* [BUG?] memory allocation profiling: the counters are not non-decreasing.
@ 2024-06-05 12:37 David Wang
2024-06-05 20:00 ` Suren Baghdasaryan
0 siblings, 1 reply; 3+ messages in thread
From: David Wang @ 2024-06-05 12:37 UTC (permalink / raw)
To: kent.overstreet, surenb, keescook; +Cc: linux-kernel
Hi,
I was playing with this new feature of memory allocation profiling, when I plot the data, grouped by caller's source dir,
I noticed that the counters are not non-decreasing.
If I understand this feature correctly, all counters should be non-decreasing, right?
Or is there some TTL for stale items?
Following is a simpler script demonstrating my sayings
$ for i in {1..100}; do sudo cat /proc/allocinfo | grep " mm\/" | awk '{s+=$2}END{print s}'; sleep 10; done
1194453
1196935
1198957
1199716
1200770
1190795 <--- decrease here
1192702
1193070
1193812
1195069
1196491
...
FYI
David
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [BUG?] memory allocation profiling: the counters are not non-decreasing.
2024-06-05 12:37 [BUG?] memory allocation profiling: the counters are not non-decreasing David Wang
@ 2024-06-05 20:00 ` Suren Baghdasaryan
2024-06-06 3:42 ` David Wang
0 siblings, 1 reply; 3+ messages in thread
From: Suren Baghdasaryan @ 2024-06-05 20:00 UTC (permalink / raw)
To: David Wang; +Cc: kent.overstreet, keescook, linux-kernel
On Wed, Jun 5, 2024 at 5:37 AM David Wang <00107082@163.com> wrote:
>
> Hi,
> I was playing with this new feature of memory allocation profiling, when I plot the data, grouped by caller's source dir,
> I noticed that the counters are not non-decreasing.
> If I understand this feature correctly, all counters should be non-decreasing, right?
> Or is there some TTL for stale items?
Hi David,
Counters can decrease once an allocation is freed.
Thanks,
Suren.
>
> Following is a simpler script demonstrating my sayings
>
> $ for i in {1..100}; do sudo cat /proc/allocinfo | grep " mm\/" | awk '{s+=$2}END{print s}'; sleep 10; done
> 1194453
> 1196935
> 1198957
> 1199716
> 1200770
> 1190795 <--- decrease here
> 1192702
> 1193070
> 1193812
> 1195069
> 1196491
> ...
>
> FYI
> David
>
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [BUG?] memory allocation profiling: the counters are not non-decreasing.
2024-06-05 20:00 ` Suren Baghdasaryan
@ 2024-06-06 3:42 ` David Wang
0 siblings, 0 replies; 3+ messages in thread
From: David Wang @ 2024-06-06 3:42 UTC (permalink / raw)
To: Suren Baghdasaryan; +Cc: kent.overstreet, keescook, linux-kernel
At 2024-06-06 04:00:17, "Suren Baghdasaryan" <surenb@google.com> wrote:
>On Wed, Jun 5, 2024 at 5:37 AM David Wang <00107082@163.com> wrote:
>>
>> Hi,
>> I was playing with this new feature of memory allocation profiling, when I plot the data, grouped by caller's source dir,
>> I noticed that the counters are not non-decreasing.
>> If I understand this feature correctly, all counters should be non-decreasing, right?
>> Or is there some TTL for stale items?
>
>Hi David,
>Counters can decrease once an allocation is freed.
>Thanks,
>Suren.
>
Hi Suren,
Thanks for the quick response.
I guess I misunderstood this feature, no wonder I saw several "zero" values.
If "freed" memory is also tracked accurately, I feel this would be a very powerful feature!
Thx~
David
>>
>> Following is a simpler script demonstrating my sayings
>>
>> $ for i in {1..100}; do sudo cat /proc/allocinfo | grep " mm\/" | awk '{s+=$2}END{print s}'; sleep 10; done
>> 1194453
>> 1196935
>> 1198957
>> 1199716
>> 1200770
>> 1190795 <--- decrease here
>> 1192702
>> 1193070
>> 1193812
>> 1195069
>> 1196491
>> ...
>>
>> FYI
>> David
>>
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2024-06-06 3:42 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2024-06-05 12:37 [BUG?] memory allocation profiling: the counters are not non-decreasing David Wang
2024-06-05 20:00 ` Suren Baghdasaryan
2024-06-06 3:42 ` David Wang
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox