Re: [PATCH 2/5] input: Use driver hint to compute the evdev buffer size (rev3)

[Ping Cheng]

Hi Henrik,

I like this patchset. Along with the mtdev patchset submitted to x.org
by Chase, I see a great collaboration for the MT support.  Keep up the
good work.

I have a minor comment inline.

Thank you.

Ping

On Wed, Jun 23, 2010 at 4:14 AM, Henrik Rydberg <rydberg@xxxxxxxxxxx> wrote:
> Some devices, in particular MT devices, produce a lot of data.  This
> leads to a high frequency of lost packets in evdev, which by default
> uses a fairly small event buffer. Let the drivers hint the average
> number of events per packet for the device by calling the
> input_set_events_per_packet(), and use that information when computing
> the evdev buffer size.
>
> Signed-off-by: Henrik Rydberg <rydberg@xxxxxxxxxxx>
> ---
>  drivers/input/evdev.c |    5 ++++-
>  include/linux/input.h |   17 +++++++++++++++++
>  2 files changed, 21 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/input/evdev.c b/drivers/input/evdev.c
> index 5d84e59..728802f 100644
> --- a/drivers/input/evdev.c
> +++ b/drivers/input/evdev.c
> @@ -11,6 +11,7 @@
>  #define EVDEV_MINOR_BASE       64
>  #define EVDEV_MINORS           32
>  #define EVDEV_MIN_BUFFER_SIZE  64
> +#define EVDEV_BUF_PACKETS      8
>
>  #include <linux/poll.h>
>  #include <linux/sched.h>
> @@ -51,7 +52,9 @@ static DEFINE_MUTEX(evdev_table_mutex);
>
>  static int evdev_compute_buffer_size(struct input_dev *dev)
>  {
> -       return EVDEV_MIN_BUFFER_SIZE;
> +       int nev = dev->hint_events_per_packet * EVDEV_BUF_PACKETS;
> +       nev = max(nev, EVDEV_MIN_BUFFER_SIZE);
> +       return roundup_pow_of_two(nev);

I think we have a backward compatibility issue here. This routine will
return 7 when nev falls to the default value
(EVDEV_MIN_BUFFER_SIZE/64).  This could happen to those drivers that
don't report MT events or forget/don't feel the need to set
hint_events_per_packet since the old BUFFER_SIZE worked perfectly for
them.  We need to keep the return value for those drivers as 64 so we
could allocate the same space as it was in [PATCH 1/5].

>  }
>
>  static void evdev_pass_event(struct evdev_client *client,
> diff --git a/include/linux/input.h b/include/linux/input.h
> index 20e4eac..9e024b6 100644
> --- a/include/linux/input.h
> +++ b/include/linux/input.h
> @@ -1162,6 +1162,8 @@ struct input_dev {
>        unsigned long ffbit[BITS_TO_LONGS(FF_CNT)];
>        unsigned long swbit[BITS_TO_LONGS(SW_CNT)];
>
> +       unsigned int hint_events_per_packet;
> +
>        unsigned int keycodemax;
>        unsigned int keycodesize;
>        void *keycode;
> @@ -1439,6 +1441,21 @@ static inline void input_mt_slot(struct input_dev *dev, int slot)
>
>  void input_set_capability(struct input_dev *dev, unsigned int type, unsigned int code);
>
> +/**
> + * input_set_events_per_packet - tell handlers about the driver event rate
> + * @dev: the input device used by the driver
> + * @nev: the average number of events between calls to input_sync()
> + *
> + * If the event rate sent from a device is unusually large, use this
> + * function to set the expected event rate. This will allow handlers
> + * to set up an approriate buffer size for the event stream, in order
> + * to minimize information loss.
> + */
> +static inline void input_set_events_per_packet(struct input_dev *dev, int nev)
> +{
> +       dev->hint_events_per_packet = nev;
> +}
> +
>  static inline void input_set_abs_params(struct input_dev *dev, int axis, int min, int max, int fuzz, int flat)
>  {
>        dev->absmin[axis] = min;
> --
> 1.6.3.3
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-input" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
--
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/