DragonFly BSD
DragonFly commits List (threaded) for 2010-02
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: git: kernel - Fix some rare pmap races in i386 and x86_64.


From: Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx>
Date: Mon, 22 Feb 2010 11:32:40 -0800 (PST)

:Err, this warning is totally meaningless to an administrator or pretty
:much anybody who didn't read this patch. Wouldn't something like
:
:kprintf("Diagnostic: pmap_interlock %08x. Please report this!\n",
:pmap->pm_active)
:
:or even
:
:kprintf("Wow. pmap_interlock %08x. Please report this!\n", pmap->pm_active)
:
:be more useful to both sides?
:
:Aggelos

    It's really just something for my own debugging, to make sure that
    the case actually occurs.  I suspect I'll remove it by the release
    but if you want to change the message please be my guest :-)

    --

    I can produce the warning when I start blogbench on x86_64.
    Interestingly enough it only happens after a fresh machine start,
    presumably due to the paging in of program binary pages that are not
    primed not in the cache yet.  Every subsequent run of blogbench failed
    to produce any new warnings.

    I have hopes that either this or the missing critical sections patched
    at the same time might also be the cause of the rare seg-faults seen
    on x86_64 on things like cc1 (not stuff like conftest from pkgsrc),
    which again only appear to occur soon after the machine has booted.
    It's really hard to test the x86_64 issue because I have to do
    a reboot, then make -j 4 buildworld, then reboot, etc.

					-Matt
					Matthew Dillon 
					<dillon@backplane.com>



[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]