Re: MSI messages

[Michael Ellerman]

On Mon, 2008-12-22 at 19:31 +0400, Manu Abraham wrote:
> Michael Ellerman wrote:
> > On Mon, 2008-12-22 at 02:56 +0400, Manu Abraham wrote:
> >> Michael Ellerman wrote:
> >>> On Sun, 2008-12-21 at 01:13 +0400, Manu Abraham wrote:
> >>>> Grant Grundler wrote:
> >>>>> On Sun, Dec 14, 2008 at 03:15:15PM +0400, Manu Abraham wrote:
> >>>>> ...
> >>>>>>> A "GSI" (Generic Sys Interrupt?) is associated with each entry in
> >>>>>>> the MSI-X table. Driver then calls request_irq() to bind an interrupt
> >>>>>>> handler to each GSI. So the driver never directly sees the "message".
> >>>>>> Oh, you mean the array of irq_handlers in the MSI-X table should 
> >>>>>> correspond to a particular message ?
> >>>>> No. I mean each MSI-X entry has an address+message pair and each MSI-X entry
> >>>>> is associated with the parameters passed to each call of request_irq().
> >>>>> Pass in unique private data to each call of request_irq() and the driver
> >>>>> can determine which message was delivered when the ISR gets called.
> >>>>> (ISR == Interrupt Service Routine, aka driver IRQ handler)
> >>>> Is it possible that even when the config space says that multiple messages 
> >>>> are supported, you cannot enable MSI-X ?
> >>> Yes, absolutely. Have a look at pci_msi_check_device() for starters. MSI
> >>> can be disabled globally, or per-device by a quirk, the bridges above
> >>> your device might not support MSI and the arch code may prevent
> >>> MSI/MSI-X for some reason (ie. platform doesn't support it).
> >>>
> >>> There's also a check in pci_enable_msix() to make sure the entries you
> >>> pass in are valid.
> >>>
> >>>>  ------------- with MSI-X -------------
> >>>>
> >>>> saa716x_pci_init (0): found a NEMO reference board PCIe card
> >>>> ACPI: PCI Interrupt 0000:05:00.0[A] -> GSI 19 (level, low) -> IRQ 19
> >>>> PCI: Setting latency timer of device 0000:05:00.0 to 64
> >>>> saa716x_request_irq (0): Using MSI-X mode
> >>>> saa716x_enable_msix (0): MSI-X request failed
> >>> For starters you should print the error code returned by
> >>> pci_enable_msix() - unfortunately it returns EINVAL for many different
> >>> reasons, but it will narrow it down a bit.
> >> It does return -22
> > 
> > If you're curious you could try this patch.
> 
> 
> There's a missing opening brace. I did add it and added more of kernel hacking
> for debugging and loaded the modules after that, which did cause a complete 
> freeze of the machine.

Yeah sorry I didn't compile it. I don't see how it would have caused the
kernel to hang though.

cheers

-- 
Michael Ellerman
OzLabs, IBM Australia Development Lab

wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)

We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person

Attachment: signature.asc
Description: This is a digitally signed message part