[S390] qdio: clear shared DSCI before scheduling the queue handler
authorJan Glauber <jang@linux.vnet.ibm.com>
Sun, 24 Jul 2011 08:48:00 +0000 (10:48 +0200)
committerMartin Schwidefsky <schwidefsky@de.ibm.com>
Sun, 24 Jul 2011 08:48:00 +0000 (10:48 +0200)
commitb02f0c2ea25781e0f94b4fc8f6f85582057857b3
tree688ed8fea4f857f5280be3f4d8adae663fea3d4b
parent50a15981a1fac7e019ff7c3cba87531fb580f065
[S390] qdio: clear shared DSCI before scheduling the queue handler

The following race can occur with qdio devices that use the shared device
state change indicator:

Device (Shared DSCI)    CPU0                            CPU1
===============================================================================

1. DSCI 0 => 1,
   INT pending
                        2. Thinint handler
                        * si_used = 1
                        * Inbound tasklet_schedule
                        * DSCI 1 => 0

3. DSCI 0 => 1,
   INT pending

                                                        4. Thinint handler
                                                        * si_used = 1
                                                        * Inbound tasklet_schedu
le
                                                           => NOP

                        5. Inbound tasklet run

6. DSCI = 1,
   INT surpressed

                                                        7. DSCI 1 => 0

The race would lead to a stall where new data in the input queue is
not recognized so the device stops working in case of no further traffic.

Fix the race by resetting the DSCI before scheduling the inbound tasklet
so the device generates an interrupt if new data arrives in the above
scenario in step 6.

Reviewed-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: Jan Glauber <jang@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
drivers/s390/cio/qdio_thinint.c