Bug 214931

Created attachment 102582 [details]
hwinfo of this machine

This looks like a regression of VIA irq quirk
    http://bugzilla.kernel.org/show_bug.cgi?id=6419

Could you try the patch there?  Also, please attach the whole kernel boot log.

[Added Greg to Cc since he already committed it to 2.6.19-rc.]

Created attachment 102588 [details]
boot.msg of the system

I don't want to build kernels right now ... this is nothing i regularly do, and would not tell anything about the problem in the end.

OK, then I reassign this bug to Greg.

I'm confused.  Takashi, which patch did you apply to the kernel?  I don't see a patch in that bugreport on bugzilla.kernel.org

The fixes are already in 2.6.20:

09d6029f43ebbe7307854abdae204c25d711ff94:
   PCI: VIA IRQ quirk behaviour change

1597cacbe39802d86656d1f2e6329895bd2ef531:
   PCI: Fix multiple problems with VIA hardware

So, I guess updating to the STABLE/FACTORY kernel based on 2.6.20 should fix these problems.  Berthold, could you try the latest STABLE kernel?

I just tried 2.6.20-rc4-git6-2-default, and the behavior is broken again. I admit, that this should be different when looking at the comments above, but it is broken right now.

Takashi, reassigning back to you :)

I revert the relevent patch from 2.6.20 and built kernel packages.  Please try the one at mbuild/bari-tiwai-62 (for STABLE).  bari-tiwai-63 is kernel-source.rpm.

If this works, it's a regression by the patch...

that one works flawlessly.

Ok, then it's a breakage by the patch:

1597cacbe39802d86656d1f2e6329895bd2ef531:
   PCI: Fix multiple problems with VIA hardware

I give you this bug back to you, Greg :)

(Added Jean to Cc since he appears in the patch, too.)

Created attachment 114657 [details]
This patch breaks some VIA system

I'm a bit confused, what kernels were tested exactly, which one worked, which one failed?

Berthold, can you please provide the boot log from a working kernel on the same machine, for comparison?

well, the kernel that is broken was from STABLE. I just took that directly from dist.

Takashi built a working kernel for me, which can be found at the internal mbuild directory. Unfortunately, the boot.log will not help you, because I just have a SLES machine where I installed that kernel, and normally, alsa is not configured on there. However, a rcalsa restart makes the sound system working.

The working kernel is 2.6.20-rc5-git1-2-default. Obviously, in addition to removing the via patch, there has been an update to rc5-git1 in this kerenl:

* Tue Jan 23 2007 - olh@suse.de
- update to 2.6.20-rc5-git1
  scsi, usb, powerpc, ocfs fixes

* Mon Jan 22 2007 - jblunck@suse.de
- rpm/kernel-binary.spec.in: fix "Requires:" and config-subst call.

When doing the rcalsa restart, dmesg tells me:

via82xx: Assuming DXS channels with 48k fixed sample rate.
         Please try dxs_support=5 option
         and report if it works on your machine.
         For more details, read ALSA-Configuration.txt.
ACPI: PCI Interrupt Link [LNKC] enabled at IRQ 11
ACPI: PCI Interrupt 0000:00:11.5[C] -> Link [LNKC] -> GSI 11 (level, low) -> IRQ 11
PCI: VIA IRQ fixup for 0000:00:11.5, from 9 to 11
PCI: Setting latency timer of device 0000:00:11.5 to 64
ACPI: PCI interrupt for device 0000:00:11.5 disabled
pnp: Device 00:0e disabled.
via82xx: Assuming DXS channels with 48k fixed sample rate.
         Please try dxs_support=5 option
         and report if it works on your machine.
         For more details, read ALSA-Configuration.txt.
ACPI: PCI Interrupt 0000:00:11.5[C] -> Link [LNKC] -> GSI 11 (level, low) -> IRQ 11
PCI: Setting latency timer of device 0000:00:11.5 to 64

The faulty patch is a mess, it should have been two or three separate patches. Shame on Andrew and Alan :( That being said, I can't see any obvious problem with it. All I can offer is to build a kernel with additional debugging messages, and see what shows in the logs when running it.

Please set the Hardware field to the right value.

Created attachment 114809 [details]
VIA IRQ quirk debugging patch

With the latest kernel, I get the following messages. This kernel also does not play the sound correctly:

via82xx: Assuming DXS channels with 48k fixed sample rate.
         Please try dxs_support=5 option
         and report if it works on your machine.
         For more details, read ALSA-Configuration.txt.
ACPI: PCI Interrupt Link [LNKC] enabled at IRQ 11
ACPI: PCI Interrupt 0000:00:11.5[C] -> Link [LNKC] -> GSI 11 (level, low) -> IRQ 11
PCI: Setting latency timer of device 0000:00:11.5 to 64
irq 9: nobody cared (try booting with the "irqpoll" option)
 [<c0148f5a>] __report_bad_irq+0x36/0x7d
 [<c0149146>] note_interrupt+0x1a5/0x1e3
 [<c01e9d72>] acpi_irq+0xb/0x14
 [<c01486a9>] handle_IRQ_event+0x1a/0x3f
 [<c0149a05>] handle_level_irq+0xa3/0xda
 [<c010638e>] do_IRQ+0x82/0x9c
 [<c0104d13>] common_interrupt+0x23/0x30
 [<c01258f4>] __do_softirq+0x54/0xba
 [<c012598c>] do_softirq+0x32/0x36
 [<c0106393>] do_IRQ+0x87/0x9c
 [<c0104d13>] common_interrupt+0x23/0x30
 [<c0102c2e>] default_idle+0x0/0x3e
 [<c0102c5a>] default_idle+0x2c/0x3e
 [<c010238f>] cpu_idle+0x9d/0xb6
 [<c036870e>] start_kernel+0x35e/0x366
 [<c03681ae>] unknown_bootoption+0x0/0x202
 =======================
handlers:
[<c01e9d67>] (acpi_irq+0x0/0x14)
Disabling IRQ #9

No "VIA quirk" messages in the logs at all?

I just rechecked. The previous output was all I got in dmesg, but also in syslog there are no further messages.

I logged into the machine to take a look. Earlier in the logs I see the following:
VIA quirk 0000:00:01.0: No supported south bridge found
At this point the VIA quirk is disabled and will not be attempted anymore.

Good news is that I didn't expect that, so at least I have something new to investigate ;)

Nick, it looks like you've been bitten by this on too?
http://lkml.org/lkml/2007/1/23/254

Your patch looks totally weird to me and probably only works by accident. I suspect that the root cause of the problem is that pci_find_present() does not work yet at the time the PCI quirks are run.

Created attachment 115902 [details]
Fix VIA quirks breakage

Fix VIA quirks that were recently broken by Alan Cox in the upstream kernel (commit 1597cacbe39802d86656d1f2e6329895bd2ef531).

My understanding is that pci_find_present() doesn't work yet at the time the quirks are run. So I used a two-step quirk as is done for some other quirks already. First we detect the VIA south bridges and set the right low and high device limits, then we are ready to actually run the quirks on the affected devices.

I will submit this patch upstream for comments now.

Created attachment 116115 [details]
Fix VIA quirks breakage (v2)

My first patch didn't fix Nick's system, because the quirkable device range wasn't correct for his old VT82C686 south bridge. This second version of the patch works; I've submitted it upstream as well.

Created attachment 116184 [details]
Fix VIA quirks breakage (v3)

One more iteration, after Sergey Vlasov exlained to me that the VT82C686 device number is different from one system to the next.

Fixed upstream in 2.6.20-rc7, and fixed in our HEAD tree as a result.

*** Bug 216398 has been marked as a duplicate of this bug. ***