This patch schedules a USB reset device call following a USB clear
halt error. The issues solved, and patch implementation,
are similar to those found in
drivers/hid/usbhid/hid-core.c.
As seen on very rare occasions approximately one time per month
(mceusb device 2304:0225 in this sample)
Jul 27 2018 15:09:39
[59388.696941] mceusb 1-1.1.2:1.0: Error: urb status = -32 (RX HALT)
[59388.698838] mceusb 1-1.1.2:1.0: rx clear halt error -32
the device can get into RX or TX HALT state where usb_clear_halt()
also fails and also returns -EPIPE (HALT/STALL). After which, all
further mceusb device control and data I/O always fail with HALT/STALL.
Subsequently, the entire mceusb device no longer functions.
Cause and problem replication conditions remain unknown.
Further troubleshooting reveals usb_reset_device()
restores mceusb device operation.
Patch test 1:
Hot unplugging the mceusb device triggers USB RX HALT and
USB clear halt errors. A mceusb_dev_disconnect() call follows unplug.
This patch's reset device call invokes an extra
mceusb_dev_probe()
mceusb_dev_disconnect()
cycle, before the mceusb driver detaches.
The additional probe/disconnect verifies the patch's device reset
code executed.
But note this patch is for USB clear halt error cases not caused by
unplugging the mceusb device.
Patch test 2:
Simulate a RX HALT and a clear halt error with instrumented code in
the driver.
Jul 12 2019 19:41:18
[522745.263104] mceusb 1-1.3:1.0: set rx halt retval, 0
[522745.263943] mceusb 1-1.3:1.0: Error: rx urb status = -32 (RX HALT)
[522745.263970] mceusb 1-1.3:1.0: kevent 1 scheduled
[522745.264016] mceusb 1-1.3:1.0: kevent handler called (flags 0x2)
[522745.272883] mceusb 1-1.3:1.0: rx clear halt status = 0
[522745.272917] mceusb 1-1.3:1.0: stuck RX HALT state requires USB Reset Device to clear
[522745.273005] mceusb 1-1.3:1.0: mceusb_dev_disconnect called
[522745.702815] usb 1-1.3: reset full-speed USB device number 14 using dwc_otg
[522745.836812] mceusb 1-1.3:1.0: mceusb_dev_probe called
[522745.836823] mceusb 1-1.3:1.0: acceptable bulk inbound endpoint found
[522745.836832] mceusb 1-1.3:1.0: acceptable bulk outbound endpoint found
...
The result matches what is expected when the device gets into
a real rx clear halt error case by itself.
This is the same sequence of messages when manually invoking
the ./usbreset command line utility with an unpatched mceusb driver.
Signed-off-by: A Sun <as1033x@comcast.net>
Signed-off-by: Sean Young <sean@mess.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>