Skip to contentSkip to conversation

2.6.27: 0 MHz CPU and wrong system time on AMD Geode system

DISCUSSION Antipov Dmitry 4 participants 4 messages Oct 17, 2008 – Oct 20, 2008 x86

lore
Share & tools
lore
#0 lore raw
Quoted 2 lines · Jordan Crouse

Could you try a 2.6.26 kernel? That will confirm that this is a new addition to 2.6.27.

Both 2.6.26 and 2.6.26.6 looks good (with CONFIG_GEODE_MFGPT_TIMER=y).

Dmitry

#1 lore raw
Quoted 6 lines · Antipov Dmitry
Quoted 2 lines · Jordan Crouse

Could you try a 2.6.26 kernel? That will confirm that this is a new addition to 2.6.27.

Both 2.6.26 and 2.6.26.6 looks good (with CONFIG_GEODE_MFGPT_TIMER=y).

Can you try the attached debug patch, it puts a printk message and will let us know what exaclty is failing during the calibration process.
There were some major changes in the 2.6.27 cycle related to tsc calibration. Apart from the unification for 32 and 64bit, some checks were added in the code to make the calibration process more robust, maybe this specific machine is more prone to SMI's than other, as a result the checks are a little to harsh for this one.

Along with the dmesg with this patch, could you also let us know how much is the frequency that the (2.6.26) kernel calibrates and what the actual freuqency is on that box ?

Thanks,
Alok

Quoted 8 lines

Dmitry
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

arch/x86/kernel/tsc.c +2 -0
Index: linux-2.6/arch/x86/kernel/tsc.c
===================================================================
@@ -163,6 +163,8 @@
 		pitcnt++;
 	}
 
+	printk("DEBUG: pitcnt %d, tscmin %lu, tscmax %lu\n",
+				pitcnt, tscmin, tscmax);
 	/*
 	 * Sanity checks:
 	 *
#2 lore raw
Quoted 17 lines · Alok Kataria
Quoted 6 lines · Antipov Dmitry
Quoted 2 lines · Jordan Crouse

Could you try a 2.6.26 kernel? That will confirm that this is a new addition to 2.6.27.

Both 2.6.26 and 2.6.26.6 looks good (with CONFIG_GEODE_MFGPT_TIMER=y).

Can you try the attached debug patch, it puts a printk message and will let us know what exaclty is failing during the calibration process.
There were some major changes in the 2.6.27 cycle related to tsc calibration. Apart from the unification for 32 and 64bit, some checks were added in the code to make the calibration process more robust, maybe this specific machine is more prone to SMI's than other, as a result the checks are a little to harsh for this one.

Quoted 3 lines

Along with the dmesg with this patch, could you also let us know how much is the frequency that the (2.6.26) kernel calibrates and what the actual freuqency is on that box ?

I just booted with 2.6.27.1 with high res timers on, and I can't recreate on a generic Geode box. The only difference is that my system doesn't have any MFGPT timers available, but that shouldn't matter if you have CONFIG_GEODE_MFGPT_TIMER turned off.

Jordan

Signature · 3 lines
Jordan Crouse
Systems Software Development Engineer 
Advanced Micro Devices, Inc.
#3 lore raw
In reply to Alok Kataria
Quoted 8 lines · Alok Kataria

Can you try the attached debug patch, it puts a printk message and will let us know what exaclty is failing during the calibration process.
There were some major changes in the 2.6.27 cycle related to tsc calibration. Apart from the unification for 32 and 64bit, some checks were added in the code to make the calibration process more robust, maybe this specific machine is more prone to SMI's than other, as a result the checks are a little to harsh for this one.

I recently had to disable the timer generating an SMI on our platform, because it would cause FIFO overflows on serial ports and other infrequent problems. SMM/SMI is just a pain in the rear.

I haven't tried 2.6.27 yet, but have not had issues with 2.6.26 so far (other than it seemingly breaking the SC1200's ide controller with certain compact flash cards which I am still trying to track down, but that isn't a Geode LX problem, that's a Geode SC1200 problem).

Signature · 1 line
Len Sorensen