On Wednesday, February 20, 2008 10:37 am Linus Torvalds wrote:
quoted text
> This *sounds* like some part of the suspend-to-disk sequence is doing
> something stupid like trying to access the screen after it has been turned
> off, which doesn't surprise me at all. My oft-stated opinion has been that
> suspend-to-disk isn't a suspend at all, and should never have been
> confused with "suspending" anything.
>
> It's "snapshot-and-restore", and my opinion is that:
>
>  - it should *never* call "suspend()"/"resume()" at all (that should be
>    reserved purely for suspend-to-RAM and has real power management
>    issues!)
>
>  - it should have a totally separate "halt/unhalt/restore" thing
>    that has nothing what-so-ever to do with power management, and is
>    purely about stopping the hardware for things like USB and network
>    cards (which otherwise do things like scan their command lists
>    asynchronously) and making sure that the driver state is consistent
>    with that stopped hw state.
>
>  - the people who confuse snapshot/restore with suspend/resume are
>    horrible people that cause problems exactly because driver people then
>    get those things mixed up, and something like the video suspend/resume
>    should probably never have impacted suspend-to-disk in the first place!

Totally agreed.  I remember when I started getting hibernation bug reports 
against this new code and boggling at how hibernate was actually done.  The 
driver actually gets its ->suspend routine called twice with two different 
pm_message_t values.  We tried to do different stuff depending on the 
pm_message_t (like only putting the device in D3hot if PM_EVENT_SUSPEND), but 
it appears we're not doing enough...

quoted text
> So there seems to be two (probably largely independent) problems:
>
>  - the hang at shutdown that requires you to press-and-hold the power
>    button to actually cut power.
>
>    At a guess: putting the VGA device into D3hot makes the ACPI code that
>    actually does the shutoff unhappy. Probably because it wants to access
>    the device, and ends up not ever getting the replies it wants, since
>    the hardware has been turned off.

Sounds like a good theory... now if we could just use set_power_state in the 
suspend case only.  That's what the latest code *tries* to do...

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