Should message signaled interrupts (MSI) work with Intel GMA 950?

[Roland Dreier]

> We don't know of any interrupt breakage related to MSI, but OTOH we
> haven't tested it either.

If should be easy to test -- just add pci_enable_msi() to
drm_fill_in_dev() like my
patch did.  (I got fancy and added a driver feature to turn it on, and
also cleared
the shared IRQ feature flag if setting MSI succeeded).  If you get a chance, to
try, I'd be curious to hear whether it works for you (or whether I'm
making a silly
mistake in how I enabled MSI).

> For sending vblanks, it doesn't seem like
> there's much of an advantage to using MSIs, but (assuming they work) if
> we modified the driver to use MSIs for other events it could simplify
> the interrupt handler a bit, which might be nice...

Yeah, the only reason I tried to enable it was to be able to separate the i915
interrupt from the other interrupts it ends up sharing an IRQ with, so
that I could
see who was really waking up the CPU in powertop.  From a quick look at
i915_driver_irq_handler(), I actually don't see how MSI would let you
simplify it at
all; it seems you have to read a register to see which interrupt you're getting
anyway, so knowing the interrupt is non-shared doesn't really help.

 - R.