Skip to contentSkip to conversation

perf: Skip and warn on unknown format 'configN' attrs

PATCH Superseded Rob Herring 2 participants 4 messages Sep 9, 2022 – Sep 13, 2022 perf events

v2 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 perf: Skip and warn on unknown format 'configN' attrs

If the kernel exposes a new perf_event_attr field in a format attr, perf will return an error stating the specified PMU can't be found. For example, a format attr with 'config3:0-63' causes an error as config3 is unknown to perf. This causes a compatibility issue between a newer kernel with older perf tool.

Before this change with a kernel adding 'config3' I get:

$ perf record -e arm_spe// -- true
event syntax error: 'arm_spe//'
                     \___ Cannot find PMU `arm_spe'. Missing kernel support?
Run 'perf list' for a list of valid events

Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]

-e, --event <event> event selector. use 'perf list' to list
available events

After this change, I get:

$ perf record -e arm_spe// -- true
WARNING: format 'inv_event_filter' requires 'config3' which is not supported by this version of perf!
[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.091 MB perf.data ]

To support unknown configN formats, rework the YACC implementation to pass any config[0-9]+ format to perf_pmu__new_format() to handle with a warning.

Note that the user will get the warning if *any* PMU has an unsupported format attr even if that PMU isn't used. This is because perf tool scans all the PMUs.

---
v2:
- Rework YACC code to handle configN formats in C code
- Add a warning when an unknown configN attr is found

v1: https://lore.kernel.org/all/[email protected]/

---
 tools/perf/util/pmu.c |  6 ++++++
 tools/perf/util/pmu.l |  2 --
 tools/perf/util/pmu.y | 15 ++++-----------
 3 files changed, 10 insertions(+), 13 deletions(-)
tools/perf/util/pmu.c +6 -0
index 89655d53117a..6757db7d559c 100644
@@ -1475,6 +1475,12 @@ int perf_pmu__new_format(struct list_head *list, char *name,
 {
 	struct perf_pmu_format *format;
 
+	if (config > PERF_PMU_FORMAT_VALUE_CONFIG2) {
+		pr_warning("WARNING: format '%s' requires 'config%d' which is not supported by this version of perf!\n",
+			   name, config);
+		return 0;
+	}
+
 	format = zalloc(sizeof(*format));
 	if (!format)
 		return -ENOMEM;
tools/perf/util/pmu.l +0 -2
index a15d9fbd7c0e..58b4926cfaca 100644
@@ -27,8 +27,6 @@ num_dec         [0-9]+
 
 {num_dec}	{ return value(10); }
 config		{ return PP_CONFIG; }
-config1		{ return PP_CONFIG1; }
-config2		{ return PP_CONFIG2; }
 -		{ return '-'; }
 :		{ return ':'; }
 ,		{ return ','; }
tools/perf/util/pmu.y +4 -11
index bfd7e8509869..283efe059819 100644
@@ -20,7 +20,7 @@ do { \
 
 %}
 
-%token PP_CONFIG PP_CONFIG1 PP_CONFIG2
+%token PP_CONFIG
 %token PP_VALUE PP_ERROR
 %type <num> PP_VALUE
 %type <bits> bit_term
@@ -47,18 +47,11 @@ PP_CONFIG ':' bits
 				      $3));
 }
 |
-PP_CONFIG1 ':' bits
+PP_CONFIG PP_VALUE ':' bits
 {
 	ABORT_ON(perf_pmu__new_format(format, name,
-				      PERF_PMU_FORMAT_VALUE_CONFIG1,
-				      $3));
-}
-|
-PP_CONFIG2 ':' bits
-{
-	ABORT_ON(perf_pmu__new_format(format, name,
-				      PERF_PMU_FORMAT_VALUE_CONFIG2,
-				      $3));
+				      $2,
+				      $4));
 }
 
 bits:
Signature · 1 line
2.34.1
#1 lore raw

Hi Rob,

Quoted 26 lines · Rob Herring

If the kernel exposes a new perf_event_attr field in a format attr, perf will return an error stating the specified PMU can't be found. For example, a format attr with 'config3:0-63' causes an error as config3 is unknown to perf. This causes a compatibility issue between a newer kernel with older perf tool.

Before this change with a kernel adding 'config3' I get:

$ perf record -e arm_spe// -- true
event syntax error: 'arm_spe//'
                     \___ Cannot find PMU `arm_spe'. Missing kernel support?
Run 'perf list' for a list of valid events

Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]

-e, --event <event> event selector. use 'perf list' to list
available events

After this change, I get:

$ perf record -e arm_spe// -- true
WARNING: format 'inv_event_filter' requires 'config3' which is not supported by this version of perf!
[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.091 MB perf.data ]
Quoted 7 lines

To support unknown configN formats, rework the YACC implementation to pass any config[0-9]+ format to perf_pmu__new_format() to handle with a warning.

Note that the user will get the warning if *any* PMU has an unsupported format attr even if that PMU isn't used. This is because perf tool scans all the PMUs.

I think essentially you want to provide a bug fixing and the fixing can be back ported on long term supported kernels? If this is the case, it's good to add a fixes tag like below?

Fixes cd82a32e9924 ("perf tools: Add perf pmu object to access pmu format definition")
Quoted 18 lines

---
v2:
- Rework YACC code to handle configN formats in C code
- Add a warning when an unknown configN attr is found

v1: https://lore.kernel.org/all/[email protected]/

---
 tools/perf/util/pmu.c |  6 ++++++
 tools/perf/util/pmu.l |  2 --
 tools/perf/util/pmu.y | 15 ++++-----------
 3 files changed, 10 insertions(+), 13 deletions(-)
tools/perf/util/pmu.c +0 -0
index 89655d53117a..6757db7d559c 100644

tools/perf/util/pmu.c int perf_pmu__new_format(struct list_head *list, char *name, · line 1478

Quoted 4 lines
tools/perf/util/pmu.c +1 -0
@@ -1475,6 +1475,12 @@ int perf_pmu__new_format(struct list_head *list, char *name,
 {
 	struct perf_pmu_format *format;
 
+	if (config > PERF_PMU_FORMAT_VALUE_CONFIG2) {

It's good to add a new macro PERF_PMU_FORMAT_VALUE_CONFIG_END in util/pmu.h. Then at here we can check the condition:

       if (config >= PERF_PMU_FORMAT_VALUE_CONFIG_END) {

tools/perf/util/pmu.c int perf_pmu__new_format(struct list_head *list, char *name, · line 1480

Quoted 2 lines
tools/perf/util/pmu.c +2 -0
+		pr_warning("WARNING: format '%s' requires 'config%d' which is not supported by this version of perf!\n",
+			   name, config);

... so at here we can print log like:

pr_warning("WARNING: format '%s' requires 'config%d' which is not "
"supported by this version of perf (maximum config%d)!\n", name, config, PERF_PMU_FORMAT_VALUE_CONFIG_END - 1);

The rest of this patch is fine for me.

As a side topic, I know you want to support the SPEv1.2 feature with config3, seems to me a complete patch set series should include the changes for supporting config3 as well. This can give more clear view for how to fix incompatibility issue between old and new kernels, and how to support config3 in the latest kernel, but it's fine for me if you want to split into two patch sets.

Thanks,
Leo

Quoted 57 lines
tools/perf/util/pmu.c +3 -0
+		return 0;
+	}
+
 	format = zalloc(sizeof(*format));
 	if (!format)
 		return -ENOMEM;
tools/perf/util/pmu.l +0 -2
index a15d9fbd7c0e..58b4926cfaca 100644
@@ -27,8 +27,6 @@ num_dec         [0-9]+
 
 {num_dec}	{ return value(10); }
 config		{ return PP_CONFIG; }
-config1		{ return PP_CONFIG1; }
-config2		{ return PP_CONFIG2; }
 -		{ return '-'; }
 :		{ return ':'; }
 ,		{ return ','; }
tools/perf/util/pmu.y +4 -11
index bfd7e8509869..283efe059819 100644
@@ -20,7 +20,7 @@ do { \
 
 %}
 
-%token PP_CONFIG PP_CONFIG1 PP_CONFIG2
+%token PP_CONFIG
 %token PP_VALUE PP_ERROR
 %type <num> PP_VALUE
 %type <bits> bit_term
@@ -47,18 +47,11 @@ PP_CONFIG ':' bits
 				      $3));
 }
 |
-PP_CONFIG1 ':' bits
+PP_CONFIG PP_VALUE ':' bits
 {
 	ABORT_ON(perf_pmu__new_format(format, name,
-				      PERF_PMU_FORMAT_VALUE_CONFIG1,
-				      $3));
-}
-|
-PP_CONFIG2 ':' bits
-{
-	ABORT_ON(perf_pmu__new_format(format, name,
-				      PERF_PMU_FORMAT_VALUE_CONFIG2,
-				      $3));
+				      $2,
+				      $4));
 }
 
 bits:
Signature · 1 line
2.34.1
#2 lore raw
Quoted 34 lines · Leo Yan

Hi Rob,

Quoted 30 lines · Rob Herring

If the kernel exposes a new perf_event_attr field in a format attr, perf will return an error stating the specified PMU can't be found. For example, a format attr with 'config3:0-63' causes an error as config3 is unknown to perf. This causes a compatibility issue between a newer kernel with older perf tool.

Before this change with a kernel adding 'config3' I get:

$ perf record -e arm_spe// -- true
event syntax error: 'arm_spe//'
                     \___ Cannot find PMU `arm_spe'. Missing kernel support?
Run 'perf list' for a list of valid events

Usage: perf record [<options>] [<command>]
or: perf record [<options>] -- <command> [<options>]

-e, --event <event> event selector. use 'perf list' to list
available events

After this change, I get:

$ perf record -e arm_spe// -- true
WARNING: format 'inv_event_filter' requires 'config3' which is not supported by this version of perf!
[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.091 MB perf.data ]

To support unknown configN formats, rework the YACC implementation to pass any config[0-9]+ format to perf_pmu__new_format() to handle with a warning.

Quoted 6 lines
Quoted 3 lines

Note that the user will get the warning if *any* PMU has an unsupported format attr even if that PMU isn't used. This is because perf tool scans all the PMUs.

I think essentially you want to provide a bug fixing and the fixing can be back ported on long term supported kernels?

Yes, certainly. I forgot to note that on this version.

Quoted 4 lines

If this is the case,
it's good to add a fixes tag like below?

Fixes cd82a32e9924 ("perf tools: Add perf pmu object to access pmu format definition")

Probably not too important given this is from 2012.

Quoted 22 lines
Quoted 22 lines

---
v2:
- Rework YACC code to handle configN formats in C code
- Add a warning when an unknown configN attr is found

v1: https://lore.kernel.org/all/[email protected]/

---
 tools/perf/util/pmu.c |  6 ++++++
 tools/perf/util/pmu.l |  2 --
 tools/perf/util/pmu.y | 15 ++++-----------
 3 files changed, 10 insertions(+), 13 deletions(-)
tools/perf/util/pmu.c +0 -0
index 89655d53117a..6757db7d559c 100644
@@ -1475,6 +1475,12 @@ int perf_pmu__new_format(struct list_head *list, char *name,
 {
      struct perf_pmu_format *format;
 
Quoted 4 lines

tools/perf/util/pmu.c int perf_pmu__new_format(struct list_head *list, char *name, · line 1478

Quoted 1 line
tools/perf/util/pmu.c +1 -0
+     if (config > PERF_PMU_FORMAT_VALUE_CONFIG2) {

It's good to add a new macro PERF_PMU_FORMAT_VALUE_CONFIG_END in util/pmu.h. Then at here we can check the condition:

Sure.

Quoted 11 lines
       if (config >= PERF_PMU_FORMAT_VALUE_CONFIG_END) {

tools/perf/util/pmu.c int perf_pmu__new_format(struct list_head *list, char *name, · line 1480

Quoted 2 lines
tools/perf/util/pmu.c +2 -0
+             pr_warning("WARNING: format '%s' requires 'config%d' which is not supported by this version of perf!\n",
+                        name, config);

... so at here we can print log like:

pr_warning("WARNING: format '%s' requires 'config%d' which is not "
"supported by this version of perf (maximum config%d)!\n", name, config, PERF_PMU_FORMAT_VALUE_CONFIG_END - 1);

I was trying to keep it to one line and given configN isn't expanded frequently it should be simple enough to figure out what version you need.

Quoted 9 lines

The rest of this patch is fine for me.

As a side topic, I know you want to support the SPEv1.2 feature with config3, seems to me a complete patch set series should include the changes for supporting config3 as well. This can give more clear view for how to fix incompatibility issue between old and new kernels, and how to support config3 in the latest kernel, but it's fine for me if you want to split into two patch sets.

I've sent out the SPE kernel support separately. I was told by Arnaldo to split perf tool and kernel side changes.

Rob

#3 lore raw

On Mon, Sep 12, 2022 at 08:55:32AM -0500, Rob Herring wrote:

[...]

Quoted 6 lines
Quoted 4 lines

If this is the case,
it's good to add a fixes tag like below?

Fixes cd82a32e9924 ("perf tools: Add perf pmu object to access pmu format definition")

Probably not too important given this is from 2012.

Which means since 2012 there have no requirement for config3 :)

AFAICT, perf tool should be forward compatible, so I think it's good for this patch to be landed on (old) LTS kernels.

Quoted 36 lines
Quoted 21 lines
Quoted 21 lines

---
v2:
- Rework YACC code to handle configN formats in C code
- Add a warning when an unknown configN attr is found

v1: https://lore.kernel.org/all/[email protected]/

---
 tools/perf/util/pmu.c |  6 ++++++
 tools/perf/util/pmu.l |  2 --
 tools/perf/util/pmu.y | 15 ++++-----------
 3 files changed, 10 insertions(+), 13 deletions(-)
tools/perf/util/pmu.c +0 -0
index 89655d53117a..6757db7d559c 100644
@@ -1475,6 +1475,12 @@ int perf_pmu__new_format(struct list_head *list, char *name,
 {
      struct perf_pmu_format *format;
 
Quoted 4 lines

tools/perf/util/pmu.c int perf_pmu__new_format(struct list_head *list, char *name, · line 1478

Quoted 1 line
tools/perf/util/pmu.c +1 -0
+     if (config > PERF_PMU_FORMAT_VALUE_CONFIG2) {

It's good to add a new macro PERF_PMU_FORMAT_VALUE_CONFIG_END in util/pmu.h. Then at here we can check the condition:

Sure.

Quoted 8 lines
       if (config >= PERF_PMU_FORMAT_VALUE_CONFIG_END) {

tools/perf/util/pmu.c int perf_pmu__new_format(struct list_head *list, char *name, · line 1480

Quoted 2 lines
tools/perf/util/pmu.c +2 -0
+             pr_warning("WARNING: format '%s' requires 'config%d' which is not supported by this version of perf!\n",
+                        name, config);

... so at here we can print log like:

Quoted 7 lines
Quoted 3 lines

pr_warning("WARNING: format '%s' requires 'config%d' which is not "
"supported by this version of perf (maximum config%d)!\n", name, config, PERF_PMU_FORMAT_VALUE_CONFIG_END - 1);

I was trying to keep it to one line and given configN isn't expanded frequently it should be simple enough to figure out what version you need.

It's fine for me.

Quoted 11 lines
Quoted 8 lines

The rest of this patch is fine for me.

As a side topic, I know you want to support the SPEv1.2 feature with config3, seems to me a complete patch set series should include the changes for supporting config3 as well. This can give more clear view for how to fix incompatibility issue between old and new kernels, and how to support config3 in the latest kernel, but it's fine for me if you want to split into two patch sets.

I've sent out the SPE kernel support separately. I was told by Arnaldo to split perf tool and kernel side changes.

To be clear, it's right to split kernel and userspace parts into two patch set, but for userspace change I cannot find you sent patch to support config3 in perf tool, something in the [1] is missed, otherwise I don't think you provide complete support for config3 (and thus we have no chance to use new event format).

Thanks,
Leo

[1] https://termbin.com/vdph