The hcall H_GET_PERF_COUNTER_INFO can be used to get data related to
chips, dimms and system topology, by passing different counter request
values.
Patchset adds sysfs files to "/sys/devices/hv_gpci/interface/"
of hv_gpci pmu driver, which will expose system topology information
using H_GET_PERF_COUNTER_INFO hcall. The added sysfs files are
available for power10 and above platforms and needs root access
to read the data.
Patches 1,3,5,7,9 adds sysfs interface files to the hv_gpci
pmu driver, to get system topology information.
List of added sysfs files:
-> processor_bus_topology (Counter request value : 0xD0)
-> processor_config (Counter request value : 0x90)
-> affinity_domain_via_virtual_processor (Counter request value : 0xA0)
-> affinity_domain_via_domain (Counter request value : 0xB0)
-> affinity_domain_via_partition (Counter request value : 0xB1)
Patches 2,4,6,8,10 adds details of the newly added hv_gpci
interface files listed above in the ABI documentation.
Patches 2,4,6,8,10 adds details of the newly added hv_gpci
interface files listed above in the ABI documentation.
Changelog:
v3 -> v4
-> Add Reviewed-by tag from Athira Rajeev.
-> Correct typo by changing processor_bug_topology to processor_bus_topology
v2 -> v3
-> Make nit changes in documentation patches as suggested by Randy Dunlap.
v1 -> v2
-> Incase the HCALL fails with errors that can be resolve during runtime,
then only add sysinfo interface attributes to the interface_attrs
attribute array. Even if one of the counter request value HCALL fails,
don't add any sysinfo attribute to the interface_attrs attribute array.
Add the code changes to make sure sysinfo interface added only when all
the requirements met as suggested by Michael Ellerman.
-> Make changes in documentation, adds detail of errors type
which can be resolved at runtime as suggested by Michael Ellerman.
-> Add new enum and sysinfo_counter_request array to get required
counter request value in hv-gpci.c file.
-> Move the macros for interface attribute array index to hv-gpci.c, as
these macros currently only used in hv-gpci.c file.
Kajol Jain (10):
powerpc/hv_gpci: Add sysfs file inside hv_gpci device to show
processor bus topology information
docs: ABI: sysfs-bus-event_source-devices-hv_gpci: Document
processor_bus_topology sysfs interface file
powerpc/hv_gpci: Add sysfs file inside hv_gpci device to show
processor config information
docs: ABI: sysfs-bus-event_source-devices-hv_gpci: Document
processor_config sysfs interface file
powerpc/hv_gpci: Add sysfs file inside hv_gpci device to show affinity
domain via virtual processor information
docs: ABI: sysfs-bus-event_source-devices-hv_gpci: Document
affinity_domain_via_virtual_processor sysfs interface file
powerpc/hv_gpci: Add sysfs file inside hv_gpci device to show affinity
domain via domain information
docs: ABI: sysfs-bus-event_source-devices-hv_gpci: Document
affinity_domain_via_domain sysfs interface file
powerpc/hv_gpci: Add sysfs file inside hv_gpci device to show affinity
domain via partition information
docs: ABI: sysfs-bus-event_source-devices-hv_gpci: Document
affinity_domain_via_partition sysfs interface file
.../sysfs-bus-event_source-devices-hv_gpci | 160 +++++
arch/powerpc/perf/hv-gpci.c | 640 +++++++++++++++++-
2 files changed, 798 insertions(+), 2 deletions(-)
--
2.39.3
The hcall H_GET_PERF_COUNTER_INFO with counter request value as
PROCESSOR_CONFIG(0X90), can be used to get the system
processor configuration information. To expose the system
processor config information, patch adds sysfs file called
"processor_config" to the "/sys/devices/hv_gpci/interface/"
of hv_gpci pmu driver.
Add enum and sysinfo_counter_request array to get required
counter request value in hv-gpci.c file.
Also add a new function called "sysinfo_device_attr_create",
which will create and return required device attribute to the
add_sysinfo_interface_files function.
The processor_config sysfs file is only available for power10
and above platforms. Add a new macro called
INTERFACE_PROCESSOR_CONFIG_ATTR, which points to the index of
NULL placefolder, for processor_config attribute in the interface_attrs
array. Also add macro INTERFACE_NULL_ATTR which points to index of NULL
attribute in interface_attrs array.
Reviewed-by: Athira Rajeev <[email protected]>
Signed-off-by: Kajol Jain <[email protected]>
---
arch/powerpc/perf/hv-gpci.c | 168 ++++++++++++++++++++++++++++++++----
1 file changed, 153 insertions(+), 15 deletions(-)
diff --git a/arch/powerpc/perf/hv-gpci.c b/arch/powerpc/perf/hv-gpci.c
index 225f148f75fd..c74076d3c7a7 100644
--- a/arch/powerpc/perf/hv-gpci.c
+++ b/arch/powerpc/perf/hv-gpci.c
@@ -102,11 +102,21 @@ static ssize_t cpumask_show(struct device *dev,
return cpumap_print_to_pagebuf(true, buf, &hv_gpci_cpumask);
}
-/* Counter request value to retrieve system information */
-#define PROCESSOR_BUS_TOPOLOGY 0XD0
-
/* Interface attribute array index to store system information */
#define INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR 6
+#define INTERFACE_PROCESSOR_CONFIG_ATTR 7
+#define INTERFACE_NULL_ATTR 8
+
+/* Counter request value to retrieve system information */
+enum {
+ PROCESSOR_BUS_TOPOLOGY,
+ PROCESSOR_CONFIG
+};
+
+static int sysinfo_counter_request[] = {
+ [PROCESSOR_BUS_TOPOLOGY] = 0xD0,
+ [PROCESSOR_CONFIG] = 0x90,
+};
static DEFINE_PER_CPU(char, hv_gpci_reqb[HGPCI_REQ_BUFFER_SIZE]) __aligned(sizeof(uint64_t));
@@ -187,7 +197,8 @@ static ssize_t processor_bus_topology_show(struct device *dev, struct device_att
* starting_index value implies the starting hardware
* chip id.
*/
- ret = systeminfo_gpci_request(PROCESSOR_BUS_TOPOLOGY, 0, 0, buf, &n, arg);
+ ret = systeminfo_gpci_request(sysinfo_counter_request[PROCESSOR_BUS_TOPOLOGY],
+ 0, 0, buf, &n, arg);
if (!ret)
return n;
@@ -220,8 +231,76 @@ static ssize_t processor_bus_topology_show(struct device *dev, struct device_att
memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
- ret = systeminfo_gpci_request(PROCESSOR_BUS_TOPOLOGY, starting_index,
- 0, buf, &n, arg);
+ ret = systeminfo_gpci_request(sysinfo_counter_request[PROCESSOR_BUS_TOPOLOGY],
+ starting_index, 0, buf, &n, arg);
+
+ if (!ret)
+ return n;
+
+ if (ret != H_PARAMETER)
+ goto out;
+ }
+
+ return n;
+
+out:
+ put_cpu_var(hv_gpci_reqb);
+ return ret;
+}
+
+static ssize_t processor_config_show(struct device *dev, struct device_attribute *attr,
+ char *buf)
+{
+ struct hv_gpci_request_buffer *arg;
+ unsigned long ret;
+ size_t n = 0;
+
+ arg = (void *)get_cpu_var(hv_gpci_reqb);
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+
+ /*
+ * Pass the counter request value 0x90 corresponds to request
+ * type 'Processor_config', to retrieve
+ * the system processor information.
+ * starting_index value implies the starting hardware
+ * processor index.
+ */
+ ret = systeminfo_gpci_request(sysinfo_counter_request[PROCESSOR_CONFIG],
+ 0, 0, buf, &n, arg);
+
+ if (!ret)
+ return n;
+
+ if (ret != H_PARAMETER)
+ goto out;
+
+ /*
+ * ret value as 'H_PARAMETER' corresponds to 'GEN_BUF_TOO_SMALL', which
+ * implies that buffer can't accommodate all information, and a partial buffer
+ * returned. To handle that, we need to take subsequent requests
+ * with next starting index to retrieve additional (missing) data.
+ * Below loop do subsequent hcalls with next starting index and add it
+ * to buffer util we get all the information.
+ */
+ while (ret == H_PARAMETER) {
+ int returned_values = be16_to_cpu(arg->params.returned_values);
+ int elementsize = be16_to_cpu(arg->params.cv_element_size);
+ int last_element = (returned_values - 1) * elementsize;
+
+ /*
+ * Since the starting index is part of counter_value
+ * buffer elements, use the starting index value in the last
+ * element and add 1 to subsequent hcalls.
+ */
+ u32 starting_index = arg->bytes[last_element + 3] +
+ (arg->bytes[last_element + 2] << 8) +
+ (arg->bytes[last_element + 1] << 16) +
+ (arg->bytes[last_element] << 24) + 1;
+
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+
+ ret = systeminfo_gpci_request(sysinfo_counter_request[PROCESSOR_CONFIG],
+ starting_index, 0, buf, &n, arg);
if (!ret)
return n;
@@ -258,6 +337,11 @@ static struct attribute *interface_attrs[] = {
* attribute, set in init function if applicable.
*/
NULL,
+ /*
+ * This NULL is a placeholder for the processor_config
+ * attribute, set in init function if applicable.
+ */
+ NULL,
NULL,
};
@@ -463,17 +547,24 @@ static int hv_gpci_cpu_hotplug_init(void)
ppc_hv_gpci_cpu_offline);
}
-static void add_sysinfo_interface_files(void)
+static struct device_attribute *sysinfo_device_attr_create(int
+ sysinfo_interface_group_index, u32 req)
{
struct device_attribute *attr = NULL;
unsigned long ret;
struct hv_gpci_request_buffer *arg;
- /* Check for counter request type PROCESSOR_BUS_TOPOLOGY support */
+ if (sysinfo_interface_group_index < INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR ||
+ sysinfo_interface_group_index >= INTERFACE_NULL_ATTR) {
+ pr_info("Wrong interface group index for system information\n");
+ return NULL;
+ }
+
+ /* Check for given counter request value support */
arg = (void *)get_cpu_var(hv_gpci_reqb);
memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
- arg->params.counter_request = cpu_to_be32(PROCESSOR_BUS_TOPOLOGY);
+ arg->params.counter_request = cpu_to_be32(req);
ret = plpar_hcall_norets(H_GET_PERF_COUNTER_INFO,
virt_to_phys(arg), HGPCI_REQ_BUFFER_SIZE);
@@ -481,24 +572,71 @@ static void add_sysinfo_interface_files(void)
put_cpu_var(hv_gpci_reqb);
/*
- * Add processor_bus_topology attribute in the interface_attrs
+ * Add given counter request value attribute in the interface_attrs
* attribute array, only for valid return types.
*/
if (!ret || ret == H_AUTHORITY || ret == H_PARAMETER) {
attr = kzalloc(sizeof(*attr), GFP_KERNEL);
if (!attr) {
- pr_info("Memory allocation failed for sysinfo interface files\n");
- return;
+ pr_info("Mem allocation failed, for sysinfo interface file");
+ return NULL;
}
sysfs_attr_init(&attr->attr);
- attr->attr.name = "processor_bus_topology";
attr->attr.mode = 0444;
- attr->show = processor_bus_topology_show;
- interface_attrs[INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR] = &attr->attr;
+
+ switch (sysinfo_interface_group_index) {
+ case INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR:
+ attr->attr.name = "processor_bus_topology";
+ attr->show = processor_bus_topology_show;
+ break;
+ case INTERFACE_PROCESSOR_CONFIG_ATTR:
+ attr->attr.name = "processor_config";
+ attr->show = processor_config_show;
+ break;
+ }
} else
pr_devel("hcall failed, with error: 0x%lx\n", ret);
+
+ return attr;
+}
+
+static void add_sysinfo_interface_files(void)
+{
+ int sysfs_count;
+ struct device_attribute *attr[INTERFACE_NULL_ATTR - INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR];
+ int i;
+
+ sysfs_count = INTERFACE_NULL_ATTR - INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR;
+
+ /* Get device attribute for a given counter request value */
+ for (i = 0; i < sysfs_count; i++) {
+ attr[i] = sysinfo_device_attr_create(i + INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR,
+ sysinfo_counter_request[i]);
+
+ if (!attr[i])
+ goto out;
+ }
+
+ /* Add sysinfo interface attributes in the interface_attrs attribute array */
+ for (i = 0; i < sysfs_count; i++)
+ interface_attrs[i + INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR] = &attr[i]->attr;
+
+ return;
+
+out:
+ /*
+ * The sysinfo interface attributes will be added, only if hcall passed for
+ * all the counter request values. Free the device attribute array incase
+ * of any hcall failure.
+ */
+ if (i > 0) {
+ while (i >= 0) {
+ kfree(attr[i]);
+ i--;
+ }
+ }
}
static int hv_gpci_init(void)
--
2.39.3
Add details of the new hv-gpci interface file called
"processor_config" in the ABI documentation.
Reviewed-by: Athira Rajeev <[email protected]>
Signed-off-by: Kajol Jain <[email protected]>
---
.../sysfs-bus-event_source-devices-hv_gpci | 32 +++++++++++++++++++
1 file changed, 32 insertions(+)
diff --git a/Documentation/ABI/testing/sysfs-bus-event_source-devices-hv_gpci b/Documentation/ABI/testing/sysfs-bus-event_source-devices-hv_gpci
index ba3f9aa3d68e..9e81de18142f 100644
--- a/Documentation/ABI/testing/sysfs-bus-event_source-devices-hv_gpci
+++ b/Documentation/ABI/testing/sysfs-bus-event_source-devices-hv_gpci
@@ -112,3 +112,35 @@ Description: admin read only
more information.
* "-EFBIG" : System information exceeds PAGE_SIZE.
+
+What: /sys/devices/hv_gpci/interface/processor_config
+Date: July 2023
+Contact: Linux on PowerPC Developer List <[email protected]>
+Description: admin read only
+ This sysfs file exposes the system topology information by making HCALL
+ H_GET_PERF_COUNTER_INFO. The HCALL is made with counter request value
+ PROCESSOR_CONFIG(0x90).
+
+ * This sysfs file will be created only for power10 and above platforms.
+
+ * User needs root privileges to read data from this sysfs file.
+
+ * This sysfs file will be created, only when the HCALL returns "H_SUCCESS",
+ "H_AUTHORITY" or "H_PARAMETER" as the return type.
+
+ HCALL with return error type "H_AUTHORITY" can be resolved during
+ runtime by setting "Enable Performance Information Collection" option.
+
+ * The end user reading this sysfs file must decode the content as per
+ underlying platform/firmware.
+
+ Possible error codes while reading this sysfs file:
+
+ * "-EPERM" : Partition is not permitted to retrieve performance information,
+ required to set "Enable Performance Information Collection" option.
+
+ * "-EIO" : Can't retrieve system information because of invalid buffer length/invalid address
+ or because of some hardware error. Refer to getPerfCountInfo documentation for
+ more information.
+
+ * "-EFBIG" : System information exceeds PAGE_SIZE.
--
2.39.3
The hcall H_GET_PERF_COUNTER_INFO with counter request value as
PROCESSOR_BUS_TOPOLOGY(0XD0), can be used to get the system
topology information. To expose the system topology information,
patch adds sysfs file called "processor_bus_topology" to the
"/sys/devices/hv_gpci/interface/" of hv_gpci pmu driver.
Add macro for PROCESSOR_BUS_TOPOLOGY counter request value
in hv-gpci.c file. Also add a new function called
"systeminfo_gpci_request", to make the H_GET_PERF_COUNTER_INFO hcall
with added macro and populates the output buffer.
The processor_bus_topology sysfs file is only available for power10
and above platforms. Add a new function called
"add_sysinfo_interface_files", which will add processor_bus_topology
attribute in the interface_attrs array, only for power10 and
above platforms.
Also add macro INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR in hv-gpci.c
file, which points to the index of NULL placefolder, for
processor_bus_topology attribute.
Reviewed-by: Athira Rajeev <[email protected]>
Signed-off-by: Kajol Jain <[email protected]>
---
arch/powerpc/perf/hv-gpci.c | 184 +++++++++++++++++++++++++++++++++++-
1 file changed, 182 insertions(+), 2 deletions(-)
diff --git a/arch/powerpc/perf/hv-gpci.c b/arch/powerpc/perf/hv-gpci.c
index 7ff8ff3509f5..225f148f75fd 100644
--- a/arch/powerpc/perf/hv-gpci.c
+++ b/arch/powerpc/perf/hv-gpci.c
@@ -102,6 +102,141 @@ static ssize_t cpumask_show(struct device *dev,
return cpumap_print_to_pagebuf(true, buf, &hv_gpci_cpumask);
}
+/* Counter request value to retrieve system information */
+#define PROCESSOR_BUS_TOPOLOGY 0XD0
+
+/* Interface attribute array index to store system information */
+#define INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR 6
+
+static DEFINE_PER_CPU(char, hv_gpci_reqb[HGPCI_REQ_BUFFER_SIZE]) __aligned(sizeof(uint64_t));
+
+static unsigned long systeminfo_gpci_request(u32 req, u32 starting_index,
+ u16 secondary_index, char *buf,
+ size_t *n, struct hv_gpci_request_buffer *arg)
+{
+ unsigned long ret;
+ size_t i, j;
+
+ arg->params.counter_request = cpu_to_be32(req);
+ arg->params.starting_index = cpu_to_be32(starting_index);
+ arg->params.secondary_index = cpu_to_be16(secondary_index);
+
+ ret = plpar_hcall_norets(H_GET_PERF_COUNTER_INFO,
+ virt_to_phys(arg), HGPCI_REQ_BUFFER_SIZE);
+
+ /*
+ * ret value as 'H_PARAMETER' corresponds to 'GEN_BUF_TOO_SMALL',
+ * which means that the current buffer size cannot accommodate
+ * all the information and a partial buffer returned.
+ * hcall fails incase of ret value other than H_SUCCESS or H_PARAMETER.
+ *
+ * ret value as H_AUTHORITY implies that partition is not permitted to retrieve
+ * performance information, and required to set
+ * "Enable Performance Information Collection" option.
+ */
+ if (ret == H_AUTHORITY)
+ return -EPERM;
+
+ /*
+ * hcall can fail with other possible ret value like H_PRIVILEGE/H_HARDWARE
+ * because of invalid buffer-length/address or due to some hardware
+ * error.
+ */
+ if (ret && (ret != H_PARAMETER))
+ return -EIO;
+
+ /*
+ * hcall H_GET_PERF_COUNTER_INFO populates the 'returned_values'
+ * to show the total number of counter_value array elements
+ * returned via hcall.
+ * hcall also populates 'cv_element_size' corresponds to individual
+ * counter_value array element size. Below loop go through all
+ * counter_value array elements as per their size and add it to
+ * the output buffer.
+ */
+ for (i = 0; i < be16_to_cpu(arg->params.returned_values); i++) {
+ j = i * be16_to_cpu(arg->params.cv_element_size);
+
+ for (; j < (i + 1) * be16_to_cpu(arg->params.cv_element_size); j++)
+ *n += sprintf(buf + *n, "%02x", (u8)arg->bytes[j]);
+ *n += sprintf(buf + *n, "\n");
+ }
+
+ if (*n >= PAGE_SIZE) {
+ pr_info("System information exceeds PAGE_SIZE\n");
+ return -EFBIG;
+ }
+
+ return ret;
+}
+
+static ssize_t processor_bus_topology_show(struct device *dev, struct device_attribute *attr,
+ char *buf)
+{
+ struct hv_gpci_request_buffer *arg;
+ unsigned long ret;
+ size_t n = 0;
+
+ arg = (void *)get_cpu_var(hv_gpci_reqb);
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+
+ /*
+ * Pass the counter request value 0xD0 corresponds to request
+ * type 'Processor_bus_topology', to retrieve
+ * the system topology information.
+ * starting_index value implies the starting hardware
+ * chip id.
+ */
+ ret = systeminfo_gpci_request(PROCESSOR_BUS_TOPOLOGY, 0, 0, buf, &n, arg);
+
+ if (!ret)
+ return n;
+
+ if (ret != H_PARAMETER)
+ goto out;
+
+ /*
+ * ret value as 'H_PARAMETER' corresponds to 'GEN_BUF_TOO_SMALL', which
+ * implies that buffer can't accommodate all information, and a partial buffer
+ * returned. To handle that, we need to make subsequent requests
+ * with next starting index to retrieve additional (missing) data.
+ * Below loop do subsequent hcalls with next starting index and add it
+ * to buffer util we get all the information.
+ */
+ while (ret == H_PARAMETER) {
+ int returned_values = be16_to_cpu(arg->params.returned_values);
+ int elementsize = be16_to_cpu(arg->params.cv_element_size);
+ int last_element = (returned_values - 1) * elementsize;
+
+ /*
+ * Since the starting index value is part of counter_value
+ * buffer elements, use the starting index value in the last
+ * element and add 1 to make subsequent hcalls.
+ */
+ u32 starting_index = arg->bytes[last_element + 3] +
+ (arg->bytes[last_element + 2] << 8) +
+ (arg->bytes[last_element + 1] << 16) +
+ (arg->bytes[last_element] << 24) + 1;
+
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+
+ ret = systeminfo_gpci_request(PROCESSOR_BUS_TOPOLOGY, starting_index,
+ 0, buf, &n, arg);
+
+ if (!ret)
+ return n;
+
+ if (ret != H_PARAMETER)
+ goto out;
+ }
+
+ return n;
+
+out:
+ put_cpu_var(hv_gpci_reqb);
+ return ret;
+}
+
static DEVICE_ATTR_RO(kernel_version);
static DEVICE_ATTR_RO(cpumask);
@@ -118,6 +253,11 @@ static struct attribute *interface_attrs[] = {
&hv_caps_attr_expanded.attr,
&hv_caps_attr_lab.attr,
&hv_caps_attr_collect_privileged.attr,
+ /*
+ * This NULL is a placeholder for the processor_bus_topology
+ * attribute, set in init function if applicable.
+ */
+ NULL,
NULL,
};
@@ -143,8 +283,6 @@ static const struct attribute_group *attr_groups[] = {
NULL,
};
-static DEFINE_PER_CPU(char, hv_gpci_reqb[HGPCI_REQ_BUFFER_SIZE]) __aligned(sizeof(uint64_t));
-
static unsigned long single_gpci_request(u32 req, u32 starting_index,
u16 secondary_index, u8 version_in, u32 offset, u8 length,
u64 *value)
@@ -325,6 +463,44 @@ static int hv_gpci_cpu_hotplug_init(void)
ppc_hv_gpci_cpu_offline);
}
+static void add_sysinfo_interface_files(void)
+{
+ struct device_attribute *attr = NULL;
+ unsigned long ret;
+ struct hv_gpci_request_buffer *arg;
+
+ /* Check for counter request type PROCESSOR_BUS_TOPOLOGY support */
+ arg = (void *)get_cpu_var(hv_gpci_reqb);
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+
+ arg->params.counter_request = cpu_to_be32(PROCESSOR_BUS_TOPOLOGY);
+
+ ret = plpar_hcall_norets(H_GET_PERF_COUNTER_INFO,
+ virt_to_phys(arg), HGPCI_REQ_BUFFER_SIZE);
+
+ put_cpu_var(hv_gpci_reqb);
+
+ /*
+ * Add processor_bus_topology attribute in the interface_attrs
+ * attribute array, only for valid return types.
+ */
+ if (!ret || ret == H_AUTHORITY || ret == H_PARAMETER) {
+ attr = kzalloc(sizeof(*attr), GFP_KERNEL);
+
+ if (!attr) {
+ pr_info("Memory allocation failed for sysinfo interface files\n");
+ return;
+ }
+
+ sysfs_attr_init(&attr->attr);
+ attr->attr.name = "processor_bus_topology";
+ attr->attr.mode = 0444;
+ attr->show = processor_bus_topology_show;
+ interface_attrs[INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR] = &attr->attr;
+ } else
+ pr_devel("hcall failed, with error: 0x%lx\n", ret);
+}
+
static int hv_gpci_init(void)
{
int r;
@@ -388,6 +564,10 @@ static int hv_gpci_init(void)
if (r)
return r;
+ /* sysinfo interface files are only available for power10 and above platforms */
+ if (PVR_VER(mfspr(SPRN_PVR)) >= PVR_POWER10)
+ add_sysinfo_interface_files();
+
return 0;
}
--
2.39.3
The hcall H_GET_PERF_COUNTER_INFO with counter request value as
AFFINITY_DOMAIN_INFORMATION_BY_VIRTUAL_PROCESSOR(0XA0), can be used to get
the system affinity domain via virtual processor information. To expose
the system affinity domain via virtual processor information, patch adds
sysfs file called "affinity_domain_via_virtual_processor" to the
"/sys/devices/hv_gpci/interface/" of hv_gpci pmu driver.
The affinity_domain_via_virtual_processor sysfs file is only available for
power10 and above platforms. Add a macro called
INTERFACE_AFFINITY_DOMAIN_VIA_VP_ATTR, which points to the index of NULL
placeholder, for affinity_domain_via_virtual_processor attribute in
interface_attrs array. Also updated the value of INTERFACE_NULL_ATTR macro
in hv-gpci.c file.
Reviewed-by: Athira Rajeev <[email protected]>
Signed-off-by: Kajol Jain <[email protected]>
---
arch/powerpc/perf/hv-gpci.c | 86 ++++++++++++++++++++++++++++++++++++-
1 file changed, 84 insertions(+), 2 deletions(-)
diff --git a/arch/powerpc/perf/hv-gpci.c b/arch/powerpc/perf/hv-gpci.c
index c74076d3c7a7..68502cb18262 100644
--- a/arch/powerpc/perf/hv-gpci.c
+++ b/arch/powerpc/perf/hv-gpci.c
@@ -105,17 +105,20 @@ static ssize_t cpumask_show(struct device *dev,
/* Interface attribute array index to store system information */
#define INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR 6
#define INTERFACE_PROCESSOR_CONFIG_ATTR 7
-#define INTERFACE_NULL_ATTR 8
+#define INTERFACE_AFFINITY_DOMAIN_VIA_VP_ATTR 8
+#define INTERFACE_NULL_ATTR 9
/* Counter request value to retrieve system information */
enum {
PROCESSOR_BUS_TOPOLOGY,
- PROCESSOR_CONFIG
+ PROCESSOR_CONFIG,
+ AFFINITY_DOMAIN_VIA_VP, /* affinity domain via virtual processor */
};
static int sysinfo_counter_request[] = {
[PROCESSOR_BUS_TOPOLOGY] = 0xD0,
[PROCESSOR_CONFIG] = 0x90,
+ [AFFINITY_DOMAIN_VIA_VP] = 0xA0,
};
static DEFINE_PER_CPU(char, hv_gpci_reqb[HGPCI_REQ_BUFFER_SIZE]) __aligned(sizeof(uint64_t));
@@ -316,6 +319,76 @@ static ssize_t processor_config_show(struct device *dev, struct device_attribute
return ret;
}
+static ssize_t affinity_domain_via_virtual_processor_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ struct hv_gpci_request_buffer *arg;
+ unsigned long ret;
+ size_t n = 0;
+
+ arg = (void *)get_cpu_var(hv_gpci_reqb);
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+
+ /*
+ * Pass the counter request 0xA0 corresponds to request
+ * type 'Affinity_domain_information_by_virutal_processor',
+ * to retrieve the system affinity domain information.
+ * starting_index value refers to the starting hardware
+ * processor index.
+ */
+ ret = systeminfo_gpci_request(sysinfo_counter_request[AFFINITY_DOMAIN_VIA_VP],
+ 0, 0, buf, &n, arg);
+
+ if (!ret)
+ return n;
+
+ if (ret != H_PARAMETER)
+ goto out;
+
+ /*
+ * ret value as 'H_PARAMETER' corresponds to 'GEN_BUF_TOO_SMALL', which
+ * implies that buffer can't accommodate all information, and a partial buffer
+ * returned. To handle that, we need to take subsequent requests
+ * with next secondary index to retrieve additional (missing) data.
+ * Below loop do subsequent hcalls with next secondary index and add it
+ * to buffer util we get all the information.
+ */
+ while (ret == H_PARAMETER) {
+ int returned_values = be16_to_cpu(arg->params.returned_values);
+ int elementsize = be16_to_cpu(arg->params.cv_element_size);
+ int last_element = (returned_values - 1) * elementsize;
+
+ /*
+ * Since the starting index and secondary index type is part of the
+ * counter_value buffer elements, use the starting index value in the
+ * last array element as subsequent starting index, and use secondary index
+ * value in the last array element plus 1 as subsequent secondary index.
+ * For counter request '0xA0', starting index points to partition id
+ * and secondary index points to corresponding virtual processor index.
+ */
+ u32 starting_index = arg->bytes[last_element + 1] + (arg->bytes[last_element] << 8);
+ u16 secondary_index = arg->bytes[last_element + 3] +
+ (arg->bytes[last_element + 2] << 8) + 1;
+
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+
+ ret = systeminfo_gpci_request(sysinfo_counter_request[AFFINITY_DOMAIN_VIA_VP],
+ starting_index, secondary_index, buf, &n, arg);
+
+ if (!ret)
+ return n;
+
+ if (ret != H_PARAMETER)
+ goto out;
+ }
+
+ return n;
+
+out:
+ put_cpu_var(hv_gpci_reqb);
+ return ret;
+}
+
static DEVICE_ATTR_RO(kernel_version);
static DEVICE_ATTR_RO(cpumask);
@@ -342,6 +415,11 @@ static struct attribute *interface_attrs[] = {
* attribute, set in init function if applicable.
*/
NULL,
+ /*
+ * This NULL is a placeholder for the affinity_domain_via_virtual_processor
+ * attribute, set in init function if applicable.
+ */
+ NULL,
NULL,
};
@@ -595,6 +673,10 @@ static struct device_attribute *sysinfo_device_attr_create(int
attr->attr.name = "processor_config";
attr->show = processor_config_show;
break;
+ case INTERFACE_AFFINITY_DOMAIN_VIA_VP_ATTR:
+ attr->attr.name = "affinity_domain_via_virtual_processor";
+ attr->show = affinity_domain_via_virtual_processor_show;
+ break;
}
} else
pr_devel("hcall failed, with error: 0x%lx\n", ret);
--
2.39.3
The hcall H_GET_PERF_COUNTER_INFO with counter request value as
AFFINITY_DOMAIN_INFORMATION_BY_DOMAIN(0XB0), can be used to get
the system affinity domain via domain information. To expose the system
affinity domain via domain information, patch adds sysfs file called
"affinity_domain_via_domain" to the "/sys/devices/hv_gpci/interface/"
of hv_gpci pmu driver.
Add new entry for AFFINITY_DOMAIN_VIA_DOM in sysinfo_counter_request
array, which points to the counter request value
"affinity_domain_via_domain" in hv-gpci.c file.
The affinity_domain_via_domain sysfs file is only available for power10
and above platforms. Add a macro called
INTERFACE_AFFINITY_DOMAIN_VIA_DOM_ATTR, which points to the index of NULL
placeholder, for affinity_domain_via_domain attribute in interface_attrs
array. Also updated the value of INTERFACE_NULL_ATTR macro in hv-gpci.c
file.
Reviewed-by: Athira Rajeev <[email protected]>
Signed-off-by: Kajol Jain <[email protected]>
---
arch/powerpc/perf/hv-gpci.c | 80 ++++++++++++++++++++++++++++++++++++-
1 file changed, 79 insertions(+), 1 deletion(-)
diff --git a/arch/powerpc/perf/hv-gpci.c b/arch/powerpc/perf/hv-gpci.c
index 68502cb18262..326b758df7c8 100644
--- a/arch/powerpc/perf/hv-gpci.c
+++ b/arch/powerpc/perf/hv-gpci.c
@@ -106,19 +106,22 @@ static ssize_t cpumask_show(struct device *dev,
#define INTERFACE_PROCESSOR_BUS_TOPOLOGY_ATTR 6
#define INTERFACE_PROCESSOR_CONFIG_ATTR 7
#define INTERFACE_AFFINITY_DOMAIN_VIA_VP_ATTR 8
-#define INTERFACE_NULL_ATTR 9
+#define INTERFACE_AFFINITY_DOMAIN_VIA_DOM_ATTR 9
+#define INTERFACE_NULL_ATTR 10
/* Counter request value to retrieve system information */
enum {
PROCESSOR_BUS_TOPOLOGY,
PROCESSOR_CONFIG,
AFFINITY_DOMAIN_VIA_VP, /* affinity domain via virtual processor */
+ AFFINITY_DOMAIN_VIA_DOM, /* affinity domain via domain */
};
static int sysinfo_counter_request[] = {
[PROCESSOR_BUS_TOPOLOGY] = 0xD0,
[PROCESSOR_CONFIG] = 0x90,
[AFFINITY_DOMAIN_VIA_VP] = 0xA0,
+ [AFFINITY_DOMAIN_VIA_DOM] = 0xB0,
};
static DEFINE_PER_CPU(char, hv_gpci_reqb[HGPCI_REQ_BUFFER_SIZE]) __aligned(sizeof(uint64_t));
@@ -389,6 +392,72 @@ static ssize_t affinity_domain_via_virtual_processor_show(struct device *dev,
return ret;
}
+static ssize_t affinity_domain_via_domain_show(struct device *dev, struct device_attribute *attr,
+ char *buf)
+{
+ struct hv_gpci_request_buffer *arg;
+ unsigned long ret;
+ size_t n = 0;
+
+ arg = (void *)get_cpu_var(hv_gpci_reqb);
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+
+ /*
+ * Pass the counter request 0xB0 corresponds to request
+ * type 'Affinity_domain_information_by_domain',
+ * to retrieve the system affinity domain information.
+ * starting_index value refers to the starting hardware
+ * processor index.
+ */
+ ret = systeminfo_gpci_request(sysinfo_counter_request[AFFINITY_DOMAIN_VIA_DOM],
+ 0, 0, buf, &n, arg);
+
+ if (!ret)
+ return n;
+
+ if (ret != H_PARAMETER)
+ goto out;
+
+ /*
+ * ret value as 'H_PARAMETER' corresponds to 'GEN_BUF_TOO_SMALL', which
+ * implies that buffer can't accommodate all information, and a partial buffer
+ * returned. To handle that, we need to take subsequent requests
+ * with next starting index to retrieve additional (missing) data.
+ * Below loop do subsequent hcalls with next starting index and add it
+ * to buffer util we get all the information.
+ */
+ while (ret == H_PARAMETER) {
+ int returned_values = be16_to_cpu(arg->params.returned_values);
+ int elementsize = be16_to_cpu(arg->params.cv_element_size);
+ int last_element = (returned_values - 1) * elementsize;
+
+ /*
+ * Since the starting index value is part of counter_value
+ * buffer elements, use the starting index value in the last
+ * element and add 1 to make subsequent hcalls.
+ */
+ u32 starting_index = arg->bytes[last_element + 1] +
+ (arg->bytes[last_element] << 8) + 1;
+
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+
+ ret = systeminfo_gpci_request(sysinfo_counter_request[AFFINITY_DOMAIN_VIA_DOM],
+ starting_index, 0, buf, &n, arg);
+
+ if (!ret)
+ return n;
+
+ if (ret != H_PARAMETER)
+ goto out;
+ }
+
+ return n;
+
+out:
+ put_cpu_var(hv_gpci_reqb);
+ return ret;
+}
+
static DEVICE_ATTR_RO(kernel_version);
static DEVICE_ATTR_RO(cpumask);
@@ -420,6 +489,11 @@ static struct attribute *interface_attrs[] = {
* attribute, set in init function if applicable.
*/
NULL,
+ /*
+ * This NULL is a placeholder for the affinity_domain_via_domain
+ * attribute, set in init function if applicable.
+ */
+ NULL,
NULL,
};
@@ -677,6 +751,10 @@ static struct device_attribute *sysinfo_device_attr_create(int
attr->attr.name = "affinity_domain_via_virtual_processor";
attr->show = affinity_domain_via_virtual_processor_show;
break;
+ case INTERFACE_AFFINITY_DOMAIN_VIA_DOM_ATTR:
+ attr->attr.name = "affinity_domain_via_domain";
+ attr->show = affinity_domain_via_domain_show;
+ break;
}
} else
pr_devel("hcall failed, with error: 0x%lx\n", ret);
--
2.39.3
The hcall H_GET_PERF_COUNTER_INFO with counter request value as
AFFINITY_DOMAIN_INFORMATION_BY_PARTITION(0XB1), can be used to get
the system affinity domain via partition information. To expose the system
affinity domain via partition information, patch adds sysfs file called
"affinity_domain_via_partition" to the "/sys/devices/hv_gpci/interface/"
of hv_gpci pmu driver.
Add new entry for AFFINITY_DOMAIN_VIA_PAR in sysinfo_counter_request
array, which points to the counter request value
"affinity_domain_via_partition" in hv-gpci.c file. Also add a
new function called "affinity_domain_via_partition_result_parse" to parse
the hcall result and store it in output buffer.
The affinity_domain_via_partition sysfs file is only available for power10
and above platforms. Add a macro called
INTERFACE_AFFINITY_DOMAIN_VIA_PAR_ATTR, which points to the index of NULL
placeholder, for affinity_domain_via_partition attribute in
interface_attrs array. Also updated the value of INTERFACE_NULL_ATTR
macro in hv-gpci.c file.
Reviewed-by: Athira Rajeev <[email protected]>
Signed-off-by: Kajol Jain <[email protected]>
---
arch/powerpc/perf/hv-gpci.c | 160 +++++++++++++++++++++++++++++++++++-
1 file changed, 159 insertions(+), 1 deletion(-)
diff --git a/arch/powerpc/perf/hv-gpci.c b/arch/powerpc/perf/hv-gpci.c
index 326b758df7c8..f2fff166290b 100644
--- a/arch/powerpc/perf/hv-gpci.c
+++ b/arch/powerpc/perf/hv-gpci.c
@@ -107,7 +107,8 @@ static ssize_t cpumask_show(struct device *dev,
#define INTERFACE_PROCESSOR_CONFIG_ATTR 7
#define INTERFACE_AFFINITY_DOMAIN_VIA_VP_ATTR 8
#define INTERFACE_AFFINITY_DOMAIN_VIA_DOM_ATTR 9
-#define INTERFACE_NULL_ATTR 10
+#define INTERFACE_AFFINITY_DOMAIN_VIA_PAR_ATTR 10
+#define INTERFACE_NULL_ATTR 11
/* Counter request value to retrieve system information */
enum {
@@ -115,6 +116,7 @@ enum {
PROCESSOR_CONFIG,
AFFINITY_DOMAIN_VIA_VP, /* affinity domain via virtual processor */
AFFINITY_DOMAIN_VIA_DOM, /* affinity domain via domain */
+ AFFINITY_DOMAIN_VIA_PAR, /* affinity domain via partition */
};
static int sysinfo_counter_request[] = {
@@ -122,6 +124,7 @@ static int sysinfo_counter_request[] = {
[PROCESSOR_CONFIG] = 0x90,
[AFFINITY_DOMAIN_VIA_VP] = 0xA0,
[AFFINITY_DOMAIN_VIA_DOM] = 0xB0,
+ [AFFINITY_DOMAIN_VIA_PAR] = 0xB1,
};
static DEFINE_PER_CPU(char, hv_gpci_reqb[HGPCI_REQ_BUFFER_SIZE]) __aligned(sizeof(uint64_t));
@@ -458,6 +461,152 @@ static ssize_t affinity_domain_via_domain_show(struct device *dev, struct device
return ret;
}
+static void affinity_domain_via_partition_result_parse(int returned_values,
+ int element_size, char *buf, size_t *last_element,
+ size_t *n, struct hv_gpci_request_buffer *arg)
+{
+ size_t i = 0, j = 0;
+ size_t k, l, m;
+ uint16_t total_affinity_domain_ele, size_of_each_affinity_domain_ele;
+
+ /*
+ * hcall H_GET_PERF_COUNTER_INFO populates the 'returned_values'
+ * to show the total number of counter_value array elements
+ * returned via hcall.
+ * Unlike other request types, the data structure returned by this
+ * request is variable-size. For this counter request type,
+ * hcall populates 'cv_element_size' corresponds to minimum size of
+ * the structure returned i.e; the size of the structure with no domain
+ * information. Below loop go through all counter_value array
+ * to determine the number and size of each domain array element and
+ * add it to the output buffer.
+ */
+ while (i < returned_values) {
+ k = j;
+ for (; k < j + element_size; k++)
+ *n += sprintf(buf + *n, "%02x", (u8)arg->bytes[k]);
+ *n += sprintf(buf + *n, "\n");
+
+ total_affinity_domain_ele = (u8)arg->bytes[k - 2] << 8 | (u8)arg->bytes[k - 3];
+ size_of_each_affinity_domain_ele = (u8)arg->bytes[k] << 8 | (u8)arg->bytes[k - 1];
+
+ for (l = 0; l < total_affinity_domain_ele; l++) {
+ for (m = 0; m < size_of_each_affinity_domain_ele; m++) {
+ *n += sprintf(buf + *n, "%02x", (u8)arg->bytes[k]);
+ k++;
+ }
+ *n += sprintf(buf + *n, "\n");
+ }
+
+ *n += sprintf(buf + *n, "\n");
+ i++;
+ j = k;
+ }
+
+ *last_element = k;
+}
+
+static ssize_t affinity_domain_via_partition_show(struct device *dev, struct device_attribute *attr,
+ char *buf)
+{
+ struct hv_gpci_request_buffer *arg;
+ unsigned long ret;
+ size_t n = 0;
+ size_t last_element = 0;
+ u32 starting_index;
+
+ arg = (void *)get_cpu_var(hv_gpci_reqb);
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+
+ /*
+ * Pass the counter request value 0xB1 corresponds to counter request
+ * type 'Affinity_domain_information_by_partition',
+ * to retrieve the system affinity domain by partition information.
+ * starting_index value refers to the starting hardware
+ * processor index.
+ */
+ arg->params.counter_request = cpu_to_be32(sysinfo_counter_request[AFFINITY_DOMAIN_VIA_PAR]);
+ arg->params.starting_index = cpu_to_be32(0);
+
+ ret = plpar_hcall_norets(H_GET_PERF_COUNTER_INFO,
+ virt_to_phys(arg), HGPCI_REQ_BUFFER_SIZE);
+
+ if (!ret)
+ goto parse_result;
+
+ /*
+ * ret value as 'H_PARAMETER' implies that the current buffer size
+ * can't accommodate all the information, and a partial buffer
+ * returned. To handle that, we need to make subsequent requests
+ * with next starting index to retrieve additional (missing) data.
+ * Below loop do subsequent hcalls with next starting index and add it
+ * to buffer util we get all the information.
+ */
+ while (ret == H_PARAMETER) {
+ affinity_domain_via_partition_result_parse(
+ be16_to_cpu(arg->params.returned_values) - 1,
+ be16_to_cpu(arg->params.cv_element_size), buf,
+ &last_element, &n, arg);
+
+ if (n >= PAGE_SIZE) {
+ put_cpu_var(hv_gpci_reqb);
+ pr_debug("System information exceeds PAGE_SIZE\n");
+ return -EFBIG;
+ }
+
+ /*
+ * Since the starting index value is part of counter_value
+ * buffer elements, use the starting_index value in the last
+ * element and add 1 to make subsequent hcalls.
+ */
+ starting_index = (u8)arg->bytes[last_element] << 8 |
+ (u8)arg->bytes[last_element + 1];
+
+ memset(arg, 0, HGPCI_REQ_BUFFER_SIZE);
+ arg->params.counter_request = cpu_to_be32(
+ sysinfo_counter_request[AFFINITY_DOMAIN_VIA_PAR]);
+ arg->params.starting_index = cpu_to_be32(starting_index);
+
+ ret = plpar_hcall_norets(H_GET_PERF_COUNTER_INFO,
+ virt_to_phys(arg), HGPCI_REQ_BUFFER_SIZE);
+
+ if (ret && (ret != H_PARAMETER))
+ goto out;
+ }
+
+parse_result:
+ affinity_domain_via_partition_result_parse(
+ be16_to_cpu(arg->params.returned_values),
+ be16_to_cpu(arg->params.cv_element_size),
+ buf, &last_element, &n, arg);
+
+ put_cpu_var(hv_gpci_reqb);
+ return n;
+
+out:
+ put_cpu_var(hv_gpci_reqb);
+
+ /*
+ * ret value as 'H_PARAMETER' corresponds to 'GEN_BUF_TOO_SMALL',
+ * which means that the current buffer size cannot accommodate
+ * all the information and a partial buffer returned.
+ * hcall fails incase of ret value other than H_SUCCESS or H_PARAMETER.
+ *
+ * ret value as H_AUTHORITY implies that partition is not permitted to retrieve
+ * performance information, and required to set
+ * "Enable Performance Information Collection" option.
+ */
+ if (ret == H_AUTHORITY)
+ return -EPERM;
+
+ /*
+ * hcall can fail with other possible ret value like H_PRIVILEGE/H_HARDWARE
+ * because of invalid buffer-length/address or due to some hardware
+ * error.
+ */
+ return -EIO;
+}
+
static DEVICE_ATTR_RO(kernel_version);
static DEVICE_ATTR_RO(cpumask);
@@ -494,6 +643,11 @@ static struct attribute *interface_attrs[] = {
* attribute, set in init function if applicable.
*/
NULL,
+ /*
+ * This NULL is a placeholder for the affinity_domain_via_partition
+ * attribute, set in init function if applicable.
+ */
+ NULL,
NULL,
};
@@ -755,6 +909,10 @@ static struct device_attribute *sysinfo_device_attr_create(int
attr->attr.name = "affinity_domain_via_domain";
attr->show = affinity_domain_via_domain_show;
break;
+ case INTERFACE_AFFINITY_DOMAIN_VIA_PAR_ATTR:
+ attr->attr.name = "affinity_domain_via_partition";
+ attr->show = affinity_domain_via_partition_show;
+ break;
}
} else
pr_devel("hcall failed, with error: 0x%lx\n", ret);
--
2.39.3