Re: [RFC PATCH] mmc: Enable wakeup_sources for mmc core

From: Ulf Hansson
Date: Fri Aug 23 2013 - 04:13:02 EST

Hi Zoran,

On 13 June 2013 19:56, Zoran Markovic <zoran.markovic@xxxxxxxxxx> wrote:
> This is a reworked implementation of wakelocks for the MMC core from
> Android kernel, originally authored by Colin Cross and San Mehat.
> The patch makes sure that whenever a MMC device is inserted/removed,
> the system stays awake until it's reconfigured for the new state.
> It is assumed that 1/2 second is sufficient for the system to start
> the configuration action for the newly detected MMC device, which might
> include e.g. mounting the hosted file system(s).
>
> The implementation uses wakeup_sources instead of wake_locks.
>
> Feedback on the approach is greatly appreciated, in particular for the
> 1/2 second extension peroid.
>
> Cc: San Mehat <san@xxxxxxxxxx>
> Cc: Colin Cross <ccross@xxxxxxxxxxx>
> Cc: John Stultz <john.stultz@xxxxxxxxxx>
> Cc: Chris Ball <cjb@xxxxxxxxxx>
> Cc: Ulf Hansson <ulf.hansson@xxxxxxxxxx>
> Cc: Johan Rudholm <johan.rudholm@xxxxxxxxxxxxxx>
> Cc: Jaehoon Chung <jh80.chung@xxxxxxxxxxx>
> Cc: Konstantin Dorfman <kdorfman@xxxxxxxxxxxxxx>
> Cc: Guennadi Liakhovetski <g.liakhovetski@xxxxxx>
> Cc: Tejun Heo <tj@xxxxxxxxxx>
> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Signed-off-by: John Stultz <john.stultz@xxxxxxxxxx>
> [<zoran.markovic@xxxxxxxxxx>: tweaked commit message, reworked to use
> wakeup_source_register/unregister instead of wakeup_source_init/trash,
> added the missing __pm_relax() for non-removable devices in mmc_rescan().]
> Signed-off-by: Zoran Markovic <zoran.markovic@xxxxxxxxxx>
> ---
> drivers/mmc/core/core.c | 31 +++++++++++++++++++++++++------
> drivers/mmc/core/host.c | 7 +++++++
> include/linux/mmc/host.h | 2 ++
> 3 files changed, 34 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
> index c40396f..d5230c7 100644
> --- a/drivers/mmc/core/core.c
> +++ b/drivers/mmc/core/core.c
> @@ -23,6 +23,7 @@
> #include <linux/log2.h>
> #include <linux/regulator/consumer.h>
> #include <linux/pm_runtime.h>
> +#include <linux/pm_wakeup.h>
> #include <linux/suspend.h>
> #include <linux/fault-inject.h>
> #include <linux/random.h>
> @@ -1656,6 +1657,7 @@ void mmc_detect_change(struct mmc_host *host, unsigned long delay)
> spin_unlock_irqrestore(&host->lock, flags);
> #endif
> host->detect_change = 1;
> + __pm_stay_awake(host->ws);

There are some scenarios I want to be sure you have thought of. Please
comment on them.

1. mmc_detect_change does obviously not have to be run the same number
of times as the mmc_rescan function. In other words, the calls to
__pm_stay_awake is not paired with __pm_relay, I suppose this does not
matter?

2. mmc_detect_change can for example be called while the device
suspend sequence is progressing. At this point the rescan work is
disabled, thus __pm_relax will not be called, until the next rescan
work as executed which is after the complete resume cycle
(mmc_pm_notify:PM_POST_SUSPEND). Is that an issue?

> mmc_schedule_delayed_work(&host->detect, delay);
> }
>
> @@ -2351,13 +2353,16 @@ void mmc_rescan(struct work_struct *work)
> struct mmc_host *host =
> container_of(work, struct mmc_host, detect.work);
> int i;
> + bool extend_wakeup = false;
>
> if (host->rescan_disable)
> return;
>
> /* If there is a non-removable card registered, only scan once */
> - if ((host->caps & MMC_CAP_NONREMOVABLE) && host->rescan_entered)
> + if ((host->caps & MMC_CAP_NONREMOVABLE) && host->rescan_entered) {
> + __pm_relax(host->ws);

By calling __pm_relax here, this indicates to me that is seems like
you might have prevented, even for a very small timeslot, with a
MMC_CAP_NONREMOVABLE card/host from the system to suspend.

For sure, you must not prevent the suspend even for small timeslots,
when MMC_CAP_NONREMOVABLE is set.

> return;
> + }
> host->rescan_entered = 1;
>
> mmc_bus_get(host);
> @@ -2400,16 +2405,27 @@ void mmc_rescan(struct work_struct *work)
>
> mmc_claim_host(host);
> for (i = 0; i < ARRAY_SIZE(freqs); i++) {
> - if (!mmc_rescan_try_freq(host, max(freqs[i], host->f_min)))
> + if (!mmc_rescan_try_freq(host, max(freqs[i], host->f_min))) {
> + /* stay awake extra time to process detected device */
> + extend_wakeup = true;
> break;
> + }
> if (freqs[i] <= host->f_min)
> break;
> }
> mmc_release_host(host);
>
> out:
> - if (host->caps & MMC_CAP_NEEDS_POLL)
> + if (extend_wakeup)
> + /* extra 1/2 second should be enough, hopefully */
> + __pm_wakeup_event(host->ws, MSEC_PER_SEC/2);
> + else
> + __pm_relax(host->ws);
> +
> + if (host->caps & MMC_CAP_NEEDS_POLL) {
> + __pm_stay_awake(host->ws);

This does not make sense.

So when using polling mode to detect card insert/remove, you will
prevent suspend forever? Maybe I missed a point somewhere?

> mmc_schedule_delayed_work(&host->detect, HZ);
> + }
> }
>
> void mmc_start_host(struct mmc_host *host)
> @@ -2433,7 +2449,8 @@ void mmc_stop_host(struct mmc_host *host)
> #endif
>
> host->rescan_disable = 1;
> - cancel_delayed_work_sync(&host->detect);
> + if (cancel_delayed_work_sync(&host->detect))
> + __pm_relax(host->ws);
> mmc_flush_scheduled_work();
>
> /* clear pm flags now and let card drivers set them as needed */
> @@ -2628,7 +2645,8 @@ int mmc_suspend_host(struct mmc_host *host)
> {

This function has become deprecated. You need to rebase this patch and
please do not add some new code in here.

> int err = 0;
>
> - cancel_delayed_work(&host->detect);
> + if (cancel_delayed_work(&host->detect))
> + __pm_relax(host->ws);
> mmc_flush_scheduled_work();
>
> mmc_bus_get(host);
> @@ -2741,7 +2759,8 @@ int mmc_pm_notify(struct notifier_block *notify_block,
> spin_lock_irqsave(&host->lock, flags);
> host->rescan_disable = 1;
> spin_unlock_irqrestore(&host->lock, flags);
> - cancel_delayed_work_sync(&host->detect);
> + if (cancel_delayed_work_sync(&host->detect))
> + __pm_relax(host->ws);
>
> if (!host->bus_ops || host->bus_ops->suspend)
> break;
> diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
> index 2a3593d..3cbb3d7 100644
> --- a/drivers/mmc/core/host.c
> +++ b/drivers/mmc/core/host.c
> @@ -452,6 +452,11 @@ struct mmc_host *mmc_alloc_host(int extra, struct device *dev)
> host->class_dev.class = &mmc_host_class;
> device_initialize(&host->class_dev);
>
> + host->ws = wakeup_source_register(
> + kasprintf(GFP_KERNEL, "%s_detect", mmc_hostname(host)));
> + if (!host->ws)
> + goto free;
> +
> mmc_host_clk_init(host);
>
> mutex_init(&host->slot.lock);
> @@ -555,6 +560,8 @@ void mmc_free_host(struct mmc_host *host)
> spin_lock(&mmc_host_lock);
> idr_remove(&mmc_host_idr, host->index);
> spin_unlock(&mmc_host_lock);
> + wakeup_source_unregister(host->ws);
> + host->ws = NULL;
>
> put_device(&host->class_dev);
> }
> diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
> index e326ae2..9dc2dd6 100644
> --- a/include/linux/mmc/host.h
> +++ b/include/linux/mmc/host.h
> @@ -15,6 +15,7 @@
> #include <linux/sched.h>
> #include <linux/device.h>
> #include <linux/fault-inject.h>
> +#include <linux/pm_wakeup.h>
>
> #include <linux/mmc/core.h>
> #include <linux/mmc/pm.h>
> @@ -329,6 +330,7 @@ struct mmc_host {
> int claim_cnt; /* "claim" nesting count */
>
> struct delayed_work detect;
> + struct wakeup_source *ws;
> int detect_change; /* card detect flag */
> struct mmc_slot slot;
>
> --
> 1.7.9.5
>

Kind regards
Ulf Hansson
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/