From: Dmitry Torokhov <dmitry.torokhov@gmail.com>
To: Anshul Garg <aksgarg1989@gmail.com>
Cc: linux-input@vger.kernel.org,
"anshul.g@samsung.com" <anshul.g@samsung.com>
Subject: Re: [PATCH] Input:Flush client events on clk_type change
Date: Thu, 8 Jan 2015 13:41:09 -0800 [thread overview]
Message-ID: <20150108214109.GD23256@dtor-ws> (raw)
In-Reply-To: <CA+HOOshSijYZnshdKhV+uPW5NKhbQoOZfv3yptsR7Bv2CD7cUA@mail.gmail.com>
On Thu, Jan 08, 2015 at 07:27:59PM +0530, Anshul Garg wrote:
> Dear Mr. Dmitry ,
>
> Thanks a lot for your suggestions.
>
> > + spin_lock_irq(&dev->event_lock);
> > + spin_lock(&client->buffer_lock);
> > + spin_unlock(&dev->event_lock);
>
> Umm, why?
>
> Yes, there is no need of event_lock as we are modifying client
> specific data structure only.
>
> Hence only buffer_lock will guarantee atomicity for flushing of client
> pending events buffer.
>
> So i will modify locking mechanism to use buffer_lock only.
>
> > +
> > + /* Flush clients events after clk_type is changed
> > + * and queue SYN_DROPPED event.*/
> > + client->packet_head = client->head = client->tail;
> > + spin_unlock_irq(&client->buffer_lock);
> > +
> > + evdev_queue_syn_dropped(client);
>
> This is still racy. I'd rather we passed a flag to
> evdev_queue_syn_dropped() to indicate it should also clear queue.
>
> Can you please tell me in which scenario's this patch is prone to race
> condition's?
> As i think we are modifying the client's buffer indexes so buffer_lock
> would be sufficient.
New events may come up between resetting the queue and queuing EV_SYN
and client would not really know if they contain valid time or not.
>
>
> Yes by adding one more parameter in evdev_queue_syn_dropped function
> on the basis
> of which we can flush the buffer.
>
> Example ::
>
> static void evdev_queue_syn_dropped(struct evdev_client *client)
>
> It can be changed to
>
> static void evdev_queue_syn_dropped(struct evdev_client *client , bool flush)
> {
> spin_lock(buffer_lock);
>
> if(flush)
> client->packet_head = client->head = client->tail;
>
>
> .........
>
> spin_unlock(buffer_lock);
> }
>
> OR
> Similarly we can extend__evdev_flush_queue function to support
> flushing of client event queue.
> As currently this function flushes single type of events only.
>
> I think 2nd way is better.
>
> Please give your insignt on above suggested changes.
I do not see how make evdev_flush_queue() to flush all types of events
without adding another parameter that would "override" type, which is
ugly.
It looks like we can make evdev_queue_syn_dropped() zap the old events
unconditionally, so I'd rather do that.
Thanks.
--
Dmitry
next prev parent reply other threads:[~2015-01-08 21:41 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-01-07 18:56 [PATCH] Input:Flush client events on clk_type change Anshul Garg
2015-01-07 19:32 ` Dmitry Torokhov
2015-01-08 13:57 ` Anshul Garg
2015-01-08 21:41 ` Dmitry Torokhov [this message]
-- strict thread matches above, loose matches on Subject: below --
2015-01-07 18:26 Anshul Garg
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20150108214109.GD23256@dtor-ws \
--to=dmitry.torokhov@gmail.com \
--cc=aksgarg1989@gmail.com \
--cc=anshul.g@samsung.com \
--cc=linux-input@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).