[mythtv-users] Hardware Supporting Suspend to RAM

[Leighton Brough]

> I'm building a MythFrontend for my living room.  The two key ingredients
> are on board Svideo, and Suspend to RAM support in linux with a way to
> wake-up from remote.  The suspend must be silent, I.E. all fans need to
> shut off, including the PSU.  I have a Harmony remote, and was thinking of
> getting an IR keyboard so I can do a wake on keypress.  Anyone know of
> motherboards that support this in Linux (Fedora Core 5 or KnoppMyth,
> undecided as of now)?
>
>   
Hi Ben,

I've got suspend to RAM working on my box, almost completely to my 
satisfaction. This is a far from trivial problem. What you will have to 
do to get this going, and whether it will work correctly or not depends 
on a lot of factors. But I can summarise my experience which might help.

After many, many hours, I have managed to get my combined front/back end 
system to automatically wake up from S3 (the suspended to RAM state) 2 
minutes before recordings, and automatically return to S3 when the 
frontend isn't active and nothing else (recording, commercial flagging 
etc) is going on. It's been more or less reliably doing this for a few 
weeks now, without a reboot or full shutdown (which is _not_ to say that 
I don't experience crashes of MythTV from time to time). The startup and 
shutdown is fast, taking just a few seconds, the power consumption in S3 
is trivial and the box is silent.

I can shut the frontend down with the remote (in this case an MCE unit), 
but as yet I cannot get it to wake this system from S3. This is 
infuriating, as I've previously installed Windows XP on the same 
hardware, and _could_ wake it with the remote. I have yet to get to the 
bottom of this problem - I assume the 'fix' will be in the ACPI related 
portions of the kernel. Anyway, the point is I have to suffer the 
indignity of pressing the power button on the case to wake the box manually.

I'm using an ASUS P4P800S-X motherboard in this system. Here are some of 
the things I needed to do to get the Linux end of the problem working:

* With my (Debian) distro, I downloaded and compiled a newer 2.6.x 
kernel, with suitable settings to get ACPI support. You will need at 
least  2.6.x kernel, so this will limit your choice of Linux distro, 
unless you are prepared to build the kernel. You may have to do this 
anyway, if it isn't configured for ACPI. Sadly, as far as I can tell, 
with 2.6.x there is no SMP support, so my P4 cannot be run in 
hyperthreaded mode, which I would very much like to do.

* As I'm using a NVidia 6200 video card, I had to make a few hacks to 
the some of the source provided with the proprietary binary as this was 
throwing errors preventing a transition to S3 (the suspended to RAM 
state). With these, I can go to S3 without stopping X, which makes the 
resume considerably quicker, as X doesn't need to be restarted.

* I've developed some scripts which do the work of transitioning the 
system to S3. These stop a number of daemons, and unload a number of 
kernel modules, sync disks etc before the transition to S3, and restart 
them again afterwards. I haven't got access to my MythTV box at the 
moment, so I can't give you precise details. The kernel modules I needed 
to unload included hot-swap, USB support and the DVB drivers (I have 3 
Twinhan DVB tuners). Basically, you've got to keep ripping out bits of 
the OS until you can get the suspend (and then resume) to work; then 
you've found the drivers and servers which don't cope with S3.

* I had to setup Linux to interpret the real-time clock value as local 
time, not UTC, or the times pushed into the BIOS with ACPI when setting 
the wakeup time were not correct.

I'm using 0.19-fixes svn source to compile MythTV now. To get this to 
use the ACPI suspend to RAM support correctly, I had to:

* Setup the backend with mythtv-setup correctly to shutdown the when 
it's not busy, and wait for a client to connect before shutdown. You 
need to set the time it will wait rather than shutdown if a recording is 
coming up to 15 minutes. With this value is configurable in the backend, 
it's hard-coded in MythWelcome (more below), so you're stuck with this 
value in practice. It's a shame, as the suspend to S3 and resume is so 
fast I would prefer a value more like 2 minutes.

* I'm using MythWelcome, which helps, but it assumes that you're using 
nvwakeup for the wake on real-time clock functionality, rather than 
ACPI, so you have to configure this appropriately too - sorry about the 
lack of details again - the point is it is possible. I ended up writing 
a little Perl script which takes a command line of the form assumed by 
MythWelcome for setting the wakeup time, and translates this to an 
appropriate setting for /proc/acpi/wakeup.

In a nutshell, pretty much any motherboard you buy today I would guess 
supports S3 and ACPI. However the Linux support for this isn't great yet 
(it's not perfect in Windows either, but it's better). So unless you can 
find positive proof of a working example for a particular configuration, 
you can't be sure it will work. There are also issues to do with video 
restarting correctly on resume. Most motherboards have some BIOS 
settings relating to this which can help (didn't work for me), and there 
some tools like vbetool which can also help (didn't work for me either). 
So what happens for me after a suspend and resume is that the X video 
state is correct, but the consoles' video states are trashed. This isn't 
the end of the world, as I don't need to use the console. But it does 
make it harder to debug the system, as you can't see all of the shutdown 
messages.

I also find that I cannot do a reboot or 'shutdown -r' after a resume 
from S3. If I attempt this, the system hangs on shutdown, and after a 
power-cycle, the BIOS date and time have been trashed (ughhh!). So I 
avoid that too!

I hope that gives you a bit of a picture. Others may have had much 
easier results.

Cheers,
Leighton