Sorry for the delayed report, the fact that I didn't report anything for three weeks was a surprise for myself, I thought it was "just" two weeks.
== Highlights ==
* Spent quite a lot of time debugging KDB/FIQ random hangs. The hangs were a challenge to debug, although resulted in three lines of code to change (the notorious SVC exit path). But hopefully it's all over and KDB/FIQ now survives all my most strict testcases, and no single hang observed;
* So I sent out v4 of KDB/NMI/FIQ debugger:
- The new set includes NMI tty and console driver. KDB 'nmi_console' command is used to make debugger shell usable for normal Linux console, but with ability to enter the magic sequence. KDB command 'disable_nmi' is more powerful, it releases NMI, and thus handles serial port back to normal serial driver; - Fixed amba-pl1011 driver polling initialization (and that also fixed another issue the driver, i.e. division by zero if console= and kgdboc= point to different devices). - Some more work on FIQ code cleanup: a new patch that removes init_FIQ() completely (plus maybe fixes a real bug, but yet not sure, let's see how review goes); - Rebased on top of the latest Linus' tree.
* Some pstore related work:
- I started receiving pstore/ram related patches, so to track the flow I created linux-pstore.git tree. I wanted to merge the fixes into v3.6-rc2, but it didn't make it. So, this is now postponed till v3.7 (linux-pstore tree is now in linux-next, btw).
- Got an 'OK' for pstore/ftrace debugfs rework, yay. The patch will be merged into linux-pstore.git tree soon, just need to finish tests.
* Announced userland lowmemory killer daemon. Almost no feedback;
* Recently there was a flood of battery/charger-related patches for review, so had dedicate some time to it as well;
* I looked into latest cgroups kmem/slab accounting work, and I still don't see any way to account slab for root cgroups (i.e. account drivers' memory). It seems that the fact that the root cgroup does not account true kernel memory is the design decision. Time to move on, I guess. We'll probably need some alternative to cgroups anyway.
== Plans ==
* Actually, I started looking into it months ago, but didn't end up with touching any code: I wanted to add 'deferrable' timer into posix timers or timerfd API (or any timers API that is nowadays considered 'mainstream' and would be suitable for these kind of things).
The deferred timer means that we want to be notified about the timeout, but only if the system is running, the timer itself won't cause system to wake up. The "problem" with the timer is that I don't see any use-case for it apart from LMK (i.e. polling /proc/vmstat without wasting battery power). And the fact that I could not see other usecases makes me wonder if there's something wrong with the idea.
Although, the idea is pretty close to ITIMER_PROF, except that we want to decrement the timer based on the wall clock, but 'when the system is executing'.
John, I believe you know a lot more about timers-related things, so maybe you have any advices/directions, based on my thoughts above?
* Apart from LMK work, I don't have anything for real 'development'. My main task for now is to get KDB/FIQ into 3.7, but hopefully I can do it in "maintenance mode", thus I guess it is time to start looking at picking other tasks. Is there anything on the priority list?