This patch series introduce support for _CCA object, which is currently
used mainly by ARM64 platform to specify DMA coherency attribute for
devices when booting with ACPI.
A copy of ACPIv6 can be found here:
http://www.uefi.org/sites/default/files/resources/ACPI_6.0.pdf
This patch also introduces 2 new APIS:
1. acpi_dma_is_coherent() as part of ACPI API.
2. device_dma_is_coherent() as part of unified device property API.
This simplifies the logic in device drivers to determine device coherency
attribute regardless of booting with DT vs ACPI.
This has been tested on AMD-Seattle platform, which implements _CCA
object as described in the AMD Opteron A1100 Series Processor ACPI Porting Guide:
http://amd-dev.wpengine.netdna-cdn.com/wordpress/media/2012/10/Seattle_ACPI_Guide.pdf
Changes from V1 (https://lkml.org/lkml/2015/4/29/290):
* Remove supports for 32-bit ARM since doesn't currently
supporting ACPI (Per Catalin suggestions.)
* Do not call arch_setup_dma_ops() and when _CCA is missing.
(per Arnd suggestion)
* Add CONFIG_ACPI_SUPPORT_CCA_ZERO kernel config flag to
allow architectures to specify the behavior when _CCA=0.
* Add dummy_dma_ops for ARM64 (per Catalin suggestions).
* Fixed build error when ACPI is not configured by defining
acpi_dma_is_coherent() for when CONFIG_ACPI is not set.
* Introduce device_dma_is_coherent().
* Use device_dma_is_coherent in crypto/ccp and amd-xgbe driver.
Changes from RFC: (https://lkml.org/lkml/2015/4/1/389)
* New logic for deriving and propagating coherent attribute from
parent devices. (by Mark)
* Introducing acpi_dma_is_coherent() API (Per Tom suggestion)
* Introducing CONFIG_ACPI_MUST_HAVE_CCA kernel configuration.
* Rebased to linux-4.1-rc1
Suravee Suthikulpanit (5):
ACPI / scan: Parse _CCA and setup device coherency
arm64 : Introduce support for ACPI _CCA object
device property: Introduces device_dma_is_coherent()
crypto: ccp - Unify coherency checking logic with
device_dma_is_coherent()
amd-xgbe: Unify coherency checking logic with device_dma_is_coherent()
arch/arm64/Kconfig | 2 +
arch/arm64/include/asm/dma-mapping.h | 18 +++++-
arch/arm64/mm/dma-mapping.c | 104 ++++++++++++++++++++++++++++++
drivers/acpi/Kconfig | 6 ++
drivers/acpi/acpi_platform.c | 4 +-
drivers/acpi/scan.c | 62 ++++++++++++++++++
drivers/base/property.c | 12 ++++
drivers/crypto/ccp/ccp-platform.c | 60 +----------------
drivers/net/ethernet/amd/xgbe/xgbe-main.c | 27 +-------
include/acpi/acpi_bus.h | 11 +++-
include/linux/acpi.h | 5 ++
include/linux/property.h | 2 +
12 files changed, 224 insertions(+), 89 deletions(-)
--
2.1.0
This patch implements support for ACPI _CCA object, which is introduced in
ACPIv5.1, can be used for specifying device DMA coherency attribute.
The parsing logic traverses device namespace to parse coherency
information, and stores it in acpi_device_flags. Then uses it to call
arch_setup_dma_ops() when creating each device enumerated in DSDT
during ACPI scan.
This patch also introduces acpi_dma_is_coherent(), which provides
an interface for device drivers to check the coherency information
similarly to the of_dma_is_coherent().
Signed-off-by: Mark Salter <[email protected]>
Signed-off-by: Suravee Suthikulpanit <[email protected]>
---
NOTE:
* Since there seem to be conflict opinions regarding how
architecture should handle _CCA=0. So, I am proposing the
CONFIG_ARCH_SUPPORT_CCA_ZERO, which can be specified by
for each architecture to define behavior of the ACPI
scanning code when _CCA=0. Let me know if this is acceptable.
drivers/acpi/Kconfig | 6 +++++
drivers/acpi/acpi_platform.c | 4 ++-
drivers/acpi/scan.c | 62 ++++++++++++++++++++++++++++++++++++++++++++
include/acpi/acpi_bus.h | 11 +++++++-
include/linux/acpi.h | 5 ++++
5 files changed, 86 insertions(+), 2 deletions(-)
diff --git a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
index ab2cbb5..dd386e9 100644
--- a/drivers/acpi/Kconfig
+++ b/drivers/acpi/Kconfig
@@ -54,6 +54,12 @@ config ACPI_GENERIC_GSI
config ACPI_SYSTEM_POWER_STATES_SUPPORT
bool
+config ACPI_MUST_HAVE_CCA
+ bool
+
+config ACPI_SUPPORT_CCA_ZERO
+ bool
+
config ACPI_SLEEP
bool
depends on SUSPEND || HIBERNATION
diff --git a/drivers/acpi/acpi_platform.c b/drivers/acpi/acpi_platform.c
index 4bf7559..a6feca4 100644
--- a/drivers/acpi/acpi_platform.c
+++ b/drivers/acpi/acpi_platform.c
@@ -108,9 +108,11 @@ struct platform_device *acpi_create_platform_device(struct acpi_device *adev)
if (IS_ERR(pdev))
dev_err(&adev->dev, "platform device creation failed: %ld\n",
PTR_ERR(pdev));
- else
+ else {
+ acpi_setup_device_dma(adev, &pdev->dev);
dev_dbg(&adev->dev, "created platform device %s\n",
dev_name(&pdev->dev));
+ }
kfree(resources);
return pdev;
diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
index 849b699..ac33b29 100644
--- a/drivers/acpi/scan.c
+++ b/drivers/acpi/scan.c
@@ -11,6 +11,7 @@
#include <linux/kthread.h>
#include <linux/dmi.h>
#include <linux/nls.h>
+#include <linux/dma-mapping.h>
#include <asm/pgtable.h>
@@ -2137,6 +2138,66 @@ void acpi_free_pnp_ids(struct acpi_device_pnp *pnp)
kfree(pnp->unique_id);
}
+void acpi_setup_device_dma(struct acpi_device *adev, struct device *dev)
+{
+ int coherent = acpi_dma_is_coherent(adev);
+
+ /**
+ * Currently, we only support DMA for devices that _CCA=1
+ * since this seems to be the case on most ACPI platforms.
+ *
+ * For the case when _CCA=0 (i.e. is_coherent=0 && cca_seen=1),
+ * we would rely on arch-specific cache maintenance for
+ * non-coherence DMA operations if architecture enables
+ * CONFIG_ACPI_SUPPORT_CCA_ZERO.
+ *
+ * For the case when _CCA is missing but platform requires it
+ * (i.e. is_coherent=0 && cca_seen=0), we do not call
+ * arch_setup_dma_ops() and fallback to arch-specific default
+ * handling.
+ */
+ if (adev->flags.cca_seen) {
+ if (!coherent && !IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO))
+ return;
+ arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
+ }
+}
+
+static void acpi_init_coherency(struct acpi_device *adev)
+{
+ unsigned long long cca = 0;
+ acpi_status status;
+ struct acpi_device *parent = adev->parent;
+
+ if (parent && parent->flags.cca_seen) {
+ /*
+ * From ACPI spec, OSPM will ignore _CCA if an ancestor
+ * already saw one.
+ */
+ adev->flags.cca_seen = 1;
+ cca = acpi_dma_is_coherent(parent);
+ } else {
+ status = acpi_evaluate_integer(adev->handle, "_CCA",
+ NULL, &cca);
+ if (ACPI_SUCCESS(status)) {
+ adev->flags.cca_seen = 1;
+ } else if (!IS_ENABLED(CONFIG_ACPI_MUST_HAVE_CCA)) {
+ /*
+ * If architecture does not specify that _CCA is
+ * required for DMA-able devices (e.g. x86),
+ * we default to _CCA=1.
+ */
+ cca = 1;
+ } else {
+ dev_err(&adev->dev, FW_BUG
+ "DMA is not setup due to missing _CCA.\n");
+ }
+ }
+
+ adev->flags.is_coherent = cca;
+ acpi_setup_device_dma(adev, &adev->dev);
+}
+
void acpi_init_device_object(struct acpi_device *device, acpi_handle handle,
int type, unsigned long long sta)
{
@@ -2155,6 +2216,7 @@ void acpi_init_device_object(struct acpi_device *device, acpi_handle handle,
device->flags.visited = false;
device_initialize(&device->dev);
dev_set_uevent_suppress(&device->dev, true);
+ acpi_init_coherency(device);
}
void acpi_device_add_finalize(struct acpi_device *device)
diff --git a/include/acpi/acpi_bus.h b/include/acpi/acpi_bus.h
index 8de4fa9..b804183 100644
--- a/include/acpi/acpi_bus.h
+++ b/include/acpi/acpi_bus.h
@@ -208,7 +208,9 @@ struct acpi_device_flags {
u32 visited:1;
u32 hotplug_notify:1;
u32 is_dock_station:1;
- u32 reserved:23;
+ u32 is_coherent:1;
+ u32 cca_seen:1;
+ u32 reserved:21;
};
/* File System */
@@ -380,6 +382,13 @@ struct acpi_device {
void (*remove)(struct acpi_device *);
};
+static inline bool acpi_dma_is_coherent(struct acpi_device *adev)
+{
+ return adev && adev->flags.is_coherent;
+}
+
+void acpi_setup_device_dma(struct acpi_device *adev, struct device *dev);
+
static inline bool is_acpi_node(struct fwnode_handle *fwnode)
{
return fwnode && fwnode->type == FWNODE_ACPI;
diff --git a/include/linux/acpi.h b/include/linux/acpi.h
index b10c4a6..d14e777 100644
--- a/include/linux/acpi.h
+++ b/include/linux/acpi.h
@@ -583,6 +583,11 @@ static inline int acpi_device_modalias(struct device *dev,
return -ENODEV;
}
+static inline bool acpi_dma_is_coherent(struct acpi_device *adev)
+{
+ return false;
+}
+
#define ACPI_PTR(_ptr) (NULL)
#endif /* !CONFIG_ACPI */
--
2.1.0
>From http://www.uefi.org/sites/default/files/resources/ACPI_6.0.pdf,
section 6.2.17 _CCA states that ARM platforms require ACPI _CCA
object to be specified for DMA-cabpable devices. This patch introduces
ACPI_MUST_HAVE_CCA in arm64 Kconfig to specify such requirement.
In this case of missing _CCA, arm64 would assign dummy_dma_ops
to disable DMA capability of the device.
Signed-off-by: Mark Salter <[email protected]>
Signed-off-by: Suravee Suthikulpanit <[email protected]>
---
arch/arm64/Kconfig | 2 +
arch/arm64/include/asm/dma-mapping.h | 18 +++++-
arch/arm64/mm/dma-mapping.c | 104 +++++++++++++++++++++++++++++++++++
3 files changed, 122 insertions(+), 2 deletions(-)
diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
index 4269dba..4f4bbaaf 100644
--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig
@@ -1,7 +1,9 @@
config ARM64
def_bool y
select ACPI_GENERIC_GSI if ACPI
+ select ACPI_MUST_HAVE_CCA if ACPI
select ACPI_REDUCED_HARDWARE_ONLY if ACPI
+ select ACPI_SUPPORT_CCA_ZERO if ACPI
select ARCH_HAS_ATOMIC64_DEC_IF_POSITIVE
select ARCH_HAS_ELF_RANDOMIZE
select ARCH_HAS_GCOV_PROFILE_ALL
diff --git a/arch/arm64/include/asm/dma-mapping.h b/arch/arm64/include/asm/dma-mapping.h
index 9437e3d..f0d6d0b 100644
--- a/arch/arm64/include/asm/dma-mapping.h
+++ b/arch/arm64/include/asm/dma-mapping.h
@@ -18,6 +18,7 @@
#ifdef __KERNEL__
+#include <linux/acpi.h>
#include <linux/types.h>
#include <linux/vmalloc.h>
@@ -28,13 +29,23 @@
#define DMA_ERROR_CODE (~(dma_addr_t)0)
extern struct dma_map_ops *dma_ops;
+extern struct dma_map_ops dummy_dma_ops;
static inline struct dma_map_ops *__generic_dma_ops(struct device *dev)
{
- if (unlikely(!dev) || !dev->archdata.dma_ops)
+ if (unlikely(!dev))
return dma_ops;
- else
+ else if (dev->archdata.dma_ops)
return dev->archdata.dma_ops;
+ else if (acpi_disabled)
+ return dma_ops;
+
+ /*
+ * When ACPI is enabled, if arch_set_dma_ops is not called,
+ * we will disable device DMA capability by setting it
+ * to dummy_dma_ops.
+ */
+ return &dummy_dma_ops;
}
static inline struct dma_map_ops *get_dma_ops(struct device *dev)
@@ -48,6 +59,9 @@ static inline struct dma_map_ops *get_dma_ops(struct device *dev)
static inline void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size,
struct iommu_ops *iommu, bool coherent)
{
+ if (!acpi_disabled && !dev->archdata.dma_ops)
+ dev->archdata.dma_ops = dma_ops;
+
dev->archdata.dma_coherent = coherent;
}
#define arch_setup_dma_ops arch_setup_dma_ops
diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c
index ef7d112..31d2b52 100644
--- a/arch/arm64/mm/dma-mapping.c
+++ b/arch/arm64/mm/dma-mapping.c
@@ -415,6 +415,110 @@ out:
return -ENOMEM;
}
+/********************************************
+ * The following APIs are for dummy DMA ops *
+ ********************************************/
+
+static void *__dummy_alloc(struct device *dev, size_t size,
+ dma_addr_t *dma_handle, gfp_t flags,
+ struct dma_attrs *attrs)
+{
+ return NULL;
+}
+
+static void __dummy_free(struct device *dev, size_t size,
+ void *vaddr, dma_addr_t dma_handle,
+ struct dma_attrs *attrs)
+{
+}
+
+static int __dummy_mmap(struct device *dev,
+ struct vm_area_struct *vma,
+ void *cpu_addr, dma_addr_t dma_addr, size_t size,
+ struct dma_attrs *attrs)
+{
+ return -ENXIO;
+}
+
+static dma_addr_t __dummy_map_page(struct device *dev, struct page *page,
+ unsigned long offset, size_t size,
+ enum dma_data_direction dir,
+ struct dma_attrs *attrs)
+{
+ return DMA_ERROR_CODE;
+}
+
+static void __dummy_unmap_page(struct device *dev, dma_addr_t dev_addr,
+ size_t size, enum dma_data_direction dir,
+ struct dma_attrs *attrs)
+{
+}
+
+static int __dummy_map_sg(struct device *dev, struct scatterlist *sgl,
+ int nelems, enum dma_data_direction dir,
+ struct dma_attrs *attrs)
+{
+ return 0;
+}
+
+static void __dummy_unmap_sg(struct device *dev,
+ struct scatterlist *sgl, int nelems,
+ enum dma_data_direction dir,
+ struct dma_attrs *attrs)
+{
+}
+
+static void __dummy_sync_single_for_cpu(struct device *dev,
+ dma_addr_t dev_addr, size_t size,
+ enum dma_data_direction dir)
+{
+}
+
+static void __dummy_sync_single_for_device(struct device *dev,
+ dma_addr_t dev_addr, size_t size,
+ enum dma_data_direction dir)
+{
+}
+
+static void __dummy_sync_sg_for_cpu(struct device *dev,
+ struct scatterlist *sgl, int nelems,
+ enum dma_data_direction dir)
+{
+}
+
+static void __dummy_sync_sg_for_device(struct device *dev,
+ struct scatterlist *sgl, int nelems,
+ enum dma_data_direction dir)
+{
+}
+
+static int __dummy_mapping_error(struct device *hwdev, dma_addr_t dma_addr)
+{
+ return 1;
+}
+
+static int __dummy_dma_supported(struct device *hwdev, u64 mask)
+{
+ return 0;
+}
+
+struct dma_map_ops dummy_dma_ops = {
+ .alloc = __dummy_alloc,
+ .free = __dummy_free,
+ .mmap = __dummy_mmap,
+ .map_page = __dummy_map_page,
+ .unmap_page = __dummy_unmap_page,
+ .map_sg = __dummy_map_sg,
+ .unmap_sg = __dummy_unmap_sg,
+ .sync_single_for_cpu = __dummy_sync_single_for_cpu,
+ .sync_single_for_device = __dummy_sync_single_for_device,
+ .sync_sg_for_cpu = __dummy_sync_sg_for_cpu,
+ .sync_sg_for_device = __dummy_sync_sg_for_device,
+ .mapping_error = __dummy_mapping_error,
+ .dma_supported = __dummy_dma_supported,
+};
+EXPORT_SYMBOL(dummy_dma_ops);
+
static int __init arm64_dma_init(void)
{
int ret;
--
2.1.0
Currently, device drivers, which support both OF and ACPI,
need to call two separate APIs, of_dma_is_coherent() and
acpi_dma_is_coherent()) to determine device coherency attribute.
This patch simplifies this process by introducing a new device
property API, device_dma_is_coherent(), which calls the appropriate
interface based on the booting architecture.
Signed-off-by: Suravee Suthikulpanit <[email protected]>
---
drivers/base/property.c | 12 ++++++++++++
include/linux/property.h | 2 ++
2 files changed, 14 insertions(+)
diff --git a/drivers/base/property.c b/drivers/base/property.c
index 1d0b116..8123c6e 100644
--- a/drivers/base/property.c
+++ b/drivers/base/property.c
@@ -14,6 +14,7 @@
#include <linux/export.h>
#include <linux/kernel.h>
#include <linux/of.h>
+#include <linux/of_address.h>
#include <linux/property.h>
/**
@@ -519,3 +520,14 @@ unsigned int device_get_child_node_count(struct device *dev)
return count;
}
EXPORT_SYMBOL_GPL(device_get_child_node_count);
+
+bool device_dma_is_coherent(struct device *dev)
+{
+ if (IS_ENABLED(CONFIG_OF) && dev->of_node)
+ return of_dma_is_coherent(dev->of_node);
+ else if (has_acpi_companion(dev))
+ return acpi_dma_is_coherent(acpi_node(dev->fwnode));
+
+ return false;
+}
+EXPORT_SYMBOL_GPL(device_dma_is_coherent);
diff --git a/include/linux/property.h b/include/linux/property.h
index de8bdf4..76ebde9 100644
--- a/include/linux/property.h
+++ b/include/linux/property.h
@@ -164,4 +164,6 @@ struct property_set {
void device_add_property_set(struct device *dev, struct property_set *pset);
+bool device_dma_is_coherent(struct device *dev);
+
#endif /* _LINUX_PROPERTY_H_ */
--
2.1.0
Currently, the driver has separate logic to determine device coherency
for DT vs ACPI. This patch simplifies the code with a call to
device_dma_is_coherent().
CC: Herbert Xu <[email protected]>
CC: David S. Miller <[email protected]>
Signed-off-by: Tom Lendacky <[email protected]>
Signed-off-by: Suravee Suthikulpanit <[email protected]>
---
drivers/crypto/ccp/ccp-platform.c | 60 +--------------------------------------
1 file changed, 1 insertion(+), 59 deletions(-)
diff --git a/drivers/crypto/ccp/ccp-platform.c b/drivers/crypto/ccp/ccp-platform.c
index b1c20b2..e446781 100644
--- a/drivers/crypto/ccp/ccp-platform.c
+++ b/drivers/crypto/ccp/ccp-platform.c
@@ -90,58 +90,6 @@ static struct resource *ccp_find_mmio_area(struct ccp_device *ccp)
return NULL;
}
-#ifdef CONFIG_ACPI
-static int ccp_acpi_support(struct ccp_device *ccp)
-{
- struct ccp_platform *ccp_platform = ccp->dev_specific;
- struct acpi_device *adev = ACPI_COMPANION(ccp->dev);
- acpi_handle handle;
- acpi_status status;
- unsigned long long data;
- int cca;
-
- /* Retrieve the device cache coherency value */
- handle = adev->handle;
- do {
- status = acpi_evaluate_integer(handle, "_CCA", NULL, &data);
- if (!ACPI_FAILURE(status)) {
- cca = data;
- break;
- }
- } while (!ACPI_FAILURE(status));
-
- if (ACPI_FAILURE(status)) {
- dev_err(ccp->dev, "error obtaining acpi coherency value\n");
- return -EINVAL;
- }
-
- ccp_platform->coherent = !!cca;
-
- return 0;
-}
-#else /* CONFIG_ACPI */
-static int ccp_acpi_support(struct ccp_device *ccp)
-{
- return -EINVAL;
-}
-#endif
-
-#ifdef CONFIG_OF
-static int ccp_of_support(struct ccp_device *ccp)
-{
- struct ccp_platform *ccp_platform = ccp->dev_specific;
-
- ccp_platform->coherent = of_dma_is_coherent(ccp->dev->of_node);
-
- return 0;
-}
-#else
-static int ccp_of_support(struct ccp_device *ccp)
-{
- return -EINVAL;
-}
-#endif
-
static int ccp_platform_probe(struct platform_device *pdev)
{
struct ccp_device *ccp;
@@ -182,13 +130,7 @@ static int ccp_platform_probe(struct platform_device *pdev)
goto e_err;
}
- if (ccp_platform->use_acpi)
- ret = ccp_acpi_support(ccp);
- else
- ret = ccp_of_support(ccp);
- if (ret)
- goto e_err;
-
+ ccp_platform->coherent = device_dma_is_coherent(ccp->dev);
if (ccp_platform->coherent)
ccp->axcache = CACHE_WB_NO_ALLOC;
else
--
2.1.0
Currently, amd-xgbe driver has separate logic to determine device
coherency for DT vs. ACPI. This patch simplifies the code with
a call to device_dma_is_coherent().
CC: David S. Miller <[email protected]>
Signed-off-by: Tom Lendacky <[email protected]>
Signed-off-by: Suravee Suthikulpanit <[email protected]>
---
drivers/net/ethernet/amd/xgbe/xgbe-main.c | 27 +--------------------------
1 file changed, 1 insertion(+), 26 deletions(-)
diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-main.c b/drivers/net/ethernet/amd/xgbe/xgbe-main.c
index 7149053..6d2c702 100644
--- a/drivers/net/ethernet/amd/xgbe/xgbe-main.c
+++ b/drivers/net/ethernet/amd/xgbe/xgbe-main.c
@@ -168,13 +168,8 @@ static void xgbe_init_all_fptrs(struct xgbe_prv_data *pdata)
#ifdef CONFIG_ACPI
static int xgbe_acpi_support(struct xgbe_prv_data *pdata)
{
- struct acpi_device *adev = pdata->adev;
struct device *dev = pdata->dev;
u32 property;
- acpi_handle handle;
- acpi_status status;
- unsigned long long data;
- int cca;
int ret;
/* Obtain the system clock setting */
@@ -195,24 +190,6 @@ static int xgbe_acpi_support(struct xgbe_prv_data *pdata)
}
pdata->ptpclk_rate = property;
- /* Retrieve the device cache coherency value */
- handle = adev->handle;
- do {
- status = acpi_evaluate_integer(handle, "_CCA", NULL, &data);
- if (!ACPI_FAILURE(status)) {
- cca = data;
- break;
- }
-
- status = acpi_get_parent(handle, &handle);
- } while (!ACPI_FAILURE(status));
-
- if (ACPI_FAILURE(status)) {
- dev_err(dev, "error obtaining acpi coherency value\n");
- return -EINVAL;
- }
- pdata->coherent = !!cca;
-
return 0;
}
#else /* CONFIG_ACPI */
@@ -243,9 +220,6 @@ static int xgbe_of_support(struct xgbe_prv_data *pdata)
}
pdata->ptpclk_rate = clk_get_rate(pdata->ptpclk);
- /* Retrieve the device cache coherency value */
- pdata->coherent = of_dma_is_coherent(dev->of_node);
-
return 0;
}
#else /* CONFIG_OF */
@@ -364,6 +338,7 @@ static int xgbe_probe(struct platform_device *pdev)
goto err_io;
/* Set the DMA coherency values */
+ pdata->coherent = device_dma_is_coherent(pdata->dev);
if (pdata->coherent) {
pdata->axdomain = XGBE_DMA_OS_AXDOMAIN;
pdata->arcache = XGBE_DMA_OS_ARCACHE;
--
2.1.0
On Tuesday 05 May 2015 10:12:06 Suravee Suthikulpanit wrote:
> +struct dma_map_ops dummy_dma_ops = {
> + .alloc = __dummy_alloc,
> + .free = __dummy_free,
> + .mmap = __dummy_mmap,
> + .map_page = __dummy_map_page,
> + .unmap_page = __dummy_unmap_page,
> + .map_sg = __dummy_map_sg,
> + .unmap_sg = __dummy_unmap_sg,
> + .sync_single_for_cpu = __dummy_sync_single_for_cpu,
> + .sync_single_for_device = __dummy_sync_single_for_device,
> + .sync_sg_for_cpu = __dummy_sync_sg_for_cpu,
> + .sync_sg_for_device = __dummy_sync_sg_for_device,
> + .mapping_error = __dummy_mapping_error,
> + .dma_supported = __dummy_dma_supported,
> +};
> +EXPORT_SYMBOL(dummy_dma_ops);
> +
>
This will clearly work, but I think it's easier to just leave
the dma_mask pointer as NULL when creating the platform device,
which should let the normal dma ops fail all the callbacks.
Arnd
On 5/5/2015 10:44 AM, Arnd Bergmann wrote:
> On Tuesday 05 May 2015 10:12:06 Suravee Suthikulpanit wrote:
>> +struct dma_map_ops dummy_dma_ops = {
>> + .alloc = __dummy_alloc,
>> + .free = __dummy_free,
>> + .mmap = __dummy_mmap,
>> + .map_page = __dummy_map_page,
>> + .unmap_page = __dummy_unmap_page,
>> + .map_sg = __dummy_map_sg,
>> + .unmap_sg = __dummy_unmap_sg,
>> + .sync_single_for_cpu = __dummy_sync_single_for_cpu,
>> + .sync_single_for_device = __dummy_sync_single_for_device,
>> + .sync_sg_for_cpu = __dummy_sync_sg_for_cpu,
>> + .sync_sg_for_device = __dummy_sync_sg_for_device,
>> + .mapping_error = __dummy_mapping_error,
>> + .dma_supported = __dummy_dma_supported,
>> +};
>> +EXPORT_SYMBOL(dummy_dma_ops);
>> +
>>
>
> This will clearly work, but I think it's easier to just leave
> the dma_mask pointer as NULL when creating the platform device,
> which should let the normal dma ops fail all the callbacks.
>
> Arnd
>
However, codes in several places are making use of dma_map_ops without
checking if the ops are NULL (i.e.
include/asm-generic/dma-mapping-common.h and in arch-specific
implementation). If setting it to NULL is what we are planning to
support, we would need to scrub the current code to put NULL check.
Also, would you consider if that is safe to do going forward?
Thanks,
Suravee
On Tuesday 05 May 2015 11:09:38 Suravee Suthikulanit wrote:
>
> However, codes in several places are making use of dma_map_ops without
> checking if the ops are NULL (i.e.
> include/asm-generic/dma-mapping-common.h and in arch-specific
> implementation). If setting it to NULL is what we are planning to
> support, we would need to scrub the current code to put NULL check.
> Also, would you consider if that is safe to do going forward?
>
>
I mean the dma_mask pointer, not dma_map_ops.
Arnd
On 5/5/2015 11:12 AM, Arnd Bergmann wrote:
> On Tuesday 05 May 2015 11:09:38 Suravee Suthikulanit wrote:
>>
>> However, codes in several places are making use of dma_map_ops without
>> checking if the ops are NULL (i.e.
>> include/asm-generic/dma-mapping-common.h and in arch-specific
>> implementation). If setting it to NULL is what we are planning to
>> support, we would need to scrub the current code to put NULL check.
>> Also, would you consider if that is safe to do going forward?
>>
>>
>
> I mean the dma_mask pointer, not dma_map_ops.
>
> Arnd
>
Ah, got it. Sorry for confusion.
Suravee
On 05/05/2015 11:13 AM, Suravee Suthikulanit wrote:
> On 5/5/2015 11:12 AM, Arnd Bergmann wrote:
>> On Tuesday 05 May 2015 11:09:38 Suravee Suthikulanit wrote:
>>>
>>> However, codes in several places are making use of dma_map_ops without
>>> checking if the ops are NULL (i.e.
>>> include/asm-generic/dma-mapping-common.h and in arch-specific
>>> implementation). If setting it to NULL is what we are planning to
>>> support, we would need to scrub the current code to put NULL check.
>>> Also, would you consider if that is safe to do going forward?
>>>
>>>
>>
>> I mean the dma_mask pointer, not dma_map_ops.
Except a lot of drivers will actually set the dma_mask pointer during
probe (usually by setting dev->dma_mask = &dev->coherent_dma_mask or by
calling dma_coerce_mask_and_coherent). So I think the dummy_dma_ops
might be the safest way to go.
Thanks,
Tom
>>
>> Arnd
>>
>
> Ah, got it. Sorry for confusion.
>
> Suravee
>
> _______________________________________________
> Linaro-acpi mailing list
> [email protected]
> https://lists.linaro.org/mailman/listinfo/linaro-acpi
On Tuesday 05 May 2015 11:24:03 Tom Lendacky wrote:
> On 05/05/2015 11:13 AM, Suravee Suthikulanit wrote:
> > On 5/5/2015 11:12 AM, Arnd Bergmann wrote:
> >> On Tuesday 05 May 2015 11:09:38 Suravee Suthikulanit wrote:
> >>>
> >>> However, codes in several places are making use of dma_map_ops without
> >>> checking if the ops are NULL (i.e.
> >>> include/asm-generic/dma-mapping-common.h and in arch-specific
> >>> implementation). If setting it to NULL is what we are planning to
> >>> support, we would need to scrub the current code to put NULL check.
> >>> Also, would you consider if that is safe to do going forward?
> >>>
> >>>
> >>
> >> I mean the dma_mask pointer, not dma_map_ops.
>
> Except a lot of drivers will actually set the dma_mask pointer during
> probe (usually by setting dev->dma_mask = &dev->coherent_dma_mask or by
> calling dma_coerce_mask_and_coherent). So I think the dummy_dma_ops
> might be the safest way to go.
Those drivers are broken already, I don't think we should worry about
them. Let's just fix them properly when we run into problems with them.
Basically any use of dma_coerce_mask_and_coherent() is an annotation
for a bug where a driver used to set dev->dma_mask = &dev->coherent_dma_mask
manually.
Arnd
On Tuesday, May 05, 2015 10:12:05 AM Suravee Suthikulpanit wrote:
> This patch implements support for ACPI _CCA object, which is introduced in
> ACPIv5.1, can be used for specifying device DMA coherency attribute.
>
> The parsing logic traverses device namespace to parse coherency
> information, and stores it in acpi_device_flags. Then uses it to call
> arch_setup_dma_ops() when creating each device enumerated in DSDT
> during ACPI scan.
>
> This patch also introduces acpi_dma_is_coherent(), which provides
> an interface for device drivers to check the coherency information
> similarly to the of_dma_is_coherent().
>
> Signed-off-by: Mark Salter <[email protected]>
> Signed-off-by: Suravee Suthikulpanit <[email protected]>
> ---
> NOTE:
> * Since there seem to be conflict opinions regarding how
> architecture should handle _CCA=0. So, I am proposing the
> CONFIG_ARCH_SUPPORT_CCA_ZERO, which can be specified by
> for each architecture to define behavior of the ACPI
> scanning code when _CCA=0. Let me know if this is acceptable.
>
> drivers/acpi/Kconfig | 6 +++++
> drivers/acpi/acpi_platform.c | 4 ++-
> drivers/acpi/scan.c | 62 ++++++++++++++++++++++++++++++++++++++++++++
> include/acpi/acpi_bus.h | 11 +++++++-
> include/linux/acpi.h | 5 ++++
> 5 files changed, 86 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
> index ab2cbb5..dd386e9 100644
> --- a/drivers/acpi/Kconfig
> +++ b/drivers/acpi/Kconfig
> @@ -54,6 +54,12 @@ config ACPI_GENERIC_GSI
> config ACPI_SYSTEM_POWER_STATES_SUPPORT
> bool
>
> +config ACPI_MUST_HAVE_CCA
ACPI_CCA_REQUIRED maybe?
> + bool
> +
> +config ACPI_SUPPORT_CCA_ZERO
I guess this means "we support devices that can DMA, but are not coherent".
right?
> + bool
> +
> config ACPI_SLEEP
> bool
> depends on SUSPEND || HIBERNATION
> diff --git a/drivers/acpi/acpi_platform.c b/drivers/acpi/acpi_platform.c
> index 4bf7559..a6feca4 100644
> --- a/drivers/acpi/acpi_platform.c
> +++ b/drivers/acpi/acpi_platform.c
> @@ -108,9 +108,11 @@ struct platform_device *acpi_create_platform_device(struct acpi_device *adev)
> if (IS_ERR(pdev))
> dev_err(&adev->dev, "platform device creation failed: %ld\n",
> PTR_ERR(pdev));
> - else
> + else {
Please add braces to both branches when making such changes (as per CodingStyle).
> + acpi_setup_device_dma(adev, &pdev->dev);
Why do we need to do that here (for the second time)?
> dev_dbg(&adev->dev, "created platform device %s\n",
> dev_name(&pdev->dev));
> + }
>
> kfree(resources);
> return pdev;
> diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
> index 849b699..ac33b29 100644
> --- a/drivers/acpi/scan.c
> +++ b/drivers/acpi/scan.c
> @@ -11,6 +11,7 @@
> #include <linux/kthread.h>
> #include <linux/dmi.h>
> #include <linux/nls.h>
> +#include <linux/dma-mapping.h>
>
> #include <asm/pgtable.h>
>
> @@ -2137,6 +2138,66 @@ void acpi_free_pnp_ids(struct acpi_device_pnp *pnp)
> kfree(pnp->unique_id);
> }
>
> +void acpi_setup_device_dma(struct acpi_device *adev, struct device *dev)
> +{
> + int coherent = acpi_dma_is_coherent(adev);
> +
> + /**
> + * Currently, we only support DMA for devices that _CCA=1
> + * since this seems to be the case on most ACPI platforms.
> + *
> + * For the case when _CCA=0 (i.e. is_coherent=0 && cca_seen=1),
> + * we would rely on arch-specific cache maintenance for
> + * non-coherence DMA operations if architecture enables
> + * CONFIG_ACPI_SUPPORT_CCA_ZERO.
> + *
> + * For the case when _CCA is missing but platform requires it
> + * (i.e. is_coherent=0 && cca_seen=0), we do not call
> + * arch_setup_dma_ops() and fallback to arch-specific default
> + * handling.
> + */
> + if (adev->flags.cca_seen) {
> + if (!coherent && !IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO))
> + return;
> + arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
Oh dear.
What about
if (adev->flags.cca_seen && (coherent || IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO)))
arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
I wonder how this is going to affect x86/ia64 too?
> + }
> +}
> +
> +static void acpi_init_coherency(struct acpi_device *adev)
> +{
> + unsigned long long cca = 0;
> + acpi_status status;
> + struct acpi_device *parent = adev->parent;
> +
> + if (parent && parent->flags.cca_seen) {
> + /*
> + * From ACPI spec, OSPM will ignore _CCA if an ancestor
> + * already saw one.
> + */
> + adev->flags.cca_seen = 1;
> + cca = acpi_dma_is_coherent(parent);
> + } else {
> + status = acpi_evaluate_integer(adev->handle, "_CCA",
> + NULL, &cca);
> + if (ACPI_SUCCESS(status)) {
> + adev->flags.cca_seen = 1;
> + } else if (!IS_ENABLED(CONFIG_ACPI_MUST_HAVE_CCA)) {
> + /*
> + * If architecture does not specify that _CCA is
> + * required for DMA-able devices (e.g. x86),
> + * we default to _CCA=1.
> + */
> + cca = 1;
> + } else {
> + dev_err(&adev->dev, FW_BUG
> + "DMA is not setup due to missing _CCA.\n");
> + }
> + }
> +
> + adev->flags.is_coherent = cca;
> + acpi_setup_device_dma(adev, &adev->dev);
> +}
> +
> void acpi_init_device_object(struct acpi_device *device, acpi_handle handle,
> int type, unsigned long long sta)
> {
> @@ -2155,6 +2216,7 @@ void acpi_init_device_object(struct acpi_device *device, acpi_handle handle,
> device->flags.visited = false;
> device_initialize(&device->dev);
> dev_set_uevent_suppress(&device->dev, true);
> + acpi_init_coherency(device);
> }
>
> void acpi_device_add_finalize(struct acpi_device *device)
> diff --git a/include/acpi/acpi_bus.h b/include/acpi/acpi_bus.h
> index 8de4fa9..b804183 100644
> --- a/include/acpi/acpi_bus.h
> +++ b/include/acpi/acpi_bus.h
> @@ -208,7 +208,9 @@ struct acpi_device_flags {
> u32 visited:1;
> u32 hotplug_notify:1;
> u32 is_dock_station:1;
> - u32 reserved:23;
> + u32 is_coherent:1;
> + u32 cca_seen:1;
> + u32 reserved:21;
That will conflict with a patch I've already queued up, but never mind.
> };
>
> /* File System */
> @@ -380,6 +382,13 @@ struct acpi_device {
> void (*remove)(struct acpi_device *);
> };
>
> +static inline bool acpi_dma_is_coherent(struct acpi_device *adev)
> +{
> + return adev && adev->flags.is_coherent;
> +}
> +
> +void acpi_setup_device_dma(struct acpi_device *adev, struct device *dev);
> +
> static inline bool is_acpi_node(struct fwnode_handle *fwnode)
> {
> return fwnode && fwnode->type == FWNODE_ACPI;
> diff --git a/include/linux/acpi.h b/include/linux/acpi.h
> index b10c4a6..d14e777 100644
> --- a/include/linux/acpi.h
> +++ b/include/linux/acpi.h
> @@ -583,6 +583,11 @@ static inline int acpi_device_modalias(struct device *dev,
> return -ENODEV;
> }
>
> +static inline bool acpi_dma_is_coherent(struct acpi_device *adev)
> +{
> + return false;
> +}
> +
> #define ACPI_PTR(_ptr) (NULL)
>
> #endif /* !CONFIG_ACPI */
>
--
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
On 2015年05月05日 23:12, Suravee Suthikulpanit wrote:
> This patch implements support for ACPI _CCA object, which is introduced in
> ACPIv5.1, can be used for specifying device DMA coherency attribute.
>
> The parsing logic traverses device namespace to parse coherency
> information, and stores it in acpi_device_flags. Then uses it to call
> arch_setup_dma_ops() when creating each device enumerated in DSDT
> during ACPI scan.
>
> This patch also introduces acpi_dma_is_coherent(), which provides
> an interface for device drivers to check the coherency information
> similarly to the of_dma_is_coherent().
>
> Signed-off-by: Mark Salter <[email protected]>
> Signed-off-by: Suravee Suthikulpanit <[email protected]>
> ---
> NOTE:
> * Since there seem to be conflict opinions regarding how
> architecture should handle _CCA=0. So, I am proposing the
> CONFIG_ARCH_SUPPORT_CCA_ZERO, which can be specified by
> for each architecture to define behavior of the ACPI
> scanning code when _CCA=0. Let me know if this is acceptable.
>
> drivers/acpi/Kconfig | 6 +++++
> drivers/acpi/acpi_platform.c | 4 ++-
> drivers/acpi/scan.c | 62 ++++++++++++++++++++++++++++++++++++++++++++
> include/acpi/acpi_bus.h | 11 +++++++-
> include/linux/acpi.h | 5 ++++
> 5 files changed, 86 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
> index ab2cbb5..dd386e9 100644
> --- a/drivers/acpi/Kconfig
> +++ b/drivers/acpi/Kconfig
> @@ -54,6 +54,12 @@ config ACPI_GENERIC_GSI
> config ACPI_SYSTEM_POWER_STATES_SUPPORT
> bool
>
> +config ACPI_MUST_HAVE_CCA
> + bool
> +
> +config ACPI_SUPPORT_CCA_ZERO
> + bool
> +
> config ACPI_SLEEP
> bool
> depends on SUSPEND || HIBERNATION
> diff --git a/drivers/acpi/acpi_platform.c b/drivers/acpi/acpi_platform.c
> index 4bf7559..a6feca4 100644
> --- a/drivers/acpi/acpi_platform.c
> +++ b/drivers/acpi/acpi_platform.c
> @@ -108,9 +108,11 @@ struct platform_device *acpi_create_platform_device(struct acpi_device *adev)
> if (IS_ERR(pdev))
> dev_err(&adev->dev, "platform device creation failed: %ld\n",
> PTR_ERR(pdev));
> - else
> + else {
> + acpi_setup_device_dma(adev, &pdev->dev);
> dev_dbg(&adev->dev, "created platform device %s\n",
> dev_name(&pdev->dev));
> + }
>
> kfree(resources);
> return pdev;
> diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
> index 849b699..ac33b29 100644
> --- a/drivers/acpi/scan.c
> +++ b/drivers/acpi/scan.c
> @@ -11,6 +11,7 @@
> #include <linux/kthread.h>
> #include <linux/dmi.h>
> #include <linux/nls.h>
> +#include <linux/dma-mapping.h>
>
> #include <asm/pgtable.h>
>
> @@ -2137,6 +2138,66 @@ void acpi_free_pnp_ids(struct acpi_device_pnp *pnp)
> kfree(pnp->unique_id);
> }
>
> +void acpi_setup_device_dma(struct acpi_device *adev, struct device *dev)
I aasume adev->dev in struct *adev is the same as struct device *dev
passed here, so
> +{
> + int coherent = acpi_dma_is_coherent(adev);
> +
> + /**
> + * Currently, we only support DMA for devices that _CCA=1
> + * since this seems to be the case on most ACPI platforms.
> + *
> + * For the case when _CCA=0 (i.e. is_coherent=0 && cca_seen=1),
> + * we would rely on arch-specific cache maintenance for
> + * non-coherence DMA operations if architecture enables
> + * CONFIG_ACPI_SUPPORT_CCA_ZERO.
> + *
> + * For the case when _CCA is missing but platform requires it
> + * (i.e. is_coherent=0 && cca_seen=0), we do not call
> + * arch_setup_dma_ops() and fallback to arch-specific default
> + * handling.
> + */
> + if (adev->flags.cca_seen) {
> + if (!coherent && !IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO))
> + return;
> + arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
how about using &adev->dev here, and just pass struct acpi_device *adev
for this function?
Thanks
Hanjun
[RESEND]
On 5/5/15 15:36, Rafael J. Wysocki wrote:
> On Tuesday, May 05, 2015 10:12:05 AM Suravee Suthikulpanit wrote:
>> diff --git a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
>> index ab2cbb5..dd386e9 100644
>> --- a/drivers/acpi/Kconfig
>> +++ b/drivers/acpi/Kconfig
>> @@ -54,6 +54,12 @@ config ACPI_GENERIC_GSI
>> config ACPI_SYSTEM_POWER_STATES_SUPPORT
>> bool
>>
>> +config ACPI_MUST_HAVE_CCA
>
> ACPI_CCA_REQUIRED maybe?
Sure.
>
>> + bool
>> +
>> +config ACPI_SUPPORT_CCA_ZERO
>
> I guess this means "we support devices that can DMA, but are not coherent".
> right?
Yes, basically when _CCA=0.
>> + bool
>> +
>> config ACPI_SLEEP
>> bool
>> depends on SUSPEND || HIBERNATION
>> diff --git a/drivers/acpi/acpi_platform.c b/drivers/acpi/acpi_platform.c
>> index 4bf7559..a6feca4 100644
>> --- a/drivers/acpi/acpi_platform.c
>> +++ b/drivers/acpi/acpi_platform.c
>> @@ -108,9 +108,11 @@ struct platform_device *acpi_create_platform_device(struct acpi_device *adev)
>> if (IS_ERR(pdev))
>> dev_err(&adev->dev, "platform device creation failed: %ld\n",
>> PTR_ERR(pdev));
>> - else
>> + else {
>
> Please add braces to both branches when making such changes (as per CodingStyle).
>
OK.
>> + acpi_setup_device_dma(adev, &pdev->dev);
>
> Why do we need to do that here (for the second time)?
Because we are calling:
acpi_create_platform_device()
|--> platform_device_register_device_full()
|-->platform_device_alloc()
This creates platform_device, which allocate a new platform_device->dev.
This is not the same as the original acpi_device->dev that was created
during acpi_add_single_object(). So, we have to set up the device
coherency again.
>> diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
>> index 849b699..ac33b29 100644
>> --- a/drivers/acpi/scan.c
>> +++ b/drivers/acpi/scan.c
>> @@ -11,6 +11,7 @@
>> #include <linux/kthread.h>
>> #include <linux/dmi.h>
>> #include <linux/nls.h>
>> +#include <linux/dma-mapping.h>
>>
>> #include <asm/pgtable.h>
>>
>> @@ -2137,6 +2138,66 @@ void acpi_free_pnp_ids(struct acpi_device_pnp *pnp)
>> kfree(pnp->unique_id);
>> }
>>
>> +void acpi_setup_device_dma(struct acpi_device *adev, struct device *dev)
>> +{
>> + int coherent = acpi_dma_is_coherent(adev);
>> +
>> + /**
>> + * Currently, we only support DMA for devices that _CCA=1
>> + * since this seems to be the case on most ACPI platforms.
>> + *
>> + * For the case when _CCA=0 (i.e. is_coherent=0 && cca_seen=1),
>> + * we would rely on arch-specific cache maintenance for
>> + * non-coherence DMA operations if architecture enables
>> + * CONFIG_ACPI_SUPPORT_CCA_ZERO.
>> + *
>> + * For the case when _CCA is missing but platform requires it
>> + * (i.e. is_coherent=0 && cca_seen=0), we do not call
>> + * arch_setup_dma_ops() and fallback to arch-specific default
>> + * handling.
>> + */
>> + if (adev->flags.cca_seen) {
>> + if (!coherent && !IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO))
>> + return;
>> + arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
>
> Oh dear.
I made a mistake here. This logic should also call arch_setup_dma_ops()
when cca_seen=0 and coherent=1 (e.g. when _CCA is not required and
default to coherent when it is missing). The current logic doesn't do that.
>
> What about
>
> if (adev->flags.cca_seen && (coherent || IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO)))
> arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
What about:
if (coherent ||
(adev->flags.cca_seen &&
IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO))
arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
> I wonder how this is going to affect x86/ia64 too?
>
This should not affect x86 since arch_setup_dma_ops() is currently not
implement for x86, and default to NOP (see include/linux/dma-mapping.h).
Also, on x86, _CCA is not required and default to 1 if missing.
Thanks,
Suravee
On 5/5/15 22:13, Hanjun Guo wrote:
> On 2015年05月05日 23:12, Suravee Suthikulpanit wrote:
>> This patch implements support for ACPI _CCA object, which is
>> introduced in
>> ACPIv5.1, can be used for specifying device DMA coherency attribute.
>>
>> The parsing logic traverses device namespace to parse coherency
>> information, and stores it in acpi_device_flags. Then uses it to call
>> arch_setup_dma_ops() when creating each device enumerated in DSDT
>> during ACPI scan.
>>
>> This patch also introduces acpi_dma_is_coherent(), which provides
>> an interface for device drivers to check the coherency information
>> similarly to the of_dma_is_coherent().
>>
>> Signed-off-by: Mark Salter <[email protected]>
>> Signed-off-by: Suravee Suthikulpanit <[email protected]>
>> ---
>> NOTE:
>> * Since there seem to be conflict opinions regarding how
>> architecture should handle _CCA=0. So, I am proposing the
>> CONFIG_ARCH_SUPPORT_CCA_ZERO, which can be specified by
>> for each architecture to define behavior of the ACPI
>> scanning code when _CCA=0. Let me know if this is acceptable.
>>
>> drivers/acpi/Kconfig | 6 +++++
>> drivers/acpi/acpi_platform.c | 4 ++-
>> drivers/acpi/scan.c | 62
>> ++++++++++++++++++++++++++++++++++++++++++++
>> include/acpi/acpi_bus.h | 11 +++++++-
>> include/linux/acpi.h | 5 ++++
>> 5 files changed, 86 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
>> index ab2cbb5..dd386e9 100644
>> --- a/drivers/acpi/Kconfig
>> +++ b/drivers/acpi/Kconfig
>> @@ -54,6 +54,12 @@ config ACPI_GENERIC_GSI
>> config ACPI_SYSTEM_POWER_STATES_SUPPORT
>> bool
>>
>> +config ACPI_MUST_HAVE_CCA
>> + bool
>> +
>> +config ACPI_SUPPORT_CCA_ZERO
>> + bool
>> +
>> config ACPI_SLEEP
>> bool
>> depends on SUSPEND || HIBERNATION
>> diff --git a/drivers/acpi/acpi_platform.c b/drivers/acpi/acpi_platform.c
>> index 4bf7559..a6feca4 100644
>> --- a/drivers/acpi/acpi_platform.c
>> +++ b/drivers/acpi/acpi_platform.c
>> @@ -108,9 +108,11 @@ struct platform_device
>> *acpi_create_platform_device(struct acpi_device *adev)
>> if (IS_ERR(pdev))
>> dev_err(&adev->dev, "platform device creation failed: %ld\n",
>> PTR_ERR(pdev));
>> - else
>> + else {
>> + acpi_setup_device_dma(adev, &pdev->dev);
>> dev_dbg(&adev->dev, "created platform device %s\n",
>> dev_name(&pdev->dev));
>> + }
>>
>> kfree(resources);
>> return pdev;
>> diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
>> index 849b699..ac33b29 100644
>> --- a/drivers/acpi/scan.c
>> +++ b/drivers/acpi/scan.c
>> @@ -11,6 +11,7 @@
>> #include <linux/kthread.h>
>> #include <linux/dmi.h>
>> #include <linux/nls.h>
>> +#include <linux/dma-mapping.h>
>>
>> #include <asm/pgtable.h>
>>
>> @@ -2137,6 +2138,66 @@ void acpi_free_pnp_ids(struct acpi_device_pnp
>> *pnp)
>> kfree(pnp->unique_id);
>> }
>>
>> +void acpi_setup_device_dma(struct acpi_device *adev, struct device *dev)
>
> I aasume adev->dev in struct *adev is the same as struct device *dev
> passed here, so
>
>> +{
>> + int coherent = acpi_dma_is_coherent(adev);
>> +
>> + /**
>> + * Currently, we only support DMA for devices that _CCA=1
>> + * since this seems to be the case on most ACPI platforms.
>> + *
>> + * For the case when _CCA=0 (i.e. is_coherent=0 && cca_seen=1),
>> + * we would rely on arch-specific cache maintenance for
>> + * non-coherence DMA operations if architecture enables
>> + * CONFIG_ACPI_SUPPORT_CCA_ZERO.
>> + *
>> + * For the case when _CCA is missing but platform requires it
>> + * (i.e. is_coherent=0 && cca_seen=0), we do not call
>> + * arch_setup_dma_ops() and fallback to arch-specific default
>> + * handling.
>> + */
>> + if (adev->flags.cca_seen) {
>> + if (!coherent && !IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO))
>> + return;
>> + arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
>
> how about using &adev->dev here, and just pass struct acpi_device *adev
> for this function?
Actually, I was using arch_setup_device_dma() in multiple places, and
adev->dev is not necessary the same as *dev. However, I am refactoring
this function in V3. Anyways, thanks for reviewing.
Suravee
Hi Suravee,
On 05/05/15 16:12, Suravee Suthikulpanit wrote:
> From http://www.uefi.org/sites/default/files/resources/ACPI_6.0.pdf,
> section 6.2.17 _CCA states that ARM platforms require ACPI _CCA
> object to be specified for DMA-cabpable devices. This patch introduces
> ACPI_MUST_HAVE_CCA in arm64 Kconfig to specify such requirement.
>
> In this case of missing _CCA, arm64 would assign dummy_dma_ops
> to disable DMA capability of the device.
>
> Signed-off-by: Mark Salter <[email protected]>
> Signed-off-by: Suravee Suthikulpanit <[email protected]>
> ---
[...]
> +static void __dummy_sync_single_for_cpu(struct device *dev,
> + dma_addr_t dev_addr, size_t size,
> + enum dma_data_direction dir)
> +{
> +}
> +
> +static void __dummy_sync_single_for_device(struct device *dev,
> + dma_addr_t dev_addr, size_t size,
> + enum dma_data_direction dir)
> +{
> +}
Minor point, but I don't see the need to have multiple dummy functions
with identical signatures - just have a generic dummy_sync_single and
assign it to both ops.
> +static void __dummy_sync_sg_for_cpu(struct device *dev,
> + struct scatterlist *sgl, int nelems,
> + enum dma_data_direction dir)
> +{
> +}
> +
> +static void __dummy_sync_sg_for_device(struct device *dev,
> + struct scatterlist *sgl, int nelems,
> + enum dma_data_direction dir)
> +{
> +}
Ditto here with dummy_sync_sg.
I wonder if there's any argument for putting the dummy DMA ops somewhere
common, like drivers/base/dma-mapping.c?
Robin.
On 5/6/2015 5:08 AM, Robin Murphy wrote:
> [...]
>> +static void __dummy_sync_single_for_cpu(struct device *dev,
>> + dma_addr_t dev_addr, size_t size,
>> + enum dma_data_direction dir)
>> +{
>> +}
>> +
>> +static void __dummy_sync_single_for_device(struct device *dev,
>> + dma_addr_t dev_addr, size_t size,
>> + enum dma_data_direction dir)
>> +{
>> +}
>
> Minor point, but I don't see the need to have multiple dummy functions
> with identical signatures - just have a generic dummy_sync_single and
> assign it to both ops.
>
>> +static void __dummy_sync_sg_for_cpu(struct device *dev,
>> + struct scatterlist *sgl, int nelems,
>> + enum dma_data_direction dir)
>> +{
>> +}
>> +
>> +static void __dummy_sync_sg_for_device(struct device *dev,
>> + struct scatterlist *sgl, int nelems,
>> + enum dma_data_direction dir)
>> +{
>> +}
>
> Ditto here with dummy_sync_sg.
Hi Robin,
Good point. I'll take care of that in V3.
>
> I wonder if there's any argument for putting the dummy DMA ops somewhere
> common, like drivers/base/dma-mapping.c?
>
> Robin.
Hm.. If this approach will be adopted by other architectures, then it
would make sense. Currently, this is only used by arm64. So, I think it
is okay to leave this here for now.
Thanks,
Suravee
On Tuesday, May 05, 2015 11:15:37 PM Suravee Suthikulpanit wrote:
> [RESEND]
>
> On 5/5/15 15:36, Rafael J. Wysocki wrote:
> > On Tuesday, May 05, 2015 10:12:05 AM Suravee Suthikulpanit wrote:
> >> diff --git a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
> >> index ab2cbb5..dd386e9 100644
> >> --- a/drivers/acpi/Kconfig
> >> +++ b/drivers/acpi/Kconfig
> >> @@ -54,6 +54,12 @@ config ACPI_GENERIC_GSI
> >> config ACPI_SYSTEM_POWER_STATES_SUPPORT
> >> bool
> >>
> >> +config ACPI_MUST_HAVE_CCA
> >
> > ACPI_CCA_REQUIRED maybe?
>
> Sure.
>
> >
> >> + bool
> >> +
> >> +config ACPI_SUPPORT_CCA_ZERO
> >
> > I guess this means "we support devices that can DMA, but are not coherent".
> > right?
>
> Yes, basically when _CCA=0.
So what about
ARCH_SUPPORT_CACHE_INCOHERENT_DMA
or something similar?
> >> + bool
> >> +
> >> config ACPI_SLEEP
> >> bool
> >> depends on SUSPEND || HIBERNATION
> >> diff --git a/drivers/acpi/acpi_platform.c b/drivers/acpi/acpi_platform.c
> >> index 4bf7559..a6feca4 100644
> >> --- a/drivers/acpi/acpi_platform.c
> >> +++ b/drivers/acpi/acpi_platform.c
> >> @@ -108,9 +108,11 @@ struct platform_device *acpi_create_platform_device(struct acpi_device *adev)
> >> if (IS_ERR(pdev))
> >> dev_err(&adev->dev, "platform device creation failed: %ld\n",
> >> PTR_ERR(pdev));
> >> - else
> >> + else {
> >
> > Please add braces to both branches when making such changes (as per CodingStyle).
> >
>
> OK.
>
> >> + acpi_setup_device_dma(adev, &pdev->dev);
> >
> > Why do we need to do that here (for the second time)?
>
> Because we are calling:
> acpi_create_platform_device()
> |--> platform_device_register_device_full()
> |-->platform_device_alloc()
>
> This creates platform_device, which allocate a new platform_device->dev.
> This is not the same as the original acpi_device->dev that was created
> during acpi_add_single_object(). So, we have to set up the device
> coherency again.
Ah, so the second arg is different now.
Well, in that case, why do we need to set it up for the adev's dev member?
> >> diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
> >> index 849b699..ac33b29 100644
> >> --- a/drivers/acpi/scan.c
> >> +++ b/drivers/acpi/scan.c
> >> @@ -11,6 +11,7 @@
> >> #include <linux/kthread.h>
> >> #include <linux/dmi.h>
> >> #include <linux/nls.h>
> >> +#include <linux/dma-mapping.h>
> >>
> >> #include <asm/pgtable.h>
> >>
> >> @@ -2137,6 +2138,66 @@ void acpi_free_pnp_ids(struct acpi_device_pnp *pnp)
> >> kfree(pnp->unique_id);
> >> }
> >>
> >> +void acpi_setup_device_dma(struct acpi_device *adev, struct device *dev)
> >> +{
> >> + int coherent = acpi_dma_is_coherent(adev);
> >> +
> >> + /**
> >> + * Currently, we only support DMA for devices that _CCA=1
> >> + * since this seems to be the case on most ACPI platforms.
> >> + *
> >> + * For the case when _CCA=0 (i.e. is_coherent=0 && cca_seen=1),
> >> + * we would rely on arch-specific cache maintenance for
> >> + * non-coherence DMA operations if architecture enables
> >> + * CONFIG_ACPI_SUPPORT_CCA_ZERO.
> >> + *
> >> + * For the case when _CCA is missing but platform requires it
> >> + * (i.e. is_coherent=0 && cca_seen=0), we do not call
> >> + * arch_setup_dma_ops() and fallback to arch-specific default
> >> + * handling.
> >> + */
> >> + if (adev->flags.cca_seen) {
> >> + if (!coherent && !IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO))
> >> + return;
> >> + arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
> >
> > Oh dear.
>
> I made a mistake here. This logic should also call arch_setup_dma_ops()
> when cca_seen=0 and coherent=1 (e.g. when _CCA is not required and
> default to coherent when it is missing). The current logic doesn't do that.
>
> >
> > What about
> >
> > if (adev->flags.cca_seen && (coherent || IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO)))
> > arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
>
> What about:
> if (coherent ||
> (adev->flags.cca_seen &&
> IS_ENABLED(CONFIG_ACPI_SUPPORT_CCA_ZERO))
> arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
Yes, that works.
> > I wonder how this is going to affect x86/ia64 too?
> >
>
> This should not affect x86 since arch_setup_dma_ops() is currently not
> implement for x86, and default to NOP (see include/linux/dma-mapping.h).
OK
> Also, on x86, _CCA is not required and default to 1 if missing.
Well, that's the point. :-)
--
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
On 5/6/2015 5:21 PM, Rafael J. Wysocki wrote:
>>>> > >>+ bool
>>>> > >>+
>>>> > >>+config ACPI_SUPPORT_CCA_ZERO
>>> > >
>>> > >I guess this means "we support devices that can DMA, but are not coherent".
>>> > >right?
>> >
>> >Yes, basically when _CCA=0.
> So what about
>
> ARCH_SUPPORT_CACHE_INCOHERENT_DMA
Since this is specific to ACPI _CCA, I just want to be clear with the
naming.
> or something similar?
>
>>>> > >>+ bool
>>>> > >>+
>>>> > >> config ACPI_SLEEP
>>>> > >> bool
>>>> > >> depends on SUSPEND || HIBERNATION
>>>> > >>diff --git a/drivers/acpi/acpi_platform.c b/drivers/acpi/acpi_platform.c
>>>> > >>index 4bf7559..a6feca4 100644
>>>> > >>--- a/drivers/acpi/acpi_platform.c
>>>> > >>+++ b/drivers/acpi/acpi_platform.c
>>>> > >>@@ -108,9 +108,11 @@ struct platform_device *acpi_create_platform_device(struct acpi_device *adev)
>>>> > >> if (IS_ERR(pdev))
>>>> > >> dev_err(&adev->dev, "platform device creation failed: %ld\n",
>>>> > >> PTR_ERR(pdev));
>>>> > >>- else
>>>> > >>+ else {
>>> > >
>>> > >Please add braces to both branches when making such changes (as per CodingStyle).
>>> > >
>> >
>> >OK.
>> >
>>>> > >>+ acpi_setup_device_dma(adev, &pdev->dev);
>>> > >
>>> > >Why do we need to do that here (for the second time)?
>> >
>> >Because we are calling:
>> > acpi_create_platform_device()
>> > |--> platform_device_register_device_full()
>> > |-->platform_device_alloc()
>> >
>> >This creates platform_device, which allocate a new platform_device->dev.
>> >This is not the same as the original acpi_device->dev that was created
>> >during acpi_add_single_object(). So, we have to set up the device
>> >coherency again.
> Ah, so the second arg is different now.
>
> Well, in that case, why do we need to set it up for the adev's dev member?
>
Just for sanity, since I don't know if adev->dev will be referenced
anywhere else. This way, it's consistent for all copied of struct device
generated.
Lemme know if you think that is unnecessary.
Thanks,
Suravee
Rafael,
Any comments on this patch?
Thanks,
Suravee
On 5/5/2015 10:12 AM, Suravee Suthikulpanit wrote:
> Currently, device drivers, which support both OF and ACPI,
> need to call two separate APIs, of_dma_is_coherent() and
> acpi_dma_is_coherent()) to determine device coherency attribute.
>
> This patch simplifies this process by introducing a new device
> property API, device_dma_is_coherent(), which calls the appropriate
> interface based on the booting architecture.
>
> Signed-off-by: Suravee Suthikulpanit <[email protected]>
> ---
> drivers/base/property.c | 12 ++++++++++++
> include/linux/property.h | 2 ++
> 2 files changed, 14 insertions(+)
>
> diff --git a/drivers/base/property.c b/drivers/base/property.c
> index 1d0b116..8123c6e 100644
> --- a/drivers/base/property.c
> +++ b/drivers/base/property.c
> @@ -14,6 +14,7 @@
> #include <linux/export.h>
> #include <linux/kernel.h>
> #include <linux/of.h>
> +#include <linux/of_address.h>
> #include <linux/property.h>
>
> /**
> @@ -519,3 +520,14 @@ unsigned int device_get_child_node_count(struct device *dev)
> return count;
> }
> EXPORT_SYMBOL_GPL(device_get_child_node_count);
> +
> +bool device_dma_is_coherent(struct device *dev)
> +{
> + if (IS_ENABLED(CONFIG_OF) && dev->of_node)
> + return of_dma_is_coherent(dev->of_node);
> + else if (has_acpi_companion(dev))
> + return acpi_dma_is_coherent(acpi_node(dev->fwnode));
> +
> + return false;
> +}
> +EXPORT_SYMBOL_GPL(device_dma_is_coherent);
> diff --git a/include/linux/property.h b/include/linux/property.h
> index de8bdf4..76ebde9 100644
> --- a/include/linux/property.h
> +++ b/include/linux/property.h
> @@ -164,4 +164,6 @@ struct property_set {
>
> void device_add_property_set(struct device *dev, struct property_set *pset);
>
> +bool device_dma_is_coherent(struct device *dev);
> +
> #endif /* _LINUX_PROPERTY_H_ */
>
On Thu, May 7, 2015 at 1:52 AM, Suravee Suthikulanit
<[email protected]> wrote:
> Rafael,
>
> Any comments on this patch?
Well, as long as acpi_dma_is_coherent() does the right thing on all
architectures, I have no objections.
Thanks,
Rafael
> On 5/5/2015 10:12 AM, Suravee Suthikulpanit wrote:
>>
>> Currently, device drivers, which support both OF and ACPI,
>> need to call two separate APIs, of_dma_is_coherent() and
>> acpi_dma_is_coherent()) to determine device coherency attribute.
>>
>> This patch simplifies this process by introducing a new device
>> property API, device_dma_is_coherent(), which calls the appropriate
>> interface based on the booting architecture.
>>
>> Signed-off-by: Suravee Suthikulpanit <[email protected]>
>> ---
>> drivers/base/property.c | 12 ++++++++++++
>> include/linux/property.h | 2 ++
>> 2 files changed, 14 insertions(+)
>>
>> diff --git a/drivers/base/property.c b/drivers/base/property.c
>> index 1d0b116..8123c6e 100644
>> --- a/drivers/base/property.c
>> +++ b/drivers/base/property.c
>> @@ -14,6 +14,7 @@
>> #include <linux/export.h>
>> #include <linux/kernel.h>
>> #include <linux/of.h>
>> +#include <linux/of_address.h>
>> #include <linux/property.h>
>>
>> /**
>> @@ -519,3 +520,14 @@ unsigned int device_get_child_node_count(struct
>> device *dev)
>> return count;
>> }
>> EXPORT_SYMBOL_GPL(device_get_child_node_count);
>> +
>> +bool device_dma_is_coherent(struct device *dev)
>> +{
>> + if (IS_ENABLED(CONFIG_OF) && dev->of_node)
>> + return of_dma_is_coherent(dev->of_node);
>> + else if (has_acpi_companion(dev))
>> + return acpi_dma_is_coherent(acpi_node(dev->fwnode));
>> +
>> + return false;
>> +}
>> +EXPORT_SYMBOL_GPL(device_dma_is_coherent);
>> diff --git a/include/linux/property.h b/include/linux/property.h
>> index de8bdf4..76ebde9 100644
>> --- a/include/linux/property.h
>> +++ b/include/linux/property.h
>> @@ -164,4 +164,6 @@ struct property_set {
>>
>> void device_add_property_set(struct device *dev, struct property_set
>> *pset);
>>
>> +bool device_dma_is_coherent(struct device *dev);
>> +
>> #endif /* _LINUX_PROPERTY_H_ */
>>
>
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wednesday 06 May 2015 17:16:35 Suravee Suthikulanit wrote:
> On 5/6/2015 5:21 PM, Rafael J. Wysocki wrote:
> >>>> > >>+ bool
> >>>> > >>+
> >>>> > >>+config ACPI_SUPPORT_CCA_ZERO
> >>> > >
> >>> > >I guess this means "we support devices that can DMA, but are not coherent".
> >>> > >right?
> >> >
> >> >Yes, basically when _CCA=0.
> > So what about
> >
> > ARCH_SUPPORT_CACHE_INCOHERENT_DMA
>
> Since this is specific to ACPI _CCA, I just want to be clear with the
> naming.
How about directly using the architecture names here, this is inherently
architecture specific, and it's more likely that if another architecture
gets added in the future that it will have other requirements.
Arnd
On Thursday, May 07, 2015 11:07:08 AM Arnd Bergmann wrote:
> On Wednesday 06 May 2015 17:16:35 Suravee Suthikulanit wrote:
> > On 5/6/2015 5:21 PM, Rafael J. Wysocki wrote:
> > >>>> > >>+ bool
> > >>>> > >>+
> > >>>> > >>+config ACPI_SUPPORT_CCA_ZERO
> > >>> > >
> > >>> > >I guess this means "we support devices that can DMA, but are not coherent".
> > >>> > >right?
> > >> >
> > >> >Yes, basically when _CCA=0.
> > > So what about
> > >
> > > ARCH_SUPPORT_CACHE_INCOHERENT_DMA
> >
> > Since this is specific to ACPI _CCA, I just want to be clear with the
> > naming.
>
> How about directly using the architecture names here, this is inherently
> architecture specific, and it's more likely that if another architecture
> gets added in the future that it will have other requirements.
Sounds reasonable to me.
Rafael