Skip to contentSkip to conversation

AMD64 Machine hardlocks when using memset

DISCUSSION Robert Hancock 4 participants 6 messages Apr 6, 2005 8 messages not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

#8 lore raw
Quoted 8 lines · Alan Cox
Quoted 1 line · Robert Hancock

I'm wondering if one does a ton of these cache-bypassing stores whether something gets hosed because of that. Not sure what that could be though. I don't imagine the chipset is involved with any of that on the Athlon 64 - either the CPU or RAM seems the most likely suspect to me

The glibc version is essentially the "perfect" copy function for the CPU. If you have any bus/memory problems or chipset bugs it will bite you.

Anyone have any suggestions on how to track this further? It seems fairly clear what circumstances are causing it, but as for figuring out what's at fault..

Signature · 3 lines
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from [email protected]
Home Page: http://www.roberthancock.com/
#9 lore raw

Hi,

Quoted 9 lines · Robert Hancock
Quoted 8 lines · Alan Cox
Quoted 4 lines · Robert Hancock

I'm wondering if one does a ton of these cache-bypassing stores whether something gets hosed because of that. Not sure what that could be though. I don't imagine the chipset is involved with any of that on the Athlon 64 - either the CPU or RAM seems the most likely suspect to me

Quoted 7 lines
Quoted 3 lines

The glibc version is essentially the "perfect" copy function for the CPU. If you have any bus/memory problems or chipset bugs it will bite you.

Anyone have any suggestions on how to track this further? It seems fairly clear what circumstances are causing it, but as for figuring out what's at fault..

Well, I would start from changing memory modules.

Greets,
Rafael

Signature · 3 lines
- Would you tell me, please, which way I ought to go from here?
- That depends a good deal on where you want to get to.
		-- Lewis Carroll "Alice's Adventures in Wonderland"
#10 lore raw
Quoted 6 lines · Rafael J. Wysocki
Quoted 3 lines

Anyone have any suggestions on how to track this further? It seems fairly clear what circumstances are causing it, but as for figuring out what's at fault..

Well, I would start from changing memory modules.

As I wrote earlier, I tried 4 different (but same brand) modules, 2 Infineon and 2 Samsung ones. No difference.

Btw, I've been working (stressing) the machine for one week now and never had any problems, the system seems rock solid (until I start my memory stresser).

kind regards Philip

#11 lore raw
In reply to Robert Hancock
Quoted 12 lines · Robert Hancock
Quoted 10 lines · Alan Cox
Quoted 5 lines · Robert Hancock

I'm wondering if one does a ton of these cache-bypassing stores whether something gets hosed because of that. Not sure what that could be though. I don't imagine the chipset is involved with any of that on the Athlon 64 - either the CPU or RAM seems the most likely suspect to me

Quoted 8 lines
Quoted 3 lines

The glibc version is essentially the "perfect" copy function for the CPU. If you have any bus/memory problems or chipset bugs it will bite you.

Anyone have any suggestions on how to track this further? It seems fairly clear what circumstances are causing it, but as for figuring out what's at fault..

Digging through my glibc's source if found that if you memset arrays <120000 bytes it will use good old mov instructions to do the job. In case of arrays larger than 120000 bytes it will use movnti instructions to do the job.

Thus I refined my test code to use mov for memset regardless of the size (simply abused glibcs code a little bit)

-> No crash!

Then, changing the all the mov to movnti and my machine frags again :(

It seems that mov'ing does not kill my machine while simply using movnti does.

kind regards Philip

#12 lore raw
Quoted 30 lines · Philip Lawatsch
Quoted 12 lines · Robert Hancock
Quoted 10 lines · Alan Cox
Quoted 5 lines · Robert Hancock

I'm wondering if one does a ton of these cache-bypassing stores whether something gets hosed because of that. Not sure what that could be though. I don't imagine the chipset is involved with any of that on the Athlon 64 - either the CPU or RAM seems the most likely suspect to me

Quoted 8 lines
Quoted 3 lines

The glibc version is essentially the "perfect" copy function for the CPU. If you have any bus/memory problems or chipset bugs it will bite you.

Anyone have any suggestions on how to track this further? It seems fairly clear what circumstances are causing it, but as for figuring out what's at fault..

Digging through my glibc's source if found that if you memset arrays <120000 bytes it will use good old mov instructions to do the job. In case of arrays larger than 120000 bytes it will use movnti instructions to do the job.

Thus I refined my test code to use mov for memset regardless of the size (simply abused glibcs code a little bit)

Quoted 6 lines

-> No crash!

Then, changing the all the mov to movnti and my machine frags again :(

It seems that mov'ing does not kill my machine while simply using movnti does.

movnti also gets a higher bandwidth so that doesn't rule out too much..

#13 lore raw
In reply to Philip Lawatsch
Quoted 7 lines · Philip Lawatsch
Quoted 3 lines

Anyone have any suggestions on how to track this further? It seems fairly clear what circumstances are causing it, but as for figuring out what's at fault..

It seems that mov'ing does not kill my machine while simply using movnti does.

Forget about what I just wrote, I've been able to reproduce this in 32bit mode too although it did take a long while to happen.

And glibc in 32bit mode simply uses mov in a normal loop to write to the memory.

Looks like using mov in 64bit mode polluted my cache and crippled performance (have been running some other programs in the background) and thus perhaps didnt trigger the problem.

I'm going nuts with this.

kind regards Philip