via-cuda: Avoid TREQ race condition
authorFinn Thain <fthain@telegraphics.com.au>
Sun, 1 Jan 2017 00:56:26 +0000 (19:56 -0500)
committerMichael Ellerman <mpe@ellerman.id.au>
Tue, 7 Feb 2017 05:56:21 +0000 (16:56 +1100)
commita64662432200f8af6f67cd3664885a323f6a2f2d
treeaac25458e8f5b21ec642515b52d8aed8d240cf0b
parentcfbf99801bcaf8398492ebc16af72259ad7aa146
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: Stan Johnson <userm57@yahoo.com>
Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
drivers/macintosh/via-cuda.c