[PATCH v2 0/3] perf/x86/amd/uncore: Update AMD uncore driver

[PATCH v2 0/3] perf/x86/amd/uncore: Update AMD uncore driver

[Janakarajan Natarajan]

This patchset updates the AMD uncore driver to support AMD Family17h processors.

First, the current L2 counters are renamed to Last level cache (LLC) counter to 
reflect either L2 or L3 counters depending on the family. Next, the maximum 
number of counters available is increased and the counters per uncore type is 
now based on the family.

Also, in Family17h, the NorthBridge counters are re-purposed as Data Fabric (DF) 
counters. Therefore, this patch exports the pmu name accordingly based on the 
processor family.

v1->v2
* Removed dynamic memory allocation for counters, based on Peter Zijlstra's 
  feedback.

Janakarajan Natarajan (3):
  perf/x86/amd/uncore: Update AMD uncore to rename L2 to LLC
  perf/x86/amd/uncore: Update number of uncore counters
  perf/x86/amd/uncore: Update sysfs attributes for Family17h processors

 arch/x86/events/amd/uncore.c | 204 ++++++++++++++++++++++++++++---------------
 1 file changed, 132 insertions(+), 72 deletions(-)

-- 
2.7.4

[PATCH v2 1/3] perf/x86/amd/uncore: Update AMD uncore to rename L2 to LLC

[Janakarajan Natarajan]

This patch renames L2 counters to LLC counters. In AMD Family17h
processors, L3 cache counter is supported.

Since older families have at most L2 counters, last level cache (LLC)
indicates L2/L3 based on the family.

Signed-off-by: Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
---
 arch/x86/events/amd/uncore.c | 98 ++++++++++++++++++++++----------------------
 1 file changed, 49 insertions(+), 49 deletions(-)

diff --git a/arch/x86/events/amd/uncore.c b/arch/x86/events/amd/uncore.c
index a0b1bdb..24c8537 100644
--- a/arch/x86/events/amd/uncore.c
+++ b/arch/x86/events/amd/uncore.c
@@ -25,7 +25,7 @@
 #define MAX_COUNTERS		NUM_COUNTERS_NB
 
 #define RDPMC_BASE_NB		6
-#define RDPMC_BASE_L2		10
+#define RDPMC_BASE_LLC		10
 
 #define COUNTER_SHIFT		16
 
@@ -45,30 +45,30 @@ struct amd_uncore {
 };
 
 static struct amd_uncore * __percpu *amd_uncore_nb;
-static struct amd_uncore * __percpu *amd_uncore_l2;
+static struct amd_uncore * __percpu *amd_uncore_llc;
 
 static struct pmu amd_nb_pmu;
-static struct pmu amd_l2_pmu;
+static struct pmu amd_llc_pmu;
 
 static cpumask_t amd_nb_active_mask;
-static cpumask_t amd_l2_active_mask;
+static cpumask_t amd_llc_active_mask;
 
 static bool is_nb_event(struct perf_event *event)
 {
 	return event->pmu->type == amd_nb_pmu.type;
 }
 
-static bool is_l2_event(struct perf_event *event)
+static bool is_llc_event(struct perf_event *event)
 {
-	return event->pmu->type == amd_l2_pmu.type;
+	return event->pmu->type == amd_llc_pmu.type;
 }
 
 static struct amd_uncore *event_to_amd_uncore(struct perf_event *event)
 {
 	if (is_nb_event(event) && amd_uncore_nb)
 		return *per_cpu_ptr(amd_uncore_nb, event->cpu);
-	else if (is_l2_event(event) && amd_uncore_l2)
-		return *per_cpu_ptr(amd_uncore_l2, event->cpu);
+	else if (is_llc_event(event) && amd_uncore_llc)
+		return *per_cpu_ptr(amd_uncore_llc, event->cpu);
 
 	return NULL;
 }
@@ -183,16 +183,16 @@ static int amd_uncore_event_init(struct perf_event *event)
 		return -ENOENT;
 
 	/*
-	 * NB and L2 counters (MSRs) are shared across all cores that share the
-	 * same NB / L2 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.
+	 * 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.
 	 */
 	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
 		return -EINVAL;
 
-	/* NB and L2 counters do not have usr/os/guest/host bits */
+	/* NB and Last level cache counters do not have usr/os/guest/host bits */
 	if (event->attr.exclude_user || event->attr.exclude_kernel ||
 	    event->attr.exclude_host || event->attr.exclude_guest)
 		return -EINVAL;
@@ -226,8 +226,8 @@ static ssize_t amd_uncore_attr_show_cpumask(struct device *dev,
 
 	if (pmu->type == amd_nb_pmu.type)
 		active_mask = &amd_nb_active_mask;
-	else if (pmu->type == amd_l2_pmu.type)
-		active_mask = &amd_l2_active_mask;
+	else if (pmu->type == amd_llc_pmu.type)
+		active_mask = &amd_llc_active_mask;
 	else
 		return 0;
 
@@ -276,7 +276,7 @@ static struct pmu amd_nb_pmu = {
 	.read		= amd_uncore_read,
 };
 
-static struct pmu amd_l2_pmu = {
+static struct pmu amd_llc_pmu = {
 	.task_ctx_nr	= perf_invalid_context,
 	.attr_groups	= amd_uncore_attr_groups,
 	.name		= "amd_l2",
@@ -296,7 +296,7 @@ static struct amd_uncore *amd_uncore_alloc(unsigned int cpu)
 
 static int amd_uncore_cpu_up_prepare(unsigned int cpu)
 {
-	struct amd_uncore *uncore_nb = NULL, *uncore_l2;
+	struct amd_uncore *uncore_nb = NULL, *uncore_llc;
 
 	if (amd_uncore_nb) {
 		uncore_nb = amd_uncore_alloc(cpu);
@@ -312,18 +312,18 @@ static int amd_uncore_cpu_up_prepare(unsigned int cpu)
 		*per_cpu_ptr(amd_uncore_nb, cpu) = uncore_nb;
 	}
 
-	if (amd_uncore_l2) {
-		uncore_l2 = amd_uncore_alloc(cpu);
-		if (!uncore_l2)
+	if (amd_uncore_llc) {
+		uncore_llc = amd_uncore_alloc(cpu);
+		if (!uncore_llc)
 			goto fail;
-		uncore_l2->cpu = cpu;
-		uncore_l2->num_counters = NUM_COUNTERS_L2;
-		uncore_l2->rdpmc_base = RDPMC_BASE_L2;
-		uncore_l2->msr_base = MSR_F16H_L2I_PERF_CTL;
-		uncore_l2->active_mask = &amd_l2_active_mask;
-		uncore_l2->pmu = &amd_l2_pmu;
-		uncore_l2->id = -1;
-		*per_cpu_ptr(amd_uncore_l2, cpu) = uncore_l2;
+		uncore_llc->cpu = cpu;
+		uncore_llc->num_counters = NUM_COUNTERS_L2;
+		uncore_llc->rdpmc_base = RDPMC_BASE_LLC;
+		uncore_llc->msr_base = MSR_F16H_L2I_PERF_CTL;
+		uncore_llc->active_mask = &amd_llc_active_mask;
+		uncore_llc->pmu = &amd_llc_pmu;
+		uncore_llc->id = -1;
+		*per_cpu_ptr(amd_uncore_llc, cpu) = uncore_llc;
 	}
 
 	return 0;
@@ -376,17 +376,17 @@ static int amd_uncore_cpu_starting(unsigned int cpu)
 		*per_cpu_ptr(amd_uncore_nb, cpu) = uncore;
 	}
 
-	if (amd_uncore_l2) {
+	if (amd_uncore_llc) {
 		unsigned int apicid = cpu_data(cpu).apicid;
 		unsigned int nshared;
 
-		uncore = *per_cpu_ptr(amd_uncore_l2, cpu);
+		uncore = *per_cpu_ptr(amd_uncore_llc, cpu);
 		cpuid_count(0x8000001d, 2, &eax, &ebx, &ecx, &edx);
 		nshared = ((eax >> 14) & 0xfff) + 1;
 		uncore->id = apicid - (apicid % nshared);
 
-		uncore = amd_uncore_find_online_sibling(uncore, amd_uncore_l2);
-		*per_cpu_ptr(amd_uncore_l2, cpu) = uncore;
+		uncore = amd_uncore_find_online_sibling(uncore, amd_uncore_llc);
+		*per_cpu_ptr(amd_uncore_llc, cpu) = uncore;
 	}
 
 	return 0;
@@ -419,8 +419,8 @@ static int amd_uncore_cpu_online(unsigned int cpu)
 	if (amd_uncore_nb)
 		uncore_online(cpu, amd_uncore_nb);
 
-	if (amd_uncore_l2)
-		uncore_online(cpu, amd_uncore_l2);
+	if (amd_uncore_llc)
+		uncore_online(cpu, amd_uncore_llc);
 
 	return 0;
 }
@@ -456,8 +456,8 @@ static int amd_uncore_cpu_down_prepare(unsigned int cpu)
 	if (amd_uncore_nb)
 		uncore_down_prepare(cpu, amd_uncore_nb);
 
-	if (amd_uncore_l2)
-		uncore_down_prepare(cpu, amd_uncore_l2);
+	if (amd_uncore_llc)
+		uncore_down_prepare(cpu, amd_uncore_llc);
 
 	return 0;
 }
@@ -479,8 +479,8 @@ static int amd_uncore_cpu_dead(unsigned int cpu)
 	if (amd_uncore_nb)
 		uncore_dead(cpu, amd_uncore_nb);
 
-	if (amd_uncore_l2)
-		uncore_dead(cpu, amd_uncore_l2);
+	if (amd_uncore_llc)
+		uncore_dead(cpu, amd_uncore_llc);
 
 	return 0;
 }
@@ -510,16 +510,16 @@ static int __init amd_uncore_init(void)
 	}
 
 	if (boot_cpu_has(X86_FEATURE_PERFCTR_L2)) {
-		amd_uncore_l2 = alloc_percpu(struct amd_uncore *);
-		if (!amd_uncore_l2) {
+		amd_uncore_llc = alloc_percpu(struct amd_uncore *);
+		if (!amd_uncore_llc) {
 			ret = -ENOMEM;
-			goto fail_l2;
+			goto fail_llc;
 		}
-		ret = perf_pmu_register(&amd_l2_pmu, amd_l2_pmu.name, -1);
+		ret = perf_pmu_register(&amd_llc_pmu, amd_llc_pmu.name, -1);
 		if (ret)
-			goto fail_l2;
+			goto fail_llc;
 
-		pr_info("perf: AMD L2I counters detected\n");
+		pr_info("perf: AMD LLC counters detected\n");
 		ret = 0;
 	}
 
@@ -529,7 +529,7 @@ static int __init amd_uncore_init(void)
 	if (cpuhp_setup_state(CPUHP_PERF_X86_AMD_UNCORE_PREP,
 			      "perf/x86/amd/uncore:prepare",
 			      amd_uncore_cpu_up_prepare, amd_uncore_cpu_dead))
-		goto fail_l2;
+		goto fail_llc;
 
 	if (cpuhp_setup_state(CPUHP_AP_PERF_X86_AMD_UNCORE_STARTING,
 			      "perf/x86/amd/uncore:starting",
@@ -546,11 +546,11 @@ static int __init amd_uncore_init(void)
 	cpuhp_remove_state(CPUHP_AP_PERF_X86_AMD_UNCORE_STARTING);
 fail_prep:
 	cpuhp_remove_state(CPUHP_PERF_X86_AMD_UNCORE_PREP);
-fail_l2:
+fail_llc:
 	if (boot_cpu_has(X86_FEATURE_PERFCTR_NB))
 		perf_pmu_unregister(&amd_nb_pmu);
-	if (amd_uncore_l2)
-		free_percpu(amd_uncore_l2);
+	if (amd_uncore_llc)
+		free_percpu(amd_uncore_llc);
 fail_nb:
 	if (amd_uncore_nb)
 		free_percpu(amd_uncore_nb);
-- 
2.7.4

[tip:perf/core] perf/x86/amd/uncore: Rename 'L2' to 'LLC'

[tip-bot for Janakarajan Natarajan]

Commit-ID:  a83f4c00dd6a646ac3c7604ee255d732fc5e0e0b
Gitweb:     http://git.kernel.org/tip/a83f4c00dd6a646ac3c7604ee255d732fc5e0e0b
Author:     Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
AuthorDate: Mon, 16 Jan 2017 17:36:21 -0600
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Mon, 30 Jan 2017 12:01:16 +0100

perf/x86/amd/uncore: Rename 'L2' to 'LLC'

This patch renames L2 counters to LLC counters. In AMD Family17h
processors, L3 cache counter is supported.

Since older families have at most L2 counters, last level cache (LLC)
indicates L2/L3 based on the family.

Signed-off-by: Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/5d8cd8736d8d578354597a548e64ff16210c319b.1484598705.git.Janakarajan.Natarajan@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/events/amd/uncore.c | 98 ++++++++++++++++++++++----------------------
 1 file changed, 49 insertions(+), 49 deletions(-)

diff --git a/arch/x86/events/amd/uncore.c b/arch/x86/events/amd/uncore.c
index a0b1bdb..a53bfbe 100644
--- a/arch/x86/events/amd/uncore.c
+++ b/arch/x86/events/amd/uncore.c
@@ -25,7 +25,7 @@
 #define MAX_COUNTERS		NUM_COUNTERS_NB
 
 #define RDPMC_BASE_NB		6
-#define RDPMC_BASE_L2		10
+#define RDPMC_BASE_LLC		10
 
 #define COUNTER_SHIFT		16
 
@@ -45,30 +45,30 @@ struct amd_uncore {
 };
 
 static struct amd_uncore * __percpu *amd_uncore_nb;
-static struct amd_uncore * __percpu *amd_uncore_l2;
+static struct amd_uncore * __percpu *amd_uncore_llc;
 
 static struct pmu amd_nb_pmu;
-static struct pmu amd_l2_pmu;
+static struct pmu amd_llc_pmu;
 
 static cpumask_t amd_nb_active_mask;
-static cpumask_t amd_l2_active_mask;
+static cpumask_t amd_llc_active_mask;
 
 static bool is_nb_event(struct perf_event *event)
 {
 	return event->pmu->type == amd_nb_pmu.type;
 }
 
-static bool is_l2_event(struct perf_event *event)
+static bool is_llc_event(struct perf_event *event)
 {
-	return event->pmu->type == amd_l2_pmu.type;
+	return event->pmu->type == amd_llc_pmu.type;
 }
 
 static struct amd_uncore *event_to_amd_uncore(struct perf_event *event)
 {
 	if (is_nb_event(event) && amd_uncore_nb)
 		return *per_cpu_ptr(amd_uncore_nb, event->cpu);
-	else if (is_l2_event(event) && amd_uncore_l2)
-		return *per_cpu_ptr(amd_uncore_l2, event->cpu);
+	else if (is_llc_event(event) && amd_uncore_llc)
+		return *per_cpu_ptr(amd_uncore_llc, event->cpu);
 
 	return NULL;
 }
@@ -183,16 +183,16 @@ static int amd_uncore_event_init(struct perf_event *event)
 		return -ENOENT;
 
 	/*
-	 * NB and L2 counters (MSRs) are shared across all cores that share the
-	 * same NB / L2 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.
+	 * 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.
 	 */
 	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
 		return -EINVAL;
 
-	/* NB and L2 counters do not have usr/os/guest/host bits */
+	/* NB and Last level cache counters do not have usr/os/guest/host bits */
 	if (event->attr.exclude_user || event->attr.exclude_kernel ||
 	    event->attr.exclude_host || event->attr.exclude_guest)
 		return -EINVAL;
@@ -226,8 +226,8 @@ static ssize_t amd_uncore_attr_show_cpumask(struct device *dev,
 
 	if (pmu->type == amd_nb_pmu.type)
 		active_mask = &amd_nb_active_mask;
-	else if (pmu->type == amd_l2_pmu.type)
-		active_mask = &amd_l2_active_mask;
+	else if (pmu->type == amd_llc_pmu.type)
+		active_mask = &amd_llc_active_mask;
 	else
 		return 0;
 
@@ -276,7 +276,7 @@ static struct pmu amd_nb_pmu = {
 	.read		= amd_uncore_read,
 };
 
-static struct pmu amd_l2_pmu = {
+static struct pmu amd_llc_pmu = {
 	.task_ctx_nr	= perf_invalid_context,
 	.attr_groups	= amd_uncore_attr_groups,
 	.name		= "amd_l2",
@@ -296,7 +296,7 @@ static struct amd_uncore *amd_uncore_alloc(unsigned int cpu)
 
 static int amd_uncore_cpu_up_prepare(unsigned int cpu)
 {
-	struct amd_uncore *uncore_nb = NULL, *uncore_l2;
+	struct amd_uncore *uncore_nb = NULL, *uncore_llc;
 
 	if (amd_uncore_nb) {
 		uncore_nb = amd_uncore_alloc(cpu);
@@ -312,18 +312,18 @@ static int amd_uncore_cpu_up_prepare(unsigned int cpu)
 		*per_cpu_ptr(amd_uncore_nb, cpu) = uncore_nb;
 	}
 
-	if (amd_uncore_l2) {
-		uncore_l2 = amd_uncore_alloc(cpu);
-		if (!uncore_l2)
+	if (amd_uncore_llc) {
+		uncore_llc = amd_uncore_alloc(cpu);
+		if (!uncore_llc)
 			goto fail;
-		uncore_l2->cpu = cpu;
-		uncore_l2->num_counters = NUM_COUNTERS_L2;
-		uncore_l2->rdpmc_base = RDPMC_BASE_L2;
-		uncore_l2->msr_base = MSR_F16H_L2I_PERF_CTL;
-		uncore_l2->active_mask = &amd_l2_active_mask;
-		uncore_l2->pmu = &amd_l2_pmu;
-		uncore_l2->id = -1;
-		*per_cpu_ptr(amd_uncore_l2, cpu) = uncore_l2;
+		uncore_llc->cpu = cpu;
+		uncore_llc->num_counters = NUM_COUNTERS_L2;
+		uncore_llc->rdpmc_base = RDPMC_BASE_LLC;
+		uncore_llc->msr_base = MSR_F16H_L2I_PERF_CTL;
+		uncore_llc->active_mask = &amd_llc_active_mask;
+		uncore_llc->pmu = &amd_llc_pmu;
+		uncore_llc->id = -1;
+		*per_cpu_ptr(amd_uncore_llc, cpu) = uncore_llc;
 	}
 
 	return 0;
@@ -376,17 +376,17 @@ static int amd_uncore_cpu_starting(unsigned int cpu)
 		*per_cpu_ptr(amd_uncore_nb, cpu) = uncore;
 	}
 
-	if (amd_uncore_l2) {
+	if (amd_uncore_llc) {
 		unsigned int apicid = cpu_data(cpu).apicid;
 		unsigned int nshared;
 
-		uncore = *per_cpu_ptr(amd_uncore_l2, cpu);
+		uncore = *per_cpu_ptr(amd_uncore_llc, cpu);
 		cpuid_count(0x8000001d, 2, &eax, &ebx, &ecx, &edx);
 		nshared = ((eax >> 14) & 0xfff) + 1;
 		uncore->id = apicid - (apicid % nshared);
 
-		uncore = amd_uncore_find_online_sibling(uncore, amd_uncore_l2);
-		*per_cpu_ptr(amd_uncore_l2, cpu) = uncore;
+		uncore = amd_uncore_find_online_sibling(uncore, amd_uncore_llc);
+		*per_cpu_ptr(amd_uncore_llc, cpu) = uncore;
 	}
 
 	return 0;
@@ -419,8 +419,8 @@ static int amd_uncore_cpu_online(unsigned int cpu)
 	if (amd_uncore_nb)
 		uncore_online(cpu, amd_uncore_nb);
 
-	if (amd_uncore_l2)
-		uncore_online(cpu, amd_uncore_l2);
+	if (amd_uncore_llc)
+		uncore_online(cpu, amd_uncore_llc);
 
 	return 0;
 }
@@ -456,8 +456,8 @@ static int amd_uncore_cpu_down_prepare(unsigned int cpu)
 	if (amd_uncore_nb)
 		uncore_down_prepare(cpu, amd_uncore_nb);
 
-	if (amd_uncore_l2)
-		uncore_down_prepare(cpu, amd_uncore_l2);
+	if (amd_uncore_llc)
+		uncore_down_prepare(cpu, amd_uncore_llc);
 
 	return 0;
 }
@@ -479,8 +479,8 @@ static int amd_uncore_cpu_dead(unsigned int cpu)
 	if (amd_uncore_nb)
 		uncore_dead(cpu, amd_uncore_nb);
 
-	if (amd_uncore_l2)
-		uncore_dead(cpu, amd_uncore_l2);
+	if (amd_uncore_llc)
+		uncore_dead(cpu, amd_uncore_llc);
 
 	return 0;
 }
@@ -510,16 +510,16 @@ static int __init amd_uncore_init(void)
 	}
 
 	if (boot_cpu_has(X86_FEATURE_PERFCTR_L2)) {
-		amd_uncore_l2 = alloc_percpu(struct amd_uncore *);
-		if (!amd_uncore_l2) {
+		amd_uncore_llc = alloc_percpu(struct amd_uncore *);
+		if (!amd_uncore_llc) {
 			ret = -ENOMEM;
-			goto fail_l2;
+			goto fail_llc;
 		}
-		ret = perf_pmu_register(&amd_l2_pmu, amd_l2_pmu.name, -1);
+		ret = perf_pmu_register(&amd_llc_pmu, amd_llc_pmu.name, -1);
 		if (ret)
-			goto fail_l2;
+			goto fail_llc;
 
-		pr_info("perf: AMD L2I counters detected\n");
+		pr_info("perf: AMD LLC counters detected\n");
 		ret = 0;
 	}
 
@@ -529,7 +529,7 @@ static int __init amd_uncore_init(void)
 	if (cpuhp_setup_state(CPUHP_PERF_X86_AMD_UNCORE_PREP,
 			      "perf/x86/amd/uncore:prepare",
 			      amd_uncore_cpu_up_prepare, amd_uncore_cpu_dead))
-		goto fail_l2;
+		goto fail_llc;
 
 	if (cpuhp_setup_state(CPUHP_AP_PERF_X86_AMD_UNCORE_STARTING,
 			      "perf/x86/amd/uncore:starting",
@@ -546,11 +546,11 @@ fail_start:
 	cpuhp_remove_state(CPUHP_AP_PERF_X86_AMD_UNCORE_STARTING);
 fail_prep:
 	cpuhp_remove_state(CPUHP_PERF_X86_AMD_UNCORE_PREP);
-fail_l2:
+fail_llc:
 	if (boot_cpu_has(X86_FEATURE_PERFCTR_NB))
 		perf_pmu_unregister(&amd_nb_pmu);
-	if (amd_uncore_l2)
-		free_percpu(amd_uncore_l2);
+	if (amd_uncore_llc)
+		free_percpu(amd_uncore_llc);
 fail_nb:
 	if (amd_uncore_nb)
 		free_percpu(amd_uncore_nb);

[PATCH v2 2/3] perf/x86/amd/uncore: Update number of uncore counters

[Janakarajan Natarajan]

This patch updates the AMD uncore driver to support AMD Family17h
processors. In Family17h, there are two extra last level cache counters.

The maximum available counters is increased and the number of counters
for each uncore type is now based on the family.

Signed-off-by: Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
---
 arch/x86/events/amd/uncore.c | 31 ++++++++++++++++++++++++++++---
 1 file changed, 28 insertions(+), 3 deletions(-)

diff --git a/arch/x86/events/amd/uncore.c b/arch/x86/events/amd/uncore.c
index 24c8537..20af57f 100644
--- a/arch/x86/events/amd/uncore.c
+++ b/arch/x86/events/amd/uncore.c
@@ -22,13 +22,17 @@
 
 #define NUM_COUNTERS_NB		4
 #define NUM_COUNTERS_L2		4
-#define MAX_COUNTERS		NUM_COUNTERS_NB
+#define NUM_COUNTERS_L3		6
+#define MAX_COUNTERS		6
 
 #define RDPMC_BASE_NB		6
 #define RDPMC_BASE_LLC		10
 
 #define COUNTER_SHIFT		16
 
+static int num_counters_llc;
+static int num_counters_nb;
+
 static HLIST_HEAD(uncore_unused_list);
 
 struct amd_uncore {
@@ -303,7 +307,7 @@ static int amd_uncore_cpu_up_prepare(unsigned int cpu)
 		if (!uncore_nb)
 			goto fail;
 		uncore_nb->cpu = cpu;
-		uncore_nb->num_counters = NUM_COUNTERS_NB;
+		uncore_nb->num_counters = num_counters_nb;
 		uncore_nb->rdpmc_base = RDPMC_BASE_NB;
 		uncore_nb->msr_base = MSR_F15H_NB_PERF_CTL;
 		uncore_nb->active_mask = &amd_nb_active_mask;
@@ -317,7 +321,7 @@ static int amd_uncore_cpu_up_prepare(unsigned int cpu)
 		if (!uncore_llc)
 			goto fail;
 		uncore_llc->cpu = cpu;
-		uncore_llc->num_counters = NUM_COUNTERS_L2;
+		uncore_llc->num_counters = num_counters_llc;
 		uncore_llc->rdpmc_base = RDPMC_BASE_LLC;
 		uncore_llc->msr_base = MSR_F16H_L2I_PERF_CTL;
 		uncore_llc->active_mask = &amd_llc_active_mask;
@@ -492,6 +496,27 @@ static int __init amd_uncore_init(void)
 	if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
 		goto fail_nodev;
 
+	switch(boot_cpu_data.x86) {
+		case 23:
+			/* Family 17h */
+			num_counters_nb = NUM_COUNTERS_NB;
+			num_counters_llc = NUM_COUNTERS_L3;
+			break;
+		case 22:
+			/* Family 16h. May change */
+			num_counters_nb = NUM_COUNTERS_NB;
+			num_counters_llc = NUM_COUNTERS_L2;
+			break;
+		default:
+			/*
+			 * All prior families have the same number of
+			 * NorthBridge and Last level cache counters
+			 */
+			num_counters_nb = NUM_COUNTERS_NB;
+			num_counters_llc = NUM_COUNTERS_L2;
+			break;
+	}
+
 	if (!boot_cpu_has(X86_FEATURE_TOPOEXT))
 		goto fail_nodev;
 
-- 
2.7.4

[tip:perf/core] perf/x86/amd/uncore: Update the number of uncore counters

[tip-bot for Janakarajan Natarajan]

Commit-ID:  bc1daef6b5da574bca0a2ec7f9b4d0c5fe0c7d11
Gitweb:     http://git.kernel.org/tip/bc1daef6b5da574bca0a2ec7f9b4d0c5fe0c7d11
Author:     Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
AuthorDate: Mon, 16 Jan 2017 17:36:22 -0600
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Mon, 30 Jan 2017 12:01:17 +0100

perf/x86/amd/uncore: Update the number of uncore counters

This patch updates the AMD uncore driver to support AMD Family17h
processors. In Family17h, there are two extra last level cache counters.

The maximum available counters is increased and the number of counters
for each uncore type is now based on the family.

Signed-off-by: Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/799f9c5be8963cc209d9169a08f4a2643b748dc7.1484598705.git.Janakarajan.Natarajan@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/events/amd/uncore.c | 31 ++++++++++++++++++++++++++++---
 1 file changed, 28 insertions(+), 3 deletions(-)

diff --git a/arch/x86/events/amd/uncore.c b/arch/x86/events/amd/uncore.c
index a53bfbe..e6a2eb5 100644
--- a/arch/x86/events/amd/uncore.c
+++ b/arch/x86/events/amd/uncore.c
@@ -22,13 +22,17 @@
 
 #define NUM_COUNTERS_NB		4
 #define NUM_COUNTERS_L2		4
-#define MAX_COUNTERS		NUM_COUNTERS_NB
+#define NUM_COUNTERS_L3		6
+#define MAX_COUNTERS		6
 
 #define RDPMC_BASE_NB		6
 #define RDPMC_BASE_LLC		10
 
 #define COUNTER_SHIFT		16
 
+static int num_counters_llc;
+static int num_counters_nb;
+
 static HLIST_HEAD(uncore_unused_list);
 
 struct amd_uncore {
@@ -303,7 +307,7 @@ static int amd_uncore_cpu_up_prepare(unsigned int cpu)
 		if (!uncore_nb)
 			goto fail;
 		uncore_nb->cpu = cpu;
-		uncore_nb->num_counters = NUM_COUNTERS_NB;
+		uncore_nb->num_counters = num_counters_nb;
 		uncore_nb->rdpmc_base = RDPMC_BASE_NB;
 		uncore_nb->msr_base = MSR_F15H_NB_PERF_CTL;
 		uncore_nb->active_mask = &amd_nb_active_mask;
@@ -317,7 +321,7 @@ static int amd_uncore_cpu_up_prepare(unsigned int cpu)
 		if (!uncore_llc)
 			goto fail;
 		uncore_llc->cpu = cpu;
-		uncore_llc->num_counters = NUM_COUNTERS_L2;
+		uncore_llc->num_counters = num_counters_llc;
 		uncore_llc->rdpmc_base = RDPMC_BASE_LLC;
 		uncore_llc->msr_base = MSR_F16H_L2I_PERF_CTL;
 		uncore_llc->active_mask = &amd_llc_active_mask;
@@ -492,6 +496,27 @@ static int __init amd_uncore_init(void)
 	if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
 		goto fail_nodev;
 
+	switch(boot_cpu_data.x86) {
+		case 23:
+			/* Family 17h: */
+			num_counters_nb = NUM_COUNTERS_NB;
+			num_counters_llc = NUM_COUNTERS_L3;
+			break;
+		case 22:
+			/* Family 16h - may change: */
+			num_counters_nb = NUM_COUNTERS_NB;
+			num_counters_llc = NUM_COUNTERS_L2;
+			break;
+		default:
+			/*
+			 * All prior families have the same number of
+			 * NorthBridge and Last Level Cache counters
+			 */
+			num_counters_nb = NUM_COUNTERS_NB;
+			num_counters_llc = NUM_COUNTERS_L2;
+			break;
+	}
+
 	if (!boot_cpu_has(X86_FEATURE_TOPOEXT))
 		goto fail_nodev;
 

[PATCH v2 3/3] perf/x86/amd/uncore: Update sysfs attributes for Family17h processors

[Janakarajan Natarajan]

This patch updates the sysfs attributes for AMD Family17h processors. In
Family17h, the event bit position is changed for both the NorthBridge
and Last level cache counters.

The sysfs attributes are assigned based on the family and the type of
the counter.

Signed-off-by: Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
---
 arch/x86/events/amd/uncore.c | 77 ++++++++++++++++++++++++++++++++------------
 1 file changed, 56 insertions(+), 21 deletions(-)

diff --git a/arch/x86/events/amd/uncore.c b/arch/x86/events/amd/uncore.c
index 20af57f..18f1068 100644
--- a/arch/x86/events/amd/uncore.c
+++ b/arch/x86/events/amd/uncore.c
@@ -248,30 +248,47 @@ static struct attribute_group amd_uncore_attr_group = {
 	.attrs = amd_uncore_attrs,
 };
 
-PMU_FORMAT_ATTR(event, "config:0-7,32-35");
-PMU_FORMAT_ATTR(umask, "config:8-15");
-
-static struct attribute *amd_uncore_format_attr[] = {
-	&format_attr_event.attr,
-	&format_attr_umask.attr,
-	NULL,
-};
-
-static struct attribute_group amd_uncore_format_group = {
-	.name = "format",
-	.attrs = amd_uncore_format_attr,
+/*
+ * Similar to PMU_FORMAT_ATTR but allowing for format_attr to be assigned based
+ * on family
+ */
+#define AMD_FORMAT_ATTR(_dev, _name, _format)				     \
+static ssize_t								     \
+_dev##_show##_name(struct device *dev,					     \
+		struct device_attribute *attr,				     \
+		char *page)						     \
+{									     \
+	BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE);			     \
+	return sprintf(page, _format "\n");				     \
+}									     \
+static struct device_attribute format_attr_##_dev##_name = __ATTR_RO(_dev);
+
+/* Used for each uncore counter type */
+#define AMD_ATTRIBUTE(_name)						     \
+static struct attribute *amd_uncore_format_attr_##_name[] = {		     \
+	&format_attr_event_##_name.attr,				     \
+	&format_attr_umask.attr,					     \
+	NULL,								     \
+};									     \
+static struct attribute_group amd_uncore_format_group_##_name = {	     \
+	.name = "format",						     \
+	.attrs = amd_uncore_format_attr_##_name,			     \
+};									     \
+static const struct attribute_group *amd_uncore_attr_groups_##_name[] = {    \
+	&amd_uncore_attr_group,						     \
+	&amd_uncore_format_group_##_name,				     \
+	NULL,								     \
 };
 
-static const struct attribute_group *amd_uncore_attr_groups[] = {
-	&amd_uncore_attr_group,
-	&amd_uncore_format_group,
-	NULL,
-};
+AMD_FORMAT_ATTR(event, , "config:0-7,32-35");
+AMD_FORMAT_ATTR(umask, , "config:8-15");
+AMD_FORMAT_ATTR(event, _df, "config:0-7,32-35,59-60");
+AMD_FORMAT_ATTR(event, _l3, "config:0-7");
+AMD_ATTRIBUTE(df);
+AMD_ATTRIBUTE(l3);
 
 static struct pmu amd_nb_pmu = {
 	.task_ctx_nr	= perf_invalid_context,
-	.attr_groups	= amd_uncore_attr_groups,
-	.name		= "amd_nb",
 	.event_init	= amd_uncore_event_init,
 	.add		= amd_uncore_add,
 	.del		= amd_uncore_del,
@@ -282,8 +299,6 @@ static struct pmu amd_nb_pmu = {
 
 static struct pmu amd_llc_pmu = {
 	.task_ctx_nr	= perf_invalid_context,
-	.attr_groups	= amd_uncore_attr_groups,
-	.name		= "amd_l2",
 	.event_init	= amd_uncore_event_init,
 	.add		= amd_uncore_add,
 	.del		= amd_uncore_del,
@@ -501,11 +516,25 @@ static int __init amd_uncore_init(void)
 			/* Family 17h */
 			num_counters_nb = NUM_COUNTERS_NB;
 			num_counters_llc = NUM_COUNTERS_L3;
+			/*
+			 * For Family17h, the NorthBridge counters are
+			 * re-purposed as Data Fabric counters. Also, support is
+			 * added for L3 counters. The pmus are exported based on
+			 * family as either L2 or L3 and NB or DF.
+			 */
+			amd_nb_pmu.name = "amd_df";
+			amd_llc_pmu.name = "amd_l3";
+			format_attr_event_df.show = &event_show_df;
+			format_attr_event_l3.show = &event_show_l3;
 			break;
 		case 22:
 			/* Family 16h. May change */
 			num_counters_nb = NUM_COUNTERS_NB;
 			num_counters_llc = NUM_COUNTERS_L2;
+			amd_nb_pmu.name = "amd_nb";
+			amd_llc_pmu.name = "amd_l2";
+			format_attr_event_df = format_attr_event;
+			format_attr_event_l3 = format_attr_event;
 			break;
 		default:
 			/*
@@ -514,8 +543,14 @@ static int __init amd_uncore_init(void)
 			 */
 			num_counters_nb = NUM_COUNTERS_NB;
 			num_counters_llc = NUM_COUNTERS_L2;
+			amd_nb_pmu.name = "amd_nb";
+			amd_llc_pmu.name = "amd_l2";
+			format_attr_event_df = format_attr_event;
+			format_attr_event_l3 = format_attr_event;
 			break;
 	}
+	amd_nb_pmu.attr_groups = amd_uncore_attr_groups_df;
+	amd_llc_pmu.attr_groups = amd_uncore_attr_groups_l3;
 
 	if (!boot_cpu_has(X86_FEATURE_TOPOEXT))
 		goto fail_nodev;
-- 
2.7.4

[tip:perf/core] perf/x86/amd/uncore: Update sysfs attributes for Family17h processors

[tip-bot for Janakarajan Natarajan]

Commit-ID:  da6adaea2b7ef658c61a557c28508668eac29fe1
Gitweb:     http://git.kernel.org/tip/da6adaea2b7ef658c61a557c28508668eac29fe1
Author:     Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
AuthorDate: Mon, 16 Jan 2017 17:36:23 -0600
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Mon, 30 Jan 2017 12:01:18 +0100

perf/x86/amd/uncore: Update sysfs attributes for Family17h processors

This patch updates the sysfs attributes for AMD Family17h processors. In
Family17h, the event bit position is changed for both the NorthBridge
and Last level cache counters.

The sysfs attributes are assigned based on the family and the type of
the counter.

Signed-off-by: Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/617570ed3634e804991f95db62c3cf3856a9d2a7.1484598705.git.Janakarajan.Natarajan@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/events/amd/uncore.c | 77 ++++++++++++++++++++++++++++++++------------
 1 file changed, 56 insertions(+), 21 deletions(-)

diff --git a/arch/x86/events/amd/uncore.c b/arch/x86/events/amd/uncore.c
index e6a2eb5..4d1f7f2d 100644
--- a/arch/x86/events/amd/uncore.c
+++ b/arch/x86/events/amd/uncore.c
@@ -248,30 +248,47 @@ static struct attribute_group amd_uncore_attr_group = {
 	.attrs = amd_uncore_attrs,
 };
 
-PMU_FORMAT_ATTR(event, "config:0-7,32-35");
-PMU_FORMAT_ATTR(umask, "config:8-15");
-
-static struct attribute *amd_uncore_format_attr[] = {
-	&format_attr_event.attr,
-	&format_attr_umask.attr,
-	NULL,
-};
-
-static struct attribute_group amd_uncore_format_group = {
-	.name = "format",
-	.attrs = amd_uncore_format_attr,
+/*
+ * Similar to PMU_FORMAT_ATTR but allowing for format_attr to be assigned based
+ * on family
+ */
+#define AMD_FORMAT_ATTR(_dev, _name, _format)				     \
+static ssize_t								     \
+_dev##_show##_name(struct device *dev,					     \
+		struct device_attribute *attr,				     \
+		char *page)						     \
+{									     \
+	BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE);			     \
+	return sprintf(page, _format "\n");				     \
+}									     \
+static struct device_attribute format_attr_##_dev##_name = __ATTR_RO(_dev);
+
+/* Used for each uncore counter type */
+#define AMD_ATTRIBUTE(_name)						     \
+static struct attribute *amd_uncore_format_attr_##_name[] = {		     \
+	&format_attr_event_##_name.attr,				     \
+	&format_attr_umask.attr,					     \
+	NULL,								     \
+};									     \
+static struct attribute_group amd_uncore_format_group_##_name = {	     \
+	.name = "format",						     \
+	.attrs = amd_uncore_format_attr_##_name,			     \
+};									     \
+static const struct attribute_group *amd_uncore_attr_groups_##_name[] = {    \
+	&amd_uncore_attr_group,						     \
+	&amd_uncore_format_group_##_name,				     \
+	NULL,								     \
 };
 
-static const struct attribute_group *amd_uncore_attr_groups[] = {
-	&amd_uncore_attr_group,
-	&amd_uncore_format_group,
-	NULL,
-};
+AMD_FORMAT_ATTR(event, , "config:0-7,32-35");
+AMD_FORMAT_ATTR(umask, , "config:8-15");
+AMD_FORMAT_ATTR(event, _df, "config:0-7,32-35,59-60");
+AMD_FORMAT_ATTR(event, _l3, "config:0-7");
+AMD_ATTRIBUTE(df);
+AMD_ATTRIBUTE(l3);
 
 static struct pmu amd_nb_pmu = {
 	.task_ctx_nr	= perf_invalid_context,
-	.attr_groups	= amd_uncore_attr_groups,
-	.name		= "amd_nb",
 	.event_init	= amd_uncore_event_init,
 	.add		= amd_uncore_add,
 	.del		= amd_uncore_del,
@@ -282,8 +299,6 @@ static struct pmu amd_nb_pmu = {
 
 static struct pmu amd_llc_pmu = {
 	.task_ctx_nr	= perf_invalid_context,
-	.attr_groups	= amd_uncore_attr_groups,
-	.name		= "amd_l2",
 	.event_init	= amd_uncore_event_init,
 	.add		= amd_uncore_add,
 	.del		= amd_uncore_del,
@@ -501,11 +516,25 @@ static int __init amd_uncore_init(void)
 			/* Family 17h: */
 			num_counters_nb = NUM_COUNTERS_NB;
 			num_counters_llc = NUM_COUNTERS_L3;
+			/*
+			 * For Family17h, the NorthBridge counters are
+			 * re-purposed as Data Fabric counters. Also, support is
+			 * added for L3 counters. The pmus are exported based on
+			 * family as either L2 or L3 and NB or DF.
+			 */
+			amd_nb_pmu.name = "amd_df";
+			amd_llc_pmu.name = "amd_l3";
+			format_attr_event_df.show = &event_show_df;
+			format_attr_event_l3.show = &event_show_l3;
 			break;
 		case 22:
 			/* Family 16h - may change: */
 			num_counters_nb = NUM_COUNTERS_NB;
 			num_counters_llc = NUM_COUNTERS_L2;
+			amd_nb_pmu.name = "amd_nb";
+			amd_llc_pmu.name = "amd_l2";
+			format_attr_event_df = format_attr_event;
+			format_attr_event_l3 = format_attr_event;
 			break;
 		default:
 			/*
@@ -514,8 +543,14 @@ static int __init amd_uncore_init(void)
 			 */
 			num_counters_nb = NUM_COUNTERS_NB;
 			num_counters_llc = NUM_COUNTERS_L2;
+			amd_nb_pmu.name = "amd_nb";
+			amd_llc_pmu.name = "amd_l2";
+			format_attr_event_df = format_attr_event;
+			format_attr_event_l3 = format_attr_event;
 			break;
 	}
+	amd_nb_pmu.attr_groups = amd_uncore_attr_groups_df;
+	amd_llc_pmu.attr_groups = amd_uncore_attr_groups_l3;
 
 	if (!boot_cpu_has(X86_FEATURE_TOPOEXT))
 		goto fail_nodev;

Re: [PATCH v2 0/3] perf/x86/amd/uncore: Update AMD uncore driver

[Peter Zijlstra]

On Mon, Jan 16, 2017 at 05:36:20PM -0600, Janakarajan Natarajan wrote:
> This patchset updates the AMD uncore driver to support AMD Family17h processors.
> 
> First, the current L2 counters are renamed to Last level cache (LLC) counter to 
> reflect either L2 or L3 counters depending on the family. Next, the maximum 
> number of counters available is increased and the counters per uncore type is 
> now based on the family.
> 
> Also, in Family17h, the NorthBridge counters are re-purposed as Data Fabric (DF) 
> counters. Therefore, this patch exports the pmu name accordingly based on the 
> processor family.
> 

Thanks!