A Weird K6 Bug : News

97/09/23:
- I have received a disturbing email that shows that there are some K6 with serial number B9730xxx that still have the bug. That's strange because there is a B9729xxx that has no problem.
97/09/22:
- I have made a patch, for Linux 2.1.56, that tries to work around the 2.6.2 bug. See the Tests page for details.
- The k6bugs.zip package has been modified to be usable with almost any C compiler. It should make Windows users life a lot easier.
97/09/11:
- For technical reasons, these pages have been moved from a NT machine to a Solaris machine. It is so much better...
- PC World Online has published an article about the K6 bug : AMD K6 Sprouts Rare Bug.
97/09/10:
- I have updated the FAQ as promised a few days ago.
- Minor updates everywhere.
97/09/06:
- I have prepared a much smaller k6bugs.zip, for Micro$oft non-operating systems. The previous one was swallowed by a NT crash.
97/09/05:
- I have received a convincing e-mail from AMD that explains how the 2.6.2 erratum is related to the 32MB limit. I have modified the Tests page accordingly. The FAQ will be updated soon.
- I would like to thank AMD for its great support.
97/09/04:
- The FAQ has been updated with AMD's reponse.
- A new test about self-modifying code as been added to the Tests page.
97/08/30:
- Yet another K6 that does not have the problem. Once again a high serial number. That's not surprising anymore.
- The problem has been confirmed on non-Intel chipsets. That's also not surprising anymore.
- Quote of the week, found two clicks away from this page :
```
If the bug occurs on a NT system, it's not that 
there is a problem with the K6 and NT, but rather 
the kernels are the one's compiled by Linux.
```
  No comment ;-)
- the linux-stack patch, suggested on the linux-kernel mailing list, does not work. See the Testspage for details.
97/08/29:
- AMD says :
  1. the bug is the one documented in section 2.6.2 of the AMD-K6 MMX Enhanced Processor Revision Guide.
  2. it has been corrected in later chips produced in the B stepping.
  3. current machines are shipped with the corrected chips.
- I agree with points 2 and 3... Read the Is AMD aware of the problem ? FAQ for details.
- The CreaWeb server has been unavailable for long hours yesterday, because of system crashs. It uses MicroS**t unreliable non-operating systems. Send a note to webmaster@creaweb.fr if these problems continue.
97/08/27:
- I have received the serial number of a chip that looks ok. It is again a high serial number. Strange...
- There are at least two reports of the bug with VIA chipsets. This will be confirmed after a few tests, just to be sure it is really the same problem.
97/08/25:
- This page is now referenced by TOM'S AMD INFO LINKS.
- According to the AMD-K6(TM) Appreciation Center more and more people are wondering about this bug.
- As far as I know, Jeff Wiegley was the first one to send a public report of the bug, so he has been designated to find a name for it ;-).
- We are searching for volunteers to run the Disable the L1 cache test. This is a long test, because the K6 is ten times slower without the L1 cache...
  First results seem to indicate that the bug disappear when the L1 cache is turned off.
97/08/24:
- I have added the The chessboard test, which is, IMHO, a major progress in the understanding of the problem.
97/08/22:
- The burnit script contained a small bug that made it create it too many directories in /usr/src/results. It made detection of silent failures (i.e. no signal, but corrupted object files) more difficult. It has been fixed.
- I have received another report of a bug-free K6. Once again, the CPU is very recent. I don't know its serial number yet.
- I have added a new page : Tests. The purpose of this page is to help understanding the bug. It probably already contains something interesting...
- AMD has been contacted by a few guys from the black list about the bug. Still no reaction, as far as I know...
97/08/19:
- I have updated the FAQ with a few K6 serial numbers.
- It looks like the FreeBSD guys are discussing the very same problem on the freebsd-hardware mailing list.
- The problem (someone has a good name for this bug ?) has been reproduced on Windows 95 OSR2, with the same test as on NT.

[Main page] [FAQ page] [Tests page]

Updated: 97/09/23.
Send comments to poulot@wanadoo.fr