Skip to contentSkip to conversation

Lost Ticks on x86_64

DISCUSSION Andi Kleen 5 participants 12 messages Aug 7, 2005 – Aug 8, 2005 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): 5348b8ba050806204453392f7f@mail.gmail.com.suse.lists.linux.kernel

Andi Kleen

author
#1 lore raw
Quoted 7 lines · Erick Turnquist

Hi, I'm running an Athlon64 X2 4400+ (a dual core model) with an nVidia GeForce 6800 Ultra on a Gigabyte GA-K8NXP-SLI motherboard and getting nasty messages like these in my dmesg:

warning: many lost ticks.
Your time source seems to be instable or some driver is hogging interupts rip default_idle+0x20/0x30

It's most likely bad SMM code in the BIOS that blocks the CPU too long and is triggered in idle. You can verify that by using idle=poll (not recommended for production, just for testing) and see if it goes away.

No way to fix this, but you can work around it with very new kernels by compiling with a lower HZ than 1000.

-Andi

#2 lore raw
Quoted 2 lines

It's most likely bad SMM code in the BIOS that blocks the CPU too long and is triggered in idle.

Might a BIOS flash help, or is this something that's there to stay?

Quoted 2 lines

No way to fix this, but you can work around it with very new kernels by compiling with a lower HZ than 1000.

Actually, it was already running at 250Hz. I must have turned it down a while ago while I was trying to find the cause of the problem.

#3 lore raw
In reply to Andi Kleen
Quoted 10 lines · Andi Kleen
Quoted 7 lines · Erick Turnquist

Hi, I'm running an Athlon64 X2 4400+ (a dual core model) with an nVidia GeForce 6800 Ultra on a Gigabyte GA-K8NXP-SLI motherboard and getting nasty messages like these in my dmesg:

warning: many lost ticks.
Your time source seems to be instable or some driver is hogging interupts rip default_idle+0x20/0x30

Quoted 6 lines

It's most likely bad SMM code in the BIOS that blocks the CPU too long and is triggered in idle. You can verify that by using idle=poll (not recommended for production, just for testing) and see if it goes away.

No way to fix this, but you can work around it with very new kernels by compiling with a lower HZ than 1000.

Some BIOSes do not lock SMM, and you *could* turn it off at the chipset level.

#4 lore raw
Quoted 2 lines

Some BIOSes do not lock SMM, and you *could* turn it off at the chipset level.

I don't see anything about SMM in my BIOS configuration even with the advanced options enabled... Turning it off at the chipset level sounds like a hardware hack - is it?

The gettimeofday patch for 2.6.13-rc3 won't apply. My source tree has no ntp.h or ntp.c, and I get a malformed patch error at line 560. I'm using the patch Google found for me: http://lwn.net/Articles/143953/
on a 2.6.13-rc3 source tree.

#5 lore raw
Quoted 6 lines · Erick Turnquist
Quoted 2 lines

Some BIOSes do not lock SMM, and you *could* turn it off at the chipset level.

I don't see anything about SMM in my BIOS configuration even with the advanced options enabled... Turning it off at the chipset level sounds like a hardware hack - is it?

No, it's usually just a PCI register you can change. Depends on your chipset, though.

author
#6 lore raw
In reply to Tim Hockin
Quoted 2 lines

Some BIOSes do not lock SMM, and you *could* turn it off at the chipset level.

Doing so would be wasteful though. Both AMD and Intel CPUs need SMM code for the deeper C* sleep states.

-Andi

#7 lore raw
Quoted 5 lines · Andi Kleen
Quoted 2 lines

Some BIOSes do not lock SMM, and you *could* turn it off at the chipset level.

Doing so would be wasteful though. Both AMD and Intel CPUs need SMM code for the deeper C* sleep states.

Really? I'm not too familiar with the deeper C states - what role does SMM play?

author
#8 lore raw
Quoted 9 lines · Tim Hockin
Quoted 5 lines · Andi Kleen
Quoted 2 lines

Some BIOSes do not lock SMM, and you *could* turn it off at the chipset level.

Doing so would be wasteful though. Both AMD and Intel CPUs need SMM code for the deeper C* sleep states.

Really? I'm not too familiar with the deeper C states - what role does SMM play?

It does all the hard work of powering down and up the CPU in C2 and C3. On AMD K8 even C1 (=HLT) executes SMM code depending on the BIOS setup.

-Andi

#9 lore raw
In reply to Andi Kleen
Quoted 6 lines · Andi Kleen
Quoted 2 lines

Some BIOSes do not lock SMM, and you *could* turn it off at the chipset level.

Doing so would be wasteful though. Both AMD and Intel CPUs need SMM code for the deeper C* sleep states.

Wouldn't it be useful for !CONFIG_PM? Many multimedia users run this way because they want their machines running full speed all the time (they need the horsepower, plus frequency scaling interferes with the TSC based timing used by JACK ) and because ACPI has a history of causing nasty latency blips.

Lee

#10 lore raw
In reply to Andi Kleen
Quoted 6 lines · Andi Kleen
Quoted 4 lines · Erick Turnquist

Hi, I'm running an Athlon64 X2 4400+ (a dual core model) with an nVidia GeForce 6800 Ultra on a Gigabyte GA-K8NXP-SLI motherboard and getting nasty messages like these in my dmesg:

Quoted 8 lines
Quoted 3 lines

warning: many lost ticks.
Your time source seems to be instable or some driver is hogging interupts rip default_idle+0x20/0x30

It's most likely bad SMM code in the BIOS that blocks the CPU too long and is triggered in idle. You can verify that by using idle=poll (not recommended for production, just for testing) and see if it goes away.

WTF, since when do *desktops* use SMM? Are you telling me that we have to worry about these stupid ACPI/SMM hardware bugs on the desktop too?

Lee

#11 lore raw
Quoted 7 lines · Lee Revell
Quoted 4 lines

It's most likely bad SMM code in the BIOS that blocks the CPU too long and is triggered in idle. You can verify that by using idle=poll (not recommended for production, just for testing) and see if it goes away.

WTF, since when do *desktops* use SMM? Are you telling me that we have to worry about these stupid ACPI/SMM hardware bugs on the desktop too?

It's a general platform thing and has been around for ages now... Intel's ICH* definitely use it for example and those are on a lot of desktop systems. For example, turning on "Legacy USB support" will generate periodic SMIs.

#12 lore raw
In reply to Lee Revell
Quoted 7 lines · Lee Revell
Quoted 4 lines

It's most likely bad SMM code in the BIOS that blocks the CPU too long and is triggered in idle. You can verify that by using idle=poll (not recommended for production, just for testing) and see if it goes away.

WTF, since when do *desktops* use SMM? Are you telling me that we have to worry about these stupid ACPI/SMM hardware bugs on the desktop too?

SMM is how BIOSes do legacy support (which stops at OS-handover). It's also how some BIOSes do ECC reporting and logging.

We just do pci tweaks to turn it off in the OS.