Re: MSI and driver APIs

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Thu, 2005-12-15 at 13:18 -0800, Roland Dreier wrote:
>     Benjamin> You can have multiple MSIs too (they just have to be
>     Benjamin> contiguous in numbers and aligned on the nearest power
>     Benjamin> of 2).
> 
> Good point.  I just got too used to the x86-ism of the current MSI
> code, which can't handle allocating contiguous vectors.
> 
>     Benjamin> I'm tempted to leave them enabled and only disable them
>     Benjamin> when request_irq() is done on the legacy INTx... Does
>     Benjamin> anybody see a problem with this approach ?
> 
> You might run into trouble on hardware (think tg3 or its ilk again)
> where you might have to do something beyond disabling MSI in the PCI
> header to switch the chip out of MSI mode.

But won't the driver call pci_enable/disable_msi() in those cases ? If
not, it's easy enough to add (explicit disable rather than not-enabled).

I'm mostly concerned about "dumb" drivers that don't know about MSI at
all...

Ben.