On Thu, Mar 27, 2008 at 10:53:43AM -0700, Brandeburg, Jesse wrote:
quoted text
> Kok, Auke wrote:
> > Jeff Garzik wrote:
> >> Auke Kok wrote:
> >>> From: Jesse Brandeburg <jesse.brandeburg@intel.com>
> >>> 
> >>> Test the MSI interrupt physically once before assuming that it
> >>> actually works. Several platforms have already come across that
> >>> have non-functional MSI interrupts and this code will attempt
> >>> to detect those safely. Once the test succeeds MSI interrupts
> >>> will be enabled.
> >>> 
> >>> Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
> >>> Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
> 
> >> Ah, the perennial add-same-test-to-every-driver conundrum.
> >> 
> >> I think we are far enough along with MSI to _not_ do this anymore in
> >> drivers. 
> 
> Actually, I'm hoping you'll allow this Jeff, we have a production system
> (see below) we know about that doesn't like the way 82571 formats MSI
> interrupt messages.  All other systems seem to be okay with this format
> of MSI messages, but this system implemented a stricter interpretation
> of the spec, and so even though that system doesn't need a quirk for MSI
> because MSI works in general, we still MUST test the MSI vector to make
> sure it works *for us*  In this case it comes down to being an errata
> workaround.
> 
> Since there is no way to "test" generation of an interrupt from any
> specific hardware device without internal knowledge of said device,
> there isn't a way for us to help the kernel by writing a generic "test
> MSI" routine.
> 
> I would prefer this "generic test" code be in the driver rather than
> having to identify all the chipsets that fail and have the driver do
> *specific chipset* detection ala bnx2.c's 8132 bridge workaround.
> 
> >> The platforms with MSI problems should be discovered, made public,
> >> and worked around. 
> 
> This is our workaround, it is our fault, the incompatible chipset is in
> an x3850 IBM system.
> 

I've seem similar problems on other systems, so this would be a nice bit
of code to have somewhere.  I can see Jeff's argument for having this
outside of drivers, but to make my life easier I'd like to see this in
e1000e (and e1000!).  :-)

I suppose an more general alternative would be to make this code run as
an ethtool test (current or new one) and then have a module option to
disable/enable MSI so all could be done in userspace, but something
automatic like this sure would be great.

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html