5_cluster type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster)
> +{
> + int tcm_bank_count, tcm_node;
> + int i = 0, j;
> + struct zynqmp_r5_core *r5_core;
> + const struct mem_bank_data *tcm = zynqmp_tcm_banks;
> + struct device *dev = cluster->dev;
> +
> + /*
> + * ToDo: Use predefined TCM address space values from driver until
> + * system-dt spec is not final for TCM
> + */
> + tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks);
> +
> + /* count per core tcm banks */
> + tcm_bank_count = tcm_bank_count / cluster->core_count;
> +
> + /* r5 core 0 will use all of TCM banks in lockstep mode.
> + * In split mode, r5 core0 will use 128k and r5 core1 will use another
> + * 128k. Assign TCM banks to each core accordingly
> + */
> + tcm_node = 0;
> + for (i = 0; i < cluster->core_count; i++) {
> + r5_core = cluster->r5_cores[i];
> + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count,
> + sizeof(struct mem_bank_data),
> + GFP_KERNEL);
> + if (!r5_core->tcm_banks)
> + return -ENOMEM;
> +
> + for (j = 0; j < tcm_bank_count; j++) {
> + /* Use pre-defined TCM reg values.
> + * Eventually this should be replaced by values
> + * parsed from dts.
> + */
> + r5_core->tcm_banks[j].addr = tcm[tcm_node].addr;
> + r5_core->tcm_banks[j].size = tcm[tcm_node].size;
> + r5_core->tcm_banks[j].pm_domain_id = tcm[tcm_node].pm_domain_id;
> + r5_core->tcm_banks[j].bank_name = tcm[tcm_node].bank_name;
> + tcm_node++;
> + }
> +
> + r5_core->tcm_bank_count = tcm_bank_count;
> + }
> +
> + return 0;
> +}
> +
> +/**
> + * zynqmp_r5_get_mem_region_node()
> + * parse memory-region property from dt node and add
> + * memory region carveouts
> + *
> + * @r5_core: pointer to zynqmp_r5_core type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core *r5_core)
> +{
> + int res_mem_count, i, ret;
> + struct device *dev;
> + struct device_node *np, *rmem_np;
> + struct reserved_mem *rmem;
> +
> + dev = r5_core->dev;
> +
> + np = r5_core->np;
> +
> + res_mem_count = of_property_count_elems_of_size(np, "memory-region",
> + sizeof(phandle));
> + if (res_mem_count <= 0) {
> + dev_warn(dev, "failed to get memory-region property %d\n",
> + res_mem_count);
> + return -EINVAL;
> + }
> +
> + for (i = 0; i < res_mem_count; i++) {
> + rmem_np = of_parse_phandle(np, "memory-region", i);
> + if (!rmem_np)
> + return -EINVAL;
> +
> + rmem = of_reserved_mem_lookup(rmem_np);
> + if (!rmem) {
> + of_node_put(rmem_np);
> + return -EINVAL;
> + }
> +
> + ret = add_mem_regions_carveout(r5_core, rmem);
> + if (ret)
> + dev_warn(dev, "failed to get reserve mem regions %d\n",
> + ret);
> +
> + of_node_put(rmem_np);
> + }
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_core_init()
> + * Create and initialize zynqmp_r5_core type object
> + *
> + * @cluster: pointer to zynqmp_r5_cluster type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
> +{
> + int ret, i;
> + struct zynqmp_r5_core *r5_core;
> + struct device *dev = cluster->dev;
> +
> + ret = zynqmp_r5_get_tcm_node(cluster);
> + if (ret < 0) {
> + dev_err(dev, "can't get tcm node, err %d\n", ret);
> + return ret;
> + }
> +
> + for (i = 0; i < cluster->core_count; i++) {
> + r5_core = cluster->r5_cores[i];
> +
> + ret = zynqmp_r5_get_mem_region_node(r5_core);
> + if (ret)
> + dev_warn(dev, "memory-region prop failed %d\n", ret);
> +
> + /* Initialize r5 cores with power-domains parsed from dts */
> + ret = of_property_read_u32_index(r5_core->np, "power-domains",
> + 1, &r5_core->pm_domain_id);
> + if (ret) {
> + dev_err(dev, "failed to get power-domains property\n");
> + return ret;
> + }
> +
> + ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
> + if (ret) {
> + dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
> + cluster->mode, ret);
> + return ret;
> + }
> + }
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_cluster_init()
> + * Create and initialize zynqmp_r5_cluster type object
> + *
> + * @cluster: pointer to zynqmp_r5_cluster type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
> +{
> + struct device *dev = cluster->dev;
> + struct device_node *dev_node = dev_of_node(dev);
> + struct device_node *child;
> + struct platform_device *child_pdev;
> + int core_count = 0, ret, i;
> + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
> + struct zynqmp_r5_core **r5_cores;
> +
> + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
> +
> + /*
> + * on success returns 0, if not defined then returns -EINVAL,
> + * In that case, default is LOCKSTEP mode
> + */
> + if (ret != -EINVAL && ret != 0) {
> + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * For now driver only supports split mode and lockstep mode.
> + * fail driver probe if either of that is not set in dts
> + */
> + if (cluster_mode == SINGLE_CPU_MODE) {
> + dev_err(dev, "driver does not support single cpu mode\n");
> + return -EINVAL;
> + } else if ((cluster_mode != SPLIT_MODE &&
> + cluster_mode != LOCKSTEP_MODE)) {
> + dev_err(dev, "Invalid cluster mode\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * Number of cores is decided by number of child nodes of
> + * r5f subsystem node in dts. If Split mode is used in dts
> + * 2 child nodes are expected.
> + * In lockstep mode if two child nodes are available,
> + * only use first child node and consider it as core0
> + * and ignore core1 dt node.
> + */
> + core_count = of_get_available_child_count(dev_node);
> + if (core_count <= 0) {
> + dev_err(dev, "Invalid number of r5 cores %d", core_count);
> + return -EINVAL;
> + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
> + dev_err(dev, "Invalid number of r5 cores for split mode\n");
> + return -EINVAL;
> + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
> + dev_warn(dev, "Only r5 core0 will be used\n");
> + core_count = 1;
> + }
> +
> + r5_cores = devm_kcalloc(dev, core_count,
> + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
> + if (!r5_cores)
> + return -ENOMEM;
> +
> + i = 0;
> + for_each_available_child_of_node(dev_node, child) {
> + child_pdev = of_find_device_by_node(child);
> + if (!child_pdev) {
> + of_node_put(child);
> + return -ENODEV;
> + }
> +
> + /* create and add remoteproc instance of type struct rproc */
> + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
> + r5_cores[i]->dev = &child_pdev->dev;
> + if (!r5_cores[i]->dev) {
> + dev_err(dev, "can't get device for r5 core %d\n", i);
> + of_node_put(child);
> + return -ENODEV;
> + }
> +
> + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
> + if (!r5_cores[i]->np) {
> + dev_err(dev, "can't get device node for r5 core %d\n", i);
> + of_node_put(child);
> + return -ENODEV;
> + }
> +
> + i++;
> +
> + /*
> + * If two child nodes are available in dts in lockstep mode,
> + * then ignore second child node.
> + */
> + if (i == core_count) {
> + of_node_put(child);
> + break;
> + }
> + of_node_put(child);
> + }
> +
> + cluster->mode = cluster_mode;
> + cluster->core_count = core_count;
> + cluster->r5_cores = r5_cores;
> +
> + ret = zynqmp_r5_core_init(cluster);
> + if (ret < 0) {
> + dev_err(dev, "failed to init r5 core err %d\n", ret);
> + return ret;
> + }
> +
> + return 0;
> +}
> +
> +static void zynqmp_r5_cluster_exit(void *data)
> +{
> + struct platform_device *pdev = (struct platform_device *)data;
> +
> + platform_set_drvdata(pdev, NULL);
> +
> + pr_info("Exit r5f subsystem driver\n");
> +}
> +
> +/*
> + * zynqmp_r5_remoteproc_probe()
> + *
> + * @pdev: domain platform device for R5 cluster
> + *
> + * called when driver is probed, for each R5 core specified in DT,
> + * setup as needed to do remoteproc-related operations
> + *
> + * Return: 0 for success, negative value for failure.
> + */
> +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
> +{
> + int ret;
> + struct zynqmp_r5_cluster *cluster;
> + struct device *dev = &pdev->dev;
> +
> + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
> + if (!cluster)
> + return -ENOMEM;
> +
> + cluster->dev = dev;
> +
> + ret = devm_of_platform_populate(dev);
> + if (ret) {
> + dev_err(dev, "failed to populate platform dev %d\n", ret);
> + return ret;
Please use dev_err_probe()
> + }
> +
> + /* wire in so each core can be cleaned up at driver remove */
> + platform_set_drvdata(pdev, cluster);
> +
> + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
> + if (ret)
> + return ret;
Not sure this is needed. The TI guys used that mechanic to release acquired
resources but merely setting driver data to NULL doesn't mandate all this.
> +
> + ret = zynqmp_r5_cluster_init(cluster);
> + if (ret) {
> + dev_err(dev, "Invalid r5f subsystem device tree\n");
> + return ret;
Here to.
More comments to come tomorrow.
Thanks,
Mathieu
> + }
> +
> + return 0;
> +}
> +
> +/* Match table for OF platform binding */
> +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
> + { .compatible = "xlnx,zynqmp-r5fss", },
> + { /* end of list */ },
> +};
> +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
> +
> +static struct platform_driver zynqmp_r5_remoteproc_driver = {
> + .probe = zynqmp_r5_remoteproc_probe,
> + .driver = {
> + .name = "zynqmp_r5_remoteproc",
> + .of_match_table = zynqmp_r5_remoteproc_match,
> + },
> +};
> +module_platform_driver(zynqmp_r5_remoteproc_driver);
> +
> +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
> +MODULE_AUTHOR("Xilinx Inc.");
> +MODULE_LICENSE("GPL v2");
> --
> 2.25.1
>
On Thu, Feb 10, 2022 at 03:28:24AM -0800, Tanmay Shah wrote:
> This driver enables r5f dual core Real time Processing Unit subsystem
> available on Xilinx Zynq Ultrascale MPSoC Platform. RPU subsystem
> (cluster) can be configured in different modes e.g. split mode in which
> two r5f cores work independent of each other and lock-step mode in which
> both r5f cores execute same code clock-for-clock and notify if the
> result is different.
>
> The Xilinx r5 Remoteproc Driver boots the RPU cores via calls to the Xilinx
> Platform Management Unit that handles the R5 configuration, memory access
> and R5 lifecycle management. The interface to this manager is done in this
> driver via zynqmp_pm_* function calls.
>
> Signed-off-by: Ben Levinsky <[email protected]>
> Signed-off-by: Tanmay Shah <[email protected]>
> ---
>
> Changes in v3:
> - Fix checkpatch script indentation warning
> - Remove unused variable from xilinx remoteproc driver
> - use C style comments, i.e /*...*/
> - Remove redundant debug information which can be derived using /proc/device-tree
> - Fix multilined comment format
> - s/"final fot TCM"/"final for TCM"
> - Function devm_kzalloc() does not return an code on error, just NULL.
> Remove redundant error check for this function throughout the driver.
> - Fix RPU mode configuration and add documentation accordingly
> - Get rid of the indentations to match function documentation style with rest of the driver
> - Fix memory leak by only using r5_rproc->priv and not replace it with new instance
> - Use 'i' for the outer loop and 'j' for the inner one as per convention
> - Remove redundant error and NULL checks throughout the driver
> - Use devm_kcalloc() when more than one element is required
> - Add memory-regions carveouts during driver probe instead of parse_fw call
> This removes redundant copy of reserved_mem object in r5_core structure.
> - Fix memory leak by using of_node_put()
> - Fix indentation of tcm_mem_map function args
> - Remove redundant init of variables
> - Initialize tcm bank size variable for lockstep mode
> - Replace u32 with phys_addr_t for variable stroing memory bank address
> - Add documentation of TCM behavior in lockstep mode
> - Use dev_get_drvdata instead of platform driver API
> - Remove info level messages
> - Fix checkpatch.pl warnings
> - Add documentation for the Xilinx r5f platform to understand driver design
>
> drivers/remoteproc/Kconfig | 12 +
> drivers/remoteproc/Makefile | 1 +
> drivers/remoteproc/xlnx_r5_remoteproc.c | 973 ++++++++++++++++++++++++
> 3 files changed, 986 insertions(+)
> create mode 100644 drivers/remoteproc/xlnx_r5_remoteproc.c
>
> diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
> index 166019786653..5637a71c0677 100644
> --- a/drivers/remoteproc/Kconfig
> +++ b/drivers/remoteproc/Kconfig
> @@ -352,6 +352,18 @@ config TI_K3_R5_REMOTEPROC
> It's safe to say N here if you're not interested in utilizing
> a slave processor.
>
> +config XLNX_R5_REMOTEPROC
> + tristate "Xilinx R5 remoteproc support"
> + depends on PM && ARCH_ZYNQMP
> + depends on ZYNQMP_FIRMWARE
> + select RPMSG_VIRTIO
> + select ZYNQMP_IPI_MBOX
> + help
> + Say y or m here to support Xilinx R5 remote processors via the remote
> + processor framework.
> +
> + It's safe to say N if not interested in using RPU r5f cores.
> +
> endif # REMOTEPROC
>
> endmenu
> diff --git a/drivers/remoteproc/Makefile b/drivers/remoteproc/Makefile
> index 5478c7cb9e07..91314a9b43ce 100644
> --- a/drivers/remoteproc/Makefile
> +++ b/drivers/remoteproc/Makefile
> @@ -38,3 +38,4 @@ obj-$(CONFIG_ST_SLIM_REMOTEPROC) += st_slim_rproc.o
> obj-$(CONFIG_STM32_RPROC) += stm32_rproc.o
> obj-$(CONFIG_TI_K3_DSP_REMOTEPROC) += ti_k3_dsp_remoteproc.o
> obj-$(CONFIG_TI_K3_R5_REMOTEPROC) += ti_k3_r5_remoteproc.o
> +obj-$(CONFIG_XLNX_R5_REMOTEPROC) += xlnx_r5_remoteproc.o
> diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c
> new file mode 100644
> index 000000000000..ce2ac364b995
> --- /dev/null
> +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c
> @@ -0,0 +1,973 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * ZynqMP R5 Remote Processor driver
> + *
> + */
> +
> +#include <dt-bindings/power/xlnx-zynqmp-power.h>
> +#include <linux/firmware/xlnx-zynqmp.h>
> +#include <linux/interrupt.h>
> +#include <linux/kernel.h>
> +#include <linux/list.h>
> +#include <linux/mailbox_client.h>
> +#include <linux/mailbox/zynqmp-ipi-message.h>
> +#include <linux/module.h>
> +#include <linux/of_address.h>
> +#include <linux/of_platform.h>
> +#include <linux/of_reserved_mem.h>
> +#include <linux/platform_device.h>
> +#include <linux/remoteproc.h>
> +#include <linux/skbuff.h>
> +#include <linux/sysfs.h>
> +
> +#include "remoteproc_internal.h"
> +
> +/* settings for RPU cluster mode */
> +enum zynqmp_r5_cluster_mode {
> + SPLIT_MODE = 0, /* When cores run as separate processor */
> + LOCKSTEP_MODE = 1, /* cores execute same code in lockstep,clk-for-clk */
> + SINGLE_CPU_MODE = 2, /* core0 is held in reset and only core1 runs */
> +};
> +
> +/**
> + * struct mem_bank_data - Memory Bank description
> + *
> + * @addr: Start address of memory bank
> + * @size: Size of Memory bank
> + * @pm_domain_id: Power-domains id of memory bank for firmware to turn on/off
> + * @bank_name: name of the bank for remoteproc framework
> + */
> +struct mem_bank_data {
> + phys_addr_t addr;
> + size_t size;
> + u32 pm_domain_id;
> + char *bank_name;
> +};
> +
> +static const struct mem_bank_data zynqmp_tcm_banks[] = {
> + {0xffe00000UL, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */
> + {0xffe20000UL, 0x10000UL, PD_R5_0_BTCM, "btcm0"},
> + {0xffe90000UL, 0x10000UL, PD_R5_1_ATCM, "atcm1"},
> + {0xffeb0000UL, 0x10000UL, PD_R5_1_BTCM, "btcm1"},
> +};
> +
> +/**
> + * struct zynqmp_r5_core - ZynqMP R5 core structure
> + *
> + * @dev: device of RPU instance
> + * @np: device node of RPU instance
> + * @tcm_bank_count: number TCM banks accessible to this RPU
> + * @tcm_banks: array of each TCM bank data
> + * @rproc: rproc handle
> + * @pm_domain_id: RPU CPU power domain id
> + */
> +struct zynqmp_r5_core {
> + struct device *dev;
> + struct device_node *np;
> + int tcm_bank_count;
> + struct mem_bank_data *tcm_banks;
> + struct rproc *rproc;
> + u32 pm_domain_id;
> +};
> +
> +/**
> + * struct zynqmp_r5_cluster - ZynqMP R5 cluster structure
> + *
> + * @dev: r5f subsystem cluster device node
> + * @mode: cluster mode of type zynqmp_r5_cluster_mode
> + * @core_count: number of r5 cores used for this cluster mode
> + * @r5_cores: Array of pointers pointing to r5 core
> + */
> +struct zynqmp_r5_cluster {
> + struct device *dev;
> + enum zynqmp_r5_cluster_mode mode;
> + int core_count;
> + struct zynqmp_r5_core **r5_cores;
> +};
> +
> +/*
> + * zynqmp_r5_set_mode - set RPU operation mode
> + *
> + * set RPU operation mode
> + *
> + * Return: 0 for success, negative value for failure
> + */
> +static int zynqmp_r5_set_mode(struct zynqmp_r5_core *r5_core,
> + enum zynqmp_r5_cluster_mode rpu_mode)
> +{
> + enum rpu_tcm_comb tcm_mode;
> + int ret, reg_val;
> +
> + /*
> + * SINGLE CPU mode is not supported in driver yet.
> + * For now we only set lockstep mode or split mode.
> + */
> + if (rpu_mode == LOCKSTEP_MODE) {
> + reg_val = 0;
> + } else if (rpu_mode == SPLIT_MODE) {
> + reg_val = 1;
> + } else if (rpu_mode == SINGLE_CPU_MODE) {
> + pr_err("RPU single cpu mode is not supported\n");
> + return -EINVAL;
> + }
> +
> + ret = zynqmp_pm_set_rpu_mode(r5_core->pm_domain_id, reg_val);
> + if (ret < 0) {
> + pr_err("failed to set RPU mode\n");
> + return ret;
> + }
> +
> + /*
> + * At this point RPU is assumed to be configured only in
> + * lockstep mode or split mode. Configure TCM accordingly.
> + */
> + tcm_mode = (rpu_mode == LOCKSTEP_MODE) ?
> + PM_RPU_TCM_COMB : PM_RPU_TCM_SPLIT;
> + ret = zynqmp_pm_set_tcm_config(r5_core->pm_domain_id, tcm_mode);
> + if (ret < 0)
> + pr_err("failed to configure TCM\n");
> +
> + return ret;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_start
> + * @rproc: single R5 core's corresponding rproc instance
> + *
> + * Start R5 Core from designated boot address.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_start(struct rproc *rproc)
> +{
> + struct zynqmp_r5_core *r5_core = rproc->priv;
> + enum rpu_boot_mem bootmem;
> + int ret;
> +
> + /*
> + * The exception vector pointers (EVP) refer to the base-address of
> + * exception vectors (for reset, IRQ, FIQ, etc). The reset-vector
> + * starts at the base-address and subsequent vectors are on 4-byte
> + * boundaries.
> + *
> + * Exception vectors can start either from 0x0000_0000 (LOVEC) or
> + * from 0xFFFF_0000 (HIVEC) which is mapped in the OCM (On-Chip Memory)
> + *
> + * Usually firmware will put Exception vectors at LOVEC.
> + *
> + * It is not recommend that you change the exception vector.
> + * Changing the EVP to HIVEC will result in increased interrupt latency
> + * and jitter. Also, if the OCM is secured and the Cortex-R5F processor
> + * is non-secured, then the Cortex-R5F processor cannot access the
> + * HIVEC exception vectors in the OCM.
> + */
> + bootmem = (rproc->bootaddr >= 0xFFFC0000) ?
> + PM_RPU_BOOTMEM_HIVEC : PM_RPU_BOOTMEM_LOVEC;
> +
> + dev_dbg(r5_core->dev, "RPU boot addr 0x%llx from %s.", rproc->bootaddr,
> + bootmem == PM_RPU_BOOTMEM_HIVEC ? "OCM" : "TCM");
> +
> + ret = zynqmp_pm_request_wake(r5_core->pm_domain_id, 1,
> + bootmem, ZYNQMP_PM_REQUEST_ACK_NO);
> + if (ret)
> + pr_err("failed to start RPU = 0x%x\n", r5_core->pm_domain_id);
> + return ret;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_stop
> + * @rproc: single R5 core's corresponding rproc instance
> + *
> + * Power down R5 Core.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_stop(struct rproc *rproc)
> +{
> + struct zynqmp_r5_core *r5_core = rproc->priv;
> + int ret;
> +
> + ret = zynqmp_pm_force_pwrdwn(r5_core->pm_domain_id,
> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
> + if (ret)
> + pr_err("failed to stop remoteproc RPU %d\n", ret);
> +
> + return ret;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_mem_map
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to map
> + *
> + * Callback to map va for memory-region's carveout.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_mem_map(struct rproc *rproc,
> + struct rproc_mem_entry *mem)
> +{
> + void __iomem *va;
> +
> + va = ioremap_wc(mem->dma, mem->len);
> + if (IS_ERR_OR_NULL(va))
> + return -ENOMEM;
> +
> + mem->va = (void *)va;
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_mem_unmap
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to unmap
> + *
> + * Unmap memory-region carveout
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_mem_unmap(struct rproc *rproc,
> + struct rproc_mem_entry *mem)
> +{
> + iounmap((void __iomem *)mem->va);
> + return 0;
> +}
> +
> +/*
> + * add_mem_regions
> + * @r5_core: single R5 core's corresponding zynqmp_r5_core type instance
> + * @rmem: reserved mem region parsed from dt node
> + *
> + * Construct rproc mem carveouts from carveout provided in
> + * memory-region property
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int add_mem_regions_carveout(struct zynqmp_r5_core *r5_core,
> + struct reserved_mem *rmem)
> +{
> + struct device *dev;
> + struct rproc_mem_entry *mem;
> + struct rproc *rproc;
> +
> + rproc = r5_core->rproc;
> + dev = r5_core->dev;
> +
> + /* Register associated reserved memory regions */
> + mem = rproc_mem_entry_init(dev, NULL,
> + (dma_addr_t)rmem->base,
> + rmem->size, rmem->base,
> + zynqmp_r5_rproc_mem_map,
> + zynqmp_r5_rproc_mem_unmap,
> + rmem->name);
> + if (!mem)
> + return -ENOMEM;
> +
> + rproc_add_carveout(rproc, mem);
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_mem_unmap
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to unmap
> + *
> + * Unmap TCM banks when powering down R5 core.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int tcm_mem_unmap(struct rproc *rproc, struct rproc_mem_entry *mem)
> +{
> + struct zynqmp_r5_core *r5_core;
> + int i;
> + u32 pm_domain_id;
> +
> + r5_core = rproc->priv;
> + if (!r5_core) {
> + pr_err("r5 core is not available\n");
> + return -EINVAL;
> + }
> +
> + iounmap((void __iomem *)mem->va);
> +
> + for (i = 0; i < r5_core->tcm_bank_count; i++) {
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> + if (zynqmp_pm_release_node(pm_domain_id))
> + pr_warn("can't turn off TCM bank 0x%x", pm_domain_id);
> + }
> +
> + return 0;
> +}
> +
> +/*
> + * tcm_mem_map
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to initialize the va and da fields of
> + *
> + * Given TCM bank entry, this callback will set device address for R5
> + * running on TCM and also setup virtual address for TCM bank
> + * remoteproc carveout.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int tcm_mem_map(struct rproc *rproc,
> + struct rproc_mem_entry *mem)
> +{
> + void __iomem *va;
> +
> + va = ioremap_wc(mem->dma, mem->len);
> + if (IS_ERR_OR_NULL(va))
> + return -ENOMEM;
> +
> + /* Update memory entry va */
> + mem->va = (void *)va;
> +
> + /* clear TCMs */
> + memset_io(va, 0, mem->len);
> +
> + /*
> + * The R5s expect their TCM banks to be at address 0x0 and 0x2000,
> + * while on the Linux side they are at 0xffexxxxx.
> + *
> + * Zero out the high 12 bits of the address. This will give
> + * expected values for TCM Banks 0A and 0B (0x0 and 0x20000).
> + */
> + mem->da &= 0x000fffff;
> +
> + /*
> + * TCM Banks 1A and 1B still have to be translated.
> + *
> + * Below handle these two banks' absolute addresses (0xffe90000 and
> + * 0xffeb0000) and convert to the expected relative addresses
> + * (0x0 and 0x20000).
> + */
> + if (mem->da == 0x90000 || mem->da == 0xB0000)
> + mem->da -= 0x90000;
> +
> + /* if translated TCM bank address is not valid report error */
> + if (mem->da != 0x0 && mem->da != 0x20000) {
> + dev_err(&rproc->dev, "invalid TCM address: %x\n", mem->da);
> + return -EINVAL;
> + }
> + return 0;
> +}
> +
> +static int add_tcm_carveout_split_mode(struct rproc *rproc)
> +{
> + int i, num_banks, ret;
> + struct rproc_mem_entry *mem;
> + u32 pm_domain_id;
> + phys_addr_t bank_addr;
> + size_t bank_size;
> + char *bank_name;
> + struct device *dev;
> + struct zynqmp_r5_core *r5_core;
> +
> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
> +
> + dev = r5_core->dev;
> +
> + /* go through zynqmp banks for r5 node */
> + num_banks = r5_core->tcm_bank_count;
> + if (num_banks <= 0) {
> + dev_err(dev, "need to specify TCM banks\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * Power-on Each 64KB TCM,
> + * register its address space, map and unmap functions
> + * and add carveouts accordingly
> + */
> + for (i = 0; i < num_banks; i++) {
> + bank_addr = r5_core->tcm_banks[i].addr;
> + bank_name = r5_core->tcm_banks[i].bank_name;
> + bank_size = r5_core->tcm_banks[i].size;
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> +
> + ret = zynqmp_pm_request_node(pm_domain_id,
> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
> + if (ret < 0) {
> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
> + return ret;
> + }
> +
> + dev_dbg(dev, "TCM carveout split mode %s addr=%llx, size=0x%lx",
> + bank_name, bank_addr, bank_size);
> +
> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
> + bank_size, bank_addr,
> + tcm_mem_map, tcm_mem_unmap,
> + bank_name);
> + if (!mem) {
> + /* If failed, Turn off all TCM banks turned on before */
> + do {
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> + ret = zynqmp_pm_release_node(pm_domain_id);
> + if (ret)
> + dev_warn(dev,
> + "fail to release node: 0x%x, %x\n",
> + pm_domain_id, ret);
> + } while (i--);
> + return -ENOMEM;
> + }
> +
> + rproc_add_carveout(rproc, mem);
> + }
> +
> + return 0;
> +}
> +
> +static int add_tcm_carveout_lockstep_mode(struct rproc *rproc)
> +{
> + int i, num_banks, ret;
> + struct rproc_mem_entry *mem;
> + u32 pm_domain_id;
> + phys_addr_t bank_addr;
> + size_t bank_size = 0;
> + char *bank_name;
> + struct device *dev;
> + struct zynqmp_r5_core *r5_core;
> +
> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
> + dev = r5_core->dev;
> +
> + /* Go through zynqmp banks for r5 node */
> + num_banks = r5_core->tcm_bank_count;
> + if (num_banks <= 0) {
> + dev_err(dev, "need to specify TCM banks\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * In lockstep mode, TCM is contiguous memory block
> + * However, each TCM block still needs to be enabled individually.
> + * So, Enable each TCM block individually, but add their size
> + * to create contiguous memory region.
> + */
> + bank_addr = r5_core->tcm_banks[0].addr;
> + bank_name = r5_core->tcm_banks[0].bank_name;
> +
> + for (i = 0; i < num_banks; i++) {
> + bank_size += r5_core->tcm_banks[i].size;
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> +
> + /* Turn on each TCM bank individually */
> + ret = zynqmp_pm_request_node(pm_domain_id,
> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
> + if (ret < 0) {
> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
> + return ret;
> + }
> + }
> +
> + dev_dbg(dev, "TCM add carveout lockstep mode %s addr=0x%llx, size=0x%lx",
> + bank_name, bank_addr, bank_size);
> +
> + /* Register TCM address range, TCM map and unmap functions */
> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
> + bank_size, bank_addr,
> + tcm_mem_map, tcm_mem_unmap,
> + bank_name);
> + if (!mem) {
> + /* In case of error, turn off TCM banks */
> + for (i = 0; i < num_banks; i++) {
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> + ret = zynqmp_pm_release_node(pm_domain_id);
> + if (ret)
> + dev_warn(dev,
> + "fail to release node: 0x%x ret: %x\n",
> + pm_domain_id, ret);
> + }
> + return -ENOMEM;
> + }
> +
> + /* If registration is success, add carveouts */
> + rproc_add_carveout(rproc, mem);
> +
> + return 0;
> +}
> +
> +/*
> + * add_tcm_banks()
> + * @rproc: single R5 core's corresponding rproc instance
> + *
> + * Given R5 node in remoteproc instance
> + * allocate remoteproc carveout for TCM memory
> + * needed for firmware to be loaded
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int add_tcm_banks(struct rproc *rproc)
> +{
> + struct device *dev;
> + struct platform_device *parent_pdev;
> + struct zynqmp_r5_cluster *cluster;
> + struct zynqmp_r5_core *r5_core;
> +
> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
> + if (!r5_core)
> + return -EINVAL;
> +
> + dev = r5_core->dev;
> +
> + cluster = dev_get_drvdata(dev->parent);
> + if (!cluster) {
> + dev_err(&parent_pdev->dev, "Invalid driver data\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * In lockstep mode TCM banks are one contiguous memory region of 256Kb
> + * In split mode, each TCM bank is 64Kb and not contiguous.
> + * We add memory carveouts accordingly.
> + */
> + if (cluster->mode == SPLIT_MODE)
> + return add_tcm_carveout_split_mode(rproc);
> + else if (cluster->mode == LOCKSTEP_MODE)
> + return add_tcm_carveout_lockstep_mode(rproc);
> +
> + dev_err(cluster->dev, "invalid cluster mode\n");
> + return -EINVAL;
> +}
> +
> +/*
> + * zynqmp_r5_parse_fw()
> + * @rproc: single R5 core's corresponding rproc instance
> + * @fw: ptr to firmware to be loaded onto r5 core
> + *
> + * When loading firmware, ensure the necessary carveouts are in remoteproc
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct firmware *fw)
> +{
> + int ret;
> + struct zynqmp_r5_core *r5_core;
> + struct device *dev;
> +
> + r5_core = rproc->priv;
> +
> + dev = r5_core->dev;
> +
> + ret = add_tcm_banks(rproc);
> + if (ret) {
> + dev_err(dev, "failed to get TCM banks, err %d\n", ret);
> + return ret;
> + }
> +
> + ret = rproc_elf_load_rsc_table(rproc, fw);
> + if (ret == -EINVAL) {
> + /*
> + * resource table only required for IPC.
> + * if not present, this is not necessarily an error;
> + * for example, loading r5 hello world application
> + * so simply inform user and keep going.
> + */
> + dev_info(&rproc->dev, "no resource table found.\n");
> + ret = 0;
> + }
> + return ret;
> +}
> +
> +static struct rproc_ops zynqmp_r5_rproc_ops = {
> + .start = zynqmp_r5_rproc_start,
> + .stop = zynqmp_r5_rproc_stop,
> + .load = rproc_elf_load_segments,
> + .parse_fw = zynqmp_r5_parse_fw,
> + .find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
> + .sanity_check = rproc_elf_sanity_check,
> + .get_boot_addr = rproc_elf_get_boot_addr,
> +};
> +
> +/**
> + * zynqmp_r5_add_rproc_core()
> + * Probes ZynqMP R5 processor device node
> + * this is called for each individual R5 core
> + *
> + * @cdev: Device node of each r5 core
> + *
> + * Return: zynqmp_r5_core object for success, error pointer in case of error.
> + */
> +static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev)
> +{
> + int ret;
> + struct rproc *r5_rproc;
> + struct zynqmp_r5_core *r5_core;
> +
> + /* Set up DMA mask */
> + ret = dma_set_coherent_mask(cdev, DMA_BIT_MASK(32));
> + if (ret)
> + return ERR_PTR(ret);
> +
> + /* Allocate remoteproc instance */
> + r5_rproc = devm_rproc_alloc(cdev, dev_name(cdev),
> + &zynqmp_r5_rproc_ops,
> + NULL, sizeof(struct zynqmp_r5_core));
> + if (!r5_rproc) {
> + dev_err(cdev, "failed to allocate memory for rproc instance\n");
> + return ERR_PTR(-ENOMEM);
> + }
> +
> + r5_rproc->auto_boot = false;
> + r5_core = (struct zynqmp_r5_core *)r5_rproc->priv;
> + r5_core->dev = cdev;
> +
> + /* Add R5 remoteproc core */
> + ret = devm_rproc_add(cdev, r5_rproc);
> + if (ret) {
> + dev_err(cdev, "failed to add r5 remoteproc\n");
> + return ERR_PTR(ret);
> + }
> +
> + r5_core->rproc = r5_rproc;
> + return r5_core;
> +}
> +
> +/**
> + * zynqmp_r5_get_tcm_node()
> + * Ideally this function should parse tcm node and store information
> + * in r5_core instance. We will use hardcoded TCM information from
> + * driver for now in this function.
> + *
> + * @cluster: pointer to zynqmp_r5_cluster type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster)
> +{
> + int tcm_bank_count, tcm_node;
> + int i = 0, j;
> + struct zynqmp_r5_core *r5_core;
> + const struct mem_bank_data *tcm = zynqmp_tcm_banks;
> + struct device *dev = cluster->dev;
> +
> + /*
> + * ToDo: Use predefined TCM address space values from driver until
> + * system-dt spec is not final for TCM
> + */
> + tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks);
> +
> + /* count per core tcm banks */
> + tcm_bank_count = tcm_bank_count / cluster->core_count;
> +
> + /* r5 core 0 will use all of TCM banks in lockstep mode.
> + * In split mode, r5 core0 will use 128k and r5 core1 will use another
> + * 128k. Assign TCM banks to each core accordingly
> + */
> + tcm_node = 0;
> + for (i = 0; i < cluster->core_count; i++) {
> + r5_core = cluster->r5_cores[i];
> + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count,
> + sizeof(struct mem_bank_data),
> + GFP_KERNEL);
> + if (!r5_core->tcm_banks)
> + return -ENOMEM;
> +
> + for (j = 0; j < tcm_bank_count; j++) {
> + /* Use pre-defined TCM reg values.
> + * Eventually this should be replaced by values
> + * parsed from dts.
> + */
> + r5_core->tcm_banks[j].addr = tcm[tcm_node].addr;
> + r5_core->tcm_banks[j].size = tcm[tcm_node].size;
> + r5_core->tcm_banks[j].pm_domain_id = tcm[tcm_node].pm_domain_id;
> + r5_core->tcm_banks[j].bank_name = tcm[tcm_node].bank_name;
> + tcm_node++;
> + }
> +
> + r5_core->tcm_bank_count = tcm_bank_count;
> + }
> +
> + return 0;
> +}
> +
> +/**
> + * zynqmp_r5_get_mem_region_node()
> + * parse memory-region property from dt node and add
> + * memory region carveouts
> + *
> + * @r5_core: pointer to zynqmp_r5_core type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core *r5_core)
> +{
> + int res_mem_count, i, ret;
> + struct device *dev;
> + struct device_node *np, *rmem_np;
> + struct reserved_mem *rmem;
> +
> + dev = r5_core->dev;
> +
> + np = r5_core->np;
> +
> + res_mem_count = of_property_count_elems_of_size(np, "memory-region",
> + sizeof(phandle));
> + if (res_mem_count <= 0) {
> + dev_warn(dev, "failed to get memory-region property %d\n",
> + res_mem_count);
> + return -EINVAL;
> + }
> +
> + for (i = 0; i < res_mem_count; i++) {
> + rmem_np = of_parse_phandle(np, "memory-region", i);
> + if (!rmem_np)
> + return -EINVAL;
> +
> + rmem = of_reserved_mem_lookup(rmem_np);
> + if (!rmem) {
> + of_node_put(rmem_np);
> + return -EINVAL;
> + }
> +
> + ret = add_mem_regions_carveout(r5_core, rmem);
> + if (ret)
> + dev_warn(dev, "failed to get reserve mem regions %d\n",
> + ret);
> +
> + of_node_put(rmem_np);
> + }
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_core_init()
> + * Create and initialize zynqmp_r5_core type object
> + *
> + * @cluster: pointer to zynqmp_r5_cluster type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
> +{
> + int ret, i;
> + struct zynqmp_r5_core *r5_core;
> + struct device *dev = cluster->dev;
> +
> + ret = zynqmp_r5_get_tcm_node(cluster);
> + if (ret < 0) {
> + dev_err(dev, "can't get tcm node, err %d\n", ret);
> + return ret;
> + }
> +
> + for (i = 0; i < cluster->core_count; i++) {
> + r5_core = cluster->r5_cores[i];
> +
> + ret = zynqmp_r5_get_mem_region_node(r5_core);
> + if (ret)
> + dev_warn(dev, "memory-region prop failed %d\n", ret);
> +
> + /* Initialize r5 cores with power-domains parsed from dts */
> + ret = of_property_read_u32_index(r5_core->np, "power-domains",
> + 1, &r5_core->pm_domain_id);
> + if (ret) {
> + dev_err(dev, "failed to get power-domains property\n");
> + return ret;
> + }
> +
> + ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
> + if (ret) {
> + dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
> + cluster->mode, ret);
> + return ret;
> + }
> + }
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_cluster_init()
> + * Create and initialize zynqmp_r5_cluster type object
> + *
> + * @cluster: pointer to zynqmp_r5_cluster type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
> +{
> + struct device *dev = cluster->dev;
> + struct device_node *dev_node = dev_of_node(dev);
> + struct device_node *child;
> + struct platform_device *child_pdev;
> + int core_count = 0, ret, i;
> + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
> + struct zynqmp_r5_core **r5_cores;
> +
> + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
> +
> + /*
> + * on success returns 0, if not defined then returns -EINVAL,
> + * In that case, default is LOCKSTEP mode
> + */
> + if (ret != -EINVAL && ret != 0) {
> + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * For now driver only supports split mode and lockstep mode.
> + * fail driver probe if either of that is not set in dts
> + */
> + if (cluster_mode == SINGLE_CPU_MODE) {
> + dev_err(dev, "driver does not support single cpu mode\n");
> + return -EINVAL;
> + } else if ((cluster_mode != SPLIT_MODE &&
> + cluster_mode != LOCKSTEP_MODE)) {
> + dev_err(dev, "Invalid cluster mode\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * Number of cores is decided by number of child nodes of
> + * r5f subsystem node in dts. If Split mode is used in dts
> + * 2 child nodes are expected.
> + * In lockstep mode if two child nodes are available,
> + * only use first child node and consider it as core0
> + * and ignore core1 dt node.
> + */
> + core_count = of_get_available_child_count(dev_node);
> + if (core_count <= 0) {
> + dev_err(dev, "Invalid number of r5 cores %d", core_count);
> + return -EINVAL;
> + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
> + dev_err(dev, "Invalid number of r5 cores for split mode\n");
> + return -EINVAL;
> + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
> + dev_warn(dev, "Only r5 core0 will be used\n");
> + core_count = 1;
> + }
> +
> + r5_cores = devm_kcalloc(dev, core_count,
> + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
> + if (!r5_cores)
> + return -ENOMEM;
> +
> + i = 0;
> + for_each_available_child_of_node(dev_node, child) {
> + child_pdev = of_find_device_by_node(child);
> + if (!child_pdev) {
> + of_node_put(child);
> + return -ENODEV;
> + }
> +
> + /* create and add remoteproc instance of type struct rproc */
> + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
Function zynqmp_r5_add_rproc_core() returns an error code on error that is never
checked.
> + r5_cores[i]->dev = &child_pdev->dev;
This will result in a kernel stack trace if zynqmp_r5_add_rproc_core() fails.
> + if (!r5_cores[i]->dev) {
> + dev_err(dev, "can't get device for r5 core %d\n", i);
> + of_node_put(child);
> + return -ENODEV;
> + }
And here the validity of child_pdev->dev is checked _after_ it has been passed
to zynqmp_r5_add_rproc_core(). This check should not be needed if proper
error handling is done above.
> +
> + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
> + if (!r5_cores[i]->np) {
> + dev_err(dev, "can't get device node for r5 core %d\n", i);
> + of_node_put(child);
As mention in the documentation for of_find_device_by_node(), the function takes
a reference to child_pdev->dev that needs to be released with put_device(). In
fact I don't see the reference dropped anywhere, even when the driver is
released.
Moreover, if we end up here for r5_cores[1], resources acquired for r5_core[0]
are not released.
> + return -ENODEV;
> + }
> +
> + i++;
> +
> + /*
> + * If two child nodes are available in dts in lockstep mode,
> + * then ignore second child node.
> + */
> + if (i == core_count) {
> + of_node_put(child);
> + break;
> + }
> + of_node_put(child);
This one is not needed as it is already done by the
for_each_available_child_of_node() loop.
More comments tomorrow.
Thanks,
Mathieu
> + }
> +
> + cluster->mode = cluster_mode;
> + cluster->core_count = core_count;
> + cluster->r5_cores = r5_cores;
> +
> + ret = zynqmp_r5_core_init(cluster);
> + if (ret < 0) {
> + dev_err(dev, "failed to init r5 core err %d\n", ret);
> + return ret;
> + }
> +
> + return 0;
> +}
> +
> +static void zynqmp_r5_cluster_exit(void *data)
> +{
> + struct platform_device *pdev = (struct platform_device *)data;
> +
> + platform_set_drvdata(pdev, NULL);
> +
> + pr_info("Exit r5f subsystem driver\n");
> +}
> +
> +/*
> + * zynqmp_r5_remoteproc_probe()
> + *
> + * @pdev: domain platform device for R5 cluster
> + *
> + * called when driver is probed, for each R5 core specified in DT,
> + * setup as needed to do remoteproc-related operations
> + *
> + * Return: 0 for success, negative value for failure.
> + */
> +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
> +{
> + int ret;
> + struct zynqmp_r5_cluster *cluster;
> + struct device *dev = &pdev->dev;
> +
> + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
> + if (!cluster)
> + return -ENOMEM;
> +
> + cluster->dev = dev;
> +
> + ret = devm_of_platform_populate(dev);
> + if (ret) {
> + dev_err(dev, "failed to populate platform dev %d\n", ret);
> + return ret;
> + }
> +
> + /* wire in so each core can be cleaned up at driver remove */
> + platform_set_drvdata(pdev, cluster);
> +
> + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
> + if (ret)
> + return ret;
> +
> + ret = zynqmp_r5_cluster_init(cluster);
> + if (ret) {
> + dev_err(dev, "Invalid r5f subsystem device tree\n");
> + return ret;
> + }
> +
> + return 0;
> +}
> +
> +/* Match table for OF platform binding */
> +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
> + { .compatible = "xlnx,zynqmp-r5fss", },
> + { /* end of list */ },
> +};
> +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
> +
> +static struct platform_driver zynqmp_r5_remoteproc_driver = {
> + .probe = zynqmp_r5_remoteproc_probe,
> + .driver = {
> + .name = "zynqmp_r5_remoteproc",
> + .of_match_table = zynqmp_r5_remoteproc_match,
> + },
> +};
> +module_platform_driver(zynqmp_r5_remoteproc_driver);
> +
> +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
> +MODULE_AUTHOR("Xilinx Inc.");
> +MODULE_LICENSE("GPL v2");
> --
> 2.25.1
>
On Thu, Feb 10, 2022 at 03:28:24AM -0800, Tanmay Shah wrote:
> This driver enables r5f dual core Real time Processing Unit subsystem
> available on Xilinx Zynq Ultrascale MPSoC Platform. RPU subsystem
> (cluster) can be configured in different modes e.g. split mode in which
> two r5f cores work independent of each other and lock-step mode in which
> both r5f cores execute same code clock-for-clock and notify if the
> result is different.
>
> The Xilinx r5 Remoteproc Driver boots the RPU cores via calls to the Xilinx
> Platform Management Unit that handles the R5 configuration, memory access
> and R5 lifecycle management. The interface to this manager is done in this
> driver via zynqmp_pm_* function calls.
>
> Signed-off-by: Ben Levinsky <[email protected]>
> Signed-off-by: Tanmay Shah <[email protected]>
> ---
>
> Changes in v3:
> - Fix checkpatch script indentation warning
> - Remove unused variable from xilinx remoteproc driver
> - use C style comments, i.e /*...*/
> - Remove redundant debug information which can be derived using /proc/device-tree
> - Fix multilined comment format
> - s/"final fot TCM"/"final for TCM"
> - Function devm_kzalloc() does not return an code on error, just NULL.
> Remove redundant error check for this function throughout the driver.
> - Fix RPU mode configuration and add documentation accordingly
> - Get rid of the indentations to match function documentation style with rest of the driver
> - Fix memory leak by only using r5_rproc->priv and not replace it with new instance
> - Use 'i' for the outer loop and 'j' for the inner one as per convention
> - Remove redundant error and NULL checks throughout the driver
> - Use devm_kcalloc() when more than one element is required
> - Add memory-regions carveouts during driver probe instead of parse_fw call
> This removes redundant copy of reserved_mem object in r5_core structure.
> - Fix memory leak by using of_node_put()
> - Fix indentation of tcm_mem_map function args
> - Remove redundant init of variables
> - Initialize tcm bank size variable for lockstep mode
> - Replace u32 with phys_addr_t for variable stroing memory bank address
> - Add documentation of TCM behavior in lockstep mode
> - Use dev_get_drvdata instead of platform driver API
> - Remove info level messages
> - Fix checkpatch.pl warnings
> - Add documentation for the Xilinx r5f platform to understand driver design
>
> drivers/remoteproc/Kconfig | 12 +
> drivers/remoteproc/Makefile | 1 +
> drivers/remoteproc/xlnx_r5_remoteproc.c | 973 ++++++++++++++++++++++++
> 3 files changed, 986 insertions(+)
> create mode 100644 drivers/remoteproc/xlnx_r5_remoteproc.c
>
> diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
> index 166019786653..5637a71c0677 100644
> --- a/drivers/remoteproc/Kconfig
> +++ b/drivers/remoteproc/Kconfig
> @@ -352,6 +352,18 @@ config TI_K3_R5_REMOTEPROC
> It's safe to say N here if you're not interested in utilizing
> a slave processor.
>
> +config XLNX_R5_REMOTEPROC
> + tristate "Xilinx R5 remoteproc support"
> + depends on PM && ARCH_ZYNQMP
> + depends on ZYNQMP_FIRMWARE
> + select RPMSG_VIRTIO
> + select ZYNQMP_IPI_MBOX
> + help
> + Say y or m here to support Xilinx R5 remote processors via the remote
> + processor framework.
> +
> + It's safe to say N if not interested in using RPU r5f cores.
> +
> endif # REMOTEPROC
>
> endmenu
> diff --git a/drivers/remoteproc/Makefile b/drivers/remoteproc/Makefile
> index 5478c7cb9e07..91314a9b43ce 100644
> --- a/drivers/remoteproc/Makefile
> +++ b/drivers/remoteproc/Makefile
> @@ -38,3 +38,4 @@ obj-$(CONFIG_ST_SLIM_REMOTEPROC) += st_slim_rproc.o
> obj-$(CONFIG_STM32_RPROC) += stm32_rproc.o
> obj-$(CONFIG_TI_K3_DSP_REMOTEPROC) += ti_k3_dsp_remoteproc.o
> obj-$(CONFIG_TI_K3_R5_REMOTEPROC) += ti_k3_r5_remoteproc.o
> +obj-$(CONFIG_XLNX_R5_REMOTEPROC) += xlnx_r5_remoteproc.o
> diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c
> new file mode 100644
> index 000000000000..ce2ac364b995
> --- /dev/null
> +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c
> @@ -0,0 +1,973 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * ZynqMP R5 Remote Processor driver
> + *
> + */
> +
> +#include <dt-bindings/power/xlnx-zynqmp-power.h>
> +#include <linux/firmware/xlnx-zynqmp.h>
> +#include <linux/interrupt.h>
> +#include <linux/kernel.h>
> +#include <linux/list.h>
> +#include <linux/mailbox_client.h>
> +#include <linux/mailbox/zynqmp-ipi-message.h>
> +#include <linux/module.h>
> +#include <linux/of_address.h>
> +#include <linux/of_platform.h>
> +#include <linux/of_reserved_mem.h>
> +#include <linux/platform_device.h>
> +#include <linux/remoteproc.h>
> +#include <linux/skbuff.h>
> +#include <linux/sysfs.h>
> +
> +#include "remoteproc_internal.h"
> +
> +/* settings for RPU cluster mode */
> +enum zynqmp_r5_cluster_mode {
> + SPLIT_MODE = 0, /* When cores run as separate processor */
> + LOCKSTEP_MODE = 1, /* cores execute same code in lockstep,clk-for-clk */
> + SINGLE_CPU_MODE = 2, /* core0 is held in reset and only core1 runs */
> +};
> +
> +/**
> + * struct mem_bank_data - Memory Bank description
> + *
> + * @addr: Start address of memory bank
> + * @size: Size of Memory bank
> + * @pm_domain_id: Power-domains id of memory bank for firmware to turn on/off
> + * @bank_name: name of the bank for remoteproc framework
> + */
> +struct mem_bank_data {
> + phys_addr_t addr;
> + size_t size;
> + u32 pm_domain_id;
> + char *bank_name;
> +};
> +
> +static const struct mem_bank_data zynqmp_tcm_banks[] = {
> + {0xffe00000UL, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */
> + {0xffe20000UL, 0x10000UL, PD_R5_0_BTCM, "btcm0"},
> + {0xffe90000UL, 0x10000UL, PD_R5_1_ATCM, "atcm1"},
> + {0xffeb0000UL, 0x10000UL, PD_R5_1_BTCM, "btcm1"},
> +};
> +
> +/**
> + * struct zynqmp_r5_core - ZynqMP R5 core structure
> + *
> + * @dev: device of RPU instance
> + * @np: device node of RPU instance
> + * @tcm_bank_count: number TCM banks accessible to this RPU
> + * @tcm_banks: array of each TCM bank data
> + * @rproc: rproc handle
> + * @pm_domain_id: RPU CPU power domain id
> + */
> +struct zynqmp_r5_core {
> + struct device *dev;
> + struct device_node *np;
> + int tcm_bank_count;
> + struct mem_bank_data *tcm_banks;
> + struct rproc *rproc;
> + u32 pm_domain_id;
> +};
> +
> +/**
> + * struct zynqmp_r5_cluster - ZynqMP R5 cluster structure
> + *
> + * @dev: r5f subsystem cluster device node
> + * @mode: cluster mode of type zynqmp_r5_cluster_mode
> + * @core_count: number of r5 cores used for this cluster mode
> + * @r5_cores: Array of pointers pointing to r5 core
> + */
> +struct zynqmp_r5_cluster {
> + struct device *dev;
> + enum zynqmp_r5_cluster_mode mode;
> + int core_count;
> + struct zynqmp_r5_core **r5_cores;
> +};
> +
> +/*
> + * zynqmp_r5_set_mode - set RPU operation mode
> + *
> + * set RPU operation mode
> + *
> + * Return: 0 for success, negative value for failure
> + */
> +static int zynqmp_r5_set_mode(struct zynqmp_r5_core *r5_core,
> + enum zynqmp_r5_cluster_mode rpu_mode)
> +{
> + enum rpu_tcm_comb tcm_mode;
> + int ret, reg_val;
> +
> + /*
> + * SINGLE CPU mode is not supported in driver yet.
> + * For now we only set lockstep mode or split mode.
> + */
> + if (rpu_mode == LOCKSTEP_MODE) {
> + reg_val = 0;
> + } else if (rpu_mode == SPLIT_MODE) {
> + reg_val = 1;
> + } else if (rpu_mode == SINGLE_CPU_MODE) {
> + pr_err("RPU single cpu mode is not supported\n");
> + return -EINVAL;
> + }
I thought we had done all those checks in zynqmp_r5_cluster_init()? Please
choose a single place to do the checks and leave it at that.
> +
> + ret = zynqmp_pm_set_rpu_mode(r5_core->pm_domain_id, reg_val);
> + if (ret < 0) {
> + pr_err("failed to set RPU mode\n");
> + return ret;
> + }
> +
> + /*
> + * At this point RPU is assumed to be configured only in
> + * lockstep mode or split mode. Configure TCM accordingly.
> + */
> + tcm_mode = (rpu_mode == LOCKSTEP_MODE) ?
> + PM_RPU_TCM_COMB : PM_RPU_TCM_SPLIT;
> + ret = zynqmp_pm_set_tcm_config(r5_core->pm_domain_id, tcm_mode);
> + if (ret < 0)
> + pr_err("failed to configure TCM\n");
> +
> + return ret;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_start
> + * @rproc: single R5 core's corresponding rproc instance
> + *
> + * Start R5 Core from designated boot address.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_start(struct rproc *rproc)
> +{
> + struct zynqmp_r5_core *r5_core = rproc->priv;
> + enum rpu_boot_mem bootmem;
> + int ret;
> +
> + /*
> + * The exception vector pointers (EVP) refer to the base-address of
> + * exception vectors (for reset, IRQ, FIQ, etc). The reset-vector
> + * starts at the base-address and subsequent vectors are on 4-byte
> + * boundaries.
> + *
> + * Exception vectors can start either from 0x0000_0000 (LOVEC) or
> + * from 0xFFFF_0000 (HIVEC) which is mapped in the OCM (On-Chip Memory)
> + *
> + * Usually firmware will put Exception vectors at LOVEC.
> + *
> + * It is not recommend that you change the exception vector.
> + * Changing the EVP to HIVEC will result in increased interrupt latency
> + * and jitter. Also, if the OCM is secured and the Cortex-R5F processor
> + * is non-secured, then the Cortex-R5F processor cannot access the
> + * HIVEC exception vectors in the OCM.
> + */
> + bootmem = (rproc->bootaddr >= 0xFFFC0000) ?
> + PM_RPU_BOOTMEM_HIVEC : PM_RPU_BOOTMEM_LOVEC;
> +
> + dev_dbg(r5_core->dev, "RPU boot addr 0x%llx from %s.", rproc->bootaddr,
> + bootmem == PM_RPU_BOOTMEM_HIVEC ? "OCM" : "TCM");
> +
> + ret = zynqmp_pm_request_wake(r5_core->pm_domain_id, 1,
> + bootmem, ZYNQMP_PM_REQUEST_ACK_NO);
> + if (ret)
> + pr_err("failed to start RPU = 0x%x\n", r5_core->pm_domain_id);
> + return ret;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_stop
> + * @rproc: single R5 core's corresponding rproc instance
> + *
> + * Power down R5 Core.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_stop(struct rproc *rproc)
> +{
> + struct zynqmp_r5_core *r5_core = rproc->priv;
> + int ret;
> +
> + ret = zynqmp_pm_force_pwrdwn(r5_core->pm_domain_id,
> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
> + if (ret)
> + pr_err("failed to stop remoteproc RPU %d\n", ret);
> +
> + return ret;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_mem_map
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to map
> + *
> + * Callback to map va for memory-region's carveout.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_mem_map(struct rproc *rproc,
> + struct rproc_mem_entry *mem)
> +{
> + void __iomem *va;
> +
> + va = ioremap_wc(mem->dma, mem->len);
> + if (IS_ERR_OR_NULL(va))
> + return -ENOMEM;
> +
> + mem->va = (void *)va;
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_mem_unmap
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to unmap
> + *
> + * Unmap memory-region carveout
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_rproc_mem_unmap(struct rproc *rproc,
> + struct rproc_mem_entry *mem)
> +{
> + iounmap((void __iomem *)mem->va);
> + return 0;
> +}
> +
> +/*
> + * add_mem_regions
> + * @r5_core: single R5 core's corresponding zynqmp_r5_core type instance
> + * @rmem: reserved mem region parsed from dt node
> + *
> + * Construct rproc mem carveouts from carveout provided in
> + * memory-region property
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int add_mem_regions_carveout(struct zynqmp_r5_core *r5_core,
> + struct reserved_mem *rmem)
> +{
> + struct device *dev;
> + struct rproc_mem_entry *mem;
> + struct rproc *rproc;
> +
> + rproc = r5_core->rproc;
> + dev = r5_core->dev;
> +
> + /* Register associated reserved memory regions */
> + mem = rproc_mem_entry_init(dev, NULL,
> + (dma_addr_t)rmem->base,
> + rmem->size, rmem->base,
> + zynqmp_r5_rproc_mem_map,
> + zynqmp_r5_rproc_mem_unmap,
> + rmem->name);
> + if (!mem)
> + return -ENOMEM;
> +
> + rproc_add_carveout(rproc, mem);
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_rproc_mem_unmap
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to unmap
> + *
> + * Unmap TCM banks when powering down R5 core.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int tcm_mem_unmap(struct rproc *rproc, struct rproc_mem_entry *mem)
> +{
> + struct zynqmp_r5_core *r5_core;
> + int i;
> + u32 pm_domain_id;
> +
> + r5_core = rproc->priv;
> + if (!r5_core) {
> + pr_err("r5 core is not available\n");
> + return -EINVAL;
> + }
> +
> + iounmap((void __iomem *)mem->va);
> +
> + for (i = 0; i < r5_core->tcm_bank_count; i++) {
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> + if (zynqmp_pm_release_node(pm_domain_id))
> + pr_warn("can't turn off TCM bank 0x%x", pm_domain_id);
> + }
> +
> + return 0;
> +}
> +
> +/*
> + * tcm_mem_map
> + * @rproc: single R5 core's corresponding rproc instance
> + * @mem: mem entry to initialize the va and da fields of
> + *
> + * Given TCM bank entry, this callback will set device address for R5
> + * running on TCM and also setup virtual address for TCM bank
> + * remoteproc carveout.
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int tcm_mem_map(struct rproc *rproc,
> + struct rproc_mem_entry *mem)
> +{
> + void __iomem *va;
> +
> + va = ioremap_wc(mem->dma, mem->len);
> + if (IS_ERR_OR_NULL(va))
> + return -ENOMEM;
> +
> + /* Update memory entry va */
> + mem->va = (void *)va;
> +
> + /* clear TCMs */
> + memset_io(va, 0, mem->len);
> +
> + /*
> + * The R5s expect their TCM banks to be at address 0x0 and 0x2000,
> + * while on the Linux side they are at 0xffexxxxx.
> + *
> + * Zero out the high 12 bits of the address. This will give
> + * expected values for TCM Banks 0A and 0B (0x0 and 0x20000).
> + */
> + mem->da &= 0x000fffff;
> +
> + /*
> + * TCM Banks 1A and 1B still have to be translated.
> + *
> + * Below handle these two banks' absolute addresses (0xffe90000 and
> + * 0xffeb0000) and convert to the expected relative addresses
> + * (0x0 and 0x20000).
> + */
> + if (mem->da == 0x90000 || mem->da == 0xB0000)
> + mem->da -= 0x90000;
> +
> + /* if translated TCM bank address is not valid report error */
> + if (mem->da != 0x0 && mem->da != 0x20000) {
> + dev_err(&rproc->dev, "invalid TCM address: %x\n", mem->da);
> + return -EINVAL;
> + }
> + return 0;
> +}
> +
> +static int add_tcm_carveout_split_mode(struct rproc *rproc)
> +{
> + int i, num_banks, ret;
> + struct rproc_mem_entry *mem;
> + u32 pm_domain_id;
> + phys_addr_t bank_addr;
> + size_t bank_size;
> + char *bank_name;
> + struct device *dev;
> + struct zynqmp_r5_core *r5_core;
> +
> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
> +
> + dev = r5_core->dev;
> +
> + /* go through zynqmp banks for r5 node */
> + num_banks = r5_core->tcm_bank_count;
> + if (num_banks <= 0) {
> + dev_err(dev, "need to specify TCM banks\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * Power-on Each 64KB TCM,
> + * register its address space, map and unmap functions
> + * and add carveouts accordingly
> + */
> + for (i = 0; i < num_banks; i++) {
> + bank_addr = r5_core->tcm_banks[i].addr;
> + bank_name = r5_core->tcm_banks[i].bank_name;
> + bank_size = r5_core->tcm_banks[i].size;
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> +
> + ret = zynqmp_pm_request_node(pm_domain_id,
> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
> + if (ret < 0) {
> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
> + return ret;
> + }
> +
> + dev_dbg(dev, "TCM carveout split mode %s addr=%llx, size=0x%lx",
> + bank_name, bank_addr, bank_size);
> +
> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
> + bank_size, bank_addr,
> + tcm_mem_map, tcm_mem_unmap,
> + bank_name);
> + if (!mem) {
> + /* If failed, Turn off all TCM banks turned on before */
> + do {
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> + ret = zynqmp_pm_release_node(pm_domain_id);
> + if (ret)
> + dev_warn(dev,
> + "fail to release node: 0x%x, %x\n",
> + pm_domain_id, ret);
> + } while (i--);
> + return -ENOMEM;
> + }
> +
> + rproc_add_carveout(rproc, mem);
> + }
> +
> + return 0;
> +}
> +
> +static int add_tcm_carveout_lockstep_mode(struct rproc *rproc)
> +{
> + int i, num_banks, ret;
> + struct rproc_mem_entry *mem;
> + u32 pm_domain_id;
> + phys_addr_t bank_addr;
> + size_t bank_size = 0;
> + char *bank_name;
> + struct device *dev;
> + struct zynqmp_r5_core *r5_core;
> +
> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
> + dev = r5_core->dev;
> +
> + /* Go through zynqmp banks for r5 node */
> + num_banks = r5_core->tcm_bank_count;
> + if (num_banks <= 0) {
> + dev_err(dev, "need to specify TCM banks\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * In lockstep mode, TCM is contiguous memory block
> + * However, each TCM block still needs to be enabled individually.
> + * So, Enable each TCM block individually, but add their size
> + * to create contiguous memory region.
> + */
> + bank_addr = r5_core->tcm_banks[0].addr;
> + bank_name = r5_core->tcm_banks[0].bank_name;
> +
> + for (i = 0; i < num_banks; i++) {
> + bank_size += r5_core->tcm_banks[i].size;
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> +
> + /* Turn on each TCM bank individually */
> + ret = zynqmp_pm_request_node(pm_domain_id,
> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
> + if (ret < 0) {
> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
> + return ret;
> + }
> + }
> +
> + dev_dbg(dev, "TCM add carveout lockstep mode %s addr=0x%llx, size=0x%lx",
> + bank_name, bank_addr, bank_size);
> +
> + /* Register TCM address range, TCM map and unmap functions */
> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
> + bank_size, bank_addr,
> + tcm_mem_map, tcm_mem_unmap,
> + bank_name);
> + if (!mem) {
> + /* In case of error, turn off TCM banks */
> + for (i = 0; i < num_banks; i++) {
> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
> + ret = zynqmp_pm_release_node(pm_domain_id);
> + if (ret)
> + dev_warn(dev,
> + "fail to release node: 0x%x ret: %x\n",
> + pm_domain_id, ret);
> + }
> + return -ENOMEM;
> + }
> +
> + /* If registration is success, add carveouts */
> + rproc_add_carveout(rproc, mem);
> +
> + return 0;
> +}
> +
> +/*
> + * add_tcm_banks()
> + * @rproc: single R5 core's corresponding rproc instance
> + *
> + * Given R5 node in remoteproc instance
> + * allocate remoteproc carveout for TCM memory
> + * needed for firmware to be loaded
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int add_tcm_banks(struct rproc *rproc)
> +{
> + struct device *dev;
> + struct platform_device *parent_pdev;
> + struct zynqmp_r5_cluster *cluster;
> + struct zynqmp_r5_core *r5_core;
> +
> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
> + if (!r5_core)
> + return -EINVAL;
> +
> + dev = r5_core->dev;
> +
> + cluster = dev_get_drvdata(dev->parent);
> + if (!cluster) {
> + dev_err(&parent_pdev->dev, "Invalid driver data\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * In lockstep mode TCM banks are one contiguous memory region of 256Kb
> + * In split mode, each TCM bank is 64Kb and not contiguous.
> + * We add memory carveouts accordingly.
> + */
> + if (cluster->mode == SPLIT_MODE)
> + return add_tcm_carveout_split_mode(rproc);
> + else if (cluster->mode == LOCKSTEP_MODE)
> + return add_tcm_carveout_lockstep_mode(rproc);
> +
> + dev_err(cluster->dev, "invalid cluster mode\n");
> + return -EINVAL;
> +}
> +
> +/*
> + * zynqmp_r5_parse_fw()
> + * @rproc: single R5 core's corresponding rproc instance
> + * @fw: ptr to firmware to be loaded onto r5 core
> + *
> + * When loading firmware, ensure the necessary carveouts are in remoteproc
> + *
> + * return 0 on success, otherwise non-zero value on failure
> + */
> +static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct firmware *fw)
> +{
> + int ret;
> + struct zynqmp_r5_core *r5_core;
> + struct device *dev;
> +
> + r5_core = rproc->priv;
> +
> + dev = r5_core->dev;
> +
> + ret = add_tcm_banks(rproc);
> + if (ret) {
> + dev_err(dev, "failed to get TCM banks, err %d\n", ret);
> + return ret;
> + }
> +
> + ret = rproc_elf_load_rsc_table(rproc, fw);
> + if (ret == -EINVAL) {
> + /*
> + * resource table only required for IPC.
> + * if not present, this is not necessarily an error;
> + * for example, loading r5 hello world application
> + * so simply inform user and keep going.
> + */
> + dev_info(&rproc->dev, "no resource table found.\n");
> + ret = 0;
> + }
> + return ret;
> +}
> +
> +static struct rproc_ops zynqmp_r5_rproc_ops = {
> + .start = zynqmp_r5_rproc_start,
> + .stop = zynqmp_r5_rproc_stop,
> + .load = rproc_elf_load_segments,
> + .parse_fw = zynqmp_r5_parse_fw,
> + .find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
> + .sanity_check = rproc_elf_sanity_check,
> + .get_boot_addr = rproc_elf_get_boot_addr,
> +};
> +
> +/**
> + * zynqmp_r5_add_rproc_core()
> + * Probes ZynqMP R5 processor device node
> + * this is called for each individual R5 core
> + *
> + * @cdev: Device node of each r5 core
> + *
> + * Return: zynqmp_r5_core object for success, error pointer in case of error.
> + */
> +static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev)
> +{
> + int ret;
> + struct rproc *r5_rproc;
> + struct zynqmp_r5_core *r5_core;
> +
> + /* Set up DMA mask */
> + ret = dma_set_coherent_mask(cdev, DMA_BIT_MASK(32));
> + if (ret)
> + return ERR_PTR(ret);
> +
> + /* Allocate remoteproc instance */
> + r5_rproc = devm_rproc_alloc(cdev, dev_name(cdev),
> + &zynqmp_r5_rproc_ops,
> + NULL, sizeof(struct zynqmp_r5_core));
> + if (!r5_rproc) {
> + dev_err(cdev, "failed to allocate memory for rproc instance\n");
> + return ERR_PTR(-ENOMEM);
> + }
> +
> + r5_rproc->auto_boot = false;
> + r5_core = (struct zynqmp_r5_core *)r5_rproc->priv;
> + r5_core->dev = cdev;
> +
> + /* Add R5 remoteproc core */
> + ret = devm_rproc_add(cdev, r5_rproc);
> + if (ret) {
> + dev_err(cdev, "failed to add r5 remoteproc\n");
> + return ERR_PTR(ret);
> + }
> +
> + r5_core->rproc = r5_rproc;
> + return r5_core;
> +}
> +
> +/**
> + * zynqmp_r5_get_tcm_node()
> + * Ideally this function should parse tcm node and store information
> + * in r5_core instance. We will use hardcoded TCM information from
> + * driver for now in this function.
> + *
> + * @cluster: pointer to zynqmp_r5_cluster type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster)
> +{
> + int tcm_bank_count, tcm_node;
> + int i = 0, j;
Variable @i doesn't need to be initialised.
> + struct zynqmp_r5_core *r5_core;
> + const struct mem_bank_data *tcm = zynqmp_tcm_banks;
> + struct device *dev = cluster->dev;
> +
> + /*
> + * ToDo: Use predefined TCM address space values from driver until
> + * system-dt spec is not final for TCM
> + */
> + tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks);
> +
> + /* count per core tcm banks */
> + tcm_bank_count = tcm_bank_count / cluster->core_count;
> +
> + /* r5 core 0 will use all of TCM banks in lockstep mode.
> + * In split mode, r5 core0 will use 128k and r5 core1 will use another
> + * 128k. Assign TCM banks to each core accordingly
> + */
Comment format problem.
> + tcm_node = 0;
> + for (i = 0; i < cluster->core_count; i++) {
> + r5_core = cluster->r5_cores[i];
> + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count,
> + sizeof(struct mem_bank_data),
> + GFP_KERNEL);
> + if (!r5_core->tcm_banks)
> + return -ENOMEM;
> +
> + for (j = 0; j < tcm_bank_count; j++) {
> + /* Use pre-defined TCM reg values.
> + * Eventually this should be replaced by values
> + * parsed from dts.
> + */
Same. I commented on that in my previous review of this set.
> + r5_core->tcm_banks[j].addr = tcm[tcm_node].addr;
> + r5_core->tcm_banks[j].size = tcm[tcm_node].size;
> + r5_core->tcm_banks[j].pm_domain_id = tcm[tcm_node].pm_domain_id;
> + r5_core->tcm_banks[j].bank_name = tcm[tcm_node].bank_name;
> + tcm_node++;
> + }
> +
> + r5_core->tcm_bank_count = tcm_bank_count;
Why do we need to copy the exact same information from zynqmp_tcm_banks to ->tcm_banks? Why can't ->tcm_banks point to an offset in zynqmp_tcm_banks and ->tcm_bank_count set accordingly?
> + }
> +
> + return 0;
> +}
> +
> +/**
> + * zynqmp_r5_get_mem_region_node()
> + * parse memory-region property from dt node and add
> + * memory region carveouts
> + *
> + * @r5_core: pointer to zynqmp_r5_core type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core *r5_core)
> +{
> + int res_mem_count, i, ret;
> + struct device *dev;
> + struct device_node *np, *rmem_np;
> + struct reserved_mem *rmem;
> +
> + dev = r5_core->dev;
> +
> + np = r5_core->np;
> +
> + res_mem_count = of_property_count_elems_of_size(np, "memory-region",
> + sizeof(phandle));
> + if (res_mem_count <= 0) {
> + dev_warn(dev, "failed to get memory-region property %d\n",
> + res_mem_count);
> + return -EINVAL;
> + }
> +
> + for (i = 0; i < res_mem_count; i++) {
> + rmem_np = of_parse_phandle(np, "memory-region", i);
> + if (!rmem_np)
> + return -EINVAL;
> +
> + rmem = of_reserved_mem_lookup(rmem_np);
> + if (!rmem) {
> + of_node_put(rmem_np);
> + return -EINVAL;
What happens to previously allocated regions when there is a failure?
> + }
> +
> + ret = add_mem_regions_carveout(r5_core, rmem);
> + if (ret)
> + dev_warn(dev, "failed to get reserve mem regions %d\n",
> + ret);
Same here.
I am out of time for this set. Please address comments provided up to here and
we will see about the rest in a future revision.
Thanks,
Mathieu
> +
> + of_node_put(rmem_np);
> + }
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_core_init()
> + * Create and initialize zynqmp_r5_core type object
> + *
> + * @cluster: pointer to zynqmp_r5_cluster type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
> +{
> + int ret, i;
> + struct zynqmp_r5_core *r5_core;
> + struct device *dev = cluster->dev;
> +
> + ret = zynqmp_r5_get_tcm_node(cluster);
> + if (ret < 0) {
> + dev_err(dev, "can't get tcm node, err %d\n", ret);
> + return ret;
> + }
> +
> + for (i = 0; i < cluster->core_count; i++) {
> + r5_core = cluster->r5_cores[i];
> +
> + ret = zynqmp_r5_get_mem_region_node(r5_core);
> + if (ret)
> + dev_warn(dev, "memory-region prop failed %d\n", ret);
> +
> + /* Initialize r5 cores with power-domains parsed from dts */
> + ret = of_property_read_u32_index(r5_core->np, "power-domains",
> + 1, &r5_core->pm_domain_id);
> + if (ret) {
> + dev_err(dev, "failed to get power-domains property\n");
> + return ret;
> + }
> +
> + ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
> + if (ret) {
> + dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
> + cluster->mode, ret);
> + return ret;
> + }
> + }
> +
> + return 0;
> +}
> +
> +/*
> + * zynqmp_r5_cluster_init()
> + * Create and initialize zynqmp_r5_cluster type object
> + *
> + * @cluster: pointer to zynqmp_r5_cluster type object
> + *
> + * Return: 0 for success and error code for failure.
> + */
> +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
> +{
> + struct device *dev = cluster->dev;
> + struct device_node *dev_node = dev_of_node(dev);
> + struct device_node *child;
> + struct platform_device *child_pdev;
> + int core_count = 0, ret, i;
> + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
> + struct zynqmp_r5_core **r5_cores;
> +
> + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
> +
> + /*
> + * on success returns 0, if not defined then returns -EINVAL,
> + * In that case, default is LOCKSTEP mode
> + */
> + if (ret != -EINVAL && ret != 0) {
> + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * For now driver only supports split mode and lockstep mode.
> + * fail driver probe if either of that is not set in dts
> + */
> + if (cluster_mode == SINGLE_CPU_MODE) {
> + dev_err(dev, "driver does not support single cpu mode\n");
> + return -EINVAL;
> + } else if ((cluster_mode != SPLIT_MODE &&
> + cluster_mode != LOCKSTEP_MODE)) {
> + dev_err(dev, "Invalid cluster mode\n");
> + return -EINVAL;
> + }
> +
> + /*
> + * Number of cores is decided by number of child nodes of
> + * r5f subsystem node in dts. If Split mode is used in dts
> + * 2 child nodes are expected.
> + * In lockstep mode if two child nodes are available,
> + * only use first child node and consider it as core0
> + * and ignore core1 dt node.
> + */
> + core_count = of_get_available_child_count(dev_node);
> + if (core_count <= 0) {
> + dev_err(dev, "Invalid number of r5 cores %d", core_count);
> + return -EINVAL;
> + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
> + dev_err(dev, "Invalid number of r5 cores for split mode\n");
> + return -EINVAL;
> + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
> + dev_warn(dev, "Only r5 core0 will be used\n");
> + core_count = 1;
> + }
> +
> + r5_cores = devm_kcalloc(dev, core_count,
> + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
> + if (!r5_cores)
> + return -ENOMEM;
> +
> + i = 0;
> + for_each_available_child_of_node(dev_node, child) {
> + child_pdev = of_find_device_by_node(child);
> + if (!child_pdev) {
> + of_node_put(child);
> + return -ENODEV;
> + }
> +
> + /* create and add remoteproc instance of type struct rproc */
> + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
> + r5_cores[i]->dev = &child_pdev->dev;
> + if (!r5_cores[i]->dev) {
> + dev_err(dev, "can't get device for r5 core %d\n", i);
> + of_node_put(child);
> + return -ENODEV;
> + }
> +
> + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
> + if (!r5_cores[i]->np) {
> + dev_err(dev, "can't get device node for r5 core %d\n", i);
> + of_node_put(child);
> + return -ENODEV;
> + }
> +
> + i++;
> +
> + /*
> + * If two child nodes are available in dts in lockstep mode,
> + * then ignore second child node.
> + */
> + if (i == core_count) {
> + of_node_put(child);
> + break;
> + }
> + of_node_put(child);
> + }
> +
> + cluster->mode = cluster_mode;
> + cluster->core_count = core_count;
> + cluster->r5_cores = r5_cores;
> +
> + ret = zynqmp_r5_core_init(cluster);
> + if (ret < 0) {
> + dev_err(dev, "failed to init r5 core err %d\n", ret);
> + return ret;
> + }
> +
> + return 0;
> +}
> +
> +static void zynqmp_r5_cluster_exit(void *data)
> +{
> + struct platform_device *pdev = (struct platform_device *)data;
> +
> + platform_set_drvdata(pdev, NULL);
> +
> + pr_info("Exit r5f subsystem driver\n");
> +}
> +
> +/*
> + * zynqmp_r5_remoteproc_probe()
> + *
> + * @pdev: domain platform device for R5 cluster
> + *
> + * called when driver is probed, for each R5 core specified in DT,
> + * setup as needed to do remoteproc-related operations
> + *
> + * Return: 0 for success, negative value for failure.
> + */
> +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
> +{
> + int ret;
> + struct zynqmp_r5_cluster *cluster;
> + struct device *dev = &pdev->dev;
> +
> + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
> + if (!cluster)
> + return -ENOMEM;
> +
> + cluster->dev = dev;
> +
> + ret = devm_of_platform_populate(dev);
> + if (ret) {
> + dev_err(dev, "failed to populate platform dev %d\n", ret);
> + return ret;
> + }
> +
> + /* wire in so each core can be cleaned up at driver remove */
> + platform_set_drvdata(pdev, cluster);
> +
> + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
> + if (ret)
> + return ret;
> +
> + ret = zynqmp_r5_cluster_init(cluster);
> + if (ret) {
> + dev_err(dev, "Invalid r5f subsystem device tree\n");
> + return ret;
> + }
> +
> + return 0;
> +}
> +
> +/* Match table for OF platform binding */
> +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
> + { .compatible = "xlnx,zynqmp-r5fss", },
> + { /* end of list */ },
> +};
> +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
> +
> +static struct platform_driver zynqmp_r5_remoteproc_driver = {
> + .probe = zynqmp_r5_remoteproc_probe,
> + .driver = {
> + .name = "zynqmp_r5_remoteproc",
> + .of_match_table = zynqmp_r5_remoteproc_match,
> + },
> +};
> +module_platform_driver(zynqmp_r5_remoteproc_driver);
> +
> +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
> +MODULE_AUTHOR("Xilinx Inc.");
> +MODULE_LICENSE("GPL v2");
> --
> 2.25.1
>
On 2/18/22 11:11 AM, Mathieu Poirier wrote:
> On Thu, Feb 10, 2022 at 03:28:24AM -0800, Tanmay Shah wrote:
>> This driver enables r5f dual core Real time Processing Unit subsystem
>> available on Xilinx Zynq Ultrascale MPSoC Platform. RPU subsystem
>> (cluster) can be configured in different modes e.g. split mode in which
>> two r5f cores work independent of each other and lock-step mode in which
>> both r5f cores execute same code clock-for-clock and notify if the
>> result is different.
>>
>> The Xilinx r5 Remoteproc Driver boots the RPU cores via calls to the Xilinx
>> Platform Management Unit that handles the R5 configuration, memory access
>> and R5 lifecycle management. The interface to this manager is done in this
>> driver via zynqmp_pm_* function calls.
>>
>> Signed-off-by: Ben Levinsky<[email protected]>
>> Signed-off-by: Tanmay Shah<[email protected]>
>> ---
>>
>> Changes in v3:
>> - Fix checkpatch script indentation warning
>> - Remove unused variable from xilinx remoteproc driver
>> - use C style comments, i.e /*...*/
>> - Remove redundant debug information which can be derived using /proc/device-tree
>> - Fix multilined comment format
>> - s/"final fot TCM"/"final for TCM"
>> - Function devm_kzalloc() does not return an code on error, just NULL.
>> Remove redundant error check for this function throughout the driver.
>> - Fix RPU mode configuration and add documentation accordingly
>> - Get rid of the indentations to match function documentation style with rest of the driver
>> - Fix memory leak by only using r5_rproc->priv and not replace it with new instance
>> - Use 'i' for the outer loop and 'j' for the inner one as per convention
>> - Remove redundant error and NULL checks throughout the driver
>> - Use devm_kcalloc() when more than one element is required
>> - Add memory-regions carveouts during driver probe instead of parse_fw call
>> This removes redundant copy of reserved_mem object in r5_core structure.
>> - Fix memory leak by using of_node_put()
>> - Fix indentation of tcm_mem_map function args
>> - Remove redundant init of variables
>> - Initialize tcm bank size variable for lockstep mode
>> - Replace u32 with phys_addr_t for variable stroing memory bank address
>> - Add documentation of TCM behavior in lockstep mode
>> - Use dev_get_drvdata instead of platform driver API
>> - Remove info level messages
>> - Fix checkpatch.pl warnings
>> - Add documentation for the Xilinx r5f platform to understand driver design
>>
>> drivers/remoteproc/Kconfig | 12 +
>> drivers/remoteproc/Makefile | 1 +
>> drivers/remoteproc/xlnx_r5_remoteproc.c | 973 ++++++++++++++++++++++++
>> 3 files changed, 986 insertions(+)
>> create mode 100644 drivers/remoteproc/xlnx_r5_remoteproc.c
>>
>> diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
>> index 166019786653..5637a71c0677 100644
>> --- a/drivers/remoteproc/Kconfig
>> +++ b/drivers/remoteproc/Kconfig
>> @@ -352,6 +352,18 @@ config TI_K3_R5_REMOTEPROC
>> It's safe to say N here if you're not interested in utilizing
>> a slave processor.
>>
>> +config XLNX_R5_REMOTEPROC
>> + tristate "Xilinx R5 remoteproc support"
>> + depends on PM && ARCH_ZYNQMP
>> + depends on ZYNQMP_FIRMWARE
>> + select RPMSG_VIRTIO
>> + select ZYNQMP_IPI_MBOX
>> + help
>> + Say y or m here to support Xilinx R5 remote processors via the remote
>> + processor framework.
>> +
>> + It's safe to say N if not interested in using RPU r5f cores.
>> +
>> endif # REMOTEPROC
>>
>> endmenu
>> diff --git a/drivers/remoteproc/Makefile b/drivers/remoteproc/Makefile
>> index 5478c7cb9e07..91314a9b43ce 100644
>> --- a/drivers/remoteproc/Makefile
>> +++ b/drivers/remoteproc/Makefile
>> @@ -38,3 +38,4 @@ obj-$(CONFIG_ST_SLIM_REMOTEPROC) += st_slim_rproc.o
>> obj-$(CONFIG_STM32_RPROC) += stm32_rproc.o
>> obj-$(CONFIG_TI_K3_DSP_REMOTEPROC) += ti_k3_dsp_remoteproc.o
>> obj-$(CONFIG_TI_K3_R5_REMOTEPROC) += ti_k3_r5_remoteproc.o
>> +obj-$(CONFIG_XLNX_R5_REMOTEPROC) += xlnx_r5_remoteproc.o
>> diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c
>> new file mode 100644
>> index 000000000000..ce2ac364b995
>> --- /dev/null
>> +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c
>> @@ -0,0 +1,973 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * ZynqMP R5 Remote Processor driver
>> + *
>> + */
>> +
>> +#include <dt-bindings/power/xlnx-zynqmp-power.h>
>> +#include <linux/firmware/xlnx-zynqmp.h>
>> +#include <linux/interrupt.h>
>> +#include <linux/kernel.h>
>> +#include <linux/list.h>
>> +#include <linux/mailbox_client.h>
>> +#include <linux/mailbox/zynqmp-ipi-message.h>
>> +#include <linux/module.h>
>> +#include <linux/of_address.h>
>> +#include <linux/of_platform.h>
>> +#include <linux/of_reserved_mem.h>
>> +#include <linux/platform_device.h>
>> +#include <linux/remoteproc.h>
>> +#include <linux/skbuff.h>
>> +#include <linux/sysfs.h>
>> +
>> +#include "remoteproc_internal.h"
>> +
>> +/* settings for RPU cluster mode */
>> +enum zynqmp_r5_cluster_mode {
>> + SPLIT_MODE = 0, /* When cores run as separate processor */
>> + LOCKSTEP_MODE = 1, /* cores execute same code in lockstep,clk-for-clk */
>> + SINGLE_CPU_MODE = 2, /* core0 is held in reset and only core1 runs */
>> +};
>> +
>> +/**
>> + * struct mem_bank_data - Memory Bank description
>> + *
>> + * @addr: Start address of memory bank
>> + * @size: Size of Memory bank
>> + * @pm_domain_id: Power-domains id of memory bank for firmware to turn on/off
>> + * @bank_name: name of the bank for remoteproc framework
>> + */
>> +struct mem_bank_data {
>> + phys_addr_t addr;
>> + size_t size;
>> + u32 pm_domain_id;
>> + char *bank_name;
>> +};
>> +
>> +static const struct mem_bank_data zynqmp_tcm_banks[] = {
>> + {0xffe00000UL, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */
>> + {0xffe20000UL, 0x10000UL, PD_R5_0_BTCM, "btcm0"},
>> + {0xffe90000UL, 0x10000UL, PD_R5_1_ATCM, "atcm1"},
>> + {0xffeb0000UL, 0x10000UL, PD_R5_1_BTCM, "btcm1"},
>> +};
>> +
>> +/**
>> + * struct zynqmp_r5_core - ZynqMP R5 core structure
>> + *
>> + * @dev: device of RPU instance
>> + * @np: device node of RPU instance
>> + * @tcm_bank_count: number TCM banks accessible to this RPU
>> + * @tcm_banks: array of each TCM bank data
>> + * @rproc: rproc handle
>> + * @pm_domain_id: RPU CPU power domain id
>> + */
>> +struct zynqmp_r5_core {
>> + struct device *dev;
>> + struct device_node *np;
>> + int tcm_bank_count;
>> + struct mem_bank_data *tcm_banks;
>> + struct rproc *rproc;
>> + u32 pm_domain_id;
>> +};
>> +
>> +/**
>> + * struct zynqmp_r5_cluster - ZynqMP R5 cluster structure
>> + *
>> + * @dev: r5f subsystem cluster device node
>> + * @mode: cluster mode of type zynqmp_r5_cluster_mode
>> + * @core_count: number of r5 cores used for this cluster mode
>> + * @r5_cores: Array of pointers pointing to r5 core
>> + */
>> +struct zynqmp_r5_cluster {
>> + struct device *dev;
>> + enum zynqmp_r5_cluster_mode mode;
>> + int core_count;
>> + struct zynqmp_r5_core **r5_cores;
>> +};
>> +
>> +/*
>> + * zynqmp_r5_set_mode - set RPU operation mode
>> + *
>> + * set RPU operation mode
>> + *
>> + * Return: 0 for success, negative value for failure
>> + */
>> +static int zynqmp_r5_set_mode(struct zynqmp_r5_core *r5_core,
>> + enum zynqmp_r5_cluster_mode rpu_mode)
>> +{
>> + enum rpu_tcm_comb tcm_mode;
>> + int ret, reg_val;
>> +
>> + /*
>> + * SINGLE CPU mode is not supported in driver yet.
>> + * For now we only set lockstep mode or split mode.
>> + */
>> + if (rpu_mode == LOCKSTEP_MODE) {
>> + reg_val = 0;
>> + } else if (rpu_mode == SPLIT_MODE) {
>> + reg_val = 1;
>> + } else if (rpu_mode == SINGLE_CPU_MODE) {
>> + pr_err("RPU single cpu mode is not supported\n");
>> + return -EINVAL;
>> + }
> I thought we had done all those checks in zynqmp_r5_cluster_init
> ()? Please
> choose a single place to do the checks and leave it at that.
Sure. I will add two more variables in zynqmp_r5_cluster i.e. r5_reg_val
and tcm_mode
and set them in zynqmp_r5_cluster_init
>> +
>> + ret = zynqmp_pm_set_rpu_mode(r5_core->pm_domain_id, reg_val);
>> + if (ret < 0) {
>> + pr_err("failed to set RPU mode\n");
>> + return ret;
>> + }
>> +
>> + /*
>> + * At this point RPU is assumed to be configured only in
>> + * lockstep mode or split mode. Configure TCM accordingly.
>> + */
>> + tcm_mode = (rpu_mode == LOCKSTEP_MODE) ?
>> + PM_RPU_TCM_COMB : PM_RPU_TCM_SPLIT;
>> + ret = zynqmp_pm_set_tcm_config(r5_core->pm_domain_id, tcm_mode);
>> + if (ret < 0)
>> + pr_err("failed to configure TCM\n");
>> +
>> + return ret;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_start
>> + * @rproc: single R5 core's corresponding rproc instance
>> + *
>> + * Start R5 Core from designated boot address.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_start(struct rproc *rproc)
>> +{
>> + struct zynqmp_r5_core *r5_core = rproc->priv;
>> + enum rpu_boot_mem bootmem;
>> + int ret;
>> +
>> + /*
>> + * The exception vector pointers (EVP) refer to the base-address of
>> + * exception vectors (for reset, IRQ, FIQ, etc). The reset-vector
>> + * starts at the base-address and subsequent vectors are on 4-byte
>> + * boundaries.
>> + *
>> + * Exception vectors can start either from 0x0000_0000 (LOVEC) or
>> + * from 0xFFFF_0000 (HIVEC) which is mapped in the OCM (On-Chip Memory)
>> + *
>> + * Usually firmware will put Exception vectors at LOVEC.
>> + *
>> + * It is not recommend that you change the exception vector.
>> + * Changing the EVP to HIVEC will result in increased interrupt latency
>> + * and jitter. Also, if the OCM is secured and the Cortex-R5F processor
>> + * is non-secured, then the Cortex-R5F processor cannot access the
>> + * HIVEC exception vectors in the OCM.
>> + */
>> + bootmem = (rproc->bootaddr >= 0xFFFC0000) ?
>> + PM_RPU_BOOTMEM_HIVEC : PM_RPU_BOOTMEM_LOVEC;
>> +
>> + dev_dbg(r5_core->dev, "RPU boot addr 0x%llx from %s.", rproc->bootaddr,
>> + bootmem == PM_RPU_BOOTMEM_HIVEC ? "OCM" : "TCM");
>> +
>> + ret = zynqmp_pm_request_wake(r5_core->pm_domain_id, 1,
>> + bootmem, ZYNQMP_PM_REQUEST_ACK_NO);
>> + if (ret)
>> + pr_err("failed to start RPU = 0x%x\n", r5_core->pm_domain_id);
>> + return ret;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_stop
>> + * @rproc: single R5 core's corresponding rproc instance
>> + *
>> + * Power down R5 Core.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_stop(struct rproc *rproc)
>> +{
>> + struct zynqmp_r5_core *r5_core = rproc->priv;
>> + int ret;
>> +
>> + ret = zynqmp_pm_force_pwrdwn(r5_core->pm_domain_id,
>> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
>> + if (ret)
>> + pr_err("failed to stop remoteproc RPU %d\n", ret);
>> +
>> + return ret;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_mem_map
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to map
>> + *
>> + * Callback to map va for memory-region's carveout.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_mem_map(struct rproc *rproc,
>> + struct rproc_mem_entry *mem)
>> +{
>> + void __iomem *va;
>> +
>> + va = ioremap_wc(mem->dma, mem->len);
>> + if (IS_ERR_OR_NULL(va))
>> + return -ENOMEM;
>> +
>> + mem->va = (void *)va;
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_mem_unmap
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to unmap
>> + *
>> + * Unmap memory-region carveout
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_mem_unmap(struct rproc *rproc,
>> + struct rproc_mem_entry *mem)
>> +{
>> + iounmap((void __iomem *)mem->va);
>> + return 0;
>> +}
>> +
>> +/*
>> + * add_mem_regions
>> + * @r5_core: single R5 core's corresponding zynqmp_r5_core type instance
>> + * @rmem: reserved mem region parsed from dt node
>> + *
>> + * Construct rproc mem carveouts from carveout provided in
>> + * memory-region property
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int add_mem_regions_carveout(struct zynqmp_r5_core *r5_core,
>> + struct reserved_mem *rmem)
>> +{
>> + struct device *dev;
>> + struct rproc_mem_entry *mem;
>> + struct rproc *rproc;
>> +
>> + rproc = r5_core->rproc;
>> + dev = r5_core->dev;
>> +
>> + /* Register associated reserved memory regions */
>> + mem = rproc_mem_entry_init(dev, NULL,
>> + (dma_addr_t)rmem->base,
>> + rmem->size, rmem->base,
>> + zynqmp_r5_rproc_mem_map,
>> + zynqmp_r5_rproc_mem_unmap,
>> + rmem->name);
>> + if (!mem)
>> + return -ENOMEM;
>> +
>> + rproc_add_carveout(rproc, mem);
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_mem_unmap
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to unmap
>> + *
>> + * Unmap TCM banks when powering down R5 core.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int tcm_mem_unmap(struct rproc *rproc, struct rproc_mem_entry *mem)
>> +{
>> + struct zynqmp_r5_core *r5_core;
>> + int i;
>> + u32 pm_domain_id;
>> +
>> + r5_core = rproc->priv;
>> + if (!r5_core) {
>> + pr_err("r5 core is not available\n");
>> + return -EINVAL;
>> + }
>> +
>> + iounmap((void __iomem *)mem->va);
>> +
>> + for (i = 0; i < r5_core->tcm_bank_count; i++) {
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> + if (zynqmp_pm_release_node(pm_domain_id))
>> + pr_warn("can't turn off TCM bank 0x%x", pm_domain_id);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * tcm_mem_map
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to initialize the va and da fields of
>> + *
>> + * Given TCM bank entry, this callback will set device address for R5
>> + * running on TCM and also setup virtual address for TCM bank
>> + * remoteproc carveout.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int tcm_mem_map(struct rproc *rproc,
>> + struct rproc_mem_entry *mem)
>> +{
>> + void __iomem *va;
>> +
>> + va = ioremap_wc(mem->dma, mem->len);
>> + if (IS_ERR_OR_NULL(va))
>> + return -ENOMEM;
>> +
>> + /* Update memory entry va */
>> + mem->va = (void *)va;
>> +
>> + /* clear TCMs */
>> + memset_io(va, 0, mem->len);
>> +
>> + /*
>> + * The R5s expect their TCM banks to be at address 0x0 and 0x2000,
>> + * while on the Linux side they are at 0xffexxxxx.
>> + *
>> + * Zero out the high 12 bits of the address. This will give
>> + * expected values for TCM Banks 0A and 0B (0x0 and 0x20000).
>> + */
>> + mem->da &= 0x000fffff;
>> +
>> + /*
>> + * TCM Banks 1A and 1B still have to be translated.
>> + *
>> + * Below handle these two banks' absolute addresses (0xffe90000 and
>> + * 0xffeb0000) and convert to the expected relative addresses
>> + * (0x0 and 0x20000).
>> + */
>> + if (mem->da == 0x90000 || mem->da == 0xB0000)
>> + mem->da -= 0x90000;
>> +
>> + /* if translated TCM bank address is not valid report error */
>> + if (mem->da != 0x0 && mem->da != 0x20000) {
>> + dev_err(&rproc->dev, "invalid TCM address: %x\n", mem->da);
>> + return -EINVAL;
>> + }
>> + return 0;
>> +}
>> +
>> +static int add_tcm_carveout_split_mode(struct rproc *rproc)
>> +{
>> + int i, num_banks, ret;
>> + struct rproc_mem_entry *mem;
>> + u32 pm_domain_id;
>> + phys_addr_t bank_addr;
>> + size_t bank_size;
>> + char *bank_name;
>> + struct device *dev;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
>> +
>> + dev = r5_core->dev;
>> +
>> + /* go through zynqmp banks for r5 node */
>> + num_banks = r5_core->tcm_bank_count;
>> + if (num_banks <= 0) {
>> + dev_err(dev, "need to specify TCM banks\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * Power-on Each 64KB TCM,
>> + * register its address space, map and unmap functions
>> + * and add carveouts accordingly
>> + */
>> + for (i = 0; i < num_banks; i++) {
>> + bank_addr = r5_core->tcm_banks[i].addr;
>> + bank_name = r5_core->tcm_banks[i].bank_name;
>> + bank_size = r5_core->tcm_banks[i].size;
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> +
>> + ret = zynqmp_pm_request_node(pm_domain_id,
>> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
>> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
>> + if (ret < 0) {
>> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
>> + return ret;
>> + }
>> +
>> + dev_dbg(dev, "TCM carveout split mode %s addr=%llx, size=0x%lx",
>> + bank_name, bank_addr, bank_size);
>> +
>> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
>> + bank_size, bank_addr,
>> + tcm_mem_map, tcm_mem_unmap,
>> + bank_name);
>> + if (!mem) {
>> + /* If failed, Turn off all TCM banks turned on before */
>> + do {
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> + ret = zynqmp_pm_release_node(pm_domain_id);
>> + if (ret)
>> + dev_warn(dev,
>> + "fail to release node: 0x%x, %x\n",
>> + pm_domain_id, ret);
>> + } while (i--);
>> + return -ENOMEM;
>> + }
>> +
>> + rproc_add_carveout(rproc, mem);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int add_tcm_carveout_lockstep_mode(struct rproc *rproc)
>> +{
>> + int i, num_banks, ret;
>> + struct rproc_mem_entry *mem;
>> + u32 pm_domain_id;
>> + phys_addr_t bank_addr;
>> + size_t bank_size = 0;
>> + char *bank_name;
>> + struct device *dev;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
>> + dev = r5_core->dev;
>> +
>> + /* Go through zynqmp banks for r5 node */
>> + num_banks = r5_core->tcm_bank_count;
>> + if (num_banks <= 0) {
>> + dev_err(dev, "need to specify TCM banks\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * In lockstep mode, TCM is contiguous memory block
>> + * However, each TCM block still needs to be enabled individually.
>> + * So, Enable each TCM block individually, but add their size
>> + * to create contiguous memory region.
>> + */
>> + bank_addr = r5_core->tcm_banks[0].addr;
>> + bank_name = r5_core->tcm_banks[0].bank_name;
>> +
>> + for (i = 0; i < num_banks; i++) {
>> + bank_size += r5_core->tcm_banks[i].size;
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> +
>> + /* Turn on each TCM bank individually */
>> + ret = zynqmp_pm_request_node(pm_domain_id,
>> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
>> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
>> + if (ret < 0) {
>> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
>> + return ret;
>> + }
>> + }
>> +
>> + dev_dbg(dev, "TCM add carveout lockstep mode %s addr=0x%llx, size=0x%lx",
>> + bank_name, bank_addr, bank_size);
>> +
>> + /* Register TCM address range, TCM map and unmap functions */
>> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
>> + bank_size, bank_addr,
>> + tcm_mem_map, tcm_mem_unmap,
>> + bank_name);
>> + if (!mem) {
>> + /* In case of error, turn off TCM banks */
>> + for (i = 0; i < num_banks; i++) {
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> + ret = zynqmp_pm_release_node(pm_domain_id);
>> + if (ret)
>> + dev_warn(dev,
>> + "fail to release node: 0x%x ret: %x\n",
>> + pm_domain_id, ret);
>> + }
>> + return -ENOMEM;
>> + }
>> +
>> + /* If registration is success, add carveouts */
>> + rproc_add_carveout(rproc, mem);
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * add_tcm_banks()
>> + * @rproc: single R5 core's corresponding rproc instance
>> + *
>> + * Given R5 node in remoteproc instance
>> + * allocate remoteproc carveout for TCM memory
>> + * needed for firmware to be loaded
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int add_tcm_banks(struct rproc *rproc)
>> +{
>> + struct device *dev;
>> + struct platform_device *parent_pdev;
>> + struct zynqmp_r5_cluster *cluster;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
>> + if (!r5_core)
>> + return -EINVAL;
>> +
>> + dev = r5_core->dev;
>> +
>> + cluster = dev_get_drvdata(dev->parent);
>> + if (!cluster) {
>> + dev_err(&parent_pdev->dev, "Invalid driver data\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * In lockstep mode TCM banks are one contiguous memory region of 256Kb
>> + * In split mode, each TCM bank is 64Kb and not contiguous.
>> + * We add memory carveouts accordingly.
>> + */
>> + if (cluster->mode == SPLIT_MODE)
>> + return add_tcm_carveout_split_mode(rproc);
>> + else if (cluster->mode == LOCKSTEP_MODE)
>> + return add_tcm_carveout_lockstep_mode(rproc);
>> +
>> + dev_err(cluster->dev, "invalid cluster mode\n");
>> + return -EINVAL;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_parse_fw()
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @fw: ptr to firmware to be loaded onto r5 core
>> + *
>> + * When loading firmware, ensure the necessary carveouts are in remoteproc
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct firmware *fw)
>> +{
>> + int ret;
>> + struct zynqmp_r5_core *r5_core;
>> + struct device *dev;
>> +
>> + r5_core = rproc->priv;
>> +
>> + dev = r5_core->dev;
>> +
>> + ret = add_tcm_banks(rproc);
>> + if (ret) {
>> + dev_err(dev, "failed to get TCM banks, err %d\n", ret);
>> + return ret;
>> + }
>> +
>> + ret = rproc_elf_load_rsc_table(rproc, fw);
>> + if (ret == -EINVAL) {
>> + /*
>> + * resource table only required for IPC.
>> + * if not present, this is not necessarily an error;
>> + * for example, loading r5 hello world application
>> + * so simply inform user and keep going.
>> + */
>> + dev_info(&rproc->dev, "no resource table found.\n");
>> + ret = 0;
>> + }
>> + return ret;
>> +}
>> +
>> +static struct rproc_ops zynqmp_r5_rproc_ops = {
>> + .start = zynqmp_r5_rproc_start,
>> + .stop = zynqmp_r5_rproc_stop,
>> + .load = rproc_elf_load_segments,
>> + .parse_fw = zynqmp_r5_parse_fw,
>> + .find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
>> + .sanity_check = rproc_elf_sanity_check,
>> + .get_boot_addr = rproc_elf_get_boot_addr,
>> +};
>> +
>> +/**
>> + * zynqmp_r5_add_rproc_core()
>> + * Probes ZynqMP R5 processor device node
>> + * this is called for each individual R5 core
>> + *
>> + * @cdev: Device node of each r5 core
>> + *
>> + * Return: zynqmp_r5_core object for success, error pointer in case of error.
>> + */
>> +static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev)
>> +{
>> + int ret;
>> + struct rproc *r5_rproc;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + /* Set up DMA mask */
>> + ret = dma_set_coherent_mask(cdev, DMA_BIT_MASK(32));
>> + if (ret)
>> + return ERR_PTR(ret);
>> +
>> + /* Allocate remoteproc instance */
>> + r5_rproc = devm_rproc_alloc(cdev, dev_name(cdev),
>> + &zynqmp_r5_rproc_ops,
>> + NULL, sizeof(struct zynqmp_r5_core));
>> + if (!r5_rproc) {
>> + dev_err(cdev, "failed to allocate memory for rproc instance\n");
>> + return ERR_PTR(-ENOMEM);
>> + }
>> +
>> + r5_rproc->auto_boot = false;
>> + r5_core = (struct zynqmp_r5_core *)r5_rproc->priv;
>> + r5_core->dev = cdev;
>> +
>> + /* Add R5 remoteproc core */
>> + ret = devm_rproc_add(cdev, r5_rproc);
>> + if (ret) {
>> + dev_err(cdev, "failed to add r5 remoteproc\n");
>> + return ERR_PTR(ret);
>> + }
>> +
>> + r5_core->rproc = r5_rproc;
>> + return r5_core;
>> +}
>> +
>> +/**
>> + * zynqmp_r5_get_tcm_node()
>> + * Ideally this function should parse tcm node and store information
>> + * in r5_core instance. We will use hardcoded TCM information from
>> + * driver for now in this function.
>> + *
>> + * @cluster: pointer to zynqmp_r5_cluster type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster)
>> +{
>> + int tcm_bank_count, tcm_node;
>> + int i = 0, j;
> Variable @i doesn't need to be initialised.
Agree.
>> + struct zynqmp_r5_core *r5_core;
>> + const struct mem_bank_data *tcm = zynqmp_tcm_banks;
>> + struct device *dev = cluster->dev;
>> +
>> + /*
>> + * ToDo: Use predefined TCM address space values from driver until
>> + * system-dt spec is not final for TCM
>> + */
>> + tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks);
>> +
>> + /* count per core tcm banks */
>> + tcm_bank_count = tcm_bank_count / cluster->core_count;
>> +
>> + /* r5 core 0 will use all of TCM banks in lockstep mode.
>> + * In split mode, r5 core0 will use 128k and r5 core1 will use another
>> + * 128k. Assign TCM banks to each core accordingly
>> + */
> Comment format problem.
My bad. I will fix it.
>> + tcm_node = 0;
>> + for (i = 0; i < cluster->core_count; i++) {
>> + r5_core = cluster->r5_cores[i];
>> + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count,
>> + sizeof(struct mem_bank_data),
>> + GFP_KERNEL);
>> + if (!r5_core->tcm_banks)
>> + return -ENOMEM;
>> +
>> + for (j = 0; j < tcm_bank_count; j++) {
>> + /* Use pre-defined TCM reg values.
>> + * Eventually this should be replaced by values
>> + * parsed from dts.
>> + */
> Same. I commented on that in my previous review of this set.
Yes. I will fix it.
>> + r5_core->tcm_banks[j].addr = tcm[tcm_node].addr;
>> + r5_core->tcm_banks[j].size = tcm[tcm_node].size;
>> + r5_core->tcm_banks[j].pm_domain_id = tcm[tcm_node].pm_domain_id;
>> + r5_core->tcm_banks[j].bank_name = tcm[tcm_node].bank_name;
>> + tcm_node++;
>> + }
>> +
>> + r5_core->tcm_bank_count = tcm_bank_count;
> Why do we need to copy the exact same information from zynqmp_tcm_banks to ->tcm_banks? Why can't ->tcm_banks point to an offset in zynqmp_tcm_banks and ->tcm_bank_count set accordingly?
zynqmp_tcm_banks is global variable. If I use ->tcm_banks point to
offset of zynqmp_tcm_banks and use it in driver, that section of code
won't be thread safe anymore.
I might have to use mutex to protect it. So Instead of pointing it to
global variable, I decided to copy information from global variable to
->tcm_banks.
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/**
>> + * zynqmp_r5_get_mem_region_node()
>> + * parse memory-region property from dt node and add
>> + * memory region carveouts
>> + *
>> + * @r5_core: pointer to zynqmp_r5_core type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core *r5_core)
>> +{
>> + int res_mem_count, i, ret;
>> + struct device *dev;
>> + struct device_node *np, *rmem_np;
>> + struct reserved_mem *rmem;
>> +
>> + dev = r5_core->dev;
>> +
>> + np = r5_core->np;
>> +
>> + res_mem_count = of_property_count_elems_of_size(np, "memory-region",
>> + sizeof(phandle));
>> + if (res_mem_count <= 0) {
>> + dev_warn(dev, "failed to get memory-region property %d\n",
>> + res_mem_count);
>> + return -EINVAL;
>> + }
>> +
>> + for (i = 0; i < res_mem_count; i++) {
>> + rmem_np = of_parse_phandle(np, "memory-region", i);
>> + if (!rmem_np)
>> + return -EINVAL;
>> +
>> + rmem = of_reserved_mem_lookup(rmem_np);
>> + if (!rmem) {
>> + of_node_put(rmem_np);
>> + return -EINVAL;
> What happens to previously allocated regions when there is a failure?
Right, I need to unmap memory and use kfree(rmem) to de-allocate memory.
>> + }
>> +
>> + ret = add_mem_regions_carveout(r5_core, rmem);
>> + if (ret)
>> + dev_warn(dev, "failed to get reserve mem regions %d\n",
>> + ret);
> Same here.
Here as well. I will release reserved mem dev with
of_reserved_mem_device_release API.
One more thing. I moved add_mem_regions_carveout from parse_fw to
zynqmp_r5_get_mem_region_node.
However, I believe I should move it back to parse_fw.
Following test case is failing in when add_mem_regions_carveout is not
available in parse_fw:
load_rpu_fw -> start rpu -> stop rpu -> start rpu again
In above case, during mem regions are not added again.
Is it fine, if I move this add_mem_regions_carveout back to
zynqmp_r5_parse_fw ?
> I am out of time for this set. Please address comments provided up to here and
> we will see about the rest in a future revision.
>
> Thanks,
> Mathieu
>
>> +
>> + of_node_put(rmem_np);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_core_init()
>> + * Create and initialize zynqmp_r5_core type object
>> + *
>> + * @cluster: pointer to zynqmp_r5_cluster type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
>> +{
>> + int ret, i;
>> + struct zynqmp_r5_core *r5_core;
>> + struct device *dev = cluster->dev;
>> +
>> + ret = zynqmp_r5_get_tcm_node(cluster);
>> + if (ret < 0) {
>> + dev_err(dev, "can't get tcm node, err %d\n", ret);
>> + return ret;
>> + }
>> +
>> + for (i = 0; i < cluster->core_count; i++) {
>> + r5_core = cluster->r5_cores[i];
>> +
>> + ret = zynqmp_r5_get_mem_region_node(r5_core);
>> + if (ret)
>> + dev_warn(dev, "memory-region prop failed %d\n", ret);
>> +
>> + /* Initialize r5 cores with power-domains parsed from dts */
>> + ret = of_property_read_u32_index(r5_core->np, "power-domains",
>> + 1, &r5_core->pm_domain_id);
>> + if (ret) {
>> + dev_err(dev, "failed to get power-domains property\n");
>> + return ret;
>> + }
>> +
>> + ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
>> + if (ret) {
>> + dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
>> + cluster->mode, ret);
>> + return ret;
>> + }
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_cluster_init()
>> + * Create and initialize zynqmp_r5_cluster type object
>> + *
>> + * @cluster: pointer to zynqmp_r5_cluster type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
>> +{
>> + struct device *dev = cluster->dev;
>> + struct device_node *dev_node = dev_of_node(dev);
>> + struct device_node *child;
>> + struct platform_device *child_pdev;
>> + int core_count = 0, ret, i;
>> + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
>> + struct zynqmp_r5_core **r5_cores;
>> +
>> + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
>> +
>> + /*
>> + * on success returns 0, if not defined then returns -EINVAL,
>> + * In that case, default is LOCKSTEP mode
>> + */
>> + if (ret != -EINVAL && ret != 0) {
>> + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * For now driver only supports split mode and lockstep mode.
>> + * fail driver probe if either of that is not set in dts
>> + */
>> + if (cluster_mode == SINGLE_CPU_MODE) {
>> + dev_err(dev, "driver does not support single cpu mode\n");
>> + return -EINVAL;
>> + } else if ((cluster_mode != SPLIT_MODE &&
>> + cluster_mode != LOCKSTEP_MODE)) {
>> + dev_err(dev, "Invalid cluster mode\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * Number of cores is decided by number of child nodes of
>> + * r5f subsystem node in dts. If Split mode is used in dts
>> + * 2 child nodes are expected.
>> + * In lockstep mode if two child nodes are available,
>> + * only use first child node and consider it as core0
>> + * and ignore core1 dt node.
>> + */
>> + core_count = of_get_available_child_count(dev_node);
>> + if (core_count <= 0) {
>> + dev_err(dev, "Invalid number of r5 cores %d", core_count);
>> + return -EINVAL;
>> + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
>> + dev_err(dev, "Invalid number of r5 cores for split mode\n");
>> + return -EINVAL;
>> + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
>> + dev_warn(dev, "Only r5 core0 will be used\n");
>> + core_count = 1;
>> + }
>> +
>> + r5_cores = devm_kcalloc(dev, core_count,
>> + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
>> + if (!r5_cores)
>> + return -ENOMEM;
>> +
>> + i = 0;
>> + for_each_available_child_of_node(dev_node, child) {
>> + child_pdev = of_find_device_by_node(child);
>> + if (!child_pdev) {
>> + of_node_put(child);
>> + return -ENODEV;
>> + }
>> +
>> + /* create and add remoteproc instance of type struct rproc */
>> + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
>> + r5_cores[i]->dev = &child_pdev->dev;
>> + if (!r5_cores[i]->dev) {
>> + dev_err(dev, "can't get device for r5 core %d\n", i);
>> + of_node_put(child);
>> + return -ENODEV;
>> + }
>> +
>> + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
>> + if (!r5_cores[i]->np) {
>> + dev_err(dev, "can't get device node for r5 core %d\n", i);
>> + of_node_put(child);
>> + return -ENODEV;
>> + }
>> +
>> + i++;
>> +
>> + /*
>> + * If two child nodes are available in dts in lockstep mode,
>> + * then ignore second child node.
>> + */
>> + if (i == core_count) {
>> + of_node_put(child);
>> + break;
>> + }
>> + of_node_put(child);
>> + }
>> +
>> + cluster->mode = cluster_mode;
>> + cluster->core_count = core_count;
>> + cluster->r5_cores = r5_cores;
>> +
>> + ret = zynqmp_r5_core_init(cluster);
>> + if (ret < 0) {
>> + dev_err(dev, "failed to init r5 core err %d\n", ret);
>> + return ret;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static void zynqmp_r5_cluster_exit(void *data)
>> +{
>> + struct platform_device *pdev = (struct platform_device *)data;
>> +
>> + platform_set_drvdata(pdev, NULL);
>> +
>> + pr_info("Exit r5f subsystem driver\n");
>> +}
>> +
>> +/*
>> + * zynqmp_r5_remoteproc_probe()
>> + *
>> + * @pdev: domain platform device for R5 cluster
>> + *
>> + * called when driver is probed, for each R5 core specified in DT,
>> + * setup as needed to do remoteproc-related operations
>> + *
>> + * Return: 0 for success, negative value for failure.
>> + */
>> +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
>> +{
>> + int ret;
>> + struct zynqmp_r5_cluster *cluster;
>> + struct device *dev = &pdev->dev;
>> +
>> + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
>> + if (!cluster)
>> + return -ENOMEM;
>> +
>> + cluster->dev = dev;
>> +
>> + ret = devm_of_platform_populate(dev);
>> + if (ret) {
>> + dev_err(dev, "failed to populate platform dev %d\n", ret);
>> + return ret;
>> + }
>> +
>> + /* wire in so each core can be cleaned up at driver remove */
>> + platform_set_drvdata(pdev, cluster);
>> +
>> + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
>> + if (ret)
>> + return ret;
>> +
>> + ret = zynqmp_r5_cluster_init(cluster);
>> + if (ret) {
>> + dev_err(dev, "Invalid r5f subsystem device tree\n");
>> + return ret;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/* Match table for OF platform binding */
>> +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
>> + { .compatible = "xlnx,zynqmp-r5fss", },
>> + { /* end of list */ },
>> +};
>> +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
>> +
>> +static struct platform_driver zynqmp_r5_remoteproc_driver = {
>> + .probe = zynqmp_r5_remoteproc_probe,
>> + .driver = {
>> + .name = "zynqmp_r5_remoteproc",
>> + .of_match_table = zynqmp_r5_remoteproc_match,
>> + },
>> +};
>> +module_platform_driver(zynqmp_r5_remoteproc_driver);
>> +
>> +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
>> +MODULE_AUTHOR("Xilinx Inc.");
>> +MODULE_LICENSE("GPL v2");
>> --
>> 2.25.1
>>
On 2/15/22 10:58 AM, Mathieu Poirier wrote:
> On Thu, Feb 10, 2022 at 03:28:24AM -0800, Tanmay Shah wrote:
>> This driver enables r5f dual core Real time Processing Unit subsystem
>> available on Xilinx Zynq Ultrascale MPSoC Platform. RPU subsystem
>> (cluster) can be configured in different modes e.g. split mode in which
>> two r5f cores work independent of each other and lock-step mode in which
>> both r5f cores execute same code clock-for-clock and notify if the
>> result is different.
>>
>> The Xilinx r5 Remoteproc Driver boots the RPU cores via calls to the Xilinx
>> Platform Management Unit that handles the R5 configuration, memory access
>> and R5 lifecycle management. The interface to this manager is done in this
>> driver via zynqmp_pm_* function calls.
>>
>> Signed-off-by: Ben Levinsky<[email protected]>
>> Signed-off-by: Tanmay Shah<[email protected]>
>> ---
>>
>> Changes in v3:
>> - Fix checkpatch script indentation warning
>> - Remove unused variable from xilinx remoteproc driver
>> - use C style comments, i.e /*...*/
>> - Remove redundant debug information which can be derived using /proc/device-tree
>> - Fix multilined comment format
>> - s/"final fot TCM"/"final for TCM"
>> - Function devm_kzalloc() does not return an code on error, just NULL.
>> Remove redundant error check for this function throughout the driver.
>> - Fix RPU mode configuration and add documentation accordingly
>> - Get rid of the indentations to match function documentation style with rest of the driver
>> - Fix memory leak by only using r5_rproc->priv and not replace it with new instance
>> - Use 'i' for the outer loop and 'j' for the inner one as per convention
>> - Remove redundant error and NULL checks throughout the driver
>> - Use devm_kcalloc() when more than one element is required
>> - Add memory-regions carveouts during driver probe instead of parse_fw call
>> This removes redundant copy of reserved_mem object in r5_core structure.
>> - Fix memory leak by using of_node_put()
>> - Fix indentation of tcm_mem_map function args
>> - Remove redundant init of variables
>> - Initialize tcm bank size variable for lockstep mode
>> - Replace u32 with phys_addr_t for variable stroing memory bank address
>> - Add documentation of TCM behavior in lockstep mode
>> - Use dev_get_drvdata instead of platform driver API
>> - Remove info level messages
>> - Fix checkpatch.pl warnings
>> - Add documentation for the Xilinx r5f platform to understand driver design
>>
>> drivers/remoteproc/Kconfig | 12 +
>> drivers/remoteproc/Makefile | 1 +
>> drivers/remoteproc/xlnx_r5_remoteproc.c | 973 ++++++++++++++++++++++++
>> 3 files changed, 986 insertions(+)
>> create mode 100644 drivers/remoteproc/xlnx_r5_remoteproc.c
>>
>> diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
>> index 166019786653..5637a71c0677 100644
>> --- a/drivers/remoteproc/Kconfig
>> +++ b/drivers/remoteproc/Kconfig
>> @@ -352,6 +352,18 @@ config TI_K3_R5_REMOTEPROC
>> It's safe to say N here if you're not interested in utilizing
>> a slave processor.
>>
>> +config XLNX_R5_REMOTEPROC
>> + tristate "Xilinx R5 remoteproc support"
>> + depends on PM && ARCH_ZYNQMP
>> + depends on ZYNQMP_FIRMWARE
>> + select RPMSG_VIRTIO
>> + select ZYNQMP_IPI_MBOX
>> + help
>> + Say y or m here to support Xilinx R5 remote processors via the remote
>> + processor framework.
>> +
>> + It's safe to say N if not interested in using RPU r5f cores.
>> +
>> endif # REMOTEPROC
>>
>> endmenu
>> diff --git a/drivers/remoteproc/Makefile b/drivers/remoteproc/Makefile
>> index 5478c7cb9e07..91314a9b43ce 100644
>> --- a/drivers/remoteproc/Makefile
>> +++ b/drivers/remoteproc/Makefile
>> @@ -38,3 +38,4 @@ obj-$(CONFIG_ST_SLIM_REMOTEPROC) += st_slim_rproc.o
>> obj-$(CONFIG_STM32_RPROC) += stm32_rproc.o
>> obj-$(CONFIG_TI_K3_DSP_REMOTEPROC) += ti_k3_dsp_remoteproc.o
>> obj-$(CONFIG_TI_K3_R5_REMOTEPROC) += ti_k3_r5_remoteproc.o
>> +obj-$(CONFIG_XLNX_R5_REMOTEPROC) += xlnx_r5_remoteproc.o
>> diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c
>> new file mode 100644
>> index 000000000000..ce2ac364b995
>> --- /dev/null
>> +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c
>> @@ -0,0 +1,973 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * ZynqMP R5 Remote Processor driver
>> + *
>> + */
>> +
>> +#include <dt-bindings/power/xlnx-zynqmp-power.h>
>> +#include <linux/firmware/xlnx-zynqmp.h>
>> +#include <linux/interrupt.h>
>> +#include <linux/kernel.h>
>> +#include <linux/list.h>
>> +#include <linux/mailbox_client.h>
>> +#include <linux/mailbox/zynqmp-ipi-message.h>
> No mailboxes are used in this revision and as such the above are not needed.
Sure I will remove them.
>> +#include <linux/module.h>
>> +#include <linux/of_address.h>
>> +#include <linux/of_platform.h>
>> +#include <linux/of_reserved_mem.h>
>> +#include <linux/platform_device.h>
>> +#include <linux/remoteproc.h>
>> +#include <linux/skbuff.h>
>> +#include <linux/sysfs.h>
> As far as I can tell the above two includes are also not needed. Moreover
> dma-mapping.h should be included because of dma_set_coherent_mask() and
> DMA_BIT_MASK().
Ok I will add it and remove redundant headers.
>> +
>> +#include "remoteproc_internal.h"
>> +
>> +/* settings for RPU cluster mode */
>> +enum zynqmp_r5_cluster_mode {
>> + SPLIT_MODE = 0, /* When cores run as separate processor */
>> + LOCKSTEP_MODE = 1, /* cores execute same code in lockstep,clk-for-clk */
>> + SINGLE_CPU_MODE = 2, /* core0 is held in reset and only core1 runs */
>> +};
>> +
>> +/**
>> + * struct mem_bank_data - Memory Bank description
>> + *
>> + * @addr: Start address of memory bank
>> + * @size: Size of Memory bank
>> + * @pm_domain_id: Power-domains id of memory bank for firmware to turn on/off
>> + * @bank_name: name of the bank for remoteproc framework
>> + */
>> +struct mem_bank_data {
>> + phys_addr_t addr;
>> + size_t size;
>> + u32 pm_domain_id;
>> + char *bank_name;
>> +};
>> +
>> +static const struct mem_bank_data zynqmp_tcm_banks[] = {
>> + {0xffe00000UL, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */
>> + {0xffe20000UL, 0x10000UL, PD_R5_0_BTCM, "btcm0"},
>> + {0xffe90000UL, 0x10000UL, PD_R5_1_ATCM, "atcm1"},
>> + {0xffeb0000UL, 0x10000UL, PD_R5_1_BTCM, "btcm1"},
>> +};
>> +
>> +/**
>> + * struct zynqmp_r5_core - ZynqMP R5 core structure
>> + *
>> + * @dev: device of RPU instance
>> + * @np: device node of RPU instance
>> + * @tcm_bank_count: number TCM banks accessible to this RPU
>> + * @tcm_banks: array of each TCM bank data
>> + * @rproc: rproc handle
>> + * @pm_domain_id: RPU CPU power domain id
>> + */
>> +struct zynqmp_r5_core {
>> + struct device *dev;
>> + struct device_node *np;
>> + int tcm_bank_count;
>> + struct mem_bank_data *tcm_banks;
>> + struct rproc *rproc;
>> + u32 pm_domain_id;
>> +};
>> +
>> +/**
>> + * struct zynqmp_r5_cluster - ZynqMP R5 cluster structure
>> + *
>> + * @dev: r5f subsystem cluster device node
>> + * @mode: cluster mode of type zynqmp_r5_cluster_mode
>> + * @core_count: number of r5 cores used for this cluster mode
>> + * @r5_cores: Array of pointers pointing to r5 core
>> + */
>> +struct zynqmp_r5_cluster {
>> + struct device *dev;
>> + enum zynqmp_r5_cluster_mode mode;
>> + int core_count;
>> + struct zynqmp_r5_core **r5_cores;
>> +};
>> +
>> +/*
>> + * zynqmp_r5_set_mode - set RPU operation mode
>> + *
>> + * set RPU operation mode
>> + *
>> + * Return: 0 for success, negative value for failure
>> + */
>> +static int zynqmp_r5_set_mode(struct zynqmp_r5_core *r5_core,
>> + enum zynqmp_r5_cluster_mode rpu_mode)
>> +{
>> + enum rpu_tcm_comb tcm_mode;
>> + int ret, reg_val;
>> +
>> + /*
>> + * SINGLE CPU mode is not supported in driver yet.
>> + * For now we only set lockstep mode or split mode.
>> + */
>> + if (rpu_mode == LOCKSTEP_MODE) {
>> + reg_val = 0;
>> + } else if (rpu_mode == SPLIT_MODE) {
>> + reg_val = 1;
>> + } else if (rpu_mode == SINGLE_CPU_MODE) {
>> + pr_err("RPU single cpu mode is not supported\n");
>> + return -EINVAL;
>> + }
>> +
>> + ret = zynqmp_pm_set_rpu_mode(r5_core->pm_domain_id, reg_val);
>> + if (ret < 0) {
>> + pr_err("failed to set RPU mode\n");
>> + return ret;
>> + }
>> +
>> + /*
>> + * At this point RPU is assumed to be configured only in
>> + * lockstep mode or split mode. Configure TCM accordingly.
>> + */
>> + tcm_mode = (rpu_mode == LOCKSTEP_MODE) ?
>> + PM_RPU_TCM_COMB : PM_RPU_TCM_SPLIT;
>> + ret = zynqmp_pm_set_tcm_config(r5_core->pm_domain_id, tcm_mode);
>> + if (ret < 0)
>> + pr_err("failed to configure TCM\n");
>> +
>> + return ret;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_start
>> + * @rproc: single R5 core's corresponding rproc instance
>> + *
>> + * Start R5 Core from designated boot address.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_start(struct rproc *rproc)
>> +{
>> + struct zynqmp_r5_core *r5_core = rproc->priv;
>> + enum rpu_boot_mem bootmem;
>> + int ret;
>> +
>> + /*
>> + * The exception vector pointers (EVP) refer to the base-address of
>> + * exception vectors (for reset, IRQ, FIQ, etc). The reset-vector
>> + * starts at the base-address and subsequent vectors are on 4-byte
>> + * boundaries.
>> + *
>> + * Exception vectors can start either from 0x0000_0000 (LOVEC) or
>> + * from 0xFFFF_0000 (HIVEC) which is mapped in the OCM (On-Chip Memory)
>> + *
>> + * Usually firmware will put Exception vectors at LOVEC.
>> + *
>> + * It is not recommend that you change the exception vector.
>> + * Changing the EVP to HIVEC will result in increased interrupt latency
>> + * and jitter. Also, if the OCM is secured and the Cortex-R5F processor
>> + * is non-secured, then the Cortex-R5F processor cannot access the
>> + * HIVEC exception vectors in the OCM.
>> + */
>> + bootmem = (rproc->bootaddr >= 0xFFFC0000) ?
>> + PM_RPU_BOOTMEM_HIVEC : PM_RPU_BOOTMEM_LOVEC;
>> +
>> + dev_dbg(r5_core->dev, "RPU boot addr 0x%llx from %s.", rproc->bootaddr,
>> + bootmem == PM_RPU_BOOTMEM_HIVEC ? "OCM" : "TCM");
>> +
>> + ret = zynqmp_pm_request_wake(r5_core->pm_domain_id, 1,
>> + bootmem, ZYNQMP_PM_REQUEST_ACK_NO);
>> + if (ret)
>> + pr_err("failed to start RPU = 0x%x\n", r5_core->pm_domain_id);
>> + return ret;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_stop
>> + * @rproc: single R5 core's corresponding rproc instance
>> + *
>> + * Power down R5 Core.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_stop(struct rproc *rproc)
>> +{
>> + struct zynqmp_r5_core *r5_core = rproc->priv;
>> + int ret;
>> +
>> + ret = zynqmp_pm_force_pwrdwn(r5_core->pm_domain_id,
>> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
>> + if (ret)
>> + pr_err("failed to stop remoteproc RPU %d\n", ret);
>> +
>> + return ret;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_mem_map
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to map
>> + *
>> + * Callback to map va for memory-region's carveout.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_mem_map(struct rproc *rproc,
>> + struct rproc_mem_entry *mem)
>> +{
>> + void __iomem *va;
>> +
>> + va = ioremap_wc(mem->dma, mem->len);
>> + if (IS_ERR_OR_NULL(va))
>> + return -ENOMEM;
>> +
>> + mem->va = (void *)va;
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_mem_unmap
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to unmap
>> + *
>> + * Unmap memory-region carveout
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_mem_unmap(struct rproc *rproc,
>> + struct rproc_mem_entry *mem)
>> +{
>> + iounmap((void __iomem *)mem->va);
>> + return 0;
>> +}
>> +
>> +/*
>> + * add_mem_regions
>> + * @r5_core: single R5 core's corresponding zynqmp_r5_core type instance
>> + * @rmem: reserved mem region parsed from dt node
>> + *
>> + * Construct rproc mem carveouts from carveout provided in
>> + * memory-region property
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int add_mem_regions_carveout(struct zynqmp_r5_core *r5_core,
>> + struct reserved_mem *rmem)
>> +{
>> + struct device *dev;
>> + struct rproc_mem_entry *mem;
>> + struct rproc *rproc;
>> +
>> + rproc = r5_core->rproc;
>> + dev = r5_core->dev;
>> +
>> + /* Register associated reserved memory regions */
>> + mem = rproc_mem_entry_init(dev, NULL,
>> + (dma_addr_t)rmem->base,
>> + rmem->size, rmem->base,
>> + zynqmp_r5_rproc_mem_map,
>> + zynqmp_r5_rproc_mem_unmap,
>> + rmem->name);
>> + if (!mem)
>> + return -ENOMEM;
>> +
>> + rproc_add_carveout(rproc, mem);
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_mem_unmap
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to unmap
>> + *
>> + * Unmap TCM banks when powering down R5 core.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int tcm_mem_unmap(struct rproc *rproc, struct rproc_mem_entry *mem)
>> +{
>> + struct zynqmp_r5_core *r5_core;
>> + int i;
>> + u32 pm_domain_id;
>> +
>> + r5_core = rproc->priv;
>> + if (!r5_core) {
>> + pr_err("r5 core is not available\n");
>> + return -EINVAL;
>> + }
>> +
>> + iounmap((void __iomem *)mem->va);
>> +
>> + for (i = 0; i < r5_core->tcm_bank_count; i++) {
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> + if (zynqmp_pm_release_node(pm_domain_id))
>> + pr_warn("can't turn off TCM bank 0x%x", pm_domain_id);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * tcm_mem_map
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to initialize the va and da fields of
>> + *
>> + * Given TCM bank entry, this callback will set device address for R5
>> + * running on TCM and also setup virtual address for TCM bank
>> + * remoteproc carveout.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int tcm_mem_map(struct rproc *rproc,
>> + struct rproc_mem_entry *mem)
>> +{
>> + void __iomem *va;
>> +
>> + va = ioremap_wc(mem->dma, mem->len);
>> + if (IS_ERR_OR_NULL(va))
>> + return -ENOMEM;
>> +
>> + /* Update memory entry va */
>> + mem->va = (void *)va;
>> +
>> + /* clear TCMs */
>> + memset_io(va, 0, mem->len);
>> +
>> + /*
>> + * The R5s expect their TCM banks to be at address 0x0 and 0x2000,
>> + * while on the Linux side they are at 0xffexxxxx.
>> + *
>> + * Zero out the high 12 bits of the address. This will give
>> + * expected values for TCM Banks 0A and 0B (0x0 and 0x20000).
>> + */
>> + mem->da &= 0x000fffff;
>> +
>> + /*
>> + * TCM Banks 1A and 1B still have to be translated.
>> + *
>> + * Below handle these two banks' absolute addresses (0xffe90000 and
>> + * 0xffeb0000) and convert to the expected relative addresses
>> + * (0x0 and 0x20000).
>> + */
>> + if (mem->da == 0x90000 || mem->da == 0xB0000)
>> + mem->da -= 0x90000;
>> +
>> + /* if translated TCM bank address is not valid report error */
>> + if (mem->da != 0x0 && mem->da != 0x20000) {
>> + dev_err(&rproc->dev, "invalid TCM address: %x\n", mem->da);
>> + return -EINVAL;
>> + }
>> + return 0;
>> +}
>> +
>> +static int add_tcm_carveout_split_mode(struct rproc *rproc)
>> +{
>> + int i, num_banks, ret;
>> + struct rproc_mem_entry *mem;
>> + u32 pm_domain_id;
>> + phys_addr_t bank_addr;
>> + size_t bank_size;
>> + char *bank_name;
>> + struct device *dev;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
>> +
>> + dev = r5_core->dev;
>> +
>> + /* go through zynqmp banks for r5 node */
>> + num_banks = r5_core->tcm_bank_count;
>> + if (num_banks <= 0) {
>> + dev_err(dev, "need to specify TCM banks\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * Power-on Each 64KB TCM,
>> + * register its address space, map and unmap functions
>> + * and add carveouts accordingly
>> + */
>> + for (i = 0; i < num_banks; i++) {
>> + bank_addr = r5_core->tcm_banks[i].addr;
>> + bank_name = r5_core->tcm_banks[i].bank_name;
>> + bank_size = r5_core->tcm_banks[i].size;
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> +
>> + ret = zynqmp_pm_request_node(pm_domain_id,
>> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
>> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
>> + if (ret < 0) {
>> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
>> + return ret;
>> + }
>> +
>> + dev_dbg(dev, "TCM carveout split mode %s addr=%llx, size=0x%lx",
>> + bank_name, bank_addr, bank_size);
>> +
>> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
>> + bank_size, bank_addr,
>> + tcm_mem_map, tcm_mem_unmap,
>> + bank_name);
>> + if (!mem) {
>> + /* If failed, Turn off all TCM banks turned on before */
>> + do {
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> + ret = zynqmp_pm_release_node(pm_domain_id);
>> + if (ret)
>> + dev_warn(dev,
>> + "fail to release node: 0x%x, %x\n",
>> + pm_domain_id, ret);
>> + } while (i--);
>> + return -ENOMEM;
>> + }
>> +
>> + rproc_add_carveout(rproc, mem);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int add_tcm_carveout_lockstep_mode(struct rproc *rproc)
>> +{
>> + int i, num_banks, ret;
>> + struct rproc_mem_entry *mem;
>> + u32 pm_domain_id;
>> + phys_addr_t bank_addr;
>> + size_t bank_size = 0;
>> + char *bank_name;
>> + struct device *dev;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
>> + dev = r5_core->dev;
>> +
>> + /* Go through zynqmp banks for r5 node */
>> + num_banks = r5_core->tcm_bank_count;
>> + if (num_banks <= 0) {
>> + dev_err(dev, "need to specify TCM banks\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * In lockstep mode, TCM is contiguous memory block
>> + * However, each TCM block still needs to be enabled individually.
>> + * So, Enable each TCM block individually, but add their size
>> + * to create contiguous memory region.
>> + */
>> + bank_addr = r5_core->tcm_banks[0].addr;
>> + bank_name = r5_core->tcm_banks[0].bank_name;
>> +
>> + for (i = 0; i < num_banks; i++) {
>> + bank_size += r5_core->tcm_banks[i].size;
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> +
>> + /* Turn on each TCM bank individually */
>> + ret = zynqmp_pm_request_node(pm_domain_id,
>> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
>> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
>> + if (ret < 0) {
>> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
>> + return ret;
>> + }
>> + }
>> +
>> + dev_dbg(dev, "TCM add carveout lockstep mode %s addr=0x%llx, size=0x%lx",
>> + bank_name, bank_addr, bank_size);
>> +
>> + /* Register TCM address range, TCM map and unmap functions */
>> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
>> + bank_size, bank_addr,
>> + tcm_mem_map, tcm_mem_unmap,
>> + bank_name);
>> + if (!mem) {
>> + /* In case of error, turn off TCM banks */
>> + for (i = 0; i < num_banks; i++) {
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> + ret = zynqmp_pm_release_node(pm_domain_id);
>> + if (ret)
>> + dev_warn(dev,
>> + "fail to release node: 0x%x ret: %x\n",
>> + pm_domain_id, ret);
>> + }
>> + return -ENOMEM;
>> + }
>> +
>> + /* If registration is success, add carveouts */
>> + rproc_add_carveout(rproc, mem);
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * add_tcm_banks()
>> + * @rproc: single R5 core's corresponding rproc instance
>> + *
>> + * Given R5 node in remoteproc instance
>> + * allocate remoteproc carveout for TCM memory
>> + * needed for firmware to be loaded
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int add_tcm_banks(struct rproc *rproc)
>> +{
>> + struct device *dev;
>> + struct platform_device *parent_pdev;
>> + struct zynqmp_r5_cluster *cluster;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
>> + if (!r5_core)
>> + return -EINVAL;
>> +
>> + dev = r5_core->dev;
>> +
>> + cluster = dev_get_drvdata(dev->parent);
>> + if (!cluster) {
>> + dev_err(&parent_pdev->dev, "Invalid driver data\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * In lockstep mode TCM banks are one contiguous memory region of 256Kb
>> + * In split mode, each TCM bank is 64Kb and not contiguous.
>> + * We add memory carveouts accordingly.
>> + */
>> + if (cluster->mode == SPLIT_MODE)
>> + return add_tcm_carveout_split_mode(rproc);
>> + else if (cluster->mode == LOCKSTEP_MODE)
>> + return add_tcm_carveout_lockstep_mode(rproc);
>> +
>> + dev_err(cluster->dev, "invalid cluster mode\n");
>> + return -EINVAL;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_parse_fw()
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @fw: ptr to firmware to be loaded onto r5 core
>> + *
>> + * When loading firmware, ensure the necessary carveouts are in remoteproc
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct firmware *fw)
>> +{
>> + int ret;
>> + struct zynqmp_r5_core *r5_core;
>> + struct device *dev;
>> +
>> + r5_core = rproc->priv;
>> +
>> + dev = r5_core->dev;
>> +
>> + ret = add_tcm_banks(rproc);
>> + if (ret) {
>> + dev_err(dev, "failed to get TCM banks, err %d\n", ret);
>> + return ret;
>> + }
>> +
>> + ret = rproc_elf_load_rsc_table(rproc, fw);
>> + if (ret == -EINVAL) {
>> + /*
>> + * resource table only required for IPC.
>> + * if not present, this is not necessarily an error;
>> + * for example, loading r5 hello world application
>> + * so simply inform user and keep going.
>> + */
>> + dev_info(&rproc->dev, "no resource table found.\n");
>> + ret = 0;
>> + }
>> + return ret;
>> +}
>> +
>> +static struct rproc_ops zynqmp_r5_rproc_ops = {
>> + .start = zynqmp_r5_rproc_start,
>> + .stop = zynqmp_r5_rproc_stop,
>> + .load = rproc_elf_load_segments,
>> + .parse_fw = zynqmp_r5_parse_fw,
>> + .find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
>> + .sanity_check = rproc_elf_sanity_check,
>> + .get_boot_addr = rproc_elf_get_boot_addr,
>> +};
>> +
>> +/**
>> + * zynqmp_r5_add_rproc_core()
>> + * Probes ZynqMP R5 processor device node
>> + * this is called for each individual R5 core
>> + *
>> + * @cdev: Device node of each r5 core
>> + *
>> + * Return: zynqmp_r5_core object for success, error pointer in case of error.
>> + */
>> +static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev)
>> +{
>> + int ret;
>> + struct rproc *r5_rproc;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + /* Set up DMA mask */
>> + ret = dma_set_coherent_mask(cdev, DMA_BIT_MASK(32));
>> + if (ret)
>> + return ERR_PTR(ret);
>> +
>> + /* Allocate remoteproc instance */
>> + r5_rproc = devm_rproc_alloc(cdev, dev_name(cdev),
>> + &zynqmp_r5_rproc_ops,
>> + NULL, sizeof(struct zynqmp_r5_core));
>> + if (!r5_rproc) {
>> + dev_err(cdev, "failed to allocate memory for rproc instance\n");
>> + return ERR_PTR(-ENOMEM);
>> + }
>> +
>> + r5_rproc->auto_boot = false;
>> + r5_core = (struct zynqmp_r5_core *)r5_rproc->priv;
>> + r5_core->dev = cdev;
>> +
>> + /* Add R5 remoteproc core */
>> + ret = devm_rproc_add(cdev, r5_rproc);
>> + if (ret) {
>> + dev_err(cdev, "failed to add r5 remoteproc\n");
>> + return ERR_PTR(ret);
>> + }
>> +
>> + r5_core->rproc = r5_rproc;
>> + return r5_core;
>> +}
>> +
>> +/**
>> + * zynqmp_r5_get_tcm_node()
>> + * Ideally this function should parse tcm node and store information
>> + * in r5_core instance. We will use hardcoded TCM information from
>> + * driver for now in this function.
>> + *
>> + * @cluster: pointer to zynqmp_r5_cluster type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster)
>> +{
>> + int tcm_bank_count, tcm_node;
>> + int i = 0, j;
>> + struct zynqmp_r5_core *r5_core;
>> + const struct mem_bank_data *tcm = zynqmp_tcm_banks;
>> + struct device *dev = cluster->dev;
>> +
>> + /*
>> + * ToDo: Use predefined TCM address space values from driver until
>> + * system-dt spec is not final for TCM
>> + */
>> + tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks);
>> +
>> + /* count per core tcm banks */
>> + tcm_bank_count = tcm_bank_count / cluster->core_count;
>> +
>> + /* r5 core 0 will use all of TCM banks in lockstep mode.
>> + * In split mode, r5 core0 will use 128k and r5 core1 will use another
>> + * 128k. Assign TCM banks to each core accordingly
>> + */
>> + tcm_node = 0;
>> + for (i = 0; i < cluster->core_count; i++) {
>> + r5_core = cluster->r5_cores[i];
>> + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count,
>> + sizeof(struct mem_bank_data),
>> + GFP_KERNEL);
>> + if (!r5_core->tcm_banks)
>> + return -ENOMEM;
>> +
>> + for (j = 0; j < tcm_bank_count; j++) {
>> + /* Use pre-defined TCM reg values.
>> + * Eventually this should be replaced by values
>> + * parsed from dts.
>> + */
>> + r5_core->tcm_banks[j].addr = tcm[tcm_node].addr;
>> + r5_core->tcm_banks[j].size = tcm[tcm_node].size;
>> + r5_core->tcm_banks[j].pm_domain_id = tcm[tcm_node].pm_domain_id;
>> + r5_core->tcm_banks[j].bank_name = tcm[tcm_node].bank_name;
>> + tcm_node++;
>> + }
>> +
>> + r5_core->tcm_bank_count = tcm_bank_count;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/**
>> + * zynqmp_r5_get_mem_region_node()
>> + * parse memory-region property from dt node and add
>> + * memory region carveouts
>> + *
>> + * @r5_core: pointer to zynqmp_r5_core type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core *r5_core)
>> +{
>> + int res_mem_count, i, ret;
>> + struct device *dev;
>> + struct device_node *np, *rmem_np;
>> + struct reserved_mem *rmem;
>> +
>> + dev = r5_core->dev;
>> +
>> + np = r5_core->np;
>> +
>> + res_mem_count = of_property_count_elems_of_size(np, "memory-region",
>> + sizeof(phandle));
>> + if (res_mem_count <= 0) {
>> + dev_warn(dev, "failed to get memory-region property %d\n",
>> + res_mem_count);
>> + return -EINVAL;
>> + }
>> +
>> + for (i = 0; i < res_mem_count; i++) {
>> + rmem_np = of_parse_phandle(np, "memory-region", i);
>> + if (!rmem_np)
>> + return -EINVAL;
>> +
>> + rmem = of_reserved_mem_lookup(rmem_np);
>> + if (!rmem) {
>> + of_node_put(rmem_np);
>> + return -EINVAL;
>> + }
>> +
>> + ret = add_mem_regions_carveout(r5_core, rmem);
>> + if (ret)
>> + dev_warn(dev, "failed to get reserve mem regions %d\n",
>> + ret);
>> +
>> + of_node_put(rmem_np);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_core_init()
>> + * Create and initialize zynqmp_r5_core type object
>> + *
>> + * @cluster: pointer to zynqmp_r5_cluster type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
>> +{
>> + int ret, i;
>> + struct zynqmp_r5_core *r5_core;
>> + struct device *dev = cluster->dev;
>> +
>> + ret = zynqmp_r5_get_tcm_node(cluster);
>> + if (ret < 0) {
>> + dev_err(dev, "can't get tcm node, err %d\n", ret);
>> + return ret;
>> + }
>> +
>> + for (i = 0; i < cluster->core_count; i++) {
>> + r5_core = cluster->r5_cores[i];
>> +
>> + ret = zynqmp_r5_get_mem_region_node(r5_core);
>> + if (ret)
>> + dev_warn(dev, "memory-region prop failed %d\n", ret);
>> +
>> + /* Initialize r5 cores with power-domains parsed from dts */
>> + ret = of_property_read_u32_index(r5_core->np, "power-domains",
>> + 1, &r5_core->pm_domain_id);
>> + if (ret) {
>> + dev_err(dev, "failed to get power-domains property\n");
>> + return ret;
>> + }
>> +
>> + ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
>> + if (ret) {
>> + dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
>> + cluster->mode, ret);
>> + return ret;
>> + }
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_cluster_init()
>> + * Create and initialize zynqmp_r5_cluster type object
>> + *
>> + * @cluster: pointer to zynqmp_r5_cluster type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
>> +{
>> + struct device *dev = cluster->dev;
>> + struct device_node *dev_node = dev_of_node(dev);
>> + struct device_node *child;
>> + struct platform_device *child_pdev;
>> + int core_count = 0, ret, i;
>> + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
>> + struct zynqmp_r5_core **r5_cores;
>> +
>> + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
>> +
>> + /*
>> + * on success returns 0, if not defined then returns -EINVAL,
>> + * In that case, default is LOCKSTEP mode
>> + */
>> + if (ret != -EINVAL && ret != 0) {
>> + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * For now driver only supports split mode and lockstep mode.
>> + * fail driver probe if either of that is not set in dts
>> + */
>> + if (cluster_mode == SINGLE_CPU_MODE) {
>> + dev_err(dev, "driver does not support single cpu mode\n");
>> + return -EINVAL;
>> + } else if ((cluster_mode != SPLIT_MODE &&
>> + cluster_mode != LOCKSTEP_MODE)) {
>> + dev_err(dev, "Invalid cluster mode\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * Number of cores is decided by number of child nodes of
>> + * r5f subsystem node in dts. If Split mode is used in dts
>> + * 2 child nodes are expected.
>> + * In lockstep mode if two child nodes are available,
>> + * only use first child node and consider it as core0
>> + * and ignore core1 dt node.
>> + */
>> + core_count = of_get_available_child_count(dev_node);
>> + if (core_count <= 0) {
>> + dev_err(dev, "Invalid number of r5 cores %d", core_count);
>> + return -EINVAL;
>> + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
>> + dev_err(dev, "Invalid number of r5 cores for split mode\n");
>> + return -EINVAL;
>> + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
>> + dev_warn(dev, "Only r5 core0 will be used\n");
>> + core_count = 1;
>> + }
>> +
>> + r5_cores = devm_kcalloc(dev, core_count,
>> + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
>> + if (!r5_cores)
>> + return -ENOMEM;
>> +
>> + i = 0;
>> + for_each_available_child_of_node(dev_node, child) {
>> + child_pdev = of_find_device_by_node(child);
>> + if (!child_pdev) {
>> + of_node_put(child);
>> + return -ENODEV;
>> + }
>> +
>> + /* create and add remoteproc instance of type struct rproc */
>> + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
>> + r5_cores[i]->dev = &child_pdev->dev;
>> + if (!r5_cores[i]->dev) {
>> + dev_err(dev, "can't get device for r5 core %d\n", i);
>> + of_node_put(child);
>> + return -ENODEV;
>> + }
>> +
>> + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
>> + if (!r5_cores[i]->np) {
>> + dev_err(dev, "can't get device node for r5 core %d\n", i);
>> + of_node_put(child);
>> + return -ENODEV;
>> + }
>> +
>> + i++;
>> +
>> + /*
>> + * If two child nodes are available in dts in lockstep mode,
>> + * then ignore second child node.
>> + */
>> + if (i == core_count) {
>> + of_node_put(child);
>> + break;
>> + }
>> + of_node_put(child);
>> + }
>> +
>> + cluster->mode = cluster_mode;
>> + cluster->core_count = core_count;
>> + cluster->r5_cores = r5_cores;
>> +
>> + ret = zynqmp_r5_core_init(cluster);
>> + if (ret < 0) {
>> + dev_err(dev, "failed to init r5 core err %d\n", ret);
>> + return ret;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static void zynqmp_r5_cluster_exit(void *data)
>> +{
>> + struct platform_device *pdev = (struct platform_device *)data;
>> +
>> + platform_set_drvdata(pdev, NULL);
>> +
>> + pr_info("Exit r5f subsystem driver\n");
>> +}
>> +
>> +/*
>> + * zynqmp_r5_remoteproc_probe()
>> + *
>> + * @pdev: domain platform device for R5 cluster
>> + *
>> + * called when driver is probed, for each R5 core specified in DT,
>> + * setup as needed to do remoteproc-related operations
>> + *
>> + * Return: 0 for success, negative value for failure.
>> + */
>> +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
>> +{
>> + int ret;
>> + struct zynqmp_r5_cluster *cluster;
>> + struct device *dev = &pdev->dev;
>> +
>> + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
>> + if (!cluster)
>> + return -ENOMEM;
>> +
>> + cluster->dev = dev;
>> +
>> + ret = devm_of_platform_populate(dev);
>> + if (ret) {
>> + dev_err(dev, "failed to populate platform dev %d\n", ret);
>> + return ret;
> Please use dev_err_probe()
Ok.
>> + }
>> +
>> + /* wire in so each core can be cleaned up at driver remove */
>> + platform_set_drvdata(pdev, cluster);
>> +
>> + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
>> + if (ret)
>> + return ret;
> Not sure this is needed. The TI guys used that mechanic to release acquired
> resources but merely setting driver data to NULL doesn't mandate all this.
If I use zynqmp_r5_cluster_exit to free devices of r5_core using
put_device( ) API then can I keep this?
More on this in next email.
>> +
>> + ret = zynqmp_r5_cluster_init(cluster);
>> + if (ret) {
>> + dev_err(dev, "Invalid r5f subsystem device tree\n");
>> + return ret;
> Here to.
>
> More comments to come tomorrow.
>
> Thanks,
> Mathieu
>
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/* Match table for OF platform binding */
>> +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
>> + { .compatible = "xlnx,zynqmp-r5fss", },
>> + { /* end of list */ },
>> +};
>> +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
>> +
>> +static struct platform_driver zynqmp_r5_remoteproc_driver = {
>> + .probe = zynqmp_r5_remoteproc_probe,
>> + .driver = {
>> + .name = "zynqmp_r5_remoteproc",
>> + .of_match_table = zynqmp_r5_remoteproc_match,
>> + },
>> +};
>> +module_platform_driver(zynqmp_r5_remoteproc_driver);
>> +
>> +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
>> +MODULE_AUTHOR("Xilinx Inc.");
>> +MODULE_LICENSE("GPL v2");
>> --
>> 2.25.1
>>
On 2/16/22 10:26 AM, Mathieu Poirier wrote:
> On Thu, Feb 10, 2022 at 03:28:24AM -0800, Tanmay Shah wrote:
>> This driver enables r5f dual core Real time Processing Unit subsystem
>> available on Xilinx Zynq Ultrascale MPSoC Platform. RPU subsystem
>> (cluster) can be configured in different modes e.g. split mode in which
>> two r5f cores work independent of each other and lock-step mode in which
>> both r5f cores execute same code clock-for-clock and notify if the
>> result is different.
>>
>> The Xilinx r5 Remoteproc Driver boots the RPU cores via calls to the Xilinx
>> Platform Management Unit that handles the R5 configuration, memory access
>> and R5 lifecycle management. The interface to this manager is done in this
>> driver via zynqmp_pm_* function calls.
>>
>> Signed-off-by: Ben Levinsky<[email protected]>
>> Signed-off-by: Tanmay Shah<[email protected]>
>> ---
>>
>> Changes in v3:
>> - Fix checkpatch script indentation warning
>> - Remove unused variable from xilinx remoteproc driver
>> - use C style comments, i.e /*...*/
>> - Remove redundant debug information which can be derived using /proc/device-tree
>> - Fix multilined comment format
>> - s/"final fot TCM"/"final for TCM"
>> - Function devm_kzalloc() does not return an code on error, just NULL.
>> Remove redundant error check for this function throughout the driver.
>> - Fix RPU mode configuration and add documentation accordingly
>> - Get rid of the indentations to match function documentation style with rest of the driver
>> - Fix memory leak by only using r5_rproc->priv and not replace it with new instance
>> - Use 'i' for the outer loop and 'j' for the inner one as per convention
>> - Remove redundant error and NULL checks throughout the driver
>> - Use devm_kcalloc() when more than one element is required
>> - Add memory-regions carveouts during driver probe instead of parse_fw call
>> This removes redundant copy of reserved_mem object in r5_core structure.
>> - Fix memory leak by using of_node_put()
>> - Fix indentation of tcm_mem_map function args
>> - Remove redundant init of variables
>> - Initialize tcm bank size variable for lockstep mode
>> - Replace u32 with phys_addr_t for variable stroing memory bank address
>> - Add documentation of TCM behavior in lockstep mode
>> - Use dev_get_drvdata instead of platform driver API
>> - Remove info level messages
>> - Fix checkpatch.pl warnings
>> - Add documentation for the Xilinx r5f platform to understand driver design
>>
>> drivers/remoteproc/Kconfig | 12 +
>> drivers/remoteproc/Makefile | 1 +
>> drivers/remoteproc/xlnx_r5_remoteproc.c | 973 ++++++++++++++++++++++++
>> 3 files changed, 986 insertions(+)
>> create mode 100644 drivers/remoteproc/xlnx_r5_remoteproc.c
>>
>> diff --git a/drivers/remoteproc/Kconfig b/drivers/remoteproc/Kconfig
>> index 166019786653..5637a71c0677 100644
>> --- a/drivers/remoteproc/Kconfig
>> +++ b/drivers/remoteproc/Kconfig
>> @@ -352,6 +352,18 @@ config TI_K3_R5_REMOTEPROC
>> It's safe to say N here if you're not interested in utilizing
>> a slave processor.
>>
>> +config XLNX_R5_REMOTEPROC
>> + tristate "Xilinx R5 remoteproc support"
>> + depends on PM && ARCH_ZYNQMP
>> + depends on ZYNQMP_FIRMWARE
>> + select RPMSG_VIRTIO
>> + select ZYNQMP_IPI_MBOX
>> + help
>> + Say y or m here to support Xilinx R5 remote processors via the remote
>> + processor framework.
>> +
>> + It's safe to say N if not interested in using RPU r5f cores.
>> +
>> endif # REMOTEPROC
>>
>> endmenu
>> diff --git a/drivers/remoteproc/Makefile b/drivers/remoteproc/Makefile
>> index 5478c7cb9e07..91314a9b43ce 100644
>> --- a/drivers/remoteproc/Makefile
>> +++ b/drivers/remoteproc/Makefile
>> @@ -38,3 +38,4 @@ obj-$(CONFIG_ST_SLIM_REMOTEPROC) += st_slim_rproc.o
>> obj-$(CONFIG_STM32_RPROC) += stm32_rproc.o
>> obj-$(CONFIG_TI_K3_DSP_REMOTEPROC) += ti_k3_dsp_remoteproc.o
>> obj-$(CONFIG_TI_K3_R5_REMOTEPROC) += ti_k3_r5_remoteproc.o
>> +obj-$(CONFIG_XLNX_R5_REMOTEPROC) += xlnx_r5_remoteproc.o
>> diff --git a/drivers/remoteproc/xlnx_r5_remoteproc.c b/drivers/remoteproc/xlnx_r5_remoteproc.c
>> new file mode 100644
>> index 000000000000..ce2ac364b995
>> --- /dev/null
>> +++ b/drivers/remoteproc/xlnx_r5_remoteproc.c
>> @@ -0,0 +1,973 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * ZynqMP R5 Remote Processor driver
>> + *
>> + */
>> +
>> +#include <dt-bindings/power/xlnx-zynqmp-power.h>
>> +#include <linux/firmware/xlnx-zynqmp.h>
>> +#include <linux/interrupt.h>
>> +#include <linux/kernel.h>
>> +#include <linux/list.h>
>> +#include <linux/mailbox_client.h>
>> +#include <linux/mailbox/zynqmp-ipi-message.h>
>> +#include <linux/module.h>
>> +#include <linux/of_address.h>
>> +#include <linux/of_platform.h>
>> +#include <linux/of_reserved_mem.h>
>> +#include <linux/platform_device.h>
>> +#include <linux/remoteproc.h>
>> +#include <linux/skbuff.h>
>> +#include <linux/sysfs.h>
>> +
>> +#include "remoteproc_internal.h"
>> +
>> +/* settings for RPU cluster mode */
>> +enum zynqmp_r5_cluster_mode {
>> + SPLIT_MODE = 0, /* When cores run as separate processor */
>> + LOCKSTEP_MODE = 1, /* cores execute same code in lockstep,clk-for-clk */
>> + SINGLE_CPU_MODE = 2, /* core0 is held in reset and only core1 runs */
>> +};
>> +
>> +/**
>> + * struct mem_bank_data - Memory Bank description
>> + *
>> + * @addr: Start address of memory bank
>> + * @size: Size of Memory bank
>> + * @pm_domain_id: Power-domains id of memory bank for firmware to turn on/off
>> + * @bank_name: name of the bank for remoteproc framework
>> + */
>> +struct mem_bank_data {
>> + phys_addr_t addr;
>> + size_t size;
>> + u32 pm_domain_id;
>> + char *bank_name;
>> +};
>> +
>> +static const struct mem_bank_data zynqmp_tcm_banks[] = {
>> + {0xffe00000UL, 0x10000UL, PD_R5_0_ATCM, "atcm0"}, /* TCM 64KB each */
>> + {0xffe20000UL, 0x10000UL, PD_R5_0_BTCM, "btcm0"},
>> + {0xffe90000UL, 0x10000UL, PD_R5_1_ATCM, "atcm1"},
>> + {0xffeb0000UL, 0x10000UL, PD_R5_1_BTCM, "btcm1"},
>> +};
>> +
>> +/**
>> + * struct zynqmp_r5_core - ZynqMP R5 core structure
>> + *
>> + * @dev: device of RPU instance
>> + * @np: device node of RPU instance
>> + * @tcm_bank_count: number TCM banks accessible to this RPU
>> + * @tcm_banks: array of each TCM bank data
>> + * @rproc: rproc handle
>> + * @pm_domain_id: RPU CPU power domain id
>> + */
>> +struct zynqmp_r5_core {
>> + struct device *dev;
>> + struct device_node *np;
>> + int tcm_bank_count;
>> + struct mem_bank_data *tcm_banks;
>> + struct rproc *rproc;
>> + u32 pm_domain_id;
>> +};
>> +
>> +/**
>> + * struct zynqmp_r5_cluster - ZynqMP R5 cluster structure
>> + *
>> + * @dev: r5f subsystem cluster device node
>> + * @mode: cluster mode of type zynqmp_r5_cluster_mode
>> + * @core_count: number of r5 cores used for this cluster mode
>> + * @r5_cores: Array of pointers pointing to r5 core
>> + */
>> +struct zynqmp_r5_cluster {
>> + struct device *dev;
>> + enum zynqmp_r5_cluster_mode mode;
>> + int core_count;
>> + struct zynqmp_r5_core **r5_cores;
>> +};
>> +
>> +/*
>> + * zynqmp_r5_set_mode - set RPU operation mode
>> + *
>> + * set RPU operation mode
>> + *
>> + * Return: 0 for success, negative value for failure
>> + */
>> +static int zynqmp_r5_set_mode(struct zynqmp_r5_core *r5_core,
>> + enum zynqmp_r5_cluster_mode rpu_mode)
>> +{
>> + enum rpu_tcm_comb tcm_mode;
>> + int ret, reg_val;
>> +
>> + /*
>> + * SINGLE CPU mode is not supported in driver yet.
>> + * For now we only set lockstep mode or split mode.
>> + */
>> + if (rpu_mode == LOCKSTEP_MODE) {
>> + reg_val = 0;
>> + } else if (rpu_mode == SPLIT_MODE) {
>> + reg_val = 1;
>> + } else if (rpu_mode == SINGLE_CPU_MODE) {
>> + pr_err("RPU single cpu mode is not supported\n");
>> + return -EINVAL;
>> + }
>> +
>> + ret = zynqmp_pm_set_rpu_mode(r5_core->pm_domain_id, reg_val);
>> + if (ret < 0) {
>> + pr_err("failed to set RPU mode\n");
>> + return ret;
>> + }
>> +
>> + /*
>> + * At this point RPU is assumed to be configured only in
>> + * lockstep mode or split mode. Configure TCM accordingly.
>> + */
>> + tcm_mode = (rpu_mode == LOCKSTEP_MODE) ?
>> + PM_RPU_TCM_COMB : PM_RPU_TCM_SPLIT;
>> + ret = zynqmp_pm_set_tcm_config(r5_core->pm_domain_id, tcm_mode);
>> + if (ret < 0)
>> + pr_err("failed to configure TCM\n");
>> +
>> + return ret;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_start
>> + * @rproc: single R5 core's corresponding rproc instance
>> + *
>> + * Start R5 Core from designated boot address.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_start(struct rproc *rproc)
>> +{
>> + struct zynqmp_r5_core *r5_core = rproc->priv;
>> + enum rpu_boot_mem bootmem;
>> + int ret;
>> +
>> + /*
>> + * The exception vector pointers (EVP) refer to the base-address of
>> + * exception vectors (for reset, IRQ, FIQ, etc). The reset-vector
>> + * starts at the base-address and subsequent vectors are on 4-byte
>> + * boundaries.
>> + *
>> + * Exception vectors can start either from 0x0000_0000 (LOVEC) or
>> + * from 0xFFFF_0000 (HIVEC) which is mapped in the OCM (On-Chip Memory)
>> + *
>> + * Usually firmware will put Exception vectors at LOVEC.
>> + *
>> + * It is not recommend that you change the exception vector.
>> + * Changing the EVP to HIVEC will result in increased interrupt latency
>> + * and jitter. Also, if the OCM is secured and the Cortex-R5F processor
>> + * is non-secured, then the Cortex-R5F processor cannot access the
>> + * HIVEC exception vectors in the OCM.
>> + */
>> + bootmem = (rproc->bootaddr >= 0xFFFC0000) ?
>> + PM_RPU_BOOTMEM_HIVEC : PM_RPU_BOOTMEM_LOVEC;
>> +
>> + dev_dbg(r5_core->dev, "RPU boot addr 0x%llx from %s.", rproc->bootaddr,
>> + bootmem == PM_RPU_BOOTMEM_HIVEC ? "OCM" : "TCM");
>> +
>> + ret = zynqmp_pm_request_wake(r5_core->pm_domain_id, 1,
>> + bootmem, ZYNQMP_PM_REQUEST_ACK_NO);
>> + if (ret)
>> + pr_err("failed to start RPU = 0x%x\n", r5_core->pm_domain_id);
>> + return ret;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_stop
>> + * @rproc: single R5 core's corresponding rproc instance
>> + *
>> + * Power down R5 Core.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_stop(struct rproc *rproc)
>> +{
>> + struct zynqmp_r5_core *r5_core = rproc->priv;
>> + int ret;
>> +
>> + ret = zynqmp_pm_force_pwrdwn(r5_core->pm_domain_id,
>> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
>> + if (ret)
>> + pr_err("failed to stop remoteproc RPU %d\n", ret);
>> +
>> + return ret;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_mem_map
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to map
>> + *
>> + * Callback to map va for memory-region's carveout.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_mem_map(struct rproc *rproc,
>> + struct rproc_mem_entry *mem)
>> +{
>> + void __iomem *va;
>> +
>> + va = ioremap_wc(mem->dma, mem->len);
>> + if (IS_ERR_OR_NULL(va))
>> + return -ENOMEM;
>> +
>> + mem->va = (void *)va;
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_mem_unmap
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to unmap
>> + *
>> + * Unmap memory-region carveout
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_rproc_mem_unmap(struct rproc *rproc,
>> + struct rproc_mem_entry *mem)
>> +{
>> + iounmap((void __iomem *)mem->va);
>> + return 0;
>> +}
>> +
>> +/*
>> + * add_mem_regions
>> + * @r5_core: single R5 core's corresponding zynqmp_r5_core type instance
>> + * @rmem: reserved mem region parsed from dt node
>> + *
>> + * Construct rproc mem carveouts from carveout provided in
>> + * memory-region property
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int add_mem_regions_carveout(struct zynqmp_r5_core *r5_core,
>> + struct reserved_mem *rmem)
>> +{
>> + struct device *dev;
>> + struct rproc_mem_entry *mem;
>> + struct rproc *rproc;
>> +
>> + rproc = r5_core->rproc;
>> + dev = r5_core->dev;
>> +
>> + /* Register associated reserved memory regions */
>> + mem = rproc_mem_entry_init(dev, NULL,
>> + (dma_addr_t)rmem->base,
>> + rmem->size, rmem->base,
>> + zynqmp_r5_rproc_mem_map,
>> + zynqmp_r5_rproc_mem_unmap,
>> + rmem->name);
>> + if (!mem)
>> + return -ENOMEM;
>> +
>> + rproc_add_carveout(rproc, mem);
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_rproc_mem_unmap
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to unmap
>> + *
>> + * Unmap TCM banks when powering down R5 core.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int tcm_mem_unmap(struct rproc *rproc, struct rproc_mem_entry *mem)
>> +{
>> + struct zynqmp_r5_core *r5_core;
>> + int i;
>> + u32 pm_domain_id;
>> +
>> + r5_core = rproc->priv;
>> + if (!r5_core) {
>> + pr_err("r5 core is not available\n");
>> + return -EINVAL;
>> + }
>> +
>> + iounmap((void __iomem *)mem->va);
>> +
>> + for (i = 0; i < r5_core->tcm_bank_count; i++) {
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> + if (zynqmp_pm_release_node(pm_domain_id))
>> + pr_warn("can't turn off TCM bank 0x%x", pm_domain_id);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * tcm_mem_map
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @mem: mem entry to initialize the va and da fields of
>> + *
>> + * Given TCM bank entry, this callback will set device address for R5
>> + * running on TCM and also setup virtual address for TCM bank
>> + * remoteproc carveout.
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int tcm_mem_map(struct rproc *rproc,
>> + struct rproc_mem_entry *mem)
>> +{
>> + void __iomem *va;
>> +
>> + va = ioremap_wc(mem->dma, mem->len);
>> + if (IS_ERR_OR_NULL(va))
>> + return -ENOMEM;
>> +
>> + /* Update memory entry va */
>> + mem->va = (void *)va;
>> +
>> + /* clear TCMs */
>> + memset_io(va, 0, mem->len);
>> +
>> + /*
>> + * The R5s expect their TCM banks to be at address 0x0 and 0x2000,
>> + * while on the Linux side they are at 0xffexxxxx.
>> + *
>> + * Zero out the high 12 bits of the address. This will give
>> + * expected values for TCM Banks 0A and 0B (0x0 and 0x20000).
>> + */
>> + mem->da &= 0x000fffff;
>> +
>> + /*
>> + * TCM Banks 1A and 1B still have to be translated.
>> + *
>> + * Below handle these two banks' absolute addresses (0xffe90000 and
>> + * 0xffeb0000) and convert to the expected relative addresses
>> + * (0x0 and 0x20000).
>> + */
>> + if (mem->da == 0x90000 || mem->da == 0xB0000)
>> + mem->da -= 0x90000;
>> +
>> + /* if translated TCM bank address is not valid report error */
>> + if (mem->da != 0x0 && mem->da != 0x20000) {
>> + dev_err(&rproc->dev, "invalid TCM address: %x\n", mem->da);
>> + return -EINVAL;
>> + }
>> + return 0;
>> +}
>> +
>> +static int add_tcm_carveout_split_mode(struct rproc *rproc)
>> +{
>> + int i, num_banks, ret;
>> + struct rproc_mem_entry *mem;
>> + u32 pm_domain_id;
>> + phys_addr_t bank_addr;
>> + size_t bank_size;
>> + char *bank_name;
>> + struct device *dev;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
>> +
>> + dev = r5_core->dev;
>> +
>> + /* go through zynqmp banks for r5 node */
>> + num_banks = r5_core->tcm_bank_count;
>> + if (num_banks <= 0) {
>> + dev_err(dev, "need to specify TCM banks\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * Power-on Each 64KB TCM,
>> + * register its address space, map and unmap functions
>> + * and add carveouts accordingly
>> + */
>> + for (i = 0; i < num_banks; i++) {
>> + bank_addr = r5_core->tcm_banks[i].addr;
>> + bank_name = r5_core->tcm_banks[i].bank_name;
>> + bank_size = r5_core->tcm_banks[i].size;
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> +
>> + ret = zynqmp_pm_request_node(pm_domain_id,
>> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
>> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
>> + if (ret < 0) {
>> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
>> + return ret;
>> + }
>> +
>> + dev_dbg(dev, "TCM carveout split mode %s addr=%llx, size=0x%lx",
>> + bank_name, bank_addr, bank_size);
>> +
>> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
>> + bank_size, bank_addr,
>> + tcm_mem_map, tcm_mem_unmap,
>> + bank_name);
>> + if (!mem) {
>> + /* If failed, Turn off all TCM banks turned on before */
>> + do {
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> + ret = zynqmp_pm_release_node(pm_domain_id);
>> + if (ret)
>> + dev_warn(dev,
>> + "fail to release node: 0x%x, %x\n",
>> + pm_domain_id, ret);
>> + } while (i--);
>> + return -ENOMEM;
>> + }
>> +
>> + rproc_add_carveout(rproc, mem);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static int add_tcm_carveout_lockstep_mode(struct rproc *rproc)
>> +{
>> + int i, num_banks, ret;
>> + struct rproc_mem_entry *mem;
>> + u32 pm_domain_id;
>> + phys_addr_t bank_addr;
>> + size_t bank_size = 0;
>> + char *bank_name;
>> + struct device *dev;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
>> + dev = r5_core->dev;
>> +
>> + /* Go through zynqmp banks for r5 node */
>> + num_banks = r5_core->tcm_bank_count;
>> + if (num_banks <= 0) {
>> + dev_err(dev, "need to specify TCM banks\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * In lockstep mode, TCM is contiguous memory block
>> + * However, each TCM block still needs to be enabled individually.
>> + * So, Enable each TCM block individually, but add their size
>> + * to create contiguous memory region.
>> + */
>> + bank_addr = r5_core->tcm_banks[0].addr;
>> + bank_name = r5_core->tcm_banks[0].bank_name;
>> +
>> + for (i = 0; i < num_banks; i++) {
>> + bank_size += r5_core->tcm_banks[i].size;
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> +
>> + /* Turn on each TCM bank individually */
>> + ret = zynqmp_pm_request_node(pm_domain_id,
>> + ZYNQMP_PM_CAPABILITY_ACCESS, 0,
>> + ZYNQMP_PM_REQUEST_ACK_BLOCKING);
>> + if (ret < 0) {
>> + dev_err(dev, "failed to turn on TCM 0x%x", pm_domain_id);
>> + return ret;
>> + }
>> + }
>> +
>> + dev_dbg(dev, "TCM add carveout lockstep mode %s addr=0x%llx, size=0x%lx",
>> + bank_name, bank_addr, bank_size);
>> +
>> + /* Register TCM address range, TCM map and unmap functions */
>> + mem = rproc_mem_entry_init(dev, NULL, bank_addr,
>> + bank_size, bank_addr,
>> + tcm_mem_map, tcm_mem_unmap,
>> + bank_name);
>> + if (!mem) {
>> + /* In case of error, turn off TCM banks */
>> + for (i = 0; i < num_banks; i++) {
>> + pm_domain_id = r5_core->tcm_banks[i].pm_domain_id;
>> + ret = zynqmp_pm_release_node(pm_domain_id);
>> + if (ret)
>> + dev_warn(dev,
>> + "fail to release node: 0x%x ret: %x\n",
>> + pm_domain_id, ret);
>> + }
>> + return -ENOMEM;
>> + }
>> +
>> + /* If registration is success, add carveouts */
>> + rproc_add_carveout(rproc, mem);
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * add_tcm_banks()
>> + * @rproc: single R5 core's corresponding rproc instance
>> + *
>> + * Given R5 node in remoteproc instance
>> + * allocate remoteproc carveout for TCM memory
>> + * needed for firmware to be loaded
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int add_tcm_banks(struct rproc *rproc)
>> +{
>> + struct device *dev;
>> + struct platform_device *parent_pdev;
>> + struct zynqmp_r5_cluster *cluster;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + r5_core = (struct zynqmp_r5_core *)rproc->priv;
>> + if (!r5_core)
>> + return -EINVAL;
>> +
>> + dev = r5_core->dev;
>> +
>> + cluster = dev_get_drvdata(dev->parent);
>> + if (!cluster) {
>> + dev_err(&parent_pdev->dev, "Invalid driver data\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * In lockstep mode TCM banks are one contiguous memory region of 256Kb
>> + * In split mode, each TCM bank is 64Kb and not contiguous.
>> + * We add memory carveouts accordingly.
>> + */
>> + if (cluster->mode == SPLIT_MODE)
>> + return add_tcm_carveout_split_mode(rproc);
>> + else if (cluster->mode == LOCKSTEP_MODE)
>> + return add_tcm_carveout_lockstep_mode(rproc);
>> +
>> + dev_err(cluster->dev, "invalid cluster mode\n");
>> + return -EINVAL;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_parse_fw()
>> + * @rproc: single R5 core's corresponding rproc instance
>> + * @fw: ptr to firmware to be loaded onto r5 core
>> + *
>> + * When loading firmware, ensure the necessary carveouts are in remoteproc
>> + *
>> + * return 0 on success, otherwise non-zero value on failure
>> + */
>> +static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct firmware *fw)
>> +{
>> + int ret;
>> + struct zynqmp_r5_core *r5_core;
>> + struct device *dev;
>> +
>> + r5_core = rproc->priv;
>> +
>> + dev = r5_core->dev;
>> +
>> + ret = add_tcm_banks(rproc);
>> + if (ret) {
>> + dev_err(dev, "failed to get TCM banks, err %d\n", ret);
>> + return ret;
>> + }
>> +
>> + ret = rproc_elf_load_rsc_table(rproc, fw);
>> + if (ret == -EINVAL) {
>> + /*
>> + * resource table only required for IPC.
>> + * if not present, this is not necessarily an error;
>> + * for example, loading r5 hello world application
>> + * so simply inform user and keep going.
>> + */
>> + dev_info(&rproc->dev, "no resource table found.\n");
>> + ret = 0;
>> + }
>> + return ret;
>> +}
>> +
>> +static struct rproc_ops zynqmp_r5_rproc_ops = {
>> + .start = zynqmp_r5_rproc_start,
>> + .stop = zynqmp_r5_rproc_stop,
>> + .load = rproc_elf_load_segments,
>> + .parse_fw = zynqmp_r5_parse_fw,
>> + .find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
>> + .sanity_check = rproc_elf_sanity_check,
>> + .get_boot_addr = rproc_elf_get_boot_addr,
>> +};
>> +
>> +/**
>> + * zynqmp_r5_add_rproc_core()
>> + * Probes ZynqMP R5 processor device node
>> + * this is called for each individual R5 core
>> + *
>> + * @cdev: Device node of each r5 core
>> + *
>> + * Return: zynqmp_r5_core object for success, error pointer in case of error.
>> + */
>> +static struct zynqmp_r5_core *zynqmp_r5_add_rproc_core(struct device *cdev)
>> +{
>> + int ret;
>> + struct rproc *r5_rproc;
>> + struct zynqmp_r5_core *r5_core;
>> +
>> + /* Set up DMA mask */
>> + ret = dma_set_coherent_mask(cdev, DMA_BIT_MASK(32));
>> + if (ret)
>> + return ERR_PTR(ret);
>> +
>> + /* Allocate remoteproc instance */
>> + r5_rproc = devm_rproc_alloc(cdev, dev_name(cdev),
>> + &zynqmp_r5_rproc_ops,
>> + NULL, sizeof(struct zynqmp_r5_core));
>> + if (!r5_rproc) {
>> + dev_err(cdev, "failed to allocate memory for rproc instance\n");
>> + return ERR_PTR(-ENOMEM);
>> + }
>> +
>> + r5_rproc->auto_boot = false;
>> + r5_core = (struct zynqmp_r5_core *)r5_rproc->priv;
>> + r5_core->dev = cdev;
>> +
>> + /* Add R5 remoteproc core */
>> + ret = devm_rproc_add(cdev, r5_rproc);
>> + if (ret) {
>> + dev_err(cdev, "failed to add r5 remoteproc\n");
>> + return ERR_PTR(ret);
>> + }
>> +
>> + r5_core->rproc = r5_rproc;
>> + return r5_core;
>> +}
>> +
>> +/**
>> + * zynqmp_r5_get_tcm_node()
>> + * Ideally this function should parse tcm node and store information
>> + * in r5_core instance. We will use hardcoded TCM information from
>> + * driver for now in this function.
>> + *
>> + * @cluster: pointer to zynqmp_r5_cluster type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster)
>> +{
>> + int tcm_bank_count, tcm_node;
>> + int i = 0, j;
>> + struct zynqmp_r5_core *r5_core;
>> + const struct mem_bank_data *tcm = zynqmp_tcm_banks;
>> + struct device *dev = cluster->dev;
>> +
>> + /*
>> + * ToDo: Use predefined TCM address space values from driver until
>> + * system-dt spec is not final for TCM
>> + */
>> + tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks);
>> +
>> + /* count per core tcm banks */
>> + tcm_bank_count = tcm_bank_count / cluster->core_count;
>> +
>> + /* r5 core 0 will use all of TCM banks in lockstep mode.
>> + * In split mode, r5 core0 will use 128k and r5 core1 will use another
>> + * 128k. Assign TCM banks to each core accordingly
>> + */
>> + tcm_node = 0;
>> + for (i = 0; i < cluster->core_count; i++) {
>> + r5_core = cluster->r5_cores[i];
>> + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count,
>> + sizeof(struct mem_bank_data),
>> + GFP_KERNEL);
>> + if (!r5_core->tcm_banks)
>> + return -ENOMEM;
>> +
>> + for (j = 0; j < tcm_bank_count; j++) {
>> + /* Use pre-defined TCM reg values.
>> + * Eventually this should be replaced by values
>> + * parsed from dts.
>> + */
>> + r5_core->tcm_banks[j].addr = tcm[tcm_node].addr;
>> + r5_core->tcm_banks[j].size = tcm[tcm_node].size;
>> + r5_core->tcm_banks[j].pm_domain_id = tcm[tcm_node].pm_domain_id;
>> + r5_core->tcm_banks[j].bank_name = tcm[tcm_node].bank_name;
>> + tcm_node++;
>> + }
>> +
>> + r5_core->tcm_bank_count = tcm_bank_count;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/**
>> + * zynqmp_r5_get_mem_region_node()
>> + * parse memory-region property from dt node and add
>> + * memory region carveouts
>> + *
>> + * @r5_core: pointer to zynqmp_r5_core type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core *r5_core)
>> +{
>> + int res_mem_count, i, ret;
>> + struct device *dev;
>> + struct device_node *np, *rmem_np;
>> + struct reserved_mem *rmem;
>> +
>> + dev = r5_core->dev;
>> +
>> + np = r5_core->np;
>> +
>> + res_mem_count = of_property_count_elems_of_size(np, "memory-region",
>> + sizeof(phandle));
>> + if (res_mem_count <= 0) {
>> + dev_warn(dev, "failed to get memory-region property %d\n",
>> + res_mem_count);
>> + return -EINVAL;
>> + }
>> +
>> + for (i = 0; i < res_mem_count; i++) {
>> + rmem_np = of_parse_phandle(np, "memory-region", i);
>> + if (!rmem_np)
>> + return -EINVAL;
>> +
>> + rmem = of_reserved_mem_lookup(rmem_np);
>> + if (!rmem) {
>> + of_node_put(rmem_np);
>> + return -EINVAL;
>> + }
>> +
>> + ret = add_mem_regions_carveout(r5_core, rmem);
>> + if (ret)
>> + dev_warn(dev, "failed to get reserve mem regions %d\n",
>> + ret);
>> +
>> + of_node_put(rmem_np);
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_core_init()
>> + * Create and initialize zynqmp_r5_core type object
>> + *
>> + * @cluster: pointer to zynqmp_r5_cluster type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
>> +{
>> + int ret, i;
>> + struct zynqmp_r5_core *r5_core;
>> + struct device *dev = cluster->dev;
>> +
>> + ret = zynqmp_r5_get_tcm_node(cluster);
>> + if (ret < 0) {
>> + dev_err(dev, "can't get tcm node, err %d\n", ret);
>> + return ret;
>> + }
>> +
>> + for (i = 0; i < cluster->core_count; i++) {
>> + r5_core = cluster->r5_cores[i];
>> +
>> + ret = zynqmp_r5_get_mem_region_node(r5_core);
>> + if (ret)
>> + dev_warn(dev, "memory-region prop failed %d\n", ret);
>> +
>> + /* Initialize r5 cores with power-domains parsed from dts */
>> + ret = of_property_read_u32_index(r5_core->np, "power-domains",
>> + 1, &r5_core->pm_domain_id);
>> + if (ret) {
>> + dev_err(dev, "failed to get power-domains property\n");
>> + return ret;
>> + }
>> +
>> + ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
>> + if (ret) {
>> + dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
>> + cluster->mode, ret);
>> + return ret;
>> + }
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * zynqmp_r5_cluster_init()
>> + * Create and initialize zynqmp_r5_cluster type object
>> + *
>> + * @cluster: pointer to zynqmp_r5_cluster type object
>> + *
>> + * Return: 0 for success and error code for failure.
>> + */
>> +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
>> +{
>> + struct device *dev = cluster->dev;
>> + struct device_node *dev_node = dev_of_node(dev);
>> + struct device_node *child;
>> + struct platform_device *child_pdev;
>> + int core_count = 0, ret, i;
>> + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
>> + struct zynqmp_r5_core **r5_cores;
>> +
>> + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
>> +
>> + /*
>> + * on success returns 0, if not defined then returns -EINVAL,
>> + * In that case, default is LOCKSTEP mode
>> + */
>> + if (ret != -EINVAL && ret != 0) {
>> + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * For now driver only supports split mode and lockstep mode.
>> + * fail driver probe if either of that is not set in dts
>> + */
>> + if (cluster_mode == SINGLE_CPU_MODE) {
>> + dev_err(dev, "driver does not support single cpu mode\n");
>> + return -EINVAL;
>> + } else if ((cluster_mode != SPLIT_MODE &&
>> + cluster_mode != LOCKSTEP_MODE)) {
>> + dev_err(dev, "Invalid cluster mode\n");
>> + return -EINVAL;
>> + }
>> +
>> + /*
>> + * Number of cores is decided by number of child nodes of
>> + * r5f subsystem node in dts. If Split mode is used in dts
>> + * 2 child nodes are expected.
>> + * In lockstep mode if two child nodes are available,
>> + * only use first child node and consider it as core0
>> + * and ignore core1 dt node.
>> + */
>> + core_count = of_get_available_child_count(dev_node);
>> + if (core_count <= 0) {
>> + dev_err(dev, "Invalid number of r5 cores %d", core_count);
>> + return -EINVAL;
>> + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
>> + dev_err(dev, "Invalid number of r5 cores for split mode\n");
>> + return -EINVAL;
>> + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
>> + dev_warn(dev, "Only r5 core0 will be used\n");
>> + core_count = 1;
>> + }
>> +
>> + r5_cores = devm_kcalloc(dev, core_count,
>> + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
>> + if (!r5_cores)
>> + return -ENOMEM;
>> +
>> + i = 0;
>> + for_each_available_child_of_node(dev_node, child) {
>> + child_pdev = of_find_device_by_node(child);
>> + if (!child_pdev) {
>> + of_node_put(child);
>> + return -ENODEV;
>> + }
>> +
>> + /* create and add remoteproc instance of type struct rproc */
>> + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
> Function zynqmp_r5_add_rproc_core() returns an error code on error that is never
> checked.
>
>> + r5_cores[i]->dev = &child_pdev->dev;
> This will result in a kernel stack trace if zynqmp_r5_add_rproc_core() fails.
Sure I will fix it.
>> + if (!r5_cores[i]->dev) {
>> + dev_err(dev, "can't get device for r5 core %d\n", i);
>> + of_node_put(child);
>> + return -ENODEV;
>> + }
> And here the validity of child_pdev->dev is checked _after_ it has been passed
> to zynqmp_r5_add_rproc_core(). This check should not be needed if proper
> error handling is done above.
I agree.
>> +
>> + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
>> + if (!r5_cores[i]->np) {
>> + dev_err(dev, "can't get device node for r5 core %d\n", i);
>> + of_node_put(child);
> As mention in the documentation for of_find_device_by_node(), the function takes
> a reference to child_pdev->dev that needs to be released with put_device(). In
> fact I don't see the reference dropped anywhere, even when the driver is
> released.
I agree. I want to use r5_cores[i]->dev throughout driver so,
Is it fine if I use put_device() during driver release i.e. function
zynqmp_r5_cluster_exit( ) ?
> Moreover, if we end up here for r5_cores[1], resources acquired for r5_core[0]
> are not released.
Most of the resources (such as zynqmp_r5_rproc_ops and r5_rproc objects)
for r5_core[0] are allocated using devm_* API,
so I believe it will be de-allocated when driver probe fails and driver
exits.
If r5_cores[1] fails, then r5_cores[0]->np needs to be released here. I
will find out mechanism to take care of this.
Please let me know if I am missing anything here.
>> + return -ENODEV;
>> + }
>> +
>> + i++;
>> +
>> + /*
>> + * If two child nodes are available in dts in lockstep mode,
>> + * then ignore second child node.
>> + */
>> + if (i == core_count) {
>> + of_node_put(child);
>> + break;
>> + }
>> + of_node_put(child);
> This one is not needed as it is already done by the
> for_each_available_child_of_node() loop.
Ok I will remove it.
> More comments tomorrow.
>
> Thanks,
> Mathieu
>
>> + }
>> +
>> + cluster->mode = cluster_mode;
>> + cluster->core_count = core_count;
>> + cluster->r5_cores = r5_cores;
>> +
>> + ret = zynqmp_r5_core_init(cluster);
>> + if (ret < 0) {
>> + dev_err(dev, "failed to init r5 core err %d\n", ret);
>> + return ret;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static void zynqmp_r5_cluster_exit(void *data)
>> +{
>> + struct platform_device *pdev = (struct platform_device *)data;
>> +
>> + platform_set_drvdata(pdev, NULL);
>> +
>> + pr_info("Exit r5f subsystem driver\n");
>> +}
>> +
>> +/*
>> + * zynqmp_r5_remoteproc_probe()
>> + *
>> + * @pdev: domain platform device for R5 cluster
>> + *
>> + * called when driver is probed, for each R5 core specified in DT,
>> + * setup as needed to do remoteproc-related operations
>> + *
>> + * Return: 0 for success, negative value for failure.
>> + */
>> +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
>> +{
>> + int ret;
>> + struct zynqmp_r5_cluster *cluster;
>> + struct device *dev = &pdev->dev;
>> +
>> + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
>> + if (!cluster)
>> + return -ENOMEM;
>> +
>> + cluster->dev = dev;
>> +
>> + ret = devm_of_platform_populate(dev);
>> + if (ret) {
>> + dev_err(dev, "failed to populate platform dev %d\n", ret);
>> + return ret;
>> + }
>> +
>> + /* wire in so each core can be cleaned up at driver remove */
>> + platform_set_drvdata(pdev, cluster);
>> +
>> + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
>> + if (ret)
>> + return ret;
>> +
>> + ret = zynqmp_r5_cluster_init(cluster);
>> + if (ret) {
>> + dev_err(dev, "Invalid r5f subsystem device tree\n");
>> + return ret;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +/* Match table for OF platform binding */
>> +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
>> + { .compatible = "xlnx,zynqmp-r5fss", },
>> + { /* end of list */ },
>> +};
>> +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
>> +
>> +static struct platform_driver zynqmp_r5_remoteproc_driver = {
>> + .probe = zynqmp_r5_remoteproc_probe,
>> + .driver = {
>> + .name = "zynqmp_r5_remoteproc",
>> + .of_match_table = zynqmp_r5_remoteproc_match,
>> + },
>> +};
>> +module_platform_driver(zynqmp_r5_remoteproc_driver);
>> +
>> +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
>> +MODULE_AUTHOR("Xilinx Inc.");
>> +MODULE_LICENSE("GPL v2");
>> --
>> 2.25.1
>>
[...]
> > > +/**
> > > + * zynqmp_r5_get_tcm_node()
> > > + * Ideally this function should parse tcm node and store information
> > > + * in r5_core instance. We will use hardcoded TCM information from
> > > + * driver for now in this function.
> > > + *
> > > + * @cluster: pointer to zynqmp_r5_cluster type object
> > > + *
> > > + * Return: 0 for success and error code for failure.
> > > + */
> > > +static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster)
> > > +{
> > > + int tcm_bank_count, tcm_node;
> > > + int i = 0, j;
> > Variable @i doesn't need to be initialised.
> Agree.
> > > + struct zynqmp_r5_core *r5_core;
> > > + const struct mem_bank_data *tcm = zynqmp_tcm_banks;
> > > + struct device *dev = cluster->dev;
> > > +
> > > + /*
> > > + * ToDo: Use predefined TCM address space values from driver until
> > > + * system-dt spec is not final for TCM
> > > + */
> > > + tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks);
> > > +
> > > + /* count per core tcm banks */
> > > + tcm_bank_count = tcm_bank_count / cluster->core_count;
> > > +
> > > + /* r5 core 0 will use all of TCM banks in lockstep mode.
> > > + * In split mode, r5 core0 will use 128k and r5 core1 will use another
> > > + * 128k. Assign TCM banks to each core accordingly
> > > + */
> > Comment format problem.
> My bad. I will fix it.
> > > + tcm_node = 0;
> > > + for (i = 0; i < cluster->core_count; i++) {
> > > + r5_core = cluster->r5_cores[i];
> > > + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count,
> > > + sizeof(struct mem_bank_data),
> > > + GFP_KERNEL);
> > > + if (!r5_core->tcm_banks)
> > > + return -ENOMEM;
> > > +
> > > + for (j = 0; j < tcm_bank_count; j++) {
> > > + /* Use pre-defined TCM reg values.
> > > + * Eventually this should be replaced by values
> > > + * parsed from dts.
> > > + */
> > Same. I commented on that in my previous review of this set.
> Yes. I will fix it.
> > > + r5_core->tcm_banks[j].addr = tcm[tcm_node].addr;
> > > + r5_core->tcm_banks[j].size = tcm[tcm_node].size;
> > > + r5_core->tcm_banks[j].pm_domain_id = tcm[tcm_node].pm_domain_id;
> > > + r5_core->tcm_banks[j].bank_name = tcm[tcm_node].bank_name;
> > > + tcm_node++;
> > > + }
> > > +
> > > + r5_core->tcm_bank_count = tcm_bank_count;
> > Why do we need to copy the exact same information from zynqmp_tcm_banks to ->tcm_banks? Why can't ->tcm_banks point to an offset in zynqmp_tcm_banks and ->tcm_bank_count set accordingly?
>
> zynqmp_tcm_banks is global variable. If I use ->tcm_banks point to offset of
> zynqmp_tcm_banks and use it in driver, that section of code won't be thread
> safe anymore.
>
> I might have to use mutex to protect it. So Instead of pointing it to global
> variable, I decided to copy information from global variable to ->tcm_banks.
As far as I can tell ->tcm_banks is a read-only field and as such shouldn't
need mutext protection.
Thanks,
Mathieu
>
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +/**
> > > + * zynqmp_r5_get_mem_region_node()
> > > + * parse memory-region property from dt node and add
> > > + * memory region carveouts
> > > + *
> > > + * @r5_core: pointer to zynqmp_r5_core type object
> > > + *
> > > + * Return: 0 for success and error code for failure.
> > > + */
> > > +static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core *r5_core)
> > > +{
> > > + int res_mem_count, i, ret;
> > > + struct device *dev;
> > > + struct device_node *np, *rmem_np;
> > > + struct reserved_mem *rmem;
> > > +
> > > + dev = r5_core->dev;
> > > +
> > > + np = r5_core->np;
> > > +
> > > + res_mem_count = of_property_count_elems_of_size(np, "memory-region",
> > > + sizeof(phandle));
> > > + if (res_mem_count <= 0) {
> > > + dev_warn(dev, "failed to get memory-region property %d\n",
> > > + res_mem_count);
> > > + return -EINVAL;
> > > + }
> > > +
> > > + for (i = 0; i < res_mem_count; i++) {
> > > + rmem_np = of_parse_phandle(np, "memory-region", i);
> > > + if (!rmem_np)
> > > + return -EINVAL;
> > > +
> > > + rmem = of_reserved_mem_lookup(rmem_np);
> > > + if (!rmem) {
> > > + of_node_put(rmem_np);
> > > + return -EINVAL;
> > What happens to previously allocated regions when there is a failure?
>
> Right, I need to unmap memory and use kfree(rmem) to de-allocate memory.
>
> > > + }
> > > +
> > > + ret = add_mem_regions_carveout(r5_core, rmem);
> > > + if (ret)
> > > + dev_warn(dev, "failed to get reserve mem regions %d\n",
> > > + ret);
> > Same here.
>
> Here as well. I will release reserved mem dev with
> of_reserved_mem_device_release API.
>
> One more thing. I moved add_mem_regions_carveout from parse_fw to
> zynqmp_r5_get_mem_region_node.
>
> However, I believe I should move it back to parse_fw.
>
> Following test case is failing in when add_mem_regions_carveout is not
> available in parse_fw:
>
> load_rpu_fw -> start rpu -> stop rpu -> start rpu again
>
> In above case, during mem regions are not added again.
>
> Is it fine, if I move this add_mem_regions_carveout back to
> zynqmp_r5_parse_fw ?
>
> > I am out of time for this set. Please address comments provided up to here and
> > we will see about the rest in a future revision.
> >
> > Thanks,
> > Mathieu
> >
> > > +
> > > + of_node_put(rmem_np);
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +/*
> > > + * zynqmp_r5_core_init()
> > > + * Create and initialize zynqmp_r5_core type object
> > > + *
> > > + * @cluster: pointer to zynqmp_r5_cluster type object
> > > + *
> > > + * Return: 0 for success and error code for failure.
> > > + */
> > > +static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
> > > +{
> > > + int ret, i;
> > > + struct zynqmp_r5_core *r5_core;
> > > + struct device *dev = cluster->dev;
> > > +
> > > + ret = zynqmp_r5_get_tcm_node(cluster);
> > > + if (ret < 0) {
> > > + dev_err(dev, "can't get tcm node, err %d\n", ret);
> > > + return ret;
> > > + }
> > > +
> > > + for (i = 0; i < cluster->core_count; i++) {
> > > + r5_core = cluster->r5_cores[i];
> > > +
> > > + ret = zynqmp_r5_get_mem_region_node(r5_core);
> > > + if (ret)
> > > + dev_warn(dev, "memory-region prop failed %d\n", ret);
> > > +
> > > + /* Initialize r5 cores with power-domains parsed from dts */
> > > + ret = of_property_read_u32_index(r5_core->np, "power-domains",
> > > + 1, &r5_core->pm_domain_id);
> > > + if (ret) {
> > > + dev_err(dev, "failed to get power-domains property\n");
> > > + return ret;
> > > + }
> > > +
> > > + ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
> > > + if (ret) {
> > > + dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
> > > + cluster->mode, ret);
> > > + return ret;
> > > + }
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +/*
> > > + * zynqmp_r5_cluster_init()
> > > + * Create and initialize zynqmp_r5_cluster type object
> > > + *
> > > + * @cluster: pointer to zynqmp_r5_cluster type object
> > > + *
> > > + * Return: 0 for success and error code for failure.
> > > + */
> > > +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
> > > +{
> > > + struct device *dev = cluster->dev;
> > > + struct device_node *dev_node = dev_of_node(dev);
> > > + struct device_node *child;
> > > + struct platform_device *child_pdev;
> > > + int core_count = 0, ret, i;
> > > + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
> > > + struct zynqmp_r5_core **r5_cores;
> > > +
> > > + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
> > > +
> > > + /*
> > > + * on success returns 0, if not defined then returns -EINVAL,
> > > + * In that case, default is LOCKSTEP mode
> > > + */
> > > + if (ret != -EINVAL && ret != 0) {
> > > + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
> > > + return -EINVAL;
> > > + }
> > > +
> > > + /*
> > > + * For now driver only supports split mode and lockstep mode.
> > > + * fail driver probe if either of that is not set in dts
> > > + */
> > > + if (cluster_mode == SINGLE_CPU_MODE) {
> > > + dev_err(dev, "driver does not support single cpu mode\n");
> > > + return -EINVAL;
> > > + } else if ((cluster_mode != SPLIT_MODE &&
> > > + cluster_mode != LOCKSTEP_MODE)) {
> > > + dev_err(dev, "Invalid cluster mode\n");
> > > + return -EINVAL;
> > > + }
> > > +
> > > + /*
> > > + * Number of cores is decided by number of child nodes of
> > > + * r5f subsystem node in dts. If Split mode is used in dts
> > > + * 2 child nodes are expected.
> > > + * In lockstep mode if two child nodes are available,
> > > + * only use first child node and consider it as core0
> > > + * and ignore core1 dt node.
> > > + */
> > > + core_count = of_get_available_child_count(dev_node);
> > > + if (core_count <= 0) {
> > > + dev_err(dev, "Invalid number of r5 cores %d", core_count);
> > > + return -EINVAL;
> > > + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
> > > + dev_err(dev, "Invalid number of r5 cores for split mode\n");
> > > + return -EINVAL;
> > > + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
> > > + dev_warn(dev, "Only r5 core0 will be used\n");
> > > + core_count = 1;
> > > + }
> > > +
> > > + r5_cores = devm_kcalloc(dev, core_count,
> > > + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
> > > + if (!r5_cores)
> > > + return -ENOMEM;
> > > +
> > > + i = 0;
> > > + for_each_available_child_of_node(dev_node, child) {
> > > + child_pdev = of_find_device_by_node(child);
> > > + if (!child_pdev) {
> > > + of_node_put(child);
> > > + return -ENODEV;
> > > + }
> > > +
> > > + /* create and add remoteproc instance of type struct rproc */
> > > + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
> > > + r5_cores[i]->dev = &child_pdev->dev;
> > > + if (!r5_cores[i]->dev) {
> > > + dev_err(dev, "can't get device for r5 core %d\n", i);
> > > + of_node_put(child);
> > > + return -ENODEV;
> > > + }
> > > +
> > > + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
> > > + if (!r5_cores[i]->np) {
> > > + dev_err(dev, "can't get device node for r5 core %d\n", i);
> > > + of_node_put(child);
> > > + return -ENODEV;
> > > + }
> > > +
> > > + i++;
> > > +
> > > + /*
> > > + * If two child nodes are available in dts in lockstep mode,
> > > + * then ignore second child node.
> > > + */
> > > + if (i == core_count) {
> > > + of_node_put(child);
> > > + break;
> > > + }
> > > + of_node_put(child);
> > > + }
> > > +
> > > + cluster->mode = cluster_mode;
> > > + cluster->core_count = core_count;
> > > + cluster->r5_cores = r5_cores;
> > > +
> > > + ret = zynqmp_r5_core_init(cluster);
> > > + if (ret < 0) {
> > > + dev_err(dev, "failed to init r5 core err %d\n", ret);
> > > + return ret;
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +static void zynqmp_r5_cluster_exit(void *data)
> > > +{
> > > + struct platform_device *pdev = (struct platform_device *)data;
> > > +
> > > + platform_set_drvdata(pdev, NULL);
> > > +
> > > + pr_info("Exit r5f subsystem driver\n");
> > > +}
> > > +
> > > +/*
> > > + * zynqmp_r5_remoteproc_probe()
> > > + *
> > > + * @pdev: domain platform device for R5 cluster
> > > + *
> > > + * called when driver is probed, for each R5 core specified in DT,
> > > + * setup as needed to do remoteproc-related operations
> > > + *
> > > + * Return: 0 for success, negative value for failure.
> > > + */
> > > +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
> > > +{
> > > + int ret;
> > > + struct zynqmp_r5_cluster *cluster;
> > > + struct device *dev = &pdev->dev;
> > > +
> > > + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
> > > + if (!cluster)
> > > + return -ENOMEM;
> > > +
> > > + cluster->dev = dev;
> > > +
> > > + ret = devm_of_platform_populate(dev);
> > > + if (ret) {
> > > + dev_err(dev, "failed to populate platform dev %d\n", ret);
> > > + return ret;
> > > + }
> > > +
> > > + /* wire in so each core can be cleaned up at driver remove */
> > > + platform_set_drvdata(pdev, cluster);
> > > +
> > > + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
> > > + if (ret)
> > > + return ret;
> > > +
> > > + ret = zynqmp_r5_cluster_init(cluster);
> > > + if (ret) {
> > > + dev_err(dev, "Invalid r5f subsystem device tree\n");
> > > + return ret;
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +/* Match table for OF platform binding */
> > > +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
> > > + { .compatible = "xlnx,zynqmp-r5fss", },
> > > + { /* end of list */ },
> > > +};
> > > +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
> > > +
> > > +static struct platform_driver zynqmp_r5_remoteproc_driver = {
> > > + .probe = zynqmp_r5_remoteproc_probe,
> > > + .driver = {
> > > + .name = "zynqmp_r5_remoteproc",
> > > + .of_match_table = zynqmp_r5_remoteproc_match,
> > > + },
> > > +};
> > > +module_platform_driver(zynqmp_r5_remoteproc_driver);
> > > +
> > > +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
> > > +MODULE_AUTHOR("Xilinx Inc.");
> > > +MODULE_LICENSE("GPL v2");
> > > --
> > > 2.25.1
> > >
[...]
> > > + * zynqmp_r5_cluster_init()
> > > + * Create and initialize zynqmp_r5_cluster type object
> > > + *
> > > + * @cluster: pointer to zynqmp_r5_cluster type object
> > > + *
> > > + * Return: 0 for success and error code for failure.
> > > + */
> > > +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
> > > +{
> > > + struct device *dev = cluster->dev;
> > > + struct device_node *dev_node = dev_of_node(dev);
> > > + struct device_node *child;
> > > + struct platform_device *child_pdev;
> > > + int core_count = 0, ret, i;
> > > + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
> > > + struct zynqmp_r5_core **r5_cores;
> > > +
> > > + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
> > > +
> > > + /*
> > > + * on success returns 0, if not defined then returns -EINVAL,
> > > + * In that case, default is LOCKSTEP mode
> > > + */
> > > + if (ret != -EINVAL && ret != 0) {
> > > + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
> > > + return -EINVAL;
> > > + }
> > > +
> > > + /*
> > > + * For now driver only supports split mode and lockstep mode.
> > > + * fail driver probe if either of that is not set in dts
> > > + */
> > > + if (cluster_mode == SINGLE_CPU_MODE) {
> > > + dev_err(dev, "driver does not support single cpu mode\n");
> > > + return -EINVAL;
> > > + } else if ((cluster_mode != SPLIT_MODE &&
> > > + cluster_mode != LOCKSTEP_MODE)) {
> > > + dev_err(dev, "Invalid cluster mode\n");
> > > + return -EINVAL;
> > > + }
> > > +
> > > + /*
> > > + * Number of cores is decided by number of child nodes of
> > > + * r5f subsystem node in dts. If Split mode is used in dts
> > > + * 2 child nodes are expected.
> > > + * In lockstep mode if two child nodes are available,
> > > + * only use first child node and consider it as core0
> > > + * and ignore core1 dt node.
> > > + */
> > > + core_count = of_get_available_child_count(dev_node);
> > > + if (core_count <= 0) {
> > > + dev_err(dev, "Invalid number of r5 cores %d", core_count);
> > > + return -EINVAL;
> > > + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
> > > + dev_err(dev, "Invalid number of r5 cores for split mode\n");
> > > + return -EINVAL;
> > > + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
> > > + dev_warn(dev, "Only r5 core0 will be used\n");
> > > + core_count = 1;
> > > + }
> > > +
> > > + r5_cores = devm_kcalloc(dev, core_count,
> > > + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
> > > + if (!r5_cores)
> > > + return -ENOMEM;
> > > +
> > > + i = 0;
> > > + for_each_available_child_of_node(dev_node, child) {
> > > + child_pdev = of_find_device_by_node(child);
> > > + if (!child_pdev) {
> > > + of_node_put(child);
> > > + return -ENODEV;
> > > + }
> > > +
> > > + /* create and add remoteproc instance of type struct rproc */
> > > + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
> > Function zynqmp_r5_add_rproc_core() returns an error code on error that is never
> > checked.
> >
> > > + r5_cores[i]->dev = &child_pdev->dev;
> > This will result in a kernel stack trace if zynqmp_r5_add_rproc_core() fails.
> Sure I will fix it.
> > > + if (!r5_cores[i]->dev) {
> > > + dev_err(dev, "can't get device for r5 core %d\n", i);
> > > + of_node_put(child);
> > > + return -ENODEV;
> > > + }
> > And here the validity of child_pdev->dev is checked _after_ it has been passed
> > to zynqmp_r5_add_rproc_core(). This check should not be needed if proper
> > error handling is done above.
> I agree.
> > > +
> > > + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
> > > + if (!r5_cores[i]->np) {
> > > + dev_err(dev, "can't get device node for r5 core %d\n", i);
> > > + of_node_put(child);
> > As mention in the documentation for of_find_device_by_node(), the function takes
> > a reference to child_pdev->dev that needs to be released with put_device(). In
> > fact I don't see the reference dropped anywhere, even when the driver is
> > released.
>
> I agree. I want to use r5_cores[i]->dev throughout driver so,
>
> Is it fine if I use put_device() during driver release i.e. function
> zynqmp_r5_cluster_exit( ) ?
>
Yes, that would be the right place to do it.
> > Moreover, if we end up here for r5_cores[1], resources acquired for r5_core[0]
> > are not released.
>
> Most of the resources (such as zynqmp_r5_rproc_ops and r5_rproc objects) for
> r5_core[0] are allocated using devm_* API,
>
> so I believe it will be de-allocated when driver probe fails and driver
> exits.
>
> If r5_cores[1] fails, then r5_cores[0]->np needs to be released here. I will
> find out mechanism to take care of this.
>
> Please let me know if I am missing anything here.
>
I'll take another look in your next revision.
> > > + return -ENODEV;
> > > + }
> > > +
> > > + i++;
> > > +
> > > + /*
> > > + * If two child nodes are available in dts in lockstep mode,
> > > + * then ignore second child node.
> > > + */
> > > + if (i == core_count) {
> > > + of_node_put(child);
> > > + break;
> > > + }
> > > + of_node_put(child);
> > This one is not needed as it is already done by the
> > for_each_available_child_of_node() loop.
>
> Ok I will remove it.
>
> > More comments tomorrow.
> >
> > Thanks,
> > Mathieu
> >
> > > + }
> > > +
> > > + cluster->mode = cluster_mode;
> > > + cluster->core_count = core_count;
> > > + cluster->r5_cores = r5_cores;
> > > +
> > > + ret = zynqmp_r5_core_init(cluster);
> > > + if (ret < 0) {
> > > + dev_err(dev, "failed to init r5 core err %d\n", ret);
> > > + return ret;
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +static void zynqmp_r5_cluster_exit(void *data)
> > > +{
> > > + struct platform_device *pdev = (struct platform_device *)data;
> > > +
> > > + platform_set_drvdata(pdev, NULL);
> > > +
> > > + pr_info("Exit r5f subsystem driver\n");
> > > +}
> > > +
> > > +/*
> > > + * zynqmp_r5_remoteproc_probe()
> > > + *
> > > + * @pdev: domain platform device for R5 cluster
> > > + *
> > > + * called when driver is probed, for each R5 core specified in DT,
> > > + * setup as needed to do remoteproc-related operations
> > > + *
> > > + * Return: 0 for success, negative value for failure.
> > > + */
> > > +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
> > > +{
> > > + int ret;
> > > + struct zynqmp_r5_cluster *cluster;
> > > + struct device *dev = &pdev->dev;
> > > +
> > > + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
> > > + if (!cluster)
> > > + return -ENOMEM;
> > > +
> > > + cluster->dev = dev;
> > > +
> > > + ret = devm_of_platform_populate(dev);
> > > + if (ret) {
> > > + dev_err(dev, "failed to populate platform dev %d\n", ret);
> > > + return ret;
> > > + }
> > > +
> > > + /* wire in so each core can be cleaned up at driver remove */
> > > + platform_set_drvdata(pdev, cluster);
> > > +
> > > + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
> > > + if (ret)
> > > + return ret;
> > > +
> > > + ret = zynqmp_r5_cluster_init(cluster);
> > > + if (ret) {
> > > + dev_err(dev, "Invalid r5f subsystem device tree\n");
> > > + return ret;
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +/* Match table for OF platform binding */
> > > +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
> > > + { .compatible = "xlnx,zynqmp-r5fss", },
> > > + { /* end of list */ },
> > > +};
> > > +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
> > > +
> > > +static struct platform_driver zynqmp_r5_remoteproc_driver = {
> > > + .probe = zynqmp_r5_remoteproc_probe,
> > > + .driver = {
> > > + .name = "zynqmp_r5_remoteproc",
> > > + .of_match_table = zynqmp_r5_remoteproc_match,
> > > + },
> > > +};
> > > +module_platform_driver(zynqmp_r5_remoteproc_driver);
> > > +
> > > +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
> > > +MODULE_AUTHOR("Xilinx Inc.");
> > > +MODULE_LICENSE("GPL v2");
> > > --
> > > 2.25.1
> > >
On 2/22/22 8:57 AM, Mathieu Poirier wrote:
> [...]
>
>>>> +/**
>>>> + * zynqmp_r5_get_tcm_node()
>>>> + * Ideally this function should parse tcm node and store information
>>>> + * in r5_core instance. We will use hardcoded TCM information from
>>>> + * driver for now in this function.
>>>> + *
>>>> + * @cluster: pointer to zynqmp_r5_cluster type object
>>>> + *
>>>> + * Return: 0 for success and error code for failure.
>>>> + */
>>>> +static int zynqmp_r5_get_tcm_node(struct zynqmp_r5_cluster *cluster)
>>>> +{
>>>> + int tcm_bank_count, tcm_node;
>>>> + int i = 0, j;
>>> Variable @i doesn't need to be initialised.
>> Agree.
>>>> + struct zynqmp_r5_core *r5_core;
>>>> + const struct mem_bank_data *tcm = zynqmp_tcm_banks;
>>>> + struct device *dev = cluster->dev;
>>>> +
>>>> + /*
>>>> + * ToDo: Use predefined TCM address space values from driver until
>>>> + * system-dt spec is not final for TCM
>>>> + */
>>>> + tcm_bank_count = ARRAY_SIZE(zynqmp_tcm_banks);
>>>> +
>>>> + /* count per core tcm banks */
>>>> + tcm_bank_count = tcm_bank_count / cluster->core_count;
>>>> +
>>>> + /* r5 core 0 will use all of TCM banks in lockstep mode.
>>>> + * In split mode, r5 core0 will use 128k and r5 core1 will use another
>>>> + * 128k. Assign TCM banks to each core accordingly
>>>> + */
>>> Comment format problem.
>> My bad. I will fix it.
>>>> + tcm_node = 0;
>>>> + for (i = 0; i < cluster->core_count; i++) {
>>>> + r5_core = cluster->r5_cores[i];
>>>> + r5_core->tcm_banks = devm_kcalloc(dev, tcm_bank_count,
>>>> + sizeof(struct mem_bank_data),
>>>> + GFP_KERNEL);
>>>> + if (!r5_core->tcm_banks)
>>>> + return -ENOMEM;
>>>> +
>>>> + for (j = 0; j < tcm_bank_count; j++) {
>>>> + /* Use pre-defined TCM reg values.
>>>> + * Eventually this should be replaced by values
>>>> + * parsed from dts.
>>>> + */
>>> Same. I commented on that in my previous review of this set.
>> Yes. I will fix it.
>>>> + r5_core->tcm_banks[j].addr = tcm[tcm_node].addr;
>>>> + r5_core->tcm_banks[j].size = tcm[tcm_node].size;
>>>> + r5_core->tcm_banks[j].pm_domain_id = tcm[tcm_node].pm_domain_id;
>>>> + r5_core->tcm_banks[j].bank_name = tcm[tcm_node].bank_name;
>>>> + tcm_node++;
>>>> + }
>>>> +
>>>> + r5_core->tcm_bank_count = tcm_bank_count;
>>> Why do we need to copy the exact same information from zynqmp_tcm_banks to ->tcm_banks? Why can't ->tcm_banks point to an offset in zynqmp_tcm_banks and ->tcm_bank_count set accordingly?
>> zynqmp_tcm_banks is global variable. If I use ->tcm_banks point to offset of
>> zynqmp_tcm_banks and use it in driver, that section of code won't be thread
>> safe anymore.
>>
>> I might have to use mutex to protect it. So Instead of pointing it to global
>> variable, I decided to copy information from global variable to ->tcm_banks.
> As far as I can tell ->tcm_banks is a read-only field and as such shouldn't
> need mutext protection.
>
> Thanks,
> Mathieu
Ok, in that case I will point to offset in zynqmp_tcm_banks as suggested.
Rest of the comments looks good, and I will address them in next revision.
Thanks,
Tanmay
>>>> + }
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +/**
>>>> + * zynqmp_r5_get_mem_region_node()
>>>> + * parse memory-region property from dt node and add
>>>> + * memory region carveouts
>>>> + *
>>>> + * @r5_core: pointer to zynqmp_r5_core type object
>>>> + *
>>>> + * Return: 0 for success and error code for failure.
>>>> + */
>>>> +static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core *r5_core)
>>>> +{
>>>> + int res_mem_count, i, ret;
>>>> + struct device *dev;
>>>> + struct device_node *np, *rmem_np;
>>>> + struct reserved_mem *rmem;
>>>> +
>>>> + dev = r5_core->dev;
>>>> +
>>>> + np = r5_core->np;
>>>> +
>>>> + res_mem_count = of_property_count_elems_of_size(np, "memory-region",
>>>> + sizeof(phandle));
>>>> + if (res_mem_count <= 0) {
>>>> + dev_warn(dev, "failed to get memory-region property %d\n",
>>>> + res_mem_count);
>>>> + return -EINVAL;
>>>> + }
>>>> +
>>>> + for (i = 0; i < res_mem_count; i++) {
>>>> + rmem_np = of_parse_phandle(np, "memory-region", i);
>>>> + if (!rmem_np)
>>>> + return -EINVAL;
>>>> +
>>>> + rmem = of_reserved_mem_lookup(rmem_np);
>>>> + if (!rmem) {
>>>> + of_node_put(rmem_np);
>>>> + return -EINVAL;
>>> What happens to previously allocated regions when there is a failure?
>> Right, I need to unmap memory and use kfree(rmem) to de-allocate memory.
>>
>>>> + }
>>>> +
>>>> + ret = add_mem_regions_carveout(r5_core, rmem);
>>>> + if (ret)
>>>> + dev_warn(dev, "failed to get reserve mem regions %d\n",
>>>> + ret);
>>> Same here.
>> Here as well. I will release reserved mem dev with
>> of_reserved_mem_device_release API.
>>
>> One more thing. I moved add_mem_regions_carveout from parse_fw to
>> zynqmp_r5_get_mem_region_node.
>>
>> However, I believe I should move it back to parse_fw.
>>
>> Following test case is failing in when add_mem_regions_carveout is not
>> available in parse_fw:
>>
>> load_rpu_fw -> start rpu -> stop rpu -> start rpu again
>>
>> In above case, during mem regions are not added again.
>>
>> Is it fine, if I move this add_mem_regions_carveout back to
>> zynqmp_r5_parse_fw ?
>>
>>> I am out of time for this set. Please address comments provided up to here and
>>> we will see about the rest in a future revision.
>>>
>>> Thanks,
>>> Mathieu
>>>
>>>> +
>>>> + of_node_put(rmem_np);
>>>> + }
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +/*
>>>> + * zynqmp_r5_core_init()
>>>> + * Create and initialize zynqmp_r5_core type object
>>>> + *
>>>> + * @cluster: pointer to zynqmp_r5_cluster type object
>>>> + *
>>>> + * Return: 0 for success and error code for failure.
>>>> + */
>>>> +static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
>>>> +{
>>>> + int ret, i;
>>>> + struct zynqmp_r5_core *r5_core;
>>>> + struct device *dev = cluster->dev;
>>>> +
>>>> + ret = zynqmp_r5_get_tcm_node(cluster);
>>>> + if (ret < 0) {
>>>> + dev_err(dev, "can't get tcm node, err %d\n", ret);
>>>> + return ret;
>>>> + }
>>>> +
>>>> + for (i = 0; i < cluster->core_count; i++) {
>>>> + r5_core = cluster->r5_cores[i];
>>>> +
>>>> + ret = zynqmp_r5_get_mem_region_node(r5_core);
>>>> + if (ret)
>>>> + dev_warn(dev, "memory-region prop failed %d\n", ret);
>>>> +
>>>> + /* Initialize r5 cores with power-domains parsed from dts */
>>>> + ret = of_property_read_u32_index(r5_core->np, "power-domains",
>>>> + 1, &r5_core->pm_domain_id);
>>>> + if (ret) {
>>>> + dev_err(dev, "failed to get power-domains property\n");
>>>> + return ret;
>>>> + }
>>>> +
>>>> + ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
>>>> + if (ret) {
>>>> + dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
>>>> + cluster->mode, ret);
>>>> + return ret;
>>>> + }
>>>> + }
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +/*
>>>> + * zynqmp_r5_cluster_init()
>>>> + * Create and initialize zynqmp_r5_cluster type object
>>>> + *
>>>> + * @cluster: pointer to zynqmp_r5_cluster type object
>>>> + *
>>>> + * Return: 0 for success and error code for failure.
>>>> + */
>>>> +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
>>>> +{
>>>> + struct device *dev = cluster->dev;
>>>> + struct device_node *dev_node = dev_of_node(dev);
>>>> + struct device_node *child;
>>>> + struct platform_device *child_pdev;
>>>> + int core_count = 0, ret, i;
>>>> + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
>>>> + struct zynqmp_r5_core **r5_cores;
>>>> +
>>>> + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode", &cluster_mode);
>>>> +
>>>> + /*
>>>> + * on success returns 0, if not defined then returns -EINVAL,
>>>> + * In that case, default is LOCKSTEP mode
>>>> + */
>>>> + if (ret != -EINVAL && ret != 0) {
>>>> + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
>>>> + return -EINVAL;
>>>> + }
>>>> +
>>>> + /*
>>>> + * For now driver only supports split mode and lockstep mode.
>>>> + * fail driver probe if either of that is not set in dts
>>>> + */
>>>> + if (cluster_mode == SINGLE_CPU_MODE) {
>>>> + dev_err(dev, "driver does not support single cpu mode\n");
>>>> + return -EINVAL;
>>>> + } else if ((cluster_mode != SPLIT_MODE &&
>>>> + cluster_mode != LOCKSTEP_MODE)) {
>>>> + dev_err(dev, "Invalid cluster mode\n");
>>>> + return -EINVAL;
>>>> + }
>>>> +
>>>> + /*
>>>> + * Number of cores is decided by number of child nodes of
>>>> + * r5f subsystem node in dts. If Split mode is used in dts
>>>> + * 2 child nodes are expected.
>>>> + * In lockstep mode if two child nodes are available,
>>>> + * only use first child node and consider it as core0
>>>> + * and ignore core1 dt node.
>>>> + */
>>>> + core_count = of_get_available_child_count(dev_node);
>>>> + if (core_count <= 0) {
>>>> + dev_err(dev, "Invalid number of r5 cores %d", core_count);
>>>> + return -EINVAL;
>>>> + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
>>>> + dev_err(dev, "Invalid number of r5 cores for split mode\n");
>>>> + return -EINVAL;
>>>> + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
>>>> + dev_warn(dev, "Only r5 core0 will be used\n");
>>>> + core_count = 1;
>>>> + }
>>>> +
>>>> + r5_cores = devm_kcalloc(dev, core_count,
>>>> + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
>>>> + if (!r5_cores)
>>>> + return -ENOMEM;
>>>> +
>>>> + i = 0;
>>>> + for_each_available_child_of_node(dev_node, child) {
>>>> + child_pdev = of_find_device_by_node(child);
>>>> + if (!child_pdev) {
>>>> + of_node_put(child);
>>>> + return -ENODEV;
>>>> + }
>>>> +
>>>> + /* create and add remoteproc instance of type struct rproc */
>>>> + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
>>>> + r5_cores[i]->dev = &child_pdev->dev;
>>>> + if (!r5_cores[i]->dev) {
>>>> + dev_err(dev, "can't get device for r5 core %d\n", i);
>>>> + of_node_put(child);
>>>> + return -ENODEV;
>>>> + }
>>>> +
>>>> + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
>>>> + if (!r5_cores[i]->np) {
>>>> + dev_err(dev, "can't get device node for r5 core %d\n", i);
>>>> + of_node_put(child);
>>>> + return -ENODEV;
>>>> + }
>>>> +
>>>> + i++;
>>>> +
>>>> + /*
>>>> + * If two child nodes are available in dts in lockstep mode,
>>>> + * then ignore second child node.
>>>> + */
>>>> + if (i == core_count) {
>>>> + of_node_put(child);
>>>> + break;
>>>> + }
>>>> + of_node_put(child);
>>>> + }
>>>> +
>>>> + cluster->mode = cluster_mode;
>>>> + cluster->core_count = core_count;
>>>> + cluster->r5_cores = r5_cores;
>>>> +
>>>> + ret = zynqmp_r5_core_init(cluster);
>>>> + if (ret < 0) {
>>>> + dev_err(dev, "failed to init r5 core err %d\n", ret);
>>>> + return ret;
>>>> + }
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +static void zynqmp_r5_cluster_exit(void *data)
>>>> +{
>>>> + struct platform_device *pdev = (struct platform_device *)data;
>>>> +
>>>> + platform_set_drvdata(pdev, NULL);
>>>> +
>>>> + pr_info("Exit r5f subsystem driver\n");
>>>> +}
>>>> +
>>>> +/*
>>>> + * zynqmp_r5_remoteproc_probe()
>>>> + *
>>>> + * @pdev: domain platform device for R5 cluster
>>>> + *
>>>> + * called when driver is probed, for each R5 core specified in DT,
>>>> + * setup as needed to do remoteproc-related operations
>>>> + *
>>>> + * Return: 0 for success, negative value for failure.
>>>> + */
>>>> +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
>>>> +{
>>>> + int ret;
>>>> + struct zynqmp_r5_cluster *cluster;
>>>> + struct device *dev = &pdev->dev;
>>>> +
>>>> + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
>>>> + if (!cluster)
>>>> + return -ENOMEM;
>>>> +
>>>> + cluster->dev = dev;
>>>> +
>>>> + ret = devm_of_platform_populate(dev);
>>>> + if (ret) {
>>>> + dev_err(dev, "failed to populate platform dev %d\n", ret);
>>>> + return ret;
>>>> + }
>>>> +
>>>> + /* wire in so each core can be cleaned up at driver remove */
>>>> + platform_set_drvdata(pdev, cluster);
>>>> +
>>>> + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
>>>> + if (ret)
>>>> + return ret;
>>>> +
>>>> + ret = zynqmp_r5_cluster_init(cluster);
>>>> + if (ret) {
>>>> + dev_err(dev, "Invalid r5f subsystem device tree\n");
>>>> + return ret;
>>>> + }
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +/* Match table for OF platform binding */
>>>> +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
>>>> + { .compatible = "xlnx,zynqmp-r5fss", },
>>>> + { /* end of list */ },
>>>> +};
>>>> +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
>>>> +
>>>> +static struct platform_driver zynqmp_r5_remoteproc_driver = {
>>>> + .probe = zynqmp_r5_remoteproc_probe,
>>>> + .driver = {
>>>> + .name = "zynqmp_r5_remoteproc",
>>>> + .of_match_table = zynqmp_r5_remoteproc_match,
>>>> + },
>>>> +};
>>>> +module_platform_driver(zynqmp_r5_remoteproc_driver);
>>>> +
>>>> +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
>>>> +MODULE_AUTHOR("Xilinx Inc.");
>>>> +MODULE_LICENSE("GPL v2");
>>>> --
>>>> 2.25.1
>>>>
On 2/21/22 6:13 PM, Tanmay Shah wrote:
>
> On 2/18/22 11:11 AM, Mathieu Poirier wrote:
>> On Thu, Feb 10, 2022 at 03:28:24AM -0800, Tanmay Shah wrote:
>>> This driver enables r5f dual core Real time Processing Unit subsystem
>>> available on Xilinx Zynq Ultrascale MPSoC Platform. RPU subsystem
>>> (cluster) can be configured in different modes e.g. split mode in which
>>> two r5f cores work independent of each other and lock-step mode in
>>> which
>>> both r5f cores execute same code clock-for-clock and notify if the
>>> result is different.
>>>
>>> The Xilinx r5 Remoteproc Driver boots the RPU cores via calls to the
>>> Xilinx
>>> Platform Management Unit that handles the R5 configuration, memory
>>> access
>>> and R5 lifecycle management. The interface to this manager is done
>>> in this
>>> driver via zynqmp_pm_* function calls.
>>>
>>> Signed-off-by: Ben Levinsky<[email protected]>
>>> Signed-off-by: Tanmay Shah<[email protected]>
>>> ---
[ .... ]
>>> +
>>> +/*
>>> + * zynqmp_r5_rproc_mem_map
>>> + * @rproc: single R5 core's corresponding rproc instance
>>> + * @mem: mem entry to map
>>> + *
>>> + * Callback to map va for memory-region's carveout.
>>> + *
>>> + * return 0 on success, otherwise non-zero value on failure
>>> + */
>>> +static int zynqmp_r5_rproc_mem_map(struct rproc *rproc,
>>> + struct rproc_mem_entry *mem)
>>> +{
>>> + void __iomem *va;
>>> +
>>> + va = ioremap_wc(mem->dma, mem->len);
>>> + if (IS_ERR_OR_NULL(va))
>>> + return -ENOMEM;
>>> +
>>> + mem->va = (void *)va;
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +/*
>>> + * zynqmp_r5_rproc_mem_unmap
>>> + * @rproc: single R5 core's corresponding rproc instance
>>> + * @mem: mem entry to unmap
>>> + *
>>> + * Unmap memory-region carveout
>>> + *
>>> + * return 0 on success, otherwise non-zero value on failure
>>> + */
>>> +static int zynqmp_r5_rproc_mem_unmap(struct rproc *rproc,
>>> + struct rproc_mem_entry *mem)
>>> +{
>>> + iounmap((void __iomem *)mem->va);
>>> + return 0;
>>> +}
>>> +
>>> +/*
>>> + * add_mem_regions
>>> + * @r5_core: single R5 core's corresponding zynqmp_r5_core type
>>> instance
>>> + * @rmem: reserved mem region parsed from dt node
>>> + *
>>> + * Construct rproc mem carveouts from carveout provided in
>>> + * memory-region property
>>> + *
>>> + * return 0 on success, otherwise non-zero value on failure
>>> + */
>>> +static int add_mem_regions_carveout(struct zynqmp_r5_core *r5_core,
>>> + struct reserved_mem *rmem)
>>> +{
>>> + struct device *dev;
>>> + struct rproc_mem_entry *mem;
>>> + struct rproc *rproc;
>>> +
>>> + rproc = r5_core->rproc;
>>> + dev = r5_core->dev;
>>> +
>>> + /* Register associated reserved memory regions */
>>> + mem = rproc_mem_entry_init(dev, NULL,
>>> + (dma_addr_t)rmem->base,
>>> + rmem->size, rmem->base,
>>> + zynqmp_r5_rproc_mem_map,
>>> + zynqmp_r5_rproc_mem_unmap,
>>> + rmem->name);
>>> + if (!mem)
>>> + return -ENOMEM;
>>> +
>>> + rproc_add_carveout(rproc, mem);
>>> +
>>> + return 0;
>>> +}
>>> +
>>>
[ ... ]
>>> +
>>> +/*
>>> + * zynqmp_r5_parse_fw()
>>> + * @rproc: single R5 core's corresponding rproc instance
>>> + * @fw: ptr to firmware to be loaded onto r5 core
>>> + *
>>> + * When loading firmware, ensure the necessary carveouts are in
>>> remoteproc
>>> + *
>>> + * return 0 on success, otherwise non-zero value on failure
>>> + */
>>> +static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct
>>> firmware *fw)
>>> +{
>>> + int ret;
>>> + struct zynqmp_r5_core *r5_core;
>>> + struct device *dev;
>>> +
>>> + r5_core = rproc->priv;
>>> +
>>> + dev = r5_core->dev;
>>> +
>>> + ret = add_tcm_banks(rproc);
>>> + if (ret) {
>>> + dev_err(dev, "failed to get TCM banks, err %d\n", ret);
>>> + return ret;
>>> + }
>>> +
>>> + ret = rproc_elf_load_rsc_table(rproc, fw);
>>> + if (ret == -EINVAL) {
>>> + /*
>>> + * resource table only required for IPC.
>>> + * if not present, this is not necessarily an error;
>>> + * for example, loading r5 hello world application
>>> + * so simply inform user and keep going.
>>> + */
>>> + dev_info(&rproc->dev, "no resource table found.\n");
>>> + ret = 0;
>>> + }
>>> + return ret;
>>> +}
>>> +
>>> +static struct rproc_ops zynqmp_r5_rproc_ops = {
>>> + .start = zynqmp_r5_rproc_start,
>>> + .stop = zynqmp_r5_rproc_stop,
>>> + .load = rproc_elf_load_segments,
>>> + .parse_fw = zynqmp_r5_parse_fw,
>>> + .find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
>>> + .sanity_check = rproc_elf_sanity_check,
>>> + .get_boot_addr = rproc_elf_get_boot_addr,
>>> +};
>>> +
>>>
[ ... ]
>>> +
>>> +/**
>>> + * zynqmp_r5_get_mem_region_node()
>>> + * parse memory-region property from dt node and add
>>> + * memory region carveouts
>>> + *
>>> + * @r5_core: pointer to zynqmp_r5_core type object
>>> + *
>>> + * Return: 0 for success and error code for failure.
>>> + */
>>> +static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core
>>> *r5_core)
>>> +{
>>> + int res_mem_count, i, ret;
>>> + struct device *dev;
>>> + struct device_node *np, *rmem_np;
>>> + struct reserved_mem *rmem;
>>> +
>>> + dev = r5_core->dev;
>>> +
>>> + np = r5_core->np;
>>> +
>>> + res_mem_count = of_property_count_elems_of_size(np,
>>> "memory-region",
>>> + sizeof(phandle));
>>> + if (res_mem_count <= 0) {
>>> + dev_warn(dev, "failed to get memory-region property %d\n",
>>> + res_mem_count);
>>> + return -EINVAL;
>>> + }
>>> +
>>> + for (i = 0; i < res_mem_count; i++) {
>>> + rmem_np = of_parse_phandle(np, "memory-region", i);
>>> + if (!rmem_np)
>>> + return -EINVAL;
>>> +
>>> + rmem = of_reserved_mem_lookup(rmem_np);
>>> + if (!rmem) {
>>> + of_node_put(rmem_np);
>>> + return -EINVAL;
>> What happens to previously allocated regions when there is a failure?
>
> Right, I need to unmap memory and use kfree(rmem) to de-allocate memory.
>
>>> + }
>>> +
>>> + ret = add_mem_regions_carveout(r5_core, rmem);
>>> + if (ret)
>>> + dev_warn(dev, "failed to get reserve mem regions %d\n",
>>> + ret);
>> Same here.
>
> Here as well. I will release reserved mem dev with
> of_reserved_mem_device_release API.
>
> One more thing. I moved add_mem_regions_carveout from parse_fw to
> zynqmp_r5_get_mem_region_node.
>
> However, I believe I should move it back to parse_fw.
>
> Following test case is failing in when add_mem_regions_carveout is not
> available in parse_fw:
>
> load_rpu_fw -> start rpu -> stop rpu -> start rpu again
>
> In above case, during mem regions are not added again.
>
> Is it fine, if I move this add_mem_regions_carveout back to
> zynqmp_r5_parse_fw ?
Hi Mathieu,
I have addressed most of the comments from this revision.
However, I wanted your opinion on this.
I think mapping memory-regions during parse_fw just like tcm mapping is
easily
maintainable rather than mapping during probe. Also in my case, I am not
able to start RPU again because memory regions are unmapped
during RPU stop and can't be mapped again. So, is it ok if I move back
add_mem_regions_carveout from driver probe to parse_fw?
Thanks,
Tanmay
>> I am out of time for this set. Please address comments provided up
>> to here and
>> we will see about the rest in a future revision.
>>
>> Thanks,
>> Mathieu
>>
>>> +
>>> + of_node_put(rmem_np);
>>> + }
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +/*
>>> + * zynqmp_r5_core_init()
>>> + * Create and initialize zynqmp_r5_core type object
>>> + *
>>> + * @cluster: pointer to zynqmp_r5_cluster type object
>>> + *
>>> + * Return: 0 for success and error code for failure.
>>> + */
>>> +static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
>>> +{
>>> + int ret, i;
>>> + struct zynqmp_r5_core *r5_core;
>>> + struct device *dev = cluster->dev;
>>> +
>>> + ret = zynqmp_r5_get_tcm_node(cluster);
>>> + if (ret < 0) {
>>> + dev_err(dev, "can't get tcm node, err %d\n", ret);
>>> + return ret;
>>> + }
>>> +
>>> + for (i = 0; i < cluster->core_count; i++) {
>>> + r5_core = cluster->r5_cores[i];
>>> +
>>> + ret = zynqmp_r5_get_mem_region_node(r5_core);
>>> + if (ret)
>>> + dev_warn(dev, "memory-region prop failed %d\n", ret);
>>> +
>>> + /* Initialize r5 cores with power-domains parsed from dts */
>>> + ret = of_property_read_u32_index(r5_core->np, "power-domains",
>>> + 1, &r5_core->pm_domain_id);
>>> + if (ret) {
>>> + dev_err(dev, "failed to get power-domains property\n");
>>> + return ret;
>>> + }
>>> +
>>> + ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
>>> + if (ret) {
>>> + dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
>>> + cluster->mode, ret);
>>> + return ret;
>>> + }
>>> + }
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +/*
>>> + * zynqmp_r5_cluster_init()
>>> + * Create and initialize zynqmp_r5_cluster type object
>>> + *
>>> + * @cluster: pointer to zynqmp_r5_cluster type object
>>> + *
>>> + * Return: 0 for success and error code for failure.
>>> + */
>>> +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
>>> +{
>>> + struct device *dev = cluster->dev;
>>> + struct device_node *dev_node = dev_of_node(dev);
>>> + struct device_node *child;
>>> + struct platform_device *child_pdev;
>>> + int core_count = 0, ret, i;
>>> + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
>>> + struct zynqmp_r5_core **r5_cores;
>>> +
>>> + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode",
>>> &cluster_mode);
>>> +
>>> + /*
>>> + * on success returns 0, if not defined then returns -EINVAL,
>>> + * In that case, default is LOCKSTEP mode
>>> + */
>>> + if (ret != -EINVAL && ret != 0) {
>>> + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
>>> + return -EINVAL;
>>> + }
>>> +
>>> + /*
>>> + * For now driver only supports split mode and lockstep mode.
>>> + * fail driver probe if either of that is not set in dts
>>> + */
>>> + if (cluster_mode == SINGLE_CPU_MODE) {
>>> + dev_err(dev, "driver does not support single cpu mode\n");
>>> + return -EINVAL;
>>> + } else if ((cluster_mode != SPLIT_MODE &&
>>> + cluster_mode != LOCKSTEP_MODE)) {
>>> + dev_err(dev, "Invalid cluster mode\n");
>>> + return -EINVAL;
>>> + }
>>> +
>>> + /*
>>> + * Number of cores is decided by number of child nodes of
>>> + * r5f subsystem node in dts. If Split mode is used in dts
>>> + * 2 child nodes are expected.
>>> + * In lockstep mode if two child nodes are available,
>>> + * only use first child node and consider it as core0
>>> + * and ignore core1 dt node.
>>> + */
>>> + core_count = of_get_available_child_count(dev_node);
>>> + if (core_count <= 0) {
>>> + dev_err(dev, "Invalid number of r5 cores %d", core_count);
>>> + return -EINVAL;
>>> + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
>>> + dev_err(dev, "Invalid number of r5 cores for split mode\n");
>>> + return -EINVAL;
>>> + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
>>> + dev_warn(dev, "Only r5 core0 will be used\n");
>>> + core_count = 1;
>>> + }
>>> +
>>> + r5_cores = devm_kcalloc(dev, core_count,
>>> + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
>>> + if (!r5_cores)
>>> + return -ENOMEM;
>>> +
>>> + i = 0;
>>> + for_each_available_child_of_node(dev_node, child) {
>>> + child_pdev = of_find_device_by_node(child);
>>> + if (!child_pdev) {
>>> + of_node_put(child);
>>> + return -ENODEV;
>>> + }
>>> +
>>> + /* create and add remoteproc instance of type struct rproc */
>>> + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
>>> + r5_cores[i]->dev = &child_pdev->dev;
>>> + if (!r5_cores[i]->dev) {
>>> + dev_err(dev, "can't get device for r5 core %d\n", i);
>>> + of_node_put(child);
>>> + return -ENODEV;
>>> + }
>>> +
>>> + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
>>> + if (!r5_cores[i]->np) {
>>> + dev_err(dev, "can't get device node for r5 core %d\n", i);
>>> + of_node_put(child);
>>> + return -ENODEV;
>>> + }
>>> +
>>> + i++;
>>> +
>>> + /*
>>> + * If two child nodes are available in dts in lockstep mode,
>>> + * then ignore second child node.
>>> + */
>>> + if (i == core_count) {
>>> + of_node_put(child);
>>> + break;
>>> + }
>>> + of_node_put(child);
>>> + }
>>> +
>>> + cluster->mode = cluster_mode;
>>> + cluster->core_count = core_count;
>>> + cluster->r5_cores = r5_cores;
>>> +
>>> + ret = zynqmp_r5_core_init(cluster);
>>> + if (ret < 0) {
>>> + dev_err(dev, "failed to init r5 core err %d\n", ret);
>>> + return ret;
>>> + }
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +static void zynqmp_r5_cluster_exit(void *data)
>>> +{
>>> + struct platform_device *pdev = (struct platform_device *)data;
>>> +
>>> + platform_set_drvdata(pdev, NULL);
>>> +
>>> + pr_info("Exit r5f subsystem driver\n");
>>> +}
>>> +
>>> +/*
>>> + * zynqmp_r5_remoteproc_probe()
>>> + *
>>> + * @pdev: domain platform device for R5 cluster
>>> + *
>>> + * called when driver is probed, for each R5 core specified in DT,
>>> + * setup as needed to do remoteproc-related operations
>>> + *
>>> + * Return: 0 for success, negative value for failure.
>>> + */
>>> +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
>>> +{
>>> + int ret;
>>> + struct zynqmp_r5_cluster *cluster;
>>> + struct device *dev = &pdev->dev;
>>> +
>>> + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
>>> + if (!cluster)
>>> + return -ENOMEM;
>>> +
>>> + cluster->dev = dev;
>>> +
>>> + ret = devm_of_platform_populate(dev);
>>> + if (ret) {
>>> + dev_err(dev, "failed to populate platform dev %d\n", ret);
>>> + return ret;
>>> + }
>>> +
>>> + /* wire in so each core can be cleaned up at driver remove */
>>> + platform_set_drvdata(pdev, cluster);
>>> +
>>> + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
>>> + if (ret)
>>> + return ret;
>>> +
>>> + ret = zynqmp_r5_cluster_init(cluster);
>>> + if (ret) {
>>> + dev_err(dev, "Invalid r5f subsystem device tree\n");
>>> + return ret;
>>> + }
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +/* Match table for OF platform binding */
>>> +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
>>> + { .compatible = "xlnx,zynqmp-r5fss", },
>>> + { /* end of list */ },
>>> +};
>>> +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
>>> +
>>> +static struct platform_driver zynqmp_r5_remoteproc_driver = {
>>> + .probe = zynqmp_r5_remoteproc_probe,
>>> + .driver = {
>>> + .name = "zynqmp_r5_remoteproc",
>>> + .of_match_table = zynqmp_r5_remoteproc_match,
>>> + },
>>> +};
>>> +module_platform_driver(zynqmp_r5_remoteproc_driver);
>>> +
>>> +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
>>> +MODULE_AUTHOR("Xilinx Inc.");
>>> +MODULE_LICENSE("GPL v2");
>>> --
>>> 2.25.1
>>>
On Wed, 9 Mar 2022 at 00:18, Tanmay Shah <[email protected]> wrote:
>
>
> On 2/21/22 6:13 PM, Tanmay Shah wrote:
> >
> > On 2/18/22 11:11 AM, Mathieu Poirier wrote:
> >> On Thu, Feb 10, 2022 at 03:28:24AM -0800, Tanmay Shah wrote:
> >>> This driver enables r5f dual core Real time Processing Unit subsystem
> >>> available on Xilinx Zynq Ultrascale MPSoC Platform. RPU subsystem
> >>> (cluster) can be configured in different modes e.g. split mode in which
> >>> two r5f cores work independent of each other and lock-step mode in
> >>> which
> >>> both r5f cores execute same code clock-for-clock and notify if the
> >>> result is different.
> >>>
> >>> The Xilinx r5 Remoteproc Driver boots the RPU cores via calls to the
> >>> Xilinx
> >>> Platform Management Unit that handles the R5 configuration, memory
> >>> access
> >>> and R5 lifecycle management. The interface to this manager is done
> >>> in this
> >>> driver via zynqmp_pm_* function calls.
> >>>
> >>> Signed-off-by: Ben Levinsky<[email protected]>
> >>> Signed-off-by: Tanmay Shah<[email protected]>
> >>> ---
>
> [ .... ]
>
>
> >>> +
> >>> +/*
> >>> + * zynqmp_r5_rproc_mem_map
> >>> + * @rproc: single R5 core's corresponding rproc instance
> >>> + * @mem: mem entry to map
> >>> + *
> >>> + * Callback to map va for memory-region's carveout.
> >>> + *
> >>> + * return 0 on success, otherwise non-zero value on failure
> >>> + */
> >>> +static int zynqmp_r5_rproc_mem_map(struct rproc *rproc,
> >>> + struct rproc_mem_entry *mem)
> >>> +{
> >>> + void __iomem *va;
> >>> +
> >>> + va = ioremap_wc(mem->dma, mem->len);
> >>> + if (IS_ERR_OR_NULL(va))
> >>> + return -ENOMEM;
> >>> +
> >>> + mem->va = (void *)va;
> >>> +
> >>> + return 0;
> >>> +}
> >>> +
> >>> +/*
> >>> + * zynqmp_r5_rproc_mem_unmap
> >>> + * @rproc: single R5 core's corresponding rproc instance
> >>> + * @mem: mem entry to unmap
> >>> + *
> >>> + * Unmap memory-region carveout
> >>> + *
> >>> + * return 0 on success, otherwise non-zero value on failure
> >>> + */
> >>> +static int zynqmp_r5_rproc_mem_unmap(struct rproc *rproc,
> >>> + struct rproc_mem_entry *mem)
> >>> +{
> >>> + iounmap((void __iomem *)mem->va);
> >>> + return 0;
> >>> +}
> >>> +
> >>> +/*
> >>> + * add_mem_regions
> >>> + * @r5_core: single R5 core's corresponding zynqmp_r5_core type
> >>> instance
> >>> + * @rmem: reserved mem region parsed from dt node
> >>> + *
> >>> + * Construct rproc mem carveouts from carveout provided in
> >>> + * memory-region property
> >>> + *
> >>> + * return 0 on success, otherwise non-zero value on failure
> >>> + */
> >>> +static int add_mem_regions_carveout(struct zynqmp_r5_core *r5_core,
> >>> + struct reserved_mem *rmem)
> >>> +{
> >>> + struct device *dev;
> >>> + struct rproc_mem_entry *mem;
> >>> + struct rproc *rproc;
> >>> +
> >>> + rproc = r5_core->rproc;
> >>> + dev = r5_core->dev;
> >>> +
> >>> + /* Register associated reserved memory regions */
> >>> + mem = rproc_mem_entry_init(dev, NULL,
> >>> + (dma_addr_t)rmem->base,
> >>> + rmem->size, rmem->base,
> >>> + zynqmp_r5_rproc_mem_map,
> >>> + zynqmp_r5_rproc_mem_unmap,
> >>> + rmem->name);
> >>> + if (!mem)
> >>> + return -ENOMEM;
> >>> +
> >>> + rproc_add_carveout(rproc, mem);
> >>> +
> >>> + return 0;
> >>> +}
> >>> +
> >>>
> [ ... ]
>
> >>> +
> >>> +/*
> >>> + * zynqmp_r5_parse_fw()
> >>> + * @rproc: single R5 core's corresponding rproc instance
> >>> + * @fw: ptr to firmware to be loaded onto r5 core
> >>> + *
> >>> + * When loading firmware, ensure the necessary carveouts are in
> >>> remoteproc
> >>> + *
> >>> + * return 0 on success, otherwise non-zero value on failure
> >>> + */
> >>> +static int zynqmp_r5_parse_fw(struct rproc *rproc, const struct
> >>> firmware *fw)
> >>> +{
> >>> + int ret;
> >>> + struct zynqmp_r5_core *r5_core;
> >>> + struct device *dev;
> >>> +
> >>> + r5_core = rproc->priv;
> >>> +
> >>> + dev = r5_core->dev;
> >>> +
> >>> + ret = add_tcm_banks(rproc);
> >>> + if (ret) {
> >>> + dev_err(dev, "failed to get TCM banks, err %d\n", ret);
> >>> + return ret;
> >>> + }
> >>> +
> >>> + ret = rproc_elf_load_rsc_table(rproc, fw);
> >>> + if (ret == -EINVAL) {
> >>> + /*
> >>> + * resource table only required for IPC.
> >>> + * if not present, this is not necessarily an error;
> >>> + * for example, loading r5 hello world application
> >>> + * so simply inform user and keep going.
> >>> + */
> >>> + dev_info(&rproc->dev, "no resource table found.\n");
> >>> + ret = 0;
> >>> + }
> >>> + return ret;
> >>> +}
> >>> +
> >>> +static struct rproc_ops zynqmp_r5_rproc_ops = {
> >>> + .start = zynqmp_r5_rproc_start,
> >>> + .stop = zynqmp_r5_rproc_stop,
> >>> + .load = rproc_elf_load_segments,
> >>> + .parse_fw = zynqmp_r5_parse_fw,
> >>> + .find_loaded_rsc_table = rproc_elf_find_loaded_rsc_table,
> >>> + .sanity_check = rproc_elf_sanity_check,
> >>> + .get_boot_addr = rproc_elf_get_boot_addr,
> >>> +};
> >>> +
> >>>
> [ ... ]
>
>
> >>> +
> >>> +/**
> >>> + * zynqmp_r5_get_mem_region_node()
> >>> + * parse memory-region property from dt node and add
> >>> + * memory region carveouts
> >>> + *
> >>> + * @r5_core: pointer to zynqmp_r5_core type object
> >>> + *
> >>> + * Return: 0 for success and error code for failure.
> >>> + */
> >>> +static int zynqmp_r5_get_mem_region_node(struct zynqmp_r5_core
> >>> *r5_core)
> >>> +{
> >>> + int res_mem_count, i, ret;
> >>> + struct device *dev;
> >>> + struct device_node *np, *rmem_np;
> >>> + struct reserved_mem *rmem;
> >>> +
> >>> + dev = r5_core->dev;
> >>> +
> >>> + np = r5_core->np;
> >>> +
> >>> + res_mem_count = of_property_count_elems_of_size(np,
> >>> "memory-region",
> >>> + sizeof(phandle));
> >>> + if (res_mem_count <= 0) {
> >>> + dev_warn(dev, "failed to get memory-region property %d\n",
> >>> + res_mem_count);
> >>> + return -EINVAL;
> >>> + }
> >>> +
> >>> + for (i = 0; i < res_mem_count; i++) {
> >>> + rmem_np = of_parse_phandle(np, "memory-region", i);
> >>> + if (!rmem_np)
> >>> + return -EINVAL;
> >>> +
> >>> + rmem = of_reserved_mem_lookup(rmem_np);
> >>> + if (!rmem) {
> >>> + of_node_put(rmem_np);
> >>> + return -EINVAL;
> >> What happens to previously allocated regions when there is a failure?
> >
> > Right, I need to unmap memory and use kfree(rmem) to de-allocate memory.
> >
> >>> + }
> >>> +
> >>> + ret = add_mem_regions_carveout(r5_core, rmem);
> >>> + if (ret)
> >>> + dev_warn(dev, "failed to get reserve mem regions %d\n",
> >>> + ret);
> >> Same here.
> >
> > Here as well. I will release reserved mem dev with
> > of_reserved_mem_device_release API.
> >
> > One more thing. I moved add_mem_regions_carveout from parse_fw to
> > zynqmp_r5_get_mem_region_node.
> >
> > However, I believe I should move it back to parse_fw.
> >
> > Following test case is failing in when add_mem_regions_carveout is not
> > available in parse_fw:
> >
> > load_rpu_fw -> start rpu -> stop rpu -> start rpu again
> >
> > In above case, during mem regions are not added again.
> >
> > Is it fine, if I move this add_mem_regions_carveout back to
> > zynqmp_r5_parse_fw ?
>
>
> Hi Mathieu,
>
> I have addressed most of the comments from this revision.
>
> However, I wanted your opinion on this.
>
> I think mapping memory-regions during parse_fw just like tcm mapping is
> easily
>
> maintainable rather than mapping during probe. Also in my case, I am not
> able to start RPU again because memory regions are unmapped
>
> during RPU stop and can't be mapped again. So, is it ok if I move back
> add_mem_regions_carveout from driver probe to parse_fw?
If you intend to support attaching to the remote processor at some
point in the future then I suggest adding the carveouts in
ops::prepare(), otherwise ops::parse_fw() is fine.
>
> Thanks,
>
> Tanmay
>
>
> >> I am out of time for this set. Please address comments provided up
> >> to here and
> >> we will see about the rest in a future revision.
> >>
> >> Thanks,
> >> Mathieu
> >>
> >>> +
> >>> + of_node_put(rmem_np);
> >>> + }
> >>> +
> >>> + return 0;
> >>> +}
> >>> +
> >>> +/*
> >>> + * zynqmp_r5_core_init()
> >>> + * Create and initialize zynqmp_r5_core type object
> >>> + *
> >>> + * @cluster: pointer to zynqmp_r5_cluster type object
> >>> + *
> >>> + * Return: 0 for success and error code for failure.
> >>> + */
> >>> +static int zynqmp_r5_core_init(struct zynqmp_r5_cluster *cluster)
> >>> +{
> >>> + int ret, i;
> >>> + struct zynqmp_r5_core *r5_core;
> >>> + struct device *dev = cluster->dev;
> >>> +
> >>> + ret = zynqmp_r5_get_tcm_node(cluster);
> >>> + if (ret < 0) {
> >>> + dev_err(dev, "can't get tcm node, err %d\n", ret);
> >>> + return ret;
> >>> + }
> >>> +
> >>> + for (i = 0; i < cluster->core_count; i++) {
> >>> + r5_core = cluster->r5_cores[i];
> >>> +
> >>> + ret = zynqmp_r5_get_mem_region_node(r5_core);
> >>> + if (ret)
> >>> + dev_warn(dev, "memory-region prop failed %d\n", ret);
> >>> +
> >>> + /* Initialize r5 cores with power-domains parsed from dts */
> >>> + ret = of_property_read_u32_index(r5_core->np, "power-domains",
> >>> + 1, &r5_core->pm_domain_id);
> >>> + if (ret) {
> >>> + dev_err(dev, "failed to get power-domains property\n");
> >>> + return ret;
> >>> + }
> >>> +
> >>> + ret = zynqmp_r5_set_mode(r5_core, cluster->mode);
> >>> + if (ret) {
> >>> + dev_err(dev, "failed to set r5 cluster mode %d, err %d\n",
> >>> + cluster->mode, ret);
> >>> + return ret;
> >>> + }
> >>> + }
> >>> +
> >>> + return 0;
> >>> +}
> >>> +
> >>> +/*
> >>> + * zynqmp_r5_cluster_init()
> >>> + * Create and initialize zynqmp_r5_cluster type object
> >>> + *
> >>> + * @cluster: pointer to zynqmp_r5_cluster type object
> >>> + *
> >>> + * Return: 0 for success and error code for failure.
> >>> + */
> >>> +static int zynqmp_r5_cluster_init(struct zynqmp_r5_cluster *cluster)
> >>> +{
> >>> + struct device *dev = cluster->dev;
> >>> + struct device_node *dev_node = dev_of_node(dev);
> >>> + struct device_node *child;
> >>> + struct platform_device *child_pdev;
> >>> + int core_count = 0, ret, i;
> >>> + enum zynqmp_r5_cluster_mode cluster_mode = LOCKSTEP_MODE;
> >>> + struct zynqmp_r5_core **r5_cores;
> >>> +
> >>> + ret = of_property_read_u32(dev_node, "xlnx,cluster-mode",
> >>> &cluster_mode);
> >>> +
> >>> + /*
> >>> + * on success returns 0, if not defined then returns -EINVAL,
> >>> + * In that case, default is LOCKSTEP mode
> >>> + */
> >>> + if (ret != -EINVAL && ret != 0) {
> >>> + dev_err(dev, "Invalid xlnx,cluster-mode property\n");
> >>> + return -EINVAL;
> >>> + }
> >>> +
> >>> + /*
> >>> + * For now driver only supports split mode and lockstep mode.
> >>> + * fail driver probe if either of that is not set in dts
> >>> + */
> >>> + if (cluster_mode == SINGLE_CPU_MODE) {
> >>> + dev_err(dev, "driver does not support single cpu mode\n");
> >>> + return -EINVAL;
> >>> + } else if ((cluster_mode != SPLIT_MODE &&
> >>> + cluster_mode != LOCKSTEP_MODE)) {
> >>> + dev_err(dev, "Invalid cluster mode\n");
> >>> + return -EINVAL;
> >>> + }
> >>> +
> >>> + /*
> >>> + * Number of cores is decided by number of child nodes of
> >>> + * r5f subsystem node in dts. If Split mode is used in dts
> >>> + * 2 child nodes are expected.
> >>> + * In lockstep mode if two child nodes are available,
> >>> + * only use first child node and consider it as core0
> >>> + * and ignore core1 dt node.
> >>> + */
> >>> + core_count = of_get_available_child_count(dev_node);
> >>> + if (core_count <= 0) {
> >>> + dev_err(dev, "Invalid number of r5 cores %d", core_count);
> >>> + return -EINVAL;
> >>> + } else if (cluster_mode == SPLIT_MODE && core_count != 2) {
> >>> + dev_err(dev, "Invalid number of r5 cores for split mode\n");
> >>> + return -EINVAL;
> >>> + } else if (cluster_mode == LOCKSTEP_MODE && core_count == 2) {
> >>> + dev_warn(dev, "Only r5 core0 will be used\n");
> >>> + core_count = 1;
> >>> + }
> >>> +
> >>> + r5_cores = devm_kcalloc(dev, core_count,
> >>> + sizeof(struct zynqmp_r5_core *), GFP_KERNEL);
> >>> + if (!r5_cores)
> >>> + return -ENOMEM;
> >>> +
> >>> + i = 0;
> >>> + for_each_available_child_of_node(dev_node, child) {
> >>> + child_pdev = of_find_device_by_node(child);
> >>> + if (!child_pdev) {
> >>> + of_node_put(child);
> >>> + return -ENODEV;
> >>> + }
> >>> +
> >>> + /* create and add remoteproc instance of type struct rproc */
> >>> + r5_cores[i] = zynqmp_r5_add_rproc_core(&child_pdev->dev);
> >>> + r5_cores[i]->dev = &child_pdev->dev;
> >>> + if (!r5_cores[i]->dev) {
> >>> + dev_err(dev, "can't get device for r5 core %d\n", i);
> >>> + of_node_put(child);
> >>> + return -ENODEV;
> >>> + }
> >>> +
> >>> + r5_cores[i]->np = dev_of_node(r5_cores[i]->dev);
> >>> + if (!r5_cores[i]->np) {
> >>> + dev_err(dev, "can't get device node for r5 core %d\n", i);
> >>> + of_node_put(child);
> >>> + return -ENODEV;
> >>> + }
> >>> +
> >>> + i++;
> >>> +
> >>> + /*
> >>> + * If two child nodes are available in dts in lockstep mode,
> >>> + * then ignore second child node.
> >>> + */
> >>> + if (i == core_count) {
> >>> + of_node_put(child);
> >>> + break;
> >>> + }
> >>> + of_node_put(child);
> >>> + }
> >>> +
> >>> + cluster->mode = cluster_mode;
> >>> + cluster->core_count = core_count;
> >>> + cluster->r5_cores = r5_cores;
> >>> +
> >>> + ret = zynqmp_r5_core_init(cluster);
> >>> + if (ret < 0) {
> >>> + dev_err(dev, "failed to init r5 core err %d\n", ret);
> >>> + return ret;
> >>> + }
> >>> +
> >>> + return 0;
> >>> +}
> >>> +
> >>> +static void zynqmp_r5_cluster_exit(void *data)
> >>> +{
> >>> + struct platform_device *pdev = (struct platform_device *)data;
> >>> +
> >>> + platform_set_drvdata(pdev, NULL);
> >>> +
> >>> + pr_info("Exit r5f subsystem driver\n");
> >>> +}
> >>> +
> >>> +/*
> >>> + * zynqmp_r5_remoteproc_probe()
> >>> + *
> >>> + * @pdev: domain platform device for R5 cluster
> >>> + *
> >>> + * called when driver is probed, for each R5 core specified in DT,
> >>> + * setup as needed to do remoteproc-related operations
> >>> + *
> >>> + * Return: 0 for success, negative value for failure.
> >>> + */
> >>> +static int zynqmp_r5_remoteproc_probe(struct platform_device *pdev)
> >>> +{
> >>> + int ret;
> >>> + struct zynqmp_r5_cluster *cluster;
> >>> + struct device *dev = &pdev->dev;
> >>> +
> >>> + cluster = devm_kzalloc(dev, sizeof(*cluster), GFP_KERNEL);
> >>> + if (!cluster)
> >>> + return -ENOMEM;
> >>> +
> >>> + cluster->dev = dev;
> >>> +
> >>> + ret = devm_of_platform_populate(dev);
> >>> + if (ret) {
> >>> + dev_err(dev, "failed to populate platform dev %d\n", ret);
> >>> + return ret;
> >>> + }
> >>> +
> >>> + /* wire in so each core can be cleaned up at driver remove */
> >>> + platform_set_drvdata(pdev, cluster);
> >>> +
> >>> + ret = devm_add_action_or_reset(dev, zynqmp_r5_cluster_exit, pdev);
> >>> + if (ret)
> >>> + return ret;
> >>> +
> >>> + ret = zynqmp_r5_cluster_init(cluster);
> >>> + if (ret) {
> >>> + dev_err(dev, "Invalid r5f subsystem device tree\n");
> >>> + return ret;
> >>> + }
> >>> +
> >>> + return 0;
> >>> +}
> >>> +
> >>> +/* Match table for OF platform binding */
> >>> +static const struct of_device_id zynqmp_r5_remoteproc_match[] = {
> >>> + { .compatible = "xlnx,zynqmp-r5fss", },
> >>> + { /* end of list */ },
> >>> +};
> >>> +MODULE_DEVICE_TABLE(of, zynqmp_r5_remoteproc_match);
> >>> +
> >>> +static struct platform_driver zynqmp_r5_remoteproc_driver = {
> >>> + .probe = zynqmp_r5_remoteproc_probe,
> >>> + .driver = {
> >>> + .name = "zynqmp_r5_remoteproc",
> >>> + .of_match_table = zynqmp_r5_remoteproc_match,
> >>> + },
> >>> +};
> >>> +module_platform_driver(zynqmp_r5_remoteproc_driver);
> >>> +
> >>> +MODULE_DESCRIPTION("Xilinx R5F remote processor driver");
> >>> +MODULE_AUTHOR("Xilinx Inc.");
> >>> +MODULE_LICENSE("GPL v2");
> >>> --
> >>> 2.25.1
> >>>