Skip to contentSkip to conversation

dwc2 fixes for edge cases on hikey

PATCH In review John Stultz 2 participants 7 messages Oct 23, 2017 – Oct 24, 2017

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

v2 cover dwc2 fixes for edge cases on hikey

So here is v2 of a few dwc2 fixes that I've been using with HiKey. I'm not totally sure these are all ideal, but they avoid edge case issues that we have been running into with switching between gadget mode and host mode.

I'd guess the first two are potentially -stable material, and the last might be worth sending to -stable too, as its a relatively simple fix, but to my understanding the UDC state tracking has always been broken so its not really a regression. But still.

thanks
-john

Cc Wei Xu <[email protected]>
Cc Guodong Xu <[email protected]>
Cc Amit Pundir <[email protected]>
Cc YongQin Liu <[email protected]>
Cc John Youn <[email protected]>
Cc Minas Harutyunyan <[email protected]>
Cc Douglas Anderson <[email protected]>
Cc Chen Yu <[email protected]>
Cc Felipe Balbi <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>

John Stultz (3):
usb: dwc2: Improve gadget state disconnection handling
usb: dwc2: Error out of dwc2_hsotg_ep_disable() if we're in host mode
usb: dwc2: Fix UDC state tracking

 drivers/usb/dwc2/gadget.c | 7 +++++++
 drivers/usb/dwc2/hcd.c    | 7 +++++--
 2 files changed, 12 insertions(+), 2 deletions(-)
Signature · 1 line
2.7.4
#1 lore raw

PATCH v2 1/3 usb: dwc2: Improve gadget state disconnection handling

In the earlier commit dad3f793f20f ("usb: dwc2: Make sure we disconnect the gadget state"), I was trying to fix up the fact that we somehow weren't disconnecting the gadget state, so that when the OTG port was plugged in the second time we would get warnings about the state tracking being wrong.

(This seems to be due to a quirk of the HiKey board where we do not ever get any otg interrupts, particularly the session end detected signal. Instead we only see status change interrupt.)

The fix there was somewhat simple, as it just made sure to call dwc2_hsotg_disconnect() before we connected things up in OTG mode, ensuring the state handling didn't throw errors.

But in looking at a different issue I was seeing with UDC state handling, I realized that it would be much better to call dwc2_hsotg_disconnect when we get the state change signal moving to host mode.

Thus, this patch removes the earlier disconnect call I added and moves it (and the needed locking) to the host mode transition.

Cc Wei Xu <[email protected]>
Cc Guodong Xu <[email protected]>
Cc Amit Pundir <[email protected]>
Cc YongQin Liu <[email protected]>
Cc John Youn <[email protected]>
Cc Minas Harutyunyan <[email protected]>
Cc Douglas Anderson <[email protected]>
Cc Chen Yu <[email protected]>
Cc Felipe Balbi <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>

---
v2: Remove the extra dwc2_hsotg_core_init_disconnected() call
I had added, as suggested by Minas.

---
 drivers/usb/dwc2/hcd.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/usb/dwc2/hcd.c +5 -2
index c263114..9bd60ec 100644
@@ -3277,7 +3277,6 @@ static void dwc2_conn_id_status_change(struct work_struct *work)
 		dwc2_core_init(hsotg, false);
 		dwc2_enable_global_interrupts(hsotg);
 		spin_lock_irqsave(&hsotg->lock, flags);
-		dwc2_hsotg_disconnect(hsotg);
 		dwc2_hsotg_core_init_disconnected(hsotg, false);
 		spin_unlock_irqrestore(&hsotg->lock, flags);
 		dwc2_hsotg_core_connect(hsotg);
@@ -3296,8 +3295,12 @@ static void dwc2_conn_id_status_change(struct work_struct *work)
 		if (count > 250)
 			dev_err(hsotg->dev,
 				"Connection id status change timed out\n");
-		hsotg->op_state = OTG_STATE_A_HOST;
 
+		spin_lock_irqsave(&hsotg->lock, flags);
+		dwc2_hsotg_disconnect(hsotg);
+		spin_unlock_irqrestore(&hsotg->lock, flags);
+
+		hsotg->op_state = OTG_STATE_A_HOST;
 		/* Initialize the Core for Host mode */
 		dwc2_core_init(hsotg, false);
 		dwc2_enable_global_interrupts(hsotg);
Signature · 1 line
2.7.4
Acked-by Tested-by on v2 1/3
#2 lore raw

Re: [PATCH 1/3 v2] usb: dwc2: Improve gadget state disconnection handling

Quoted 60 lines · John Stultz

In the earlier commit dad3f793f20f ("usb: dwc2: Make sure we disconnect the gadget state"), I was trying to fix up the fact that we somehow weren't disconnecting the gadget state, so that when the OTG port was plugged in the second time we would get warnings about the state tracking being wrong.

(This seems to be due to a quirk of the HiKey board where we do not ever get any otg interrupts, particularly the session end detected signal. Instead we only see status change interrupt.)

The fix there was somewhat simple, as it just made sure to call dwc2_hsotg_disconnect() before we connected things up in OTG mode, ensuring the state handling didn't throw errors.

But in looking at a different issue I was seeing with UDC state handling, I realized that it would be much better to call dwc2_hsotg_disconnect when we get the state change signal moving to host mode.

Thus, this patch removes the earlier disconnect call I added and moves it (and the needed locking) to the host mode transition.

Cc Wei Xu <[email protected]>
Cc Guodong Xu <[email protected]>
Cc Amit Pundir <[email protected]>
Cc YongQin Liu <[email protected]>
Cc John Youn <[email protected]>
Cc Minas Harutyunyan <[email protected]>
Cc Douglas Anderson <[email protected]>
Cc Chen Yu <[email protected]>
Cc Felipe Balbi <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>

---
v2: Remove the extra dwc2_hsotg_core_init_disconnected() call
I had added, as suggested by Minas.

---
  drivers/usb/dwc2/hcd.c | 7 +++++--
  1 file changed, 5 insertions(+), 2 deletions(-)
drivers/usb/dwc2/hcd.c +0 -1
index c263114..9bd60ec 100644
@@ -3277,7 +3277,6 @@ static void dwc2_conn_id_status_change(struct work_struct *work)
  		dwc2_core_init(hsotg, false);
  		dwc2_enable_global_interrupts(hsotg);
  		spin_lock_irqsave(&hsotg->lock, flags);
-		dwc2_hsotg_disconnect(hsotg);
  		dwc2_hsotg_core_init_disconnected(hsotg, false);
  		spin_unlock_irqrestore(&hsotg->lock, flags);
  		dwc2_hsotg_core_connect(hsotg);
@@ -3296,8 +3295,12 @@ static void dwc2_conn_id_status_change(struct work_struct *work)
  		if (count > 250)
  			dev_err(hsotg->dev,
  				"Connection id status change timed out\n");

drivers/usb/dwc2/hcd.c static void dwc2_conn_id_status_change(struct work_struct *work) · line 3306

Quoted 10 lines
drivers/usb/dwc2/hcd.c +5 -1
-		hsotg->op_state = OTG_STATE_A_HOST;
 
+		spin_lock_irqsave(&hsotg->lock, flags);
+		dwc2_hsotg_disconnect(hsotg);
+		spin_unlock_irqrestore(&hsotg->lock, flags);
+
+		hsotg->op_state = OTG_STATE_A_HOST;
  		/* Initialize the Core for Host mode */
  		dwc2_core_init(hsotg, false);
  		dwc2_enable_global_interrupts(hsotg);

This patch is required for the HiKey platform, because the assertion of the "Connector ID status change" interrupt is different: asserting on B connector unplug and goes to Host mode.
No any side effect on SNPS HAPS-DX platform where ConnIDStsChng assertion is correct.

Thanks,
Minas

#3 lore raw

PATCH v2 2/3 usb: dwc2: Error out of dwc2_hsotg_ep_disable() if we're in host mode

In reply to John Stultz

We've found that while in host mode, using Android, if one runs the command:
stop adbd

The existing usb devices being utilized in host mode are disconnected. This is most visible with usb networking devices.

This seems to be due to adbd closing the file:
/dev/usb-ffs/adb/ep0
Which calls ffs_ep0_release() and the following backtrace:

[<ffffff800875a430>] dwc2_hsotg_ep_disable+0x148/0x150
[<ffffff800875a498>] dwc2_hsotg_udc_stop+0x60/0x110
[<ffffff8008787950>] usb_gadget_remove_driver+0x58/0x78
[<ffffff80087879e4>] usb_gadget_unregister_driver+0x74/0xe8
[<ffffff80087850c0>] unregister_gadget+0x28/0x58
[<ffffff800878511c>] unregister_gadget_item+0x2c/0x40
[<ffffff8008790ea8>] ffs_data_clear+0xe8/0xf8
[<ffffff8008790ed8>] ffs_data_reset+0x20/0x58
[<ffffff8008793218>] ffs_data_closed+0x98/0xe8
[<ffffff80087932d8>] ffs_ep0_release+0x20/0x30

Then when dwc2_hsotg_ep_disable() is called, we call kill_all_requests() which causes a bunch of the following messages:

dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
init: Service 'adbd' (pid 1915) killed by signal 9
init: Sending signal 9 to service 'adbd' (pid 1915) process group...
init: Successfully killed process cgroup uid 0 pid 1915 in 0ms
init: processing action (init.svc.adbd=stopped) from (/init.usb.configfs.rc:15) dwc2 f72c0000.usb: dwc2_hc_chhltd_intr_dma: Channel 8 - ChHltd set, but reason is unknown dwc2 f72c0000.usb: hcint 0x00000002, intsts 0x04200029
dwc2 f72c0000.usb: dwc2_hc_chhltd_intr_dma: Channel 12 - ChHltd set, but reason is unknown dwc2 f72c0000.usb: hcint 0x00000002, intsts 0x04200029
dwc2 f72c0000.usb: dwc2_hc_chhltd_intr_dma: Channel 15 - ChHltd set, but reason is unknown dwc2 f72c0000.usb: hcint 0x00000002, intsts 0x04200029
dwc2 f72c0000.usb: dwc2_hc_chhltd_intr_dma: Channel 3 - ChHltd set, but reason is unknown dwc2 f72c0000.usb: hcint 0x00000002, intsts 0x04200029
dwc2 f72c0000.usb: dwc2_hc_chhltd_intr_dma: Channel 4 - ChHltd set, but reason is unknown dwc2 f72c0000.usb: hcint 0x00000002, intsts 0x04200029
dwc2 f72c0000.usb: dwc2_update_urb_state_abn(): trimming xfer length

And the usb devices connected are basically hung at this point.

It seems like if we're in host mode, we probably shouldn't run the dwc2_hostg_ep_disable logic, so this patch returns an error in that case.

With this patch (along with the previous patch in this set), we avoid the mismatched interrupts and connected usb devices continue to function.

I'm not sure if some other solution would be better here, but this seems to work, so I wanted to send it out for input on what the right approach should be.

Cc: Wei Xu <[email protected]>
Cc: Guodong Xu <[email protected]>
Cc: Amit Pundir <[email protected]>
Cc: YongQin Liu <[email protected]>
Cc: John Youn <[email protected]>
Cc: Minas Harutyunyan <[email protected]>
Cc: Douglas Anderson <[email protected]>
Cc: Chen Yu <[email protected]>
Cc: Felipe Balbi <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: [email protected]
Tested by: Minas Harutyunyan <[email protected]>

---
 drivers/usb/dwc2/gadget.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/usb/dwc2/gadget.c +5 -0
index 0d8e09c..7fd0e38 100644
@@ -4004,6 +4004,11 @@ static int dwc2_hsotg_ep_disable(struct usb_ep *ep)
 		return -EINVAL;
 	}
 
+	if (hsotg->op_state != OTG_STATE_B_PERIPHERAL) {
+		dev_err(hsotg->dev, "%s: called in host mode?\n", __func__);
+		return -EINVAL;
+	}
+
 	epctrl_reg = dir_in ? DIEPCTL(index) : DOEPCTL(index);
 
 	spin_lock_irqsave(&hsotg->lock, flags);
Signature · 1 line
2.7.4
Acked-by Tested-by on v2 2/3
#4 lore raw

Re: [PATCH 2/3 v2] usb: dwc2: Error out of dwc2_hsotg_ep_disable() if we're in host mode

Quoted 88 lines · John Stultz

We've found that while in host mode, using Android, if one runs the command:
stop adbd

The existing usb devices being utilized in host mode are disconnected. This is most visible with usb networking devices.

This seems to be due to adbd closing the file:

   /dev/usb-ffs/adb/ep0

Which calls ffs_ep0_release() and the following backtrace:

[<ffffff800875a430>] dwc2_hsotg_ep_disable+0x148/0x150
[<ffffff800875a498>] dwc2_hsotg_udc_stop+0x60/0x110
[<ffffff8008787950>] usb_gadget_remove_driver+0x58/0x78
[<ffffff80087879e4>] usb_gadget_unregister_driver+0x74/0xe8
[<ffffff80087850c0>] unregister_gadget+0x28/0x58
[<ffffff800878511c>] unregister_gadget_item+0x2c/0x40
[<ffffff8008790ea8>] ffs_data_clear+0xe8/0xf8
[<ffffff8008790ed8>] ffs_data_reset+0x20/0x58
[<ffffff8008793218>] ffs_data_closed+0x98/0xe8
[<ffffff80087932d8>] ffs_ep0_release+0x20/0x30

Then when dwc2_hsotg_ep_disable() is called, we call kill_all_requests() which causes a bunch of the following messages:

dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
dwc2 f72c0000.usb: Mode Mismatch Interrupt: currently in Host mode
init: Service 'adbd' (pid 1915) killed by signal 9
init: Sending signal 9 to service 'adbd' (pid 1915) process group...
init: Successfully killed process cgroup uid 0 pid 1915 in 0ms
init: processing action (init.svc.adbd=stopped) from (/init.usb.configfs.rc:15) dwc2 f72c0000.usb: dwc2_hc_chhltd_intr_dma: Channel 8 - ChHltd set, but reason is unknown dwc2 f72c0000.usb: hcint 0x00000002, intsts 0x04200029
dwc2 f72c0000.usb: dwc2_hc_chhltd_intr_dma: Channel 12 - ChHltd set, but reason is unknown dwc2 f72c0000.usb: hcint 0x00000002, intsts 0x04200029
dwc2 f72c0000.usb: dwc2_hc_chhltd_intr_dma: Channel 15 - ChHltd set, but reason is unknown dwc2 f72c0000.usb: hcint 0x00000002, intsts 0x04200029
dwc2 f72c0000.usb: dwc2_hc_chhltd_intr_dma: Channel 3 - ChHltd set, but reason is unknown dwc2 f72c0000.usb: hcint 0x00000002, intsts 0x04200029
dwc2 f72c0000.usb: dwc2_hc_chhltd_intr_dma: Channel 4 - ChHltd set, but reason is unknown dwc2 f72c0000.usb: hcint 0x00000002, intsts 0x04200029
dwc2 f72c0000.usb: dwc2_update_urb_state_abn(): trimming xfer length

And the usb devices connected are basically hung at this point.

It seems like if we're in host mode, we probably shouldn't run the dwc2_hostg_ep_disable logic, so this patch returns an error in that case.

With this patch (along with the previous patch in this set), we avoid the mismatched interrupts and connected usb devices continue to function.

I'm not sure if some other solution would be better here, but this seems to work, so I wanted to send it out for input on what the right approach should be.

Cc: Wei Xu <[email protected]>
Cc: Guodong Xu <[email protected]>
Cc: Amit Pundir <[email protected]>
Cc: YongQin Liu <[email protected]>
Cc: John Youn <[email protected]>
Cc: Minas Harutyunyan <[email protected]>
Cc: Douglas Anderson <[email protected]>
Cc: Chen Yu <[email protected]>
Cc: Felipe Balbi <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: [email protected]
Tested by: Minas Harutyunyan <[email protected]>

---
  drivers/usb/dwc2/gadget.c | 5 +++++
  1 file changed, 5 insertions(+)
drivers/usb/dwc2/gadget.c +0 -0
index 0d8e09c..7fd0e38 100644
@@ -4004,6 +4004,11 @@ static int dwc2_hsotg_ep_disable(struct usb_ep *ep)
  		return -EINVAL;

drivers/usb/dwc2/gadget.c static int dwc2_hsotg_ep_disable(struct usb_ep *ep) · line 4014

Quoted 10 lines
drivers/usb/dwc2/gadget.c +5 -0
  	}
 
+	if (hsotg->op_state != OTG_STATE_B_PERIPHERAL) {
+		dev_err(hsotg->dev, "%s: called in host mode?\n", __func__);
+		return -EINVAL;
+	}
+
  	epctrl_reg = dir_in ? DIEPCTL(index) : DOEPCTL(index);
 
  	spin_lock_irqsave(&hsotg->lock, flags);
#5 lore raw

PATCH v2 3/3 usb: dwc2: Fix UDC state tracking

In reply to John Stultz

It has been noticed that the dwc2 udc state reporting doesn't seem to work (at least on HiKey boards). Where after the initial setup, the sysfs /sys/class/udc/f72c0000.usb/state file would report "configured" no matter the state of the OTG port.

This patch adds a call so that we report to the UDC layer when the gadget device is disconnected.

This patch does depend on the previous patch ("usb: dwc2:
Improve gadget state disconnection handling") in this patch set in order to properly work.

Cc: Wei Xu <[email protected]>
Cc: Guodong Xu <[email protected]>
Cc: Amit Pundir <[email protected]>
Cc: YongQin Liu <[email protected]>
Cc: John Youn <[email protected]>
Cc: Minas Harutyunyan <[email protected]>
Cc: Douglas Anderson <[email protected]>
Cc: Chen Yu <[email protected]>
Cc: Felipe Balbi <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: [email protected]
Tested by: Minas Harutyunyan <[email protected]>

---
 drivers/usb/dwc2/gadget.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/usb/dwc2/gadget.c +2 -0
index 7fd0e38..603c216 100644
@@ -3202,6 +3202,8 @@ void dwc2_hsotg_disconnect(struct dwc2_hsotg *hsotg)
 
 	call_gadget(hsotg, disconnect);
 	hsotg->lx_state = DWC2_L3;
+
+	usb_gadget_set_state(&hsotg->gadget, USB_STATE_NOTATTACHED);
 }
 
 /**
Signature · 1 line
2.7.4
Acked-by Tested-by on v2 3/3
#6 lore raw

Re: [PATCH 3/3 v2] usb: dwc2: Fix UDC state tracking

Quoted 36 lines · John Stultz

It has been noticed that the dwc2 udc state reporting doesn't seem to work (at least on HiKey boards). Where after the initial setup, the sysfs /sys/class/udc/f72c0000.usb/state file would report "configured" no matter the state of the OTG port.

This patch adds a call so that we report to the UDC layer when the gadget device is disconnected.

This patch does depend on the previous patch ("usb: dwc2:
Improve gadget state disconnection handling") in this patch set in order to properly work.

Cc: Wei Xu <[email protected]>
Cc: Guodong Xu <[email protected]>
Cc: Amit Pundir <[email protected]>
Cc: YongQin Liu <[email protected]>
Cc: John Youn <[email protected]>
Cc: Minas Harutyunyan <[email protected]>
Cc: Douglas Anderson <[email protected]>
Cc: Chen Yu <[email protected]>
Cc: Felipe Balbi <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: [email protected]
Tested by: Minas Harutyunyan <[email protected]>

---
  drivers/usb/dwc2/gadget.c | 2 ++
  1 file changed, 2 insertions(+)
drivers/usb/dwc2/gadget.c +0 -0
index 7fd0e38..603c216 100644

drivers/usb/dwc2/gadget.c void dwc2_hsotg_disconnect(struct dwc2_hsotg *hsotg) · line 3209

Quoted 8 lines
drivers/usb/dwc2/gadget.c +2 -0
@@ -3202,6 +3202,8 @@ void dwc2_hsotg_disconnect(struct dwc2_hsotg *hsotg)
 
  	call_gadget(hsotg, disconnect);
  	hsotg->lx_state = DWC2_L3;
+
+	usb_gadget_set_state(&hsotg->gadget, USB_STATE_NOTATTACHED);
  }
 
  /**