linux-mmc.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Ulf Hansson <ulf.hansson@linaro.org>
To: Tony Lindgren <tony@atomide.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>,
	Chunyan Zhang <zhang.chunyan@linaro.org>,
	Faiz Abbas <faiz_abbas@ti.com>,
	Kishon Vijay Abraham I <kishon@ti.com>,
	Santosh Shilimkar <ssantosh@kernel.org>,
	linux-mmc <linux-mmc@vger.kernel.org>,
	linux-omap <linux-omap@vger.kernel.org>,
	Rob Herring <robh@kernel.org>, DTML <devicetree@vger.kernel.org>
Subject: Re: [PATCH 4/5] mmc: sdhci-omap: Implement PM runtime functions
Date: Fri, 8 Oct 2021 16:43:29 +0200	[thread overview]
Message-ID: <CAPDyKFpybVPeYy-FsXnzDXNri+f7rhPmKa6vBF8NMUc3dQCZRw@mail.gmail.com> (raw)
In-Reply-To: <20210930065733.31943-5-tony@atomide.com>

On Thu, 30 Sept 2021 at 08:57, Tony Lindgren <tony@atomide.com> wrote:
>
> Implement PM runtime functions and enable MMC_CAP_AGGRESSIVE_PM.

I suggest you split this change into two pieces. MMC_CAP_AGGRESSIVE_PM
is about enabling runtime PM management for the eMMC/SD card device,
which is perfectly fine to use independently of whether runtime PM is
supported for the host device.

>
> Note that we save context in probe to avoid restoring invalid context
> on the first resume. For system suspend, we have the new PM runtime
> functions do most of the work.
>
> Signed-off-by: Tony Lindgren <tony@atomide.com>
> ---
>  drivers/mmc/host/sdhci-omap.c | 66 +++++++++++++++++++++++++++++------
>  1 file changed, 56 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/mmc/host/sdhci-omap.c b/drivers/mmc/host/sdhci-omap.c
> --- a/drivers/mmc/host/sdhci-omap.c
> +++ b/drivers/mmc/host/sdhci-omap.c
> @@ -117,6 +117,9 @@ struct sdhci_omap_host {
>
>         struct pinctrl          *pinctrl;
>         struct pinctrl_state    **pinctrl_state;
> +       unsigned long           context_valid:1;
> +       unsigned long           is_runtime_suspended:1;
> +       unsigned long           needs_resume:1;
>         bool                    is_tuning;
>
>         /* Offset for omap specific registers from base */
> @@ -1207,6 +1210,8 @@ static const struct soc_device_attribute sdhci_omap_soc_devices[] = {
>         }
>  };
>
> +static void sdhci_omap_context_save(struct sdhci_omap_host *omap_host);
> +
>  static int sdhci_omap_probe(struct platform_device *pdev)
>  {
>         int ret;
> @@ -1338,6 +1343,8 @@ static int sdhci_omap_probe(struct platform_device *pdev)
>         /* R1B responses is required to properly manage HW busy detection. */
>         mmc->caps |= MMC_CAP_NEED_RSP_BUSY;
>
> +       mmc->caps |= MMC_CAP_AGGRESSIVE_PM;
> +
>         ret = sdhci_setup_host(host);
>         if (ret)
>                 goto err_put_sync;
> @@ -1350,6 +1357,11 @@ static int sdhci_omap_probe(struct platform_device *pdev)
>         if (ret)
>                 goto err_cleanup_host;
>
> +       sdhci_omap_context_save(omap_host);
> +       omap_host->context_valid = 1;

Looks like you can remove this flag, it's not being used.

> +
> +       pm_runtime_put_sync(dev);

I recommend to use the PM runtime autosuspend feature, as to avoid an
initial latency for every I/O request to the host driver. The mmc core
already supports that, see mmc_release_host().

The typical default timeout value for autosuspend, is usually set
~50-200ms, by host drivers (if I recall correctly).

> +
>         return 0;
>
>  err_cleanup_host:
> @@ -1371,6 +1383,7 @@ static int sdhci_omap_remove(struct platform_device *pdev)
>         struct device *dev = &pdev->dev;
>         struct sdhci_host *host = platform_get_drvdata(pdev);
>
> +       pm_runtime_get_sync(dev);
>         sdhci_remove_host(host, true);
>         pm_runtime_put_sync(dev);

There is no guarantee that this triggers a call to
->sdhci_omap_runtime_suspend(), which I guess is what we want.
Userspace via sysfs may have increase the RPM usage count
(pm_runtime_forbid(), for example.

To address this, I would call pm_runtime_disable() first and then
explicitly put the device into low power state, rather than relying on
runtime PM to do it. Another option could be to use
pm_runtime_force_suspend().

>         pm_runtime_disable(dev);
> @@ -1402,42 +1415,75 @@ static void sdhci_omap_context_restore(struct sdhci_omap_host *omap_host)
>         sdhci_omap_writel(omap_host, SDHCI_OMAP_ISE, omap_host->ise);
>  }
>
> -static int __maybe_unused sdhci_omap_suspend(struct device *dev)
> +static int __maybe_unused sdhci_omap_runtime_suspend(struct device *dev)
>  {
>         struct sdhci_host *host = dev_get_drvdata(dev);
>         struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
>         struct sdhci_omap_host *omap_host = sdhci_pltfm_priv(pltfm_host);
>
> -       sdhci_suspend_host(host);
> -

Shouldn't you call sdhci_runtime_suspend_host() somewhere here?

>         sdhci_omap_context_save(omap_host);
>
>         pinctrl_pm_select_idle_state(dev);
>
> -       pm_runtime_force_suspend(dev);
> +       omap_host->is_runtime_suspended = 1;
>
>         return 0;
>  }
>
> -static int __maybe_unused sdhci_omap_resume(struct device *dev)
> +static int __maybe_unused sdhci_omap_runtime_resume(struct device *dev)
>  {
>         struct sdhci_host *host = dev_get_drvdata(dev);
>         struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
>         struct sdhci_omap_host *omap_host = sdhci_pltfm_priv(pltfm_host);
>
> -       pm_runtime_force_resume(dev);
> -
>         pinctrl_pm_select_default_state(dev);
>
> -       sdhci_omap_context_restore(omap_host);
> +       if (omap_host->context_valid)
> +               sdhci_omap_context_restore(omap_host);
> +
> +       omap_host->is_runtime_suspended = 0;

Shouldn't you call sdhci_runtime_resume_host() somewhere here?

> +
> +       return 0;
> +}
> +
> +static int __maybe_unused sdhci_omap_suspend(struct device *dev)
> +{
> +       struct sdhci_host *host = dev_get_drvdata(dev);
> +       struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
> +       struct sdhci_omap_host *omap_host = sdhci_pltfm_priv(pltfm_host);
> +
> +       if (omap_host->is_runtime_suspended)
> +               return 0;

So if the host is already runtime suspended, it's okay to just leave it as is?

In a way that sounds like you could call pm_runtime_force_suspend()
instead, assuming the sdhci_omap_runtime_suspend() can be extended to
do the right thing for system suspend as well.

It looks a bit odd that sdhci_suspend_host() is called only when the
host is runtime resumed. Perhaps you can elaborate a bit more on why
this is, so I can understand better what you want to achieve here.

> +
> +       sdhci_suspend_host(host);
> +       sdhci_omap_runtime_suspend(dev);
> +       omap_host->needs_resume = 1;
>
> +       return 0;
> +}
> +
> +static int __maybe_unused sdhci_omap_resume(struct device *dev)
> +{
> +       struct sdhci_host *host = dev_get_drvdata(dev);
> +       struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
> +       struct sdhci_omap_host *omap_host = sdhci_pltfm_priv(pltfm_host);
> +
> +       if (!omap_host->needs_resume)
> +               return 0;
> +
> +       sdhci_omap_runtime_resume(dev);
>         sdhci_resume_host(host);
> +       omap_host->needs_resume = 0;
>
>         return 0;
>  }
>  #endif
> -static SIMPLE_DEV_PM_OPS(sdhci_omap_dev_pm_ops, sdhci_omap_suspend,
> -                        sdhci_omap_resume);
> +
> +static const struct dev_pm_ops sdhci_omap_dev_pm_ops = {
> +       SET_RUNTIME_PM_OPS(sdhci_omap_runtime_suspend,
> +                          sdhci_omap_runtime_resume, NULL)
> +       SET_SYSTEM_SLEEP_PM_OPS(sdhci_omap_suspend, sdhci_omap_resume)
> +};
>
>  static struct platform_driver sdhci_omap_driver = {
>         .probe = sdhci_omap_probe,
> --
> 2.33.0

Kind regards
Uffe

  reply	other threads:[~2021-10-08 14:44 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-09-30  6:57 [PATCHv2 0/5] More SoCs for sdhci-omap to deprecate omap_hsmmc Tony Lindgren
2021-09-30  6:57 ` [PATCH 1/5] dt-bindings: sdhci-omap: Update binding for legacy SoCs Tony Lindgren
2021-10-02 13:29   ` Adam Ford
2021-10-05  8:04     ` Tony Lindgren
2021-10-05 10:45       ` Adam Ford
2021-10-06  5:03         ` Tony Lindgren
2021-10-02 16:15   ` Adam Ford
2021-09-30  6:57 ` [PATCH 2/5] mmc: sdhci-omap: Handle voltages to add support omap4 Tony Lindgren
2021-09-30  6:57 ` [PATCH 3/5] mmc: sdhci-omap: Add omap_offset to support omap3 and earlier Tony Lindgren
2021-09-30  6:57 ` [PATCH 4/5] mmc: sdhci-omap: Implement PM runtime functions Tony Lindgren
2021-10-08 14:43   ` Ulf Hansson [this message]
2021-10-11  5:23     ` Tony Lindgren
2021-10-12  9:05       ` Ulf Hansson
2021-10-12  9:18         ` Tony Lindgren
2021-09-30  6:57 ` [PATCH 5/5] mmc: sdhci-omap: Configure optional wakeirq Tony Lindgren
  -- strict thread matches above, loose matches on Subject: below --
2021-09-21 11:15 [PATCH 0/5] More SoCs for sdhci-omap to deprecate omap_hsmmc Tony Lindgren
2021-09-21 11:15 ` [PATCH 4/5] mmc: sdhci-omap: Implement PM runtime functions Tony Lindgren

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=CAPDyKFpybVPeYy-FsXnzDXNri+f7rhPmKa6vBF8NMUc3dQCZRw@mail.gmail.com \
    --to=ulf.hansson@linaro.org \
    --cc=adrian.hunter@intel.com \
    --cc=devicetree@vger.kernel.org \
    --cc=faiz_abbas@ti.com \
    --cc=kishon@ti.com \
    --cc=linux-mmc@vger.kernel.org \
    --cc=linux-omap@vger.kernel.org \
    --cc=robh@kernel.org \
    --cc=ssantosh@kernel.org \
    --cc=tony@atomide.com \
    --cc=zhang.chunyan@linaro.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).