Skip to contentSkip to conversation

perf/amd/uncore: Replace manual sampling check with CAP_NO_INTERRUPT flag

PATCH RESEND Applied Kim Phillips 4 participants 7 messages Mar 11, 2020 – Mar 17, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH RESEND 1/3 perf/amd/uncore: Replace manual sampling check with CAP_NO_INTERRUPT flag

This enables the sampling check in kernel/events/core.c's perf_event_open, which returns the more appropriate -EOPNOTSUPP.

BEFORE:

$ sudo perf record -a -e instructions,l3_request_g1.caching_l3_cache_accesses true
Error:
The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (l3_request_g1.caching_l3_cache_accesses).
/bin/dmesg | grep -i perf may provide additional information.

With nothing relevant in dmesg.

AFTER:

$ sudo perf record -a -e instructions,l3_request_g1.caching_l3_cache_accesses true
Error:
l3_request_g1.caching_l3_cache_accesses: PMU Hardware doesn't support sampling/overflow-interrupts. Try 'perf stat'
Cc Alexander Shishkin <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mark Rutland <[email protected]>
Cc Michael Petlan <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Fixes c43ca5091a37 ("perf/x86/amd: Add support for AMD NB and L2I "uncore" counters")

---
RESEND. No changes since original submission 19 Feb 2020:

https://lkml.org/lkml/2020/2/19/1194

 arch/x86/events/amd/uncore.c | 17 +++++++----------
 1 file changed, 7 insertions(+), 10 deletions(-)
arch/x86/events/amd/uncore.c +7 -10
index a6ea07f2aa84..4d867a752f0e 100644
@@ -190,15 +190,12 @@ static int amd_uncore_event_init(struct perf_event *event)
 
 	/*
 	 * NB and Last level cache counters (MSRs) are shared across all cores
-	 * that share the same NB / Last level cache. Interrupts can be directed
-	 * to a single target core, however, event counts generated by processes
-	 * running on other cores cannot be masked out. So we do not support
-	 * sampling and per-thread events.
+	 * that share the same NB / Last level cache.  On family 16h and below,
+	 * Interrupts can be directed to a single target core, however, event
+	 * counts generated by processes running on other cores cannot be masked
+	 * out. So we do not support sampling and per-thread events via
+	 * CAP_NO_INTERRUPT, and we do not enable counter overflow interrupts:
 	 */
-	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
-		return -EINVAL;
-
-	/* and we do not enable counter overflow interrupts */
 	hwc->config = event->attr.config & AMD64_RAW_EVENT_MASK_NB;
 	hwc->idx = -1;
 
@@ -306,7 +303,7 @@ static struct pmu amd_nb_pmu = {
 	.start		= amd_uncore_start,
 	.stop		= amd_uncore_stop,
 	.read		= amd_uncore_read,
-	.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	.capabilities	= PERF_PMU_CAP_NO_EXCLUDE | PERF_PMU_CAP_NO_INTERRUPT,
 };
 
 static struct pmu amd_llc_pmu = {
@@ -317,7 +314,7 @@ static struct pmu amd_llc_pmu = {
 	.start		= amd_uncore_start,
 	.stop		= amd_uncore_stop,
 	.read		= amd_uncore_read,
-	.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	.capabilities	= PERF_PMU_CAP_NO_EXCLUDE | PERF_PMU_CAP_NO_INTERRUPT,
 };
 
 static struct amd_uncore *amd_uncore_alloc(unsigned int cpu)
Signature · 1 line
2.25.1
#1 lore raw

PATCH RESEND 2/3 perf/amd/uncore: Prepare L3 thread mask code for Family 19h support

In order to better accommodate the upcoming Family 19h support, given the 80-char line limit, we move the existing code into a new l3_thread_slice_mask function, and convert it to use the more readable topology_* helper functions.

Cc Alexander Shishkin <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mark Rutland <[email protected]>
Cc Michael Petlan <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>

---
RESEND. No changes since original submission 19 Feb 2020:

https://lkml.org/lkml/2020/2/19/1192

 arch/x86/events/amd/uncore.c | 28 +++++++++++++++++++---------
 1 file changed, 19 insertions(+), 9 deletions(-)
arch/x86/events/amd/uncore.c +19 -9
index 4d867a752f0e..e635c40ca9c4 100644
@@ -180,6 +180,23 @@ static void amd_uncore_del(struct perf_event *event, int flags)
 	hwc->idx = -1;
 }
 
+/*
+ * Convert logical cpu number to L3 PMC Config ThreadMask format
+ */
+static u64 l3_thread_slice_mask(int cpu)
+{
+	unsigned int shift, thread = 0;
+	u64 thread_mask, core = topology_core_id(cpu);
+
+	if (topology_smt_supported() && !topology_is_primary_thread(cpu))
+		thread = 1;
+
+	shift = AMD64_L3_THREAD_SHIFT + 2 * (core % 4) + thread;
+	thread_mask = BIT_ULL(shift);
+
+	return AMD64_L3_SLICE_MASK | thread_mask;
+}
+
 static int amd_uncore_event_init(struct perf_event *event)
 {
 	struct amd_uncore *uncore;
@@ -206,15 +223,8 @@ static int amd_uncore_event_init(struct perf_event *event)
 	 * SliceMask and ThreadMask need to be set for certain L3 events in
 	 * Family 17h. For other events, the two fields do not affect the count.
 	 */
-	if (l3_mask && is_llc_event(event)) {
-		int thread = 2 * (cpu_data(event->cpu).cpu_core_id % 4);
-
-		if (smp_num_siblings > 1)
-			thread += cpu_data(event->cpu).apicid & 1;
-
-		hwc->config |= (1ULL << (AMD64_L3_THREAD_SHIFT + thread) &
-				AMD64_L3_THREAD_MASK) | AMD64_L3_SLICE_MASK;
-	}
+	if (l3_mask && is_llc_event(event))
+		hwc->config |= l3_thread_slice_mask(event->cpu);
 
 	uncore = event_to_amd_uncore(event);
 	if (!uncore)
Signature · 1 line
2.25.1
maintainer
#2 lore raw

Re: [PATCH 2/3 RESEND] perf/amd/uncore: Prepare L3 thread mask code for Family 19h support

Quoted 65 lines · Kim Phillips

In order to better accommodate the upcoming Family 19h support, given the 80-char line limit, we move the existing code into a new l3_thread_slice_mask function, and convert it to use the more readable topology_* helper functions.

Cc Alexander Shishkin <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mark Rutland <[email protected]>
Cc Michael Petlan <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>

---
RESEND. No changes since original submission 19 Feb 2020:

https://lkml.org/lkml/2020/2/19/1192

 arch/x86/events/amd/uncore.c | 28 +++++++++++++++++++---------
 1 file changed, 19 insertions(+), 9 deletions(-)
arch/x86/events/amd/uncore.c +17 -4
index 4d867a752f0e..e635c40ca9c4 100644
@@ -180,6 +180,23 @@ static void amd_uncore_del(struct perf_event *event, int flags)
 	hwc->idx = -1;
 }
 
+/*
+ * Convert logical cpu number to L3 PMC Config ThreadMask format
+ */
+static u64 l3_thread_slice_mask(int cpu)
+{
+	unsigned int shift, thread = 0;
+	u64 thread_mask, core = topology_core_id(cpu);
+
+	if (topology_smt_supported() && !topology_is_primary_thread(cpu))
+		thread = 1;
+
+	shift = AMD64_L3_THREAD_SHIFT + 2 * (core % 4) + thread;
+	thread_mask = BIT_ULL(shift);
+
+	return AMD64_L3_SLICE_MASK | thread_mask;
+}
+
 static int amd_uncore_event_init(struct perf_event *event)
 {
 	struct amd_uncore *uncore;
@@ -206,15 +223,8 @@ static int amd_uncore_event_init(struct perf_event *event)
 	 * SliceMask and ThreadMask need to be set for certain L3 events in
 	 * Family 17h. For other events, the two fields do not affect the count.
 	 */
-	if (l3_mask && is_llc_event(event)) {
-		int thread = 2 * (cpu_data(event->cpu).cpu_core_id % 4);
-
-		if (smp_num_siblings > 1)

arch/x86/events/amd/uncore.c static int amd_uncore_event_init(struct perf_event *event) · line 230

Quoted 10 lines
arch/x86/events/amd/uncore.c +2 -5
-			thread += cpu_data(event->cpu).apicid & 1;
-
-		hwc->config |= (1ULL << (AMD64_L3_THREAD_SHIFT + thread) &
-				AMD64_L3_THREAD_MASK) | AMD64_L3_SLICE_MASK;
-	}
+	if (l3_mask && is_llc_event(event))
+		hwc->config |= l3_thread_slice_mask(event->cpu);
 
 	uncore = event_to_amd_uncore(event);
 	if (!uncore)

If you carve out functionality into a separate function and then do changes to that functionality, you do two patches: the first one is doing only the mechanical move only and the second one does the changes.

Please do that with that one too.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#3 lore raw

PATCH RESEND 3/3 perf/amd/uncore: Add support for Family 19h L3 PMU

In reply to Kim Phillips

Family 19h introduces a change in slice, core and thread specification in its L3 Performance Event Select (ChL3PmcCfg) register. We implement the new bitmask conversions in a new path in l3_thread_slice_mask() based on a family 19h-and-above check.

We also change the uncore_init() family check to 19h-and-above, so as to not revert to the Family 16h-or-below L2/NB code paths in the driver.

Cc Alexander Shishkin <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mark Rutland <[email protected]>
Cc Michael Petlan <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>

---
RESEND. No changes since original submission:

https://lkml.org/lkml/2020/2/19/1193

 arch/x86/events/amd/uncore.c      | 20 ++++++++++++++------
 arch/x86/include/asm/perf_event.h | 15 +++++++++++++--
 2 files changed, 27 insertions(+), 8 deletions(-)
arch/x86/events/amd/uncore.c +14 -6
index e635c40ca9c4..78b4fb917ad6 100644
@@ -191,10 +191,18 @@ static u64 l3_thread_slice_mask(int cpu)
 	if (topology_smt_supported() && !topology_is_primary_thread(cpu))
 		thread = 1;
 
-	shift = AMD64_L3_THREAD_SHIFT + 2 * (core % 4) + thread;
+	if (boot_cpu_data.x86 <= 0x18) {
+		shift = AMD64_L3_THREAD_SHIFT + 2 * (core % 4) + thread;
+		thread_mask = BIT_ULL(shift);
+
+		return AMD64_L3_SLICE_MASK | thread_mask;
+	}
+
+	core = (core << AMD64_L3_COREID_SHIFT) & AMD64_L3_COREID_MASK;
+	shift = AMD64_L3_THREAD_SHIFT + thread;
 	thread_mask = BIT_ULL(shift);
 
-	return AMD64_L3_SLICE_MASK | thread_mask;
+	return AMD64_L3_EN_ALL_SLICES | core | thread_mask;
 }
 
 static int amd_uncore_event_init(struct perf_event *event)
@@ -220,8 +228,8 @@ static int amd_uncore_event_init(struct perf_event *event)
 		return -EINVAL;
 
 	/*
-	 * SliceMask and ThreadMask need to be set for certain L3 events in
-	 * Family 17h. For other events, the two fields do not affect the count.
+	 * SliceMask and ThreadMask need to be set for certain L3 events.
+	 * For other events, the two fields do not affect the count.
 	 */
 	if (l3_mask && is_llc_event(event))
 		hwc->config |= l3_thread_slice_mask(event->cpu);
@@ -530,9 +538,9 @@ static int __init amd_uncore_init(void)
 	if (!boot_cpu_has(X86_FEATURE_TOPOEXT))
 		return -ENODEV;
 
-	if (boot_cpu_data.x86 == 0x17 || boot_cpu_data.x86 == 0x18) {
+	if (boot_cpu_data.x86 >= 0x17) {
 		/*
-		 * For F17h or F18h, the Northbridge counters are
+		 * For F17h and above, the Northbridge counters are
 		 * repurposed as Data Fabric counters. Also, L3
 		 * counters are supported too. The PMUs are exported
 		 * based on family as either L2 or L3 and NB or DF.
arch/x86/include/asm/perf_event.h +13 -2
index 29964b0e1075..e855e9cf2c37 100644
@@ -50,11 +50,22 @@
 
 #define AMD64_L3_SLICE_SHIFT				48
 #define AMD64_L3_SLICE_MASK				\
-	((0xFULL) << AMD64_L3_SLICE_SHIFT)
+	(0xFULL << AMD64_L3_SLICE_SHIFT)
+#define AMD64_L3_SLICEID_MASK				\
+	(0x7ULL << AMD64_L3_SLICE_SHIFT)
 
 #define AMD64_L3_THREAD_SHIFT				56
 #define AMD64_L3_THREAD_MASK				\
-	((0xFFULL) << AMD64_L3_THREAD_SHIFT)
+	(0xFFULL << AMD64_L3_THREAD_SHIFT)
+#define AMD64_L3_F19H_THREAD_MASK			\
+	(0x3ULL << AMD64_L3_THREAD_SHIFT)
+
+#define AMD64_L3_EN_ALL_CORES				BIT_ULL(47)
+#define AMD64_L3_EN_ALL_SLICES				BIT_ULL(46)
+
+#define AMD64_L3_COREID_SHIFT				42
+#define AMD64_L3_COREID_MASK				\
+	(0x7ULL << AMD64_L3_COREID_SHIFT)
 
 #define X86_RAW_EVENT_MASK		\
 	(ARCH_PERFMON_EVENTSEL_EVENT |	\
Signature · 1 line
2.25.1
maintainer
#4 lore raw

Re: [PATCH 3/3 RESEND] perf/amd/uncore: Add support for Family 19h L3 PMU

Quoted 7 lines · Kim Phillips

Family 19h introduces a change in slice, core and thread specification in its L3 Performance Event Select (ChL3PmcCfg) register. We implement the new bitmask conversions in a new path in l3_thread_slice_mask() based on a family 19h-and-above check.

We also change the uncore_init() family check to 19h-and-above, so as to not revert to the Family 16h-or-below L2/NB code paths in the driver.

Pls use passive voice in your commit message: no "we" or "I", etc, and describe your changes in imperative mood.

Also, pls read section "2) Describe your changes" in Documentation/process/submitting-patches.rst for more details.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
merge bot Applied
#5 lore raw
In reply to Kim Phillips
The following commit has been merged into the perf/urgent branch of tip:

The following commit has been merged into the perf/urgent branch of tip:

Commit-ID:     f967140dfb7442e2db0868b03b961f9c59418a1b
Gitweb:        https://git.kernel.org/tip/f967140dfb7442e2db0868b03b961f9c59418a1b
Author:        Kim Phillips <[email protected]>
AuthorDate:    Wed, 11 Mar 2020 14:13:21 -05:00
Committer:     Borislav Petkov <[email protected]>
CommitterDate: Thu, 12 Mar 2020 14:08:50 +01:00

perf/amd/uncore: Replace manual sampling check with CAP_NO_INTERRUPT flag

Enable the sampling check in kernel/events/core.c::perf_event_open(), which returns the more appropriate -EOPNOTSUPP.

BEFORE:

  $ sudo perf record -a -e instructions,l3_request_g1.caching_l3_cache_accesses true
  Error:
  The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (l3_request_g1.caching_l3_cache_accesses).
  /bin/dmesg | grep -i perf may provide additional information.

With nothing relevant in dmesg.

AFTER:

  $ sudo perf record -a -e instructions,l3_request_g1.caching_l3_cache_accesses true
  Error:
  l3_request_g1.caching_l3_cache_accesses: PMU Hardware doesn't support sampling/overflow-interrupts. Try 'perf stat'
Fixes c43ca5091a37 ("perf/x86/amd: Add support for AMD NB and L2I "uncore" counters")
---
 arch/x86/events/amd/uncore.c | 17 +++++++----------
 1 file changed, 7 insertions(+), 10 deletions(-)
arch/x86/events/amd/uncore.c +7 -10
index a6ea07f..4d867a7 100644
@@ -190,15 +190,12 @@ static int amd_uncore_event_init(struct perf_event *event)
 
 	/*
 	 * NB and Last level cache counters (MSRs) are shared across all cores
-	 * that share the same NB / Last level cache. Interrupts can be directed
-	 * to a single target core, however, event counts generated by processes
-	 * running on other cores cannot be masked out. So we do not support
-	 * sampling and per-thread events.
+	 * that share the same NB / Last level cache.  On family 16h and below,
+	 * Interrupts can be directed to a single target core, however, event
+	 * counts generated by processes running on other cores cannot be masked
+	 * out. So we do not support sampling and per-thread events via
+	 * CAP_NO_INTERRUPT, and we do not enable counter overflow interrupts:
 	 */
-	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
-		return -EINVAL;
-
-	/* and we do not enable counter overflow interrupts */
 	hwc->config = event->attr.config & AMD64_RAW_EVENT_MASK_NB;
 	hwc->idx = -1;
 
@@ -306,7 +303,7 @@ static struct pmu amd_nb_pmu = {
 	.start		= amd_uncore_start,
 	.stop		= amd_uncore_stop,
 	.read		= amd_uncore_read,
-	.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	.capabilities	= PERF_PMU_CAP_NO_EXCLUDE | PERF_PMU_CAP_NO_INTERRUPT,
 };
 
 static struct pmu amd_llc_pmu = {
@@ -317,7 +314,7 @@ static struct pmu amd_llc_pmu = {
 	.start		= amd_uncore_start,
 	.stop		= amd_uncore_stop,
 	.read		= amd_uncore_read,
-	.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	.capabilities	= PERF_PMU_CAP_NO_EXCLUDE | PERF_PMU_CAP_NO_INTERRUPT,
 };
 
 static struct amd_uncore *amd_uncore_alloc(unsigned int cpu)
maintainer
#6 lore raw
In reply to Kim Phillips

Hi

[This is an automated email]

This commit has been processed because it contains a "Fixes:" tag
fixing commit: c43ca5091a37 ("perf/x86/amd: Add support for AMD NB and L2I "uncore" counters").

The bot has tested the following trees: v5.5.9, v5.4.25, v4.19.109, v4.14.173, v4.9.216, v4.4.216.

v5.5.9: Build OK!
v5.4.25: Build OK!
v4.19.109: Failed to apply! Possible dependencies:
88dbe3c94e27 ("perf/core, arch/x86: Strengthen exclusion checks with PERF_PMU_CAP_NO_EXCLUDE")

v4.14.173: Failed to apply! Possible dependencies:
88dbe3c94e27 ("perf/core, arch/x86: Strengthen exclusion checks with PERF_PMU_CAP_NO_EXCLUDE")

v4.9.216: Failed to apply! Possible dependencies:
1650dfd1a9bc ("x86/events, drivers/amd/iommu: Prepare for multiple IOMMUs support") 25df39f2cfd0 ("x86/events/amd/iommu: Enable support for multiple IOMMUs") 51686546304f ("x86/events/amd/iommu: Fix sysfs perf attribute groups") 6aad0c626905 ("x86/events/amd/iommu: Clean up bitwise operations") 88dbe3c94e27 ("perf/core, arch/x86: Strengthen exclusion checks with PERF_PMU_CAP_NO_EXCLUDE") cf25f904ef75 ("x86/events/amd/iommu: Add IOMMU-specific hw_perf_event struct") f5863a00e73c ("x86/events/amd/iommu.c: Modify functions to query max banks and counters") f9573e53f123 ("x86/events/amd/iommu: Declare pr_fmt() format")

v4.4.216: Failed to apply! Possible dependencies:

    1229735b290d ("perf/x86/intel/uncore: Make code more readable")
    1a246b9f58c6 ("perf/x86/intel/uncore: Convert to hotplug state machine")
    31d50c551e30 ("perf/x86/amd/uncore: Do not register a task ctx for uncore PMUs")
    4f089678d071 ("perf/x86/intel/uncore: Fix error handling")
    5485592c1795 ("perf/x86/intel/uncore: Make PCI and MSR uncore independent")
    77af0037de0a ("perf/x86/intel/uncore: Add Knights Landing uncore PMU support")
    7d762e49c211 ("perf/x86/amd/uncore: Prevent use after free")
    83f8ebd2eb45 ("perf/x86/intel/uncore: Add sanity checks for PCI dev package id")
    96b2bd3866a0 ("perf/x86/amd/uncore: Convert to hotplug state machine")
    a46195f1782e ("perf/x86/intel/uncore: Clean up hardware on exit")
    a83f4c00dd6a ("perf/x86/amd/uncore: Rename 'L2' to 'LLC'")
    cf6d445f6897 ("perf/x86/uncore: Track packages, not per CPU data")
    ffeda0038032 ("perf/x86/intel/uncore: Simplify error rollback")

NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

Signature · 2 lines
Thanks
Sasha