diff mbox

pci: Add a acs_disable option for pci kernel parameter

Message ID 1509071869-18771-1-git-send-email-sbates@raithlin.com (mailing list archive)
State New, archived
Delegated to: Bjorn Helgaas
Headers show

Commit Message

Stephen Bates Oct. 27, 2017, 2:37 a.m. UTC
From: Stephen Bates <sbates@raithlin.com>

On some servers the BIOS sets up ACS on any valid pci_dev in the
system. The kernel has no way of backing this out since the kernel
only turns ACS capabilities on.

This patch adds a new boot option to the pci kernel parameter called
"acs_disable" that will disable ACS. This is useful for PCI peer to
peer communication but can cause problems when IOVA isolation is
required and an IOMMU is enabled. Use with care.

Signed-off-by: Stephen Bates <sbates@raithlin.com>
---
 Documentation/admin-guide/kernel-parameters.txt |  4 ++++
 drivers/pci/pci.c                               | 23 ++++++++++++++++++++---
 drivers/pci/pci.h                               |  2 +-
 drivers/pci/probe.c                             |  4 ++--
 4 files changed, 27 insertions(+), 6 deletions(-)

Comments

Christoph Hellwig Oct. 27, 2017, 6:37 a.m. UTC | #1
On Thu, Oct 26, 2017 at 08:37:49PM -0600, sbates@raithlin.com wrote:
> From: Stephen Bates <sbates@raithlin.com>
> 
> On some servers the BIOS sets up ACS on any valid pci_dev in the
> system. The kernel has no way of backing this out since the kernel
> only turns ACS capabilities on.
> 
> This patch adds a new boot option to the pci kernel parameter called
> "acs_disable" that will disable ACS. This is useful for PCI peer to
> peer communication but can cause problems when IOVA isolation is
> required and an IOMMU is enabled. Use with care.

Eww.  Can we please add smbios quirks for the systems where you've
observed this? (we probably also want to keep the option just in case).
Stephen Bates Oct. 30, 2017, 1:04 a.m. UTC | #2
>> This patch adds a new boot option to the pci kernel parameter called

>> "acs_disable" that will disable ACS. This is useful for PCI peer to

>> peer communication but can cause problems when IOVA isolation is

>> required and an IOMMU is enabled. Use with care.


> Eww.


Thanks for the feedback Christoph. My sentiments exactly ;-).  

> Can we please add smbios quirks for the systems where you've

> observed this? 


I can look at doing this. The issue with this approach is that it will require a kernel patch for each new system that is detected.

> we probably also want to keep the option just in case).


Agreed. At least then operators have a path to ACS disable prior to a quirk being accepted.

Stephen
diff mbox

Patch

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 0549662..695eb12 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -2907,6 +2907,10 @@ 
 		earlydump	[X86] dump PCI config space before the kernel
 			        changes anything
 		off		[X86] don't probe for the PCI bus
+		acs_disable     [PCIE] disable access control services. Note
+				this can interfere with IOVA isolation if an IOMMU
+				is enabled but can be necessary when doing PCI
+				peer to peer communication. Use with care.
 		bios		[X86-32] force use of PCI BIOS, don't access
 				the hardware directly. Use this if your machine
 				has a non-standard PCI host bridge.
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index 6078dfc..ce33608 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -110,6 +110,9 @@  unsigned int pcibios_max_latency = 255;
 /* If set, the PCIe ARI capability will not be used. */
 static bool pcie_ari_disabled;
 
+/* If set, the PCIe ACS capability will be disabled. */
+static bool pci_acs_disable;
+
 /* Disable bridge_d3 for all PCIe ports */
 static bool pci_bridge_d3_disable;
 /* Force bridge_d3 for all PCIe ports */
@@ -1182,7 +1185,7 @@  void pci_restore_state(struct pci_dev *dev)
 	pci_restore_msi_state(dev);
 
 	/* Restore ACS and IOV configuration state */
-	pci_enable_acs(dev);
+	pci_config_acs(dev);
 	pci_restore_iov_state(dev);
 
 	dev->state_saved = false;
@@ -2821,11 +2824,23 @@  static void pci_std_enable_acs(struct pci_dev *dev)
 }
 
 /**
- * pci_enable_acs - enable ACS if hardware support it
+ * pci_config_acs - configure ACS
  * @dev: the PCI device
  */
-void pci_enable_acs(struct pci_dev *dev)
+void pci_config_acs(struct pci_dev *dev)
 {
+	int pos;
+
+	if (pci_acs_disable) {
+		pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
+		if (!pos)
+			return;
+
+		dev_warn_ratelimited(&dev->dev,
+				     "disabling ACS via pci_acs_disable\n");
+		pci_write_config_word(dev, pos + PCI_ACS_CTRL, 0);
+	}
+
 	if (!pci_acs_enable)
 		return;
 
@@ -5471,6 +5486,8 @@  static int __init pci_setup(char *str)
 		if (*str && (str = pcibios_setup(str)) && *str) {
 			if (!strcmp(str, "nomsi")) {
 				pci_no_msi();
+			} else if (!strcmp(str, "acs_disable")) {
+				pci_acs_disable = true;
 			} else if (!strcmp(str, "noaer")) {
 				pci_no_aer();
 			} else if (!strncmp(str, "realloc=", 8)) {
diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index a6560c9..16d94d3 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -338,7 +338,7 @@  static inline resource_size_t pci_resource_alignment(struct pci_dev *dev,
 	return resource_alignment(res);
 }
 
-void pci_enable_acs(struct pci_dev *dev);
+void pci_config_acs(struct pci_dev *dev);
 
 #ifdef CONFIG_PCIE_PTM
 void pci_ptm_init(struct pci_dev *dev);
diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index ff94b69..86c3299 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -2020,8 +2020,8 @@  static void pci_init_capabilities(struct pci_dev *dev)
 	/* Address Translation Services */
 	pci_ats_init(dev);
 
-	/* Enable ACS P2P upstream forwarding */
-	pci_enable_acs(dev);
+	/* Configure ACS P2P upstream forwarding */
+	pci_config_acs(dev);
 
 	/* Precision Time Measurement */
 	pci_ptm_init(dev);