Re: tty breakage in X (Was: tty vs workqueue oddities)

[Felipe Balbi <balbi@ti.com>]

[-- Attachment #1: Type: text/plain, Size: 1753 bytes --]

Hi,

On Tue, Jun 07, 2011 at 07:44:48PM -0700, Linus Torvalds wrote:
> On Mon, Jun 6, 2011 at 7:24 AM, Guillaume Chazarain <guichaz@gmail.com> wrote:
> >
> > After reverting http://git.kernel.org/linus/a5660b4 "tty: fix endless
> > work loop when the buffer fills up" I cannot reproduce the hangs on
> > SMP anymore but it brings back the busy loop on UP.
> 
> Hmm. The n_tty layer has some rather distressing locking, and doesn't
> lock "tty->receive_room" changes at all, for example (and uses
> multiple locks for some other things).
> 
> It may well be that there is some SMP race there. The n_tty line
> discipline has its own locking for its counts, and the tty buffer code
> has its own locking, and "receive_room" kind o fends up being in the
> middle between them.
> 
> The sad part is that the patch that made receive_buf() return the
> amout of bytes received was actually trying to do the right thing, it
> just did it entirely in the wrong way (re-introducing the crazy
> re-arming of the workqueue from within itself, and using all the wrong
> sign issues).

yeah sorry about that. I originally wrote that patch over one year ago
and had to send it three times until it was finally noticed. Then I had
to do a quick rebase and things went pretty bad.

Should've checked better and forget about loosing another merge window.

> I'd love to get rid of receive_room entirely - and just letting the
> tty line discipline handler say how much it actually received. in
> other words, having receive_buf() just tell us how much it used, and
> not looking at receive_room in the caller is absolutely the right
> thing.

that was the idea, unfortunately I missed the last part. Sorry about
that.

-- 
balbi

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 490 bytes --]