Currently all controller IP/revisions except DWC3_usb3 >= 310a wait 1ms unconditionally for ENDXFER completion when IOC is not set. This is because DWC_usb3 controller revisions >= 3.10a supports GUCTL2[14: Rst_actbitlater] bit which allows polling CMDACT bit to know whether ENDXFER command is completed.
Consider a case where an IN request was queued, and parallelly soft_disconnect was called (due to ffs_epfile_release). This eventually calls stop_active_transfer with IOC cleared, hence send_gadget_ep_cmd() skips waiting for CMDACT cleared during EndXfer. For DWC3 controllers with revisions >= 310a, we don't forcefully wait for 1ms either, and we proceed by unmapping the requests. If ENDXFER didn't complete by this time, it leads to SMMU faults since the controller would still be accessing those requests.
Fix this by ensuring ENDXFER completion by adding 1ms delay in __dwc3_stop_active_transfer() unconditionally.
Cc: stable@vger.kernel.org Fixes: b353eb6dc285 ("usb: dwc3: gadget: Skip waiting for CMDACT cleared during endxfer") Signed-off-by: Prashanth K quic_prashk@quicinc.com --- Changes in v2: Changed the patch logic from CMDACT polling to 1ms mdelay. Updated subject and commit accordingly. Link to v1: https://lore.kernel.org/all/20240422090539.3986723-1-quic_prashk@quicinc.com...
drivers/usb/dwc3/gadget.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c index 4df2661f6675..666eae94524f 100644 --- a/drivers/usb/dwc3/gadget.c +++ b/drivers/usb/dwc3/gadget.c @@ -1724,8 +1724,7 @@ static int __dwc3_stop_active_transfer(struct dwc3_ep *dep, bool force, bool int dep->resource_index = 0;
if (!interrupt) { - if (!DWC3_IP_IS(DWC3) || DWC3_VER_IS_PRIOR(DWC3, 310A)) - mdelay(1); + mdelay(1); dep->flags &= ~DWC3_EP_TRANSFER_STARTED; } else if (!ret) { dep->flags |= DWC3_EP_END_TRANSFER_PENDING;