--- Begin Message ---
- To: Jonathan Nieder <jrnieder@gmail.com>
- Subject: Re: indefinite soft lockup on rm
- From: Egon Eckert <egon.eckert@heaven-industries.com>
- Date: Mon, 12 Dec 2011 10:26:13 +0100
- Message-id: <20111212092613.GA7157@heaven-industries.com>
- In-reply-to: <20111210031058.GD6243@elie.hsd1.il.comcast.net>
- References: <20111203005736.GA8547@elie.hsd1.il.comcast.net> <20111006205828.5925.34899.reportbug@tornado.autokelly.local> <handler.644550.D644550.132287386830367.notifdone@bugs.debian.org> <20111205091426.GC27425@heaven-industries.com> <20111205100427.GA9080@elie.hsd1.il.comcast.net> <20111206220448.GB27974@heaven-industries.com> <[🔎] 20111206230034.GB5821@elie.hsd1.il.comcast.net> <[🔎] 20111209092630.GA6728@heaven-industries.com> <20111210031058.GD6243@elie.hsd1.il.comcast.net>
> Version: 3.2~rc4-1~experimental.1
>
> Egon Eckert wrote:
>
> > Actually, 3.2.0rc4 seems to run well! The big (and relevant) change since
> > the squeeze kernel is that the new has the intel_idle driver (which kicks in
> > instead of the acpi_idle):
>
> Thanks for checking. Marking accordingly.
>
> The intel_idle driver was introduced upstream in the 2.6.36 merge
> window and enabled in Debian kernels in 3.1.0-1~experimental.1.
> It would be interesting to hear how a recent kernel built with
> CONFIG_INTEL_IDLE=n behaves, to confirm that that is actually what
> fixed it.
After disabling the intel_idle driver (booting with intel_idle.max_cstate=0,
which effectively disables it) the 3.2.0rc4 kernel installs acpi_idle driver
and crashes again. The crash, however, doesn't look like the squeeze kernel
crash, so it's not certain they both relate to these C-states. It's very
similar to the 3.0.0 crash I reported on Oct 25 (in the same bug report
thread).
I also confirmed the squeeze kernel crashes with (C-states enabled in setup)
processor.max_cstate=2
but runs fine with
processor.max_cstate=1
which, interestingly, seems to avoid C-states completely--the
/sys/devices/system/cpu/cpu*/cpuidle directories vanish and idle power
consumption increases to that with C-states disabled in BIOS (the difference
is 20W with/without them, by the way).
Thanks,
--
Egon
--- End Message ---