Subject: Re: gcc 2.95.2 sanity check, please
To: NetBSD/alpha Discussion List <port-alpha@NetBSD.ORG>
From: Eric Schnoebelen <eric@cirr.com>
List: port-alpha
Date: 01/26/2004 11:22:07
"Greg A. Woods" writes:
- [ On Friday, January 23, 2004 at 22:17:37 (-0600), Eric Schnoebelen wrote: 
- ]
- > Subject: Re: gcc 2.95.2 sanity check, please 
- >
- > 
- > "Greg A. Woods" writes:
- > - Expressions as simple as that above also occur in a bazillion other
- > - places in the whole code-base -- it's just an integer mask and test.
- > 
- > 	Yup, I'm aware of that.. Which is why I was so suprised
- > to see the truth path being taken when it was obvious it should
- > be false..
- 
- very strange indeed!
- 
- BTW, did you really mean "2.95.2", or was that supposed to be "2.95.3"?

	I meant 2.95.3.. Gosh, I don't know where I gathered
2.95.2 from..

- > 	Now, it's entirely possible that I was remembering
- > something from the FreeBSD/alpha system...
- > 
- > 	All the more annoying, since changing the clients from
- > using soft nfs mounts to hard (non-interruptable) mounts, nary a
- > vnode starvation has been seen.  Running the same kernel.
- 
- Hmm....  I've got my sources on NFS, and my NFS server suffers from an
- occasional "le0: device timeout" problem when under load (it should be
- using pcn(4), but this machine suffers the pcn MII panic-on-boot-
- if-link-enabled bug).  With most clients the result is that things seem
- to hang for a second or two, then all's well.  However on the alpha the
- build sometimes comes to a stop until I trigger some other NFS activity
- from anther session.  It did that when I was having supper and it stayed
- stuck until I tickled it with an "ls" in my NFS-mounted home dir a
- couple of hours later, and it's done it once more since.  On the console
- I get the expected "nfs server proven.weird.com:/work: not responding"
- error, and "nfs server proven.weird.com:/work: is alive again" when it
- starts working again (and sometimes, presumably depending on the file
- operation, I see the same messages in the window where the build is
- running).

	What does it say if you ask for a status (^T, frequently) on
the hung session?  We see it waiting with [vnlock].

	Aare you using hard mounts or soft mounts? Interrupts
permitted?

- I wonder if this is all just my imagination, or whether there could be
- anything to it that's related to what you're mentioning.

	You've got me wondering too..

	Unfortunately, our current configuration has the one _huge_
filesystem where everything interesting (aka /usr/pkg/*) is stored
mounted via NFS, and then the users home directories also mounted
via NFS (on a seperate mount point.)

--
Eric Schnoebelen		eric@cirr.com		http://www.cirr.com
    "Beware of bugs in the above code; I have only proved it correct, not
                   tried it." -- Donald Knuth