This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] sysconf: Use conservative default for _SC_NPROCESSORS_ONLN [BZ #21542]


On Fri, Jun 9, 2017 at 5:22 AM, Florian Weimer <fweimer@redhat.com> wrote:
> On 06/09/2017 02:17 PM, Zack Weinberg wrote:
>> On Fri, Jun 9, 2017 at 8:11 AM, Florian Weimer <fweimer@redhat.com> wrote:
>>> On 06/02/2017 04:21 PM, Florian Weimer wrote:
>>>> 2017-06-02  Florian Weimer  <fweimer@redhat.com>
>>>>
>>>>       [BZ #21542]
>>>>       * sysdeps/unix/sysv/linux/getsysstats.c (__get_nprocs): Default to
>>>>       two if no accurate information is available.
>>>>       * posix/tst-sysconf-empty-chroot.c: New file.
>>>>       * posix/Makefile (tests): Add it.
>>>
>>> Ping?
>>>
>>>   <https://sourceware.org/ml/libc-alpha/2017-06/msg00101.html>
>>
>> It seems reasonable to me but I'm not sure I understand all of the implications.
>
> The implication is that the Hotspot JVM no longer crashes in a chroot on
> an SMP host because it assumes a UP host and elides atomics.
>
> The downside is that some other application could choose to use
> spinlocks on a UP system because glibc suggests it's SMP, which could
> result in terrible performance.
>
> All this is just for the corner case of an improperly setup
> chroot/contain (lacking key parts of /sys and /proc).
>

On x86, we compute the number of threads which share L3
cache in a package via CPUID.  We should use this info as
default on x86.

-- 
H.J.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]