On 3/22/22 08:23, Sondhauß, Jan wrote:
On 22/03/2022 13:43, Guenter Roeck wrote:
On 3/22/22 01:38, Sondhauß, Jan wrote: > When the watchdog is
initialized and triggered before the kernel runs, > it must be ensured
that the kernel uses a different trigger pattern. > Otherwise the
watchdog cannot be kicked. > ZjQcmQRYFpfptBannerStart
This Message Is From an External Sender
Please use caution when clicking on links or opening attachments!
ZjQcmQRYFpfptBannerEnd
On 3/22/22 01:38, Sondhauß, Jan wrote:
When the watchdog is initialized and triggered before the kernel runs,
it must be ensured that the kernel uses a different trigger pattern.
Otherwise the watchdog cannot be kicked.
Reading the current counter reload trigger pattern from the watchdog
hardware during probing makes sure that the trigger pattern is different
from the one previously used.
It is kind of annoying that u-boot uses the same trigger pattern. Question
though: What happens if the register was not initialized by the ROM monitor
and contains the power-up value. Is the chip ok with using that as base ?
In the current implementation the value is negated before writing to the
chip. After reset the register contains 0. In this case the driver would
alternate between ~0 and 0.
From the technical reference manual (SPRUH73P): Writing a different
value than the one already written in the Watchdog Trigger Register does
a watchdog counter reload.
As far as I can tell it should work. But I just noticed a different
problem with my patch: The access to the watchdog happens before the
hardware is enabled via pm_runtime enable. So in the current state that
could lead to an access violation..
If not it might be easier to just use some other pattern like 0xfeedface
or even some pseudo-random value.
The would be easier indeed. The idea behind reading out the current
value is that maybe the driver could get copied from the kernel again.
What do you suggest as a V2 patch? What kind of testing should I
provide? I only have a am335x soc at hand.
I can see two possibilities:
- Use get_random_int() to initialize wdt_trgr_pattern
- Instead of using wdt_trgr_pattern, do something like
writel_relaxed(~readl_relaxed(base + OMAP_WATCHDOG_TGR), (base + OMAP_WATCHDOG_TGR));
My personal preference would be to use get_random_int().
Thanks,
Guenter
