https://bugzilla.novell.com/show_bug.cgi?id=325917#c42
Thomas Renninger changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |acpi@linux.intel.com,
| |venkatesh.pallipadi@intel.com
--- Comment #42 from Thomas Renninger 2007-10-10 03:49:45 MST ---
I also had the idea that the break comes from:
nohz/clocksource/highres patches + C3/C4
means the time keeping or timer interrupts could be broken and therefore C3/C4
is entered wrongly.
IIRC we already tried several time sources, not sure anymore, because I then
came more to the conclusion that it could have to do with ata DMA setup.
We also tried to go to C3/C4 more conservative by increasing the bm_history
bitmask -> this did also not help.
I also don't know for sure how to test whether this works appropriate.
IIRC the timer interrupts nearly (or at all?) stopped firing on IRQ0 sometimes
or with some clocksources, but the system itself still went smootly (e.g. watch
-n1 date). AFAIK this is a nohz feature, but as said I don't know well about
these details. It seems as if the system get stuck on (some) disk accesses. If
stuck, everything seem to be frozen (as far as I can tell from remote access)
irqs seem to be disabled.
Adding Venki (and some more ACPI people), at least we have a broken ICH chipset
here, maybe he already knows about this one or has an idea...
Venki: This system gets stuck on disk access with ata_piix and latest 2.6.22
kernel with nohz enabled. The system worked fine with piix and 10.2 kernel.
Tejun says there is not much ata_piix could have done wrongly with DMA setup.
As disabling C3/C4 helps we expect we have a DMA vs C-state problem here,
possibly caused through nohz patches.
Srinivasa: In comment #13 you stated that you tried nohz=disable parameter, do
you remember whether this parameter helped to make the machine fast again?
I think we also checked that: Could you double check that C3/C4 is used in
10.2, pls. "cat /proc/acpi/processor/*/power" will tell us.
--
Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.