There are no reads in these functions, so if MMIO writes are posted,
the writes in enable/disable may not have completed by the time these
functions return. If the functions run from different CPUs, it's
in theory possible for the writes to be interleaved, which would be
disastrous for this driver.
At the very least, we need an mmiowb() before releasing the lock, but
since it seems desirable for the watchdog timer to be actually stopped
or reset when these functions return, read the lock register to force
the writes out.
Signed-off-by: Nick Bowler <nbowler@elliptictech.com>
Signed-off-by: Wim Van Sebroeck <wim@iguana.be>
At least on the Versatile Express' V2M, calling wdt_disable followed by
wdt_enable, for instance by running the following sequence:
echo V > /dev/watchdog; echo V > /dev/watchdog
results in an immediate reset. The wdt_disable function writes 0 to the
load register; while the watchdog interrupts are disabled at this point,
this special value is defined to trigger an interrupt immediately. It
appears that in this instance, the reset happens when the interrupts
are subsequently enabled by wdt_enable.
Putting in a short delay after writing a new load value in wdt_enable
solves the issue, but it seems cleaner to simply never write 0 to the
load register at all: according to the hardware docs, writing 0 to the
control register suffices to stop the counter, and the write of 0 to
the load register is questionable anyway since this register resets to
0xffffffff.
Signed-off-by: Nick Bowler <nbowler@elliptictech.com>
Signed-off-by: Wim Van Sebroeck <wim@iguana.be>
Make Primecell driver probe functions take a const pointer to their
ID tables. Drivers should never modify their ID tables in their
probe handler.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
