PCI VPD size

PCI VPD size

[Stephen Hemminger]

Why did you choose to limit VPD access by using PCI quirks:
  http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=99cb233d60cbe644203f19938c729ea2bb004d70

Rather than controlling size in the device driver which is what my
patches (which seem to have been ignored) did:
  http://marc.info/?l=linux-netdev&m=122393195716363&w=2

The problem with the quirk scheme is that it means putting more
per device settings in a location farther away in the source code
than the underlying device driver.

Re: PCI VPD size

[Ben Hutchings]

Stephen Hemminger wrote:
> Why did you choose to limit VPD access by using PCI quirks:
>   http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=99cb233d60cbe644203f19938c729ea2bb004d70
> 
> Rather than controlling size in the device driver which is what my
> patches (which seem to have been ignored) did:
>   http://marc.info/?l=linux-netdev&m=122393195716363&w=2
> 
> The problem with the quirk scheme is that it means putting more
> per device settings in a location farther away in the source code
> than the underlying device driver.

The limit needs to be set in quirks because VPD I/O beyond that limit
causes those devices to stop responding.  If I understand correctly,
you're only reducing the limit as an optimisation.

Ben.

-- 
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.

Re: PCI VPD size

[Jesse Barnes]

On Friday, October 24, 2008 12:50 pm Stephen Hemminger wrote:
> Why did you choose to limit VPD access by using PCI quirks:
>  
> http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h
>=99cb233d60cbe644203f19938c729ea2bb004d70
>
> Rather than controlling size in the device driver which is what my
> patches (which seem to have been ignored) did:
>   http://marc.info/?l=linux-netdev&m=122393195716363&w=2
>
> The problem with the quirk scheme is that it means putting more
> per device settings in a location farther away in the source code
> than the underlying device driver.

The last status I received on that was a comment from Jeff about assuming the 
maximum size for the VPD space.  That's the whole reason we have the quirks: 
on many devices accessing beyond the end of VPD space can cause a device hang.  
And since that space is available from a sysfs file we need to be extra 
careful.

Other than that, your patches looked fine, I was just waiting for you to 
address that last comment.

-- 
Jesse Barnes, Intel Open Source Technology Center

Re: PCI VPD size

[Stephen Hemminger]

On Fri, 24 Oct 2008 14:18:38 -0700
Jesse Barnes <jbarnes@virtuousgeek.org> wrote:

> On Friday, October 24, 2008 12:50 pm Stephen Hemminger wrote:
> > Why did you choose to limit VPD access by using PCI quirks:
> >  
> > http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h
> >=99cb233d60cbe644203f19938c729ea2bb004d70
> >
> > Rather than controlling size in the device driver which is what my
> > patches (which seem to have been ignored) did:
> >   http://marc.info/?l=linux-netdev&m=122393195716363&w=2
> >
> > The problem with the quirk scheme is that it means putting more
> > per device settings in a location farther away in the source code
> > than the underlying device driver.
> 
> The last status I received on that was a comment from Jeff about assuming the 
> maximum size for the VPD space.  That's the whole reason we have the quirks: 
> on many devices accessing beyond the end of VPD space can cause a device hang.  
> And since that space is available from a sysfs file we need to be extra 
> careful.
> 
> Other than that, your patches looked fine, I was just waiting for you to 
> address that last comment.
> 

The max size of vpd issue was addressed in last round (I thought).
I'll respin against 2.6.28-rc1