Skip to contentSkip to conversation

usb: pci-quirks: disable D3cold on xhci suspend for s2idle on AMD Renoir

PATCH In review Werner Sembach 3 participants 6 messages Jul 16, 2021 1 message not in the archive

Subsystems (3)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

PATCH v2 1/1 usb: pci-quirks: disable D3cold on xhci suspend for s2idle on AMD Renoir

From: Mario Limonciello <[email protected]>

The XHCI controller is required to enter D3hot rather than D3cold for AMD s2idle on this hardware generation.

Otherwise, the 'Controller Not Ready' (CNR) bit is not being cleared by host in resume and eventually this results in xhci resume failures during the s2idle wakeup.

---
 drivers/usb/host/xhci-pci.c | 7 ++++++-
 drivers/usb/host/xhci.h     | 1 +
 2 files changed, 7 insertions(+), 1 deletion(-)
drivers/usb/host/xhci-pci.c +6 -1
index 9568b907fd05..ceccd27cf4d0 100644
@@ -59,6 +59,7 @@
 #define PCI_DEVICE_ID_INTEL_MAPLE_RIDGE_XHCI		0x1138
 #define PCI_DEVICE_ID_INTEL_ALDER_LAKE_XHCI		0x461e
 
+#define PCI_DEVICE_ID_AMD_RENOIR_XHCI			0x1639
 #define PCI_DEVICE_ID_AMD_PROMONTORYA_4			0x43b9
 #define PCI_DEVICE_ID_AMD_PROMONTORYA_3			0x43ba
 #define PCI_DEVICE_ID_AMD_PROMONTORYA_2			0x43bb
@@ -182,6 +183,10 @@ static void xhci_pci_quirks(struct device *dev, struct xhci_hcd *xhci)
 		(pdev->device == PCI_DEVICE_ID_AMD_PROMONTORYA_1)))
 		xhci->quirks |= XHCI_U2_DISABLE_WAKE;
 
+	if (pdev->vendor == PCI_VENDOR_ID_AMD &&
+		pdev->device == PCI_DEVICE_ID_AMD_RENOIR_XHCI)
+		xhci->quirks |= XHCI_BROKEN_D3COLD;
+
 	if (pdev->vendor == PCI_VENDOR_ID_INTEL) {
 		xhci->quirks |= XHCI_LPM_SUPPORT;
 		xhci->quirks |= XHCI_INTEL_HOST;
@@ -548,7 +553,7 @@ static int xhci_pci_suspend(struct usb_hcd *hcd, bool do_wakeup)
 	 * Systems with the TI redriver that loses port status change events
 	 * need to have the registers polled during D3, so avoid D3cold.
 	 */
-	if (xhci->quirks & XHCI_COMP_MODE_QUIRK)
+	if (xhci->quirks & (XHCI_COMP_MODE_QUIRK | XHCI_BROKEN_D3COLD))
 		pci_d3cold_disable(pdev);
 
 	if (xhci->quirks & XHCI_PME_STUCK_QUIRK)
drivers/usb/host/xhci.h +1 -0
index 3190fd570c57..80a43b843af7 100644
@@ -1884,6 +1884,7 @@ struct xhci_hcd {
 #define XHCI_DISABLE_SPARSE	BIT_ULL(38)
 #define XHCI_SG_TRB_CACHE_SIZE_QUIRK	BIT_ULL(39)
 #define XHCI_NO_SOFT_RETRY	BIT_ULL(40)
+#define XHCI_BROKEN_D3COLD	BIT_ULL(41)
 
 	unsigned int		num_active_eps;
 	unsigned int		limit_active_eps;
Signature · 1 line
2.25.1
#2 lore raw
Quoted 19 lines · Werner Sembach

From: Mario Limonciello <[email protected]>

The XHCI controller is required to enter D3hot rather than D3cold for AMD s2idle on this hardware generation.

Otherwise, the 'Controller Not Ready' (CNR) bit is not being cleared by host in resume and eventually this results in xhci resume failures during the s2idle wakeup.

Quoted 5 lines
---
 drivers/usb/host/xhci-pci.c | 7 ++++++-
 drivers/usb/host/xhci.h     | 1 +
 2 files changed, 7 insertions(+), 1 deletion(-)

Any reason you resent us a patch that is already in a stable release?

And why not just use the stable kernel trees as-is? Why attempt to cherry-pick random portions of them?

thanks,

greg k-h

#3 lore raw

Am 16.07.21 um 14:50 schrieb Greg Kroah-Hartman:

Quoted 25 lines · Werner Sembach
Quoted 19 lines

From: Mario Limonciello <[email protected]>

The XHCI controller is required to enter D3hot rather than D3cold for AMD s2idle on this hardware generation.

Otherwise, the 'Controller Not Ready' (CNR) bit is not being cleared by host in resume and eventually this results in xhci resume failures during the s2idle wakeup.

Quoted 5 lines
---
 drivers/usb/host/xhci-pci.c | 7 ++++++-
 drivers/usb/host/xhci.h     | 1 +
 2 files changed, 7 insertions(+), 1 deletion(-)
Quoted 8 lines

Any reason you resent us a patch that is already in a stable release?

And why not just use the stable kernel trees as-is? Why attempt to cherry-pick random portions of them?

thanks,

greg k-h

I didn't add the mailing list as recipent for my last replies so here again:

I only checked the Ubuntu 5.11 tree where the patch is actually missing.

The 5.8 kernel has other issues because of outdated amdgpu, that's why we never checked the 5.4 kernel.

Testing for 5.4: often hangs on boot before display manager shows up

5.4 + amdgpu-dkms from here: https://www.amd.com/en/support/kb/release-notes/rn-amdgpu-unified-linux-21-20 : Hang on boot issue gone, but does not suspend anymore, and has graphic glitches.

Should I add these findings to the SRU?

#4 lore raw
Quoted 40 lines · Werner Sembach

Am 16.07.21 um 14:50 schrieb Greg Kroah-Hartman:

Quoted 25 lines · Werner Sembach
Quoted 19 lines

From: Mario Limonciello <[email protected]>

The XHCI controller is required to enter D3hot rather than D3cold for AMD s2idle on this hardware generation.

Otherwise, the 'Controller Not Ready' (CNR) bit is not being cleared by host in resume and eventually this results in xhci resume failures during the s2idle wakeup.

Quoted 5 lines
---
 drivers/usb/host/xhci-pci.c | 7 ++++++-
 drivers/usb/host/xhci.h     | 1 +
 2 files changed, 7 insertions(+), 1 deletion(-)
Quoted 8 lines

Any reason you resent us a patch that is already in a stable release?

And why not just use the stable kernel trees as-is? Why attempt to cherry-pick random portions of them?

thanks,

greg k-h

I didn't add the mailing list as recipent for my last replies so here again:

I only checked the Ubuntu 5.11 tree where the patch is actually missing.

Quoted 8 lines

The 5.8 kernel has other issues because of outdated amdgpu, that's why we never checked the 5.4 kernel.

Testing for 5.4: often hangs on boot before display manager shows up

5.4 + amdgpu-dkms from here: https://www.amd.com/en/support/kb/release-notes/rn-amdgpu-unified-linux-21-20 : Hang on boot issue gone, but does not suspend anymore, and has graphic glitches.

Should I add these findings to the SRU?

Hi Werner,

This patch is specific to Ubuntu so it should not have been sent to LKML (stable, Greg and others), which you did. I guess that's what Greg meant.

When sending such backports from your tree, be sure to set "suppresscc=all" in your .git/config or use appropriate argument for git send-email.

Best regards,
Krzysztof

#5 lore raw

Am 16.07.21 um 17:30 schrieb Krzysztof Kozlowski:

Quoted 52 lines · Werner Sembach
Quoted 38 lines

Am 16.07.21 um 14:50 schrieb Greg Kroah-Hartman:

Quoted 25 lines · Werner Sembach
Quoted 19 lines

From: Mario Limonciello <[email protected]>

The XHCI controller is required to enter D3hot rather than D3cold for AMD s2idle on this hardware generation.

Otherwise, the 'Controller Not Ready' (CNR) bit is not being cleared by host in resume and eventually this results in xhci resume failures during the s2idle wakeup.

Quoted 5 lines
---
  drivers/usb/host/xhci-pci.c | 7 ++++++-
  drivers/usb/host/xhci.h     | 1 +
  2 files changed, 7 insertions(+), 1 deletion(-)
Quoted 8 lines

Any reason you resent us a patch that is already in a stable release?

And why not just use the stable kernel trees as-is? Why attempt to cherry-pick random portions of them?

thanks,

greg k-h

I didn't add the mailing list as recipent for my last replies so here again:

I only checked the Ubuntu 5.11 tree where the patch is actually missing.

Quoted 8 lines

The 5.8 kernel has other issues because of outdated amdgpu, that's why we never checked the 5.4 kernel.

Testing for 5.4: often hangs on boot before display manager shows up

5.4 + amdgpu-dkms from here: https://www.amd.com/en/support/kb/release-notes/rn-amdgpu-unified-linux-21-20 : Hang on boot issue gone, but does not suspend anymore, and has graphic glitches.

Should I add these findings to the SRU?

Hi Werner,

This patch is specific to Ubuntu so it should not have been sent to LKML (stable, Greg and others), which you did. I guess that's what Greg meant.

Quoted 6 lines

When sending such backports from your tree, be sure to set "suppresscc=all" in your .git/config or use appropriate argument for git send-email.

Best regards,
Krzysztof

Hi,

now I see, thanks for spelling it out, I was a little bit slow today. I didn't realize that git send-email added non Ubuntu mailings lists because of the cc in the commit message (besides already should know this because I did it once before by accident).

I'm really really sorry, for all the confusion I caused.

Kind regards

Werner Sembach

#6 lore raw
In reply to Werner Sembach
Quoted 40 lines · Werner Sembach

Am 16.07.21 um 14:50 schrieb Greg Kroah-Hartman:

Quoted 25 lines · Werner Sembach
Quoted 19 lines

From: Mario Limonciello <[email protected]>

The XHCI controller is required to enter D3hot rather than D3cold for AMD s2idle on this hardware generation.

Otherwise, the 'Controller Not Ready' (CNR) bit is not being cleared by host in resume and eventually this results in xhci resume failures during the s2idle wakeup.

Quoted 5 lines
---
 drivers/usb/host/xhci-pci.c | 7 ++++++-
 drivers/usb/host/xhci.h     | 1 +
 2 files changed, 7 insertions(+), 1 deletion(-)
Quoted 8 lines

Any reason you resent us a patch that is already in a stable release?

And why not just use the stable kernel trees as-is? Why attempt to cherry-pick random portions of them?

thanks,

greg k-h

I didn't add the mailing list as recipent for my last replies so here again:

I only checked the Ubuntu 5.11 tree where the patch is actually missing.

Quoted 8 lines

The 5.8 kernel has other issues because of outdated amdgpu, that's why we never checked the 5.4 kernel.

Testing for 5.4: often hangs on boot before display manager shows up

5.4 + amdgpu-dkms from here: https://www.amd.com/en/support/kb/release-notes/rn-amdgpu-unified-linux-21-20 : Hang on boot issue gone, but does not suspend anymore, and has graphic glitches.

Should I add these findings to the SRU?

What does any of this have to do with the [email protected] developers?