Skip to contentSkip to conversation

perf/x86/uncore: Correct the number of CHAs on EMR

PATCH Superseded [email protected] 2 participants 5 messages Sep 1, 2023 – Sep 5, 2023

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf/x86/uncore: Correct the number of CHAs on EMR

From: Kan Liang <[email protected]>

The MSR UNC_CBO_CONFIG, which was used to detect the number of CHAs on SPR, is broken on EMR XCC. It always returns 0.

Roll back to the discovery method, which can give the correct number for this case.

Fixes 38776cc45eb7 ("perf/x86/uncore: Correct the number of CHAs on SPR")
---
 arch/x86/events/intel/uncore_snbep.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/events/intel/uncore_snbep.c +3 -1
index d49e90dc04a4..c41d7d46481c 100644
@@ -6475,7 +6475,9 @@ void spr_uncore_cpu_init(void)
 	type = uncore_find_type_by_id(uncore_msr_uncores, UNCORE_SPR_CHA);
 	if (type) {
 		rdmsrl(SPR_MSR_UNC_CBO_CONFIG, num_cbo);
-		type->num_boxes = num_cbo;
+		/* The MSR doesn't work on the EMR XCC. Roll back to the discovery method. */
+		if (num_cbo)
+			type->num_boxes = num_cbo;
 	}
 	spr_uncore_iio_free_running.num_boxes = uncore_type_max_boxes(uncore_msr_uncores, UNCORE_SPR_IIO);
 }
Signature · 1 line
2.35.1
#1 lore raw
Quoted 22 lines · * [email protected] <[email protected]> wrote:

From: Kan Liang <[email protected]>

The MSR UNC_CBO_CONFIG, which was used to detect the number of CHAs on SPR, is broken on EMR XCC. It always returns 0.

Roll back to the discovery method, which can give the correct number for this case.

Fixes 38776cc45eb7 ("perf/x86/uncore: Correct the number of CHAs on SPR")
---
 arch/x86/events/intel/uncore_snbep.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/events/intel/uncore_snbep.c +0 -0
index d49e90dc04a4..c41d7d46481c 100644

arch/x86/events/intel/uncore_snbep.c void spr_uncore_cpu_init(void) · line 6480

Quoted 7 lines
arch/x86/events/intel/uncore_snbep.c +3 -1
@@ -6475,7 +6475,9 @@ void spr_uncore_cpu_init(void)
 	type = uncore_find_type_by_id(uncore_msr_uncores, UNCORE_SPR_CHA);
 	if (type) {
 		rdmsrl(SPR_MSR_UNC_CBO_CONFIG, num_cbo);
-		type->num_boxes = num_cbo;
+		/* The MSR doesn't work on the EMR XCC. Roll back to the discovery method. */
+		if (num_cbo)
+			type->num_boxes = num_cbo;

So in the zero case we don't write type->num_boxes and leave it as-is.

How does this fall back to the discovery method, is the existing (default?) value of type->num_boxes some special value?

Thanks,

Ingo

author
#2 lore raw
Quoted 35 lines · Ingo Molnar
Quoted 22 lines · * [email protected] <[email protected]> wrote:

From: Kan Liang <[email protected]>

The MSR UNC_CBO_CONFIG, which was used to detect the number of CHAs on SPR, is broken on EMR XCC. It always returns 0.

Roll back to the discovery method, which can give the correct number for this case.

Fixes 38776cc45eb7 ("perf/x86/uncore: Correct the number of CHAs on SPR")
---
 arch/x86/events/intel/uncore_snbep.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/events/intel/uncore_snbep.c +0 -0
index d49e90dc04a4..c41d7d46481c 100644

arch/x86/events/intel/uncore_snbep.c void spr_uncore_cpu_init(void) · line 6480

Quoted 7 lines
arch/x86/events/intel/uncore_snbep.c +3 -1
@@ -6475,7 +6475,9 @@ void spr_uncore_cpu_init(void)
 	type = uncore_find_type_by_id(uncore_msr_uncores, UNCORE_SPR_CHA);
 	if (type) {
 		rdmsrl(SPR_MSR_UNC_CBO_CONFIG, num_cbo);
-		type->num_boxes = num_cbo;
+		/* The MSR doesn't work on the EMR XCC. Roll back to the discovery method. */
+		if (num_cbo)
+			type->num_boxes = num_cbo;

So in the zero case we don't write type->num_boxes and leave it as-is.

Quoted 3 lines

How does this fall back to the discovery method, is the existing (default?) value of type->num_boxes some special value?

Starts from SPR, the basic uncore PMON information are retrieved from the discovery table (resides in an MMIO space populated by BIOS.). It is called the discovery method. The existing value of the type->num_boxes is from the discovery table.

On some SPR variants, there is a firmware bug. So the value from the discovery table is incorrect. We use the value from SPR_MSR_UNC_CBO_CONFIG to replace the one from the discovery table. 38776cc45eb7 ("perf/x86/uncore: Correct the number of CHAs on SPR")

Unfortunately, the SPR_MSR_UNC_CBO_CONFIG isn't available for the EMR XCC (It works well for other EMR variants). But the above firmware bug doesn't impact the EMR XCC. So this patch NOT lets the value from the SPR_MSR_UNC_CBO_CONFIG replace the existing value from the discovery table.

Thanks,
Kan

#3 lore raw
Quoted 52 lines · Liang, Kan
Quoted 35 lines · Ingo Molnar
Quoted 22 lines · * [email protected] <[email protected]> wrote:

From: Kan Liang <[email protected]>

The MSR UNC_CBO_CONFIG, which was used to detect the number of CHAs on SPR, is broken on EMR XCC. It always returns 0.

Roll back to the discovery method, which can give the correct number for this case.

Fixes 38776cc45eb7 ("perf/x86/uncore: Correct the number of CHAs on SPR")
---
 arch/x86/events/intel/uncore_snbep.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/events/intel/uncore_snbep.c +0 -0
index d49e90dc04a4..c41d7d46481c 100644

arch/x86/events/intel/uncore_snbep.c void spr_uncore_cpu_init(void) · line 6480

Quoted 7 lines
arch/x86/events/intel/uncore_snbep.c +3 -1
@@ -6475,7 +6475,9 @@ void spr_uncore_cpu_init(void)
 	type = uncore_find_type_by_id(uncore_msr_uncores, UNCORE_SPR_CHA);
 	if (type) {
 		rdmsrl(SPR_MSR_UNC_CBO_CONFIG, num_cbo);
-		type->num_boxes = num_cbo;
+		/* The MSR doesn't work on the EMR XCC. Roll back to the discovery method. */
+		if (num_cbo)
+			type->num_boxes = num_cbo;

So in the zero case we don't write type->num_boxes and leave it as-is.

Quoted 3 lines

How does this fall back to the discovery method, is the existing (default?) value of type->num_boxes some special value?

Starts from SPR, the basic uncore PMON information are retrieved from the discovery table (resides in an MMIO space populated by BIOS.). It is called the discovery method. The existing value of the type->num_boxes is from the discovery table.

On some SPR variants, there is a firmware bug. So the value from the discovery table is incorrect. We use the value from SPR_MSR_UNC_CBO_CONFIG to replace the one from the discovery table. 38776cc45eb7 ("perf/x86/uncore: Correct the number of CHAs on SPR")

Quoted 4 lines

Unfortunately, the SPR_MSR_UNC_CBO_CONFIG isn't available for the EMR XCC (It works well for other EMR variants). But the above firmware bug doesn't impact the EMR XCC. So this patch NOT lets the value from the SPR_MSR_UNC_CBO_CONFIG replace the existing value from the discovery table.

Thanks - the comment & changelog should probably reflect this background.

Ingo

author
#4 lore raw
Quoted 55 lines · Ingo Molnar
Quoted 52 lines · Liang, Kan
Quoted 35 lines · Ingo Molnar
Quoted 22 lines · * [email protected] <[email protected]> wrote:

From: Kan Liang <[email protected]>

The MSR UNC_CBO_CONFIG, which was used to detect the number of CHAs on SPR, is broken on EMR XCC. It always returns 0.

Roll back to the discovery method, which can give the correct number for this case.

Fixes 38776cc45eb7 ("perf/x86/uncore: Correct the number of CHAs on SPR")
---
 arch/x86/events/intel/uncore_snbep.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/events/intel/uncore_snbep.c +0 -0
index d49e90dc04a4..c41d7d46481c 100644

arch/x86/events/intel/uncore_snbep.c void spr_uncore_cpu_init(void) · line 6480

Quoted 7 lines
arch/x86/events/intel/uncore_snbep.c +3 -1
@@ -6475,7 +6475,9 @@ void spr_uncore_cpu_init(void)
 	type = uncore_find_type_by_id(uncore_msr_uncores, UNCORE_SPR_CHA);
 	if (type) {
 		rdmsrl(SPR_MSR_UNC_CBO_CONFIG, num_cbo);
-		type->num_boxes = num_cbo;
+		/* The MSR doesn't work on the EMR XCC. Roll back to the discovery method. */
+		if (num_cbo)
+			type->num_boxes = num_cbo;

So in the zero case we don't write type->num_boxes and leave it as-is.

Quoted 3 lines

How does this fall back to the discovery method, is the existing (default?) value of type->num_boxes some special value?

Starts from SPR, the basic uncore PMON information are retrieved from the discovery table (resides in an MMIO space populated by BIOS.). It is called the discovery method. The existing value of the type->num_boxes is from the discovery table.

On some SPR variants, there is a firmware bug. So the value from the discovery table is incorrect. We use the value from SPR_MSR_UNC_CBO_CONFIG to replace the one from the discovery table. 38776cc45eb7 ("perf/x86/uncore: Correct the number of CHAs on SPR")

Quoted 7 lines
Quoted 4 lines

Unfortunately, the SPR_MSR_UNC_CBO_CONFIG isn't available for the EMR XCC (It works well for other EMR variants). But the above firmware bug doesn't impact the EMR XCC. So this patch NOT lets the value from the SPR_MSR_UNC_CBO_CONFIG replace the existing value from the discovery table.

Thanks - the comment & changelog should probably reflect this background.

I will update the comment & changelog and send a V2.

Thanks,
Kan