Re: Time to make PCI_MSI default y ?

Re: Time to make PCI_MSI default y ?

[Ralf Baechle]

On Tue, Nov 24, 2009 at 01:39:48PM +1100, Michael Ellerman wrote:

> Having just hit a build-break caused by a distro building with
> PCI_MSI=n, I set out to make it default y for powerpc. Unfortunately
> that's not possible, because it's in drivers/pci/Kconfig.
> 
> So is it time to make it default y for everyone? It seems to me having
> it off is more likely to cause problems than having it on these days,
> though I'm not sure if that is true for all archs.
> 
> An arch that really didn't want it default y could conditionally select
> ARCH_SUPPORTS_MSI, like x86 does already.

On MIPS the age of MSI only recently started; once single platform (Cavium)
out of all the many uses it.  Cavium does a "select ARCH_SUPPORTS_MSI" but
not "select PCI_MSI" because not all platform variants actually have PCI.

We should  not give a user a chance to select something wrong in kconfig
thus automatically as many options for a platform as possible is a good
thing - after all the kconfig dialog for any given platfrom has become
painfully long.  And we really should have to avoid users having to know
that the Frobnic 2000 they're trying to upgrade the kernel for requires
MSI to work ...

  Ralf

Re: Time to make PCI_MSI default y ?

[Benjamin Herrenschmidt]

On Tue, 2009-11-24 at 12:53 +0000, Ralf Baechle wrote:
> On Tue, Nov 24, 2009 at 01:39:48PM +1100, Michael Ellerman wrote:
> 
> > Having just hit a build-break caused by a distro building with
> > PCI_MSI=n, I set out to make it default y for powerpc. Unfortunately
> > that's not possible, because it's in drivers/pci/Kconfig.
> > 
> > So is it time to make it default y for everyone? It seems to me having
> > it off is more likely to cause problems than having it on these days,
> > though I'm not sure if that is true for all archs.
> > 
> > An arch that really didn't want it default y could conditionally select
> > ARCH_SUPPORTS_MSI, like x86 does already.
> 
> On MIPS the age of MSI only recently started; once single platform (Cavium)
> out of all the many uses it.  Cavium does a "select ARCH_SUPPORTS_MSI" but
> not "select PCI_MSI" because not all platform variants actually have PCI.
> 
> We should  not give a user a chance to select something wrong in kconfig
> thus automatically as many options for a platform as possible is a good
> thing - after all the kconfig dialog for any given platfrom has become
> painfully long.  And we really should have to avoid users having to know
> that the Frobnic 2000 they're trying to upgrade the kernel for requires
> MSI to work ...

Still... select has nasty issues. I think default y is fine here. For
platforms that don't need it, make sure their defconfigs don't have it
set...

Or maybe default y if (X86 || PPC)

Cheers,
Ben.

Re: Time to make PCI_MSI default y ?

[Michael Ellerman]

[-- Attachment #1: Type: text/plain, Size: 1796 bytes --]

On Wed, 2009-11-25 at 08:09 +1100, Benjamin Herrenschmidt wrote:
> On Tue, 2009-11-24 at 12:53 +0000, Ralf Baechle wrote:
> > On Tue, Nov 24, 2009 at 01:39:48PM +1100, Michael Ellerman wrote:
> > 
> > > Having just hit a build-break caused by a distro building with
> > > PCI_MSI=n, I set out to make it default y for powerpc. Unfortunately
> > > that's not possible, because it's in drivers/pci/Kconfig.
> > > 
> > > So is it time to make it default y for everyone? It seems to me having
> > > it off is more likely to cause problems than having it on these days,
> > > though I'm not sure if that is true for all archs.
> > > 
> > > An arch that really didn't want it default y could conditionally select
> > > ARCH_SUPPORTS_MSI, like x86 does already.
> > 
> > On MIPS the age of MSI only recently started; once single platform (Cavium)
> > out of all the many uses it.  Cavium does a "select ARCH_SUPPORTS_MSI" but
> > not "select PCI_MSI" because not all platform variants actually have PCI.
> > 
> > We should  not give a user a chance to select something wrong in kconfig
> > thus automatically as many options for a platform as possible is a good
> > thing - after all the kconfig dialog for any given platfrom has become
> > painfully long.  And we really should have to avoid users having to know
> > that the Frobnic 2000 they're trying to upgrade the kernel for requires
> > MSI to work ...
> 
> Still... select has nasty issues. I think default y is fine here. For
> platforms that don't need it, make sure their defconfigs don't have it
> set...
> 
> Or maybe default y if (X86 || PPC)

Oh yeah I forgot we could do that. So we could either do that, or add a
ARCH_MSI_DEFAULT and use that as the if condition. If people really
care.

cheers



[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 197 bytes --]