Re: [PATCH RFC v2] mmc: sdhci_am654: Add support for PM suspend/resume

From: Aswath Govindraju
Date: Wed Apr 13 2022 - 08:07:31 EST


Hi Uffe,

On 12/04/22 20:08, Ulf Hansson wrote:
> On Fri, 8 Apr 2022 at 14:43, Aswath Govindraju <a-govindraju@xxxxxx> wrote:
>>
>> Add support for suspend/resume and pm_runtime resume/suspend.
>>
>> Signed-off-by: Aswath Govindraju <a-govindraju@xxxxxx>
>> ---
>>
>> Changes since v1:
>> - Removed System pm calls and used force_runtime_pm calls instead
>> - Added error handling at required places
>> - Added pm_runtime_sync call in the sdhci_am654_remove function
>> - Replaced pm_runtime_*_sync calls in probe with noidle and noresume
>> calls
>> - Used MACRO SET_RUNTIME_PM_OPS to avoid build errors when CONFIG_PM
>> is disabled
>>
>> drivers/mmc/host/sdhci_am654.c | 146 +++++++++++++++++++++++++++++----
>> 1 file changed, 132 insertions(+), 14 deletions(-)
>>
>> diff --git a/drivers/mmc/host/sdhci_am654.c b/drivers/mmc/host/sdhci_am654.c
>> index e54fe24d47e7..131b2b9ae0e7 100644
>> --- a/drivers/mmc/host/sdhci_am654.c
>> +++ b/drivers/mmc/host/sdhci_am654.c
>> @@ -84,6 +84,7 @@
>> #define DRIVER_STRENGTH_40_OHM 0x4
>>
>> #define CLOCK_TOO_SLOW_HZ 50000000
>> +#define SDHCI_AM654_AUTOSUSPEND_DELAY 100
>>
>> /* Command Queue Host Controller Interface Base address */
>> #define SDHCI_AM654_CQE_BASE_ADDR 0x200
>> @@ -791,16 +792,10 @@ static int sdhci_am654_probe(struct platform_device *pdev)
>>
>> pltfm_host->clk = clk_xin;
>>
>> - /* Clocks are enabled using pm_runtime */
>> - pm_runtime_enable(dev);
>> - ret = pm_runtime_resume_and_get(dev);
>> - if (ret)
>> - goto pm_runtime_disable;
>> -
>> base = devm_platform_ioremap_resource(pdev, 1);
>> if (IS_ERR(base)) {
>> ret = PTR_ERR(base);
>> - goto pm_runtime_put;
>> + goto err_pltfm_free;
>> }
>>
>> sdhci_am654->base = devm_regmap_init_mmio(dev, base,
>> @@ -808,31 +803,47 @@ static int sdhci_am654_probe(struct platform_device *pdev)
>> if (IS_ERR(sdhci_am654->base)) {
>> dev_err(dev, "Failed to initialize regmap\n");
>> ret = PTR_ERR(sdhci_am654->base);
>> - goto pm_runtime_put;
>> + goto err_pltfm_free;
>> }
>>
>> ret = sdhci_am654_get_of_property(pdev, sdhci_am654);
>> if (ret)
>> - goto pm_runtime_put;
>> + goto err_pltfm_free;
>>
>> ret = mmc_of_parse(host->mmc);
>> if (ret) {
>> dev_err(dev, "parsing dt failed (%d)\n", ret);
>> - goto pm_runtime_put;
>> + goto err_pltfm_free;
>> }
>>
>> host->mmc_host_ops.execute_tuning = sdhci_am654_execute_tuning;
>>
>> + pm_runtime_get_noresume(dev);
>> + ret = pm_runtime_set_active(dev);
>> + if (ret)
>> + goto pm_put;
>> + pm_runtime_enable(dev);
>> + ret = clk_prepare_enable(pltfm_host->clk);
>> + if (ret)
>> + goto pm_disable;
>> +
>> ret = sdhci_am654_init(host);
>> if (ret)
>> - goto pm_runtime_put;
>> + goto clk_disable;
>>
>> + /* Setting up autosuspend */
>> + pm_runtime_set_autosuspend_delay(dev, SDHCI_AM654_AUTOSUSPEND_DELAY);
>> + pm_runtime_use_autosuspend(dev);
>> + pm_runtime_mark_last_busy(dev);
>> + pm_runtime_put_autosuspend(dev);
>> return 0;
>>
>> -pm_runtime_put:
>> - pm_runtime_put_sync(dev);
>> -pm_runtime_disable:
>> +clk_disable:
>> + clk_disable_unprepare(pltfm_host->clk);
>> +pm_disable:
>> pm_runtime_disable(dev);
>> +pm_put:
>> + pm_runtime_put_noidle(dev);
>> err_pltfm_free:
>> sdhci_pltfm_free(pdev);
>> return ret;
>> @@ -841,23 +852,130 @@ static int sdhci_am654_probe(struct platform_device *pdev)
>> static int sdhci_am654_remove(struct platform_device *pdev)
>> {
>> struct sdhci_host *host = platform_get_drvdata(pdev);
>> + struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
>> int ret;
>>
>> + ret = pm_runtime_get_sync(&pdev->dev);
>
> pm_runtime_resume_and_get() is better suited here, as it helps us to
> restore the usage count in the error path.
> >> + if (ret < 0)
>> + return ret;
>> +
>> sdhci_remove_host(host, true);
>> ret = pm_runtime_put_sync(&pdev->dev);
>
> We need to drop the above call to pm_runtime_put_sync(). Otherwise you
> may potentially end up gating the clock, in which case the below call
> to clk_disable_unprepare() becomes wrong.
> >> if (ret < 0)
>> return ret;
>>
>> + clk_disable_unprepare(pltfm_host->clk);
>> pm_runtime_disable(&pdev->dev);
>
> You need a call to pm_runtime_put_noidle() here to restore the RPM
> usage count to zero.
>
>> sdhci_pltfm_free(pdev);
>> + return 0;
>> +}
>
> [...]
>
> Other than the rather minor things above, this looks good to me!
>

Thank you for the review.

Regards,
Aswath

> Kind regards
> Uffe