Re: suspending to disk on FC6 not working

[Dave Jones <davej@redhat.com>]

On Sat, Sep 30, 2006 at 10:07:16PM -0700, Pallipadi, Venkatesh wrote:

 > >I had not noticed it was not loaded. This maybe a bigger problem. With
 > >the latest FC6 kernel this module gets loaded and thus suspend fails. I
 > >rebuilt the same kernel but plopped this patch in. The module does not
 > >want to load:
 > >
 > ># /sbin/modprobe acpi-cpufreq
 > >FATAL: Error inserting acpi_cpufreq
 > >(/lib/modules/2.6.18-1.2708.cpufreq/kernel/arch/i386/kernel/cpu
 > >/cpufreq/acpi-cpufreq.ko): No such device
 > >
 > >This is why suspend works. But now this module won't load.
 > >
 > 
 > The module was not working on your laptop even when suspend was failing.
 > That is what the absence of /sys.../cpufreq says. If this module was
 > working for you earlier and stopped working recently or even otherwise
 > both this and speedstep-centrino does not work on your platform, that is
 > a separate issue. Was this module working and cpufreq supported on your
 > system with any earlier kernels?

One problem acpi-cpufreq faces when built modular compared to the
other cpufreq drivers is that there's no way of knowing before modprobe'ing
whether or not it's going to do anything or not, so userspace never knows
if it's really 'safe' to load this module.

In Fedora, we modprobe the acpi-cpufreq module if all other scaling
drivers have failed their init routines. Ie, it's treated as a fallback
"Well, nothing else worked, lets try this, and if this don't work out, bail..."
When we added this userspace 'fallback', we started getting a bunch of
reports like the above, from users who had CPUs that couldn't do any scaling
at all. The STICKY tag was added to the driver so that it would stop
making noise for those users.

Looking at the problem differently, we *could* add back the sticky tag,
and add some explicit checks for !data in the suspend/resume paths
to abort cleanly. It's a bit of a hack though.

hmm?

	Dave