Re: [PATCH 2/2] usb: host: xhci: Handle the right timeout command

From: Mathias Nyman
Date: Mon Dec 12 2016 - 10:51:37 EST

On 05.12.2016 09:51, Baolin Wang wrote:
If a command event is found on the event ring during an interrupt,
we need to stop the command timer with del_timer(). Since del_timer()
can fail if the timer is running and waiting on the xHCI lock, then
it maybe get the wrong timeout command in xhci_handle_command_timeout()
if host fetched a new command and updated the xhci->current_cmd in
handle_cmd_completion(). For this situation, we need a way to signal
to the command timer that everything is fine and it should exit.

Ah, right, this could actually happen.

We should introduce a counter (xhci->current_cmd_pending) for the number
of pending commands. If we need to cancel the command timer and del_timer()
succeeds, we decrement the number of pending commands. If del_timer() fails,
we leave the number of pending commands alone.

For handling timeout command, in xhci_handle_command_timeout() we will check
the counter after decrementing it, if the counter (xhci->current_cmd_pending)
is 0, which means xhci->current_cmd is the right timeout command. If the
counter (xhci->current_cmd_pending) is greater than 0, which means current
timeout command has been handled by host and host has fetched new command as
xhci->current_cmd, then just return and wait for new current command.

A counter like this could work.

Writing the abort bit can generate either ABORT+STOP, or just STOP
event, this seems to cover both.

quick check, case 1: timeout and cmd completion at the same time.

cpu1 cpu2

queue_command(first), p++ (=1)
--completion irq fires-- -- timer times out at same time--
handle_cmd_completion() handle_cmd_timeout(),)
lock(xhci_lock ) spin_on(xhci_lock)
del_timer() fail, p (=1, nochange)
cur_cmd = list_next(), p++ (=2)
p-- (=1)
if (p > 0), exit
OK works

case 2: normal timeout case with ABORT+STOP, no race.

cpu1 cpu2

queue_command(first), p++ (=1)
p-- (P=0), don't exit
mod_timer(), p++ (P=1)
del_timer(), ok, p-- (p = 0)
del_timer(), fail, (P=0)
cur_cmd = list_next(), p++ (=1)

OK, works, and same for just STOP case, with the only difference that
during handle_cmd_completion(STOP) p is decremented (p--)

So unless there is a way to find out if cur_cmd is valid in command timeout
in command timeout with the help of existing flags and lists this would be a working