| Message ID | 201304042158.r34LwGPg010714@d03av02.boulder.ibm.com |
|---|---|
| State | Not Applicable |
| Headers | show |
Hi Brian, On Thu, Apr 04, 2013 at 04:58:17PM -0500, Brian King wrote: > > Add a PCI quirk for VGA devices on Power to set the default VGA device. > Ensures a default VGA is always set if a graphics adapter is present, > even if firmware did not initialize it. If more than one graphics > adapter is present, ensure the one initialized by firmware is set > as the default VGA device. This ensures that X autoconfiguration > will work. So a few things: - You are doing this on all power systems, not just pseries which is I assume what you're testing on - that seems OK to me, but just checking. - What is the "initialized by firmware" test? Just that IO & MEM are enabled? - You potentially override an existing default, is that a problem? Can the user set the default? (no AFAICS). - The x86 code is slightly different, they don't override an existing default, why do we? cheers -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 04/05/2013 01:52 AM, Michael Ellerman wrote: > Hi Brian, > > > On Thu, Apr 04, 2013 at 04:58:17PM -0500, Brian King wrote: >> >> Add a PCI quirk for VGA devices on Power to set the default VGA device. >> Ensures a default VGA is always set if a graphics adapter is present, >> even if firmware did not initialize it. If more than one graphics >> adapter is present, ensure the one initialized by firmware is set >> as the default VGA device. This ensures that X autoconfiguration >> will work. > > So a few things: > > - You are doing this on all power systems, not just pseries which is I > assume what you're testing on - that seems OK to me, but just > checking. Correct. I've only tested on pseries, but figured it would make sense for this to be more generic. I'm happy to make this pseries specific if that is preferred. > - What is the "initialized by firmware" test? Just that IO & MEM are > enabled? Correct. This is what the x86 code does. Alternatively, its possible there is a chosen attribute in the device tree we could look at. > - You potentially override an existing default, is that a problem? Can > the user set the default? (no AFAICS). I couldn't find anywhere that this could be set by the user and wanted to be able to handle both the case of a single adapter that wasn't initialized by firmware as well as the multi adapter case where only one of the adapters was initialized by firmware. I could have made this smarter so that we only override the default if the previous default was not initialized by firmware, but opted for the simpler patch. > - The x86 code is slightly different, they don't override an existing > default, why do we? I wanted to be able to handle the case of a single graphics adapter installed in a system that has not been initialized by firmware, since I have a system like this. This generally isn't something that x86 would need to do since all graphics adapters should have x86 boot code in them, but very few current graphics adapters have fcode in them today. Thanks, Brian
On Thu, Apr 4, 2013 at 3:58 PM, Brian King <brking@linux.vnet.ibm.com> wrote: > > Add a PCI quirk for VGA devices on Power to set the default VGA device. > Ensures a default VGA is always set if a graphics adapter is present, > even if firmware did not initialize it. If more than one graphics > adapter is present, ensure the one initialized by firmware is set > as the default VGA device. This ensures that X autoconfiguration > will work. > > Signed-off-by: Brian King <brking@linux.vnet.ibm.com> > --- > > arch/powerpc/kernel/pci-common.c | 13 +++++++++++++ > 1 file changed, 13 insertions(+) > > diff -puN arch/powerpc/kernel/pci-common.c~powerpc_vga_default_device arch/powerpc/kernel/pci-common.c > --- linux/arch/powerpc/kernel/pci-common.c~powerpc_vga_default_device 2013-04-03 09:50:33.000000000 -0500 > +++ linux-bjking1/arch/powerpc/kernel/pci-common.c 2013-04-03 09:50:33.000000000 -0500 > @@ -30,6 +30,7 @@ > #include <linux/irq.h> > #include <linux/vmalloc.h> > #include <linux/slab.h> > +#include <linux/vgaarb.h> > > #include <asm/processor.h> > #include <asm/io.h> > @@ -1725,3 +1726,15 @@ static void fixup_hide_host_resource_fsl > } > DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_MOTOROLA, PCI_ANY_ID, fixup_hide_host_resource_fsl); > DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_FREESCALE, PCI_ANY_ID, fixup_hide_host_resource_fsl); > + > +static void fixup_vga(struct pci_dev *pdev) > +{ > + u16 cmd; > + > + pci_read_config_word(pdev, PCI_COMMAND, &cmd); > + if ((cmd & (PCI_COMMAND_IO | PCI_COMMAND_MEMORY)) || !vga_default_device()) > + vga_set_default_device(pdev); > + > +} > +DECLARE_PCI_FIXUP_CLASS_FINAL(PCI_ANY_ID, PCI_ANY_ID, > + PCI_CLASS_DISPLAY_VGA, 8, fixup_vga); This not really an arch-specific issue, so it's a shame to add another arch-specific quirk for it (in addition to the x86 and ia64 ones we already have). In b5e4efe7e0, Eiichiro Oiwa <eiichiro.oiwa.nm@hitachi.com> tried to make this quirk generic, but the implementation was naive and it didn't work for sparc64. There's a good thread about the sparc issue at https://lkml.org/lkml/2006/10/19/17 The outcome was to basically revert back to arch-specific quirks with 6b5c76b8e2, but I think the generic version probably could have been made to work, possibly with a pcibios hook or something for anything that really is arch-dependent, like the shadowed ROM image. This particular patch is arch-specific, and I'm not going to try to nack it because I'm not the powerpc maintainer, but I will certainly try to help you if you want to push on making a generic version. Bjorn -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, Apr 05, 2013 at 02:24:09PM -0600, Bjorn Helgaas wrote: > On Thu, Apr 4, 2013 at 3:58 PM, Brian King <brking@linux.vnet.ibm.com> wrote: > > @@ -1725,3 +1726,15 @@ static void fixup_hide_host_resource_fsl > > } > > DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_MOTOROLA, PCI_ANY_ID, fixup_hide_host_resource_fsl); > > DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_FREESCALE, PCI_ANY_ID, fixup_hide_host_resource_fsl); > > + > > +static void fixup_vga(struct pci_dev *pdev) > > +{ > > + u16 cmd; > > + > > + pci_read_config_word(pdev, PCI_COMMAND, &cmd); > > + if ((cmd & (PCI_COMMAND_IO | PCI_COMMAND_MEMORY)) || !vga_default_device()) > > + vga_set_default_device(pdev); > > + > > +} > > +DECLARE_PCI_FIXUP_CLASS_FINAL(PCI_ANY_ID, PCI_ANY_ID, > > + PCI_CLASS_DISPLAY_VGA, 8, fixup_vga); > > This not really an arch-specific issue, so it's a shame to add another > arch-specific quirk for it (in addition to the x86 and ia64 ones we > already have). > > In b5e4efe7e0, Eiichiro Oiwa <eiichiro.oiwa.nm@hitachi.com> tried to > make this quirk generic, but the implementation was naive and it > didn't work for sparc64. There's a good thread about the sparc issue > at https://lkml.org/lkml/2006/10/19/17 > > The outcome was to basically revert back to arch-specific quirks with > 6b5c76b8e2, but I think the generic version probably could have been > made to work, possibly with a pcibios hook or something for anything > that really is arch-dependent, like the shadowed ROM image. > > This particular patch is arch-specific, and I'm not going to try to > nack it because I'm not the powerpc maintainer, but I will certainly > try to help you if you want to push on making a generic version. OK. Looking at x86, ia64 and ours, there's a fair bit of difference. x86/ia64 both walk up the parents checking PCI_BRIDGE_CTL_VGA, which powerpc doesn't (though maybe it should?). x86/ia64 set IORESOURCE_ROM_SHADOW, which powerpc doesn't. ia64 doesn't call vga_set_default_device(), x86 and powerpc do. So we'll merge this and maybe someone can tease out the common bits, but personally I don't see that there's an obvious chunk of generic logic. cheers -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, 2013-04-08 at 15:21 +1000, Michael Ellerman wrote: > Looking at x86, ia64 and ours, there's a fair bit of difference. > > x86/ia64 both walk up the parents checking PCI_BRIDGE_CTL_VGA, which > powerpc doesn't (though maybe it should?). Unclear for several reasons. That bit indicates that the bridge forwards the hard coded legacy VGA IO ports and memory aperture. These are not necessary to get modern video cards going. Only if legacy modes are used. On x86, that tends to be the case (is it always even with EFI nowadays ?). On other architectures that is not necessarily the case. The firmware can (and will) initialize the card using MMIO entirely and even if possible disabling the legacy stuff, which means that turning those bits on in the bridge is also unnecessary. In fact, on such setups, the isn't really a concept of a "primary" video card to begin with. On the other hand, that also means that a video card initialized like that is pretty much out of the grasp of the vga arbiter which has no effect on it either. Also be careful that while it may be relevant on x86, the VGA fwd bit is not on the PCIe root complex on IBM machines. Finally, P8 has no IO space at all... > x86/ia64 set IORESOURCE_ROM_SHADOW, which powerpc doesn't. > > ia64 doesn't call vga_set_default_device(), x86 and powerpc do. > > So we'll merge this and maybe someone can tease out the common bits, but > personally I don't see that there's an obvious chunk of generic logic. Cheers, Ben. -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff -puN arch/powerpc/kernel/pci-common.c~powerpc_vga_default_device arch/powerpc/kernel/pci-common.c --- linux/arch/powerpc/kernel/pci-common.c~powerpc_vga_default_device 2013-04-03 09:50:33.000000000 -0500 +++ linux-bjking1/arch/powerpc/kernel/pci-common.c 2013-04-03 09:50:33.000000000 -0500 @@ -30,6 +30,7 @@ #include <linux/irq.h> #include <linux/vmalloc.h> #include <linux/slab.h> +#include <linux/vgaarb.h> #include <asm/processor.h> #include <asm/io.h> @@ -1725,3 +1726,15 @@ static void fixup_hide_host_resource_fsl } DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_MOTOROLA, PCI_ANY_ID, fixup_hide_host_resource_fsl); DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_FREESCALE, PCI_ANY_ID, fixup_hide_host_resource_fsl); + +static void fixup_vga(struct pci_dev *pdev) +{ + u16 cmd; + + pci_read_config_word(pdev, PCI_COMMAND, &cmd); + if ((cmd & (PCI_COMMAND_IO | PCI_COMMAND_MEMORY)) || !vga_default_device()) + vga_set_default_device(pdev); + +} +DECLARE_PCI_FIXUP_CLASS_FINAL(PCI_ANY_ID, PCI_ANY_ID, + PCI_CLASS_DISPLAY_VGA, 8, fixup_vga);
Add a PCI quirk for VGA devices on Power to set the default VGA device. Ensures a default VGA is always set if a graphics adapter is present, even if firmware did not initialize it. If more than one graphics adapter is present, ensure the one initialized by firmware is set as the default VGA device. This ensures that X autoconfiguration will work. Signed-off-by: Brian King <brking@linux.vnet.ibm.com> --- arch/powerpc/kernel/pci-common.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)