Re: MSI failure on Nvidia nForce

Re: MSI failure on Nvidia nForce

[Jeff Garzik]

Stephen Hemminger wrote:
> I got a report of sky2 driver irq test failing on x86_64 using
> the following configuration.  Is this a known problem?
> Should workaround be done at PCI layer?
> 
> What the driver does is setup MSI handler, then do a software generated
> IRQ and check that it was received (similar to tg3).  If IRQ test fails
> it falls back to INTx.

Please describe precisely -how- it fails.

pci_enable_msi() does not fail properly on systems that do not support 
MSI.  This is a major unresolved problem that is preventing MSI 
deployment, and causing every driver writer to include a does-MSI-work 
test in their driver.

We need to find a good generic test, or if that fails, adopt an 
ACPI-like rule:  whitelist systems with working MSI before $X date, and 
blacklist systems with broken MSI after $X date.

	Jeff

Re: MSI failure on Nvidia nForce

[Stephen Hemminger]

On Tue, 18 Apr 2006 14:37:44 -0400
Jeff Garzik <jeff@garzik.org> wrote:

> Stephen Hemminger wrote:
> > I got a report of sky2 driver irq test failing on x86_64 using
> > the following configuration.  Is this a known problem?
> > Should workaround be done at PCI layer?
> > 
> > What the driver does is setup MSI handler, then do a software generated
> > IRQ and check that it was received (similar to tg3).  If IRQ test fails
> > it falls back to INTx.
> 
> Please describe precisely -how- it fails.
> 
> pci_enable_msi() does not fail properly on systems that do not support 
> MSI.  This is a major unresolved problem that is preventing MSI 
> deployment, and causing every driver writer to include a does-MSI-work 
> test in their driver.
> 
> We need to find a good generic test, or if that fails, adopt an 
> ACPI-like rule:  whitelist systems with working MSI before $X date, and 
> blacklist systems with broken MSI after $X date.
> 
> 	Jeff
> 

The message from the driver reported was:

> I'me currently getting a lot of hang up with my computer (x86_64,
> ati,...). I've saw this message in my dmesg :
> 
> sky2: 0000:02:00.0: No interupt was generated using MSI, switching to
> INTx mode. Please report this failure to the PCI maintener and include
> system chipset information.

Which means that. pci_enable_msi succeeded, but the IRQ routing was
screwed up.. There maybe more of a BIOS problem on this system because
the IRQ is showing up as edge triggered as well. Edge triggered interrupts
don't work with NAPI.

cat /proc/interrupts
           CPU0
  0:     333581    IO-APIC-edge  timer
  1:       1461    IO-APIC-edge  i8042
  2:          0          XT-PIC  cascade
  7:          2    IO-APIC-edge  ehci_hcd:usb1, NVidia CK804
  8:          0    IO-APIC-edge  rtc
  9:          0    IO-APIC-edge  acpi
 10:      55872    IO-APIC-edge  sky2
 11:      21025    IO-APIC-edge  libata
 15:         49    IO-APIC-edge  ide1
NMI:        587
LOC:     333569
ERR:          0
MIS:          0