[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAPDyKFrBSOwzn--m321VhRKM-J-9x9BkoPw7+TBWcOE5LpS4HQ@mail.gmail.com>
Date: Fri, 30 Aug 2024 12:27:23 +0200
From: Ulf Hansson <ulf.hansson@...aro.org>
To: zhangxirui <xirui.zhang@...o.com>, Stephen Boyd <sboyd@...nel.org>
Cc: adrian.hunter@...el.com, bastien.curutchet@...tlin.com, festevam@...il.com,
imx@...ts.linux.dev, jh80.chung@...sung.com, kernel@...gutronix.de,
linus.walleij@...aro.org, linux-arm-kernel@...ts.infradead.org,
linux-kernel@...r.kernel.org, linux-mmc@...r.kernel.org, michal.simek@....com,
nico@...xnic.net, orito.takao@...ionext.com, rric@...nel.org,
s.hauer@...gutronix.de, shawnguo@...nel.org, sugaya.taichi@...ionext.com,
u.kleine-koenig@...gutronix.de
Subject: Re: [PATCH 0/9] mmc: convert to devm_clk_get_enabled() API
+ Stephen
On Fri, 30 Aug 2024 at 04:32, zhangxirui <xirui.zhang@...o.com> wrote:
>
> On Wed, 28 Aug 2024 at 17:11, Ulf Hansson <ulf.hansson@...aro.org> wrote:
>
> > On Tue, 27 Aug 2024 at 09:34, zhangxirui <xirui.zhang@...o.com> wrote:
> > >
> > > This series use devm_clk_get_enabled() to simplify code
> > > and avoids the calls to clk_disable_unprepare()
> >
> > I agree that it simplifies code - but it also changes the behaviour,
> > in which order clocks are getting unprepared/disabled during the
> > ->remove() phase. In other words, this needs to be thoroughly tested
> > and not just considered as a trivial cleanup series.
> >
> > For example, if there is a PM domain attached to the mmc host device,
> > is it really okay to allow powering-off the PM domain before the
> > clocks are being gated? This could potentially happen if we apply the
> > $subject series.
>
> Thanks for the reply, are you saying that merging the above patch will
> lead to the following issue?
>
> before:
> bus_remove -> driver_remove -> clk unprepare
> -> dev_pm_domain_detach
> -> device_unbind_cleanup
> after:
> bus_remove -> driver_remove (delete clk unprepare)
> -> dev_pm_domain_detach
> -> device_unbind_cleanup (devm_clk_get_enbaled ->relase(clk unprepare))
Correct!
>
> But I think this issue is not only specific to the MMC host, it will also
> occur with other devices, if there is a PM domain attachded an use devm_clk_get_enbaled API.
Right. Which kind of questions whether the API is really that useful,
from a cleanup point of view.
>
> So, can we solve this problem by swap device_ubind_cleanup and dev_pm_domain_detach ?
> clk unprepare -> power off not power off -> clk unprepare
That may work, I don't know.
Another option is simply to avoid using devm_clk_get_enabled(), unless
you know that the HW can cope with the potentially changed behaviour.
I have looped in Stephen to allow him to provide us his thoughts
around this too.
>
> bus_remove -> driver_remove
> -> device_unbind_cleanup
> -> dev_pm_domain_detach
>
> Thanks.
>
> > >
> > > zhangxirui (9):
> > > mmc: cavium-thunderx: Use devm_clk_get_enabled() helpers
> > > mmc: davinci_mmc: Use devm_clk_get_enabled() helpers
> > > mmc: dw_mmc-hi3798cv200: Use devm_clk_get_enabled() helpers
> > > mmc: mvsdio: Use devm_clk_get_enabled() helpers
> > > mmc: mxcmmc: Use devm_clk_get_enabled() helpers
> > > mmc: mxs-mmc: Use devm_clk_get_enabled() helpers
> > > mmc: sdhci: milbeaut: Use devm_clk_get_enabled() helpers
> > > mmc: sdhci-of-arasan: Use devm_clk_get_enabled() helpers
> > > mmc: sdhci_f_sdh30: Use devm_clk_get_enabled() helpers
Kind regards
Uffe
Powered by blists - more mailing lists