In order to maximise testing and reporting on failed suspend/hibernate/resume cycles we are proposing to modify processing such it is possible to track whether a suspend/hibernate was in progress. This allows us to detect this on normal boot and then report it via apport to launchpad. Additionally we are delivering a suspend stress test. Change fall into three parts:
- kernel: shipping the suspend stress test script with the kernel image itself,
- pm-utils: modifications to actually track the what operations are in progress, and
- apport: adding a new type of report for failed operations.
---
KERNEL:
The changes here are already applied and released.
PM-UTILS:
simple new sleep.d script to expose the current 'in-progress' operations such that external scripts can detect it indicating a failure during that operation.
APPORT:
There are changes required to apport in order to support reporting of suspend/resume problems. During suspend and hibernate, pm-utils will create a 'flag' file while going down, and remove it while coming up. When the apport init script start function runs, it will check for the existence of this flag file. If it is present, it is assumed that a previous resume failed and that the user restarted the machine.
These changes to apport add the check in the init script. If the flag file exists, then a python script is called which creates a crash report file in /var/crash for later processing by apport, and the flag file is deleted.
Martin Pitt has reviewed these [changes to apport] and approved them, but he's on holiday this week, so could someone else please push them into Jaunty?
Attached to this comment are the proposed changes to pm-utils to record the current operation. Currently this is done via a new sleep.d script.