[PATCH v2 1/2] powerpc/perf: consolidate GPCI hcall structs into asm/hvcall.h

The H_GetPerformanceCounterInfo (GPCI) hypercall input/output structs are
useful to modules outside of perf/, so move them into asm/hvcall.h to live
alongside the other powerpc hypercall structs.

Leave the perf-specific GPCI stuff in perf/hv-gpci.h.

Signed-off-by: Scott Cheloha <redacted>
---
 arch/powerpc/include/asm/hvcall.h | 36 +++++++++++++++++++++++++++++++
 arch/powerpc/perf/hv-gpci.c       |  9 --------
 arch/powerpc/perf/hv-gpci.h       | 27 -----------------------
 3 files changed, 36 insertions(+), 36 deletions(-)

diff --git a/arch/powerpc/include/asm/hvcall.h b/arch/powerpc/include/asm/hvcall.h
index e90c073e437e..c338480b4551 100644
--- a/arch/powerpc/include/asm/hvcall.h
+++ b/arch/powerpc/include/asm/hvcall.h

@@ -527,6 +527,42 @@ struct hv_guest_state {
 /* Latest version of hv_guest_state structure */
 #define HV_GUEST_STATE_VERSION	1
 
+/*
+ * From the document "H_GetPerformanceCounterInfo Interface" v1.07
+ *
+ * H_GET_PERF_COUNTER_INFO argument
+ */
+struct hv_get_perf_counter_info_params {
+	__be32 counter_request; /* I */
+	__be32 starting_index;  /* IO */
+	__be16 secondary_index; /* IO */
+	__be16 returned_values; /* O */
+	__be32 detail_rc; /* O, only needed when called via *_norets() */
+
+	/*
+	 * O, size each of counter_value element in bytes, only set for version
+	 * >= 0x3
+	 */
+	__be16 cv_element_size;
+
+	/* I, 0 (zero) for versions < 0x3 */
+	__u8 counter_info_version_in;
+
+	/* O, 0 (zero) if version < 0x3. Must be set to 0 when making hcall */
+	__u8 counter_info_version_out;
+	__u8 reserved[0xC];
+	__u8 counter_value[];
+} __packed;
+
+#define HGPCI_REQ_BUFFER_SIZE	4096
+#define HGPCI_MAX_DATA_BYTES \
+	(HGPCI_REQ_BUFFER_SIZE - sizeof(struct hv_get_perf_counter_info_params))
+
+struct hv_gpci_request_buffer {
+	struct hv_get_perf_counter_info_params params;
+	uint8_t bytes[HGPCI_MAX_DATA_BYTES];
+} __packed;
+
 #endif /* __ASSEMBLY__ */
 #endif /* __KERNEL__ */
 #endif /* _ASM_POWERPC_HVCALL_H */

diff --git a/arch/powerpc/perf/hv-gpci.c b/arch/powerpc/perf/hv-gpci.c
index 6884d16ec19b..1667315b82e9 100644
--- a/arch/powerpc/perf/hv-gpci.c
+++ b/arch/powerpc/perf/hv-gpci.c

@@ -123,17 +123,8 @@ static const struct attribute_group *attr_groups[] = {
 	NULL,
 };
 
-#define HGPCI_REQ_BUFFER_SIZE	4096
-#define HGPCI_MAX_DATA_BYTES \
-	(HGPCI_REQ_BUFFER_SIZE - sizeof(struct hv_get_perf_counter_info_params))
-
 static DEFINE_PER_CPU(char, hv_gpci_reqb[HGPCI_REQ_BUFFER_SIZE]) __aligned(sizeof(uint64_t));
 
-struct hv_gpci_request_buffer {
-	struct hv_get_perf_counter_info_params params;
-	uint8_t bytes[HGPCI_MAX_DATA_BYTES];
-} __packed;
-
 static unsigned long single_gpci_request(u32 req, u32 starting_index,
 		u16 secondary_index, u8 version_in, u32 offset, u8 length,
 		u64 *value)

diff --git a/arch/powerpc/perf/hv-gpci.h b/arch/powerpc/perf/hv-gpci.h
index a3053eda5dcc..4d108262bed7 100644
--- a/arch/powerpc/perf/hv-gpci.h
+++ b/arch/powerpc/perf/hv-gpci.h

@@ -2,33 +2,6 @@
 #ifndef LINUX_POWERPC_PERF_HV_GPCI_H_
 #define LINUX_POWERPC_PERF_HV_GPCI_H_
 
-#include <linux/types.h>
-
-/* From the document "H_GetPerformanceCounterInfo Interface" v1.07 */
-
-/* H_GET_PERF_COUNTER_INFO argument */
-struct hv_get_perf_counter_info_params {
-	__be32 counter_request; /* I */
-	__be32 starting_index;  /* IO */
-	__be16 secondary_index; /* IO */
-	__be16 returned_values; /* O */
-	__be32 detail_rc; /* O, only needed when called via *_norets() */
-
-	/*
-	 * O, size each of counter_value element in bytes, only set for version
-	 * >= 0x3
-	 */
-	__be16 cv_element_size;
-
-	/* I, 0 (zero) for versions < 0x3 */
-	__u8 counter_info_version_in;
-
-	/* O, 0 (zero) if version < 0x3. Must be set to 0 when making hcall */
-	__u8 counter_info_version_out;
-	__u8 reserved[0xC];
-	__u8 counter_value[];
-} __packed;
-
 /*
  * counter info version => fw version/reference (spec version)
  *

-- 
2.24.1

The H_GetPerformanceCounterInfo (GPCI) PHYP hypercall has a subcall,
Affinity_Domain_Info_By_Partition, which returns, among other things,
a "partition affinity score" for a given LPAR.  This score, a value on
[0-100], represents the processor-memory affinity for the LPAR in
question.  A score of 0 indicates the worst possible affinity while a
score of 100 indicates perfect affinity.  The score can be used to
reason about performance.

This patch adds the score for the local LPAR to the lparcfg procfile
under a new 'partition_affinity_score' key.

Signed-off-by: Scott Cheloha <redacted>
---
 arch/powerpc/platforms/pseries/lparcfg.c | 35 ++++++++++++++++++++++++
 1 file changed, 35 insertions(+)

diff --git a/arch/powerpc/platforms/pseries/lparcfg.c b/arch/powerpc/platforms/pseries/lparcfg.c
index b8d28ab88178..e278390ab28d 100644
--- a/arch/powerpc/platforms/pseries/lparcfg.c
+++ b/arch/powerpc/platforms/pseries/lparcfg.c

@@ -136,6 +136,39 @@ static unsigned int h_get_ppp(struct hvcall_ppp_data *ppp_data)
 	return rc;
 }
 
+static void show_gpci_data(struct seq_file *m)
+{
+	struct hv_gpci_request_buffer *buf;
+	unsigned int affinity_score;
+	long ret;
+
+	buf = kmalloc(sizeof(*buf), GFP_KERNEL);
+	if (buf == NULL)
+		return;
+
+	/*
+	 * Show the local LPAR's affinity score.
+	 *
+	 * 0xB1 selects the Affinity_Domain_Info_By_Partition subcall.
+	 * The score is at byte 0xB in the output buffer.
+	 */
+	memset(&buf->params, 0, sizeof(buf->params));
+	buf->params.counter_request = cpu_to_be32(0xB1);
+	buf->params.starting_index = cpu_to_be32(-1);	/* local LPAR */
+	buf->params.counter_info_version_in = 0x5;	/* v5+ for score */
+	ret = plpar_hcall_norets(H_GET_PERF_COUNTER_INFO, virt_to_phys(buf),
+				 sizeof(*buf));
+	if (ret != H_SUCCESS) {
+		pr_debug("hcall failed: H_GET_PERF_COUNTER_INFO: %ld, %x\n",
+			 ret, be32_to_cpu(buf->params.detail_rc));
+		goto out;
+	}
+	affinity_score = buf->bytes[0xB];
+	seq_printf(m, "partition_affinity_score=%u\n", affinity_score);
+out:
+	kfree(buf);
+}
+
 static unsigned h_pic(unsigned long *pool_idle_time,
 		      unsigned long *num_procs)
 {

@@ -487,6 +520,8 @@ static int pseries_lparcfg_data(struct seq_file *m, void *v)
 			   partition_active_processors * 100);
 	}
 
+	show_gpci_data(m);
+
 	seq_printf(m, "partition_active_processors=%d\n",
 		   partition_active_processors);
 

-- 
2.24.1

On 7/27/20 11:46 AM, Scott Cheloha wrote:

Hmm, this pretty much guts this header which normally I'd be inclined to suggest
moving the whole thing. The remainder of that header for context:

 /*
  * counter info version => fw version/reference (spec version)
  *

 * 8 => power8 (1.07)
 * [7 is skipped by spec 1.07]
 * 6 => TLBIE (1.07)
 * 5 => v7r7m0.phyp (1.05)
 * [4 skipped]
 * 3 => v7r6m0.phyp (?)
 * [1,2 skipped]
 * 0 => v7r{2,3,4}m0.phyp (?)
 */
#define COUNTER_INFO_VERSION_CURRENT 0x8

/* capability mask masks. */
enum {
        HV_GPCI_CM_GA = (1 << 7),
        HV_GPCI_CM_EXPANDED = (1 << 6),
        HV_GPCI_CM_LAB = (1 << 5)
};

#define REQUEST_FILE "../hv-gpci-requests.h"
#define NAME_LOWER hv_gpci
#define NAME_UPPER HV_GPCI
#include "req-gen/perf.h"
#undef REQUEST_FILE
#undef NAME_LOWER
#undef NAME_UPPER

The side effect of moving seems that we would have to drag "hv-gpci-requests.h"
along as well. So, maybe its best just moving the struct as you've done so it
can be used by code outside of perf.

-Tyrel

On 7/27/20 11:46 AM, Scott Cheloha wrote:

The H_GetPerformanceCounterInfo (GPCI) PHYP hypercall has a subcall,
Affinity_Domain_Info_By_Partition, which returns, among other things,
a "partition affinity score" for a given LPAR.  This score, a value on
[0-100], represents the processor-memory affinity for the LPAR in
question.  A score of 0 indicates the worst possible affinity while a
score of 100 indicates perfect affinity.  The score can be used to
reason about performance.

This patch adds the score for the local LPAR to the lparcfg procfile
under a new 'partition_affinity_score' key.

Signed-off-by: Scott Cheloha <redacted>

I was hoping Michael would chime in the first time around on this patch series
about adding another key/value pair to lparcfg. So, barring a NACK from mpe:

Reviewed-by: Tyrel Datwyler <tyreld@linux.ibm.com>

Tyrel Datwyler [off-list ref] writes:

On 7/27/20 11:46 AM, Scott Cheloha wrote:

quoted

The H_GetPerformanceCounterInfo (GPCI) PHYP hypercall has a subcall,
Affinity_Domain_Info_By_Partition, which returns, among other things,
a "partition affinity score" for a given LPAR.  This score, a value on
[0-100], represents the processor-memory affinity for the LPAR in
question.  A score of 0 indicates the worst possible affinity while a
score of 100 indicates perfect affinity.  The score can be used to
reason about performance.

This patch adds the score for the local LPAR to the lparcfg procfile
under a new 'partition_affinity_score' key.

Signed-off-by: Scott Cheloha <redacted>

I was hoping Michael would chime in the first time around on this patch series
about adding another key/value pair to lparcfg.

That guy is so unreliable.

I don't love adding new stuff in lparcfg, but given the file already
exists and there's no prospect of removing it, it's probably not worth
the effort to put the new field anywhere else.

My other query with this was how on earth anyone is meant to interpret
the metric. ie. if my metric is 50, what does that mean? If it's 90
should I worry?

Which makes me realise we have no documentation for lparcfg in the
kernel at all.

So it would be nice to have it mentioned somewhere in Documentation,
even if it just points to the manpage in powerpc-ibm-utils.

cheers

So, barring a NACK from mpe:

Reviewed-by: Tyrel Datwyler <tyreld@linux.ibm.com>

quoted

---
 arch/powerpc/platforms/pseries/lparcfg.c | 35 ++++++++++++++++++++++++
 1 file changed, 35 insertions(+)

diff --git a/arch/powerpc/platforms/pseries/lparcfg.c b/arch/powerpc/platforms/pseries/lparcfg.c
index b8d28ab88178..e278390ab28d 100644
--- a/arch/powerpc/platforms/pseries/lparcfg.c
+++ b/arch/powerpc/platforms/pseries/lparcfg.c

@@ -136,6 +136,39 @@ static unsigned int h_get_ppp(struct hvcall_ppp_data *ppp_data)
 	return rc;
 }
 
+static void show_gpci_data(struct seq_file *m)
+{
+	struct hv_gpci_request_buffer *buf;
+	unsigned int affinity_score;
+	long ret;
+
+	buf = kmalloc(sizeof(*buf), GFP_KERNEL);
+	if (buf == NULL)
+		return;
+
+	/*
+	 * Show the local LPAR's affinity score.
+	 *
+	 * 0xB1 selects the Affinity_Domain_Info_By_Partition subcall.
+	 * The score is at byte 0xB in the output buffer.
+	 */
+	memset(&buf->params, 0, sizeof(buf->params));
+	buf->params.counter_request = cpu_to_be32(0xB1);
+	buf->params.starting_index = cpu_to_be32(-1);	/* local LPAR */
+	buf->params.counter_info_version_in = 0x5;	/* v5+ for score */
+	ret = plpar_hcall_norets(H_GET_PERF_COUNTER_INFO, virt_to_phys(buf),
+				 sizeof(*buf));
+	if (ret != H_SUCCESS) {
+		pr_debug("hcall failed: H_GET_PERF_COUNTER_INFO: %ld, %x\n",
+			 ret, be32_to_cpu(buf->params.detail_rc));
+		goto out;
+	}
+	affinity_score = buf->bytes[0xB];
+	seq_printf(m, "partition_affinity_score=%u\n", affinity_score);
+out:
+	kfree(buf);
+}
+
 static unsigned h_pic(unsigned long *pool_idle_time,
 		      unsigned long *num_procs)
 {

@@ -487,6 +520,8 @@ static int pseries_lparcfg_data(struct seq_file *m, void *v)
 			   partition_active_processors * 100);
 	}
 
+	show_gpci_data(m);
+
 	seq_printf(m, "partition_active_processors=%d\n",
 		   partition_active_processors);
 

Michael Ellerman [off-list ref] writes:

Tyrel Datwyler [off-list ref] writes:

quoted

On 7/27/20 11:46 AM, Scott Cheloha wrote:

quoted

The H_GetPerformanceCounterInfo (GPCI) PHYP hypercall has a subcall,
Affinity_Domain_Info_By_Partition, which returns, among other things,
a "partition affinity score" for a given LPAR.  This score, a value on
[0-100], represents the processor-memory affinity for the LPAR in
question.  A score of 0 indicates the worst possible affinity while a
score of 100 indicates perfect affinity.  The score can be used to
reason about performance.

This patch adds the score for the local LPAR to the lparcfg procfile
under a new 'partition_affinity_score' key.

Signed-off-by: Scott Cheloha <redacted>

I was hoping Michael would chime in the first time around on this patch series
about adding another key/value pair to lparcfg.

That guy is so unreliable.

I don't love adding new stuff in lparcfg, but given the file already
exists and there's no prospect of removing it, it's probably not worth
the effort to put the new field anywhere else.

My other query with this was how on earth anyone is meant to interpret
the metric. ie. if my metric is 50, what does that mean? If it's 90
should I worry?

Here's some more background.

This interface is just passing up what the platform provides, and it's
identical to the partition affinity score described in the documentation
for the management console's lsmemopt command:

https://www.ibm.com/support/knowledgecenter/POWER9/p9edm/lsmemopt.html

The score is 0-100, higher values are better. To illustrate: I believe a
partition's score will be 100 (or very close to it) if all of its CPUs
and memory reside within one node. It will be lower than that when a
partition has some memory without local CPUs, and lower still when there
is no CPU-memory affinity within the partition. Beyond that I don't have
more specific information and the algorithm and scale are set by the
platform.

The intent is for this to be a metric to gather during problem
determination e.g. via sosreport or similar, but as far as Linux is
concerned this should be treated as an opaque value.

Scott Cheloha [off-list ref] writes:

Acked-by: Nathan Lynch <redacted>

Scott Cheloha [off-list ref] writes:

The H_GetPerformanceCounterInfo (GPCI) hypercall input/output structs are
useful to modules outside of perf/, so move them into asm/hvcall.h to live
alongside the other powerpc hypercall structs.

Leave the perf-specific GPCI stuff in perf/hv-gpci.h.

Signed-off-by: Scott Cheloha <redacted>

Acked-by: Nathan Lynch <redacted>

On Mon, 27 Jul 2020 13:46:04 -0500, Scott Cheloha wrote:

The H_GetPerformanceCounterInfo (GPCI) hypercall input/output structs are
useful to modules outside of perf/, so move them into asm/hvcall.h to live
alongside the other powerpc hypercall structs.

Leave the perf-specific GPCI stuff in perf/hv-gpci.h.

Applied to powerpc/next.

[1/2] powerpc/perf: consolidate GPCI hcall structs into asm/hvcall.h
      https://git.kernel.org/powerpc/c/59562b5c33d6ff3685509ed58b2ed3c5b5712704
[2/2] powerpc/pseries: new lparcfg key/value pair: partition_affinity_score
      https://git.kernel.org/powerpc/c/5d1bc776428f34941a6237afb9454061b5b5e1e1

cheers