james@CSD4.MILW.WISC.EDU (Jim Lowe) (11/11/87)
What follows is a recap of the microcode problem that several sites have been seeing and converstions that I have had with people from CCI & Unisys. Currently, we (UWM) have 3 versions of microcode. One version (4/30/86) has a bug in the EMUL instruction. Another version (8/26/86) has an addressing bug that under certain circumstances can cause the machine to crash (reproducible). The third version (6/29/87) seems to work fine and is what we are currently running at UWM. From what I have been able to gather, the way to determine the versions of the microcode is by using "od -x /fppwcs". Check the address at location 264. This should contain the date of the microcode. You can also use "sum /fppwcs", but I think the "od" gives you a bit more information. Here is the output from "od" & "sum" on our system. We are running Berkeley's 4.3 tahoe-beta release from last April. The fppwcs with the bug in the EMUL intruction: od -x: 0000260 3040 010e 4308 6105 0040 010f 0540 0ff0 ^^^^^^^^^ version 4/30/86 105 sum: 36613 32 The fppwcs with the addressing bug that causes crashes: od -x : 0000260 3040 010e 8268 6100 0040 010f 0540 0ff0 ^^^^^^^^^ version 8/26/86 100 sum: 57552 32 New microcode from CCI: od -x: 0000260 3040 010e 6298 7102 0040 010f 0540 0ff0 ^^^^^^^^^ version 6/29/87 102 sum: 58295 32 The version of the microcode that we are currently running and that works is dated 6/29/87 #102. If you have a Unisys 7000/40 and would like to get the new microcode, the SUR number that we submitted the problem under was 50233707. If you purchased the machine from CCI, the HRR number is 31593. I don't know if this problem exists on the Harris HCX/7 or HCX/9. In any case, I would suggest picking up the 6/29/87 version of the microcode for your Tahoe. It is also suppose to have the interger multiply fixed for Sys V. - Jim Internet: james@csd4.milw.wisc.edu Uucp: ihnp4!uwmcsd1!james Csnet: james@uwm Phone: +1 (414) 229-6431
phillips@CS.BUFFALO.EDU (Gretchen Phillips) (11/11/87)
As a side note about the microcode. We have received the third version of the microcode and have a program which produces the following: trap type 8, code = d50ae10, pc = c000e572 panic: Segmentation fault trap type 8, code = d50ae10, pc = c000e572 panic: Segmentation fault And the machine dies. The program is a hypercube simulator with multiple forks and is thus rather hard to trace through. I have a couple of students working on it in an effort to isolate the exact configuration that crashes the machine. If we can get it to a small, reproducable example, I will forward it for others to try. Gretchen