pci error recovery procedure

Zhang, Yanmin yanmin_zhang at linux.intel.com
Wed Sep 6 11:35:21 EST 2006


On Wed, 2006-09-06 at 02:50, Linas Vepstas wrote:
> On Mon, Sep 04, 2006 at 07:03:12PM +1000, Benjamin Herrenschmidt wrote:
> > 
> > > As you know, all functions of a device share the same bus number and 5 bit dev number.
> > > They just have different 3 bit function number. We could deduce if functions are in the same
> > > device (slot).
> > 
> > Until you have a P2P bridge ...
> 
> And this is not theoretical: for example, the matrox graphics cards:
> 
> 0000:c8:01.0 PCI bridge: Hint Corp HB6 Universal PCI-PCI bridge (non-transparent mode) (rev 13)
> 0000:c9:00.0 VGA compatible controller: Matrox Graphics, Inc. MGA G400 AGP (rev 85)
> 
> Now, I could have sworn there was another device behind this bridge, 
> some serial or joystick controller or something, although this
> particular card doesn't seem to have it.
Thanks. My comments above in this email is just to try to find a method
to judge if 2 or more functions belongs to the same device. If it's
not right, it still doesn't hurt the new API pci_error_handlers.

> 
> ------
> It's not clear to me what hardware may show up in the future. 
> For example, someone may build a 32x PCI-E card that will act 
> as a bridge to a drawer with half-a-dozen ordinary PCI-X slots 
> in it. This is perhaps a bit hypothetical, but changing the API 
> will make it harder to implement eror recovery for such a system.
I agree that it's difficult to predict the future. At least the new API
could process the example.

> FWIW, there is at least one pSeries system in the lab which has
> several hundred PCI slots attached to it, although I've never 
> done testing on it. Hmm. Maybe its time I did ... 
> 
> --linas



More information about the Linuxppc-dev mailing list