Re: [Penance PATCH] PCI: clean up the MSI code a bit

[Andi Kleen <ak@suse.de>]

On Wed, Jun 08, 2005 at 10:45:48AM -0700, Greg KH wrote:
> On Wed, Jun 08, 2005 at 11:14:40AM -0600, Grant Grundler wrote:
> > On Wed, Jun 08, 2005 at 03:41:09PM +0200, Andi Kleen wrote:
> > > I disagree it should stay as it is. Basically you are trading
> > > a bit less complexity in Infiniband now for a lot of code everywhere.
> > 
> > It's not just infiniband. It's tg3 and e1000 as well.
> 
> Yes, it's every device that wants to enable MSI.  So far, only one

No, it is every device that wants to use MSI-X. 

Also see the proposals from Jeff/Stefan et.al. for simpler interfaces
for this. No need to actually make it messy if we have nice helpers.

> driver that wants to enable MSI, has to handle broken devices.  And odds
> are, that driver just isn't tested properly yet :)
> 
> So I stand by my decision now, it's just too complex to enable MSI for

I think you decision is wrong here.

> everyone and expect drivers to disable it properly if they need to.  The
> logic is just convoluted (see the patch for details.)  As proof, I got
> it completly wrong the first time, and I'm still not sure that I got it
> correct after working on this for a while. :)
> 
> In the end, the pci_enable/pci_disable interface is the way to go.

No, it isnt.

-Andi