calloc question about clearing allocated memory

calloc question about clearing allocated memory

[Vicente Bergas]

Hello,
when compiling a program against the musl libc and running it under
valgrind, then it reports lots of uninitialized value errors.
I have tracked them down to only two sources:
http://git.musl-libc.org/cgit/musl/tree/src/string/strlen.c?id=e6def544#n15
http://git.musl-libc.org/cgit/musl/tree/src/malloc/malloc.c?id=e6def544#n375

The first one about strlen is clearly an optimization everybody would
like to have, but the second one in __malloc0 is not that obvious.
Please, could that be explained: why is *z read just before being
cleared?
In fact it is counter-intuitive and looks like an overhead.
In case there is a reason for it, it would be nice to put an
explanatory comment there. Otherwise, that extra read could be
removed to make valgrind happy.

Regards,
  Vicenç.

P.S.: Please, CC me as I am not subscribed to the list.

Re: calloc question about clearing allocated memory

[Szabolcs Nagy]

* Vicente Bergas <vicencb@gmail.com> [2017-06-26 23:30:53 +0200]:
> Hello,
> when compiling a program against the musl libc and running it under
> valgrind, then it reports lots of uninitialized value errors.
> I have tracked them down to only two sources:
> http://git.musl-libc.org/cgit/musl/tree/src/string/strlen.c?id=e6def544#n15
> http://git.musl-libc.org/cgit/musl/tree/src/malloc/malloc.c?id=e6def544#n375
> 
> The first one about strlen is clearly an optimization everybody would
> like to have, but the second one in __malloc0 is not that obvious.
> Please, could that be explained: why is *z read just before being
> cleared?
> In fact it is counter-intuitive and looks like an overhead.
> In case there is a reason for it, it would be nice to put an
> explanatory comment there. Otherwise, that extra read could be
> removed to make valgrind happy.

this should be the faq of the month..

a zero page does not take up resources, however when
it is written (even if that's just 0) the page gets
dirty and it does take up resources (and the write
operation would cause a page fault introducing
significant delays)

so if(*z) *z=0; should be an obvious optimization.
that said the current code is not optimal and an
improvement was just posted:
http://www.openwall.com/lists/musl/2017/06/26/1
but that won't fix the valgrind issue: to fix that
valgrind just needs to be taught not to report
uninitialized error for calloced memory.

> 
> Regards,
>   Vicenç.
> 
> P.S.: Please, CC me as I am not subscribed to the list.

Re: calloc question about clearing allocated memory

[Vicente Bergas]

Thanks for the explanation, now it is clear!

On Tue, Jun 27, 2017 at 3:13 PM, Szabolcs Nagy <nsz@port70.net> wrote:
> * Vicente Bergas <vicencb@gmail.com> [2017-06-26 23:30:53 +0200]:
>> Hello,
>> when compiling a program against the musl libc and running it under
>> valgrind, then it reports lots of uninitialized value errors.
>> I have tracked them down to only two sources:
>> http://git.musl-libc.org/cgit/musl/tree/src/string/strlen.c?id=e6def544#n15
>> http://git.musl-libc.org/cgit/musl/tree/src/malloc/malloc.c?id=e6def544#n375
>>
>> The first one about strlen is clearly an optimization everybody would
>> like to have, but the second one in __malloc0 is not that obvious.
>> Please, could that be explained: why is *z read just before being
>> cleared?
>> In fact it is counter-intuitive and looks like an overhead.
>> In case there is a reason for it, it would be nice to put an
>> explanatory comment there. Otherwise, that extra read could be
>> removed to make valgrind happy.
>
> this should be the faq of the month..
>
> a zero page does not take up resources, however when
> it is written (even if that's just 0) the page gets
> dirty and it does take up resources (and the write
> operation would cause a page fault introducing
> significant delays)
>
> so if(*z) *z=0; should be an obvious optimization.
> that said the current code is not optimal and an
> improvement was just posted:
> http://www.openwall.com/lists/musl/2017/06/26/1
> but that won't fix the valgrind issue: to fix that
> valgrind just needs to be taught not to report
> uninitialized error for calloced memory.
>
>>
>> Regards,
>>   Vicenç.
>>
>> P.S.: Please, CC me as I am not subscribed to the list.