Bugzilla – Bug 128648
APM sleep/auto-wake after 3 mins ...
Last modified: 2005-12-23 11:11:51 UTC
So - APM works wonderfully with NLD on the same machine - so, there is ~clearly no serious H/W problem here that can't be worked around: HW is IBM Thinkpad R40. I enable APM & turn ACPI off on boot; ACPI just doesn't cut it wrt. sleep, suspend-to-disk, special-key handling etc. Using Fn-F4 (sleep/suspend-to-ram) I get a wonderful laptop/suspend experience - both suspend & resume work perfectly. *Unfortunately* after 3 minutes suspended - the laptop automagically wakes up & starts hammering the disk, which is not ideal for the disk eg. when you're running down a railway platform [ of course it also burns battery ]. Why 3 minutes ?.why does it need to wake up ? what is the delta from NLD here ? is there a fixed kernel/whatever to test somewhere ?
The delta is really huge regarding the kernel. Holger, Seife, did you notice such a wake-up? Michael: I know APM with newer kernels sometimes make problems :( BTW, I think with ACPI susped-to-ram and suspend-to-disk might work if you add acpi_sleep=s3_bios,s3_mode to the boot line. At least suspend-to-disk should work with ACPI.
I cannot reproduce the wakeup with an IBM ThinkPad T20. There are also no code changes which could result in such a behaviour IMO. I think it is a kernel issue or an external event (like network) that causes the wakeup.
So - the wake-up occurs with no network at all - eg. on the station platform [ I can't believe there is wlan-wakup-on-network support: it'd be too power intensive ;-] It also wakes up with full-power & sitting on my (fire-walled & no-wake-ups) network - where an equivalent NLD machine doesn't wake up. It also woke up & annoyed me last night shoved under my bed with no power, & wlan etc. ;-) Wrt. ACPI - AFAICS ACPI has the following down-sides for me: + no suspend-to-ram out of the box + suspend-to-disk crashes on resume with 3D accelerated graphics + no extended key support - the following set which I use are non-functional (AFAIR etc.): + suspend-to-ram, LCD brightness up/down + critically - cycle output (for projectors) The latter is broken anyway it seems (most likely) by the new X server anyway so ... will file another bug about that.
I'd really hate to debug APM... You may try to put NLD kernel into SL10, but... Or try to binary-search over mainline kernels, to see where it broke. But you should really try to debug ACPI. BTW this is definitely not severity=major. Sorry.
We really want to fix ACPI. Try reading Doc*/power/video.txt to get suspend-to-RAM working. We probably should get some DMI-based solution up and running. s-t-d 3D crash: is it with open-source drivers? Use ibmacpi to get LCD brightness and projector output (/proc/acpi/ibm, IIRC).
Anything new here?
Not really - I switched to ACPI & installed 'ibmacpi' - which indeed seems to make some of the keys work & (potentially) the projector output stuff work nicely - which is great [ not tested that in detail ]. I hope we can ship & enable the ibmacpi stuff in NLD 10. Wrt. suspend-to-disk that works fine - it's just the resuming; that regularly crashes - perhaps 1 time in 2-3. It's not clear to me how to help debug that - advice appreciated. Clearly if it was more repeatable - it'd be easier to help ;-)
Do you see any special messages during the crash? Also try it with minimal modules... to see if it is driver causing problems.
No special messages no; what I see is that it (apparently) loads to the point where it restores the previous memory image - ie. if it continues at this stage - the screen gets over-written with what I believe is prolly the contents of the screen at suspend time. However - we never get that new screen contents & it just hangs there - I imagine that may well be a somewhat risky stage. Is there no kernel option for more verbose debugging dumpage during this process ? Wrt. detailed kernel crash dumps / register information - I get none of that - just a hang with no output.
Okay, so I guess we should close this as "wontfix" (debugging apm is hard and you switched to ACPI, anyway). swsusp problems should go to separate bugreport. No messages... that's bad. Still there's chance that with minimum system it will work reliably. If you only get one crash out of three, binary search for module breaking it will take long, but it should be still doable...
No answer from reporter for some time. I close the bug as WONTFIX now. Please reopen if there is something new.