On 6/29/2018 8:49 PM, Bjorn Helgaas wrote:

On Tue, Jun 19, 2018 at 10:14:46PM -0400, Sinan Kaya wrote:

quoted

A PCIe endpoint carries the process address space identifier (PASID) in
the TLP prefix as part of the memory read/write transaction. The address
information in the TLP is relevant only for a given PASID context.

An IOMMU takes PASID value and the address information from the
TLP to look up the physical address in the system.

If a bridge drops the TLP prefix, the translation agent can resolve the
address to an incorrect location and cause data corruption. Prevent
this condition by requiring End-to-End TLP prefix to be supported on the
entire data path between the endpoint and the root port.

PASID is an End-End TLP Prefix (PCIe r4.0, sec 6.20).  Sec 2.2.10.2 says

  It is an error to receive a TLP with an End-End TLP Prefix by a
  Receiver that does not support End-End TLP Prefixes. A TLP in
  violation of this rule is handled as a Malformed TLP. This is a
  reported error associated with the Receiving Port (see Section 6.2).

So I agree that we shouldn't enable PASID in an endpoint unless all
the switch ports leading to it support End-End prefixes.  But I don't
see how a bridge can drop a prefix and cause data corruption -- if it
doesn't support End-End prefixes, shouldn't the bridge raise a
Malformed TLP error instead of forwarding the TLP?

It should under normal circumstances. 

I remember reading that most PCIe switches don't support TLP prefixes.
I don't know if it is because of buggy behavior or if it is just plain
unsupported while dropping the request as Malformed TLP.

I was trying to be proactive and not enable PASID if the entire path
is incapable.

quoted

Signed-off-by: Sinan Kaya <redacted>
---
 drivers/pci/ats.c             |  9 +++++++++
 drivers/pci/probe.c           | 17 +++++++++++++++++
 include/linux/pci.h           |  1 +
 include/uapi/linux/pci_regs.h |  1 +
 4 files changed, 28 insertions(+)

diff --git a/drivers/pci/ats.c b/drivers/pci/ats.c
index 4923a2a..e1b2e6d 100644
--- a/drivers/pci/ats.c
+++ b/drivers/pci/ats.c

@@ -268,6 +268,7 @@ EXPORT_SYMBOL_GPL(pci_reset_pri);
 int pci_enable_pasid(struct pci_dev *pdev, int features)
 {
 	u16 control, supported;
+	struct pci_dev *bridge;
 	int pos;
 
 	if (WARN_ON(pdev->pasid_enabled))

@@ -277,6 +278,14 @@ int pci_enable_pasid(struct pci_dev *pdev, int features)
 	if (!pos)
 		return -EINVAL;
 
+	bridge = pci_upstream_bridge(pdev);
+	while (bridge) {
+		if (!bridge->eetlp_prefix)
+			return -EINVAL;
+
+		bridge = pci_upstream_bridge(bridge);
+	}

I was hoping to avoid even this loop by having the eetlp_prefix bit
indicate that "End-End TLP Prefixes are supported from the Root Port
to here".

I see.

quoted

 	pci_read_config_word(pdev, pos + PCI_PASID_CAP, &supported);
 	supported &= PCI_PASID_CAP_EXEC | PCI_PASID_CAP_PRIV;
 

diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index ac876e3..a7f7ac1 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c

@@ -2042,6 +2042,22 @@ static void pci_configure_ltr(struct pci_dev *dev)
 #endif
 }
 
+static void pci_configure_eetlp_prefix(struct pci_dev *dev)
+{
+#ifdef CONFIG_PCI_PASID
+	u32 cap;
+
+	if (!pci_is_pcie(dev))
+		return;
+
+	pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap);
+	if (!(cap & PCI_EXP_DEVCAP2_E2ETLP))
+		return;
+
+	dev->eetlp_prefix = 1;

I.e., here we would do:

  if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
    dev->eetlp_prefix_path = 1;
  else {
    bridge = pci_upstream_bridge(dev);
    if (bridge && bridge->eetlp_prefix_path)
      dev->eetlp_prefix_path = 1;
  }

Sure, let me make the changes and post a new version.

quoted

+#endif
+}
+
 static void pci_configure_device(struct pci_dev *dev)
 {
 	struct hotplug_params hpp;

@@ -2051,6 +2067,7 @@ static void pci_configure_device(struct pci_dev *dev)
 	pci_configure_extended_tags(dev, NULL);
 	pci_configure_relaxed_ordering(dev);
 	pci_configure_ltr(dev);
+	pci_configure_eetlp_prefix(dev);
 
 	memset(&hpp, 0, sizeof(hpp));
 	ret = pci_get_hp_params(dev, &hpp);

diff --git a/include/linux/pci.h b/include/linux/pci.h
index 340029b..cf88d47 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h

@@ -350,6 +350,7 @@ struct pci_dev {
 	unsigned int	ltr_path:1;	/* Latency Tolerance Reporting
 					   supported from root to here */
 #endif
+	unsigned int	eetlp_prefix:1;	/* End-to-End TLP Prefix */
 
 	pci_channel_state_t error_state;	/* Current connectivity state */
 	struct device	dev;			/* Generic device interface */

diff --git a/include/uapi/linux/pci_regs.h b/include/uapi/linux/pci_regs.h
index 4da87e2..a617ab2 100644
--- a/include/uapi/linux/pci_regs.h
+++ b/include/uapi/linux/pci_regs.h

@@ -636,6 +636,7 @@
 #define  PCI_EXP_DEVCAP2_OBFF_MASK	0x000c0000 /* OBFF support mechanism */
 #define  PCI_EXP_DEVCAP2_OBFF_MSG	0x00040000 /* New message signaling */
 #define  PCI_EXP_DEVCAP2_OBFF_WAKE	0x00080000 /* Re-use WAKE# for OBFF */
+#define PCI_EXP_DEVCAP2_E2ETLP		0x00200000 /* End-to-End TLP Prefix */

It looks like lspci doesn't decode this bit (and several others in
DevCap2).  Would you be interested in adding that?  The source is at
git://git.kernel.org/pub/scm/utils/pciutils/pciutils.git

quoted

 #define PCI_EXP_DEVCTL2		40	/* Device Control 2 */
 #define  PCI_EXP_DEVCTL2_COMP_TIMEOUT	0x000f	/* Completion Timeout Value */
 #define  PCI_EXP_DEVCTL2_COMP_TMOUT_DIS	0x0010	/* Completion Timeout Disable */
-- 
2.7.4


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.