Memory increase using glibc - 2.35

Memory increase using glibc - 2.35

[Ronny Meeus]

Hello

we recently did an upstep to glibc-2.35 (from glibc-2.34) and we
observed a substantial memory increase in 1 application.
Our embedded SW is running on Linux 4.1.52 on ARM CPUs and consists of
many applications.

All apps, except 1, are consuming more or less the same amount of memory.
Note that these apps are executing a malloc_trim periodically (no clue
whether that has something to do with it).
For the problematic application (which is not executing malloc_trim),
the memory consumption increased from 250Mb to 300Mb, with the same
configuration.

We did already a "pmap" of the both apps and the observation is that
there are a lot more (and bigger in size) anon pages allocated when
running 2.35.
The complete increase of 50Mb can be traced back to the increase of these pages.

Can this be explained by one of the changes in 2.35?
We are not using any tuneables.

Thanks.

Best regards,
Ronny

Re: Memory increase using glibc - 2.35

[Szabolcs Nagy]

The 02/07/2023 09:21, Ronny Meeus via Libc-help wrote:
> we recently did an upstep to glibc-2.35 (from glibc-2.34) and we
> observed a substantial memory increase in 1 application.
> Our embedded SW is running on Linux 4.1.52 on ARM CPUs and consists of
> many applications.
> 
> All apps, except 1, are consuming more or less the same amount of memory.
> Note that these apps are executing a malloc_trim periodically (no clue
> whether that has something to do with it).
> For the problematic application (which is not executing malloc_trim),
> the memory consumption increased from 250Mb to 300Mb, with the same
> configuration.
> 
> We did already a "pmap" of the both apps and the observation is that
> there are a lot more (and bigger in size) anon pages allocated when
> running 2.35.
> The complete increase of 50Mb can be traced back to the increase of these pages.
> 
> Can this be explained by one of the changes in 2.35?
> We are not using any tuneables.

the 2.35 development window was between 2021-08-01 and 2022-02-01.
there were malloc changes, maybe the huge page support had some sideeffect:
https://sourceware.org/git/?p=glibc.git;a=history;f=malloc;hb=refs/tags/glibc-2.35

you may have to bisect this.