Message ID | 1418929903-8506-2-git-send-email-l.stach@pengutronix.de |
---|---|
State | Superseded |
Headers | show |
On Thu, Dec 18, 2014 at 08:11:43PM +0100, Lucas Stach wrote: > The fixup to enable relaxed ordering on all PCI devices was > executed unconditionally if the Tegra PCI host driver was > built into the kernel. This doesn't play nice with a > multiplatform kernel executed on other platforms which > may not need this fixup. > > Make sure to only apply the fixup if the root port is > a Tegra. > > Signed-off-by: Lucas Stach <l.stach@pengutronix.de> > --- > v2: > - split out PCI hierarchy walk > - separate code from data by moving PCI IDs into own structure > --- > drivers/pci/host/pci-tegra.c | 34 +++++++++++++++++++++++++++++++++- > 1 file changed, 33 insertions(+), 1 deletion(-) > > diff --git a/drivers/pci/host/pci-tegra.c b/drivers/pci/host/pci-tegra.c > index 333a57afacc4..b77f417e1a3c 100644 > --- a/drivers/pci/host/pci-tegra.c > +++ b/drivers/pci/host/pci-tegra.c > @@ -635,10 +635,42 @@ DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0bf1, tegra_pcie_fixup_class); > DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0e1c, tegra_pcie_fixup_class); > DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0e1d, tegra_pcie_fixup_class); > > +static const struct pci_device_id tegra_rootport_ids[] = { > + { > + /* Tegra20 4 lane root port */ > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0bf0, > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > + }, { > + /* Tegra20 2 lane root port */ > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0bf1, > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID The number of lanes is configurable, so I'm not sure exactly what this comment is supposed to indicate. Are you saying that port 0 has 0x0bf0 and port 1 has 0x0bf1 as device IDs. > + }, { > + /* Tegra30 4 lane root port */ > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e1c, > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > + }, { > + /* Tegra30 2 lane root port */ > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e1d, > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID Tegra30 has three ports, so does this second entry (0x0e1d) apply to ports 1 and 2, whereas the previous entry (0x0e1c) applies only to port 0? > + }, { > + /* Tegra124 4 lane root port */ > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e12, > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > + }, { > + /* Tegra124 1 lane root port */ > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e13, > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID Or perhaps what this signifies is that the first port is actually a different device because it supports up to 4 lanes, whereas the others support up to 2 lanes (or only 1 on Tegra124)? In that case: Acked-by: Thierry Reding <treding@nvidia.com>
Am Freitag, den 09.01.2015, 12:32 +0100 schrieb Thierry Reding: > On Thu, Dec 18, 2014 at 08:11:43PM +0100, Lucas Stach wrote: > > The fixup to enable relaxed ordering on all PCI devices was > > executed unconditionally if the Tegra PCI host driver was > > built into the kernel. This doesn't play nice with a > > multiplatform kernel executed on other platforms which > > may not need this fixup. > > > > Make sure to only apply the fixup if the root port is > > a Tegra. > > > > Signed-off-by: Lucas Stach <l.stach@pengutronix.de> > > --- > > v2: > > - split out PCI hierarchy walk > > - separate code from data by moving PCI IDs into own structure > > --- > > drivers/pci/host/pci-tegra.c | 34 +++++++++++++++++++++++++++++++++- > > 1 file changed, 33 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/pci/host/pci-tegra.c b/drivers/pci/host/pci-tegra.c > > index 333a57afacc4..b77f417e1a3c 100644 > > --- a/drivers/pci/host/pci-tegra.c > > +++ b/drivers/pci/host/pci-tegra.c > > @@ -635,10 +635,42 @@ DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0bf1, tegra_pcie_fixup_class); > > DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0e1c, tegra_pcie_fixup_class); > > DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0e1d, tegra_pcie_fixup_class); > > > > +static const struct pci_device_id tegra_rootport_ids[] = { > > + { > > + /* Tegra20 4 lane root port */ > > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0bf0, > > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > > + }, { > > + /* Tegra20 2 lane root port */ > > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0bf1, > > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > > The number of lanes is configurable, so I'm not sure exactly what this > comment is supposed to indicate. Are you saying that port 0 has 0x0bf0 > and port 1 has 0x0bf1 as device IDs. > No, the device ID of the root port is dependent on the number of lanes configured for the specific port. So if you have a 4x1 configuration you will get to see a single device with ID 0x0bf0, in a 2x2 configuration you will see 2 devices with ID 0x0bf1. > > + }, { > > + /* Tegra30 4 lane root port */ > > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e1c, > > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > > + }, { > > + /* Tegra30 2 lane root port */ > > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e1d, > > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > > Tegra30 has three ports, so does this second entry (0x0e1d) apply to > ports 1 and 2, whereas the previous entry (0x0e1c) applies only to port > 0? > > > + }, { > > + /* Tegra124 4 lane root port */ > > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e12, > > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > > + }, { > > + /* Tegra124 1 lane root port */ > > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e13, > > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > > Or perhaps what this signifies is that the first port is actually a > different device because it supports up to 4 lanes, whereas the others > support up to 2 lanes (or only 1 on Tegra124)? In that case: > > Acked-by: Thierry Reding <treding@nvidia.com> -- 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, Jan 09, 2015 at 12:45:18PM +0100, Lucas Stach wrote: > Am Freitag, den 09.01.2015, 12:32 +0100 schrieb Thierry Reding: > > On Thu, Dec 18, 2014 at 08:11:43PM +0100, Lucas Stach wrote: > > > The fixup to enable relaxed ordering on all PCI devices was > > > executed unconditionally if the Tegra PCI host driver was > > > built into the kernel. This doesn't play nice with a > > > multiplatform kernel executed on other platforms which > > > may not need this fixup. > > > > > > Make sure to only apply the fixup if the root port is > > > a Tegra. > > > > > > Signed-off-by: Lucas Stach <l.stach@pengutronix.de> > > > --- > > > v2: > > > - split out PCI hierarchy walk > > > - separate code from data by moving PCI IDs into own structure > > > --- > > > drivers/pci/host/pci-tegra.c | 34 +++++++++++++++++++++++++++++++++- > > > 1 file changed, 33 insertions(+), 1 deletion(-) > > > > > > diff --git a/drivers/pci/host/pci-tegra.c b/drivers/pci/host/pci-tegra.c > > > index 333a57afacc4..b77f417e1a3c 100644 > > > --- a/drivers/pci/host/pci-tegra.c > > > +++ b/drivers/pci/host/pci-tegra.c > > > @@ -635,10 +635,42 @@ DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0bf1, tegra_pcie_fixup_class); > > > DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0e1c, tegra_pcie_fixup_class); > > > DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0e1d, tegra_pcie_fixup_class); > > > > > > +static const struct pci_device_id tegra_rootport_ids[] = { > > > + { > > > + /* Tegra20 4 lane root port */ > > > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0bf0, > > > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > > > + }, { > > > + /* Tegra20 2 lane root port */ > > > + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0bf1, > > > + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID > > > > The number of lanes is configurable, so I'm not sure exactly what this > > comment is supposed to indicate. Are you saying that port 0 has 0x0bf0 > > and port 1 has 0x0bf1 as device IDs. > > > > No, the device ID of the root port is dependent on the number of lanes > configured for the specific port. So if you have a 4x1 configuration you > will get to see a single device with ID 0x0bf0, in a 2x2 configuration > you will see 2 devices with ID 0x0bf1. Okay, that's interesting to know. My ack remains valid in that case: Acked-by: Thierry Reding <treding@nvidia.com>
diff --git a/drivers/pci/host/pci-tegra.c b/drivers/pci/host/pci-tegra.c index 333a57afacc4..b77f417e1a3c 100644 --- a/drivers/pci/host/pci-tegra.c +++ b/drivers/pci/host/pci-tegra.c @@ -635,10 +635,42 @@ DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0bf1, tegra_pcie_fixup_class); DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0e1c, tegra_pcie_fixup_class); DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_NVIDIA, 0x0e1d, tegra_pcie_fixup_class); +static const struct pci_device_id tegra_rootport_ids[] = { + { + /* Tegra20 4 lane root port */ + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0bf0, + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID + }, { + /* Tegra20 2 lane root port */ + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0bf1, + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID + }, { + /* Tegra30 4 lane root port */ + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e1c, + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID + }, { + /* Tegra30 2 lane root port */ + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e1d, + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID + }, { + /* Tegra124 4 lane root port */ + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e12, + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID + }, { + /* Tegra124 1 lane root port */ + .vendor = PCI_VENDOR_ID_NVIDIA, .device = 0x0e13, + .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID + }, { + /* sentinel */ + } +}; + /* Tegra PCIE requires relaxed ordering */ static void tegra_pcie_relax_enable(struct pci_dev *dev) { - pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN); + if (pci_match_id(tegra_rootport_ids, pci_get_rootport(dev))) + pcie_capability_set_word(dev, PCI_EXP_DEVCTL, + PCI_EXP_DEVCTL_RELAX_EN); } DECLARE_PCI_FIXUP_FINAL(PCI_ANY_ID, PCI_ANY_ID, tegra_pcie_relax_enable);
The fixup to enable relaxed ordering on all PCI devices was executed unconditionally if the Tegra PCI host driver was built into the kernel. This doesn't play nice with a multiplatform kernel executed on other platforms which may not need this fixup. Make sure to only apply the fixup if the root port is a Tegra. Signed-off-by: Lucas Stach <l.stach@pengutronix.de> --- v2: - split out PCI hierarchy walk - separate code from data by moving PCI IDs into own structure --- drivers/pci/host/pci-tegra.c | 34 +++++++++++++++++++++++++++++++++- 1 file changed, 33 insertions(+), 1 deletion(-)