Subject: ACPI userland issues
To: None <email@example.com>
From: Steven M. Bellovin <firstname.lastname@example.org>
Date: 06/20/2006 13:57:12
First, my thanks and congratulations to everyone who's been working on
ACPI -- I love some of the new abilities, and I can't wait until I can
switch to it. It's almost there.
I do have some issues involving the user interface, though.
There don't seem to be any. I do a lot in my /etc/apm/suspend
and resume scripts; acpi needs that ability
Do we support other ACPI states? Are we going to? Standby
What does acpi do on a lid open event? Yes, there's a
script -- but how does the script run if the machine is
still in state 3? If it's not in state 3 at that time, is
there a purpose to the script? On my T42, I can suspend
with the proper sysctl, and wake up by hitting the Fn key.
If I suspend and close the lid, though, opening it doesn't
wake the machine up. I haven't tried writing a lid_switch
script that will do the sysctl when I close the lid, to
see if that makes for useful behavior when I open it.
'envstat' gives a lot of useful information, but it's
missing some things. The battery model and serial number
really should be available that way; I'd love to write a
daemon that tracked battery behavior over time. (Windows
has such a thing, at least for Thinkpads.) There's no
indication of 'time left', though arguably that should be
calculated at a different layer.
The values for the design capacity of the battery vs. the
last full charge are *very* useful. (My Ultrabattery,
which I had felt was dead, really is -- it's last full-capacity
charge was 3 Wh, compared to a design capacity of 23.76
Wh... My old main battery was only managing 46.66 Wh,
compared to a 71.28 Wh design capacity.)
This does pose an interesting question about what battery
status tools should display. sysutils/asapm, which has
been patched to use the envstat interface if available,
gives battery percentage relative to the design capacity.
apm (and Windows) seem to give it relative to the last full
charge. It isn't clear to me which is correct. It's good
to know how worn out your battery is, but it's also good
to know when it's as fully charged as it can be.
acpibat did not seem to notice when I (suspended and)
switched batteries; in particular, the design and last full
charge capacities did not change. This is a bug, though
arguably I should have told it that I was doing this. Of
course, there's currently no way to tell acpi about switches,
whether it's inserting or removing an Ultrabay DVD drive
or simply switching batteries.
I wish there were a single interface that scripts could
use when talking to power management. Right now, I use
'apm' for various things, such as setting my screen background
color and logging battery state information. With acpi,
I'll have to use envstat, and it doesn't given quite the
same information (such as minutes remaining).
Preserve the kernel interface to the apm command. On acpi
systems, emulate it, keeping the current semantics. It's
simple and easy to use in scripts. Keep envstat for more
Add support for suspend/resume scripts to powerd. If we're
going to support other ACPI states, have scripts for them, too.
I'm not at all convinced that having a single lid_switch or
acadapter script is correct, since the actions to be taken
for close/open or disconnectd/connected are completely different.
Add a configuration file to let some choices be controlled without
hacking the code -- 'lidswitch_close: acpi3' versus
'lidswitch_close: ignore' or some such.
--Steven M. Bellovin, http://www.cs.columbia.edu/~smb