cvs commit: src/sys/dev/mpt mpt_pci.c

[John Baldwin]

On Wednesday 15 November 2006 22:30, mjacob at freebsd.org wrote:
> 
> MSI/MSI-X seemed work with MPT for SAS, 4Gb FibreChannel on one system 
> (SuperMicro Xeon) and 2Gb FibreChannel on another (4xdual core Opteron), 
> but failed for onboard MPT on that system (dual core Opteron)
> 
> mpt3: request 0xffffffff80e75050:96 timed out for ccb 0xffffff01433ec400 
> (req->ccb 0xffffff01433ec400)
> mpt3: completing timedout/aborted req 0xffffffff80e75050:96
> mpt3: Timedout requests already complete. Interrupts may not be 
> functioning.
> mpt2: request 0xffffffff80e67050:96 timed out for ccb 0xffffff0141c75c00 
> (req->ccb 0xffffff0141c75c00)
> mpt2: completing timedout/aborted req 0xffffffff80e67050:96
> mpt2: Timedout requests already complete. Interrupts may not be 
> functioning.
> 
> 
> MSI/MSI-X failed to work for the onboard em on the SuperMicro system.
> 
> Given the (unselective) failure rate for me and the relatively small 
> sample, but with two pretty widely used system types I'd like to suggest 
> that you reconsider making the MSI and MSI-X being 'on' as the default.

It is current, so I think having it on is ok.  When I put it in 6.x it will be 
off by default (eventually the defualt in 6.x may change after broader 
testing).  If you have chipsets that don't work, we can certainly blacklist 
those.  However, we aren't likely to find them if no one tries MSI on them.

Also, in your case it looks like you might have an HT bridge that needs the 
MSI window turned on.  I haven't committed that patch as I haven't been able 
to get it tested yet.  Please try 
http://www.freebsd.org/~jhb/patches/msi_ht.patch

-- 
John Baldwin