On 24.07.2012 21:13, Steven Bosscher wrote:
AFAIR the qsort is just for getting a stable ordering over two pools to find
the leaked regsets afterwards, the type of ordering doesn't matter.
What matters is that the compare function gives a reliable result. You
can't subtract pointers like that for qsort.
After consulting the experts on IRC, I'm going with a fix along the
lines of "return (x == y ? 0 : (x < y ? -1 : 1));".
Yeah, I agree the code was dubious, so thanks for uncovering this. If you
have already tested the patch, consider it preapproved, otherwise I can fix
it on this week.
Anyways,
how come this is related to your patch? We don't use statistics in
sel-sched... Something got miscompiled?
No, just allocated slightly differently. A bitmap_head is one pointer
bigger than before. I'm unsure how that causes this problem, though. I
suspect you would have seen the same failure with GATHER_STATISTICS
enabled.
Still interesting to know why your first patch fixed the issue in the first
place. I will try taking a look at the generated code in my spare time.
Yours,
Andrey
Ciao!
Steven
