Bug fixes MFCd to the release branch need some testing.

YONETANI Tomokazu qhwt+dfly at les.ath.cx
Sun Sep 18 07:43:00 PDT 2005


On Thu, Sep 15, 2005 at 12:16:26PM -0700, Matthew Dillon wrote:
>     Before I bump the sub-version and slip the release tag it would
>     be nice if someone running RELEASE on an SMP box could test 
>     "DragonFly_RELEASE_1_2" (not the official DragonFly_RELEASE_1_2_Slip tag).
> 
>     I don't happen to have an SMP box running release at the moment, only
>     a UP box, so I can't test the token fixes.
> 
>     Basically all you need to do to test is to run a buildworld -j 8 in a 
>     loop for a day or so, to make sure I haven't done anything aggregious
>     in MFCing the recent bug fixes.

I've been running this test on an SMP machine.  So far I haven't seen a
problem after finishing 27-th iteration, but I saw a lot of warnings in
the dmesg:
Sep 18 11:40:17 X345 kernel: Diagnostic: attempt to switch while mastering cpusync: 0xc030fe08

The hexadecimal number is consistent between warnings.

And although very occasionally, I saw the following message in dmesg
while running blogbench storm on the same machine(but I think I saw
this one on HEAD either):
Sep 18 12:16:05 X345 kernel: conflict with vnode 0xe9b4e898 ncp e_log.So
Sep 18 12:16:05 X345 kernel: Warning: inode free race avoided 1 times





More information about the Kernel mailing list