Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!cs.utexas.edu!uunet!munnari!otc!metro!basser!natmlab!ditsyda!evans
From: evans@ditsyda.oz (Bruce Evans)
Newsgroups: comp.os.minix
Subject: Re: bug in minix fs/devicce.c (causing deadlock)
Message-ID: <1954@ditsyda.oz>
Date: 24 Apr 89 19:30:04 GMT
References: <443@prles2.UUCP>
Organization: CSIRO DIT Sydney, Australia
Lines: 26

in article <443@prles2.UUCP>, meulenbr@cst.prl.philips.nl (Frans Meulenbroeks) says:
> [fs sometimes sends a CANCEL message which is never replied to]
> fix:
> The problem can be fixed in several ways:
> - always give a reply back on the CANCEL message.
>   however withing the pc tty driver a warning for races is given in the
>   do_cancel routine.
> - check in the routine rw_dev for the condition described above and deal
>   with it.
> 
> I've chosen for the second approach.

My PC TTY driver fixes it using the 1st approach. The race conditions
are avoided by avoiding duplicate cancellations. It is unreasonable for
a task not to reply to a sendrec()!

The 1.3d printer driver has the same bad behaviour. It only replies to
messages it understands. So an ioctl() to the printer will hang the
system.

> PS: this was a tough one.

Same. How can we avoid duplicating the effort of fixing bugs for the PC,
the ST, and out of date versions of both?

Bruce Evans             evans@ditsyda.oz.au