Re: [PATCH] HID: i2c-hid: add runtime PM support

[Mika Westerberg]

On Mon, Jan 27, 2014 at 10:36:25PM -0500, Benjamin Tissoires wrote:
> On Tue, Jan 14, 2014 at 5:13 AM, Mika Westerberg
> <mika.westerberg@xxxxxxxxxxxxxxx> wrote:
> > This patch adds runtime PM support for the HID over I2C driver. When the
> > i2c-hid device is first opened we power it on and on the last close we
> > power it off.
> >
> > The implementation is not the most power efficient because it needs some
> > interaction from the userspace (e.g close the device node whenever we are
> > no more interested in getting events), nevertheless it allows us to save
> > some power and works with devices that are not wake capable.
> >
> 
> Hi Mika,
> 
> I am a little bit puzzled here. The commit message just says that you
> changed the implementation of the power saving with the exact same
> behavior...  At least that's what I understand.
> Currently, the devices should be put on sleep if nobody is reading,
> and back alive if a reader arrives.
> I think there is a gain with the patch, but my knowledge of the pm
> subsystem is far too limited to see it :(

Yes, there should be gain. If there is power domain involved like, ACPI in
our case, runtime suspending the device on close will let ACPI to move the
device to D3cold (e.g full off) which saves more power.

> > Signed-off-by: Mika Westerberg <mika.westerberg@xxxxxxxxxxxxxxx>
> > ---
> >  drivers/hid/i2c-hid/i2c-hid.c | 81 ++++++++++++++++++++++++++++---------------
> >  1 file changed, 54 insertions(+), 27 deletions(-)
> >
> > diff --git a/drivers/hid/i2c-hid/i2c-hid.c b/drivers/hid/i2c-hid/i2c-hid.c
> > index d1f81f52481a..ff767d03d60e 100644
> > --- a/drivers/hid/i2c-hid/i2c-hid.c
> > +++ b/drivers/hid/i2c-hid/i2c-hid.c
> > @@ -25,6 +25,7 @@
> >  #include <linux/delay.h>
> >  #include <linux/slab.h>
> >  #include <linux/pm.h>
> > +#include <linux/pm_runtime.h>
> >  #include <linux/device.h>
> >  #include <linux/wait.h>
> >  #include <linux/err.h>
> > @@ -454,10 +455,18 @@ static void i2c_hid_init_reports(struct hid_device *hid)
> >                 return;
> >         }
> >
> > +       /*
> > +        * The device must be powered on while we fetch initial reports
> > +        * from it.
> > +        */
> > +       pm_runtime_get_sync(&client->dev);
> > +
> >         list_for_each_entry(report,
> >                 &hid->report_enum[HID_FEATURE_REPORT].report_list, list)
> >                 i2c_hid_init_report(report, inbuf, ihid->bufsize);
> >
> > +       pm_runtime_put(&client->dev);
> > +
> >         kfree(inbuf);
> >  }
> >
> > @@ -703,8 +712,8 @@ static int i2c_hid_open(struct hid_device *hid)
> >
> >         mutex_lock(&i2c_hid_open_mut);
> >         if (!hid->open++) {
> > -               ret = i2c_hid_set_power(client, I2C_HID_PWR_ON);
> > -               if (ret) {
> > +               ret = pm_runtime_get_sync(&client->dev);
> 
> dummy question (kind of late here...). Is there a counter of how many
> get/put has been called in pm_runtime which could allow us to get rid
> of the hid->open count?

There is a counter but I'm not sure if drivers are supposed to use that. I
would prefer not to use that.

> 
> > +               if (ret < 0) {
> >                         hid->open--;
> >                         goto done;
> >                 }
> > @@ -712,7 +721,7 @@ static int i2c_hid_open(struct hid_device *hid)
> >         }
> >  done:
> >         mutex_unlock(&i2c_hid_open_mut);
> > -       return ret;
> > +       return ret < 0 ? ret : 0;
> >  }
> >
> >  static void i2c_hid_close(struct hid_device *hid)
> > @@ -729,37 +738,17 @@ static void i2c_hid_close(struct hid_device *hid)
> >                 clear_bit(I2C_HID_STARTED, &ihid->flags);
> >
> >                 /* Save some power */
> > -               i2c_hid_set_power(client, I2C_HID_PWR_SLEEP);
> > +               pm_runtime_put(&client->dev);
> >         }
> >         mutex_unlock(&i2c_hid_open_mut);
> >  }
> >
> > -static int i2c_hid_power(struct hid_device *hid, int lvl)
> > -{
> > -       struct i2c_client *client = hid->driver_data;
> > -       struct i2c_hid *ihid = i2c_get_clientdata(client);
> > -       int ret = 0;
> > -
> > -       i2c_hid_dbg(ihid, "%s lvl:%d\n", __func__, lvl);
> > -
> > -       switch (lvl) {
> > -       case PM_HINT_FULLON:
> > -               ret = i2c_hid_set_power(client, I2C_HID_PWR_ON);
> > -               break;
> > -       case PM_HINT_NORMAL:
> > -               ret = i2c_hid_set_power(client, I2C_HID_PWR_SLEEP);
> > -               break;
> > -       }
> > -       return ret;
> > -}
> > -
> >  static struct hid_ll_driver i2c_hid_ll_driver = {
> >         .parse = i2c_hid_parse,
> >         .start = i2c_hid_start,
> >         .stop = i2c_hid_stop,
> >         .open = i2c_hid_open,
> >         .close = i2c_hid_close,
> > -       .power = i2c_hid_power,
> 
> If I understand correctly, here you are trying to fix hidraw (with
> i2c_hid tramsport) which used to set_power on/off twice with the first
> reader, right?

Right.

> I don't think we have other i2c_hid users of hid_hw_power, but I am a
> little bit worried of simply removing the callback.

OK.

> What if we just change i2c_hid_set_power in i2c_hid_power by the
> corresponding pm_runtime calls?

Sure, I'll change that in the next version.

> 
> >         .request = i2c_hid_request,
> >  };
> >
> > @@ -973,13 +962,17 @@ static int i2c_hid_probe(struct i2c_client *client,
> >         if (ret < 0)
> >                 goto err;
> >
> > +       pm_runtime_get_noresume(&client->dev);
> > +       pm_runtime_set_active(&client->dev);
> > +       pm_runtime_enable(&client->dev);
> > +
> >         ret = i2c_hid_fetch_hid_descriptor(ihid);
> >         if (ret < 0)
> > -               goto err;
> > +               goto err_pm;
> >
> >         ret = i2c_hid_init_irq(client);
> >         if (ret < 0)
> > -               goto err;
> > +               goto err_pm;
> >
> >         hid = hid_allocate_device();
> >         if (IS_ERR(hid)) {
> > @@ -1010,6 +1003,7 @@ static int i2c_hid_probe(struct i2c_client *client,
> >                 goto err_mem_free;
> >         }
> >
> > +       pm_runtime_put(&client->dev);
> >         return 0;
> >
> >  err_mem_free:
> > @@ -1018,6 +1012,10 @@ err_mem_free:
> >  err_irq:
> >         free_irq(client->irq, ihid);
> >
> > +err_pm:
> > +       pm_runtime_put_noidle(&client->dev);
> > +       pm_runtime_disable(&client->dev);
> > +
> >  err:
> >         i2c_hid_free_buffers(ihid);
> >         kfree(ihid);
> > @@ -1029,6 +1027,11 @@ static int i2c_hid_remove(struct i2c_client *client)
> >         struct i2c_hid *ihid = i2c_get_clientdata(client);
> >         struct hid_device *hid;
> >
> > +       pm_runtime_get_sync(&client->dev);
> > +       pm_runtime_disable(&client->dev);
> > +       pm_runtime_set_suspended(&client->dev);
> > +       pm_runtime_put_noidle(&client->dev);
> > +
> >         hid = ihid->hid;
> >         hid_destroy_device(hid);
> >
> > @@ -1074,7 +1077,31 @@ static int i2c_hid_resume(struct device *dev)
> >  }
> >  #endif
> >
> > -static SIMPLE_DEV_PM_OPS(i2c_hid_pm, i2c_hid_suspend, i2c_hid_resume);
> > +#ifdef CONFIG_PM_RUNTIME
> > +static int i2c_hid_runtime_suspend(struct device *dev)
> > +{
> > +       struct i2c_client *client = to_i2c_client(dev);
> > +
> > +       i2c_hid_set_power(client, I2C_HID_PWR_SLEEP);
> > +       disable_irq(client->irq);
> > +       return 0;
> > +}
> > +
> > +static int i2c_hid_runtime_resume(struct device *dev)
> > +{
> > +       struct i2c_client *client = to_i2c_client(dev);
> > +
> > +       enable_irq(client->irq);
> > +       i2c_hid_set_power(client, I2C_HID_PWR_ON);
> > +       return 0;
> > +}
> 
> These two functions looks very similar to i2c_hid_suspend and
> i2c_hid_resume, without the reset and the irq_wake :(
> So, my question here is can we use some common code for them?
> It may not be possible regarding CONFIG_PM_RUNTIME and
> CONFIG_PM_SLEEP, but it still looks ugly to me.

It looks ugly, I agree and we can probably reuse some code in the
callbacks. I'll look into that.

> I tested this today, and it works, so you should be right, but I'd
> like to have your opinion on this.

Thanks for testing and comments.

I'll prepare a new version with the suggested changes if you are OK with my
explanations ;-)
--
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/