Path: utzoo!attcan!uunet!tektronix!tekgen!puffin!penguin!richl
From: richl@penguin.USS.TEK.COM (Rick Lindsley)
Newsgroups: comp.unix.wizards
Subject: Re: cdc disk on uVAXII
Message-ID: <297@puffin.USS.TEK.COM>
Date: 24 May 88 20:13:45 GMT
References: <8805181321.AA02870@jade.berkeley.edu> <11569@mimsy.UUCP>
Sender: news@puffin.USS.TEK.COM
Reply-To: richl@penguin.USS.TEK.COM (Rick Lindsley)
Organization: Tektronix, Inc., Beaverton, OR.
Lines: 24

In article <11569@mimsy.UUCP> chris@mimsy.UUCP (Chris Torek) writes:

    The old driver is full of mysterious bugs.  In particular, a number
    of people claim it tends to hang when using multiple controllers on
    a uVax II.  (I have had one or two reports that my rewritten driver
    also hangs, and one that it fixes the problem.  Which is right I
    cannot say.)

Yes, the problem is real, though I cannot say if Chris's driver fixes
it, not having seen it. The fix is twofold. First, on uVaxes,
everything interrupts at spl7. If you write or use a routine that could
be called at interrupt level, it sure better not lower the priority
below spl7 unless you are really at an interruptible state. At many
places in the uda driver, you aren't, but it does an implicit spl6
(splbio) or even an explicit spl5. Problems abound once your data
structures get mucked up.

Second, since everything happens at spl7, and interrupts do not appear
to be stacked in any way, you do miss interrupts. It's unclear whether
this was due entirely to the spl problem or some other strangeness on
the part of uVaxen, but a more stringent watchdog routine fixed that.
Between the two of them, the uda driver now works fine on our uVaxen.

Rick