Re: [PATCH v3 01/11] stm class: Introduce an abstraction for System Trace Module devices

[Alexander Shishkin <alexander.shishkin@linux.intel.com>]

Chunyan Zhang <zhang.lyra@gmail.com> writes:

> Sure, I mean, the root reason of this problem is here ( i.e.
> "stm_core_up" was zero then):
>      if (!stm_core_up)
>          return -EPROBE_DEFER;
>
> Why it was zero?
> Because the function (i.e. stm_core_init() ) in which "stm_core_up"
> would be added one hasn't been executed at this moment. It would be
> executed on module_init stage for you this version of patch.

Again, this is the indented behavior.

> The reason of this warning is:
> After stm_probe() failed, clk_core_disable() would be called from
> amba_put_disable_pclk(), then WARN_ON() happened:
>          if (WARN_ON(core->enable_count == 0))
>                  return;
>
> I'm guessing the reason why "core->enable_count" was 0 at this moment is:
> I don't know who created a thread to process the
> amba_pm_runtime_suspend(), in which clk_core_disable() was already
> called, "core->enable_count" was, of course, cleared to zero then.
> And this thread run before amba_put_disable_pclk(pcdev) which is just
> the one called from amba_probe() after
> "->probe"(i.e. stm_probe in this case) returning a non-zero value.

No, this is guesswork. In amba_probe(), clocks are enabled for the
drv->probe() and then disabled afterwards and that's where the refcount
ends up unbalanced, the probe is the culprit.

I can debug your driver for you but you'll at least need to put the code
up somewhere so I can see it.

Regards,
--
Alex