In some cases, there is a small-time gap in which CMD_RING_BUSY can be cleared by controller but adding command completion event to event ring will be delayed. As the result driver will return error code. This behavior has been detected on usbtest driver (test 9) with configuration including ep1in/ep1out bulk and ep2in/ep2out isoc endpoint. Probably this gap occurred because controller was busy with adding some other events to event ring. The CMD_RING_BUSY is cleared to '0' when the Command Descriptor has been executed and not when command completion event has been added to event ring.
To fix this issue for this test the small delay is sufficient less than 10us) but to make sure the problem doesn't happen again in the future the patch introduce 3 retries to check with delay about 100us before returning error code
Fixes: 3d82904559f4 ("usb: cdnsp: cdns3 Add main part of Cadence USBSSP DRD Driver") cc: stable@vger.kernel.org Signed-off-by: Pawel Laszczak pawell@cadence.com --- drivers/usb/cdns3/cdnsp-gadget.c | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-)
diff --git a/drivers/usb/cdns3/cdnsp-gadget.c b/drivers/usb/cdns3/cdnsp-gadget.c index f773518185c9..0eb11b5dd9d3 100644 --- a/drivers/usb/cdns3/cdnsp-gadget.c +++ b/drivers/usb/cdns3/cdnsp-gadget.c @@ -547,6 +547,7 @@ int cdnsp_wait_for_cmd_compl(struct cdnsp_device *pdev) dma_addr_t cmd_deq_dma; union cdnsp_trb *event; u32 cycle_state; + u32 retry = 3; int ret, val; u64 cmd_dma; u32 flags; @@ -578,8 +579,23 @@ int cdnsp_wait_for_cmd_compl(struct cdnsp_device *pdev) flags = le32_to_cpu(event->event_cmd.flags);
/* Check the owner of the TRB. */ - if ((flags & TRB_CYCLE) != cycle_state) + if ((flags & TRB_CYCLE) != cycle_state) { + /* + *Give some extra time to get chance controller + * to finish command before returning error code. + * Checking CMD_RING_BUSY is not sufficient because + * this bit is cleared to '0' when the Command + * Descriptor has been executed by controller + * and not when command completion event has + * be added to event ring. + */ + if (retry--) { + usleep_range(90, 100); + continue; + } + return -EINVAL; + }
cmd_dma = le64_to_cpu(event->event_cmd.cmd_trb);
In some cases, there is a small-time gap in which CMD_RING_BUSY can be cleared by controller but adding command completion event to event ring will be delayed. As the result driver will return error code. This behavior has been detected on usbtest driver (test 9) with configuration including ep1in/ep1out bulk and ep2in/ep2out isoc endpoint. Probably this gap occurred because controller was busy with adding some other events to event ring. The CMD_RING_BUSY is cleared to '0' when the Command Descriptor has been executed and not when command completion event has been added to event ring.
To fix this issue for this test the small delay is sufficient less than 10us) but to make sure the problem doesn't happen again in the future the patch introduce 3 retries to check with delay about 100us before returning error code
Fixes: 3d82904559f4 ("usb: cdnsp: cdns3 Add main part of Cadence USBSSP DRD Driver") cc: stable@vger.kernel.org Signed-off-by: Pawel Laszczak pawell@cadence.com
drivers/usb/cdns3/cdnsp-gadget.c | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-)
diff --git a/drivers/usb/cdns3/cdnsp-gadget.c b/drivers/usb/cdns3/cdnsp- gadget.c index f773518185c9..0eb11b5dd9d3 100644 --- a/drivers/usb/cdns3/cdnsp-gadget.c +++ b/drivers/usb/cdns3/cdnsp-gadget.c @@ -547,6 +547,7 @@ int cdnsp_wait_for_cmd_compl(struct cdnsp_device *pdev) dma_addr_t cmd_deq_dma; union cdnsp_trb *event; u32 cycle_state;
- u32 retry = 3; int ret, val; u64 cmd_dma; u32 flags;
@@ -578,8 +579,23 @@ int cdnsp_wait_for_cmd_compl(struct cdnsp_device *pdev) flags = le32_to_cpu(event->event_cmd.flags);
/* Check the owner of the TRB. */
if ((flags & TRB_CYCLE) != cycle_state)
if ((flags & TRB_CYCLE) != cycle_state) {/**Give some extra time to get chance controller* to finish command before returning error code.* Checking CMD_RING_BUSY is not sufficient because* this bit is cleared to '0' when the Command* Descriptor has been executed by controller* and not when command completion event has* be added to event ring.*/if (retry--) {usleep_range(90, 100);
I was guided by the warning from checkpatch.pl script and changed udelay to usleep_range. It was wrong. In this place must be used udelay. I will give some time linux community for commenting and I will send it again in a few days.
Regards, Pawel
continue;}return -EINVAL;}cmd_dma = le64_to_cpu(event->event_cmd.cmd_trb);
-- 2.43.0
linux-stable-mirror@lists.linaro.org
-
Pawel Laszczak