Commit a6466243 authored by Finn Thain's avatar Finn Thain Committed by Michael Ellerman

via-cuda: Avoid TREQ race condition

When a read transaction completes, one of several things will happen:
a new transfer is started by the driver, a new transfer request
is raised by the Cuda (i.e. TREQ asserted), or both happen at once.

When both happen at once, there is a race condition between the TREQ test
in the read_done state and the same test in cuda_start(). Moreover, the
former test uses a stale TREQ value.

Theoretically, this can result in the undesirable outcome that the
interrupt handler completes with the state machine 'idle' when it should
instead start the next transaction.

Avoid this race by calling cuda_start() first and then confirming that it
succeeded. If not, test the current TREQ value before entering the
'reading' state.
Tested-by: default avatarStan Johnson <userm57@yahoo.com>
Signed-off-by: default avatarFinn Thain <fthain@telegraphics.com.au>
Signed-off-by: default avatarMichael Ellerman <mpe@ellerman.id.au>
parent cfbf9980
......@@ -605,12 +605,11 @@ cuda_interrupt(int irq, void *arg)
memcpy(ibuf, cuda_rbuf, ibuf_len);
}
reply_ptr = cuda_rbuf;
if (TREQ_asserted(status)) {
assert_TIP();
cuda_state = reading;
} else {
cuda_state = idle;
cuda_start();
if (cuda_state == idle && TREQ_asserted(in_8(&via[B]))) {
assert_TIP();
cuda_state = reading;
}
break;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment