Skip to contentSkip to conversation

Add quirk for PCIe root port on AMD systems

PATCH In review Mario Limonciello 4 participants 10 messages Sep 20, 2023 – Oct 3, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

v20 cover Add quirk for PCIe root port on AMD systems

Iain reports that USB devices can't be used to wake a Lenovo Z13 from suspend. This is because the PCIe root port has been put into D3hot and AMD's platform can't handle USB devices waking the platform from a hardware sleep state in this case.

The firmware doesn't express this limitation in a way that Linux recognizes so this series introduces a quirk for this problem.

Previous submissions:
* v19
https://lore.kernel.org/linux-pci/[email protected]/
* v18
https://lore.kernel.org/linux-usb/[email protected]/T/#m92a2654129d5f6be5439d8a94fb36734804763c1
* v17
https://lore.kernel.org/platform-driver-x86/[email protected]/
This version implemented constraints for the amd-pmc driver and introduced a veto/optin system for the PCI core as suggested by Hans. Rafael suggested not to use the veto/optin system and instead use a quirk.
* v16
https://lore.kernel.org/platform-driver-x86/[email protected]/
This version implemented constraints for the amd-pmc driver.
* v15
https://lore.kernel.org/platform-driver-x86/[email protected]/#t
This version hardcoded the quirk into amd-pmc driver as part of suspend callback.
* v14
https://lore.kernel.org/linux-pci/[email protected]/
https://lore.kernel.org/linux-pci/[email protected]/
https://lore.kernel.org/linux-pci/[email protected]/
This version implemented constraints exporting and limited the policy for

Quoted 1 line

= 2015 to Intel only. It also added support for constraints optin.

V13 was split into multiple parts to make it easier to land. 14.b was merged.
* v13
https://lore.kernel.org/linux-pci/[email protected]/
* v12
https://lore.kernel.org/linux-pci/[email protected]/
* v11
https://lore.kernel.org/linux-pci/[email protected]/
* v10
https://lore.kernel.org/linux-pci/[email protected]/
* v9
https://lore.kernel.org/linux-pci/[email protected]/
* v8
https://lore.kernel.org/linux-pci/[email protected]/
* v7
https://lore.kernel.org/linux-pci/[email protected]/
* v6
https://lore.kernel.org/linux-pci/[email protected]/
* v5
https://lore.kernel.org/linux-pci/[email protected]/
* v4
https://lore.kernel.org/linux-pci/[email protected]/
* v3
https://lore.kernel.org/linux-pci/[email protected]/
* v2
https://lore.kernel.org/linux-pci/[email protected]/
* v1
https://lore.kernel.org/linux-pci/[email protected]/

Mario Limonciello (2):
PCI: Move the PCI_CLASS_SERIAL_USB_USB4 definition to common header
PCI: Add a quirk for AMD PCIe root ports w/ USB4 controllers

 drivers/pci/quirks.c      | 71 +++++++++++++++++++++++++++++++++++++++
 drivers/thunderbolt/nhi.h |  2 --
 include/linux/pci_ids.h   |  1 +
 3 files changed, 72 insertions(+), 2 deletions(-)
Signature · 1 line
2.34.1
#1 lore raw

PATCH v20 1/2 PCI: Move the `PCI_CLASS_SERIAL_USB_USB4` definition to common header

PCI_CLASS_SERIAL_USB_USB4 may be used by code outside of thunderbolt. Move the declaration into the common pci_ids.h header.

---
 drivers/thunderbolt/nhi.h | 2 --
 include/linux/pci_ids.h   | 1 +
 2 files changed, 1 insertion(+), 2 deletions(-)
drivers/thunderbolt/nhi.h +0 -2
index 0f029ce75882..675ddefe283c 100644
@@ -91,6 +91,4 @@ extern const struct tb_nhi_ops icl_nhi_ops;
 #define PCI_DEVICE_ID_INTEL_RPL_NHI0			0xa73e
 #define PCI_DEVICE_ID_INTEL_RPL_NHI1			0xa76d
 
-#define PCI_CLASS_SERIAL_USB_USB4			0x0c0340
-
 #endif
include/linux/pci_ids.h +1 -0
index 5fb3d4c393a9..29aeac53dc41 100644
@@ -121,6 +121,7 @@
 #define PCI_CLASS_SERIAL_USB_OHCI	0x0c0310
 #define PCI_CLASS_SERIAL_USB_EHCI	0x0c0320
 #define PCI_CLASS_SERIAL_USB_XHCI	0x0c0330
+#define PCI_CLASS_SERIAL_USB_USB4	0x0c0340
 #define PCI_CLASS_SERIAL_USB_DEVICE	0x0c03fe
 #define PCI_CLASS_SERIAL_FIBER		0x0c04
 #define PCI_CLASS_SERIAL_SMBUS		0x0c05
Signature · 1 line
2.34.1
#2 lore raw

PATCH v20 2/2 PCI: Add a quirk for AMD PCIe root ports w/ USB4 controllers

In reply to Mario Limonciello

Iain reports that USB devices can't be used to wake a Lenovo Z13 from suspend. This problem occurs because the PCIe root port has been put into D3hot and AMD's platform can't handle the PME associated with USB devices waking the platform from a hardware sleep state in this case. The platform is put into a hardware sleep state by the actions of the amd-pmc driver.

Although the issue is initially reported on a single model it actually affects all Yellow Carp (Rembrandt) and Pink Sardine (Phoenix) SoCs. This problem only occurs on Linux specifically when attempting to wake the platform from a hardware sleep state.
Comparing the behavior on Windows and Linux, Windows doesn't put the root ports into D3 at this time.

Linux decides the target state to put the device into at suspend by this policy:
1. If platform_pci_power_manageable():
Use platform_pci_choose_state()
2. If the device is armed for wakeup:
Select the deepest D-state that supports a PME.
3. Else:
Use D3hot.

Devices are considered power manageable by the platform when they have one or more objects described in the table in section 7.3 of the ACPI 6.5 specification [1]. In this case the root ports are not power manageable.

If devices are not considered power manageable; specs are ambiguous as to what should happen. In this situation Windows 11 puts PCIe ports in D0 ostensibly due the policy from the "uPEP driver" which is a complimentary driver the Linux "amd-pmc" driver.

Linux chooses to allow D3 for these root ports due to the policy introduced by commit 9d26d3a8f1b0 ("PCI: Put PCIe ports into D3 during suspend"). Since Linux allows D3 for these ports, it follows the assertion that a PME can be used to wake from D3hot or D3cold and selects D3hot at suspend time.

Even though the PCIe PM capabilities advertise PME from D3hot or D3cold the Windows uPEP driver expresses the desired state that should be selected for suspend is still D30. As Linux doesn't use this information, for makin ga policy decision introduce a quirk for the problematic root ports.

The quirk removes PME support for D3hot and D3cold at system suspend time. When the port is configured for wakeup this will prevent these states from being selected in pci_target_state().

After the system is resumes the PME support is re-read from the PM capabilities register to allow opportunistic power savings at runtime by letting the root port go into D3hot or D3cold.

---
v19->v20:
* Adjust commit message (Bjorn)
* Use FIELD_GET (Ilpo)
* Use pci_walk_bus (Lukas)

---
 drivers/pci/quirks.c | 71 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 71 insertions(+)
drivers/pci/quirks.c +71 -0
index eeec1d6f9023..4159b7f20fd5 100644
@@ -6188,3 +6188,74 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5020, of_pci_make_dev_node);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5021, of_pci_make_dev_node);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_REDHAT, 0x0005, of_pci_make_dev_node);
+
+#ifdef CONFIG_SUSPEND
+/*
+ * When AMD PCIe root ports with AMD USB4 controllers attached to them are put
+ * into D3hot or D3cold downstream USB devices may fail to wakeup the system
+ * from suspend to idle.  This manifests as a missing wakeup interrupt.
+ *
+ * Prevent the associated root port from using PME to wake from D3hot or
+ * D3cold power states during suspend.
+ * This will effectively put the root port into D0 power state over suspend.
+ */
+#define PCI_PM_CAP_D3_PME_MASK	((PCI_PM_CAP_PME_D3hot|PCI_PM_CAP_PME_D3cold) \
+				>> PCI_PM_CAP_PME_SHIFT)
+static int modify_pme_amd_usb4(struct pci_dev *dev, void *data)
+{
+	bool *suspend = (bool *)data;
+	struct pci_dev *rp;
+	u16 pmc;
+
+	if (dev->vendor != PCI_VENDOR_ID_AMD ||
+	    dev->class != PCI_CLASS_SERIAL_USB_USB4)
+		return 0;
+	rp = pcie_find_root_port(dev);
+	if (!rp->pm_cap)
+		return -ENODEV;
+
+	if (*suspend) {
+		if (!(rp->pme_support & PCI_PM_CAP_D3_PME_MASK))
+			return -EINVAL;
+
+		rp->pme_support &= ~PCI_PM_CAP_D3_PME_MASK;
+		dev_info_once(&rp->dev, "quirk: disabling PME from D3hot and D3cold at suspend\n");
+
+		/* no need to check any more devices, found and applied quirk */
+		return -EEXIST;
+	}
+
+	/* already done */
+	if (rp->pme_support & PCI_PM_CAP_D3_PME_MASK)
+		return -EINVAL;
+
+	/* restore hardware defaults so runtime suspend can use it */
+	pci_read_config_word(rp, rp->pm_cap + PCI_PM_PMC, &pmc);
+	rp->pme_support = FIELD_GET(PCI_PM_CAP_PME_MASK, pmc);
+
+	return -EEXIST;
+}
+
+static void quirk_reenable_pme(struct pci_dev *dev)
+{
+	bool suspend = FALSE;
+
+	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)&suspend);
+}
+
+static void quirk_disable_pme_suspend(struct pci_dev *dev)
+{
+	bool suspend = TRUE;
+
+	/* skip for runtime suspend */
+	if (pm_suspend_target_state == PM_SUSPEND_ON)
+		return;
+
+	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)&suspend);
+}
+
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14b9, quirk_disable_pme_suspend);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14b9, quirk_reenable_pme);
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14eb, quirk_disable_pme_suspend);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14eb, quirk_reenable_pme);
+#endif /* CONFIG_SUSPEND */
Signature · 1 line
2.34.1
Reviewed-by on v20 2/2
#3 lore raw

Re: [PATCH v20 2/2] PCI: Add a quirk for AMD PCIe root ports w/ USB4 controllers

Quoted 52 lines · Mario Limonciello

Iain reports that USB devices can't be used to wake a Lenovo Z13 from suspend. This problem occurs because the PCIe root port has been put into D3hot and AMD's platform can't handle the PME associated with USB devices waking the platform from a hardware sleep state in this case. The platform is put into a hardware sleep state by the actions of the amd-pmc driver.

Although the issue is initially reported on a single model it actually affects all Yellow Carp (Rembrandt) and Pink Sardine (Phoenix) SoCs. This problem only occurs on Linux specifically when attempting to wake the platform from a hardware sleep state.
Comparing the behavior on Windows and Linux, Windows doesn't put the root ports into D3 at this time.

Linux decides the target state to put the device into at suspend by this policy:
1. If platform_pci_power_manageable():
Use platform_pci_choose_state()
2. If the device is armed for wakeup:
Select the deepest D-state that supports a PME.
3. Else:
Use D3hot.

Devices are considered power manageable by the platform when they have one or more objects described in the table in section 7.3 of the ACPI 6.5 specification [1]. In this case the root ports are not power manageable.

If devices are not considered power manageable; specs are ambiguous as to what should happen. In this situation Windows 11 puts PCIe ports in D0 ostensibly due the policy from the "uPEP driver" which is a complimentary driver the Linux "amd-pmc" driver.

Linux chooses to allow D3 for these root ports due to the policy introduced by commit 9d26d3a8f1b0 ("PCI: Put PCIe ports into D3 during suspend"). Since Linux allows D3 for these ports, it follows the assertion that a PME can be used to wake from D3hot or D3cold and selects D3hot at suspend time.

Even though the PCIe PM capabilities advertise PME from D3hot or D3cold the Windows uPEP driver expresses the desired state that should be selected for suspend is still D30. As Linux doesn't use this information, for makin ga policy decision introduce a quirk for the problematic root ports.

The quirk removes PME support for D3hot and D3cold at system suspend time. When the port is configured for wakeup this will prevent these states from being selected in pci_target_state().

After the system is resumes the PME support is re-read from the PM capabilities register to allow opportunistic power savings at runtime by letting the root port go into D3hot or D3cold.

Quoted 6 lines

One super-minor comment, no need to send a new version just for this.

Quoted 23 lines

---
v19->v20:
* Adjust commit message (Bjorn)
* Use FIELD_GET (Ilpo)
* Use pci_walk_bus (Lukas)

---
 drivers/pci/quirks.c | 71 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 71 insertions(+)
drivers/pci/quirks.c +6 -0
index eeec1d6f9023..4159b7f20fd5 100644
@@ -6188,3 +6188,74 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5020, of_pci_make_dev_node);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5021, of_pci_make_dev_node);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_REDHAT, 0x0005, of_pci_make_dev_node);
+
+#ifdef CONFIG_SUSPEND
+/*
+ * When AMD PCIe root ports with AMD USB4 controllers attached to them are put
+ * into D3hot or D3cold downstream USB devices may fail to wakeup the system
+ * from suspend to idle.  This manifests as a missing wakeup interrupt.

drivers/pci/quirks.c DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); · line 6206

Quoted 10 lines
drivers/pci/quirks.c +10 -0
+ *
+ * Prevent the associated root port from using PME to wake from D3hot or
+ * D3cold power states during suspend.
+ * This will effectively put the root port into D0 power state over suspend.
+ */
+#define PCI_PM_CAP_D3_PME_MASK	((PCI_PM_CAP_PME_D3hot|PCI_PM_CAP_PME_D3cold) \
+				>> PCI_PM_CAP_PME_SHIFT)
+static int modify_pme_amd_usb4(struct pci_dev *dev, void *data)
+{
+	bool *suspend = (bool *)data;

You could also pass the bool as value because void * can hold it so

	bool suspend = (bool)data;
Quoted 27 lines
drivers/pci/quirks.c +27 -0
+	struct pci_dev *rp;
+	u16 pmc;
+
+	if (dev->vendor != PCI_VENDOR_ID_AMD ||
+	    dev->class != PCI_CLASS_SERIAL_USB_USB4)
+		return 0;
+	rp = pcie_find_root_port(dev);
+	if (!rp->pm_cap)
+		return -ENODEV;
+
+	if (*suspend) {
+		if (!(rp->pme_support & PCI_PM_CAP_D3_PME_MASK))
+			return -EINVAL;
+
+		rp->pme_support &= ~PCI_PM_CAP_D3_PME_MASK;
+		dev_info_once(&rp->dev, "quirk: disabling PME from D3hot and D3cold at suspend\n");
+
+		/* no need to check any more devices, found and applied quirk */
+		return -EEXIST;
+	}
+
+	/* already done */
+	if (rp->pme_support & PCI_PM_CAP_D3_PME_MASK)
+		return -EINVAL;
+
+	/* restore hardware defaults so runtime suspend can use it */
+	pci_read_config_word(rp, rp->pm_cap + PCI_PM_PMC, &pmc);

drivers/pci/quirks.c DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); · line 6243

Quoted 10 lines
drivers/pci/quirks.c +10 -0
+	rp->pme_support = FIELD_GET(PCI_PM_CAP_PME_MASK, pmc);
+
+	return -EEXIST;
+}
+
+static void quirk_reenable_pme(struct pci_dev *dev)
+{
+	bool suspend = FALSE;
+
+	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)&suspend);

and here

	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)false);

drivers/pci/quirks.c DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); · line 6254

Quoted 11 lines
drivers/pci/quirks.c +11 -0
+}
+
+static void quirk_disable_pme_suspend(struct pci_dev *dev)
+{
+	bool suspend = TRUE;
+
+	/* skip for runtime suspend */
+	if (pm_suspend_target_state == PM_SUSPEND_ON)
+		return;
+
+	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)&suspend);

here

	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)true);
Quoted 9 lines
drivers/pci/quirks.c +7 -0
+}
+
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14b9, quirk_disable_pme_suspend);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14b9, quirk_reenable_pme);
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14eb, quirk_disable_pme_suspend);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14eb, quirk_reenable_pme);
+#endif /* CONFIG_SUSPEND */
Signature · 1 line
2.34.1
#4 lore raw

Re: [PATCH v20 2/2] PCI: Add a quirk for AMD PCIe root ports w/ USB4 controllers

Quoted 60 lines · Mika Westerberg
Quoted 52 lines · Mario Limonciello

Iain reports that USB devices can't be used to wake a Lenovo Z13 from suspend. This problem occurs because the PCIe root port has been put into D3hot and AMD's platform can't handle the PME associated with USB devices waking the platform from a hardware sleep state in this case. The platform is put into a hardware sleep state by the actions of the amd-pmc driver.

Although the issue is initially reported on a single model it actually affects all Yellow Carp (Rembrandt) and Pink Sardine (Phoenix) SoCs. This problem only occurs on Linux specifically when attempting to wake the platform from a hardware sleep state.
Comparing the behavior on Windows and Linux, Windows doesn't put the root ports into D3 at this time.

Linux decides the target state to put the device into at suspend by this policy:
1. If platform_pci_power_manageable():
Use platform_pci_choose_state()
2. If the device is armed for wakeup:
Select the deepest D-state that supports a PME.
3. Else:
Use D3hot.

Devices are considered power manageable by the platform when they have one or more objects described in the table in section 7.3 of the ACPI 6.5 specification [1]. In this case the root ports are not power manageable.

If devices are not considered power manageable; specs are ambiguous as to what should happen. In this situation Windows 11 puts PCIe ports in D0 ostensibly due the policy from the "uPEP driver" which is a complimentary driver the Linux "amd-pmc" driver.

Linux chooses to allow D3 for these root ports due to the policy introduced by commit 9d26d3a8f1b0 ("PCI: Put PCIe ports into D3 during suspend"). Since Linux allows D3 for these ports, it follows the assertion that a PME can be used to wake from D3hot or D3cold and selects D3hot at suspend time.

Even though the PCIe PM capabilities advertise PME from D3hot or D3cold the Windows uPEP driver expresses the desired state that should be selected for suspend is still D30. As Linux doesn't use this information, for makin ga policy decision introduce a quirk for the problematic root ports.

The quirk removes PME support for D3hot and D3cold at system suspend time. When the port is configured for wakeup this will prevent these states from being selected in pci_target_state().

After the system is resumes the PME support is re-read from the PM capabilities register to allow opportunistic power savings at runtime by letting the root port go into D3hot or D3cold.

Quoted 6 lines
Quoted 3 lines

One super-minor comment, no need to send a new version just for this.

Much appreciated!

Bjorn,

Anything else you want to see for this series?

Quoted 105 lines
Quoted 24 lines

---
v19->v20:
* Adjust commit message (Bjorn)
* Use FIELD_GET (Ilpo)
* Use pci_walk_bus (Lukas)

---
  drivers/pci/quirks.c | 71 ++++++++++++++++++++++++++++++++++++++++++++
  1 file changed, 71 insertions(+)
drivers/pci/quirks.c +6 -0
index eeec1d6f9023..4159b7f20fd5 100644
@@ -6188,3 +6188,74 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size);
  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5020, of_pci_make_dev_node);
  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5021, of_pci_make_dev_node);
  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_REDHAT, 0x0005, of_pci_make_dev_node);
+
+#ifdef CONFIG_SUSPEND
+/*
+ * When AMD PCIe root ports with AMD USB4 controllers attached to them are put
+ * into D3hot or D3cold downstream USB devices may fail to wakeup the system
+ * from suspend to idle.  This manifests as a missing wakeup interrupt.

drivers/pci/quirks.c DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); · line 6206

Quoted 10 lines
drivers/pci/quirks.c +10 -0
+ *
+ * Prevent the associated root port from using PME to wake from D3hot or
+ * D3cold power states during suspend.
+ * This will effectively put the root port into D0 power state over suspend.
+ */
+#define PCI_PM_CAP_D3_PME_MASK	((PCI_PM_CAP_PME_D3hot|PCI_PM_CAP_PME_D3cold) \
+				>> PCI_PM_CAP_PME_SHIFT)
+static int modify_pme_amd_usb4(struct pci_dev *dev, void *data)
+{
+	bool *suspend = (bool *)data;

You could also pass the bool as value because void * can hold it so

	bool suspend = (bool)data;
Quoted 27 lines
drivers/pci/quirks.c +27 -0
+	struct pci_dev *rp;
+	u16 pmc;
+
+	if (dev->vendor != PCI_VENDOR_ID_AMD ||
+	    dev->class != PCI_CLASS_SERIAL_USB_USB4)
+		return 0;
+	rp = pcie_find_root_port(dev);
+	if (!rp->pm_cap)
+		return -ENODEV;
+
+	if (*suspend) {
+		if (!(rp->pme_support & PCI_PM_CAP_D3_PME_MASK))
+			return -EINVAL;
+
+		rp->pme_support &= ~PCI_PM_CAP_D3_PME_MASK;
+		dev_info_once(&rp->dev, "quirk: disabling PME from D3hot and D3cold at suspend\n");
+
+		/* no need to check any more devices, found and applied quirk */
+		return -EEXIST;
+	}
+
+	/* already done */
+	if (rp->pme_support & PCI_PM_CAP_D3_PME_MASK)
+		return -EINVAL;
+
+	/* restore hardware defaults so runtime suspend can use it */
+	pci_read_config_word(rp, rp->pm_cap + PCI_PM_PMC, &pmc);

drivers/pci/quirks.c DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); · line 6243

Quoted 10 lines
drivers/pci/quirks.c +10 -0
+	rp->pme_support = FIELD_GET(PCI_PM_CAP_PME_MASK, pmc);
+
+	return -EEXIST;
+}
+
+static void quirk_reenable_pme(struct pci_dev *dev)
+{
+	bool suspend = FALSE;
+
+	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)&suspend);

and here

	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)false);

drivers/pci/quirks.c DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); · line 6254

Quoted 11 lines
drivers/pci/quirks.c +11 -0
+}
+
+static void quirk_disable_pme_suspend(struct pci_dev *dev)
+{
+	bool suspend = TRUE;
+
+	/* skip for runtime suspend */
+	if (pm_suspend_target_state == PM_SUSPEND_ON)
+		return;
+
+	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)&suspend);

here

	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)true);
Quoted 9 lines
drivers/pci/quirks.c +7 -0
+}
+
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14b9, quirk_disable_pme_suspend);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14b9, quirk_reenable_pme);
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14eb, quirk_disable_pme_suspend);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14eb, quirk_reenable_pme);
+#endif /* CONFIG_SUSPEND */
Signature · 1 line
2.34.1
#5 lore raw

Re: [PATCH v20 2/2] PCI: Add a quirk for AMD PCIe root ports w/ USB4 controllers

In reply to Mario Limonciello
Quoted 112 lines · Mario Limonciello

Iain reports that USB devices can't be used to wake a Lenovo Z13 from suspend. This problem occurs because the PCIe root port has been put into D3hot and AMD's platform can't handle the PME associated with USB devices waking the platform from a hardware sleep state in this case. The platform is put into a hardware sleep state by the actions of the amd-pmc driver.

Although the issue is initially reported on a single model it actually affects all Yellow Carp (Rembrandt) and Pink Sardine (Phoenix) SoCs. This problem only occurs on Linux specifically when attempting to wake the platform from a hardware sleep state.
Comparing the behavior on Windows and Linux, Windows doesn't put the root ports into D3 at this time.

Linux decides the target state to put the device into at suspend by this policy:
1. If platform_pci_power_manageable():
Use platform_pci_choose_state()
2. If the device is armed for wakeup:
Select the deepest D-state that supports a PME.
3. Else:
Use D3hot.

Devices are considered power manageable by the platform when they have one or more objects described in the table in section 7.3 of the ACPI 6.5 specification [1]. In this case the root ports are not power manageable.

If devices are not considered power manageable; specs are ambiguous as to what should happen. In this situation Windows 11 puts PCIe ports in D0 ostensibly due the policy from the "uPEP driver" which is a complimentary driver the Linux "amd-pmc" driver.

Linux chooses to allow D3 for these root ports due to the policy introduced by commit 9d26d3a8f1b0 ("PCI: Put PCIe ports into D3 during suspend"). Since Linux allows D3 for these ports, it follows the assertion that a PME can be used to wake from D3hot or D3cold and selects D3hot at suspend time.

Even though the PCIe PM capabilities advertise PME from D3hot or D3cold the Windows uPEP driver expresses the desired state that should be selected for suspend is still D30. As Linux doesn't use this information, for makin ga policy decision introduce a quirk for the problematic root ports.

The quirk removes PME support for D3hot and D3cold at system suspend time. When the port is configured for wakeup this will prevent these states from being selected in pci_target_state().

After the system is resumes the PME support is re-read from the PM capabilities register to allow opportunistic power savings at runtime by letting the root port go into D3hot or D3cold.

---
v19->v20:
* Adjust commit message (Bjorn)
* Use FIELD_GET (Ilpo)
* Use pci_walk_bus (Lukas)

---
 drivers/pci/quirks.c | 71 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 71 insertions(+)
drivers/pci/quirks.c +37 -0
index eeec1d6f9023..4159b7f20fd5 100644
@@ -6188,3 +6188,74 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5020, of_pci_make_dev_node);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5021, of_pci_make_dev_node);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_REDHAT, 0x0005, of_pci_make_dev_node);
+
+#ifdef CONFIG_SUSPEND
+/*
+ * When AMD PCIe root ports with AMD USB4 controllers attached to them are put
+ * into D3hot or D3cold downstream USB devices may fail to wakeup the system
+ * from suspend to idle.  This manifests as a missing wakeup interrupt.
+ *
+ * Prevent the associated root port from using PME to wake from D3hot or
+ * D3cold power states during suspend.
+ * This will effectively put the root port into D0 power state over suspend.
+ */
+#define PCI_PM_CAP_D3_PME_MASK	((PCI_PM_CAP_PME_D3hot|PCI_PM_CAP_PME_D3cold) \
+				>> PCI_PM_CAP_PME_SHIFT)
+static int modify_pme_amd_usb4(struct pci_dev *dev, void *data)
+{
+	bool *suspend = (bool *)data;
+	struct pci_dev *rp;
+	u16 pmc;
+
+	if (dev->vendor != PCI_VENDOR_ID_AMD ||
+	    dev->class != PCI_CLASS_SERIAL_USB_USB4)
+		return 0;
+	rp = pcie_find_root_port(dev);
+	if (!rp->pm_cap)
+		return -ENODEV;
+
+	if (*suspend) {
+		if (!(rp->pme_support & PCI_PM_CAP_D3_PME_MASK))
+			return -EINVAL;
+
+		rp->pme_support &= ~PCI_PM_CAP_D3_PME_MASK;
+		dev_info_once(&rp->dev, "quirk: disabling PME from D3hot and D3cold at suspend\n");
+
+		/* no need to check any more devices, found and applied quirk */
+		return -EEXIST;
+	}
+

drivers/pci/quirks.c DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); · line 6237

Quoted 10 lines
drivers/pci/quirks.c +10 -0
+	/* already done */
+	if (rp->pme_support & PCI_PM_CAP_D3_PME_MASK)
+		return -EINVAL;
+
+	/* restore hardware defaults so runtime suspend can use it */
+	pci_read_config_word(rp, rp->pm_cap + PCI_PM_PMC, &pmc);
+	rp->pme_support = FIELD_GET(PCI_PM_CAP_PME_MASK, pmc);
+
+	return -EEXIST;
+}

This is a lot of hassle to look for USB4 devices below the Root Port. You said earlier that these Root Ports *only* connect to xHCI and USB4 [1]. If that's the case, why even bother with the search? Why not just clear PCI_PM_CAP_PME_D3hot and PCI_PM_CAP_PME_D3cold unconditionally, e.g., the possible patch below?

I have no idea whether it's even possible to have a device other than xHCI/USB4 below these Root Ports, but I don't think we have any evidence that the PME failure is specific to USB4, so if it *is* possible, it's likely that PME from them would fail the same way.

[1] https://lore.kernel.org/r/[email protected]

Quoted 13 lines
drivers/pci/quirks.c +13 -0
+static void quirk_reenable_pme(struct pci_dev *dev)
+{
+	bool suspend = FALSE;
+
+	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)&suspend);
+}
+
+static void quirk_disable_pme_suspend(struct pci_dev *dev)
+{
+	bool suspend = TRUE;
+
+	/* skip for runtime suspend */
+	if (pm_suspend_target_state == PM_SUSPEND_ON)

drivers/pci/quirks.c DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); · line 6260

Quoted 10 lines
drivers/pci/quirks.c +10 -0
+		return;
+
+	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)&suspend);
+}
+
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14b9, quirk_disable_pme_suspend);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14b9, quirk_reenable_pme);
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14eb, quirk_disable_pme_suspend);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14eb, quirk_reenable_pme);
+#endif /* CONFIG_SUSPEND */

PCI: Avoid PME from D3hot/cold for AMD Rembrandt and Phoenix

Iain reports that USB devices can't be used to wake a Lenovo Z13 from suspend. This occurs because on some AMD platforms, even though the Root Ports advertise PME_Support for D3hot and D3cold, they don't handle PME messages and generate wakeup interrupts from those states when amd-pmc has put the platform in a hardware sleep state.

Iain reported this on an AMD Rembrandt platform, but it also affects Phoenix SoCs. On Iain's system, a USB device below the affected Root Port generates the PME, but there is no reason to believe PMEs from other kinds of devices would work differently.

To avoid this issue, remove D3hot and D3cold from the Root Port's PME_Support mask when suspending if we expect amd-pmc to put the platform in a hardware sleep state. The effect of this is that if there is a wakeup device below the Root Port, we will avoid D3hot and D3cold.

Restore the advertised PME_Support mask on resume so D3hot and D3cold can be used by runtime suspend. The amd-pmc driver doesn't put the platform in a hardware sleep state for runtime suspend, so PMEs work as advertised.

---
 drivers/pci/quirks.c | 54 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 54 insertions(+)
drivers/pci/quirks.c +54 -0
index eeec1d6f9023..cdf03eb610b0 100644
@@ -6188,3 +6188,57 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5020, of_pci_make_dev_node);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5021, of_pci_make_dev_node);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_REDHAT, 0x0005, of_pci_make_dev_node);
+
+#ifdef CONFIG_SUSPEND
+/*
+ * Root Ports on AMD Rembrandt and Phoenix SoCs advertise PME_Support for
+ * D3hot and D3cold, but if the SoC is put into a hardware sleep state by
+ * the amd-pmc driver, the Root Ports don't generate wakeup interrupts from
+ * those states.
+ *
+ * When suspending, remove D3hot and D3cold from the PME_Support advertised
+ * by the Root Port so we don't use those states if we're expecting wakeup
+ * interrupts.  Restore the advertised PME_Support when resuming.
+ */
+#define PCI_PM_CAP_D3_PME_MASK ((PCI_PM_CAP_PME_D3hot|PCI_PM_CAP_PME_D3cold) \
+				>> PCI_PM_CAP_PME_SHIFT)
+
+static void quirk_amd_pme_unsupported(struct pci_dev *dev)
+{
+	if (!dev->pm_cap)
+		return;
+
+	/*
+	 * PM_SUSPEND_ON means we're doing runtime suspend, which means
+	 * amd-pmc will not be involved so PMEs work as advertised.
+	 *
+	 * The PMEs *do* work in D3hot/D3cold if amd-pmc doesn't put the
+	 * SoC in the hardware sleep state, but we assume amd-pmc is always
+	 * present.
+	 */
+	if (pm_suspend_target_state == PM_SUSPEND_ON)
+		return;
+
+	dev->pme_support &= ~PCI_PM_CAP_D3_PME_MASK;
+	dev_info_once(&dev->dev, "D3hot/D3cold PMEs don't work in hardware sleep state\n");
+}
+
+static void quirk_amd_pme_supported(struct pci_dev *dev)
+{
+	u16 pmc;
+
+	if (!dev->pm_cap)
+		return;
+
+	if (dev->pme_support & PCI_PM_CAP_D3_PME_MASK)
+		return;
+
+	pci_read_config_word(dev, dev->pm_cap + PCI_PM_PMC, &pmc);
+	dev->pme_support = FIELD_GET(PCI_PM_CAP_PME_MASK, pmc);
+}
+
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14b9, quirk_amd_pme_unsupported);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14b9, quirk_amd_pme_supported);
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14eb, quirk_amd_pme_unsupported);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14eb, quirk_amd_pme_supported);
+#endif /* CONFIG_SUSPEND */
Signature · 1 line
2.34.1
#6 lore raw

Re: [PATCH v20 2/2] PCI: Add a quirk for AMD PCIe root ports w/ USB4 controllers

Quoted 119 lines · Bjorn Helgaas
Quoted 118 lines · Mario Limonciello

Iain reports that USB devices can't be used to wake a Lenovo Z13 from suspend. This problem occurs because the PCIe root port has been put into D3hot and AMD's platform can't handle the PME associated with USB devices waking the platform from a hardware sleep state in this case. The platform is put into a hardware sleep state by the actions of the amd-pmc driver.

Although the issue is initially reported on a single model it actually affects all Yellow Carp (Rembrandt) and Pink Sardine (Phoenix) SoCs. This problem only occurs on Linux specifically when attempting to wake the platform from a hardware sleep state.
Comparing the behavior on Windows and Linux, Windows doesn't put the root ports into D3 at this time.

Linux decides the target state to put the device into at suspend by this policy:
1. If platform_pci_power_manageable():
Use platform_pci_choose_state()
2. If the device is armed for wakeup:
Select the deepest D-state that supports a PME.
3. Else:
Use D3hot.

Devices are considered power manageable by the platform when they have one or more objects described in the table in section 7.3 of the ACPI 6.5 specification [1]. In this case the root ports are not power manageable.

If devices are not considered power manageable; specs are ambiguous as to what should happen. In this situation Windows 11 puts PCIe ports in D0 ostensibly due the policy from the "uPEP driver" which is a complimentary driver the Linux "amd-pmc" driver.

Linux chooses to allow D3 for these root ports due to the policy introduced by commit 9d26d3a8f1b0 ("PCI: Put PCIe ports into D3 during suspend"). Since Linux allows D3 for these ports, it follows the assertion that a PME can be used to wake from D3hot or D3cold and selects D3hot at suspend time.

Even though the PCIe PM capabilities advertise PME from D3hot or D3cold the Windows uPEP driver expresses the desired state that should be selected for suspend is still D30. As Linux doesn't use this information, for makin ga policy decision introduce a quirk for the problematic root ports.

The quirk removes PME support for D3hot and D3cold at system suspend time. When the port is configured for wakeup this will prevent these states from being selected in pci_target_state().

After the system is resumes the PME support is re-read from the PM capabilities register to allow opportunistic power savings at runtime by letting the root port go into D3hot or D3cold.

---
v19->v20:
* Adjust commit message (Bjorn)
* Use FIELD_GET (Ilpo)
* Use pci_walk_bus (Lukas)

---
  drivers/pci/quirks.c | 71 ++++++++++++++++++++++++++++++++++++++++++++
  1 file changed, 71 insertions(+)
drivers/pci/quirks.c +43 -0
index eeec1d6f9023..4159b7f20fd5 100644
@@ -6188,3 +6188,74 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size);
  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5020, of_pci_make_dev_node);
  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5021, of_pci_make_dev_node);
  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_REDHAT, 0x0005, of_pci_make_dev_node);
+
+#ifdef CONFIG_SUSPEND
+/*
+ * When AMD PCIe root ports with AMD USB4 controllers attached to them are put
+ * into D3hot or D3cold downstream USB devices may fail to wakeup the system
+ * from suspend to idle.  This manifests as a missing wakeup interrupt.
+ *
+ * Prevent the associated root port from using PME to wake from D3hot or
+ * D3cold power states during suspend.
+ * This will effectively put the root port into D0 power state over suspend.
+ */
+#define PCI_PM_CAP_D3_PME_MASK	((PCI_PM_CAP_PME_D3hot|PCI_PM_CAP_PME_D3cold) \
+				>> PCI_PM_CAP_PME_SHIFT)
+static int modify_pme_amd_usb4(struct pci_dev *dev, void *data)
+{
+	bool *suspend = (bool *)data;
+	struct pci_dev *rp;
+	u16 pmc;
+
+	if (dev->vendor != PCI_VENDOR_ID_AMD ||
+	    dev->class != PCI_CLASS_SERIAL_USB_USB4)
+		return 0;
+	rp = pcie_find_root_port(dev);
+	if (!rp->pm_cap)
+		return -ENODEV;
+
+	if (*suspend) {
+		if (!(rp->pme_support & PCI_PM_CAP_D3_PME_MASK))
+			return -EINVAL;
+
+		rp->pme_support &= ~PCI_PM_CAP_D3_PME_MASK;
+		dev_info_once(&rp->dev, "quirk: disabling PME from D3hot and D3cold at suspend\n");
+
+		/* no need to check any more devices, found and applied quirk */
+		return -EEXIST;
+	}
+
+	/* already done */
+	if (rp->pme_support & PCI_PM_CAP_D3_PME_MASK)
+		return -EINVAL;
+
+	/* restore hardware defaults so runtime suspend can use it */
+	pci_read_config_word(rp, rp->pm_cap + PCI_PM_PMC, &pmc);
Quoted 8 lines

drivers/pci/quirks.c DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); · line 6224

Quoted 4 lines
drivers/pci/quirks.c +4 -0
+	rp->pme_support = FIELD_GET(PCI_PM_CAP_PME_MASK, pmc);
+
+	return -EEXIST;
+}

This is a lot of hassle to look for USB4 devices below the Root Port. You said earlier that these Root Ports *only* connect to xHCI and USB4 [1].

That's correct, but the part that I believe you're missing is that there are multiple "instances" of this PCI device ID on a failing system.

On an OEM Phoenix laptop in front of me I see 3 instances of 1022:14eb:
00:08.1, 00:08.2, and 00:08.3.

Quoted 3 lines

If that's the case, why even bother with the search? Why not just clear PCI_PM_CAP_PME_D3hot and PCI_PM_CAP_PME_D3cold unconditionally, e.g., the possible patch below?

Only the one with the USB4 controller (in this system it's at 00:08.3) has the problem.

So I believe the proposal you have below will apply the policy to too many ports.

Quoted 6 lines

I have no idea whether it's even possible to have a device other than xHCI/USB4 below these Root Ports, but I don't think we have any evidence that the PME failure is specific to USB4, so if it *is* possible, it's likely that PME from them would fail the same way.

On some OEM system they only have type C ports wired up (to the USB4 controller), but on a "reference system" I have a USB-A ports that are connected directly to the XHCI controller available as well.

These work fine for wakeup without a quirk.

Quoted 114 lines

[1] https://lore.kernel.org/r/[email protected]

Quoted 13 lines
drivers/pci/quirks.c +13 -0
+static void quirk_reenable_pme(struct pci_dev *dev)
+{
+	bool suspend = FALSE;
+
+	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)&suspend);
+}
+
+static void quirk_disable_pme_suspend(struct pci_dev *dev)
+{
+	bool suspend = TRUE;
+
+	/* skip for runtime suspend */
+	if (pm_suspend_target_state == PM_SUSPEND_ON)

drivers/pci/quirks.c DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); · line 6260

Quoted 10 lines
drivers/pci/quirks.c +10 -0
+		return;
+
+	pci_walk_bus(dev->bus, modify_pme_amd_usb4, (void *)&suspend);
+}
+
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14b9, quirk_disable_pme_suspend);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14b9, quirk_reenable_pme);
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14eb, quirk_disable_pme_suspend);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14eb, quirk_reenable_pme);
+#endif /* CONFIG_SUSPEND */

PCI: Avoid PME from D3hot/cold for AMD Rembrandt and Phoenix

Iain reports that USB devices can't be used to wake a Lenovo Z13 from suspend. This occurs because on some AMD platforms, even though the Root Ports advertise PME_Support for D3hot and D3cold, they don't handle PME messages and generate wakeup interrupts from those states when amd-pmc has put the platform in a hardware sleep state.

Iain reported this on an AMD Rembrandt platform, but it also affects Phoenix SoCs. On Iain's system, a USB device below the affected Root Port generates the PME, but there is no reason to believe PMEs from other kinds of devices would work differently.

To avoid this issue, remove D3hot and D3cold from the Root Port's PME_Support mask when suspending if we expect amd-pmc to put the platform in a hardware sleep state. The effect of this is that if there is a wakeup device below the Root Port, we will avoid D3hot and D3cold.

Restore the advertised PME_Support mask on resume so D3hot and D3cold can be used by runtime suspend. The amd-pmc driver doesn't put the platform in a hardware sleep state for runtime suspend, so PMEs work as advertised.

---
  drivers/pci/quirks.c | 54 ++++++++++++++++++++++++++++++++++++++++++++
  1 file changed, 54 insertions(+)
drivers/pci/quirks.c +54 -0
index eeec1d6f9023..cdf03eb610b0 100644
@@ -6188,3 +6188,57 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size);
  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5020, of_pci_make_dev_node);
  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5021, of_pci_make_dev_node);
  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_REDHAT, 0x0005, of_pci_make_dev_node);
+
+#ifdef CONFIG_SUSPEND
+/*
+ * Root Ports on AMD Rembrandt and Phoenix SoCs advertise PME_Support for
+ * D3hot and D3cold, but if the SoC is put into a hardware sleep state by
+ * the amd-pmc driver, the Root Ports don't generate wakeup interrupts from
+ * those states.
+ *
+ * When suspending, remove D3hot and D3cold from the PME_Support advertised
+ * by the Root Port so we don't use those states if we're expecting wakeup
+ * interrupts.  Restore the advertised PME_Support when resuming.
+ */
+#define PCI_PM_CAP_D3_PME_MASK ((PCI_PM_CAP_PME_D3hot|PCI_PM_CAP_PME_D3cold) \
+				>> PCI_PM_CAP_PME_SHIFT)
+
+static void quirk_amd_pme_unsupported(struct pci_dev *dev)
+{
+	if (!dev->pm_cap)
+		return;
+
+	/*
+	 * PM_SUSPEND_ON means we're doing runtime suspend, which means
+	 * amd-pmc will not be involved so PMEs work as advertised.
+	 *
+	 * The PMEs *do* work in D3hot/D3cold if amd-pmc doesn't put the
+	 * SoC in the hardware sleep state, but we assume amd-pmc is always
+	 * present.
+	 */
+	if (pm_suspend_target_state == PM_SUSPEND_ON)
+		return;
+
+	dev->pme_support &= ~PCI_PM_CAP_D3_PME_MASK;
+	dev_info_once(&dev->dev, "D3hot/D3cold PMEs don't work in hardware sleep state\n");
+}
+
+static void quirk_amd_pme_supported(struct pci_dev *dev)
+{
+	u16 pmc;
+
+	if (!dev->pm_cap)
+		return;
+
+	if (dev->pme_support & PCI_PM_CAP_D3_PME_MASK)
+		return;
+
+	pci_read_config_word(dev, dev->pm_cap + PCI_PM_PMC, &pmc);
+	dev->pme_support = FIELD_GET(PCI_PM_CAP_PME_MASK, pmc);
+}
+
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14b9, quirk_amd_pme_unsupported);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14b9, quirk_amd_pme_supported);
+DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x14eb, quirk_amd_pme_unsupported);
+DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_AMD, 0x14eb, quirk_amd_pme_supported);
+#endif /* CONFIG_SUSPEND */
#7 lore raw

Re: [PATCH v20 2/2] PCI: Add a quirk for AMD PCIe root ports w/ USB4 controllers

Quoted 17 lines · Mario Limonciello

On 9/29/2023 14:44, Bjorn Helgaas wrote:
...

Quoted 3 lines

This is a lot of hassle to look for USB4 devices below the Root Port. You said earlier that these Root Ports *only* connect to xHCI and USB4 [1].

That's correct, but the part that I believe you're missing is that there are multiple "instances" of this PCI device ID on a failing system.

On an OEM Phoenix laptop in front of me I see 3 instances of 1022:14eb:
00:08.1, 00:08.2, and 00:08.3.

Quoted 3 lines

If that's the case, why even bother with the search? Why not just clear PCI_PM_CAP_PME_D3hot and PCI_PM_CAP_PME_D3cold unconditionally, e.g., the possible patch below?

Quoted 5 lines

Only the one with the USB4 controller (in this system it's at 00:08.3) has the problem.

So I believe the proposal you have below will apply the policy to too many ports.

Ah, thanks for clearing that up; I've been laboring under that misapprehension for a long time.

Bjorn

#8 lore raw

Re: [PATCH v20 2/2] PCI: Add a quirk for AMD PCIe root ports w/ USB4 controllers

In reply to Mario Limonciello
Quoted 2 lines · Mario Limonciello
drivers/pci/quirks.c +0 -0

[...]

Quoted 7 lines

+ * When AMD PCIe root ports with AMD USB4 controllers attached to them are put
+ * into D3hot or D3cold downstream USB devices may fail to wakeup the system
+ * from suspend to idle. This manifests as a missing wakeup interrupt.
+ *
+ * Prevent the associated root port from using PME to wake from D3hot or
+ * D3cold power states during suspend.
+ * This will effectively put the root port into D0 power state over suspend.

IIUC, the quirk matches for a Root Port, then searches for a USB controller below the Root Port, and if found, searches for the Root Port above again to clear or reinstate the PME bits.

That seems like a roundabout way of doing things. Have you considered matching for the USB controller's Device ID in the quirk, then checking whether the Root Port above has the Device ID which is known to be broken?

Also, since pci_d3cold_enable() / pci_d3cold_disable() are now fixed (can no longer be interfered with from user space), you might want to consider using them alternatively to clearing PME bits. They don't require access to config space.

Thanks,

Lukas

#9 lore raw

Re: [PATCH v20 2/2] PCI: Add a quirk for AMD PCIe root ports w/ USB4 controllers

Quoted 12 lines · Lukas Wunner
Quoted 2 lines · Mario Limonciello
drivers/pci/quirks.c +0 -0

[...]

Quoted 7 lines

+ * When AMD PCIe root ports with AMD USB4 controllers attached to them are put
+ * into D3hot or D3cold downstream USB devices may fail to wakeup the system
+ * from suspend to idle. This manifests as a missing wakeup interrupt.
+ *
+ * Prevent the associated root port from using PME to wake from D3hot or
+ * D3cold power states during suspend.
+ * This will effectively put the root port into D0 power state over suspend.

Quoted 9 lines

IIUC, the quirk matches for a Root Port, then searches for a USB controller below the Root Port, and if found, searches for the Root Port above again to clear or reinstate the PME bits.

That seems like a roundabout way of doing things. Have you considered matching for the USB controller's Device ID in the quirk, then checking whether the Root Port above has the Device ID which is known to be broken?

Yeah; this suggestion works. It's 8 quirks instead of 4, but it's much cleaner and easier to follow. I will send a v21 with this later on.

Quoted 5 lines

Also, since pci_d3cold_enable() / pci_d3cold_disable() are now fixed (can no longer be interfered with from user space), you might want to consider using them alternatively to clearing PME bits. They don't require access to config space.

I still don't like that userspace can potentially influence state selection policy. Separate from this quirk I'm going to send another patch for that.