[openib-general] ip over ib throughtput

[Grant Grundler]

On Thu, Jan 13, 2005 at 03:00:31PM -0800, Roland Dreier wrote:
> Yes, all we have to do is try a NOP FW command and see if it generates
> an interrupt on completion.  This will succeed within a millisecond in
> the extremely worst case if interrupts are working, so even the
> failure case will be quick.

Ok - that sounds great.

>     Grant> TBH, I would expect IRQ lines to work on *ALL* platforms.
>     Grant> Basic part of PCI compliance dictates this.
> 
> You would expect this but the supply of BIOS and kernel ACPI bugs
> seems inexhaustible...

Getting IRQ line routing wrong will show up for *any* device.
I'll agree this case is possibly an ACPI bug - but in an experimental
release of bleeding edge kernel version. That's no surprise.
Normal users will (almost) never see this kind of failure.

A coworker did build/boot (on rx2600) linus' bk tree 24h before 2.6.11-rc1
was tagged. I don't know of anyone with ZX1 platform following -mm tree.

>     Grant> It's clear not all platforms support MSI and the failure
>     Grant> mode may vary. So a test for MSI or MSI_X would be good.
>     Grant> And then the code can warn "MSI[-X] Failed - falling back
>     Grant> to IRQ Line".
> 
> That's a good idea but it may take me a little while to get to it
> because it's a little bit of a pain to unwind our MSI configuration
> after we get to the point where we can see if it worked or not.

Maybe that's a good reason to put the pci_enable_msi() right before
the enable_irq() call path?
(and ditto for free_irq()/pci_disable_msi())

Right now those two are not anywhere near each other and the MSI_HOWTO.txt
is very clear there is a dependency - at least for MSI support. I still
haven't studied details of MSI-X though I'm aware of some of the basic
differences.  I can submit a patch after I'm done messing with MSI
on tg3...sound ok?


> For
> now we'll have to be satisfied with a diagnostic telling you to try
> again without MSI...

ok - that's reasonable.

thanks,
grant