Re: [PATCH v2 1/2] mmc: tmio: leave clock handling to runtime PM if enabled

From: Ulf Hansson <ulf.hansson@linaro.org>
To: Ulrich Hecht <uli+renesas@fpond.eu>
Cc: Linux-Renesas <linux-renesas-soc@vger.kernel.org>,
	"linux-mmc@vger.kernel.org" <linux-mmc@vger.kernel.org>,
	"Niklas Söderlund" <niklas.soderlund@ragnatech.se>,
	"Wolfram Sang" <wsa@the-dreams.de>,
	"Masahiro Yamada" <yamada.masahiro@socionext.com>,
	"Geert Uytterhoeven" <geert@linux-m68k.org>,
	"Magnus Damm" <magnus.damm@gmail.com>
Subject: Re: [PATCH v2 1/2] mmc: tmio: leave clock handling to runtime PM if enabled
Date: Fri, 2 Aug 2019 15:24:46 +0200	[thread overview]
Message-ID: <CAPDyKFotQvTAC8OVVRiEqLc0ij3XNVdL3eyNHioK0YEWxvER0A@mail.gmail.com> (raw)
In-Reply-To: <1564589857-17720-2-git-send-email-uli+renesas@fpond.eu>

On Wed, 31 Jul 2019 at 18:17, Ulrich Hecht <uli+renesas@fpond.eu> wrote:
>
> This fixes a clock imbalance that occurs on Renesas systems because the SD
> clock is handled by both runtime PM and the hardware driver. After a
> suspend/resume cycle both enable the same clock, resulting in an enable
> count of 2 even if the clock is only used by one device.
>
> See https://www.spinics.net/lists/linux-mmc/msg44431.html for details.
>
> This patch removes the clock handling from the driver's runtime PM
> callbacks and turns the clock off after probing if the device has a power
> domain attached.
>
> Reported-by: Geert Uytterhoeven <geert+renesas@glider.be>
> Signed-off-by: Ulrich Hecht <uli+renesas@fpond.eu>
> Tested-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
> ---
>  drivers/mmc/host/tmio_mmc_core.c | 14 +++++++++++---
>  1 file changed, 11 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/mmc/host/tmio_mmc_core.c b/drivers/mmc/host/tmio_mmc_core.c
> index 31ffcc3..733ff96 100644
> --- a/drivers/mmc/host/tmio_mmc_core.c
> +++ b/drivers/mmc/host/tmio_mmc_core.c
> @@ -1260,9 +1260,14 @@ int tmio_mmc_host_probe(struct tmio_mmc_host *_host)
>         /* See if we also get DMA */
>         tmio_mmc_request_dma(_host, pdata);
>
> -       pm_runtime_set_active(&pdev->dev);
> +#ifdef CONFIG_PM
> +       /* PM handles the clock; disable it so it won't be enabled twice. */
> +       if (_host->clk_disable && pdev->dev.pm_domain)

Hmm.

This seems to work for most cases of yours, but it's fragile, because
how do you know that the pm_domain above is managing the clock? You
don't.

> +               _host->clk_disable(_host);
> +       pm_runtime_get_sync(&pdev->dev);
>         pm_runtime_set_autosuspend_delay(&pdev->dev, 50);
>         pm_runtime_use_autosuspend(&pdev->dev);
> +#endif
>
>         ret = mmc_add_host(mmc);
>         if (ret)
> @@ -1325,7 +1330,8 @@ int tmio_mmc_host_runtime_suspend(struct device *dev)
>         if (host->clk_cache)
>                 host->set_clock(host, 0);
>
> -       tmio_mmc_clk_disable(host);
> +       if (!dev->pm_domain)
> +               tmio_mmc_clk_disable(host);
>
>         return 0;
>  }
> @@ -1340,7 +1346,9 @@ int tmio_mmc_host_runtime_resume(struct device *dev)
>  {
>         struct tmio_mmc_host *host = dev_get_drvdata(dev);
>
> -       tmio_mmc_clk_enable(host);
> +       if (!dev->pm_domain)
> +               tmio_mmc_clk_enable(host);
> +
>         tmio_mmc_hw_reset(host->mmc);
>
>         if (host->clk_cache)
> --
> 2.7.4
>

I am going to think a bit more about this, but at this point, my
gut-feeling is that may need to add some helper functions to let genpd
and/or the pm_clk framework, to share some internal information with
drivers.

Kind regards
Uffe