Thursday, December 18, 2014

Waking Up is Hard to Do

(Apologies to Neil Sedaka for the weak pun in the title.)

In the continuing war of man (me) against machine (my new Acer PC, which I'm configuring as a Mythbuntu PVR - see my last post), I've spent a full day trying to get the box to wake on a BIOS real-time clock alarm, so that MythTV can convince it to wake up when a recording is scheduled. The battle took so long because two things needed to be set correctly, and I kept getting stuck on incorrect combinations (this one right, that one wrong; vice versa; both wrong) while spending huge amounts of time searching the web and beating my head against the wall.

I copied a couple of time tested shell scripts from my previous Mythbuntu box, one that sets the BIOS alarm to an arbitrary time (used by MythTV to schedule actual recordings) and another that simply sets up an alarm for five minutes in the future (useful for testing). Before trying them out, I reviewed the official ACPI Wakeup wiki for MythTV. One of the first things it suggests is looking in the kernel log to make sure the BIOS supports ACPI wakeup (although I think pretty much all modern BIOSes do). In the log, I found the good news that the BIOS supports alarms up to one month in advance and this bit of seemingly bad news: "System wakeup disabled by ACPI".

I nonetheless followed the directions on the wiki for tweaking the shutdown script (under "Disable hwclock updates", although the tweak for Ubuntu does not so much disable them as reassert the alarm time after doing them). I ran the wake-in-5 test, and it failed.

Next, I got into the BIOS and checked the power settings. My BIOS has a "Deep Power Off Mode" (enabled by default) and a "Power on by RTC Alarm" feature (disabled by default). According to the BIOS help, the latter works only if you also disable the former. With "Deep Power Off Mode" disabled and "Power on by RTC Alarm" enabled, I was able to schedule a wakeup call for a fixed time either on a give day of every month (e.g., the 15th) or every day of the month. Neither is what I want -- I want to wake up only when there is something to record -- but at least I was able to confirm that the machine could wake on its own.

I'll skip a lot of gory details here, since they pertain to wasted effort. I was doing a lot of reboots during my experiments, and I noticed that one of the system logs was showing about five hours and change between reboots when actually they were minutes apart. Since I'm in the US Eastern time zone (GMT-5), I suspected that some confusion between local time and UTC (a five hour difference) was going on. Eventually I came to look in /etc/default/rcS, where I found the setting "UTC=no", telling the operating system that the BIOS was using local time. That seems to be the default setting, since my desktop PC has it as well. The BIOS on the Acer, however, was using UTC. I changed "no" to "yes", and I think that has fixed the five hour bumps. By itself, however, it did not fix the wakeup problem.

It turns out that having "Power on by RTC Alarm" not only schedules fixed wakeups I don't want, it also blocks the system from setting any other wakeup calls. This is covered in the "Fussy BIOS" section of the ACPI Wakeup document. So I went back into the BIOS, disabled "Power on by RTC Alarm" (but left "Deep Power Off Mode" also disabled), rebooted and ran the five minute wake-up test successfully.

The next step was to program the machine to wake and record a show ... which it did not do. So I went back, followed the directions for Configuring MythTV Automatic Wakeup and Shutdown (following directions -- what a concept!), and it seems to be working now.

Configuring wakeup/shutdown included setting up MythWelcome, which I had planned to do in any case. The directions are very clear and well-written, including explanations of one or two less obvious bits. There is one thing on the page that I find a bit confusing, though. In two different locations, the instructions call for you to exit MythWelcome by starting a terminal (F12 key) and then running a command to kill it. MythWelcome has a pop-up menu (accessed via the 'M' key), and one of the choices on that menu is to exit MythWelcome. It's both easier (much less typing) and a bit more natural/graceful, so I'm not sure why the instructions have you do it the hard way. In any case, I used the menu to exit, and things worked correctly.

No comments:

Post a Comment

If this is your first time commenting on the blog, please read the Ground Rules for Comments. In particular, if you want to ask an operations research-related question not relevant to this post, consider asking it on OR-Exchange.