Re: BFS for the power kernel
 

hmm... after installing the latest kernel with bfq I got into an reboot loop. guess i have to reflash it.

Re: BFS for the power kernel
 

I have crashes immediately after turning on bfq, do I need to disable cgroups before that?

Re: BFS for the power kernel
 

Well I reflashed the kernel (I think I just forgot to install the flasher this time hence the reboot loop) so it's working fine now. BFQ is also enabled, so far no crashes but I guess I'll have to test more.

Re: BFS for the power kernel
 

Hmm, no, you don't need to disable cgroups. In fact, they should be enabled (I would imagine BFQ would crash without them). Could you run

.. and attach the file it generates?

Re: BFS for the power kernel
 

Look at [13374.992126], here I turned on bfq for eMMC, switched to desktop and opened the browser. The crash happened trying to read bookmarks file, I guess.

Re: BFS for the power kernel
 

Hmm, there are a whole lot of the null reference errors there, all caused by BFQ. Not good. Could you check that /syspart is being mounted correctly, and that bfq.ioprio, bfq.ioprio_class and bfq.weight are all present under that directory (when BFQ is the default scheduler, of course)? You could set it just for mmcblk1, if you've an external SD.. I am guessing that /usr/lib/libohm_cgroups.so is present and loaded?

Re: BFS for the power kernel
 

/usr/lib/ohm/libohm_cgroups.so is present, how can I check if it's really loaded? I guess it is, /proc/mounts contains

/syspart contains

setting bfq on mmcblk1 doesn't change anything in /syspart

Re: BFS for the power kernel
 

Yep, it seems to definitely be loaded. Did you have bfq as the default scheduler for mmcblk0, if so then setting it on mmcblk1 wouldn't change anything in /syspart.. it's also possible that bfqio is enabled regardless of whether bfq is or not, and just sits idle until called upon. Ok, could you please try one of these (depending on your setup - the modules should be unchanged), I have disabled the "queue small requests" part of BFQ v2 r1 and have been running it happily on my device for a full day without any crashes..

Multiboot deb
Standard deb (use with flasher)
U-Boot image

Re: BFS for the power kernel
 

Hmm I'm still getting crashes immediately after activating bfq.

Re: BFS for the power kernel
 

Do you have any applications etc running when you change to the BFQ scheduler? I activate it in an /etc/event.d script, but it would probably be fine to do so after reboot from X-term - not so sure if you've had the system running for several hours/days beforehand, though.

Re: BFS for the power kernel
 

The system was just started a few minutes before and I also put the "echo bfq ..." into /etc/init.d/rcS and tried it after several reboots. I was able to start xterm but as soon as I started something like cutetube or the browser it always crashed pretty fast. I also once noticed that it rebooted while the screen was turned off and the system was idling. So I guess that it still has something to do with swapping.

Thanks for the effort nonetheless!

Re: BFS for the power kernel
 

cfq is the default scheduler, /syspart/bfqio* is always being created

I always prefer to set scheduler on start of hildon-desktop, I don't want to have reboot loops :) Btw, how can I set "elevator=" for the kernel via multiboot? Does it work? My only kernel is bfs, so I'm accurate with booting.

Tried the version from your last post, with bfq for mmcblk0 UI freezes (again I switched to desktop and opened the browser) and watchdogs reboot the system after 30 seconds or so, there is no messages in crash log. Seems like Xorg gets stuck, but it doesn't seem to be related to swap: I load Xorg with the mlocker.so trick.

With bfq for mmcblk1 the kernel immediately crashes after opening some application that takes additional memory (because I have swap on mmcblk1 in addition to mmcblk0), here is the log:

Bought my external card several days ago, there must be no read/write errors. But my eMMC is not so fine, sometimes crashes happen on high fs load, have no idea if that can be related. Will examine /dev/mtd2ro on next crash, thanks for the hint!

Re: BFS for the power kernel
 

I also load Xorg into memory using libmlockall, which is the same as mlocker, basically. The solitary crash I had (yesterday morning) occurred also when the screen was off and the system idle. I have BFQ on both mmcblk0 and 1, and my swap is also on 1, so I am not convinced about it being swap at fault.

The next stage is to build the kernel without bfqio enabled, if you wouldn't mind testing it. I am unable to reproduce the instability in my own device, which is frustrating as it makes tracking down the cause difficult..

Re: BFS for the power kernel
 

Ok, here we go again with bfqio disabled this time..

Standard
Multiboot
U-Boot

Re: BFS for the power kernel
 

After some minutes messing around in cutetube, watching videos and using the browser there were no crashes so far.

Re: BFS for the power kernel
 

Same result here - about an hour of heavy use with the new kernel (bfqio disaled). BFQ is enabled on mmcblk0 and mmcblk1, and no instability yet.

For the record, I've been testing BFS kernel with swap on default emmc partition and had very bad stability issues using BFQ with the previous kernel.

Re: BFS for the power kernel
 

I've just realised I made a mistake when backporting the code. 2.6.32 and above (which the 'official' BFQ v1 was written for, when the site moved some time in 2009) calls __blk_run_queue() instead of blk_start_queueing(). As __blk_run_queue exists in 2.6.28, I assumed this was an optimization - but it turns out that in fact, blk_start_queueing was depreciated some time between .28 and .32, with its sanity checks added to __blk_run_queue; unfortunately, the .28 version (of __blk_run_queue) has almost no such checks at all. I'm just recompiling again, having changed the relevant lines back to blk_start_queueing, and will be testing in due course..

Re: BFS for the power kernel
 

Any updates on this? Hmm this thread is already 1 week inactive

Re: BFS for the power kernel
 

I've been trying to track down the cause of the distorted calls, to no avail so far, sorry to say. I implemented a subset of this patch (minus the "avoid recursion in run_workqueue()" patch, as this causes regressions, and the avenrun patch as that makes BatteryGraph go mental) and also updated the "need_resched" and "should_resched" logic to match the 2.6.29 backport of BFS, but the problem persisted. However I've since found that powertop is showing my C2 to be at 80% almost constantly, even with the power kernel (though my battery life is very good) so I'm contemplating reflashing everything and starting over, but I don't have the time for that at the moment unfortunately.

I did, however, have some joy updating the UBIFS drivers from this git tree, I can't tell if it's any faster or not though. Oh, and I did change the two places __blk_run_queue was called within BFQ (there was no discernable difference, i.e it was still as stable as before without cgroups, but as soon as I enable them, crashes aplenty - not tried since implementing the prereqs though).

So, although I haven't updated the thread, I've not been inactive ;)

EDIT: Almost forgot, reversed this patch (which has always been present within our BFS) as it's for Android's VM only and I really can't see why it would be needed on Maemo, but again I haven't noticed any difference.

Re: BFS for the power kernel
 

Hmm... the CPU in my n900 runs about 89% in C3 and I can happily report that I didn't notice any distortion in the last time.

Re: BFS for the power kernel
 

Hi would like to ask if there's a new upcoming deb file for us? :D

Re: BFS for the power kernel
 

It would be nice to see this effort continue -- this seemed like a pretty exciting thread.

Re: BFS for the power kernel
 

I and Tigerite haven't found the cause of the distortion yet. We only know it's caused by the bfs-330-to-350.patch in our tree. The problem is that after putting a lot of effort in finding the cause without result, things kind of shifted to the background (at least for me). Then we also have some other things going on, which push kernel-bfs back a little more, unfortunately.
On a side note, kernel-bfs actually is up-to-date at this moment. The latest BFS version is v0.406, which fixes an incompatibility in v0.404 with kernel 2.6.39 (which is irrelevant to us). The changes that went into kernel-power v47 are in kernel-bfs as well already. The only thing is that you have to git clone and build the kernel for yourself currently.

If you were talking about BFQ (as the last pages of this thread were about this subject): last time I spoke with Tigerite he noted that it was ready to be committed. I don't know why it isn't committed yet though. I do know that Tigerite has been busy with real life (and a bit with Meego/Harmatten), so that might be a reason.

Re: BFS for the power kernel
 

Cloned git tree and tried to build it with a few options changed in config, but every time it fails after making modules in one place without any meaningful error... Tried also building without making any changes anywhere. Yet still it fails.
Am i missing something?
I've tried to build it with your build.sh script, and also by using make modules. Results are the same...

Re: BFS for the power kernel
 

Nope, it looks like you did everything right (considering the fact that you've got the modules to build).

I'm just about to go out of the country (holidays ;)), so I won't be able to look into it next ~1.5 week, unfortunately. I'll get back to you when I'm back home.

Re: BFS for the power kernel
 

...enjoy your vacation! :D

Re: BFS for the power kernel
 

Is there a way to use bfs with kernel-power v48? I am n00b with git, code or compiling something...

Re: BFS for the power kernel
 

Okey, Have a good vacation! I'll be waiting :D

Re: BFS for the power kernel
 

OK. something weird...
Today tried to rebuild it again and... it built! and it even works!..
Now I don't know where the problem was..
Just that i used now:
And put al the source code in another sub directory, like so:
/scratchbox/users/blayder/home/blayder/bfs/linux-2.6.28/
Now recompiling with modded config, will see if it works.
Weird stuff is happening all the week...

Re: BFS for the power kernel
 

I'm really sorry for not pushing the BFQ patches. As iDont rightly says, I've been incredibly busy in real life lately :( and that's likely to continue for another week or so yet. I haven't even had much chance to look into BFS+BFQ on Harmattan, so please don't think I've deserted Maemo 5, it's not the case ;) I will try to push the patches today - unfortunately, the Harmattan SDK has borked my Fremantle scratchbox environment, which doesn't help matters!

Re: BFS for the power kernel
 

As promised, the BFQ patches are now in the GIT tree :)

Re: BFS for the power kernel
 

AWSOME! :D
Will rebuild kernel this evening again :D Yaay!

Re: BFS for the power kernel
 

please share it :D haven't updated my kernel for ages lol

Re: BFS for the power kernel
 

I could, but my rx51power_defconfig isn't stock :D
And i'm not sure it is stable, since i compiled it just yesterday :P

And i wanted to ask allso: How do you properly disable watchdog?
if i disabe them in config, it compiles fine, but phone still fails to boot.
Boots fine in R&D mode. So where's the catch?

Re: BFS for the power kernel
 

what is this bfs?is it fr normal users? what does it do?

Re: BFS for the power kernel
 

It is a different CPU scheduler
BFS
P.S. Yay - got working BFS kernel with framebuffer, next gonna be truing to get a fbcondecor working ^_^. Patch applied fine.

Re: BFS for the power kernel
 

How can I install BFS?

Re: BFS for the power kernel
 

Install kernel with BFS =)
http://talk.maemo.org/showpost.php?p...&postcount=254

Re: BFS for the power kernel
 

Is compatible with kernel-power v48?

Re: BFS for the power kernel
 

Read whole thread pls, it's not so long.