Skip to contentSkip to conversation

iommu/vt-d: Don't apply gfx quirks to untrusted devices

PATCH Superseded Rajat Jain 4 participants 5 messages Jun 2, 2020 IOMMU

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Rajat Jain

author
#0 lore raw

PATCH 1/1 iommu/vt-d: Don't apply gfx quirks to untrusted devices

Currently, an external malicious PCI device can masquerade the VID:PID of faulty gfx devices, and thus apply iommu quirks to effectively disable the IOMMU restrictions for itself.

Thus we need to ensure that the device we are applying quirks to, is indeed an internal trusted device.

---
 drivers/iommu/intel-iommu.c | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)
drivers/iommu/intel-iommu.c +28 -0
index ef0a5246700e5..f2a480168a02f 100644
@@ -6214,6 +6214,11 @@ const struct iommu_ops intel_iommu_ops = {
 
 static void quirk_iommu_igfx(struct pci_dev *dev)
 {
+	if (dev->untrusted) {
+		pci_warn(dev, "skipping iommu quirk for untrusted gfx dev\n");
+		return;
+	}
+
 	pci_info(dev, "Disabling IOMMU for graphics on this chipset\n");
 	dmar_map_gfx = 0;
 }
@@ -6255,6 +6260,11 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x163D, quirk_iommu_igfx);
 
 static void quirk_iommu_rwbf(struct pci_dev *dev)
 {
+	if (dev->untrusted) {
+		pci_warn(dev, "skipping iommu quirk for untrusted dev\n");
+		return;
+	}
+
 	/*
 	 * Mobile 4 Series Chipset neglects to set RWBF capability,
 	 * but needs it. Same seems to hold for the desktop versions.
@@ -6285,6 +6295,11 @@ static void quirk_calpella_no_shadow_gtt(struct pci_dev *dev)
 {
 	unsigned short ggc;
 
+	if (dev->untrusted) {
+		pci_warn(dev, "skipping iommu quirk for untrusted gfx dev\n");
+		return;
+	}
+
 	if (pci_read_config_word(dev, GGC, &ggc))
 		return;
 
@@ -6318,6 +6333,13 @@ static void __init check_tylersburg_isoch(void)
 	pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 0x3a3e, NULL);
 	if (!pdev)
 		return;
+
+	if (pdev->untrusted) {
+		pci_warn(pdev, "skipping iommu quirk due to untrusted dev\n");
+		pci_dev_put(pdev);
+		return;
+	}
+
 	pci_dev_put(pdev);
 
 	/* System Management Registers. Might be hidden, in which case
@@ -6327,6 +6349,12 @@ static void __init check_tylersburg_isoch(void)
 	if (!pdev)
 		return;
 
+	if (pdev->untrusted) {
+		pci_warn(pdev, "skipping iommu quirk due to untrusted dev\n");
+		pci_dev_put(pdev);
+		return;
+	}
+
 	if (pci_read_config_dword(pdev, 0x188, &vtisochctrl)) {
 		pci_dev_put(pdev);
 		return;
Signature · 1 line
2.27.0.rc2.251.g90737beb825-goog
Acked-by on v1 1/1
#1 lore raw
Quoted 7 lines · Rajat Jain

Currently, an external malicious PCI device can masquerade the VID:PID of faulty gfx devices, and thus apply iommu quirks to effectively disable the IOMMU restrictions for itself.

Thus we need to ensure that the device we are applying quirks to, is indeed an internal trusted device.

Best regards,
baolu

Quoted 73 lines
---
  drivers/iommu/intel-iommu.c | 28 ++++++++++++++++++++++++++++
  1 file changed, 28 insertions(+)
drivers/iommu/intel-iommu.c +28 -0
index ef0a5246700e5..f2a480168a02f 100644
@@ -6214,6 +6214,11 @@ const struct iommu_ops intel_iommu_ops = {
 
  static void quirk_iommu_igfx(struct pci_dev *dev)
  {
+	if (dev->untrusted) {
+		pci_warn(dev, "skipping iommu quirk for untrusted gfx dev\n");
+		return;
+	}
+
  	pci_info(dev, "Disabling IOMMU for graphics on this chipset\n");
  	dmar_map_gfx = 0;
  }
@@ -6255,6 +6260,11 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x163D, quirk_iommu_igfx);
 
  static void quirk_iommu_rwbf(struct pci_dev *dev)
  {
+	if (dev->untrusted) {
+		pci_warn(dev, "skipping iommu quirk for untrusted dev\n");
+		return;
+	}
+
  	/*
  	 * Mobile 4 Series Chipset neglects to set RWBF capability,
  	 * but needs it. Same seems to hold for the desktop versions.
@@ -6285,6 +6295,11 @@ static void quirk_calpella_no_shadow_gtt(struct pci_dev *dev)
  {
  	unsigned short ggc;
 
+	if (dev->untrusted) {
+		pci_warn(dev, "skipping iommu quirk for untrusted gfx dev\n");
+		return;
+	}
+
  	if (pci_read_config_word(dev, GGC, &ggc))
  		return;
 
@@ -6318,6 +6333,13 @@ static void __init check_tylersburg_isoch(void)
  	pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 0x3a3e, NULL);
  	if (!pdev)
  		return;
+
+	if (pdev->untrusted) {
+		pci_warn(pdev, "skipping iommu quirk due to untrusted dev\n");
+		pci_dev_put(pdev);
+		return;
+	}
+
  	pci_dev_put(pdev);
 
  	/* System Management Registers. Might be hidden, in which case
@@ -6327,6 +6349,12 @@ static void __init check_tylersburg_isoch(void)
  	if (!pdev)
  		return;
 
+	if (pdev->untrusted) {
+		pci_warn(pdev, "skipping iommu quirk due to untrusted dev\n");
+		pci_dev_put(pdev);
+		return;
+	}
+
  	if (pci_read_config_dword(pdev, 0x188, &vtisochctrl)) {
  		pci_dev_put(pdev);
  		return;
#2 lore raw
In reply to Rajat Jain
Quoted 17 lines · Rajat Jain

Currently, an external malicious PCI device can masquerade the VID:PID of faulty gfx devices, and thus apply iommu quirks to effectively disable the IOMMU restrictions for itself.

Thus we need to ensure that the device we are applying quirks to, is indeed an internal trusted device.

---
 drivers/iommu/intel-iommu.c | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)
drivers/iommu/intel-iommu.c +0 -0
index ef0a5246700e5..f2a480168a02f 100644

drivers/iommu/intel-iommu.c const struct iommu_ops intel_iommu_ops = { · line 6218

Quoted 5 lines
drivers/iommu/intel-iommu.c +2 -0
@@ -6214,6 +6214,11 @@ const struct iommu_ops intel_iommu_ops = {
 
 static void quirk_iommu_igfx(struct pci_dev *dev)
 {
+	if (dev->untrusted) {
+		pci_warn(dev, "skipping iommu quirk for untrusted gfx dev\n");

I think you should be consistent with other messages. For example iommu should be spelled IOMMU as done below.

Also this is visible to users so maybe put bit more information there:

pci_warn(dev, "Will not apply IOMMU quirk for untrusted graphics device\n");

Ditto for all the other places. Also is "untrusted" good word here? If an ordinary user sees this will it trigger some sort of panic reaction. Perhaps we should call it "potentially untrusted" or something like that?

Quoted 5 lines
drivers/iommu/intel-iommu.c +3 -0
+		return;
+	}
+
 	pci_info(dev, "Disabling IOMMU for graphics on this chipset\n");
 	dmar_map_gfx = 0;
author
#3 lore raw

Hi MIka,

Thanks for taking a look.

Quoted 28 lines · Mika Westerberg
Quoted 17 lines · Rajat Jain

Currently, an external malicious PCI device can masquerade the VID:PID of faulty gfx devices, and thus apply iommu quirks to effectively disable the IOMMU restrictions for itself.

Thus we need to ensure that the device we are applying quirks to, is indeed an internal trusted device.

---
 drivers/iommu/intel-iommu.c | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)
drivers/iommu/intel-iommu.c +0 -0
index ef0a5246700e5..f2a480168a02f 100644

drivers/iommu/intel-iommu.c const struct iommu_ops intel_iommu_ops = { · line 6218

Quoted 5 lines
drivers/iommu/intel-iommu.c +2 -0
@@ -6214,6 +6214,11 @@ const struct iommu_ops intel_iommu_ops = {
 
 static void quirk_iommu_igfx(struct pci_dev *dev)
 {
+     if (dev->untrusted) {
+             pci_warn(dev, "skipping iommu quirk for untrusted gfx dev\n");

I think you should be consistent with other messages. For example iommu should be spelled IOMMU as done below.

Quoted 8 lines

Also this is visible to users so maybe put bit more information there:

pci_warn(dev, "Will not apply IOMMU quirk for untrusted graphics device\n");

Ditto for all the other places. Also is "untrusted" good word here? If an ordinary user sees this will it trigger some sort of panic reaction. Perhaps we should call it "potentially untrusted" or something like that?

Fixed it, posted new patch at
https://lkml.org/lkml/2020/6/2/822

Thanks,

Rajat

Quoted 6 lines
Quoted 6 lines
drivers/iommu/intel-iommu.c +3 -0
+             return;
+     }
+
      pci_info(dev, "Disabling IOMMU for graphics on this chipset\n");
      dmar_map_gfx = 0;
#4 lore raw
Quoted 42 lines · Rajat Jain

Hi MIka,

Thanks for taking a look.

Quoted 36 lines · Mika Westerberg
Quoted 17 lines · Rajat Jain

Currently, an external malicious PCI device can masquerade the VID:PID of faulty gfx devices, and thus apply iommu quirks to effectively disable the IOMMU restrictions for itself.

Thus we need to ensure that the device we are applying quirks to, is indeed an internal trusted device.

---
 drivers/iommu/intel-iommu.c | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)
drivers/iommu/intel-iommu.c +0 -0
index ef0a5246700e5..f2a480168a02f 100644

drivers/iommu/intel-iommu.c const struct iommu_ops intel_iommu_ops = { · line 6218

Quoted 5 lines
drivers/iommu/intel-iommu.c +2 -0
@@ -6214,6 +6214,11 @@ const struct iommu_ops intel_iommu_ops = {
 
 static void quirk_iommu_igfx(struct pci_dev *dev)
 {
+     if (dev->untrusted) {
+             pci_warn(dev, "skipping iommu quirk for untrusted gfx dev\n");

I think you should be consistent with other messages. For example iommu should be spelled IOMMU as done below.

Also this is visible to users so maybe put bit more information there:

pci_warn(dev, "Will not apply IOMMU quirk for untrusted graphics device\n");

Ditto for all the other places. Also is "untrusted" good word here? If an ordinary user sees this will it trigger some sort of panic reaction. Perhaps we should call it "potentially untrusted" or something like that?

Wish we called it external_facing rather than untrusted attribute, so its description is consistent with the spec that defines it. Once we have Platform Component Security Enhancements.

to be correct, maybe call "Device located on an untrusted link" rather than assert blame on the device.

Since the information is harvsted from BIOS tables and there are chances this could be wrongly grouped such, add "Check with your BIOS/Platform Vendor.

Quoted 14 lines

Fixed it, posted new patch at
https://lkml.org/lkml/2020/6/2/822

Thanks,

Rajat

Quoted 6 lines
Quoted 6 lines
drivers/iommu/intel-iommu.c +3 -0
+             return;
+     }
+
      pci_info(dev, "Disabling IOMMU for graphics on this chipset\n");
      dmar_map_gfx = 0;