[PATCH 0/4] perf mem amd: Fix few logic bugs

[PATCH 0/4] perf mem amd: Fix few logic bugs

[Ravi Bangoria]

Recent pmu refactoring changes[1] introduced a notion of core vs other
pmus and made perf mem/c2c code depend only on core pmus, which is
logically wrong for AMD as perf mem/c2c on AMD depends on IBS OP pmu,
not the core pmu. Although user visible perf mem/c2c functionality is
still working fine, internal code logic is wrong. Fix those.

[1] https://lore.kernel.org/r/20230527072210.2900565-1-irogers@google.com

Ravi Bangoria (4):
  perf pmus: Describe semantics of 'core_pmus' and 'other_pmus'
  perf tool x86: Consolidate is_amd check into single function
  perf mem amd: Fix perf_pmus__num_mem_pmus()
  perf mem amd: Scan all PMUs instead of just core ones

 tools/perf/arch/x86/util/Build        |  1 +
 tools/perf/arch/x86/util/env.c        | 19 +++++++++++++++++++
 tools/perf/arch/x86/util/env.h        |  7 +++++++
 tools/perf/arch/x86/util/evsel.c      | 16 ++--------------
 tools/perf/arch/x86/util/mem-events.c | 24 +++++++-----------------
 tools/perf/arch/x86/util/pmu.c        | 15 +++++++++++++++
 tools/perf/util/mem-events.c          | 16 ++++++++++++----
 tools/perf/util/mem-events.h          |  1 +
 tools/perf/util/pmus.c                | 15 ++++++++++++++-
 9 files changed, 78 insertions(+), 36 deletions(-)
 create mode 100644 tools/perf/arch/x86/util/env.c
 create mode 100644 tools/perf/arch/x86/util/env.h

-- 
2.40.1

[PATCH 1/4] perf pmus: Describe semantics of 'core_pmus' and 'other_pmus'

[Ravi Bangoria]

'core_pmus' and 'other_pmus' are independent of hw core pmu and uncore
pmus. For example, AMD IBS PMUs are present in each SMT-thread but they
belongs to 'other_pmus'. Add a comment describing what these list
contains and how they are treated.

Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
---
 tools/perf/util/pmus.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/tools/perf/util/pmus.c b/tools/perf/util/pmus.c
index e1d0a93147e5..e505d2fef828 100644
--- a/tools/perf/util/pmus.c
+++ b/tools/perf/util/pmus.c
@@ -12,6 +12,19 @@
 #include "pmu.h"
 #include "print-events.h"
 
+/*
+ * core_pmus:  A PMU belongs to core_pmus if it's name is "cpu" or it's sysfs
+ *             directory contains "cpus" file. All PMUs belonging to core_pmus
+ *             must have pmu->is_core=1. If there are more than one PMUs in
+ *             this list, perf interprets it as a heterogeneous platform.
+ *             (FWIW, certain ARM platforms having heterogeneous cores uses
+ *             homogeneous PMU in which case it will be treated as homogeneous
+ *             platform by perf because core_pmus will have only one entry.)
+ * other_pmus: All other PMUs which are not part of core_pmus list. Does not
+ *             matter whether it is a per SMT-thread or outside of the core in
+ *             hw. i.e. PMUs belonging to other_pmus must have pmu->is_core=0
+ *             but pmu->is_uncore could be 0 or 1.
+ */
 static LIST_HEAD(core_pmus);
 static LIST_HEAD(other_pmus);
 static bool read_sysfs_core_pmus;
-- 
2.40.1

[PATCH 2/4] perf tool x86: Consolidate is_amd check into single function

[Ravi Bangoria]

There are multiple places where x86 specific code determines AMD vs
Intel arch and acts based on that. Consolidate those checks into a
single function.

Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
---
 tools/perf/arch/x86/util/Build        |  1 +
 tools/perf/arch/x86/util/env.c        | 19 +++++++++++++++++++
 tools/perf/arch/x86/util/env.h        |  7 +++++++
 tools/perf/arch/x86/util/evsel.c      | 16 ++--------------
 tools/perf/arch/x86/util/mem-events.c | 19 ++-----------------
 5 files changed, 31 insertions(+), 31 deletions(-)
 create mode 100644 tools/perf/arch/x86/util/env.c
 create mode 100644 tools/perf/arch/x86/util/env.h

diff --git a/tools/perf/arch/x86/util/Build b/tools/perf/arch/x86/util/Build
index 195ccfdef7aa..005907cb97d8 100644
--- a/tools/perf/arch/x86/util/Build
+++ b/tools/perf/arch/x86/util/Build
@@ -10,6 +10,7 @@ perf-y += evlist.o
 perf-y += mem-events.o
 perf-y += evsel.o
 perf-y += iostat.o
+perf-y += env.o
 
 perf-$(CONFIG_DWARF) += dwarf-regs.o
 perf-$(CONFIG_BPF_PROLOGUE) += dwarf-regs.o
diff --git a/tools/perf/arch/x86/util/env.c b/tools/perf/arch/x86/util/env.c
new file mode 100644
index 000000000000..33b87f8ac1cc
--- /dev/null
+++ b/tools/perf/arch/x86/util/env.c
@@ -0,0 +1,19 @@
+// SPDX-License-Identifier: GPL-2.0
+#include "linux/string.h"
+#include "util/env.h"
+#include "env.h"
+
+bool x86__is_amd_cpu(void)
+{
+	struct perf_env env = { .total_mem = 0, };
+	static int is_amd; /* 0: Uninitialized, 1: Yes, -1: No */
+
+	if (is_amd)
+		goto ret;
+
+	perf_env__cpuid(&env);
+	is_amd = env.cpuid && strstarts(env.cpuid, "AuthenticAMD") ? 1 : -1;
+
+ret:
+	return is_amd >= 1 ? true : false;
+}
diff --git a/tools/perf/arch/x86/util/env.h b/tools/perf/arch/x86/util/env.h
new file mode 100644
index 000000000000..d78f080b6b3f
--- /dev/null
+++ b/tools/perf/arch/x86/util/env.h
@@ -0,0 +1,7 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _X86_ENV_H
+#define _X86_ENV_H
+
+bool x86__is_amd_cpu(void);
+
+#endif /* _X86_ENV_H */
diff --git a/tools/perf/arch/x86/util/evsel.c b/tools/perf/arch/x86/util/evsel.c
index 25da46c8cca9..512c2d885d24 100644
--- a/tools/perf/arch/x86/util/evsel.c
+++ b/tools/perf/arch/x86/util/evsel.c
@@ -8,6 +8,7 @@
 #include "linux/string.h"
 #include "evsel.h"
 #include "util/debug.h"
+#include "env.h"
 
 #define IBS_FETCH_L3MISSONLY   (1ULL << 59)
 #define IBS_OP_L3MISSONLY      (1ULL << 16)
@@ -78,23 +79,10 @@ void arch__post_evsel_config(struct evsel *evsel, struct perf_event_attr *attr)
 {
 	struct perf_pmu *evsel_pmu, *ibs_fetch_pmu, *ibs_op_pmu;
 	static int warned_once;
-	/* 0: Uninitialized, 1: Yes, -1: No */
-	static int is_amd;
 
-	if (warned_once || is_amd == -1)
+	if (warned_once || !x86__is_amd_cpu())
 		return;
 
-	if (!is_amd) {
-		struct perf_env *env = evsel__env(evsel);
-
-		if (!perf_env__cpuid(env) || !env->cpuid ||
-		    !strstarts(env->cpuid, "AuthenticAMD")) {
-			is_amd = -1;
-			return;
-		}
-		is_amd = 1;
-	}
-
 	evsel_pmu = evsel__find_pmu(evsel);
 	if (!evsel_pmu)
 		return;
diff --git a/tools/perf/arch/x86/util/mem-events.c b/tools/perf/arch/x86/util/mem-events.c
index 32879d12a8d5..a8a782bcb121 100644
--- a/tools/perf/arch/x86/util/mem-events.c
+++ b/tools/perf/arch/x86/util/mem-events.c
@@ -5,6 +5,7 @@
 #include "map_symbol.h"
 #include "mem-events.h"
 #include "linux/string.h"
+#include "env.h"
 
 static char mem_loads_name[100];
 static bool mem_loads_name__init;
@@ -27,28 +28,12 @@ static struct perf_mem_event perf_mem_events_amd[PERF_MEM_EVENTS__MAX] = {
 	E("mem-ldst",	"ibs_op//",	"ibs_op"),
 };
 
-static int perf_mem_is_amd_cpu(void)
-{
-	struct perf_env env = { .total_mem = 0, };
-
-	perf_env__cpuid(&env);
-	if (env.cpuid && strstarts(env.cpuid, "AuthenticAMD"))
-		return 1;
-	return -1;
-}
-
 struct perf_mem_event *perf_mem_events__ptr(int i)
 {
-	/* 0: Uninitialized, 1: Yes, -1: No */
-	static int is_amd;
-
 	if (i >= PERF_MEM_EVENTS__MAX)
 		return NULL;
 
-	if (!is_amd)
-		is_amd = perf_mem_is_amd_cpu();
-
-	if (is_amd == 1)
+	if (x86__is_amd_cpu())
 		return &perf_mem_events_amd[i];
 
 	return &perf_mem_events_intel[i];
-- 
2.40.1

Re: [PATCH 2/4] perf tool x86: Consolidate is_amd check into single function

[Arnaldo Carvalho de Melo]

Em Tue, Jun 13, 2023 at 03:25:04PM +0530, Ravi Bangoria escreveu:
> There are multiple places where x86 specific code determines AMD vs
> Intel arch and acts based on that. Consolidate those checks into a
> single function.

I'm cherry picking this one now.

- Arnaldo
 
> Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
> ---
>  tools/perf/arch/x86/util/Build        |  1 +
>  tools/perf/arch/x86/util/env.c        | 19 +++++++++++++++++++
>  tools/perf/arch/x86/util/env.h        |  7 +++++++
>  tools/perf/arch/x86/util/evsel.c      | 16 ++--------------
>  tools/perf/arch/x86/util/mem-events.c | 19 ++-----------------
>  5 files changed, 31 insertions(+), 31 deletions(-)
>  create mode 100644 tools/perf/arch/x86/util/env.c
>  create mode 100644 tools/perf/arch/x86/util/env.h
> 
> diff --git a/tools/perf/arch/x86/util/Build b/tools/perf/arch/x86/util/Build
> index 195ccfdef7aa..005907cb97d8 100644
> --- a/tools/perf/arch/x86/util/Build
> +++ b/tools/perf/arch/x86/util/Build
> @@ -10,6 +10,7 @@ perf-y += evlist.o
>  perf-y += mem-events.o
>  perf-y += evsel.o
>  perf-y += iostat.o
> +perf-y += env.o
>  
>  perf-$(CONFIG_DWARF) += dwarf-regs.o
>  perf-$(CONFIG_BPF_PROLOGUE) += dwarf-regs.o
> diff --git a/tools/perf/arch/x86/util/env.c b/tools/perf/arch/x86/util/env.c
> new file mode 100644
> index 000000000000..33b87f8ac1cc
> --- /dev/null
> +++ b/tools/perf/arch/x86/util/env.c
> @@ -0,0 +1,19 @@
> +// SPDX-License-Identifier: GPL-2.0
> +#include "linux/string.h"
> +#include "util/env.h"
> +#include "env.h"
> +
> +bool x86__is_amd_cpu(void)
> +{
> +	struct perf_env env = { .total_mem = 0, };
> +	static int is_amd; /* 0: Uninitialized, 1: Yes, -1: No */
> +
> +	if (is_amd)
> +		goto ret;
> +
> +	perf_env__cpuid(&env);
> +	is_amd = env.cpuid && strstarts(env.cpuid, "AuthenticAMD") ? 1 : -1;
> +
> +ret:
> +	return is_amd >= 1 ? true : false;
> +}
> diff --git a/tools/perf/arch/x86/util/env.h b/tools/perf/arch/x86/util/env.h
> new file mode 100644
> index 000000000000..d78f080b6b3f
> --- /dev/null
> +++ b/tools/perf/arch/x86/util/env.h
> @@ -0,0 +1,7 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +#ifndef _X86_ENV_H
> +#define _X86_ENV_H
> +
> +bool x86__is_amd_cpu(void);
> +
> +#endif /* _X86_ENV_H */
> diff --git a/tools/perf/arch/x86/util/evsel.c b/tools/perf/arch/x86/util/evsel.c
> index 25da46c8cca9..512c2d885d24 100644
> --- a/tools/perf/arch/x86/util/evsel.c
> +++ b/tools/perf/arch/x86/util/evsel.c
> @@ -8,6 +8,7 @@
>  #include "linux/string.h"
>  #include "evsel.h"
>  #include "util/debug.h"
> +#include "env.h"
>  
>  #define IBS_FETCH_L3MISSONLY   (1ULL << 59)
>  #define IBS_OP_L3MISSONLY      (1ULL << 16)
> @@ -78,23 +79,10 @@ void arch__post_evsel_config(struct evsel *evsel, struct perf_event_attr *attr)
>  {
>  	struct perf_pmu *evsel_pmu, *ibs_fetch_pmu, *ibs_op_pmu;
>  	static int warned_once;
> -	/* 0: Uninitialized, 1: Yes, -1: No */
> -	static int is_amd;
>  
> -	if (warned_once || is_amd == -1)
> +	if (warned_once || !x86__is_amd_cpu())
>  		return;
>  
> -	if (!is_amd) {
> -		struct perf_env *env = evsel__env(evsel);
> -
> -		if (!perf_env__cpuid(env) || !env->cpuid ||
> -		    !strstarts(env->cpuid, "AuthenticAMD")) {
> -			is_amd = -1;
> -			return;
> -		}
> -		is_amd = 1;
> -	}
> -
>  	evsel_pmu = evsel__find_pmu(evsel);
>  	if (!evsel_pmu)
>  		return;
> diff --git a/tools/perf/arch/x86/util/mem-events.c b/tools/perf/arch/x86/util/mem-events.c
> index 32879d12a8d5..a8a782bcb121 100644
> --- a/tools/perf/arch/x86/util/mem-events.c
> +++ b/tools/perf/arch/x86/util/mem-events.c
> @@ -5,6 +5,7 @@
>  #include "map_symbol.h"
>  #include "mem-events.h"
>  #include "linux/string.h"
> +#include "env.h"
>  
>  static char mem_loads_name[100];
>  static bool mem_loads_name__init;
> @@ -27,28 +28,12 @@ static struct perf_mem_event perf_mem_events_amd[PERF_MEM_EVENTS__MAX] = {
>  	E("mem-ldst",	"ibs_op//",	"ibs_op"),
>  };
>  
> -static int perf_mem_is_amd_cpu(void)
> -{
> -	struct perf_env env = { .total_mem = 0, };
> -
> -	perf_env__cpuid(&env);
> -	if (env.cpuid && strstarts(env.cpuid, "AuthenticAMD"))
> -		return 1;
> -	return -1;
> -}
> -
>  struct perf_mem_event *perf_mem_events__ptr(int i)
>  {
> -	/* 0: Uninitialized, 1: Yes, -1: No */
> -	static int is_amd;
> -
>  	if (i >= PERF_MEM_EVENTS__MAX)
>  		return NULL;
>  
> -	if (!is_amd)
> -		is_amd = perf_mem_is_amd_cpu();
> -
> -	if (is_amd == 1)
> +	if (x86__is_amd_cpu())
>  		return &perf_mem_events_amd[i];
>  
>  	return &perf_mem_events_intel[i];
> -- 
> 2.40.1
> 

-- 

- Arnaldo

Re: [PATCH 2/4] perf tool x86: Consolidate is_amd check into single function

[Ian Rogers]

On Tue, Jun 13, 2023 at 2:56 AM Ravi Bangoria <ravi.bangoria@amd.com> wrote:
>
> There are multiple places where x86 specific code determines AMD vs
> Intel arch and acts based on that. Consolidate those checks into a
> single function.
>
> Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>

Acked-by: Ian Rogers <irogers@google.com>

Thanks,
Ian

> ---
>  tools/perf/arch/x86/util/Build        |  1 +
>  tools/perf/arch/x86/util/env.c        | 19 +++++++++++++++++++
>  tools/perf/arch/x86/util/env.h        |  7 +++++++
>  tools/perf/arch/x86/util/evsel.c      | 16 ++--------------
>  tools/perf/arch/x86/util/mem-events.c | 19 ++-----------------
>  5 files changed, 31 insertions(+), 31 deletions(-)
>  create mode 100644 tools/perf/arch/x86/util/env.c
>  create mode 100644 tools/perf/arch/x86/util/env.h
>
> diff --git a/tools/perf/arch/x86/util/Build b/tools/perf/arch/x86/util/Build
> index 195ccfdef7aa..005907cb97d8 100644
> --- a/tools/perf/arch/x86/util/Build
> +++ b/tools/perf/arch/x86/util/Build
> @@ -10,6 +10,7 @@ perf-y += evlist.o
>  perf-y += mem-events.o
>  perf-y += evsel.o
>  perf-y += iostat.o
> +perf-y += env.o
>
>  perf-$(CONFIG_DWARF) += dwarf-regs.o
>  perf-$(CONFIG_BPF_PROLOGUE) += dwarf-regs.o
> diff --git a/tools/perf/arch/x86/util/env.c b/tools/perf/arch/x86/util/env.c
> new file mode 100644
> index 000000000000..33b87f8ac1cc
> --- /dev/null
> +++ b/tools/perf/arch/x86/util/env.c
> @@ -0,0 +1,19 @@
> +// SPDX-License-Identifier: GPL-2.0
> +#include "linux/string.h"
> +#include "util/env.h"
> +#include "env.h"
> +
> +bool x86__is_amd_cpu(void)
> +{
> +       struct perf_env env = { .total_mem = 0, };
> +       static int is_amd; /* 0: Uninitialized, 1: Yes, -1: No */
> +
> +       if (is_amd)
> +               goto ret;
> +
> +       perf_env__cpuid(&env);
> +       is_amd = env.cpuid && strstarts(env.cpuid, "AuthenticAMD") ? 1 : -1;
> +
> +ret:
> +       return is_amd >= 1 ? true : false;
> +}
> diff --git a/tools/perf/arch/x86/util/env.h b/tools/perf/arch/x86/util/env.h
> new file mode 100644
> index 000000000000..d78f080b6b3f
> --- /dev/null
> +++ b/tools/perf/arch/x86/util/env.h
> @@ -0,0 +1,7 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +#ifndef _X86_ENV_H
> +#define _X86_ENV_H
> +
> +bool x86__is_amd_cpu(void);
> +
> +#endif /* _X86_ENV_H */
> diff --git a/tools/perf/arch/x86/util/evsel.c b/tools/perf/arch/x86/util/evsel.c
> index 25da46c8cca9..512c2d885d24 100644
> --- a/tools/perf/arch/x86/util/evsel.c
> +++ b/tools/perf/arch/x86/util/evsel.c
> @@ -8,6 +8,7 @@
>  #include "linux/string.h"
>  #include "evsel.h"
>  #include "util/debug.h"
> +#include "env.h"
>
>  #define IBS_FETCH_L3MISSONLY   (1ULL << 59)
>  #define IBS_OP_L3MISSONLY      (1ULL << 16)
> @@ -78,23 +79,10 @@ void arch__post_evsel_config(struct evsel *evsel, struct perf_event_attr *attr)
>  {
>         struct perf_pmu *evsel_pmu, *ibs_fetch_pmu, *ibs_op_pmu;
>         static int warned_once;
> -       /* 0: Uninitialized, 1: Yes, -1: No */
> -       static int is_amd;
>
> -       if (warned_once || is_amd == -1)
> +       if (warned_once || !x86__is_amd_cpu())
>                 return;
>
> -       if (!is_amd) {
> -               struct perf_env *env = evsel__env(evsel);
> -
> -               if (!perf_env__cpuid(env) || !env->cpuid ||
> -                   !strstarts(env->cpuid, "AuthenticAMD")) {
> -                       is_amd = -1;
> -                       return;
> -               }
> -               is_amd = 1;
> -       }
> -
>         evsel_pmu = evsel__find_pmu(evsel);
>         if (!evsel_pmu)
>                 return;
> diff --git a/tools/perf/arch/x86/util/mem-events.c b/tools/perf/arch/x86/util/mem-events.c
> index 32879d12a8d5..a8a782bcb121 100644
> --- a/tools/perf/arch/x86/util/mem-events.c
> +++ b/tools/perf/arch/x86/util/mem-events.c
> @@ -5,6 +5,7 @@
>  #include "map_symbol.h"
>  #include "mem-events.h"
>  #include "linux/string.h"
> +#include "env.h"
>
>  static char mem_loads_name[100];
>  static bool mem_loads_name__init;
> @@ -27,28 +28,12 @@ static struct perf_mem_event perf_mem_events_amd[PERF_MEM_EVENTS__MAX] = {
>         E("mem-ldst",   "ibs_op//",     "ibs_op"),
>  };
>
> -static int perf_mem_is_amd_cpu(void)
> -{
> -       struct perf_env env = { .total_mem = 0, };
> -
> -       perf_env__cpuid(&env);
> -       if (env.cpuid && strstarts(env.cpuid, "AuthenticAMD"))
> -               return 1;
> -       return -1;
> -}
> -
>  struct perf_mem_event *perf_mem_events__ptr(int i)
>  {
> -       /* 0: Uninitialized, 1: Yes, -1: No */
> -       static int is_amd;
> -
>         if (i >= PERF_MEM_EVENTS__MAX)
>                 return NULL;
>
> -       if (!is_amd)
> -               is_amd = perf_mem_is_amd_cpu();
> -
> -       if (is_amd == 1)
> +       if (x86__is_amd_cpu())
>                 return &perf_mem_events_amd[i];
>
>         return &perf_mem_events_intel[i];
> --
> 2.40.1
>

Re: [PATCH 2/4] perf tool x86: Consolidate is_amd check into single function

[Arnaldo Carvalho de Melo]

Em Tue, Jun 13, 2023 at 12:49:50PM -0700, Ian Rogers escreveu:
> On Tue, Jun 13, 2023 at 2:56 AM Ravi Bangoria <ravi.bangoria@amd.com> wrote:
> >
> > There are multiple places where x86 specific code determines AMD vs
> > Intel arch and acts based on that. Consolidate those checks into a
> > single function.
> >
> > Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
> 
> Acked-by: Ian Rogers <irogers@google.com>

Thanks, added to the cset.

- Arnaldo

[PATCH 3/4] perf mem amd: Fix perf_pmus__num_mem_pmus()

[Ravi Bangoria]

AMD cpus does not contain hybrid cores. Also, perf mem/c2c on AMD
internally uses IBS OP PMU, not the core PMU.

Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
---
 tools/perf/arch/x86/util/pmu.c | 15 +++++++++++++++
 tools/perf/util/pmus.c         |  2 +-
 2 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/tools/perf/arch/x86/util/pmu.c b/tools/perf/arch/x86/util/pmu.c
index 3c0de3370d7e..8a20d28d9927 100644
--- a/tools/perf/arch/x86/util/pmu.c
+++ b/tools/perf/arch/x86/util/pmu.c
@@ -14,6 +14,8 @@
 #include "../../../util/intel-bts.h"
 #include "../../../util/pmu.h"
 #include "../../../util/fncache.h"
+#include "../../../util/pmus.h"
+#include "env.h"
 
 struct pmu_alias {
 	char *name;
@@ -168,3 +170,16 @@ char *pmu_find_alias_name(const char *name)
 
 	return __pmu_find_alias_name(name);
 }
+
+int perf_pmus__num_mem_pmus(void)
+{
+	/*
+	 * AMD does not have hybrid cores and also uses IBS OP
+	 * pmu for perf mem/c2c.
+	 */
+	if (x86__is_amd_cpu())
+		return 1;
+
+	/* Intel uses core pmus for perf mem/c2c */
+	return perf_pmus__num_core_pmus();
+}
diff --git a/tools/perf/util/pmus.c b/tools/perf/util/pmus.c
index e505d2fef828..0ed943932945 100644
--- a/tools/perf/util/pmus.c
+++ b/tools/perf/util/pmus.c
@@ -240,7 +240,7 @@ const struct perf_pmu *perf_pmus__pmu_for_pmu_filter(const char *str)
 	return NULL;
 }
 
-int perf_pmus__num_mem_pmus(void)
+int __weak perf_pmus__num_mem_pmus(void)
 {
 	/* All core PMUs are for mem events. */
 	return perf_pmus__num_core_pmus();
-- 
2.40.1

Re: [PATCH 3/4] perf mem amd: Fix perf_pmus__num_mem_pmus()

[Ian Rogers]

On Tue, Jun 13, 2023 at 2:56 AM Ravi Bangoria <ravi.bangoria@amd.com> wrote:
>
> AMD cpus does not contain hybrid cores. Also, perf mem/c2c on AMD
> internally uses IBS OP PMU, not the core PMU.
>
> Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
> ---
>  tools/perf/arch/x86/util/pmu.c | 15 +++++++++++++++
>  tools/perf/util/pmus.c         |  2 +-
>  2 files changed, 16 insertions(+), 1 deletion(-)
>
> diff --git a/tools/perf/arch/x86/util/pmu.c b/tools/perf/arch/x86/util/pmu.c
> index 3c0de3370d7e..8a20d28d9927 100644
> --- a/tools/perf/arch/x86/util/pmu.c
> +++ b/tools/perf/arch/x86/util/pmu.c
> @@ -14,6 +14,8 @@
>  #include "../../../util/intel-bts.h"
>  #include "../../../util/pmu.h"
>  #include "../../../util/fncache.h"
> +#include "../../../util/pmus.h"
> +#include "env.h"
>
>  struct pmu_alias {
>         char *name;
> @@ -168,3 +170,16 @@ char *pmu_find_alias_name(const char *name)
>
>         return __pmu_find_alias_name(name);
>  }
> +
> +int perf_pmus__num_mem_pmus(void)
> +{
> +       /*
> +        * AMD does not have hybrid cores and also uses IBS OP
> +        * pmu for perf mem/c2c.
> +        */
> +       if (x86__is_amd_cpu())
> +               return 1;

The code and comment seem out of sync here. For the hybrid part
perf_pmus__num_core_pmus() will yield 1 if there is no hybrid, so we
can just use perf_pmus__num_core_pmus(). For the IBS OP part, does
that mean that AMD should have 2 mem pmus? Or is IBS OP a core PMU?
Can we add this as an example in the core/other documentation in patch
1, as you've done for ARM, for clarity.

Thanks,
Ian

> +
> +       /* Intel uses core pmus for perf mem/c2c */
> +       return perf_pmus__num_core_pmus();
> +}
> diff --git a/tools/perf/util/pmus.c b/tools/perf/util/pmus.c
> index e505d2fef828..0ed943932945 100644
> --- a/tools/perf/util/pmus.c
> +++ b/tools/perf/util/pmus.c
> @@ -240,7 +240,7 @@ const struct perf_pmu *perf_pmus__pmu_for_pmu_filter(const char *str)
>         return NULL;
>  }
>
> -int perf_pmus__num_mem_pmus(void)
> +int __weak perf_pmus__num_mem_pmus(void)
>  {
>         /* All core PMUs are for mem events. */
>         return perf_pmus__num_core_pmus();
> --
> 2.40.1
>

Re: [PATCH 3/4] perf mem amd: Fix perf_pmus__num_mem_pmus()

[Ravi Bangoria]

>> +int perf_pmus__num_mem_pmus(void)
>> +{
>> +       /*
>> +        * AMD does not have hybrid cores and also uses IBS OP
>> +        * pmu for perf mem/c2c.
>> +        */
>> +       if (x86__is_amd_cpu())
>> +               return 1;
> 
> The code and comment seem out of sync here. For the hybrid part
> perf_pmus__num_core_pmus() will yield 1 if there is no hybrid, so we
> can just use perf_pmus__num_core_pmus(). For the IBS OP part, does
> that mean that AMD should have 2 mem pmus? Or is IBS OP a core PMU?

Sure. Let me remove hybrid part from the comment.

There are two IBS pmus: ibs_fetch// and ibs_op//. Both of them are
independent of the core pmu (cpu//). An instance of all 3 pmus is
present in each hw SMT thread. And, perf mem/c2c internally uses
ibs_op// pmu on AMD. See tools/perf/arch/x86/util/mem-events.c

  static struct perf_mem_event perf_mem_events_amd[PERF_MEM_EVENTS__MAX] = {
          E(NULL,         NULL,           NULL),
          E(NULL,         NULL,           NULL),
          E("mem-ldst",   "ibs_op//",     "ibs_op"),
  };

Hope this clarifies.

> Can we add this as an example in the core/other documentation in patch
> 1, as you've done for ARM, for clarity.

Sure. Let me also add IBS example in the patch #1 comment.

Thanks,
Ravi

[PATCH 4/4] perf mem amd: Scan all PMUs instead of just core ones

[Ravi Bangoria]

Scanning only core PMUs is not sufficient on AMD since perf mem on
AMD uses IBS OP PMU, which is independent of core PMU.

Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
---
 tools/perf/arch/x86/util/mem-events.c |  5 +++++
 tools/perf/util/mem-events.c          | 16 ++++++++++++----
 tools/perf/util/mem-events.h          |  1 +
 3 files changed, 18 insertions(+), 4 deletions(-)

diff --git a/tools/perf/arch/x86/util/mem-events.c b/tools/perf/arch/x86/util/mem-events.c
index a8a782bcb121..43af872e89a6 100644
--- a/tools/perf/arch/x86/util/mem-events.c
+++ b/tools/perf/arch/x86/util/mem-events.c
@@ -91,3 +91,8 @@ char *perf_mem_events__name(int i, char *pmu_name)
 
 	return (char *)e->name;
 }
+
+bool perf_mem_events__via_core_pmus(void)
+{
+	return !x86__is_amd_cpu();
+}
diff --git a/tools/perf/util/mem-events.c b/tools/perf/util/mem-events.c
index be15aadb6b14..0c04f883d634 100644
--- a/tools/perf/util/mem-events.c
+++ b/tools/perf/util/mem-events.c
@@ -109,6 +109,14 @@ static bool perf_mem_event__supported(const char *mnt, char *sysfs_name)
 	return !stat(path, &st);
 }
 
+bool __weak perf_mem_events__via_core_pmus(void)
+{
+	return true;
+}
+
+#define perf_mem_scan_next_pmu(pmu) \
+	(perf_mem_events__via_core_pmus() ? perf_pmus__scan_core(pmu) : perf_pmus__scan(pmu))
+
 int perf_mem_events__init(void)
 {
 	const char *mnt = sysfs__mount();
@@ -130,7 +138,7 @@ int perf_mem_events__init(void)
 		if (!e->tag)
 			continue;
 
-		while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
+		while ((pmu = perf_mem_scan_next_pmu(pmu)) != NULL) {
 			scnprintf(sysfs_name, sizeof(sysfs_name), e->sysfs_name, pmu->name);
 			e->supported |= perf_mem_event__supported(mnt, sysfs_name);
 		}
@@ -165,7 +173,7 @@ static void perf_mem_events__print_unsupport_hybrid(struct perf_mem_event *e,
 	char sysfs_name[100];
 	struct perf_pmu *pmu = NULL;
 
-	while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
+	while ((pmu = perf_mem_scan_next_pmu(pmu)) != NULL) {
 		scnprintf(sysfs_name, sizeof(sysfs_name), e->sysfs_name,
 			  pmu->name);
 		if (!perf_mem_event__supported(mnt, sysfs_name)) {
@@ -188,7 +196,7 @@ int perf_mem_events__record_args(const char **rec_argv, int *argv_nr,
 		if (!e->record)
 			continue;
 
-		if (perf_pmus__num_core_pmus() == 1) {
+		if (perf_pmus__num_mem_pmus() == 1) {
 			if (!e->supported) {
 				pr_err("failed: event '%s' not supported\n",
 				       perf_mem_events__name(j, NULL));
@@ -203,7 +211,7 @@ int perf_mem_events__record_args(const char **rec_argv, int *argv_nr,
 				return -1;
 			}
 
-			while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
+			while ((pmu = perf_mem_scan_next_pmu(pmu)) != NULL) {
 				rec_argv[i++] = "-e";
 				s = perf_mem_events__name(j, pmu->name);
 				if (s) {
diff --git a/tools/perf/util/mem-events.h b/tools/perf/util/mem-events.h
index 12372309d60e..d650eb311113 100644
--- a/tools/perf/util/mem-events.h
+++ b/tools/perf/util/mem-events.h
@@ -36,6 +36,7 @@ enum {
 extern unsigned int perf_mem_events__loads_ldlat;
 
 int perf_mem_events__parse(const char *str);
+bool perf_mem_events__via_core_pmus(void);
 int perf_mem_events__init(void);
 
 char *perf_mem_events__name(int i, char *pmu_name);
-- 
2.40.1

Re: [PATCH 4/4] perf mem amd: Scan all PMUs instead of just core ones

[Ian Rogers]

On Tue, Jun 13, 2023 at 2:56 AM Ravi Bangoria <ravi.bangoria@amd.com> wrote:
>
> Scanning only core PMUs is not sufficient on AMD since perf mem on
> AMD uses IBS OP PMU, which is independent of core PMU.
>
> Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
> ---
>  tools/perf/arch/x86/util/mem-events.c |  5 +++++
>  tools/perf/util/mem-events.c          | 16 ++++++++++++----
>  tools/perf/util/mem-events.h          |  1 +
>  3 files changed, 18 insertions(+), 4 deletions(-)
>
> diff --git a/tools/perf/arch/x86/util/mem-events.c b/tools/perf/arch/x86/util/mem-events.c
> index a8a782bcb121..43af872e89a6 100644
> --- a/tools/perf/arch/x86/util/mem-events.c
> +++ b/tools/perf/arch/x86/util/mem-events.c
> @@ -91,3 +91,8 @@ char *perf_mem_events__name(int i, char *pmu_name)
>
>         return (char *)e->name;
>  }
> +
> +bool perf_mem_events__via_core_pmus(void)
> +{
> +       return !x86__is_amd_cpu();
> +}
> diff --git a/tools/perf/util/mem-events.c b/tools/perf/util/mem-events.c
> index be15aadb6b14..0c04f883d634 100644
> --- a/tools/perf/util/mem-events.c
> +++ b/tools/perf/util/mem-events.c
> @@ -109,6 +109,14 @@ static bool perf_mem_event__supported(const char *mnt, char *sysfs_name)
>         return !stat(path, &st);
>  }
>
> +bool __weak perf_mem_events__via_core_pmus(void)
> +{
> +       return true;
> +}
> +
> +#define perf_mem_scan_next_pmu(pmu) \
> +       (perf_mem_events__via_core_pmus() ? perf_pmus__scan_core(pmu) : perf_pmus__scan(pmu))
> +
>  int perf_mem_events__init(void)
>  {
>         const char *mnt = sysfs__mount();
> @@ -130,7 +138,7 @@ int perf_mem_events__init(void)
>                 if (!e->tag)
>                         continue;
>
> -               while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
> +               while ((pmu = perf_mem_scan_next_pmu(pmu)) != NULL) {
>                         scnprintf(sysfs_name, sizeof(sysfs_name), e->sysfs_name, pmu->name);
>                         e->supported |= perf_mem_event__supported(mnt, sysfs_name);
>                 }
> @@ -165,7 +173,7 @@ static void perf_mem_events__print_unsupport_hybrid(struct perf_mem_event *e,
>         char sysfs_name[100];
>         struct perf_pmu *pmu = NULL;
>
> -       while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
> +       while ((pmu = perf_mem_scan_next_pmu(pmu)) != NULL) {

It was my mistake to optimize this, I think we can just go back to:
perf_pmus__scan(pmu)
which would remove a lot of the weak/macros etc. here. We can have a
comment as to why this is scan not scan_core, because of AMD. I plan
to further improve overhead of PMUs so I'm not worried about losing
the small performance win from this.

Thanks,
Ian

>                 scnprintf(sysfs_name, sizeof(sysfs_name), e->sysfs_name,
>                           pmu->name);
>                 if (!perf_mem_event__supported(mnt, sysfs_name)) {
> @@ -188,7 +196,7 @@ int perf_mem_events__record_args(const char **rec_argv, int *argv_nr,
>                 if (!e->record)
>                         continue;
>
> -               if (perf_pmus__num_core_pmus() == 1) {
> +               if (perf_pmus__num_mem_pmus() == 1) {
>                         if (!e->supported) {
>                                 pr_err("failed: event '%s' not supported\n",
>                                        perf_mem_events__name(j, NULL));
> @@ -203,7 +211,7 @@ int perf_mem_events__record_args(const char **rec_argv, int *argv_nr,
>                                 return -1;
>                         }
>
> -                       while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
> +                       while ((pmu = perf_mem_scan_next_pmu(pmu)) != NULL) {
>                                 rec_argv[i++] = "-e";
>                                 s = perf_mem_events__name(j, pmu->name);
>                                 if (s) {
> diff --git a/tools/perf/util/mem-events.h b/tools/perf/util/mem-events.h
> index 12372309d60e..d650eb311113 100644
> --- a/tools/perf/util/mem-events.h
> +++ b/tools/perf/util/mem-events.h
> @@ -36,6 +36,7 @@ enum {
>  extern unsigned int perf_mem_events__loads_ldlat;
>
>  int perf_mem_events__parse(const char *str);
> +bool perf_mem_events__via_core_pmus(void);
>  int perf_mem_events__init(void);
>
>  char *perf_mem_events__name(int i, char *pmu_name);
> --
> 2.40.1
>

Re: [PATCH 4/4] perf mem amd: Scan all PMUs instead of just core ones

[Ravi Bangoria]

>> @@ -165,7 +173,7 @@ static void perf_mem_events__print_unsupport_hybrid(struct perf_mem_event *e,
>>         char sysfs_name[100];
>>         struct perf_pmu *pmu = NULL;
>>
>> -       while ((pmu = perf_pmus__scan_core(pmu)) != NULL) {
>> +       while ((pmu = perf_mem_scan_next_pmu(pmu)) != NULL) {
> 
> It was my mistake to optimize this,

Not really. I mean, there was already a bug which just got exacerbated.

> I think we can just go back to:
> perf_pmus__scan(pmu)
> which would remove a lot of the weak/macros etc. here. We can have a
> comment as to why this is scan not scan_core, because of AMD. I plan
> to further improve overhead of PMUs so I'm not worried about losing
> the small performance win from this.

Sure. Let me do that.

Thanks,
Ravi