Re: [PATCH v1 0/3] perf record: adapt NUMA awareness to machines with #CPUs > 1K

[Alexey Budankov]

On 22.11.2019 16:26, Jiri Olsa wrote:
> On Wed, Nov 20, 2019 at 12:33:10PM +0300, Alexey Budankov wrote:
>>
>> Current implementation of cpu_set_t type by glibc has internal cpu
>> mask size limitation of no more than 1024 CPUs. This limitation confines
>> NUMA awareness of Perf tool in record mode, thru --affinity option,
>> to the first 1024 CPUs on machines with larger amount of CPUs.
>>
>> This patch set enables Perf tool to overcome 1024 CPUs limitation by
>> using a dedicated struct mmap_cpu_mask type and applying tool's bitmap
>> API operations to manipulate affinity masks of the tool's thread and
>> the mmaped data buffers.
>>
>> tools bitmap API has been extended with bitmap_equal() operation
>> and its implementation is derived from the kernel one.
>>
>> ---
>> Alexey Budankov (3):
>>   tools bitmap: extend bitmap API with bitmap_equal()
>>   perf mmap: declare type for cpu mask of arbitrary length
>>   perf record: adapt affinity to machines with #CPUs > 1K
> 
> looks good to me, I sent some minor comments
> 
>>
>>  tools/include/linux/bitmap.h | 21 +++++++++++++++++++++
>>  tools/lib/bitmap.c           | 15 +++++++++++++++
>>  tools/perf/builtin-record.c  | 28 ++++++++++++++++++++++------
>>  tools/perf/util/mmap.c       | 28 ++++++++++++++++++++++------
>>  tools/perf/util/mmap.h       | 11 ++++++++++-
>>  5 files changed, 90 insertions(+), 13 deletions(-)
>>
>> ---
>> Testing:
>>
>>   $ tools/perf/perf record -v --affinity=cpu -- ls
>>   thread mask[8]: empty
>>   Using CPUID GenuineIntel-6-5E-3
>>   ...
>>   mmap size 528384B
>>   0x7f95f8f85010: mmap mask[8]: 0
>>   0x7f95f8f950d8: mmap mask[8]: 1
>>   0x7f95f8fa51a0: mmap mask[8]: 2
>>   0x7f95f8fb5268: mmap mask[8]: 3
>>   0x7f95f8fc5330: mmap mask[8]: 4
>>   0x7f95f8fd53f8: mmap mask[8]: 5
>>   0x7f95f8fe54c0: mmap mask[8]: 6
>>   0x7f95f8ff5588: mmap mask[8]: 7
> 
> could we add this to -vv? -v is poluted already

In v2.

Thanks,
Alexey

> 
> perhaps we should make some effort and try to consolidate -v output
> for some really basic verbose, the rest would be under -vv or specialized
> --debug variable .. not in scope of this patchset of course ;-)
> 
> thanks,
> jirka
> 
>