Skip to contentSkip to conversation

usb: dwc3: Disable AutoRetry controller feature for dwc_usb3 v2.00a

PATCH In review Jakub Vanek 2 participants 6 messages Jun 30, 2023 – Jul 12, 2023

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 usb: dwc3: Disable AutoRetry controller feature for dwc_usb3 v2.00a

AutoRetry has been found to cause issues in this controller revision. This feature allows the controller to send non-terminating/burst retry ACKs (Retry=1 and Nump!=0) as opposed to terminating retry ACKs (Retry=1 and Nump=0) to devices when a transaction error occurs.

Unfortunately, some USB devices do not retry transfers when the controller sends them a non-terminating retry ACK. After the transfer times out, the xHCI driver tries to resume normal operation of the controller by sending a Stop Endpoint command to it. However, this revision of the controller fails to respond to that command in this state and the xHCI driver therefore gives up. This is reported via dmesg:

[sda] tag#29 uas_eh_abort_handler 0 uas-tag 1 inflight: CMD IN [sda] tag#29 CDB: opcode=0x28 28 00 00 69 42 80 00 00 48 00
xhci-hcd: xHCI host not responding to stop endpoint command
xhci-hcd: xHCI host controller not responding, assume dead
xhci-hcd: HC died; cleaning up

This problem has been observed on Odroid HC2. This board has an integrated JMS578 USB3-to-SATA bridge. The above problem could be triggered by starting a read-heavy workload on an attached SSD. After a while, the host controller would die and the SSD would disappear from the system.

Reverting b138e23d3dff ("usb: dwc3: core: Enable AutoRetry feature in the controller") stopped the issue from occurring on Odroid HC2. As this problem hasn't been reported on other devices, disable AutoRetry just for the dwc_usb3 revision v2.00a that the board SoC (Exynos 5422) uses.

Fixes b138e23d3dff ("usb: dwc3: core: Enable AutoRetry feature in the controller")
Cc Mauro Ribeiro <[email protected]>
Cc Krzysztof Kozlowski <[email protected]>
---
 drivers/usb/dwc3/core.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/usb/dwc3/core.c +8 -2
index d68958e151a7..d742fc882d7e 100644
@@ -1218,9 +1218,15 @@ static int dwc3_core_init(struct dwc3 *dwc)
 		 * Host mode on seeing transaction errors(CRC errors or internal
 		 * overrun scenerios) on IN transfers to reply to the device
 		 * with a non-terminating retry ACK (i.e, an ACK transcation
-		 * packet with Retry=1 & Nump != 0)
+		 * packet with Retry=1 & Nump != 0).
+		 *
+		 * Do not enable this for DWC_usb3 v2.00a. This controller
+		 * revision stops responding to Stop Endpoint commands if
+		 * a USB device does not retry after a non-terminating retry
+		 * ACK is sent to it.
 		 */
-		reg |= DWC3_GUCTL_HSTINAUTORETRY;
+		if (!DWC3_VER_IS(DWC3, 200A))
+			reg |= DWC3_GUCTL_HSTINAUTORETRY;
 
 		dwc3_writel(dwc->regs, DWC3_GUCTL, reg);
 	}
Signature · 1 line
2.25.1
maintainer
#1 lore raw

Hi,

Quoted 67 lines · Jakub Vanek

AutoRetry has been found to cause issues in this controller revision. This feature allows the controller to send non-terminating/burst retry ACKs (Retry=1 and Nump!=0) as opposed to terminating retry ACKs (Retry=1 and Nump=0) to devices when a transaction error occurs.

Unfortunately, some USB devices do not retry transfers when the controller sends them a non-terminating retry ACK. After the transfer times out, the xHCI driver tries to resume normal operation of the controller by sending a Stop Endpoint command to it. However, this revision of the controller fails to respond to that command in this state and the xHCI driver therefore gives up. This is reported via dmesg:

[sda] tag#29 uas_eh_abort_handler 0 uas-tag 1 inflight: CMD IN [sda] tag#29 CDB: opcode=0x28 28 00 00 69 42 80 00 00 48 00
xhci-hcd: xHCI host not responding to stop endpoint command
xhci-hcd: xHCI host controller not responding, assume dead
xhci-hcd: HC died; cleaning up

This problem has been observed on Odroid HC2. This board has an integrated JMS578 USB3-to-SATA bridge. The above problem could be triggered by starting a read-heavy workload on an attached SSD. After a while, the host controller would die and the SSD would disappear from the system.

Reverting b138e23d3dff ("usb: dwc3: core: Enable AutoRetry feature in the controller") stopped the issue from occurring on Odroid HC2. As this problem hasn't been reported on other devices, disable AutoRetry just for the dwc_usb3 revision v2.00a that the board SoC (Exynos 5422) uses.

---
 drivers/usb/dwc3/core.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/usb/dwc3/core.c +8 -2
index d68958e151a7..d742fc882d7e 100644
@@ -1218,9 +1218,15 @@ static int dwc3_core_init(struct dwc3 *dwc)
 		 * Host mode on seeing transaction errors(CRC errors or internal
 		 * overrun scenerios) on IN transfers to reply to the device
 		 * with a non-terminating retry ACK (i.e, an ACK transcation
-		 * packet with Retry=1 & Nump != 0)
+		 * packet with Retry=1 & Nump != 0).
+		 *
+		 * Do not enable this for DWC_usb3 v2.00a. This controller
+		 * revision stops responding to Stop Endpoint commands if
+		 * a USB device does not retry after a non-terminating retry
+		 * ACK is sent to it.
 		 */
-		reg |= DWC3_GUCTL_HSTINAUTORETRY;
+		if (!DWC3_VER_IS(DWC3, 200A))
+			reg |= DWC3_GUCTL_HSTINAUTORETRY;
 
 		dwc3_writel(dwc->regs, DWC3_GUCTL, reg);
 	}
Signature · 1 line
2.25.1

I brought up this inquiry internally. Please wait as I need to review this further. The handling for this may be different depending on the team's feedback.

Thanks,
Thinh

#2 lore raw
Quoted 66 lines · Thinh Nguyen

Hi,

Quoted 63 lines · Jakub Vanek

AutoRetry has been found to cause issues in this controller revision. This feature allows the controller to send non-terminating/burst retry ACKs (Retry=1 and Nump!=0) as opposed to terminating retry ACKs (Retry=1 and Nump=0) to devices when a transaction error occurs.

Unfortunately, some USB devices do not retry transfers when the controller sends them a non-terminating retry ACK. After the transfer times out, the xHCI driver tries to resume normal operation of the controller by sending a Stop Endpoint command to it. However, this revision of the controller fails to respond to that command in this state and the xHCI driver therefore gives up. This is reported via dmesg:

[sda] tag#29 uas_eh_abort_handler 0 uas-tag 1 inflight: CMD IN [sda] tag#29 CDB: opcode=0x28 28 00 00 69 42 80 00 00 48 00
xhci-hcd: xHCI host not responding to stop endpoint command
xhci-hcd: xHCI host controller not responding, assume dead
xhci-hcd: HC died; cleaning up

This problem has been observed on Odroid HC2. This board has an integrated JMS578 USB3-to-SATA bridge. The above problem could be triggered by starting a read-heavy workload on an attached SSD. After a while, the host controller would die and the SSD would disappear from the system.

Reverting b138e23d3dff ("usb: dwc3: core: Enable AutoRetry feature in the controller") stopped the issue from occurring on Odroid HC2. As this problem hasn't been reported on other devices, disable AutoRetry just for the dwc_usb3 revision v2.00a that the board SoC (Exynos 5422) uses.

---
 drivers/usb/dwc3/core.c | 10 ++++++++--

 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/usb/dwc3/core.c +8 -2
index d68958e151a7..d742fc882d7e 100644
@@ -1218,9 +1218,15 @@ static int dwc3_core_init(struct dwc3 *dwc)
                  * Host mode on seeing transaction errors(CRC errors or internal
                  * overrun scenerios) on IN transfers to reply to the device
                  * with a non-terminating retry ACK (i.e, an ACK transcation
-                * packet with Retry=1 & Nump != 0)
+                * packet with Retry=1 & Nump != 0).
+                *
+                * Do not enable this for DWC_usb3 v2.00a. This controller
+                * revision stops responding to Stop Endpoint commands if
+                * a USB device does not retry after a non-terminating retry
+                * ACK is sent to it.
                  */
-               reg |= DWC3_GUCTL_HSTINAUTORETRY;
+               if (!DWC3_VER_IS(DWC3, 200A))
+                       reg |= DWC3_GUCTL_HSTINAUTORETRY;


                dwc3_writel(dwc->regs, DWC3_GUCTL, reg);

Quoted 9 lines
Quoted 4 lines

        }

Signature · 1 line
2.25.1

I brought up this inquiry internally. Please wait as I need to review this further. The handling for this may be different depending on the team's feedback.

OK; feel free to contact me if I could be of any help.

Quoted 2 lines

Thanks,
Thinh

Thank you,
Jakub

maintainer
#3 lore raw

Hi Jakub,

Quoted 77 lines · Jakub Vaněk
Quoted 66 lines · Thinh Nguyen

Hi,

Quoted 63 lines · Jakub Vanek

AutoRetry has been found to cause issues in this controller revision. This feature allows the controller to send non-terminating/burst retry ACKs (Retry=1 and Nump!=0) as opposed to terminating retry ACKs (Retry=1 and Nump=0) to devices when a transaction error occurs.

Unfortunately, some USB devices do not retry transfers when the controller sends them a non-terminating retry ACK. After the transfer times out, the xHCI driver tries to resume normal operation of the controller by sending a Stop Endpoint command to it. However, this revision of the controller fails to respond to that command in this state and the xHCI driver therefore gives up. This is reported via dmesg:

[sda] tag#29 uas_eh_abort_handler 0 uas-tag 1 inflight: CMD IN [sda] tag#29 CDB: opcode=0x28 28 00 00 69 42 80 00 00 48 00
xhci-hcd: xHCI host not responding to stop endpoint command
xhci-hcd: xHCI host controller not responding, assume dead
xhci-hcd: HC died; cleaning up

This problem has been observed on Odroid HC2. This board has an integrated JMS578 USB3-to-SATA bridge. The above problem could be triggered by starting a read-heavy workload on an attached SSD. After a while, the host controller would die and the SSD would disappear from the system.

Reverting b138e23d3dff ("usb: dwc3: core: Enable AutoRetry feature in the controller") stopped the issue from occurring on Odroid HC2. As this problem hasn't been reported on other devices, disable AutoRetry just for the dwc_usb3 revision v2.00a that the board SoC (Exynos 5422) uses.

---
 drivers/usb/dwc3/core.c | 10 ++++++++--

 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/usb/dwc3/core.c +8 -2
index d68958e151a7..d742fc882d7e 100644
@@ -1218,9 +1218,15 @@ static int dwc3_core_init(struct dwc3 *dwc)
                  * Host mode on seeing transaction errors(CRC errors or internal
                  * overrun scenerios) on IN transfers to reply to the device
                  * with a non-terminating retry ACK (i.e, an ACK transcation
-                * packet with Retry=1 & Nump != 0)
+                * packet with Retry=1 & Nump != 0).
+                *
+                * Do not enable this for DWC_usb3 v2.00a. This controller
+                * revision stops responding to Stop Endpoint commands if
+                * a USB device does not retry after a non-terminating retry
+                * ACK is sent to it.
                  */
-               reg |= DWC3_GUCTL_HSTINAUTORETRY;
+               if (!DWC3_VER_IS(DWC3, 200A))
+                       reg |= DWC3_GUCTL_HSTINAUTORETRY;


                dwc3_writel(dwc->regs, DWC3_GUCTL, reg);

Quoted 9 lines
Quoted 4 lines

        }

Signature · 1 line
2.25.1

I brought up this inquiry internally. Please wait as I need to review this further. The handling for this may be different depending on the team's feedback.

Quoted 2 lines

OK; feel free to contact me if I could be of any help.

Sorry for the delay. After discussion, this behavior is the same across different controller versions. The consensus is to disable this feature altoghether. There will not be any noticeable performance impact.

Would you mind revise the patch to disable this feature for all?

Many thanks,
Thinh

#4 lore raw

Hi Thinh,

Quoted 110 lines · Thinh Nguyen

Hi Jakub,

Quoted 104 lines · Jakub Vaněk
Quoted 93 lines · Thinh Nguyen

Hi,

Quoted 90 lines · Jakub Vanek

AutoRetry has been found to cause issues in this controller revision.
This feature allows the controller to send non- terminating/burst retry
ACKs (Retry=1 and Nump!=0) as opposed to terminating retry ACKs (Retry=1 and Nump=0) to devices when a transaction error occurs.

Unfortunately, some USB devices do not retry transfers when the controller sends them a non-terminating retry ACK. After the transfer
times out, the xHCI driver tries to resume normal operation of the
controller by sending a Stop Endpoint command to it. However, this
revision of the controller fails to respond to that command in this
state and the xHCI driver therefore gives up. This is reported via dmesg:

[sda] tag#29 uas_eh_abort_handler 0 uas-tag 1 inflight: CMD IN [sda] tag#29 CDB: opcode=0x28 28 00 00 69 42 80 00 00 48 00
xhci-hcd: xHCI host not responding to stop endpoint command
xhci-hcd: xHCI host controller not responding, assume dead
xhci-hcd: HC died; cleaning up

This problem has been observed on Odroid HC2. This board has an integrated JMS578 USB3-to-SATA bridge. The above problem could be
triggered by starting a read-heavy workload on an attached SSD. After a while, the host controller would die and the SSD would disappear
from the system.

Reverting b138e23d3dff ("usb: dwc3: core: Enable AutoRetry feature in
the controller") stopped the issue from occurring on Odroid HC2.
As this problem hasn't been reported on other devices, disable AutoRetry just for the dwc_usb3 revision v2.00a that the board SoC
(Exynos 5422) uses.

Fixes b138e23d3dff ("usb: dwc3: core: Enable AutoRetry feature

in the controller")
Link:
https://urldefense.com/v3/__https://lore.kernel.org/r/[email protected]/__;!!A4F2R9G_pg!YWgF6oqfuVY6xstZmroukjlrwAFEYEQE8uj_iUu4fd10mnJxEPG345k75dMLLdNFP8rT1leok-aPNkz_FuAJ1zxnmw$

Link:
https://urldefense.com/v3/__https://forum.odroid.com/viewtopic.php?t=42630__;!!A4F2R9G_pg!YWgF6oqfuVY6xstZmroukjlrwAFEYEQE8uj_iUu4fd10mnJxEPG345k75dMLLdNFP8rT1leok-aPNkz_FuCzOGIVPA$

---
 drivers/usb/dwc3/core.c | 10 ++++++++--

 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/usb/dwc3/core.c +0 -0
index d68958e151a7..d742fc882d7e 100644
@@ -1218,9 +1218,15 @@ static int dwc3_core_init(struct dwc3
 *dwc)
                  * Host mode on seeing transaction errors(CRC

errors or internal
                 * overrun scenerios) on IN transfers to reply to the device
                 * with a non-terminating retry ACK (i.e, an ACK transcation
-                * packet with Retry=1 & Nump != 0)
+                * packet with Retry=1 & Nump != 0). +                *
+                * Do not enable this for DWC_usb3 v2.00a. This controller
+                * revision stops responding to Stop Endpoint commands if
+                * a USB device does not retry after a non- terminating retry
+                * ACK is sent to it.
                 */
-               reg |= DWC3_GUCTL_HSTINAUTORETRY;
+               if (!DWC3_VER_IS(DWC3, 200A))
+                       reg |= DWC3_GUCTL_HSTINAUTORETRY;
                dwc3_writel(dwc->regs, DWC3_GUCTL, reg); }

Quoted 9 lines
Quoted 2 lines

2.25.1

I brought up this inquiry internally. Please wait as I need to review
this further. The handling for this may be different depending on the
team's feedback.

Quoted 2 lines

OK; feel free to contact me if I could be of any help.

Quoted 7 lines

Sorry for the delay. After discussion, this behavior is the same across
different controller versions. The consensus is to disable this feature
altoghether. There will not be any noticeable performance impact.

Would you mind revise the patch to disable this feature for all?

Not at all, I have sent a new revision of the patch:
https://lore.kernel.org/linux-usb/[email protected]/

Reverting the commit altogether seemed preferable to me to updating the revision-specific patch. However, if the other way would be better, I can send an updated revision.

Quoted 2 lines

Many thanks,
Thinh

Thank you,
Jakub

maintainer
#5 lore raw
Quoted 23 lines · Jakub Vaněk

Hi Thinh,

Quoted 20 lines · Thinh Nguyen

Hi Jakub,

Quoted 10 lines · Jakub Vaněk
Quoted 6 lines · Thinh Nguyen

I brought up this inquiry internally. Please wait as I need to review
this further. The handling for this may be different depending on the
team's feedback.

OK; feel free to contact me if I could be of any help.

Sorry for the delay. After discussion, this behavior is the same across
different controller versions. The consensus is to disable this feature
altoghether. There will not be any noticeable performance impact.

Quoted 9 lines
Quoted 1 line

Would you mind revise the patch to disable this feature for all?

Not at all, I have sent a new revision of the patch:
https://urldefense.com/v3/__https://lore.kernel.org/linux-usb/[email protected]/__;!!A4F2R9G_pg!euvmvJVr_AMCpyd59tiav4Co4u8p0TKzsLq0W7yKoyIlN_K7WFv2GP57wXmsddsJoxsEq6_Bhx8yp5gG8vAqebKhkw$

Reverting the commit altogether seemed preferable to me to updating the revision-specific patch. However, if the other way would be better, I can send an updated revision.

Reverting it is fine also. We'll keep the default setting from CoreConsultant.

Thanks,
Thinh