2020-11-17 20:46:40

by Ernst, Justin

[permalink] [raw]
Subject: [PATCH 0/5] Add uv_sysfs platform driver

Introduce a new platform driver to gather topology information from UV systems
and expose that information via a sysfs interface at /sys/firmware/sgi_uv/.

Justin Ernst (5):
x86/platform/uv: Remove existing /sys/firmware/sgi_uv/ interface
x86/platform/uv: Add and export uv_bios_* functions
x86/platform/uv: Add new uv_sysfs platform driver
x86/platform/uv: Update ABI documentation of /sys/firmware/sgi_uv/
x86/platform/uv: Update MAINTAINERS for uv_sysfs driver

.../ABI/testing/sysfs-firmware-sgi_uv | 137 ++-
MAINTAINERS | 6 +
arch/x86/include/asm/uv/bios.h | 49 +
arch/x86/include/asm/uv/uv_geo.h | 103 +++
arch/x86/platform/uv/Makefile | 2 +-
arch/x86/platform/uv/bios_uv.c | 54 ++
arch/x86/platform/uv/uv_sysfs.c | 63 --
drivers/platform/x86/Kconfig | 11 +
drivers/platform/x86/Makefile | 3 +
drivers/platform/x86/uv_sysfs.c | 853 ++++++++++++++++++
10 files changed, 1202 insertions(+), 79 deletions(-)
create mode 100644 arch/x86/include/asm/uv/uv_geo.h
delete mode 100644 arch/x86/platform/uv/uv_sysfs.c
create mode 100644 drivers/platform/x86/uv_sysfs.c


base-commit: 4ef8451b332662d004df269d4cdeb7d9f31419b5
--
2.26.2


2020-11-17 20:46:46

by Ernst, Justin

[permalink] [raw]
Subject: [PATCH 5/5] x86/platform/uv: Update MAINTAINERS for uv_sysfs driver

Add an entry and email address for the new uv_sysfs driver and
its maintainer.

Signed-off-by: Justin Ernst <[email protected]>
Acked-by: Steve Wahl <[email protected]>
---
MAINTAINERS | 6 ++++++
1 file changed, 6 insertions(+)

diff --git a/MAINTAINERS b/MAINTAINERS
index b43b59542d15..f693d2d97203 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -18361,6 +18361,12 @@ F: include/uapi/linux/uuid.h
F: lib/test_uuid.c
F: lib/uuid.c

+UV SYSFS DRIVER
+M: Justin Ernst <[email protected]>
+L: [email protected]
+S: Maintained
+F: drivers/platform/x86/uv_sysfs.c
+
UVESAFB DRIVER
M: Michal Januszewski <[email protected]>
L: [email protected]
--
2.26.2

2020-11-17 20:46:53

by Ernst, Justin

[permalink] [raw]
Subject: [PATCH 4/5] x86/platform/uv: Update ABI documentation of /sys/firmware/sgi_uv/

Update the ABI documentation to describe the sysfs interface provided by
the new uv_sysfs platform driver.

Signed-off-by: Justin Ernst <[email protected]>
Reviewed-by: Steve Wahl <[email protected]>
---
.../ABI/testing/sysfs-firmware-sgi_uv | 137 ++++++++++++++++--
1 file changed, 122 insertions(+), 15 deletions(-)

diff --git a/Documentation/ABI/testing/sysfs-firmware-sgi_uv b/Documentation/ABI/testing/sysfs-firmware-sgi_uv
index 66800baab096..ca5cdccdb1a8 100644
--- a/Documentation/ABI/testing/sysfs-firmware-sgi_uv
+++ b/Documentation/ABI/testing/sysfs-firmware-sgi_uv
@@ -1,27 +1,134 @@
What: /sys/firmware/sgi_uv/
-Date: August 2008
-Contact: Russ Anderson <[email protected]>
+Date: September 2020
+Contact: Justin Ernst <[email protected]>
Description:
The /sys/firmware/sgi_uv directory contains information
- about the SGI UV platform.
+ about the UV platform.

- Under that directory are a number of files::
+ Under that directory are a number of read-only attributes:

partition_id
- coherence_id
+ uv_type

The partition_id entry contains the partition id.
- SGI UV systems can be partitioned into multiple physical
+ UV systems can be partitioned into multiple physical
machines, which each partition running a unique copy
- of the operating system. Each partition will have a unique
- partition id. To display the partition id, use the command::
+ of the operating system. Each partition will have a unique
+ partition id.

- cat /sys/firmware/sgi_uv/partition_id
+ The uv_type entry contains the hub revision number.
+ This value can be used to identify the UV system version:
+ "3.0" = UV2
+ "5.0" = UV3
+ "7.0" = UV4
+ "7.1" = UV4a
+ "9.0" = UV5

- The coherence_id entry contains the coherence id.
- A partitioned SGI UV system can have one or more coherence
- domain. The coherence id indicates which coherence domain
- this partition is in. To display the coherence id, use the
- command::
+ The /sys/firmware/sgi_uv directory also contains two directories:

- cat /sys/firmware/sgi_uv/coherence_id
+ hubs/
+ pcibuses/
+
+ The hubs directory contains a number of hub objects, each representing
+ a UV Hub visible to the BIOS. Each hub object's name is appended by a
+ unique ordinal value (ex. /sys/firmware/sgi_uv/hubs/hub_5)
+
+ Each hub object directory contains a number of read-only attributes:
+
+ cnode
+ location
+ name
+ nasid
+ shared
+ this_partition
+
+ The cnode entry contains the cnode number of the corresponding hub.
+ If a cnode value is not applicable, the value returned will be -1.
+
+ The location entry contains the location string of the corresponding hub.
+ This value is used to physically identify a hub within a system.
+
+ The name entry contains the name of the corresponding hub. This name can
+ be two variants:
+ "UVHub x.x" = A 'node' ASIC, connecting a CPU to the interconnect
+ fabric. The 'x.x' value represents the ASIC revision.
+ (ex. 'UVHub 5.0')
+ "NLxRouter" = A 'router ASIC, only connecting other ASICs to
+ the interconnect fabric. The 'x' value representing
+ the fabric technology version. (ex. 'NL8Router')
+
+ The nasid entry contains the nasid number of the corresponding hub.
+ If a nasid value is not applicable, the value returned will be -1.
+
+ The shared entry contains a boolean value describing whether the
+ corresponding hub is shared between system partitions.
+
+ The this_partition entry contains a boolean value describing whether
+ the corresponding hub is local to the current partition.
+
+ Each hub object directory also contains a number of port objects,
+ each representing a fabric port on the corresponding hub.
+ A port object's name is appended by a unique ordinal value
+ (ex. /sys/firmware/sgi_uv/hubs/hub_5/port_3)
+
+ Each port object directory contains a number of read-only attributes:
+
+ conn_hub
+ conn_port
+
+ The conn_hub entry contains a value representing the unique
+ oridinal value of the hub on the other end of the fabric
+ cable plugged into the port. If the port is disconnected,
+ the value returned will be -1.
+
+ The conn_port entry contains a value representing the unique
+ oridinal value of the port on the other end of the fabric cable
+ plugged into the port. If the port is disconnected, the value
+ returned will be -1.
+
+ Ex:
+ A value of '3' is read from:
+ /sys/firmware/sgi_uv/hubs/hub_5/port_3/conn_hub
+
+ and a value of '6' is read from:
+ /sys/firmware/sgi_uv/hubs/hub_5/port_3/conn_port
+
+ representing that this port is connected to:
+ /sys/firmware/sgi_uv/hubs/hub_3/port_6
+
+
+ The pcibuses directory contains a number of PCI bus objects.
+ Each PCI bus object's name is appended by its PCI bus address.
+ (ex. pcibus_0003:80)
+
+ Each pcibus object has a number of possible read-only attributes:
+
+ type
+ location
+ slot
+ ppb_addr
+ iio_stack
+
+ The type entry contains a value describing the type of IO at
+ the corresponding PCI bus address. Known possible values
+ across all UV versions are:
+ BASE IO
+ PCIe IO
+ PCIe SLOT
+ NODE IO
+ Riser
+ PPB
+
+ The location entry contains the location string of the UV Hub
+ of the CPU physically connected to the corresponding PCI bus.
+
+ The slot entry contains the physical slot number of the
+ corresponding PCI bus. This value is used to physically locate
+ PCI cards within a system.
+
+ The ppb_addr entry contains the PCI address string of the
+ bridged PCI bus. This entry is only present when the PCI bus
+ object type is 'PPB'.
+
+ The iio_stack entry contains a value describing the IIO stack
+ number that the corresponding PCI bus object is connected to.
--
2.26.2

2020-11-17 20:49:03

by Ernst, Justin

[permalink] [raw]
Subject: [PATCH 3/5] x86/platform/uv: Add new uv_sysfs platform driver

Add the uv_sysfs driver to construct a read-only sysfs interface at
/sys/firmware/sgi_uv/ to expose information gathered from UV BIOS.
This information includes:
UV Hub descriptions, including physical location
Cabling layout between hubs on the fabric
PCI topology, including physical location of PCI cards

Together, the information provides a robust physical description of a
UV system, useful for correlating to performance data or performing
remote support.

Signed-off-by: Justin Ernst <[email protected]>
Reviewed-by: Steve Wahl <[email protected]>
---
drivers/platform/x86/Kconfig | 11 +
drivers/platform/x86/Makefile | 3 +
drivers/platform/x86/uv_sysfs.c | 853 ++++++++++++++++++++++++++++++++
3 files changed, 867 insertions(+)
create mode 100644 drivers/platform/x86/uv_sysfs.c

diff --git a/drivers/platform/x86/Kconfig b/drivers/platform/x86/Kconfig
index 0d91d136bc3b..ba34153571b8 100644
--- a/drivers/platform/x86/Kconfig
+++ b/drivers/platform/x86/Kconfig
@@ -78,6 +78,17 @@ config HUAWEI_WMI
To compile this driver as a module, choose M here: the module
will be called huawei-wmi.

+config UV_SYSFS
+ tristate "Sysfs structure for UV systems"
+ depends on X86_UV
+ depends on SYSFS
+ help
+ This driver supports a sysfs tree describing information about
+ UV systems at /sys/firmware/sgi_uv/.
+
+ To compile this driver as a module, choose M here: the module will
+ be called uv_sysfs.
+
config INTEL_WMI_SBL_FW_UPDATE
tristate "Intel WMI Slim Bootloader firmware update signaling driver"
depends on ACPI_WMI
diff --git a/drivers/platform/x86/Makefile b/drivers/platform/x86/Makefile
index 5f823f7eff45..a34875d833dd 100644
--- a/drivers/platform/x86/Makefile
+++ b/drivers/platform/x86/Makefile
@@ -62,6 +62,9 @@ obj-$(CONFIG_HP_WIRELESS) += hp-wireless.o
obj-$(CONFIG_HP_WMI) += hp-wmi.o
obj-$(CONFIG_TC1100_WMI) += tc1100-wmi.o

+# Hewlett Packard Enterprise
+obj-$(CONFIG_UV_SYSFS) += uv_sysfs.o
+
# IBM Thinkpad and Lenovo
obj-$(CONFIG_IBM_RTL) += ibm_rtl.o
obj-$(CONFIG_IDEAPAD_LAPTOP) += ideapad-laptop.o
diff --git a/drivers/platform/x86/uv_sysfs.c b/drivers/platform/x86/uv_sysfs.c
new file mode 100644
index 000000000000..6de360370f0e
--- /dev/null
+++ b/drivers/platform/x86/uv_sysfs.c
@@ -0,0 +1,853 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * This file supports the /sys/firmware/sgi_uv topology tree on HPE UV.
+ *
+ * Copyright (c) 2020 Hewlett Packard Enterprise. All Rights Reserved.
+ * Copyright (c) Justin Ernst
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/device.h>
+#include <linux/slab.h>
+#include <linux/kobject.h>
+#include <asm/uv/bios.h>
+#include <asm/uv/uv.h>
+#include <asm/uv/uv_hub.h>
+#include <asm/uv/uv_geo.h>
+
+#define INVALID_CNODE -1
+
+struct kobject *sgi_uv_kobj;
+struct kset *uv_pcibus_kset;
+struct kset *uv_hubs_kset;
+static struct uv_bios_hub_info *hub_buf;
+static struct uv_bios_port_info **port_buf;
+static struct uv_hub **uv_hubs;
+static struct uv_pci_top_obj **uv_pci_objs;
+static int num_pci_lines;
+static int num_cnodes;
+static int *prev_obj_to_cnode;
+static int uv_bios_obj_cnt;
+static signed short uv_master_nasid = -1;
+static void *uv_biosheap;
+
+static const char *uv_type_string(void)
+{
+ if (is_uv5_hub())
+ return "9.0";
+ else if (is_uv4a_hub())
+ return "7.1";
+ else if (is_uv4_hub())
+ return "7.0";
+ else if (is_uv3_hub())
+ return "5.0";
+ else if (is_uv2_hub())
+ return "3.0";
+ else
+ return "unknown";
+}
+
+static int ordinal_to_nasid(int ordinal)
+{
+ if (ordinal < num_cnodes && ordinal >= 0)
+ return UV_PNODE_TO_NASID(uv_blade_to_pnode(ordinal));
+ else
+ return -1;
+}
+
+static union geoid_u cnode_to_geoid(int cnode)
+{
+ union geoid_u geoid;
+
+ uv_bios_get_geoinfo(ordinal_to_nasid(cnode), (u64)sizeof(union geoid_u), (u64 *)&geoid);
+ return geoid;
+}
+
+static int location_to_bpos(char *location, int *rack, int *slot, int *blade)
+{
+ char type, r, b, h;
+ int idb, idh;
+
+ if (sscanf(location, "%c%03d%c%02d%c%2d%c%d",
+ &r, rack, &type, slot, &b, &idb, &h, &idh) != 8)
+ return -1;
+ *blade = idb * 2 + idh;
+
+ return 0;
+}
+
+static int cache_obj_to_cnode(struct uv_bios_hub_info *obj)
+{
+ int cnode;
+ union geoid_u geoid;
+ int obj_rack, obj_slot, obj_blade;
+ int rack, slot, blade;
+
+ if (!obj->f.fields.this_part && !obj->f.fields.is_shared)
+ return 0;
+
+ if (location_to_bpos(obj->location, &obj_rack, &obj_slot, &obj_blade))
+ return -1;
+
+ for (cnode = 0; cnode < num_cnodes; cnode++) {
+ geoid = cnode_to_geoid(cnode);
+ rack = geo_rack(geoid);
+ slot = geo_slot(geoid);
+ blade = geo_blade(geoid);
+ if (obj_rack == rack && obj_slot == slot && obj_blade == blade)
+ prev_obj_to_cnode[obj->id] = cnode;
+ }
+
+ return 0;
+}
+
+static int get_obj_to_cnode(int obj_id)
+{
+ return prev_obj_to_cnode[obj_id];
+}
+
+struct uv_hub {
+ struct kobject kobj;
+ struct uv_bios_hub_info *hub_info;
+ struct uv_port **ports;
+};
+
+#define to_uv_hub(kobj_ptr) container_of(kobj_ptr, struct uv_hub, kobj)
+
+static ssize_t hub_name_show(struct uv_bios_hub_info *hub_info, char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", hub_info->name);
+}
+
+static ssize_t hub_location_show(struct uv_bios_hub_info *hub_info, char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", hub_info->location);
+}
+
+static ssize_t hub_partition_show(struct uv_bios_hub_info *hub_info, char *buf)
+{
+ return sprintf(buf, "%d\n", hub_info->f.fields.this_part);
+}
+
+static ssize_t hub_shared_show(struct uv_bios_hub_info *hub_info, char *buf)
+{
+ return sprintf(buf, "%d\n", hub_info->f.fields.is_shared);
+}
+static ssize_t hub_nasid_show(struct uv_bios_hub_info *hub_info, char *buf)
+{
+ int cnode = get_obj_to_cnode(hub_info->id);
+
+ return sprintf(buf, "%d\n", ordinal_to_nasid(cnode));
+}
+static ssize_t hub_cnode_show(struct uv_bios_hub_info *hub_info, char *buf)
+{
+ return sprintf(buf, "%d\n", get_obj_to_cnode(hub_info->id));
+}
+
+struct hub_sysfs_entry {
+ struct attribute attr;
+ ssize_t (*show)(struct uv_bios_hub_info *hub_info, char *buf);
+ ssize_t (*store)(struct uv_bios_hub_info *hub_info, const char *buf, size_t sz);
+};
+
+static struct hub_sysfs_entry name_attribute =
+ __ATTR(name, 0444, hub_name_show, NULL);
+static struct hub_sysfs_entry location_attribute =
+ __ATTR(location, 0444, hub_location_show, NULL);
+static struct hub_sysfs_entry partition_attribute =
+ __ATTR(this_partition, 0444, hub_partition_show, NULL);
+static struct hub_sysfs_entry shared_attribute =
+ __ATTR(shared, 0444, hub_shared_show, NULL);
+static struct hub_sysfs_entry nasid_attribute =
+ __ATTR(nasid, 0444, hub_nasid_show, NULL);
+static struct hub_sysfs_entry cnode_attribute =
+ __ATTR(cnode, 0444, hub_cnode_show, NULL);
+
+static struct attribute *uv_hub_attrs[] = {
+ &name_attribute.attr,
+ &location_attribute.attr,
+ &partition_attribute.attr,
+ &shared_attribute.attr,
+ &nasid_attribute.attr,
+ &cnode_attribute.attr,
+ NULL,
+};
+
+static void hub_release(struct kobject *kobj)
+{
+ struct uv_hub *hub = to_uv_hub(kobj);
+
+ kfree(hub);
+}
+
+static ssize_t hub_type_show(struct kobject *kobj, struct attribute *attr,
+ char *buf)
+{
+ struct uv_hub *hub = to_uv_hub(kobj);
+ struct uv_bios_hub_info *bios_hub_info = hub->hub_info;
+ struct hub_sysfs_entry *entry;
+
+ entry = container_of(attr, struct hub_sysfs_entry, attr);
+
+ if (!entry->show)
+ return -EIO;
+
+ return entry->show(bios_hub_info, buf);
+}
+
+static const struct sysfs_ops hub_sysfs_ops = {
+ .show = hub_type_show,
+};
+
+static struct kobj_type hub_attr_type = {
+ .release = hub_release,
+ .sysfs_ops = &hub_sysfs_ops,
+ .default_attrs = uv_hub_attrs,
+};
+
+static int uv_hubs_init(void)
+{
+ s64 biosr;
+ u64 sz;
+ int i, ret;
+
+ prev_obj_to_cnode = kmalloc_array(uv_bios_obj_cnt, sizeof(*prev_obj_to_cnode),
+ GFP_KERNEL);
+ if (!prev_obj_to_cnode)
+ return -ENOMEM;
+
+ for (i = 0; i < uv_bios_obj_cnt; i++)
+ prev_obj_to_cnode[i] = INVALID_CNODE;
+
+ uv_hubs_kset = kset_create_and_add("hubs", NULL, sgi_uv_kobj);
+ if (!uv_hubs_kset) {
+ ret = -ENOMEM;
+ goto err_hubs_kset;
+ }
+ sz = uv_bios_obj_cnt * sizeof(*hub_buf);
+ hub_buf = kzalloc(sz, GFP_KERNEL);
+ if (!hub_buf) {
+ ret = -ENOMEM;
+ goto err_hub_buf;
+ }
+
+ biosr = uv_bios_enum_objs((u64)uv_master_nasid, sz, (u64 *)hub_buf);
+ if (biosr) {
+ ret = -EINVAL;
+ goto err_enum_objs;
+ }
+
+ uv_hubs = kcalloc(uv_bios_obj_cnt, sizeof(*uv_hubs), GFP_KERNEL);
+ if (!uv_hubs) {
+ ret = -ENOMEM;
+ goto err_enum_objs;
+ }
+
+ for (i = 0; i < uv_bios_obj_cnt; i++) {
+ uv_hubs[i] = kzalloc(sizeof(*uv_hubs[i]), GFP_KERNEL);
+ if (!uv_hubs[i]) {
+ i--;
+ goto err_hubs;
+ }
+
+ uv_hubs[i]->hub_info = &hub_buf[i];
+ cache_obj_to_cnode(uv_hubs[i]->hub_info);
+
+ uv_hubs[i]->kobj.kset = uv_hubs_kset;
+
+ ret = kobject_init_and_add(&uv_hubs[i]->kobj, &hub_attr_type,
+ NULL, "hub_%u", hub_buf[i].id);
+ if (ret)
+ goto err_hubs;
+ kobject_uevent(&uv_hubs[i]->kobj, KOBJ_ADD);
+ }
+ return 0;
+
+err_hubs:
+ for (; i >= 0; i--)
+ kobject_put(&uv_hubs[i]->kobj);
+ kfree(uv_hubs);
+err_enum_objs:
+ kfree(hub_buf);
+err_hub_buf:
+ kset_unregister(uv_hubs_kset);
+err_hubs_kset:
+ kfree(prev_obj_to_cnode);
+ return ret;
+
+}
+
+static void uv_hubs_exit(void)
+{
+ int i;
+
+ for (i = 0; i < uv_bios_obj_cnt; i++)
+ kobject_put(&uv_hubs[i]->kobj);
+
+ kfree(uv_hubs);
+ kfree(hub_buf);
+ kset_unregister(uv_hubs_kset);
+ kfree(prev_obj_to_cnode);
+}
+
+struct uv_port {
+ struct kobject kobj;
+ struct uv_bios_port_info *port_info;
+};
+
+#define to_uv_port(kobj_ptr) container_of(kobj_ptr, struct uv_port, kobj)
+
+static ssize_t uv_port_conn_hub_show(struct uv_bios_port_info *port, char *buf)
+{
+ return sprintf(buf, "%d\n", port->conn_id);
+}
+
+static ssize_t uv_port_conn_port_show(struct uv_bios_port_info *port, char *buf)
+{
+ return sprintf(buf, "%d\n", port->conn_port);
+}
+
+struct uv_port_sysfs_entry {
+ struct attribute attr;
+ ssize_t (*show)(struct uv_bios_port_info *port_info, char *buf);
+ ssize_t (*store)(struct uv_bios_port_info *port_info, const char *buf, size_t size);
+};
+
+static struct uv_port_sysfs_entry uv_port_conn_hub_attribute =
+ __ATTR(conn_hub, 0444, uv_port_conn_hub_show, NULL);
+static struct uv_port_sysfs_entry uv_port_conn_port_attribute =
+ __ATTR(conn_port, 0444, uv_port_conn_port_show, NULL);
+
+static struct attribute *uv_port_attrs[] = {
+ &uv_port_conn_hub_attribute.attr,
+ &uv_port_conn_port_attribute.attr,
+ NULL,
+};
+
+static void uv_port_release(struct kobject *kobj)
+{
+ struct uv_port *port = to_uv_port(kobj);
+
+ kfree(port);
+}
+
+static ssize_t uv_port_type_show(struct kobject *kobj, struct attribute *attr,
+ char *buf)
+{
+ struct uv_port *port = to_uv_port(kobj);
+ struct uv_bios_port_info *port_info = port->port_info;
+ struct uv_port_sysfs_entry *entry;
+
+ entry = container_of(attr, struct uv_port_sysfs_entry, attr);
+
+ if (!entry->show)
+ return -EIO;
+
+ return entry->show(port_info, buf);
+}
+
+static const struct sysfs_ops uv_port_sysfs_ops = {
+ .show = uv_port_type_show,
+};
+
+static struct kobj_type uv_port_attr_type = {
+ .release = uv_port_release,
+ .sysfs_ops = &uv_port_sysfs_ops,
+ .default_attrs = uv_port_attrs,
+};
+
+static int uv_ports_init(void)
+{
+ s64 biosr;
+ int j = 0, k = 0, ret, sz;
+
+ port_buf = kcalloc(uv_bios_obj_cnt, sizeof(*port_buf), GFP_KERNEL);
+ if (!port_buf)
+ return -ENOMEM;
+
+ for (j = 0; j < uv_bios_obj_cnt; j++) {
+ sz = hub_buf[j].ports * sizeof(*port_buf[j]);
+ port_buf[j] = kzalloc(sz, GFP_KERNEL);
+ if (!port_buf[j]) {
+ ret = -ENOMEM;
+ j--;
+ goto err_port_info;
+ }
+ biosr = uv_bios_enum_ports((u64)uv_master_nasid, (u64)hub_buf[j].id, sz,
+ (u64 *)port_buf[j]);
+ if (biosr) {
+ ret = -EINVAL;
+ goto err_port_info;
+ }
+ }
+ for (j = 0; j < uv_bios_obj_cnt; j++) {
+ uv_hubs[j]->ports = kcalloc(hub_buf[j].ports,
+ sizeof(*uv_hubs[j]->ports), GFP_KERNEL);
+ if (!uv_hubs[j]->ports) {
+ ret = -ENOMEM;
+ j--;
+ goto err_ports;
+ }
+ }
+ for (j = 0; j < uv_bios_obj_cnt; j++) {
+ for (k = 0; k < hub_buf[j].ports; k++) {
+ uv_hubs[j]->ports[k] = kzalloc(sizeof(*uv_hubs[j]->ports[k]), GFP_KERNEL);
+ if (!uv_hubs[j]->ports[k]) {
+ ret = -ENOMEM;
+ k--;
+ goto err_kobj_ports;
+ }
+ uv_hubs[j]->ports[k]->port_info = &port_buf[j][k];
+ ret = kobject_init_and_add(&uv_hubs[j]->ports[k]->kobj, &uv_port_attr_type,
+ &uv_hubs[j]->kobj, "port_%d", port_buf[j][k].port);
+ if (ret)
+ goto err_kobj_ports;
+ kobject_uevent(&uv_hubs[j]->ports[k]->kobj, KOBJ_ADD);
+ }
+ }
+ return 0;
+
+err_kobj_ports:
+ for (; j >= 0; j--) {
+ for (; k >= 0; k--)
+ kobject_put(&uv_hubs[j]->ports[k]->kobj);
+ if (j > 0)
+ k = hub_buf[j-1].ports - 1;
+ }
+ j = uv_bios_obj_cnt - 1;
+err_ports:
+ for (; j >= 0; j--)
+ kfree(uv_hubs[j]->ports);
+ j = uv_bios_obj_cnt - 1;
+err_port_info:
+ for (; j >= 0; j--)
+ kfree(port_buf[j]);
+ kfree(port_buf);
+ return ret;
+}
+
+static void uv_ports_exit(void)
+{
+ int j, k;
+
+ for (j = 0; j < uv_bios_obj_cnt; j++) {
+ for (k = hub_buf[j].ports; k >= 0; k--)
+ kobject_put(&uv_hubs[j]->ports[k]->kobj);
+ }
+ for (j = 0; j < uv_bios_obj_cnt; j++) {
+ kfree(uv_hubs[j]->ports);
+ kfree(port_buf[j]);
+ }
+ kfree(port_buf);
+}
+
+struct uv_pci_top_obj {
+ struct kobject kobj;
+ char *type;
+ char *location;
+ int iio_stack;
+ char *ppb_addr;
+ int slot;
+};
+
+#define to_uv_pci_top_obj(kobj_ptr) container_of(kobj_ptr, struct uv_pci_top_obj, kobj)
+
+static ssize_t uv_pci_type_show(struct uv_pci_top_obj *top_obj, char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", top_obj->type);
+}
+
+static ssize_t uv_pci_location_show(struct uv_pci_top_obj *top_obj, char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", top_obj->location);
+}
+
+static ssize_t uv_pci_iio_stack_show(struct uv_pci_top_obj *top_obj, char *buf)
+{
+ return sprintf(buf, "%d\n", top_obj->iio_stack);
+}
+
+static ssize_t uv_pci_ppb_addr_show(struct uv_pci_top_obj *top_obj, char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", top_obj->ppb_addr);
+}
+
+static ssize_t uv_pci_slot_show(struct uv_pci_top_obj *top_obj, char *buf)
+{
+ return sprintf(buf, "%d\n", top_obj->slot);
+}
+
+struct uv_pci_top_sysfs_entry {
+ struct attribute attr;
+ ssize_t (*show)(struct uv_pci_top_obj *top_obj, char *buf);
+ ssize_t (*store)(struct uv_pci_top_obj *top_obj, const char *buf, size_t size);
+};
+
+static struct uv_pci_top_sysfs_entry uv_pci_type_attribute =
+ __ATTR(type, 0444, uv_pci_type_show, NULL);
+static struct uv_pci_top_sysfs_entry uv_pci_location_attribute =
+ __ATTR(location, 0444, uv_pci_location_show, NULL);
+static struct uv_pci_top_sysfs_entry uv_pci_iio_stack_attribute =
+ __ATTR(iio_stack, 0444, uv_pci_iio_stack_show, NULL);
+static struct uv_pci_top_sysfs_entry uv_pci_ppb_addr_attribute =
+ __ATTR(ppb_addr, 0444, uv_pci_ppb_addr_show, NULL);
+static struct uv_pci_top_sysfs_entry uv_pci_slot_attribute =
+ __ATTR(slot, 0444, uv_pci_slot_show, NULL);
+
+static void uv_pci_top_release(struct kobject *kobj)
+{
+ struct uv_pci_top_obj *top_obj = to_uv_pci_top_obj(kobj);
+
+ kfree(top_obj->type);
+ kfree(top_obj->location);
+ kfree(top_obj->ppb_addr);
+ kfree(top_obj);
+}
+
+static ssize_t pci_top_type_show(struct kobject *kobj,
+ struct attribute *attr, char *buf)
+{
+ struct uv_pci_top_obj *top_obj = to_uv_pci_top_obj(kobj);
+ struct uv_pci_top_sysfs_entry *entry;
+
+ entry = container_of(attr, struct uv_pci_top_sysfs_entry, attr);
+
+ if (!entry->show)
+ return -EIO;
+
+ return entry->show(top_obj, buf);
+}
+
+static const struct sysfs_ops uv_pci_top_sysfs_ops = {
+ .show = pci_top_type_show,
+};
+
+static struct kobj_type uv_pci_top_attr_type = {
+ .release = uv_pci_top_release,
+ .sysfs_ops = &uv_pci_top_sysfs_ops,
+};
+
+static int init_pci_top_obj(struct uv_pci_top_obj *top_obj, char *line)
+{
+ char *start;
+ char type[11], location[14], ppb_addr[15];
+ int str_cnt, ret;
+ unsigned int tmp_match[2];
+
+ // Minimum line length
+ if (strlen(line) < 36)
+ return -EINVAL;
+
+ //Line must match format "pcibus %4x:%2x" to be valid
+ str_cnt = sscanf(line, "pcibus %4x:%2x", &tmp_match[0], &tmp_match[1]);
+ if (str_cnt < 2)
+ return -EINVAL;
+
+ /* Connect pcibus to segment:bus number with '_'
+ * to concatenate name tokens.
+ * pcibus 0000:00 ... -> pcibus_0000:00 ...
+ */
+ line[6] = '_';
+
+ /* Null terminate after the concatencated name tokens
+ * to produce kobj name string.
+ */
+ line[14] = '\0';
+
+ // Use start to index after name tokens string for remainder of line info.
+ start = &line[15];
+
+ top_obj->iio_stack = -1;
+ top_obj->slot = -1;
+
+ /* r001i01b00h0 BASE IO (IIO Stack 0)
+ * r001i01b00h1 PCIe IO (IIO Stack 1)
+ * r001i01b03h1 PCIe SLOT
+ * r001i01b00h0 NODE IO
+ * r001i01b00h0 Riser
+ * (IIO Stack #) may not be present.
+ */
+ if (start[0] == 'r') {
+ str_cnt = sscanf(start, "%13s %10[^(] %*s %*s %d)",
+ location, type, &top_obj->iio_stack);
+ if (str_cnt < 2)
+ return -EINVAL;
+ top_obj->type = kstrdup(type, GFP_KERNEL);
+ if (!top_obj->type)
+ return -ENOMEM;
+ top_obj->location = kstrdup(location, GFP_KERNEL);
+ if (!top_obj->location) {
+ kfree(top_obj->type);
+ return -ENOMEM;
+ }
+ }
+ /* PPB at 0000:80:00.00 (slot 3)
+ * (slot #) may not be present.
+ */
+ else if (start[0] == 'P') {
+ str_cnt = sscanf(start, "%10s %*s %14s %*s %d)",
+ type, ppb_addr, &top_obj->slot);
+ if (str_cnt < 2)
+ return -EINVAL;
+ top_obj->type = kstrdup(type, GFP_KERNEL);
+ if (!top_obj->type)
+ return -ENOMEM;
+ top_obj->ppb_addr = kstrdup(ppb_addr, GFP_KERNEL);
+ if (!top_obj->ppb_addr) {
+ kfree(top_obj->type);
+ return -ENOMEM;
+ }
+ } else
+ return -EINVAL;
+
+ top_obj->kobj.kset = uv_pcibus_kset;
+
+ ret = kobject_init_and_add(&top_obj->kobj, &uv_pci_top_attr_type, NULL, "%s", line);
+ if (ret)
+ goto err_add_sysfs;
+
+ if (top_obj->type) {
+ ret = sysfs_create_file(&top_obj->kobj, &uv_pci_type_attribute.attr);
+ if (ret)
+ goto err_add_sysfs;
+ }
+ if (top_obj->location) {
+ ret = sysfs_create_file(&top_obj->kobj, &uv_pci_location_attribute.attr);
+ if (ret)
+ goto err_add_sysfs;
+ }
+ if (top_obj->iio_stack >= 0) {
+ ret = sysfs_create_file(&top_obj->kobj, &uv_pci_iio_stack_attribute.attr);
+ if (ret)
+ goto err_add_sysfs;
+ }
+ if (top_obj->ppb_addr) {
+ ret = sysfs_create_file(&top_obj->kobj, &uv_pci_ppb_addr_attribute.attr);
+ if (ret)
+ goto err_add_sysfs;
+ }
+ if (top_obj->slot >= 0) {
+ ret = sysfs_create_file(&top_obj->kobj, &uv_pci_slot_attribute.attr);
+ if (ret)
+ goto err_add_sysfs;
+ }
+
+ kobject_uevent(&top_obj->kobj, KOBJ_ADD);
+ return 0;
+
+err_add_sysfs:
+ kobject_put(&top_obj->kobj);
+ return ret;
+}
+
+static int pci_topology_init(void)
+{
+ char *pci_top_str, *start, *found, *count;
+ size_t sz;
+ s64 biosr;
+ int l = 0, k = 0;
+ int len, ret;
+
+ uv_pcibus_kset = kset_create_and_add("pcibuses", NULL, sgi_uv_kobj);
+ if (!uv_pcibus_kset)
+ return -ENOMEM;
+
+ for (sz = PAGE_SIZE; sz < 16 * PAGE_SIZE; sz += PAGE_SIZE) {
+ pci_top_str = kmalloc(sz, GFP_KERNEL);
+ if (!pci_top_str) {
+ ret = -ENOMEM;
+ goto err_pci_top_str;
+ }
+ biosr = uv_bios_get_pci_topology((u64)sz, (u64 *)pci_top_str);
+ if (biosr == BIOS_STATUS_SUCCESS) {
+ len = strnlen(pci_top_str, sz);
+ for (count = pci_top_str; count < pci_top_str + len; count++) {
+ if (*count == '\n')
+ l++;
+ }
+ num_pci_lines = l;
+
+ uv_pci_objs = kcalloc(num_pci_lines,
+ sizeof(*uv_pci_objs), GFP_KERNEL);
+ if (!uv_pci_objs) {
+ kfree(pci_top_str);
+ ret = -ENOMEM;
+ goto err_pci_top_str;
+ }
+ start = pci_top_str;
+ while ((found = strsep(&start, "\n")) != NULL) {
+ uv_pci_objs[k] = kzalloc(sizeof(*uv_pci_objs[k]), GFP_KERNEL);
+ if (!uv_pci_objs[k]) {
+ ret = -ENOMEM;
+ goto err_pci_obj;
+ }
+ ret = init_pci_top_obj(uv_pci_objs[k], found);
+ if (ret)
+ goto err_pci_obj;
+ k++;
+ if (k == num_pci_lines)
+ break;
+ }
+ }
+ kfree(pci_top_str);
+ if (biosr == BIOS_STATUS_SUCCESS || biosr == BIOS_STATUS_UNIMPLEMENTED)
+ break;
+ }
+
+ return 0;
+err_pci_obj:
+ k--;
+ for (; k >= 0; k--)
+ kobject_put(&uv_pci_objs[k]->kobj);
+ kfree(uv_pci_objs);
+ kfree(pci_top_str);
+err_pci_top_str:
+ kset_unregister(uv_pcibus_kset);
+ return ret;
+}
+
+static void pci_topology_exit(void)
+{
+ int k;
+
+ for (k = 0; k < num_pci_lines; k++)
+ kobject_put(&uv_pci_objs[k]->kobj);
+ kset_unregister(uv_pcibus_kset);
+ kfree(uv_pci_objs);
+}
+
+static ssize_t partition_id_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ return sprintf(buf, "%ld\n", sn_partition_id);
+}
+
+static ssize_t uv_type_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ return scnprintf(buf, PAGE_SIZE, "%s\n", uv_type_string());
+}
+
+static struct kobj_attribute partition_id_attr =
+ __ATTR(partition_id, 0444, partition_id_show, NULL);
+static struct kobj_attribute uv_type_attr =
+ __ATTR(uv_type, 0444, uv_type_show, NULL);
+
+static struct attribute *base_attrs[] = {
+ &partition_id_attr.attr,
+ &uv_type_attr.attr,
+ NULL,
+};
+
+static struct attribute_group base_attr_group = {
+ .attrs = base_attrs
+};
+
+static int initial_bios_setup(void)
+{
+ u64 v;
+ s64 biosr;
+
+ biosr = uv_bios_get_master_nasid((u64)sizeof(uv_master_nasid), (u64 *)&uv_master_nasid);
+ if (biosr)
+ return -EINVAL;
+
+ biosr = uv_bios_get_heapsize((u64)uv_master_nasid, (u64)sizeof(u64), &v);
+ if (biosr)
+ return -EINVAL;
+
+ uv_biosheap = vmalloc(v);
+ if (!uv_biosheap)
+ return -ENOMEM;
+
+ biosr = uv_bios_install_heap((u64)uv_master_nasid, v, (u64 *)uv_biosheap);
+ if (biosr) {
+ vfree(uv_biosheap);
+ return -EINVAL;
+ }
+
+ biosr = uv_bios_obj_count((u64)uv_master_nasid, sizeof(u64), &v);
+ if (biosr) {
+ vfree(uv_biosheap);
+ return -EINVAL;
+ }
+ uv_bios_obj_cnt = (int)v;
+
+ return 0;
+}
+
+static int __init uv_sysfs_init(void)
+{
+ int ret = 0;
+
+ if (!is_uv_system())
+ return -ENODEV;
+
+ num_cnodes = uv_num_possible_blades();
+
+ if (!sgi_uv_kobj)
+ sgi_uv_kobj = kobject_create_and_add("sgi_uv", firmware_kobj);
+ if (!sgi_uv_kobj) {
+ pr_warn("kobject_create_and_add sgi_uv failed\n");
+ return -EINVAL;
+ }
+ ret = sysfs_create_group(sgi_uv_kobj, &base_attr_group);
+ if (ret) {
+ pr_warn("sysfs_create_group base_attr_group failed\n");
+ goto err_create_group;
+ }
+
+ ret = initial_bios_setup();
+ if (ret)
+ goto err_bios_setup;
+
+ ret = uv_hubs_init();
+ if (ret)
+ goto err_hubs_init;
+
+ ret = uv_ports_init();
+ if (ret)
+ goto err_ports_init;
+
+ ret = pci_topology_init();
+ if (ret)
+ goto err_pci_init;
+
+ return 0;
+
+err_pci_init:
+ uv_ports_exit();
+err_ports_init:
+ uv_hubs_exit();
+err_hubs_init:
+ vfree(uv_biosheap);
+err_bios_setup:
+ sysfs_remove_group(sgi_uv_kobj, &base_attr_group);
+err_create_group:
+ kobject_put(sgi_uv_kobj);
+ return ret;
+}
+
+static void __exit uv_sysfs_exit(void)
+{
+ if (!is_uv_system())
+ return;
+
+ pci_topology_exit();
+ uv_ports_exit();
+ uv_hubs_exit();
+ vfree(uv_biosheap);
+ sysfs_remove_group(sgi_uv_kobj, &base_attr_group);
+ kobject_put(sgi_uv_kobj);
+}
+
+#ifndef MODULE
+device_initcall(uv_sysfs_init);
+#else
+module_init(uv_sysfs_init);
+#endif
+module_exit(uv_sysfs_exit);
+
+MODULE_AUTHOR("Hewlett Packard Enterprise");
+MODULE_LICENSE("GPL");
--
2.26.2

2020-11-18 11:04:08

by Hans de Goede

[permalink] [raw]
Subject: Re: [PATCH 0/5] Add uv_sysfs platform driver

Hi,

On 11/17/20 9:42 PM, Justin Ernst wrote:
> Introduce a new platform driver to gather topology information from UV systems
> and expose that information via a sysfs interface at /sys/firmware/sgi_uv/.
>
> Justin Ernst (5):
> x86/platform/uv: Remove existing /sys/firmware/sgi_uv/ interface
> x86/platform/uv: Add and export uv_bios_* functions
> x86/platform/uv: Add new uv_sysfs platform driver
> x86/platform/uv: Update ABI documentation of /sys/firmware/sgi_uv/
> x86/platform/uv: Update MAINTAINERS for uv_sysfs driver

So patch 1/1 drops the existing

/sys/firmware/sgi_uv/coherence_id
/sys/firmware/sgi_uv/partition_id

sysfs API, then according to patch 4/5 patch 3/5 reintroduces
the /sys/firmware/sgi_uv/partition_id API, but the /sys/firmware/sgi_uv/coherence_id
file is gone for ever ?

I'm not sure what userspace bits (may) depend on this but without more info
this looks like a clear violation of the do not break userspace APIs rule.

So, based on this, I have to nack this series in its current state.

Now if there is a strong believe there are 0 (not a few, but _zero_) users
out there who rely on the /sys/firmware/sgi_uv/coherence_id file then this
might be ok. But then there needs to be a technical analysis of why this is
ok in the commit message of the patch dropping this sysfs file.

Also the commit message of patch 1/5 should mention that
/sys/firmware/sgi_uv/partition_id will be re-introduced later through
another driver.

Regards,

Hans


>
> .../ABI/testing/sysfs-firmware-sgi_uv | 137 ++-
> MAINTAINERS | 6 +
> arch/x86/include/asm/uv/bios.h | 49 +
> arch/x86/include/asm/uv/uv_geo.h | 103 +++
> arch/x86/platform/uv/Makefile | 2 +-
> arch/x86/platform/uv/bios_uv.c | 54 ++
> arch/x86/platform/uv/uv_sysfs.c | 63 --
> drivers/platform/x86/Kconfig | 11 +
> drivers/platform/x86/Makefile | 3 +
> drivers/platform/x86/uv_sysfs.c | 853 ++++++++++++++++++
> 10 files changed, 1202 insertions(+), 79 deletions(-)
> create mode 100644 arch/x86/include/asm/uv/uv_geo.h
> delete mode 100644 arch/x86/platform/uv/uv_sysfs.c
> create mode 100644 drivers/platform/x86/uv_sysfs.c
>
>
> base-commit: 4ef8451b332662d004df269d4cdeb7d9f31419b5
>

2020-11-18 16:10:53

by Ernst, Justin

[permalink] [raw]
Subject: RE: [PATCH 0/5] Add uv_sysfs platform driver

> Hi,
>
> On 11/17/20 9:42 PM, Justin Ernst wrote:
> > Introduce a new platform driver to gather topology information from UV systems
> > and expose that information via a sysfs interface at /sys/firmware/sgi_uv/.
> >
> > Justin Ernst (5):
> > x86/platform/uv: Remove existing /sys/firmware/sgi_uv/ interface
> > x86/platform/uv: Add and export uv_bios_* functions
> > x86/platform/uv: Add new uv_sysfs platform driver
> > x86/platform/uv: Update ABI documentation of /sys/firmware/sgi_uv/
> > x86/platform/uv: Update MAINTAINERS for uv_sysfs driver
>
> So patch 1/1 drops the existing
>
> /sys/firmware/sgi_uv/coherence_id
> /sys/firmware/sgi_uv/partition_id
>
> sysfs API, then according to patch 4/5 patch 3/5 reintroduces
> the /sys/firmware/sgi_uv/partition_id API, but the /sys/firmware/sgi_uv/coherence_id
> file is gone for ever ?
>
> I'm not sure what userspace bits (may) depend on this but without more info
> this looks like a clear violation of the do not break userspace APIs rule.
>
> So, based on this, I have to nack this series in its current state.
>
> Now if there is a strong believe there are 0 (not a few, but _zero_) users
> out there who rely on the /sys/firmware/sgi_uv/coherence_id file then this
> might be ok. But then there needs to be a technical analysis of why this is
> ok in the commit message of the patch dropping this sysfs file.
>
> Also the commit message of patch 1/5 should mention that
> /sys/firmware/sgi_uv/partition_id will be re-introduced later through
> another driver.

Hello Hans,

I will resubmit these patches without the API breakage, reintroducing the coherence_id file in the new driver.

Thank you for taking the time to look over my patch set.

>
> Regards,
>
> Hans