Re: [Regression, post-rc2] Commit a5ee4eb7541 breaks OpenGL on RS780 (was: Re: Linux 2.6.34-rc3)

[Alex Deucher <alexdeucher@gmail.com>]

On Thu, Apr 1, 2010 at 5:08 PM, Rafael J. Wysocki <rjw@sisk.pl> wrote:
> On Thursday 01 April 2010, Alex Deucher wrote:
>> On Thu, Apr 1, 2010 at 4:48 PM, Rafael J. Wysocki <rjw@sisk.pl> wrote:
>> > On Thursday 01 April 2010, Alex Deucher wrote:
>> >> On Thu, Apr 1, 2010 at 4:28 PM, Rafael J. Wysocki <rjw@sisk.pl> wrote:
>> >> > On Thursday 01 April 2010, Alex Deucher wrote:
>> >> >> On Thu, Apr 1, 2010 at 2:36 AM, Clemens Ladisch <clemens@ladisch.de> wrote:
>> >> >> > Alex Deucher wrote:
>> >> >> >> On Wed, Mar 31, 2010 at 9:13 PM, Rafael J. Wysocki <rjw@sisk.pl> wrote:
>> >> >> >>> On Tuesday 30 March 2010, Rafael J. Wysocki wrote:
>> >> >> >>>> >       PCI quirk: RS780/RS880: work around missing MSI initialization
>> >> >> >>>>
>> >> >> >>>> This one (commit a5ee4eb7541) broke OpenGL acceleration on my new test box
>> >> >> >>>> which happens to have a RS780.
>> >> >> >
>> >> >> > So it's better to disable MSI unconditionally.
>> >> >> >
>> >> >> > Rafael, can you check if MSI works for the HDMI audio device?
>> >> >> > (I'd guess it doesn't.)
>> >> >> >
>> >> >> >> I also have the attached patch queued in via Dave's tree to disable
>> >> >> >> MSI on all IGP chips for the time being.
>> >> >> >
>> >> >> > This disables MSI only for the graphics device.  I'd prefer to have
>> >> >> > the quirk on its bridge so that MSI gets disabled for the HDMI audio
>> >> >> > device too, to avoid having to duplicate this quirk in the snd-hda-intel
>> >> >> > driver.
>> >> >> >
>> >> >> > ==========
>> >> >> >
>> >> >> > PCI quirk: RS780/RS880: disable MSI completely
>> >> >> >
>> >> >> > The missing initialization of the nb_cntl.strap_msi_enable does not seem
>> >> >> > to be the only problem that prevents MSI, so that quirk is not
>> >> >> > sufficient to enable MSI on all machines.  To be safe, unconditionally
>> >> >> > disable MSI for the internal graphics and HDMI audio on these chipsets.
>> >> >> >
>> >> >> > Signed-off-by: Clemens Ladisch <clemens@ladisch.de>
>> >> >>
>> >> >> Works fine here.
>> >> >>
>> >> >> Tested-by: Alex Deucher <alexdeucher@gmail.com>
>> >> >
>> >> > Unfortunately it doesn't work for me without the
>> >> >
>> >> > if ((rdev->family >= CHIP_RV380) &&
>> >> >            (!(rdev->flags & RADEON_IS_IGP)))
>> >> >
>> >> > radeon quirk.
>> >>
>> >> what are your pci ids?
>> >
>> > 1022:960b
>> >
>> > I guess 1022 is AMD.
>> >
>> > OK, I'll try to add that.
>>
>> It's possible your oem has the wrong vendor id for the 0x9602 bridge.
>
> Yes, the patch below works.
>
> Thanks,
> Rafael
>
>
> ---
>  drivers/gpu/drm/radeon/radeon_irq_kms.c |    3 --
>  drivers/pci/quirks.c                    |   36 ++------------------------------
>  2 files changed, 4 insertions(+), 35 deletions(-)
>
> Index: linux-2.6/drivers/pci/quirks.c
> ===================================================================
> --- linux-2.6.orig/drivers/pci/quirks.c
> +++ linux-2.6/drivers/pci/quirks.c
> @@ -2123,6 +2123,9 @@ static void __devinit quirk_disable_msi(
>        }
>  }
>  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_8131_BRIDGE, quirk_disable_msi);
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AMD, 0x9602, quirk_disable_msi);
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_ASUSTEK, 0x9602, quirk_disable_msi);
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AI, 0x9602, quirk_disable_msi);
>  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_VIA, 0xa238, quirk_disable_msi);
>
>  /* Go through the list of Hypertransport capabilities and
> @@ -2495,39 +2498,6 @@ DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AT
>  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_ATI, 0x4375,
>                        quirk_msi_intx_disable_bug);
>
> -/*
> - * MSI does not work with the AMD RS780/RS880 internal graphics and HDMI audio
> - * devices unless the BIOS has initialized the nb_cntl.strap_msi_enable bit.
> - */
> -static void __init rs780_int_gfx_disable_msi(struct pci_dev *int_gfx_bridge)
> -{
> -       u32 nb_cntl;
> -
> -       if (!int_gfx_bridge->subordinate)
> -               return;
> -
> -       pci_bus_write_config_dword(int_gfx_bridge->bus, PCI_DEVFN(0, 0),
> -                                  0x60, 0);
> -       pci_bus_read_config_dword(int_gfx_bridge->bus, PCI_DEVFN(0, 0),
> -                                 0x64, &nb_cntl);
> -
> -       if (!(nb_cntl & BIT(10))) {
> -               dev_warn(&int_gfx_bridge->dev,
> -                        FW_WARN "RS780: MSI for internal graphics disabled\n");
> -               int_gfx_bridge->subordinate->bus_flags |= PCI_BUS_FLAGS_NO_MSI;
> -       }
> -}
> -
> -#define PCI_DEVICE_ID_AMD_RS780_P2P_INT_GFX    0x9602
> -
> -DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_AMD,
> -                       PCI_DEVICE_ID_AMD_RS780_P2P_INT_GFX,
> -                       rs780_int_gfx_disable_msi);
> -/* wrong vendor ID on M4A785TD motherboard: */
> -DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_ASUSTEK,
> -                       PCI_DEVICE_ID_AMD_RS780_P2P_INT_GFX,
> -                       rs780_int_gfx_disable_msi);
> -
>  #endif /* CONFIG_PCI_MSI */
>
>  #ifdef CONFIG_PCI_IOV
> Index: linux-2.6/drivers/gpu/drm/radeon/radeon_irq_kms.c
> ===================================================================
> --- linux-2.6.orig/drivers/gpu/drm/radeon/radeon_irq_kms.c
> +++ linux-2.6/drivers/gpu/drm/radeon/radeon_irq_kms.c
> @@ -117,8 +117,7 @@ int radeon_irq_kms_init(struct radeon_de
>        /* MSIs don't seem to work reliably on all IGP
>         * chips.  Disable MSI on them for now.
>         */
> -       if ((rdev->family >= CHIP_RV380) &&
> -           (!(rdev->flags & RADEON_IS_IGP))) {
> +       if (rdev->family >= CHIP_RV380) {
>                int ret = pci_enable_msi(rdev->pdev);
>                if (!ret) {
>                        rdev->msi_enabled = 1;
>

Let's skip this second chunk for now as there are other non-RS780 IGP
chips that could be problematic, so I'd rather just leave MSIs
disabled for now.

Alex