[PATCH v4] PCI: Support hibmc VGA cards behind a misbehaving HiSilicon bridge

[Will Deacon]

Hi Bjorn,

On Wed, Jul 12, 2017 at 03:04:30PM -0500, Bjorn Helgaas wrote:
> [+cc Catalin, Will, linux-arm-kernel]
> 
> On Wed, Jul 12, 2017 at 03:08:11PM +1000, Daniel Axtens wrote:
> > The HiSilicon D05 board has some PCI bridges (PCI ID 19e5:1610) that
> > are not spec-compliant: the VGA Enable bit is hardwired to 0 and
> > writes do not change it.
> > 
> > The HiSilicon engineers report that the bridge does not forward the
> > 0xa0000-0xbffff mem range and the 0x3b0-0x3bb and 0x3c0-0x3df I/O
> > ranges.
> > 
> > Because the VGA Enable bit is hardwired to 0, the VGA arbiter refuses
> > to mark any card behind it as the boot device. This breaks Xorg
> > auto-detection.
> > 
> > However, the hibmc VGA card (PCI ID 19e5:1711) has been tested and is
> > known to work when behind these bridges. (It does not require the
> > legacy resources to operate.)
> > 
> > Provide a quirk so that this combination of bridge and card is eligible
> > to be the default VGA card. This fixes Xorg auto-detection on the D05.
> > 
> > Cc: Xinliang Liu <z.liuxinliang at hisilicon.com>
> > Cc: Rongrong Zou <zourongrong at gmail.com>
> > Signed-off-by: Daniel Axtens <dja at axtens.net>
> > ---
> > 
> > v3: fix commit message
> > v4: fix comment (forgot to git add/git commit, sorry for the noise)
> > 
> > ---
> >  drivers/pci/quirks.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 50 insertions(+)
> > 
> > diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
> > index 16e6cd86ad71..b42324cba29e 100644
> > --- a/drivers/pci/quirks.c
> > +++ b/drivers/pci/quirks.c
> > @@ -25,6 +25,7 @@
> >  #include <linux/sched.h>
> >  #include <linux/ktime.h>
> >  #include <linux/mm.h>
> > +#include <linux/vgaarb.h>
> >  #include <asm/dma.h>	/* isa_dma_bridge_buggy */
> >  #include "pci.h"
> >  
> > @@ -4664,3 +4665,52 @@ static void quirk_intel_no_flr(struct pci_dev *dev)
> >  }
> >  DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, 0x1502, quirk_intel_no_flr);
> >  DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, 0x1503, quirk_intel_no_flr);
> > +
> > +/*
> > + * The HiSilicon D05 board has some PCI bridges (PCI ID 19e5:1610)
> > + * that are not spec-compliant: the VGA Enable bit is hardwired to 0
> > + * and writes do not change it. The bridge does not forward legacy
> > + * memory or I/O resources.
> > + *
> > + * Because the VGA Enable bit is hardwired to 0, the VGA arbiter
> > + * refuses to mark any card behind it as the boot device. However, the
> > + * hibmc VGA card (PCI ID 19e5:1711) has been tested and is known to
> > + * work when behind these bridges.
> > + *
> > + * If we have this bridge, this card, and no default card already,
> > + * mark the card as default.
> > + */
> > +static void hibmc_fixup_vgaarb(struct pci_dev *pdev)
> > +{
> > +	struct pci_dev *bridge;
> > +	struct pci_bus *bus;
> > +	u16 config;
> > +
> > +	bus = pdev->bus;
> > +	bridge = bus->self;
> > +	if (!bridge)
> > +		return;
> > +
> > +	if (!pci_is_bridge(bridge))
> > +		return;
> > +
> > +	if (bridge->vendor != PCI_VENDOR_ID_HUAWEI ||
> > +	    bridge->device != 0x1610)
> > +		return;
> > +
> > +	pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
> > +			     &config);
> > +	if (config & PCI_BRIDGE_CTL_VGA) {
> > +		/*
> > +		 * Weirdly, this bridge *is* spec compliant, so bail
> > +		 * and let vgaarb do its job
> > +		 */
> > +		return;
> > +	}
> > +
> > +	if (vga_default_device())
> > +		return;
> > +
> > +	vga_set_default_device(pdev);
> > +}
> > +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_HUAWEI, 0x1711, hibmc_fixup_vgaarb);
> 
> Is this quirk useful on any arch other than arm64?  Per
> drivers/pci/dwc/Kconfig, CONFIG_PCI_HISI depends on CONFIG_ARM64.
> 
> Would it make sense to put this quirk in arch/arm64/kernel/pci.c?

We've resisted adding PCI quirks in the arch directory so far, so I'd rather
it lived in the PCI code if you don't have a strong objection to that.
Hardware vendors have a tendency to reuse broken IP, so it's not beyond the
realms of imagination that this quirk might be needed for another part some
day (e.g. based on 32-bit ARM).

Will