[2268] in Release_7.7_team
Special instructions for 8.4 beta on R5000 Rev. 1.x Indy
daemon@ATHENA.MIT.EDU (Robert A Basch)
Wed May 17 18:48:03 2000
Message-Id: <200005172247.SAA411889@aupair.mit.edu>
To: tlyu@MIT.EDU, jlwfnord@MIT.EDU
cc: release-team@MIT.EDU
Date: Wed, 17 May 2000 18:47:55 -0400
From: Robert A Basch <rbasch@MIT.EDU>
Hi.
You are listed as the contacts for the following machines in the
SGI beta cluster:
capacitor-bank
special-forces
These machines have the R5000 Revision 1.0 processor; there is a
serious bug in IRIX 6.5.7 which breaks the ability to run binaries out
of AFS on R5000 rev 1.x machines. The bug is in kernel code which
attempts to work around a flaw in the processor.
Since Athena 8.4 (which will be in beta shortly) is based on IRIX
6.5.7, the update would fail on these machines as things stand now.
However, there is a way to disable the kernel workaround, so that AFS
binaries would run as expected. For the beta, we are asking you to
disable the workaround manually, as described below, *before* you take
the update; we expect to have a better solution (a software fix, or a
hardware replacement) in time for the general 8.4 release.
Here are the directions for disabling the software workaround which
breaks AFS under IRIX 6.5.7 on R5000 rev 1.x Indy's:
Bring the machine down, and enter the PROM command monitor. At the
">>" prompt, enter:
setenv -p _R5000_CVT_WAR 0
This sets a persistent special environment variable which will disable
the workaround in the kernel. This is required in order for the
update to the Athena 8.4 beta to succeed.
The processor revision can be obtained by running the IRIX command
"hinv -t cpu".
(The above setting can be undone by doing "unsetenv _R5000_CVT_WAR" at
PROM command level; this should be done after we have a fix installed
for the AFS problem under 6.5.7, since disabling the workaround leaves
the machine vulnerable to the processor flaw).
The mipscheck(1) man page contains some more information about the
problem, though it should be noted that the processor flaw actually
exists in all rev 1.x R5000's; according to the R5000 errata sheet,
the flaw was corrected in rev 2.1:
-cvtl[:action...]
Look for cvt.s.l and cvt.d.l instructions. These instructions
convert 64-bit integers to single or double floating point
format.
Revision [1.1] of the r5000 can misexecute cvt.s.l and cvt.d.l
instructions when the 64-bit integer input data is in either of the
following ranges:
0x7FF0 0000 0000 0000 to 0x7FFF FFFF FFFF FFFF
0x8000 0000 0000 0000 to 0x800F FFFF FFFF FFFF
When input data is in the preceding ranges, these instructions are
supposed to trap into the kernel where they will be emulated in
software. Unfortunately, they do not trap and they generate an
incorrect result. These instructions are fairly rare and are found in
mips3 and mips4 executables only; they are never in mips1 or mips2
programs. There is a work-around for this problem, implemented
entirely within the operating system kernel, which should be invisible
to all user programs. See the r5000 errata sheet for more details.
Please contact me if you have any questions about this.
Bob
