Hi Felipe, I'm running into a hang with the mass-storage gadget with later versions of the dwc3 driver. I see this with the version from your master branch from Feb 1, and also with the latest version from today. I see it with both Linux and Windows hosts, while doing my simple file corruption test. It goes like this: - Connect the device at SuperSpeed. - Copy a large (~750MB) file from the host to the device. - "Safely remove" the device and disconnect it. - Reconnect the device. - Rename the file on the device. - Copy the file back to the host. - "Safely remove" the device and disconnect it. - Compare the two files (this step is not important, you can skip it). - Reconnect the device. - Delete the file from the device. - Copy the original file from the host to the device again. Here, the file transfer hangs. Using a USB analyzer, I see that after the first few packets of a 16KB transfer, the device has sent an NRDY, and it never sends an ERDY, so the transfer hangs. The last few lines of the dmesg look like this: dwc3 dwc3.0: request ffff8801361cf5c0 from ep1out-bulk completed 16384/16384 ===> 0 dwc3 dwc3.0: ep1in-bulk: Transfer Not Ready dwc3 dwc3.0: ep1in-bulk: reason Transfer Not Active dwc3 dwc3.0: queing request ffff8801361cf2c0 to ep1in-bulk length 13 dwc3 dwc3.0: ep1in-bulk: req ffff8801361cf2c0 dma be053000 length 13 last dwc3 dwc3.0: ep1in-bulk: cmd 'Start Transfer' params 00000000 37ae3000 00000000 dwc3 dwc3.0: Command Complete --> 0 dwc3 dwc3.0: queing request ffff8801361cf5c0 to ep1out-bulk length 1024 dwc3 dwc3.0: ep1in-bulk: Transfer Complete dwc3 dwc3.0: request ffff8801361cf2c0 from ep1in-bulk completed 13/13 ===> 0 dwc3 dwc3.0: ep1out-bulk: Transfer Not Ready dwc3 dwc3.0: ep1out-bulk: reason Transfer Not Active dwc3 dwc3.0: ep1out-bulk: req ffff8801361cf5c0 dma be053800 length 1024 last dwc3 dwc3.0: ep1out-bulk: cmd 'Start Transfer' params 00000000 3b9151e0 00000000 dwc3 dwc3.0: Command Complete --> 0 dwc3 dwc3.0: ep1out-bulk: Transfer Complete dwc3 dwc3.0: request ffff8801361cf5c0 from ep1out-bulk completed 31/1024 ===> 0 dwc3 dwc3.0: queing request ffff8801361cf5c0 to ep1out-bulk length 16384 dwc3 dwc3.0: queing request ffff8801361ce600 to ep1out-bulk length 16384 dwc3 dwc3.0: ep1out-bulk: Transfer Not Ready dwc3 dwc3.0: ep1out-bulk: reason Transfer Not Active dwc3 dwc3.0: ep1out-bulk: req ffff8801361cf5c0 dma be054000 length 16384 dwc3 dwc3.0: ep1out-bulk: req ffff8801361ce600 dma be058000 length 16384 last dwc3 dwc3.0: ep1out-bulk: cmd 'Start Transfer' params 00000000 3b9151f0 00000000 dwc3 dwc3.0: Command Complete --> 0 dwc3 dwc3.0: ep1out-bulk: Transfer Not Ready dwc3 dwc3.0: ep1out-bulk: reason Transfer Active dwc3 dwc3.0: ep1out-bulk: endpoint busy and after that no more messages appear in the dmesg. The "Transfer Not Ready" / "reason Transfer Active" only appears that one time in the dmesg (there's only a few seconds worth of messages in the buffer though). After hacking around in the driver, I came up with the following workaround: diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c index 6d64351..184f2dd 100644 --- a/drivers/usb/dwc3/gadget.c +++ b/drivers/usb/dwc3/gadget.c @@ -898,7 +898,7 @@ static void dwc3_prepare_trbs(struct dwc3_ep *dep, bool starting) last_one = 1; dwc3_prepare_one_trb(dep, req, dma, length, - last_one, false); + true, false); if (last_one) break; Obviously that's not the correct fix, but it does make the mass-storage gadget work again. Any ideas? Anything I can try to narrow it down? -- Paul -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
