PCI: Turn off Request Attributes to avoid Chelsio T5 Completion erratum

The Chelsio T5 has a PCIe compliance erratum that causes Malformed TLP or
Unexpected Completion errors in some systems, which may cause device access
timeouts.

Per PCIe r3.0, sec 2.2.9, "Completion headers must supply the same values
for the Attribute as were supplied in the header of the corresponding
Request, except as explicitly allowed when IDO is used."

Instead of copying the Attributes from the Request to the Completion, the
T5 always generates Completions with zero Attributes.  The receiver of a
Completion whose Attributes don't match the Request may accept it (which
itself seems non-compliant based on sec 2.3.2), or it may handle it as a
Malformed TLP or an Unexpected Completion, which will probably lead to a
device access timeout.

Work around this by disabling "Relaxed Ordering" and "No Snoop" in the Root
Port so it always generate Requests with zero Attributes.

This does affect all other devices which are downstream of that Root Port,
but these are performance optimizations that should not make a functional
difference.

Note that Configuration Space accesses are never supposed to have TLP
Attributes, so we're safe waiting till after any Configuration Space
accesses to do the Root Port "fixup".

Based on original work by Casey Leedom <leedom@chelsio.com>

[bhelgaas: changelog, comments, rename to pci_find_pcie_root_port(), rework
to use pci_upstream_bridge() and check for Root Port device type, edit
diagnostics to clarify intent and devices affected]
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>

1 files changed, 57 insertions, 0 deletions

diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
index 6a30252cd79f..eb3c98e0fb2b 100644
--- a/drivers/pci/quirks.c
+++ b/drivers/pci/quirks.c
@@ -3692,6 +3692,63 @@ DECLARE_PCI_FIXUP_CLASS_EARLY(0x1797, 0x6869, PCI_CLASS_NOT_DEFINED, 8,
 			      quirk_tw686x_class);
 
 /*
+ * Per PCIe r3.0, sec 2.2.9, "Completion headers must supply the same
+ * values for the Attribute as were supplied in the header of the
+ * corresponding Request, except as explicitly allowed when IDO is used."
+ *
+ * If a non-compliant device generates a completion with a different
+ * attribute than the request, the receiver may accept it (which itself
+ * seems non-compliant based on sec 2.3.2), or it may handle it as a
+ * Malformed TLP or an Unexpected Completion, which will probably lead to a
+ * device access timeout.
+ *
+ * If the non-compliant device generates completions with zero attributes
+ * (instead of copying the attributes from the request), we can work around
+ * this by disabling the "Relaxed Ordering" and "No Snoop" attributes in
+ * upstream devices so they always generate requests with zero attributes.
+ *
+ * This affects other devices under the same Root Port, but since these
+ * attributes are performance hints, there should be no functional problem.
+ *
+ * Note that Configuration Space accesses are never supposed to have TLP
+ * Attributes, so we're safe waiting till after any Configuration Space
+ * accesses to do the Root Port fixup.
+ */
+static void quirk_disable_root_port_attributes(struct pci_dev *pdev)
+{
+	struct pci_dev *root_port = pci_find_pcie_root_port(pdev);
+
+	if (!root_port) {
+		dev_warn(&pdev->dev, "PCIe Completion erratum may cause device errors\n");
+		return;
+	}
+
+	dev_info(&root_port->dev, "Disabling No Snoop/Relaxed Ordering Attributes to avoid PCIe Completion erratum in %s\n",
+		 dev_name(&pdev->dev));
+	pcie_capability_clear_and_set_word(root_port, PCI_EXP_DEVCTL,
+					   PCI_EXP_DEVCTL_RELAX_EN |
+					   PCI_EXP_DEVCTL_NOSNOOP_EN, 0);
+}
+
+/*
+ * The Chelsio T5 chip fails to copy TLP Attributes from a Request to the
+ * Completion it generates.
+ */
+static void quirk_chelsio_T5_disable_root_port_attributes(struct pci_dev *pdev)
+{
+	/*
+	 * This mask/compare operation selects for Physical Function 4 on a
+	 * T5.  We only need to fix up the Root Port once for any of the
+	 * PFs.  PF[0..3] have PCI Device IDs of 0x50xx, but PF4 is uniquely
+	 * 0x54xx so we use that one,
+	 */
+	if ((pdev->device & 0xff00) == 0x5400)
+		quirk_disable_root_port_attributes(pdev);
+}
+DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_CHELSIO, PCI_ANY_ID,
+			 quirk_chelsio_T5_disable_root_port_attributes);
+
+/*
  * AMD has indicated that the devices below do not support peer-to-peer
  * in any system where they are found in the southbridge with an AMD
  * IOMMU in the system.  Multifunction devices that do not support