Suggestions for debugging ata_piix and ahci driver failures on Xen pv_ops dom0 kernel (2.6.29-rcX)

Suggestions for debugging ata_piix and ahci driver failures on Xen pv_ops dom0 kernel (2.6.29-rcX)

[Pasi Kärkkäinen]

Hello!

I'm trying to figure out what's wrong with the pv_ops dom0 kernel/patches [1]
causing at least ata_piix and ahci drivers to fail to work.

There was legacy IRQ enabling problem earlier with the pv_ops dom0 patches, 
which pretty much prevented the disks from being detected.. but that got fixed. 
It could still be something related to Xen/dom0 irq handling.. 

I'd be glad if someone could take a look at this 2.6.29-rc3-tip pv_ops dom0 kernel boot log
and comment about _possible_ causes for such behaviour of ata_piix and ahci:

http://pasik.reaktio.net/xen/pv_ops-dom0-debug/pv_ops-dom0-bootlog-11-xen331-linux-2.6.29-rc3-atadebug3.txt

In that testing box I have just one IDE hdd attached to ata_piix controller, the AHCI controller doesn't
have any disks attached.

The exact same kernel works OK without problems on baremetal without Xen
hypervisor. 

[1] http://xenbits.xen.org/paravirt_ops/patches.hg/ and
http://wiki.xensource.com/xenwiki/XenParavirtOps

-- Pasi

Re: Suggestions for debugging ata_piix and ahci driver failures on Xen pv_ops dom0 kernel (2.6.29-rcX)

[Robert Hancock]

Pasi Kärkkäinen wrote:
> Hello!
> 
> I'm trying to figure out what's wrong with the pv_ops dom0 kernel/patches [1]
> causing at least ata_piix and ahci drivers to fail to work.
> 
> There was legacy IRQ enabling problem earlier with the pv_ops dom0 patches, 
> which pretty much prevented the disks from being detected.. but that got fixed. 
> It could still be something related to Xen/dom0 irq handling.. 
> 
> I'd be glad if someone could take a look at this 2.6.29-rc3-tip pv_ops dom0 kernel boot log
> and comment about _possible_ causes for such behaviour of ata_piix and ahci:
> 
> http://pasik.reaktio.net/xen/pv_ops-dom0-debug/pv_ops-dom0-bootlog-11-xen331-linux-2.6.29-rc3-atadebug3.txt
> 
> In that testing box I have just one IDE hdd attached to ata_piix controller, the AHCI controller doesn't
> have any disks attached.
> 
> The exact same kernel works OK without problems on baremetal without Xen
> hypervisor. 

Seems like me like ata_piix is not receiving any interrupts.. the 
polling-based IDENTIFY and SET FEATURES work, but the READ NATIVE MAX 
ADDRESS times out and so does the READ DMA that gets tried next.

Note that the ata_piix controller is in legacy mode, not native mode, so 
it's using IRQ 14 and 15, not whatever the hardware may otherwise 
indicate. Could be the problem is related to that.

Re: Suggestions for debugging ata_piix and ahci driver failures on Xen pv_ops dom0 kernel (2.6.29-rcX)

[Pasi Kärkkäinen]

On Thu, Feb 05, 2009 at 06:45:02PM -0600, Robert Hancock wrote:
> Pasi Kärkkäinen wrote:
> >Hello!
> >
> >I'm trying to figure out what's wrong with the pv_ops dom0 kernel/patches 
> >[1]
> >causing at least ata_piix and ahci drivers to fail to work.
> >
> >There was legacy IRQ enabling problem earlier with the pv_ops dom0 
> >patches, which pretty much prevented the disks from being detected.. but 
> >that got fixed. It could still be something related to Xen/dom0 irq 
> >handling.. 
> >I'd be glad if someone could take a look at this 2.6.29-rc3-tip pv_ops 
> >dom0 kernel boot log
> >and comment about _possible_ causes for such behaviour of ata_piix and 
> >ahci:
> >
> >http://pasik.reaktio.net/xen/pv_ops-dom0-debug/pv_ops-dom0-bootlog-11-xen331-linux-2.6.29-rc3-atadebug3.txt
> >
> >In that testing box I have just one IDE hdd attached to ata_piix 
> >controller, the AHCI controller doesn't
> >have any disks attached.
> >
> >The exact same kernel works OK without problems on baremetal without Xen
> >hypervisor. 
> 
> Seems like me like ata_piix is not receiving any interrupts.. the 
> polling-based IDENTIFY and SET FEATURES work, but the READ NATIVE MAX 
> ADDRESS times out and so does the READ DMA that gets tried next.
> 

OK. Thanks for pointing this out. 

> Note that the ata_piix controller is in legacy mode, not native mode, so 
> it's using IRQ 14 and 15, not whatever the hardware may otherwise 
> indicate. Could be the problem is related to that.
>

Yep. With earlier version of pv_ops dom0 patches devm_request_irq() call failed 
for IRQ 14 and 15 in ata_piix driver, but that has been fixed now.. 

Calling devm_request_irq() returns success now, but maybe it doesn't actually
work and do what it should? 

Thanks for your reply!

-- Pasi