| Message ID | 20140827190125.29558.80166.stgit@gimli.home (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Aug 27, 2014 at 3:01 PM, Alex Williamson <alex.williamson@redhat.com> wrote: > If we assign a Radeon device to a virtual machine, we can no longer > assume a fixed hardware topology, like the GPU having a parent device. > This patch simply adds a few pci_is_root_bus() tests to avoid passing > a NULL pointer to PCI access functions, allowing the radeon driver to > work in a QEMU 440FX machine with an assigned HD8570 on the emulated > PCI root bus. > > Signed-off-by: Alex Williamson <alex.williamson@redhat.com> Added to my -fixes queue. thanks! Alex > --- > > drivers/gpu/drm/radeon/cik.c | 6 +++++- > drivers/gpu/drm/radeon/si.c | 6 +++++- > 2 files changed, 10 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/radeon/cik.c b/drivers/gpu/drm/radeon/cik.c > index 79a5a55..cc34308 100644 > --- a/drivers/gpu/drm/radeon/cik.c > +++ b/drivers/gpu/drm/radeon/cik.c > @@ -9555,6 +9555,9 @@ static void cik_pcie_gen3_enable(struct radeon_device *rdev) > int ret, i; > u16 tmp16; > > + if (pci_is_root_bus(rdev->pdev->bus)) > + return; > + > if (radeon_pcie_gen2 == 0) > return; > > @@ -9781,7 +9784,8 @@ static void cik_program_aspm(struct radeon_device *rdev) > if (orig != data) > WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, data); > > - if (!disable_clkreq) { > + if (!disable_clkreq && > + !pci_is_root_bus(rdev->pdev->bus)) { > struct pci_dev *root = rdev->pdev->bus->self; > u32 lnkcap; > > diff --git a/drivers/gpu/drm/radeon/si.c b/drivers/gpu/drm/radeon/si.c > index a1274a3..ffb1d4e 100644 > --- a/drivers/gpu/drm/radeon/si.c > +++ b/drivers/gpu/drm/radeon/si.c > @@ -7177,6 +7177,9 @@ static void si_pcie_gen3_enable(struct radeon_device *rdev) > int ret, i; > u16 tmp16; > > + if (pci_is_root_bus(rdev->pdev->bus)) > + return; > + > if (radeon_pcie_gen2 == 0) > return; > > @@ -7454,7 +7457,8 @@ static void si_program_aspm(struct radeon_device *rdev) > if (orig != data) > WREG32_PIF_PHY1(PB1_PIF_CNTL, data); > > - if (!disable_clkreq) { > + if (!disable_clkreq && > + !pci_is_root_bus(rdev->pdev->bus)) { > struct pci_dev *root = rdev->pdev->bus->self; > u32 lnkcap; > > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/dri-devel
> On Wed, Aug 27, 2014 at 3:01 PM, Alex Williamson > <alex.williamson@redhat.com> wrote: >> If we assign a Radeon device to a virtual machine, we can no longer >> assume a fixed hardware topology, like the GPU having a parent device. >> This patch simply adds a few pci_is_root_bus() tests to avoid passing >> a NULL pointer to PCI access functions, allowing the radeon driver to >> work in a QEMU 440FX machine with an assigned HD8570 on the emulated >> PCI root bus. >> >> Signed-off-by: Alex Williamson <alex.williamson@redhat.com> > Does this mean inside a VM we can't enable pcie 2 speeds? This could lead to a quite disappointing speed decrease for DMA transfers. Dave.
On Thu, 2014-08-28 at 11:32 +1000, Dave Airlie wrote: > > On Wed, Aug 27, 2014 at 3:01 PM, Alex Williamson > > <alex.williamson@redhat.com> wrote: > >> If we assign a Radeon device to a virtual machine, we can no longer > >> assume a fixed hardware topology, like the GPU having a parent device. > >> This patch simply adds a few pci_is_root_bus() tests to avoid passing > >> a NULL pointer to PCI access functions, allowing the radeon driver to > >> work in a QEMU 440FX machine with an assigned HD8570 on the emulated > >> PCI root bus. > >> > >> Signed-off-by: Alex Williamson <alex.williamson@redhat.com> > > > > Does this mean inside a VM we can't enable pcie 2 speeds? Shouldn't the device already be set to something reasonable before it's assigned? > This could lead to a quite disappointing speed decrease for DMA transfers. We couldn't do it before anyway, there's no decrease. Even with a Q35 QEMU chipset the best we were doing was tweaking dummy registers on the upstream port. Should VFIO be doing bus tuning prior to exposing the device? I suppose our other option would be to let the link control bits of the emulated downstream port pass through to the host. In any case, I don't see that this patch limits performance any more than it already is for a VM. Thanks, Alex
On Wed, Aug 27, 2014 at 9:32 PM, Dave Airlie <airlied@gmail.com> wrote: >> On Wed, Aug 27, 2014 at 3:01 PM, Alex Williamson >> <alex.williamson@redhat.com> wrote: >>> If we assign a Radeon device to a virtual machine, we can no longer >>> assume a fixed hardware topology, like the GPU having a parent device. >>> This patch simply adds a few pci_is_root_bus() tests to avoid passing >>> a NULL pointer to PCI access functions, allowing the radeon driver to >>> work in a QEMU 440FX machine with an assigned HD8570 on the emulated >>> PCI root bus. >>> >>> Signed-off-by: Alex Williamson <alex.williamson@redhat.com> >> > > Does this mean inside a VM we can't enable pcie 2 speeds? > > This could lead to a quite disappointing speed decrease for DMA transfers. It depends on the sbios, but most boards negotiate the highest speed at boot. Alex
diff --git a/drivers/gpu/drm/radeon/cik.c b/drivers/gpu/drm/radeon/cik.c index 79a5a55..cc34308 100644 --- a/drivers/gpu/drm/radeon/cik.c +++ b/drivers/gpu/drm/radeon/cik.c @@ -9555,6 +9555,9 @@ static void cik_pcie_gen3_enable(struct radeon_device *rdev) int ret, i; u16 tmp16; + if (pci_is_root_bus(rdev->pdev->bus)) + return; + if (radeon_pcie_gen2 == 0) return; @@ -9781,7 +9784,8 @@ static void cik_program_aspm(struct radeon_device *rdev) if (orig != data) WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, data); - if (!disable_clkreq) { + if (!disable_clkreq && + !pci_is_root_bus(rdev->pdev->bus)) { struct pci_dev *root = rdev->pdev->bus->self; u32 lnkcap; diff --git a/drivers/gpu/drm/radeon/si.c b/drivers/gpu/drm/radeon/si.c index a1274a3..ffb1d4e 100644 --- a/drivers/gpu/drm/radeon/si.c +++ b/drivers/gpu/drm/radeon/si.c @@ -7177,6 +7177,9 @@ static void si_pcie_gen3_enable(struct radeon_device *rdev) int ret, i; u16 tmp16; + if (pci_is_root_bus(rdev->pdev->bus)) + return; + if (radeon_pcie_gen2 == 0) return; @@ -7454,7 +7457,8 @@ static void si_program_aspm(struct radeon_device *rdev) if (orig != data) WREG32_PIF_PHY1(PB1_PIF_CNTL, data); - if (!disable_clkreq) { + if (!disable_clkreq && + !pci_is_root_bus(rdev->pdev->bus)) { struct pci_dev *root = rdev->pdev->bus->self; u32 lnkcap;
If we assign a Radeon device to a virtual machine, we can no longer assume a fixed hardware topology, like the GPU having a parent device. This patch simply adds a few pci_is_root_bus() tests to avoid passing a NULL pointer to PCI access functions, allowing the radeon driver to work in a QEMU 440FX machine with an assigned HD8570 on the emulated PCI root bus. Signed-off-by: Alex Williamson <alex.williamson@redhat.com> --- drivers/gpu/drm/radeon/cik.c | 6 +++++- drivers/gpu/drm/radeon/si.c | 6 +++++- 2 files changed, 10 insertions(+), 2 deletions(-)