Re: userspace pci config space accesses

[Brian King <brking@us.ibm.com>]

Greg KH wrote:
> On Wed, Apr 28, 2004 at 04:49:02PM -0500, Brian King wrote:
> 
>>I recently ran into a problem where lspci was trying to read pci config 
>>space
>>of a pci adapter while the device driver for that adapter was running BIST
>>on it. On ppc64, this resulted in a PCI error and puts the slot into an 
>>error state making it unusable for the remainder of that system boot.
>>Should there be some blocking in place so that userspace pci config
>>reads will not occur in these windows or is using tools like lspci
>>user beware?
> 
> 
> There already is a pci_config_lock that should be grabbed when accessing
> pci config space.  It sounds like the driver needs to play a bit nicer
> when it's running a self test :)

Found the lock. Unfortunately, its not exported, so a device driver can't use
it without changing that. Additionally, its a spinlock, and it takes 2 seconds
to complete BIST, which seems a bit too long to hold a spinlock.

> What driver is doing this?

The ipr driver, a scsi device driver for ppc64.

http://marc.theaimsgroup.com/?l=linux-scsi&m=108144942527994&w=2

The driver runs BIST at device initialization time to ensure that the device
is in a clean state. It will also run BIST on module unload, and in various
error scenarios.


-- 
Brian King
eServer Storage I/O
IBM Linux Technology Center