perf sched stats: Fixes and improvements

[PATCH v1 1/6] perf header: Replace hardcoded max cpus by MAX_NR_CPUS

Posted by Swapnil Sapkal 1 week, 5 days ago

cpumask and cpulist from cpu-domain header have hardcoded max_cpus value
of 1024. Current systems have more cpus than this value. Replace it with
MAX_NR_CPUS. Also define a macro to represent domain name length.

Fixes: d40c68a49f69 ("perf header: Support CPU DOMAIN relation info")
Reported-by: Shrikanth Hegde <sshegde@linux.ibm.com>
Signed-off-by: Swapnil Sapkal <swapnil.sapkal@amd.com>
---
 tools/perf/util/header.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index eefd1cd73b6a..31c3bab1b10a 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -76,6 +76,7 @@ static const u64 __perf_magic2    = 0x32454c4946524550ULL;
 static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
 
 #define PERF_MAGIC	__perf_magic2
+#define DNAME_LEN	16
 
 const char perf_version_string[] = PERF_VERSION;
 
@@ -1616,10 +1617,10 @@ static int write_pmu_caps(struct feat_fd *ff,
 
 struct cpu_domain_map **build_cpu_domain_map(u32 *schedstat_version, u32 *max_sched_domains, u32 nr)
 {
+	char dname[DNAME_LEN], cpumask[MAX_NR_CPUS];
 	struct domain_info *domain_info;
 	struct cpu_domain_map **cd_map;
-	char dname[16], cpumask[256];
-	char cpulist[1024];
+	char cpulist[MAX_NR_CPUS];
 	char *line = NULL;
 	u32 cpu, domain;
 	u32 dcount = 0;
-- 
2.43.0

Re: [PATCH v1 1/6] perf header: Replace hardcoded max cpus by MAX_NR_CPUS

Posted by Shrikanth Hegde 1 week, 5 days ago


On 1/28/26 12:19 AM, Swapnil Sapkal wrote:
> cpumask and cpulist from cpu-domain header have hardcoded max_cpus value
> of 1024. Current systems have more cpus than this value. Replace it with
> MAX_NR_CPUS. Also define a macro to represent domain name length.
> 
> Fixes: d40c68a49f69 ("perf header: Support CPU DOMAIN relation info")
> Reported-by: Shrikanth Hegde <sshegde@linux.ibm.com>
> Signed-off-by: Swapnil Sapkal <swapnil.sapkal@amd.com>
> ---
>   tools/perf/util/header.c | 5 +++--
>   1 file changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
> index eefd1cd73b6a..31c3bab1b10a 100644
> --- a/tools/perf/util/header.c
> +++ b/tools/perf/util/header.c
> @@ -76,6 +76,7 @@ static const u64 __perf_magic2    = 0x32454c4946524550ULL;
>   static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
>   
>   #define PERF_MAGIC	__perf_magic2
> +#define DNAME_LEN	16
>   
>   const char perf_version_string[] = PERF_VERSION;
>   
> @@ -1616,10 +1617,10 @@ static int write_pmu_caps(struct feat_fd *ff,
>   
>   struct cpu_domain_map **build_cpu_domain_map(u32 *schedstat_version, u32 *max_sched_domains, u32 nr)
>   {
> +	char dname[DNAME_LEN], cpumask[MAX_NR_CPUS];
>   	struct domain_info *domain_info;
>   	struct cpu_domain_map **cd_map;
> -	char dname[16], cpumask[256];
> -	char cpulist[1024];
> +	char cpulist[MAX_NR_CPUS];
>   	char *line = NULL;
>   	u32 cpu, domain;
>   	u32 dcount = 0;

Looking at
https://lore.kernel.org/all/20260119175833.340369-3-swapnil.sapkal@amd.com/

There was one more "char cpus[1024]"  in tools/perf/util/util.c.
You may need to fix that too. It is unlikely but, if one has created exclusive
cpusets comprising of only one cpu from a core, maybe you will run out the length.
So better use the MAX_NR_CPUS there as well.

Re: [PATCH v1 1/6] perf header: Replace hardcoded max cpus by MAX_NR_CPUS

Posted by Swapnil Sapkal 1 week, 4 days ago

Hi Srikanth, Arnaldo,

Thank you for reviewing the patches.

On 28-01-2026 13:09, Shrikanth Hegde wrote:
> 
> 
> On 1/28/26 12:19 AM, Swapnil Sapkal wrote:
>> cpumask and cpulist from cpu-domain header have hardcoded max_cpus value
>> of 1024. Current systems have more cpus than this value. Replace it with
>> MAX_NR_CPUS. Also define a macro to represent domain name length.
>>
>> Fixes: d40c68a49f69 ("perf header: Support CPU DOMAIN relation info")
>> Reported-by: Shrikanth Hegde <sshegde@linux.ibm.com>
>> Signed-off-by: Swapnil Sapkal <swapnil.sapkal@amd.com>
>> ---
>>   tools/perf/util/header.c | 5 +++--
>>   1 file changed, 3 insertions(+), 2 deletions(-)
>>
>> diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
>> index eefd1cd73b6a..31c3bab1b10a 100644
>> --- a/tools/perf/util/header.c
>> +++ b/tools/perf/util/header.c
>> @@ -76,6 +76,7 @@ static const u64 __perf_magic2    = 
>> 0x32454c4946524550ULL;
>>   static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
>>   #define PERF_MAGIC    __perf_magic2
>> +#define DNAME_LEN    16
>>   const char perf_version_string[] = PERF_VERSION;
>> @@ -1616,10 +1617,10 @@ static int write_pmu_caps(struct feat_fd *ff,
>>   struct cpu_domain_map **build_cpu_domain_map(u32 *schedstat_version, 
>> u32 *max_sched_domains, u32 nr)
>>   {
>> +    char dname[DNAME_LEN], cpumask[MAX_NR_CPUS];
>>       struct domain_info *domain_info;
>>       struct cpu_domain_map **cd_map;
>> -    char dname[16], cpumask[256];
>> -    char cpulist[1024];
>> +    char cpulist[MAX_NR_CPUS];
>>       char *line = NULL;
>>       u32 cpu, domain;
>>       u32 dcount = 0;
> 
> Looking at
> https://lore.kernel.org/all/20260119175833.340369-3-swapnil.sapkal@amd.com/
> 
> There was one more "char cpus[1024]"  in tools/perf/util/util.c.
> You may need to fix that too. It is unlikely but, if one has created 
> exclusive
> cpusets comprising of only one cpu from a core, maybe you will run out 
> the length.
> So better use the MAX_NR_CPUS there as well.

Yes, I missed this.

Arnaldo, can you please consider the below diff? Let me know if you'd 
like me to respin the patch.

--
Thanks and Regards,
Swapnil

diff --git a/tools/perf/util/util.c b/tools/perf/util/util.c
index c83e59e8c787..3795d5182ce8 100644
--- a/tools/perf/util/util.c
+++ b/tools/perf/util/util.c
@@ -262,7 +262,7 @@ void cpumask_to_cpulist(char *cpumask, char *cpulist)
         int i, j, bm_size, nbits;
         int len = strlen(cpumask);
         unsigned long *bm;
-       char cpus[1024];
+       char cpus[MAX_NR_CPUS];

         for (i = 0; i < len; i++) {
                 if (cpumask[i] == ',') {
diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
index 394dbfa944ac..c43f17137efd 100644
--- a/tools/perf/util/util.h
+++ b/tools/perf/util/util.h
@@ -6,6 +6,7 @@
  /* glibc 2.20 deprecates _BSD_SOURCE in favour of _DEFAULT_SOURCE */
  #define _DEFAULT_SOURCE 1

+#include "perf.h"
  #include <dirent.h>
  #include <fcntl.h>
  #include <stdbool.h>

Re: [PATCH v1 1/6] perf header: Replace hardcoded max cpus by MAX_NR_CPUS

Posted by Arnaldo Carvalho de Melo 1 week, 4 days ago

On Wed, Jan 28, 2026 at 09:55:05PM +0530, Swapnil Sapkal wrote:
> Hi Srikanth, Arnaldo,
> 
> Thank you for reviewing the patches.
> 
> On 28-01-2026 13:09, Shrikanth Hegde wrote:
> > 
> > 
> > On 1/28/26 12:19 AM, Swapnil Sapkal wrote:
> > > cpumask and cpulist from cpu-domain header have hardcoded max_cpus value
> > > of 1024. Current systems have more cpus than this value. Replace it with
> > > MAX_NR_CPUS. Also define a macro to represent domain name length.
> > > 
> > > Fixes: d40c68a49f69 ("perf header: Support CPU DOMAIN relation info")
> > > Reported-by: Shrikanth Hegde <sshegde@linux.ibm.com>
> > > Signed-off-by: Swapnil Sapkal <swapnil.sapkal@amd.com>
> > > ---
> > >   tools/perf/util/header.c | 5 +++--
> > >   1 file changed, 3 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
> > > index eefd1cd73b6a..31c3bab1b10a 100644
> > > --- a/tools/perf/util/header.c
> > > +++ b/tools/perf/util/header.c
> > > @@ -76,6 +76,7 @@ static const u64 __perf_magic2    =
> > > 0x32454c4946524550ULL;
> > >   static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
> > >   #define PERF_MAGIC    __perf_magic2
> > > +#define DNAME_LEN    16
> > >   const char perf_version_string[] = PERF_VERSION;
> > > @@ -1616,10 +1617,10 @@ static int write_pmu_caps(struct feat_fd *ff,
> > >   struct cpu_domain_map **build_cpu_domain_map(u32
> > > *schedstat_version, u32 *max_sched_domains, u32 nr)
> > >   {
> > > +    char dname[DNAME_LEN], cpumask[MAX_NR_CPUS];
> > >       struct domain_info *domain_info;
> > >       struct cpu_domain_map **cd_map;
> > > -    char dname[16], cpumask[256];
> > > -    char cpulist[1024];
> > > +    char cpulist[MAX_NR_CPUS];
> > >       char *line = NULL;
> > >       u32 cpu, domain;
> > >       u32 dcount = 0;
> > 
> > Looking at
> > https://lore.kernel.org/all/20260119175833.340369-3-swapnil.sapkal@amd.com/
> > 
> > There was one more "char cpus[1024]"  in tools/perf/util/util.c.
> > You may need to fix that too. It is unlikely but, if one has created
> > exclusive
> > cpusets comprising of only one cpu from a core, maybe you will run out
> > the length.
> > So better use the MAX_NR_CPUS there as well.
> 
> Yes, I missed this.
> 
> Arnaldo, can you please consider the below diff? Let me know if you'd like
> me to respin the patch.
> 
> --
> Thanks and Regards,
> Swapnil
> 
> diff --git a/tools/perf/util/util.c b/tools/perf/util/util.c
> index c83e59e8c787..3795d5182ce8 100644
> --- a/tools/perf/util/util.c
> +++ b/tools/perf/util/util.c
> @@ -262,7 +262,7 @@ void cpumask_to_cpulist(char *cpumask, char *cpulist)
>         int i, j, bm_size, nbits;
>         int len = strlen(cpumask);
>         unsigned long *bm;
> -       char cpus[1024];
> +       char cpus[MAX_NR_CPUS];
> 
>         for (i = 0; i < len; i++) {
>                 if (cpumask[i] == ',') {
> diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
> index 394dbfa944ac..c43f17137efd 100644
> --- a/tools/perf/util/util.h
> +++ b/tools/perf/util/util.h
> @@ -6,6 +6,7 @@
>  /* glibc 2.20 deprecates _BSD_SOURCE in favour of _DEFAULT_SOURCE */
>  #define _DEFAULT_SOURCE 1
> 
> +#include "perf.h"

Why add it to util.h? I'll add it to where it is used, util.c, ok?

- Arnaldo

>  #include <dirent.h>
>  #include <fcntl.h>
>  #include <stdbool.h>

Re: [PATCH v1 1/6] perf header: Replace hardcoded max cpus by MAX_NR_CPUS

Posted by Swapnil Sapkal 1 week, 4 days ago

Hello Arnaldo,

On 28-01-2026 23:45, Arnaldo Carvalho de Melo wrote:
> On Wed, Jan 28, 2026 at 09:55:05PM +0530, Swapnil Sapkal wrote:
>> Hi Srikanth, Arnaldo,
>>
>> Thank you for reviewing the patches.
>>
>> On 28-01-2026 13:09, Shrikanth Hegde wrote:
>>>
>>>
>>> On 1/28/26 12:19 AM, Swapnil Sapkal wrote:
>>>> cpumask and cpulist from cpu-domain header have hardcoded max_cpus value
>>>> of 1024. Current systems have more cpus than this value. Replace it with
>>>> MAX_NR_CPUS. Also define a macro to represent domain name length.
>>>>
>>>> Fixes: d40c68a49f69 ("perf header: Support CPU DOMAIN relation info")
>>>> Reported-by: Shrikanth Hegde <sshegde@linux.ibm.com>
>>>> Signed-off-by: Swapnil Sapkal <swapnil.sapkal@amd.com>
>>>> ---
>>>>    tools/perf/util/header.c | 5 +++--
>>>>    1 file changed, 3 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
>>>> index eefd1cd73b6a..31c3bab1b10a 100644
>>>> --- a/tools/perf/util/header.c
>>>> +++ b/tools/perf/util/header.c
>>>> @@ -76,6 +76,7 @@ static const u64 __perf_magic2    =
>>>> 0x32454c4946524550ULL;
>>>>    static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
>>>>    #define PERF_MAGIC    __perf_magic2
>>>> +#define DNAME_LEN    16
>>>>    const char perf_version_string[] = PERF_VERSION;
>>>> @@ -1616,10 +1617,10 @@ static int write_pmu_caps(struct feat_fd *ff,
>>>>    struct cpu_domain_map **build_cpu_domain_map(u32
>>>> *schedstat_version, u32 *max_sched_domains, u32 nr)
>>>>    {
>>>> +    char dname[DNAME_LEN], cpumask[MAX_NR_CPUS];
>>>>        struct domain_info *domain_info;
>>>>        struct cpu_domain_map **cd_map;
>>>> -    char dname[16], cpumask[256];
>>>> -    char cpulist[1024];
>>>> +    char cpulist[MAX_NR_CPUS];
>>>>        char *line = NULL;
>>>>        u32 cpu, domain;
>>>>        u32 dcount = 0;
>>>
>>> Looking at
>>> https://lore.kernel.org/all/20260119175833.340369-3-swapnil.sapkal@amd.com/
>>>
>>> There was one more "char cpus[1024]"  in tools/perf/util/util.c.
>>> You may need to fix that too. It is unlikely but, if one has created
>>> exclusive
>>> cpusets comprising of only one cpu from a core, maybe you will run out
>>> the length.
>>> So better use the MAX_NR_CPUS there as well.
>>
>> Yes, I missed this.
>>
>> Arnaldo, can you please consider the below diff? Let me know if you'd like
>> me to respin the patch.
>>
>> --
>> Thanks and Regards,
>> Swapnil
>>
>> diff --git a/tools/perf/util/util.c b/tools/perf/util/util.c
>> index c83e59e8c787..3795d5182ce8 100644
>> --- a/tools/perf/util/util.c
>> +++ b/tools/perf/util/util.c
>> @@ -262,7 +262,7 @@ void cpumask_to_cpulist(char *cpumask, char *cpulist)
>>          int i, j, bm_size, nbits;
>>          int len = strlen(cpumask);
>>          unsigned long *bm;
>> -       char cpus[1024];
>> +       char cpus[MAX_NR_CPUS];
>>
>>          for (i = 0; i < len; i++) {
>>                  if (cpumask[i] == ',') {
>> diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
>> index 394dbfa944ac..c43f17137efd 100644
>> --- a/tools/perf/util/util.h
>> +++ b/tools/perf/util/util.h
>> @@ -6,6 +6,7 @@
>>   /* glibc 2.20 deprecates _BSD_SOURCE in favour of _DEFAULT_SOURCE */
>>   #define _DEFAULT_SOURCE 1
>>
>> +#include "perf.h"
> 
> Why add it to util.h? I'll add it to where it is used, util.c, ok?
> 

This doesn't need to be in util.h. Please move it to util.c

--
Thanks and Regards,
Swapnil

> - Arnaldo
> 
>>   #include <dirent.h>
>>   #include <fcntl.h>
>>   #include <stdbool.h>

Re: [PATCH v1 1/6] perf header: Replace hardcoded max cpus by MAX_NR_CPUS

Posted by Arnaldo Carvalho de Melo 1 week, 4 days ago

On Wed, Jan 28, 2026 at 09:55:05PM +0530, Swapnil Sapkal wrote:
> Hi Srikanth, Arnaldo,
> 
> Thank you for reviewing the patches.
> 
> On 28-01-2026 13:09, Shrikanth Hegde wrote:
> > 
> > 
> > On 1/28/26 12:19 AM, Swapnil Sapkal wrote:
> > > cpumask and cpulist from cpu-domain header have hardcoded max_cpus value
> > > of 1024. Current systems have more cpus than this value. Replace it with
> > > MAX_NR_CPUS. Also define a macro to represent domain name length.
> > > 
> > > Fixes: d40c68a49f69 ("perf header: Support CPU DOMAIN relation info")
> > > Reported-by: Shrikanth Hegde <sshegde@linux.ibm.com>
> > > Signed-off-by: Swapnil Sapkal <swapnil.sapkal@amd.com>
> > > ---
> > >   tools/perf/util/header.c | 5 +++--
> > >   1 file changed, 3 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
> > > index eefd1cd73b6a..31c3bab1b10a 100644
> > > --- a/tools/perf/util/header.c
> > > +++ b/tools/perf/util/header.c
> > > @@ -76,6 +76,7 @@ static const u64 __perf_magic2    =
> > > 0x32454c4946524550ULL;
> > >   static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
> > >   #define PERF_MAGIC    __perf_magic2
> > > +#define DNAME_LEN    16
> > >   const char perf_version_string[] = PERF_VERSION;
> > > @@ -1616,10 +1617,10 @@ static int write_pmu_caps(struct feat_fd *ff,
> > >   struct cpu_domain_map **build_cpu_domain_map(u32
> > > *schedstat_version, u32 *max_sched_domains, u32 nr)
> > >   {
> > > +    char dname[DNAME_LEN], cpumask[MAX_NR_CPUS];
> > >       struct domain_info *domain_info;
> > >       struct cpu_domain_map **cd_map;
> > > -    char dname[16], cpumask[256];
> > > -    char cpulist[1024];
> > > +    char cpulist[MAX_NR_CPUS];
> > >       char *line = NULL;
> > >       u32 cpu, domain;
> > >       u32 dcount = 0;
> > 
> > Looking at
> > https://lore.kernel.org/all/20260119175833.340369-3-swapnil.sapkal@amd.com/
> > 
> > There was one more "char cpus[1024]"  in tools/perf/util/util.c.
> > You may need to fix that too. It is unlikely but, if one has created
> > exclusive
> > cpusets comprising of only one cpu from a core, maybe you will run out
> > the length.
> > So better use the MAX_NR_CPUS there as well.
> 
> Yes, I missed this.
> 
> Arnaldo, can you please consider the below diff? Let me know if you'd like
> me to respin the patch.

Sure, will add it.

- Arnaldo
 
> --
> Thanks and Regards,
> Swapnil
> 
> diff --git a/tools/perf/util/util.c b/tools/perf/util/util.c
> index c83e59e8c787..3795d5182ce8 100644
> --- a/tools/perf/util/util.c
> +++ b/tools/perf/util/util.c
> @@ -262,7 +262,7 @@ void cpumask_to_cpulist(char *cpumask, char *cpulist)
>         int i, j, bm_size, nbits;
>         int len = strlen(cpumask);
>         unsigned long *bm;
> -       char cpus[1024];
> +       char cpus[MAX_NR_CPUS];
> 
>         for (i = 0; i < len; i++) {
>                 if (cpumask[i] == ',') {
> diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
> index 394dbfa944ac..c43f17137efd 100644
> --- a/tools/perf/util/util.h
> +++ b/tools/perf/util/util.h
> @@ -6,6 +6,7 @@
>  /* glibc 2.20 deprecates _BSD_SOURCE in favour of _DEFAULT_SOURCE */
>  #define _DEFAULT_SOURCE 1
> 
> +#include "perf.h"
>  #include <dirent.h>
>  #include <fcntl.h>
>  #include <stdbool.h>