K7/Athlon optimizations and Sacrifices to the Great Ones.
I'm upset, and angry, and could go into more detail but a future employer might read this.
133Mhz FSB + KT133A chipset theory has been officially shot to hell. Not only that, but 6-4-4 (family/model/stepping) processors don't seem to be the culprit. I've now had reports of 6-4-2 experiencing problems, and 6-4-4 NOT experiencing problems, even on KT133A @ 133Mhz.
At this point, I can't even isolate a MOTHERBOARD that could be the culprit, and I don't think it's the power supply.
The *only* other theory I have left in my arsenal is cooling. Unfortunately this data is more difficult to obtain from users, and thus wouldn't gather as many responses and in the end it'd probably be useless. It would be best tested in controlled conditions, but unfortunately I don't have the money to purchase the neccisary hardware to test these issues. Any companies want to sponsor tests? I'm serious, if someone wanted to, I'd be willing to test this possibility. There is another remote possibility, and that's the fab plant that the processors came from. I didn't ask for CPU serial numbers, so I can't speak to that effect.
It's also possible that this is related to a specific batch or batches of KT133A chipsets, however I currently have one report of a guy seeing this problem on the SAME physical board, just two different processors. Both 6-4-2, the only difference is that one is 1.13Ghz and doesn't have the problem, and the other is 1.2Ghz and DOES have the problem. This of course leads me back to the clock speed theory, but again it doesn't make any SENSE because the FSB on both of them is 133Mhz and I've got at least two reports of 1.33Ghz chips running FINE! ARG!
At this point, I'm giving up on collecting data, as I just don't see a definitive pattern, all I can say for sure is that the "majority" KT133A-based motherboards seem to have problems, but not ALL. I don't know of a single report outside of the KT133A chipset of these problems.
If anyone wants to keep collecting data, I'd be happy to send all the information I have so far to that person, and any that filters in over the next couple days.
Now for the Sacrifices.
At this point, I'd like to sacrifice a Red Hat Linux 6.2 CD to Alan Cox.
I would also like to sacrifice Minix 1.3(?) installation diskettes to Linus Torvalds.
I perform these sacrifices in the hope that enlightenment comes to me.
Quoted 5 lines
At this point, I'd like to sacrifice a Red Hat Linux 6.2 CD to Alan Cox. I would also like to sacrifice Minix 1.3(?) installation diskettes to Linus Torvalds.
I perform these sacrifices in the hope that enlightenment comes to me.
A deep booming voice says "I have no idea either"
Quoted 5 lines · Alan Cox
Quoted 4 lines
At this point, I'd like to sacrifice a Red Hat Linux 6.2 CD to Alan Cox. I would also like to sacrifice Minix 1.3(?) installation diskettes to Linus Torvalds.
I perform these sacrifices in the hope that enlightenment comes to me.A deep booming voice says "I have no idea either"
We need a good tester (floppy-bootable k7-killer, something along the lines of memtest86) and many more data points.
Anyone yet verified if burnMMX2 causes the same failures the athlon-optimized kernel does?
-Dan
Signature · 1 line
[-] Omae no subete no kichi wa ore no mono da. [-]
Quoted 10 lines · Dan Hollis
Quoted 7 lines · Alan Cox
Quoted 5 lines
At this point, I'd like to sacrifice a Red Hat Linux 6.2 CD to Alan Cox. I would also like to sacrifice Minix 1.3(?) installation diskettes to Linus Torvalds.
I perform these sacrifices in the hope that enlightenment comes to me.A deep booming voice says "I have no idea either"
Quoted 7 lines
We need a good tester (floppy-bootable k7-killer, something along the lines of memtest86) and many more data points.
Anyone yet verified if burnMMX2 causes the same failures the athlon-optimized kernel does?
-Dan
MMX2 does not cause any problems for me. Robert (the guy who wrote these) has provided me with two more versions that mimic the Athlon optimized fast_page_copy and fast_page_clear functions in mmx.c. They aren't exact copies, but are close. One fails for me consistently, the other does not. The one that fails consistently is the one that mimics the fast_page_copy code. I'm still trying to provide him more datapoints about the failures to see if we can uncover anything.
Signature · 3 lines
David Hollister Driversoft Engineering: http://devicedrivers.com Digital Audio Resources: http://digitalaudioresources.org
Quoted 2 lines · Dan Hollis
Anyone yet verified if burnMMX2 causes the same failures the athlon-optimized kernel does?
several versions of Robert's burnMMX2 run on by problematic athlons without failing for several hours.
I did some trial and error modifications to mmx.c and found out that this one makes my athlons happy (but must admin i have no clue why). it seems to run stable now.
arch/i386/lib/mmx.c +3 -0
@@ -194,6 +194,9 @@ : : "r" (from), "r" (to) : "memory"); from+=64 ; to+=64 ; + __asm__ __volatile__ ( +" sfence \n" : : + ); }for (i=(4096 -320 )/64 ; i<4096 /64 ; i++) {
maybe someone with more knowledge can take a look..
ciao,
heinz
Signature · 3 lines
Heinz Deinhart <[email protected]> +43 1 58801-18321 Technische Universitaet Wien, Dept. E183/1
Quoted 8 lines
I did some trial and error modifications to
mmx.cand found out that this one makes my athlons happy (but must admin i have no clue why). it seems to run stable now.
arch/i386/lib/mmx.c line 202
Quoted 9 lines
arch/i386/lib/mmx.c +3
-0@@ -194,6 +194,9 @@ : : "r" (from), "r" (to) : "memory"); from+=64 ; to+=64 ; + __asm__ __volatile__ ( +" sfence \n" : : + ); }for (i=(4096 -320 )/64 ; i<4096 /64 ; i++) {
You are effectively continually stalling the processor so that the fast streaming memory transfers dont occur. Instead you start, block, start, block, ...
Alan
Thursday, September 06, 2001, 10:51:36 PM, Nicholas Knight <[email protected]> wrote:
NK> 133Mhz FSB + KT133A chipset theory has been officially shot to hell. NK> Not only that, but 6-4-4 (family/model/stepping) processors don't seem to NK> be the culprit. I've now had reports of 6-4-2 experiencing problems, and NK> 6-4-4 NOT experiencing problems, even on KT133A @ 133Mhz.
NK> At this point, I can't even isolate a MOTHERBOARD that could be the NK> culprit, and I don't think it's the power supply.
...
NK> At this point, I'm giving up on collecting data, as I just don't see a NK> definitive pattern, all I can say for sure is that the "majority" NK> KT133A-based motherboards seem to have problems, but not ALL. I don't NK> know of a single report outside of the KT133A chipset of these problems.
Well, why can't guys with Athlons and KT133As who did enable K7 optimizations just open their boxes and report to Nicholas:
* processor and chipset markings
* bus speed
* CPUcore/bus multiplier
* Motherboard model
* BIOS manufacturer, version, date
This is important. BIOS might fix/mask chipset bugs by programming it to stable but slow cfg
* do they see K7 related oops
("I don't see oops" is a valuable report too!)
* did oops go away with 100MHz FSB
* did oops go away with different CPU voltage
* did oops go away with smaller multiplier
* did oops go away with BIOS update
* did oops go away with any trick with mmx.c - see below. More advanced reporters might try to fiddle with arch/i386/lib/mmx.c and try to make oops disappear
* did oops go away with K7 optimization off
* results of memtest86
* results of running burnK7/burnMMX
(Last 3 *'s to make sure it is the K7 bug, not bad memory or something)
Need to check for oops? "Simen Thoresen" <[email protected]>:
Quoted 4 lines
I've determined that with the Athlon-optimized
fast_copy_page, the machine is easy to push into oopsing. Just starting a dd with blocksize 128M (half available ram) provokes an oops. This is repeatable, consistent and almost fun.
Since fast_copy_page() from arch/i386/lib/mmx.c has been isolated as a code which triggers oops, it can be instrumented to check whether page is indeed copied right by questionable K7 code and barf loudly if it is not.
Since oops are not instant, looks like interrupts might interfere with movntq instruction... On the other hand, fast_clear_page() isn't triggering oops (right?) so maybe mixing normal and cache-bypassing instructions is triggering oops...
Comparing K7 and MMX fast_copy_page...
Does replacing movntq->movq fix makes oops go avay?
If no, does this (or similar) change makes oops go away?
movq (%0), %%mm0 -> movq (%0), %%mm0 movntq %%mm0, (%1) -> movq 8(%0), %%mm1 movq 8(%0), %%mm1 -> movq 16(%0), %%mm2 movntq %%mm1, 8(%1) -> movq 24(%0), %%mm3 movq 16(%0), %%mm2 -> movntq %%mm0, (%1) movntq %%mm2, 16(%1) -> movntq %%mm1, 8(%1) movq 24(%0), %%mm3 -> movntq %%mm2, 16(%1) movntq %%mm3, 24(%1) -> movntq %%mm3, 24(%1) movq 32(%0), %%mm4 -> movq 32(%0), %%mm4 movntq %%mm4, 32(%1) -> movq 40(%0), %%mm5 movq 40(%0), %%mm5 -> movq 48(%0), %%mm6 movntq %%mm5, 40(%1) -> movq 56(%0), %%mm7 movq 48(%0), %%mm6 -> movntq %%mm4, 32(%1) movntq %%mm6, 48(%1) -> movntq %%mm5, 40(%1) movq 56(%0), %%mm7 -> movntq %%mm6, 48(%1) movntq %%mm7, 56(%1) -> movntq %%mm7, 56(%1)
No? Changing first for() loop from
for(i=0; i<(4096-320)/64; i++) into
for(i=0; i<4096/64; i++) and eliminating second for() - does this help?
One of above changes HAS to fix K7 oops, because you convert K7 fast_copy_page to MMX fast_copy_page that way :-)
So if you have an Athlon - try these and report. I don't have the hardware.
Quoted 5 lines · David Hollister
MMX2 does not cause any problems for me. Robert (the guy who wrote these) has provided me with two more versions that mimic the Athlon optimized
fast_page_copyandfast_page_clearfunctions inmmx.c. They aren't exact copies, but are close. One fails for me consistently, the other does not. The one that fails consistently is the one that mimics thefast_page_copycode.
Robert Redelmeier: [email protected] http://users.ev1.net/~redelm/
Although this tester is not on his page (yet?).
Signature · 3 lines
Best regards, VDA mailto:[email protected] http://port.imtp.ilyichevsk.odessa.ua/vda/
Quoted 7 lines · Nicholas Knight
problem on the SAME physical board, just two different processors. Both 6-4-2, the only difference is that one is 1.13Ghz and doesn't have the problem, and the other is 1.2Ghz and DOES have the problem. This of course leads me back to the clock speed theory, but again it doesn't make any SENSE because the FSB on both of them is 133Mhz and I've got at least two reports of 1.33Ghz chips running FINE! ARG!
How about the synchronization issue that came up in the power saving thread? Some bus synchronization problem (integer/non-integer multiplier)?
- Jussi Laako
Signature · 2 lines
PGP key fingerprint: 161D 6FED 6A92 39E2 EB5B 39DD A4DE 63EB C216 1E4B Available at PGP keyservers