Re: [PATCH v4] amba: Remove deferred device addition

[Sudeep Holla]

On Mon, Jul 18, 2022 at 06:51:29PM -0700, Saravana Kannan wrote:
> On Tue, Jul 12, 2022 at 11:53 PM Marek Szyprowski
> <m.szyprowski@xxxxxxxxxxx> wrote:

[..]

> Longer explanation follows:
> 
> 5a46079a9645 ("PM: domains: Delete usage of
> driver_deferred_probe_check_state()") correctly assumed fw_devlink
> will block calls to __genpd_dev_pm_attach() before the power domain
> has probed or we have given up waiting on suppliers at the driver core
> level. So, __genpd_dev_pm_attach() returning -2 was not a problem
> (well, there are other issues, but we'll pretend they don't exist for
> now).
> 
> Until this amba patch, that was true because really_probe() calls
> device_links_check_suppliers() before you'll get anywhere near
> __genpd_dev_pm_attach().
>

Last time I started looking at this patch, I was suspecting some issue
around __genpd_dev_pm_attach() but your explanation makes sense to me
now and I am more or less convinced this was what happening on Juno.

> But with this amba patch, we try to get power domains before we get to
> really_probe() and that doesn't get the device links check. So,
> amba_match() has to always return -EPROBE_DEFER on any error until we
> optimize out match() calls for devices whose suppliers aren't ready
> yet. I'm considering reverting 5a46079a9645 due to other issues, so I
> think v4 might be okay as is.
>

OK, do I need to check with 5a46079a9645 reverted then ?

-- 
Regards,
Sudeep