[ClusterLabs] Antw: Re: lrmd segfault

[Jan Pokorný]

On 31/01/17 15:04 +0100, Jan Pokorný wrote:
> On 31/01/17 10:16 +0100, Ulrich Windl wrote:
>>>> Kristoffer Grönlund <deceiver.g at gmail.com> schrieb am 31.01.2017 um 07:34
>>>> in Nachricht <87mve768lx.fsf at suse.com>:
>> 
>> [...]
>>> Just from looking at the core dump, it looks like your processor doesn't
>>> support the SSE extensions used by the newer version of the code. You'll
>>> need to recompile and disable use of those extensions.
>>> 
>>> It looks like the code is using SSE 4.2, which is relatively new:
>>> 
>>> https://en.wikipedia.org/wiki/SSE4#SSE4.2 
>> 
>> I wonder: Shouldn't this be a part of RPM's preinstallation checks? Despite of
>> that I'd expect a SIGILL on illegal instructions. In my personal experience
>> errors in any string fuction are typically the consequence of some NULL ponter
>> or unterminated C string...
> 
> My guess is that it has nothing to do with supported instruction set.

it = segfault, and the sentence should end with "per se" for more
clarity.

Anyway, there were some use-after-free issues related to inflight ops
so it's likely the root cause, perhaps fixed ever since.

> ISTR glibc chooses "fastest optimized" version in runtime according to
> what the machine supports.  And "optimized" versions are more prone to
> trigger the undefined behavior that may be otherwise tolerated or
> deterministic by less optimized siblings.
> 
> RPM is not an ABI checker or machine eligibility enforcing tool,
> at least not at instruction set granularity.

-- 
Jan (Poki)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.clusterlabs.org/pipermail/users/attachments/20170131/abf55074/attachment-0002.sig>