Skip to contentSkip to conversation

[2.6] Missing L2-cache after warm boot

DISCUSSION Jochen Hein 4 participants 9 messages Dec 1, 2003 – Dec 3, 2003

lore
Share & tools
lore
#0 lore raw

I'm running 2.6.0-test11 on an older Thinkpad 390E, When booting into 2.6.0-test11 after running Windows2000 I get:

Dec  1 14:51:56 gswi1164 kernel: Initializing CPU#0
Dec  1 14:51:56 gswi1164 kernel: PID hash table entries: 1024 (order 10: 8192 bytes)
Dec  1 14:51:56 gswi1164 kernel: Detected 298.602 MHz processor.
Dec  1 14:51:56 gswi1164 kernel: Console: colour dummy device 80x25
Dec  1 14:51:56 gswi1164 kernel: Memory: 190848k/196544k available (2008k kernel code, 5060k reserved, 762k data, 148k init, 0k
highmem)
Dec  1 14:51:56 gswi1164 kernel: Calibrating delay loop... 587.77 BogoMIPS
Dec  1 14:51:56 gswi1164 kernel: Security Scaffold v1.0.0 initialized
Dec  1 14:51:56 gswi1164 kernel: Capability LSM initialized
Dec  1 14:51:56 gswi1164 kernel: Dentry cache hash table entries: 32768 (order: 5, 131072 bytes)
Dec  1 14:51:56 gswi1164 kernel: Inode-cache hash table entries: 16384 (order: 4, 65536 bytes)
Dec  1 14:51:56 gswi1164 kernel: Mount-cache hash table entries: 512 (order: 0, 4096 bytes)
Dec  1 14:51:56 gswi1164 kernel: CPU: L1 I cache: 16K, L1 D cache: 16K
Dec  1 14:51:56 gswi1164 kernel: Intel machine check architecture supported.
Dec  1 14:51:56 gswi1164 kernel: Intel machine check reporting enabled on CPU#0.
Dec  1 14:51:56 gswi1164 kernel: CPU: Intel Mobile Pentium II stepping 0a
Dec  1 14:51:56 gswi1164 kernel: Enabling fast FPU save and restore... done.
Dec  1 14:51:56 gswi1164 kernel: Checking 'hlt' instruction... OK.

When booting cold the boot messages are:

Dec  1 14:54:57 gswi1164 kernel: Initializing CPU#0
Dec  1 14:54:57 gswi1164 kernel: PID hash table entries: 1024 (order 10: 8192 bytes)
Dec  1 14:54:57 gswi1164 kernel: Detected 298.598 MHz processor.
Dec  1 14:54:57 gswi1164 kernel: Console: colour dummy device 80x25
Dec  1 14:54:57 gswi1164 kernel: Memory: 190848k/196544k available (2008k kernel code, 5060k reserved, 762k data, 148k init, 0k
highmem)
Dec  1 14:54:57 gswi1164 kernel: Calibrating delay loop... 587.77 BogoMIPS
Dec  1 14:54:57 gswi1164 kernel: Security Scaffold v1.0.0 initialized
Dec  1 14:54:57 gswi1164 kernel: Capability LSM initialized
Dec  1 14:54:57 gswi1164 kernel: Dentry cache hash table entries: 32768 (order: 5, 131072 bytes)
Dec  1 14:54:57 gswi1164 kernel: Inode-cache hash table entries: 16384 (order: 4, 65536 bytes)
Dec  1 14:54:57 gswi1164 kernel: Mount-cache hash table entries: 512 (order: 0, 4096 bytes)
Dec  1 14:54:57 gswi1164 kernel: CPU: L1 I cache: 16K, L1 D cache: 16K
Dec  1 14:54:57 gswi1164 kernel: CPU: L2 cache: 256K
Dec  1 14:54:57 gswi1164 kernel: Intel machine check architecture supported.
Dec  1 14:54:57 gswi1164 kernel: Intel machine check reporting enabled on CPU#0.
Dec  1 14:54:57 gswi1164 kernel: CPU: Intel Mobile Pentium II stepping 0a
Dec  1 14:54:57 gswi1164 kernel: Enabling fast FPU save and restore... done.
Dec  1 14:54:57 gswi1164 kernel: Checking 'hlt' instruction... OK.

/proc/cpuinfo contains (after warm boot):
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 6
model name : Mobile Pentium II
stepping : 10
cpu MHz : 298.598
cache size : 256 KB
fdiv_bug : no
hlt_bug : no
f00f_bug : no
coma_bug : no
fpu : yes
fpu_exception : yes
cpuid level : 2
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov pat pse36 mmx fxsr
bogomips : 587.77

Is there any way to find out, why the second level cache isn't detected after a warm boot?

Jochen

Signature · 1 line
#include <~/.signature>: permission denied
#1 lore raw

On Mon, 01 Dec 2003 15:04:22 +0100 Jochen Hein <[email protected]> wrote:

| 
| I'm running 2.6.0-test11 on an older Thinkpad 390E,
| When booting into 2.6.0-test11 after running Windows2000 I get:
| 
...
| Dec  1 14:51:56 gswi1164 kernel: Inode-cache hash table entries: 16384 (order: 4, 65536 bytes)
| Dec  1 14:51:56 gswi1164 kernel: Mount-cache hash table entries: 512 (order: 0, 4096 bytes)
| Dec  1 14:51:56 gswi1164 kernel: CPU: L1 I cache: 16K, L1 D cache: 16K
...
| 
| When booting cold the boot messages are:
| 
...
| Dec  1 14:54:57 gswi1164 kernel: Inode-cache hash table entries: 16384 (order: 4, 65536 bytes)
| Dec  1 14:54:57 gswi1164 kernel: Mount-cache hash table entries: 512 (order: 0, 4096 bytes)
| Dec  1 14:54:57 gswi1164 kernel: CPU: L1 I cache: 16K, L1 D cache: 16K
| Dec  1 14:54:57 gswi1164 kernel: CPU: L2 cache: 256K

Can you apply this patch so that we can see what cache/TLB descriptors are found during the boot?

--
~Randy

description: dump cache/TLB descriptors while trying to identify them
product_versions: Linux 2.6.0-test11

diffstat:=

 arch/i386/kernel/cpu/intel.c |    1 +
 1 files changed, 1 insertion(+)
arch/i386/kernel/cpu/intel.c renamed from arch/i386/kernel/cpu/intel.c~caches +1 -0
diff -Naurp ./arch/i386/kernel/cpu/intel.c~caches ./arch/i386/kernel/cpu/intel.c
@@ -187,6 +187,7 @@ static void __init init_intel(struct cpu
 				unsigned char des = dp[j];
 				unsigned char k = 0;
 
+				printk("CPU: searching for des = 0x%x\n", des);
 				/* look up this descriptor in the table */
 				while (cache_table[k].descriptor != 0)
 				{
#2 lore raw
Quoted 2 lines · Randy.Dunlap

Can you apply this patch so that we can see what cache/TLB descriptors are found during the boot?

The last reboots (cold, warm from Linux and Win2k) all had Level2 cache enabled. I'll keep running with your patch, so we have some traces when it happens again.

What disturbs me is that the system locked hard a view times now. I've now restricted memory with mem=190m (has 196m installed). I'll keep investigating.

Jochen

Signature · 1 line
#include <~/.signature>: permission denied
#3 lore raw
Quoted 2 lines · Jochen Hein

I've now restricted memory with mem=190m (has 196m installed). I'll keep investigating.

What good does cutting 6MB off from use?

#4 lore raw
Quoted 5 lines · Mike Fedyk
Quoted 2 lines · Jochen Hein

I've now restricted memory with mem=190m (has 196m installed). I'll keep investigating.

What good does cutting 6MB off from use?

My old TP 600 had 96m installed, but due to BIOS-usage (mwave IIRC), only 95m could be used for the os. This is just a shot in the blue now, but it didn't help anway.

Jochen

Signature · 1 line
#include <~/.signature>: permission denied
#5 lore raw
In reply to Jochen Hein
Quoted 2 lines · Jochen Hein

I'm running 2.6.0-test11 on an older Thinkpad 390E, When booting into 2.6.0-test11 after running Windows2000 I get:

Do any of the previous test releases show this problem? How are you booting Linux? Full warm boot via the BIOS or some loadlin kind of boot? If it's via the BIOS then does that show the L2 cache as being present?

Quoted 1 line

When booting cold the boot messages are:

Presumably from switch on.

Quoted 9 lines

/proc/cpuinfo contains (after warm boot):
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 6
model name : Mobile Pentium II
stepping : 10
cpu MHz : 298.598
cache size : 256 KB

And shows as 0 after warm boot?

Quoted 11 lines

fdiv_bug : no
hlt_bug : no
f00f_bug : no
coma_bug : no
fpu : yes
fpu_exception : yes
cpuid level : 2
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov
pat pse36 mmx fxsr bogomips : 587.77

Quoted 2 lines

Is there any way to find out, why the second level cache isn't detected after a warm boot?

My immediate thought was a BIOS problem. IBM's web site doesn't say any BIOS updates fix L2 cache related problems, but then it doesn't seem to use technical descriptions like that. It says the latest BIOS is 1.55 - R01_C9.

http://www-1.ibm.com/support/docview.wss?rs=0&uid=psg1MIGR-4F3VKB&loc=en_US

Or maybe it's possible that something in MS Windows 2000 is turning off the L2 cache and it isn't getting reactivated by the warm boot? What happens when you do a cold boot to Linux then reboot from there?

Signature · 1 line
Ian.
#6 lore raw
Quoted 5 lines · Ian Hastie
Quoted 2 lines · Jochen Hein

I'm running 2.6.0-test11 on an older Thinkpad 390E, When booting into 2.6.0-test11 after running Windows2000 I get:

Do any of the previous test releases show this problem?

-test11 is the first release running on that machine. My older TP600 destroyed its WIndows 95 so I got a replacement. Compiling a kernel takes an hour or two, so it is not much fun trying different kernels.

Quoted 2 lines

How are you booting
Linux? Full warm boot via the BIOS or some loadlin kind of boot?

via BIOS, lilo in MBR

Quoted 2 lines

If it's
via the BIOS then does that show the L2 cache as being present?

The BIOS doesn't tell anything, the setup doesn't have a "cache enable" or "turbo" entry.

Quoted 3 lines
Quoted 1 line

When booting cold the boot messages are:

Presumably from switch on.

Yes.

Quoted 11 lines
Quoted 9 lines

/proc/cpuinfo contains (after warm boot):
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 6
model name : Mobile Pentium II
stepping : 10
cpu MHz : 298.598
cache size : 256 KB

And shows as 0 after warm boot?

Hm, can't say for sure, because it didn't happen again.

Quoted 5 lines

My immediate thought was a BIOS problem. IBM's web site doesn't say any BIOS updates fix L2 cache related problems, but then it doesn't seem to use technical descriptions like that. It says the latest BIOS is 1.55 - R01_C9.

http://www-1.ibm.com/support/docview.wss?rs=0&uid=psg1MIGR-4F3VKB&loc=en_US

I'll see what BIOS I have - it is the latest. Thanks for the hint anyway.

Quoted 2 lines

Or maybe it's possible that something in MS Windows 2000 is turning off the L2 cache and it isn't getting reactivated by the warm boot?

Is there any way to see what Windows does here? I only found a manual enable of the L2cache when using older processors.

Quoted 2 lines

What happens when
you do a cold boot to Linux then reboot from there?

That is fine.

For now the system seems to be fine, even when starting from Win2k via BIOS reboot. Hmpf.

Jochen

Signature · 1 line
#include <~/.signature>: permission denied
#7 lore raw
Quoted 10 lines · Jochen Hein
Quoted 5 lines · Ian Hastie
Quoted 2 lines · Jochen Hein

I'm running 2.6.0-test11 on an older Thinkpad 390E, When booting into 2.6.0-test11 after running Windows2000 I get:

Do any of the previous test releases show this problem?

-test11 is the first release running on that machine. My older TP600 destroyed its WIndows 95 so I got a replacement. Compiling a kernel takes an hour or two, so it is not much fun trying different kernels.

No faster machine to compile on? Still think this isn't a kernel problem anyway.

Quoted 5 lines
Quoted 2 lines

If it's
via the BIOS then does that show the L2 cache as being present?

The BIOS doesn't tell anything, the setup doesn't have a "cache enable" or "turbo" entry.

I was just hoping it would have a configuration screen shown after POST. I had expected that it would show any active caches. Doesn't seem so likely now though. *8/

Quoted 9 lines
Quoted 7 lines

My immediate thought was a BIOS problem. IBM's web site doesn't say any BIOS updates fix L2 cache related problems, but then it doesn't seem to use technical descriptions like that. It says the latest BIOS is 1.55 - R01_C9.

http://www-1.ibm.com/support/docview.wss?rs=0&uid=psg1MIGR-4F3VKB&loc=en_
US

I'll see what BIOS I have - it is the latest. Thanks for the hint anyway.

OK. *8)

Quoted 5 lines
Quoted 2 lines

Or maybe it's possible that something in MS Windows 2000 is turning off the L2 cache and it isn't getting reactivated by the warm boot?

Is there any way to see what Windows does here? I only found a manual enable of the L2cache when using older processors.

Who knows what a Windows programme or driver might do in order to "fix" some problem? I suppose it's possible that a programme like SiSoft Sandra might be able to tell if the L2 cache is active or not. There is almost bound to be some programme that can read the appropriate configuration bits.

Quoted 7 lines
Quoted 2 lines

What happens when
you do a cold boot to Linux then reboot from there?

That is fine.

For now the system seems to be fine, even when starting from Win2k via BIOS reboot. Hmpf.

Annoying. Can you think of anything you might have done with Windows 2000, eg programme run or device used, that you haven't done when testing?

Signature · 1 line
Ian.
#8 lore raw
Quoted 1 line · Ian Hastie

No faster machine to compile on?

An Mini-ITX with 550MHz isn't that much faster (but very quiet).

Quoted 1 line

Still think this isn't a kernel problem anyway.

That's my current guess too.

Quoted 5 lines
Quoted 2 lines

For now the system seems to be fine, even when starting from Win2k via BIOS reboot. Hmpf.

Annoying. Can you think of anything you might have done with Windows 2000, eg programme run or device used, that you haven't done when testing?

I won't have time this week, maybe next. What concerns my is that -test11 locks the machine hard after an hour or so (nmi-watchdog doesn't say anything), but 2.4 seems to be stable. Since I need a stable machine next week I'll stay with 2.4 for now and come back later.

Jochen

Signature · 1 line
#include <~/.signature>: permission denied