| Message ID | 0-v1-f96b686c625b+124-pci_acs_quirk_fix_jgg@nvidia.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | f3c3ccc4fe49dbc560b01d16bebd1b116c46c2b4 |
| Delegated to: | Bjorn Helgaas |
| Headers | show |
| Series | [rc] PCI: Fix pci_enable_acs() support for the ACS quirks | expand |
On Wed, Oct 16, 2024 at 08:52:33PM -0300, Jason Gunthorpe wrote: > There are ACS quirks that hijack the normal ACS processing and deliver to > to special quirk code. The enable path needs to call > pci_dev_specific_enable_acs() and then pci_dev_specific_acs_enabled() will > report the hidden ACS state controlled by the quirk. > > The recent rework got this out of order and we should try to call > pci_dev_specific_enable_acs() regardless of any actual ACS support in the > device. > > As before command line parameters that effect standard PCI ACS don't > interact with the quirk versions, including the new config_acs= option. > > Fixes: 47c8846a49ba ("PCI: Extend ACS configurability") > Reported-by: Jiri Slaby <jirislaby@kernel.org> > Closes: https://lore.kernel.org/all/e89107da-ac99-4d3a-9527-a4df9986e120@kernel.org > Closes: https://bugzilla.suse.com/show_bug.cgi?id=1229019 > Tested-by: Steffen Dirkwinkel <me@steffen.cc> > Signed-off-by: Jason Gunthorpe <jgg@nvidia.com> > --- > drivers/pci/pci.c | 14 +++++++++----- > 1 file changed, 9 insertions(+), 5 deletions(-) Bjorn? This was a regression? Jason
On Wed, Oct 16, 2024 at 08:52:33PM -0300, Jason Gunthorpe wrote: > There are ACS quirks that hijack the normal ACS processing and deliver to > to special quirk code. The enable path needs to call > pci_dev_specific_enable_acs() and then pci_dev_specific_acs_enabled() will > report the hidden ACS state controlled by the quirk. > > The recent rework got this out of order and we should try to call > pci_dev_specific_enable_acs() regardless of any actual ACS support in the > device. > > As before command line parameters that effect standard PCI ACS don't > interact with the quirk versions, including the new config_acs= option. > > Fixes: 47c8846a49ba ("PCI: Extend ACS configurability") > Reported-by: Jiri Slaby <jirislaby@kernel.org> > Closes: https://lore.kernel.org/all/e89107da-ac99-4d3a-9527-a4df9986e120@kernel.org > Closes: https://bugzilla.suse.com/show_bug.cgi?id=1229019 > Tested-by: Steffen Dirkwinkel <me@steffen.cc> > Signed-off-by: Jason Gunthorpe <jgg@nvidia.com> Applied to pci/for-linus for v6.12, thanks for the ping. > --- > drivers/pci/pci.c | 14 +++++++++----- > 1 file changed, 9 insertions(+), 5 deletions(-) > > diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c > index 7d85c04fbba2ae..225a6cd2e9ca3b 100644 > --- a/drivers/pci/pci.c > +++ b/drivers/pci/pci.c > @@ -1067,8 +1067,15 @@ static void pci_std_enable_acs(struct pci_dev *dev, struct pci_acs *caps) > static void pci_enable_acs(struct pci_dev *dev) > { > struct pci_acs caps; > + bool enable_acs = false; > int pos; > > + /* If an iommu is present we start with kernel default caps */ > + if (pci_acs_enable) { > + if (pci_dev_specific_enable_acs(dev)) > + enable_acs = true; > + } > + > pos = dev->acs_cap; > if (!pos) > return; > @@ -1077,11 +1084,8 @@ static void pci_enable_acs(struct pci_dev *dev) > pci_read_config_word(dev, pos + PCI_ACS_CTRL, &caps.ctrl); > caps.fw_ctrl = caps.ctrl; > > - /* If an iommu is present we start with kernel default caps */ > - if (pci_acs_enable) { > - if (pci_dev_specific_enable_acs(dev)) > - pci_std_enable_acs(dev, &caps); > - } > + if (enable_acs) > + pci_std_enable_acs(dev, &caps); > > /* > * Always apply caps from the command line, even if there is no iommu. > > base-commit: dc5006cfcf62bea88076a587344ba5e00e66d1c6 > -- > 2.46.2 >
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c index 7d85c04fbba2ae..225a6cd2e9ca3b 100644 --- a/drivers/pci/pci.c +++ b/drivers/pci/pci.c @@ -1067,8 +1067,15 @@ static void pci_std_enable_acs(struct pci_dev *dev, struct pci_acs *caps) static void pci_enable_acs(struct pci_dev *dev) { struct pci_acs caps; + bool enable_acs = false; int pos; + /* If an iommu is present we start with kernel default caps */ + if (pci_acs_enable) { + if (pci_dev_specific_enable_acs(dev)) + enable_acs = true; + } + pos = dev->acs_cap; if (!pos) return; @@ -1077,11 +1084,8 @@ static void pci_enable_acs(struct pci_dev *dev) pci_read_config_word(dev, pos + PCI_ACS_CTRL, &caps.ctrl); caps.fw_ctrl = caps.ctrl; - /* If an iommu is present we start with kernel default caps */ - if (pci_acs_enable) { - if (pci_dev_specific_enable_acs(dev)) - pci_std_enable_acs(dev, &caps); - } + if (enable_acs) + pci_std_enable_acs(dev, &caps); /* * Always apply caps from the command line, even if there is no iommu.
There are ACS quirks that hijack the normal ACS processing and deliver to to special quirk code. The enable path needs to call pci_dev_specific_enable_acs() and then pci_dev_specific_acs_enabled() will report the hidden ACS state controlled by the quirk. The recent rework got this out of order and we should try to call pci_dev_specific_enable_acs() regardless of any actual ACS support in the device. As before command line parameters that effect standard PCI ACS don't interact with the quirk versions, including the new config_acs= option. Fixes: 47c8846a49ba ("PCI: Extend ACS configurability") Reported-by: Jiri Slaby <jirislaby@kernel.org> Closes: https://lore.kernel.org/all/e89107da-ac99-4d3a-9527-a4df9986e120@kernel.org Closes: https://bugzilla.suse.com/show_bug.cgi?id=1229019 Tested-by: Steffen Dirkwinkel <me@steffen.cc> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com> --- drivers/pci/pci.c | 14 +++++++++----- 1 file changed, 9 insertions(+), 5 deletions(-) base-commit: dc5006cfcf62bea88076a587344ba5e00e66d1c6