Two Resource Leaks
joel at rtems.org
Sat Mar 27 13:53:11 UTC 2021
On Sat, Mar 27, 2021, 12:04 AM Richi Dubey <richidubey at gmail.com> wrote:
> I believe that cause of the recent commits in the last two months, these
> are the tests that have started failing on master:
> and a timeout:
The password ones are for an incorrect change to pwdgrp.c to assert if
mkdir fails creating /etc. That should account for pwdgrp, paxpasswd02, and
shell01. This code needs to turn into a (void) rather than an assert.
And looking at git makes me realise that I didn't push the fix for that. It
is in my tree. Sorry.
The issue I found is different and won't happen on every target or bsp.
Psim has 6-9 failures even after freeing the right stack address.
> On Sat, Mar 27, 2021 at 3:21 AM Joel Sherrill <joel at rtems.org> wrote:
>> Jennifer has been working on a network driver and had some odd failures
>> in libbsd. I suggested turning on rtems debug and that caused a number of
>> libbsd tests to fail. She pointed me in the right direction and I found
>> that the following patch resulted in the stack address being freed
>> including an "align up" adjustment in some cases. This looks to be from
>> something Sebastian committed early this month.
>> I am not sure how that wasn't noticed since about 40 tests were failing
>> on psim due to that.
>> I have attached a straightforward patch to address this issue.
>> Unfortunately, even with this patch and using the RTEMS hash just before
>> this patch program01 and syscalls01 in libbsd fail. I debugged into
>> syscalls01 enough to find that there are 7 blocks at the beginning of one
>> of the tests and 5 after. There is another leak and I tried using the has
>> before Sebastian's change above but it is still leaking.
>> On top of that, psxconfig01 and spconfig01 are failing on psim which
>> appears to be independent. I am not sure what these are but it is something
>> about minimum stack size not matching. Since I was looking for stack memory
>> issues, I started to investigate these but decided they were not
>> allocation/free issues.
>> Help really appreciated in addressing these leaks.
>> devel mailing list
>> devel at rtems.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the devel