xhci: dbgtty: Fix data corruption when transmitting data form DbC to host
Data read from a DbC device may be corrupted due to a race between
ongoing write and write request completion handler both queuing new
transfer blocks (TRBs) if there are remining data in the kfifo.
TRBs may be in incorrct order compared to the data in the kfifo.
Driver fails to keep lock between reading data from kfifo into a
dbc request buffer, and queuing the request to the transfer ring.
This allows completed request to re-queue itself in the middle of
an ongoing transfer loop, forcing itself between a kfifo read and
request TRB write of another request
cpu0 cpu1 (re-queue completed req2)
lock(port_lock)
dbc_start_tx()
kfifo_out(fifo, req1->buffer)
unlock(port_lock)
lock(port_lock)
dbc_write_complete(req2)
dbc_start_tx()
kfifo_out(fifo, req2->buffer)
unlock(port_lock)
lock(port_lock)
req2->trb = ring->enqueue;
ring->enqueue++
unlock(port_lock)
lock(port_lock)
req1->trb = ring->enqueue;
ring->enqueue++
unlock(port_lock)
In the above scenario a kfifo containing "12345678" would read "1234" to
req1 and "5678" to req2, but req2 is queued before req1 leading to
data being transmitted as "56781234"
Solve this by adding a flag that prevents starting a new tx if we
are already mid dbc_start_tx() during the unlocked part.
The already running dbc_do_start_tx() will make sure the newly completed
request gets re-queued as it is added to the request write_pool while
holding the lock.
Cc: stable@vger.kernel.org
Fixes: dfba2174dc42 ("usb: xhci: Add DbC support in xHCI driver")
Tested-by: Łukasz Bartosik <ukaszb@chromium.org>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
2 files changed
