[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Xen-users] Memory problem's 2GB mod


  • To: "Kevin van Kuik" <kkuik@xxxxxxxxxxxxxxxxxxxxxx>
  • From: "Petersson, Mats" <Mats.Petersson@xxxxxxx>
  • Date: Thu, 24 Aug 2006 13:16:51 +0200
  • Cc: xen-users@xxxxxxxxxxxxxxxxxxx
  • Delivery-date: Thu, 24 Aug 2006 04:17:44 -0700
  • List-id: Xen user discussion <xen-users.lists.xensource.com>
  • Thread-index: AcbG/1b5Iw7sJe+cRU2ymv7lj11z+gACT266ABa6wAAAAEjf0AAB+FoQ
  • Thread-topic: [Xen-users] Memory problem's 2GB mod

 


From: Kevin van Kuik [mailto:kkuik@xxxxxxxxxxxxxxxxxxxxxx]
Sent: 24 August 2006 11:15
To: Petersson, Mats
Subject: RE: [Xen-users] Memory problem's 2GB mod

If you put 4GB memories (2 X 2GB) xen-3.0.2-2 crashed to and if when we put 2Gb (2 X 1 GB) it works fine. So 16Gb is not the problem.!
We have tried new memory and we have the same crash.
 
So if I put 2gb memories it's crashing with xen-3.0.2-2, but it's work fine with redhat 4.1 kernel, xen-3.0.1 So or only explanation is that xen-3.0.2-2 have a problem with 2 GB memories 
 
I disagree. First and foremost because Xen itself has not knowledge of what combination of memories (type of module or otherwise) you have.
 
However, it may EXPOSE some problems that previously didn't occur, due to some other scenario... For example, the new version of Xen may have a different memory access pattern that causes a failure to happen because there's some particular sequence of memory accesses, which doesn't happen in the same way with Xen 3.0.1 or plain RedHat. I've worked with hardware long enough to know that "strange things happen" when you mix different hardware and software.
 
One thing to consider is to upgrade the BIOS on the machine, just in case there's some sort of bug in the memory configuration code in the BIOS. Here's how it works: The BIOS recognizes your memory chips (from the SPD), and then sets up the timing parameters, memory sizing parameters, number of banks and some other miscellaneous parameters. Once this setup has been done, nothing touches the memory configuration registers ever again. Xen has absolutely NO CODE to touch these registers, so how can Xen itself fail with one memory configuration when another similar one works - well, because the ORIGINAL setup didn't work right. But it's not Xen's fault.
 
Please feel free to pursue Xen as the culprit if you wish, but I can 100% guarantee you that it's going to end with you fiinding that it's the memory and/or BIOS and/or motherboard that is ACTUALLY causing the problem. I know this, because not only have I worked with Xen for the last year and a bit, but I've also written memory sizing/configuration algorithms for startup of the processor (like what you find in the BIOS).
 
Try running Memtest86 on the 2GB memory sticks and see if that shows anything. That is NOT GUARANTEED to show a problem, but it's a good starting point for figuring out that the memory is good or bad.
 
Just to confirm that software can influence the memory errors, I ran memtest86 as a HVM instance under Xen - so running the virtual machine doing memory testing. I got memory errors. I then ran memtest86 on the machine itself, no errors. Changing to a different memory module (from a different manufacturer), fixed the problem. It's still not Xen's fault that the memory fails, it just changes the access pattern by locating things in a different way, and in the case of HVM models adding memory mapping to the not-usually-memory-mapped test-case, so there will be a different memory access pattern when running the HVM guest compared to running the bare-metal version.
 
--
Mats
Ragard Kevin

 
 -----Oorspronkelijk bericht-----
Van: Petersson, Mats [mailto:Mats.Petersson@xxxxxxx]
Verzonden: Thursday, August 24, 2006 12:00 PM
Aan: Kevin van Kuik; xen-users@xxxxxxxxxxxxxxxxxxx
Onderwerp: RE: [Xen-users] Memory problem's 2GB mod

Xen has absolutely no idea what type of memory modules you're using, so obviously this is not DIRECTLY causing the problem.
 
Have you tried runnign a memory test on the 16GB memories? It may just be that a slightly different setup of memory (caused by arbitrary changes in Xen 3.0.1->3.0.2) causes you to hit some bad spot in the memory.
 
To utterly prove that it's not a "module" issue, you could also try running 4 * 2GB instead of 8 * 1GB, which should have the same effect.
 
The only place where the actual configuration of the memory is known is within the BIOS that sets up the memory configuration early on in the boot process (before you see any text on your monitor). Beyond that, no software will touch anything that is related to what type or model of memory you're using [except for some utilities that are able to READ the memory configuration registers to tell you what type of memories you're using, etc, etc].
 
--
Mats


From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Kevin van Kuik
Sent: 24 August 2006 00:15
To: xen-users@xxxxxxxxxxxxxxxxxxx
Subject: [Xen-users] Memory problem's 2GB mod

I have a HP ml370 with xen-3.0.1 and redhat 4.1 with 16 GB memory (8 x 2GB). I want to upgrade to xen-3.0.2-2 because the server crashed on the scsi driver.
 
If I boot the server with xen-3.0.2-2 it crashed on the boot. But if I put 8 x 1GB memory the server boot fine.
 
So xen-3.0.2-2 can't work with 2GB memory modules. Is there a compile option to solve this problem?
 
The memory is OK because it's work fine with redhat 4.1 (without xen) and xen-3.0.1
 
regards Kevin
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.