On Wed, Jan 26, 2005 at 10:07:35AM -0500, David Wood wrote:
> Has anyone else noticed instability when using cpufreq modules?
>
> I have a pretty stable system, but I get OOPSes within a few hours or days
> of using cpufreq_ondemand, and I've had a few crashes I was pretty sure
> were related to cpufreq_userspace/powernowd.
I'd been ignoring my Debian inbox for a long time until today...
I have an Asus K8V (basic) with an Athlon64 3200+ (newcastle core) 1.5GB of
RAM, two IDE disks, two SATA disks, and an ATI AIW Radeon 7200 (but I don't
use the TV in/out features). I run x86 2.6.12.6. I'll eventually switch to
AMD64 software when I know my hardware is stable with x86, so I can usefully
make bug reports on crashing software...
BTW, the K8V is a nice piece of hardware; the AD1980 sound hardware
supports mixing PCM streams in hardware (or at least the driver does?), so I
can have xmms, xine, and whatever other program all not interfering with
each other. (except when something is doing 4 channel output). I couldn't
decide between an Abit (I think) with a K8T800Pro chipset and my Asus with
just K8T800, but I eventually chose the Asus because it had Analog Devices
sound instead of Realtek. I was pleasantly surprised that the sound really
was good on it, esp. with the multiple opens of the sound dev :)
I've found that my machine is a lot less stable when running at lower than
max speed. Not just stuff crashing, but memtest (from sysutils, or
memtester; just mlock()s some memory to test, not like memtest86+). memtest
finds errors when the CPU is slowed down. There might be other correlated
factors, like disk access. To change speed, I've just used cpufreq-set -u
2000MHz (or 1800MHz, or 1000MHz). Max speed is 2200MHz. (Newcastle core:
from dmidecode: ID: C0 0F 00 00 FF FB 8B 07
Signature: Extended Family 0, Model C, Stepping 0
)
Unfortunately, the machine isn't perfectly stable even at max speed. It
never crashed before I upgraded the BIOS from 1.04 or something to 1.07,
which was needed for cpufreq to work. Even when running at 2.2GHz (full
speed) with only one stick of RAM (1024MB OCZ), it sometimes shows a cluster
of memory errors in memtest. It doesn't seem significantly different from
with both sticks of RAM, the other being a 512MB Infineon, IIRC. All DDR400.
Run 129 completed in 357 seconds (0 tests showed errors).
Run 130:
Test 1: Stuck Address: Testing...Passed.
Test 2: Random value: Setting...Testing...
FAILURE: 0x7ffeebc8 != 0x7efeebc8 at offset 0x01ca67f0.
Skipping to next test...
Test 3: XOR comparison: Setting...Testing...
FAILURE: 0x42ff4c7a != 0x43ff4c7a at offset 0x01ca67f0.
Skipping to next test...
Test 4: SUB comparison: Setting...Testing...
FAILURE: 0x2707802e != 0x2807802e at offset 0x01ca67f0.
Skipping to next test...
Test 5: MUL comparison: Setting...Testing...
FAILURE: 0x73c9b7ae != 0xb4c9b7ae at offset 0x01ca67f0.
Skipping to next test...
Test 6: DIV comparison: Setting...Testing...
FAILURE: 0x00000000 != 0x00000001 at offset 0x01ca67f0.
Skipping to next test...
Test 7: OR comparison: Setting...Testing...
FAILURE: 0xb2dd75dc != 0xb2dd75dd at offset 0x01ca67f0.
Skipping to next test...
Test 8: AND comparison: Setting...Testing...Passed.
Test 9: Sequential Increment: Setting...Testing...Passed.
Test 10: Solid Bits: Testing...Passed.
BIOS on all auto settings.
no other runs showed errors (140 runs)
(Running at lower CPU speeds, errors were much more frequent).
Interesting that all the errors are clustered in time and space at one
memory location... As I said, the software running is Debian i386 sid with
Linux 2.6.12.6, compiled with gcc 4.0.2 20050816 (from sid).
Does anyone have any ideas? I hate hardware I can't trust! What's the
point of digital logic if it makes mistakes!
So does anyone have any experience or advice?
--
#define X(x,y) x##y
Peter Cordes ; e-mail: X(peter@cor , des.ca)
"The gods confound the man who first found out how to distinguish the hours!
Confound him, too, who in this place set up a sundial, to cut and hack
my day so wretchedly into small pieces!" -- Plautus, 200 BC
Attachment:
signature.asc
Description: Digital signature