Re: x86: runtime waring in pcibios_fwaddrmap_lookup

[Myron Stowe <mstowe@redhat.com>]

On Sat, 2012-04-28 at 08:36 +0300, Meelis Roos wrote:
> > So, somehow, your system hit the assertion (WARN_ON) in
> > pcibios_fwaddrmap_lookup().  From the dmesg log you provided I can see
> > why pcibios_save_fw_addr() was called - there is a resource collision
> > with the video device's GART - but what I don't understand is how the
> > assertion is being triggered.
> > 
> > There are only two callers of pcibios_fwaddrmap_lookup()
> > (see ./arch/x86/pci/i386.c:: pcibios_save_fw_addr(), and
> > pcibios_retrieve_fw_addr()) and both of those are acquiring the lock
> > before making the call - so triggering the assertion just doesn't seem
> > possible but obviously it is getting triggered.
> > 
> > I don't have any ideas currently but do have some questions that
> > hopefully will start to enlighten us.
> > 
> >         Does your system encounter this consistently upon every boot or
> >         is it hit intermittently?
> 
> Every time.
> 
> >         Do you know of any kernel versions that do not encounter it?
> 
> Checked my logs, no kernel up to 3.3.0 included gave it, next tested 
> kernel was 3.4.0-rc2 and that had the warning and every kernel since has 
> had it.
> 
> >         Are you running a para-virtualized kernel?
> 
> No.

Thanks for the quick response Meelis (unlike myself; I've been consumed
with a hard deadline task recently - sorry).

I got together with Bjorn and after showing him the paths, and locking,
he also did not see how we could be hitting the WARN_ON.  I would like
to take advantage of your system since it encounters this consistently
but really have no idea how to progress currently.  I'll query a couple
of my colleagues for suggestions.  In the meantime, if anyone else -
Jesse - has any ideas, please speak up.

As for you noticing this after 3.3.0 that makes sense as the WARN_ON was
added with 3.3.0.

Myron