| Message ID | 20181001124934.7222-1-thomas.petazzoni@bootlin.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | PCI: mvebu: Fix PCI I/O mapping creation sequence | expand |
> Just a rebuild-and-boot test. We don't use any PCIe devices > yet, but I'll grab an ugly eBay converter and try this patch > with a normal PCIe card later. The patch works. I can access an i210 over a miniPCIe -> PCIe adapter. With kind regards, Jan
On Mon, Oct 01, 2018 at 02:49:34PM +0200, Thomas Petazzoni wrote: > Commit ee1604381a371 ("PCI: mvebu: Only remap I/O space if > configured") had the side effect that the PCI I/O mapping was created > much earlier than before, at a point where the probe() of the driver > could still fail. This is for example a problem if one gets an > -EPROBE_DEFER at some point during probe(), after pci_ioremap_io() has > been called. > > Indeed, there is currently no function to undo what pci_ioremap_io() > did, and switching to pci_remap_iospace() is not an option in > pci-mvebu due to the need for special memory attributes on Armada 38x. > > Reverting ee1604381a371 ("PCI: mvebu: Only remap I/O space if > configured") would be a possibility, but it would require also > reverting 42342073e38b5 ("PCI: mvebu: Convert to use pci_host_bridge > directly"). So instead, we use an open-coded version of > pci_host_probe() that creates the PCI I/O mapping at a point where we > are guaranteed not to fail anymore. > > Reported-by: Jan Kundrát <jan.kundrat@cesnet.cz> > Fixes: ee1604381a371 ("PCI: mvebu: Only remap I/O space if configured") > Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com> Applied with Lorenzo's ack and Jan's tested-by to for-linux for v4.19, thanks! > --- > drivers/pci/controller/pci-mvebu.c | 52 +++++++++++++++++++++++++++++++++++--- > 1 file changed, 48 insertions(+), 4 deletions(-) > > diff --git a/drivers/pci/controller/pci-mvebu.c b/drivers/pci/controller/pci-mvebu.c > index 50eb0729385b..a41d79b8d46a 100644 > --- a/drivers/pci/controller/pci-mvebu.c > +++ b/drivers/pci/controller/pci-mvebu.c > @@ -1145,7 +1145,6 @@ static int mvebu_pcie_parse_request_resources(struct mvebu_pcie *pcie) > { > struct device *dev = &pcie->pdev->dev; > struct device_node *np = dev->of_node; > - unsigned int i; > int ret; > > INIT_LIST_HEAD(&pcie->resources); > @@ -1179,13 +1178,58 @@ static int mvebu_pcie_parse_request_resources(struct mvebu_pcie *pcie) > resource_size(&pcie->io) - 1); > pcie->realio.name = "PCI I/O"; > > + pci_add_resource(&pcie->resources, &pcie->realio); > + } > + > + return devm_request_pci_bus_resources(dev, &pcie->resources); > +} > + > +/* > + * This is a copy of pci_host_probe(), except that it does the I/O > + * remap as the last step, once we are sure we won't fail. > + * > + * It should be removed once the I/O remap error handling issue has > + * been sorted out. > + */ > +static int mvebu_pci_host_probe(struct pci_host_bridge *bridge) > +{ > + struct mvebu_pcie *pcie; > + struct pci_bus *bus, *child; > + int ret; > + > + ret = pci_scan_root_bus_bridge(bridge); > + if (ret < 0) { > + dev_err(bridge->dev.parent, "Scanning root bridge failed"); > + return ret; > + } > + > + pcie = pci_host_bridge_priv(bridge); > + if (resource_size(&pcie->io) != 0) { > + unsigned int i; > + > for (i = 0; i < resource_size(&pcie->realio); i += SZ_64K) > pci_ioremap_io(i, pcie->io.start + i); > + } > > - pci_add_resource(&pcie->resources, &pcie->realio); > + bus = bridge->bus; > + > + /* > + * We insert PCI resources into the iomem_resource and > + * ioport_resource trees in either pci_bus_claim_resources() > + * or pci_bus_assign_resources(). > + */ > + if (pci_has_flag(PCI_PROBE_ONLY)) { > + pci_bus_claim_resources(bus); > + } else { > + pci_bus_size_bridges(bus); > + pci_bus_assign_resources(bus); > + > + list_for_each_entry(child, &bus->children, node) > + pcie_bus_configure_settings(child); > } > > - return devm_request_pci_bus_resources(dev, &pcie->resources); > + pci_bus_add_devices(bus); > + return 0; > } > > static int mvebu_pcie_probe(struct platform_device *pdev) > @@ -1268,7 +1312,7 @@ static int mvebu_pcie_probe(struct platform_device *pdev) > bridge->align_resource = mvebu_pcie_align_resource; > bridge->msi = pcie->msi; > > - return pci_host_probe(bridge); > + return mvebu_pci_host_probe(bridge); > } > > static const struct of_device_id mvebu_pcie_of_match_table[] = { > -- > 2.14.4 >
diff --git a/drivers/pci/controller/pci-mvebu.c b/drivers/pci/controller/pci-mvebu.c index 50eb0729385b..a41d79b8d46a 100644 --- a/drivers/pci/controller/pci-mvebu.c +++ b/drivers/pci/controller/pci-mvebu.c @@ -1145,7 +1145,6 @@ static int mvebu_pcie_parse_request_resources(struct mvebu_pcie *pcie) { struct device *dev = &pcie->pdev->dev; struct device_node *np = dev->of_node; - unsigned int i; int ret; INIT_LIST_HEAD(&pcie->resources); @@ -1179,13 +1178,58 @@ static int mvebu_pcie_parse_request_resources(struct mvebu_pcie *pcie) resource_size(&pcie->io) - 1); pcie->realio.name = "PCI I/O"; + pci_add_resource(&pcie->resources, &pcie->realio); + } + + return devm_request_pci_bus_resources(dev, &pcie->resources); +} + +/* + * This is a copy of pci_host_probe(), except that it does the I/O + * remap as the last step, once we are sure we won't fail. + * + * It should be removed once the I/O remap error handling issue has + * been sorted out. + */ +static int mvebu_pci_host_probe(struct pci_host_bridge *bridge) +{ + struct mvebu_pcie *pcie; + struct pci_bus *bus, *child; + int ret; + + ret = pci_scan_root_bus_bridge(bridge); + if (ret < 0) { + dev_err(bridge->dev.parent, "Scanning root bridge failed"); + return ret; + } + + pcie = pci_host_bridge_priv(bridge); + if (resource_size(&pcie->io) != 0) { + unsigned int i; + for (i = 0; i < resource_size(&pcie->realio); i += SZ_64K) pci_ioremap_io(i, pcie->io.start + i); + } - pci_add_resource(&pcie->resources, &pcie->realio); + bus = bridge->bus; + + /* + * We insert PCI resources into the iomem_resource and + * ioport_resource trees in either pci_bus_claim_resources() + * or pci_bus_assign_resources(). + */ + if (pci_has_flag(PCI_PROBE_ONLY)) { + pci_bus_claim_resources(bus); + } else { + pci_bus_size_bridges(bus); + pci_bus_assign_resources(bus); + + list_for_each_entry(child, &bus->children, node) + pcie_bus_configure_settings(child); } - return devm_request_pci_bus_resources(dev, &pcie->resources); + pci_bus_add_devices(bus); + return 0; } static int mvebu_pcie_probe(struct platform_device *pdev) @@ -1268,7 +1312,7 @@ static int mvebu_pcie_probe(struct platform_device *pdev) bridge->align_resource = mvebu_pcie_align_resource; bridge->msi = pcie->msi; - return pci_host_probe(bridge); + return mvebu_pci_host_probe(bridge); } static const struct of_device_id mvebu_pcie_of_match_table[] = {
Commit ee1604381a371 ("PCI: mvebu: Only remap I/O space if configured") had the side effect that the PCI I/O mapping was created much earlier than before, at a point where the probe() of the driver could still fail. This is for example a problem if one gets an -EPROBE_DEFER at some point during probe(), after pci_ioremap_io() has been called. Indeed, there is currently no function to undo what pci_ioremap_io() did, and switching to pci_remap_iospace() is not an option in pci-mvebu due to the need for special memory attributes on Armada 38x. Reverting ee1604381a371 ("PCI: mvebu: Only remap I/O space if configured") would be a possibility, but it would require also reverting 42342073e38b5 ("PCI: mvebu: Convert to use pci_host_bridge directly"). So instead, we use an open-coded version of pci_host_probe() that creates the PCI I/O mapping at a point where we are guaranteed not to fail anymore. Reported-by: Jan Kundrát <jan.kundrat@cesnet.cz> Fixes: ee1604381a371 ("PCI: mvebu: Only remap I/O space if configured") Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com> --- drivers/pci/controller/pci-mvebu.c | 52 +++++++++++++++++++++++++++++++++++--- 1 file changed, 48 insertions(+), 4 deletions(-)