threads/119920: fork broken in libpthread
Julian Elischer
julian at elischer.org
Thu Jan 31 11:31:47 PST 2008
Landon Fuller wrote:
>
> On Jan 31, 2008, at 11:25, Julian Elischer wrote:
>
>> Landon Fuller wrote:
>>> On Jan 31, 2008, at 10:44, Julian Elischer wrote:
>>>> the releng 6 and releng 7 versions of this patch are at:
>>>>
>>>> http://www.freebsd.org/~julian/kse-6.diff
>>>>
>>>> and
>>>>
>>>> http://www.freebsd.org/~julian/kse-7.diff
>>>>
>>>>
>>>> but the one for 7 is just speculative.
>>>>
>>>> I don't have a 7.x system to test.
>>>>
>>>>
>>>> Landon, what is your test case?
>>>>
>>> Inlined below. On our 8-core 6.3 system without this patch, the
>>> fork() threads wedge and spin in <2 minutes. With the patch, works as
>>> expected.
>>
>> when you say "wedge and spin" what do you see in top?
>>
>> which is it , wedge, or spin?
>
> The children processes spin (in either sigprocmask() or
> _lockuser_reinit()), and the parent process thread wedges on waitpid().
so if the children show up in top at all (usually they are already
exited by the time top looks) then the problem has been hit. right?
More information about the freebsd-threads
mailing list