Re: 2.6.19 -mm merge plans (input patches)

["Dmitry Torokhov" <dmitry.torokhov@gmail.com>]

On 9/21/06, James Bruce <bruce@andrew.cmu.edu> wrote:
> Dmitry Torokhov wrote:
> > On Wednesday 20 September 2006 17:55, Dave Jones wrote:
> >> On Wed, Sep 20, 2006 at 05:29:43PM -0400, Dmitry Torokhov wrote:
> >>  > On 9/20/06, Andrew Morton <akpm@osdl.org> wrote:
> >>  > > remove-silly-messages-from-input-layer.patch
> >>  >
> >>  > I firmly believe that we should keep reporting these conditions. This
> >>  > way we can explain why keyboard might be losing keypresses. I am open
> >>  > to changing the message text. Would "atkbd.c: keyboard reported error
> >>  > condition (FYI only)" be better?
> >>
> >> Q: What do you expect users to do when they see the message?
> >
> > A: Nothing. But when they tell me that sometimes they lose keystrokes I
> > can ask them if they see it in dmesg. And if they see it there is nothing
> > I can do. Again, if you could suggest a better wording that would not alarm
> > unsuspecting users that would be great.
>
> If it is needed only to answer "does my keyboard work", maybe you could
> store an error count in the driver, or put it to the event layer.
> Coupled with a way to retrieve the value (ioctl+evtest,proc,sys,etc),
> the user can get the information they need, but only when they actually
> want it.
>
> The networking subsystem seems to store a lot of its error conditions in
> proc-accessible counters rather than printing a warning.  Thus there is
> precedent for avoiding dmesg spam in this way.
>

That is actually a pretty good idea. I'll add a sysfs counter
attribute and remove that printk, unless there are objections.

-- 
Dmitry