Request Memory SWAP support

[GoogleCodeExporter]

What version of the product are you using?
0.03 - 21.10.2010 0.03-10.10.21b

Please provide any additional information below.
Could we enable SWAP support in this kernel ?

Original issue reported on code.google.com by ComFla...@gmail.com on 27 Oct 2010 at 12:28

[GoogleCodeExporter]

As you may know the kernel is not a part of this release because the Milestone 
bootloader only loads a kernel that has been signed by a (secret) key. Because 
of this fact no one else will be able to provide a kernel that is able to 
support swap support. I doubt this could be done by a module but things 
happen...

Original comment by bodorif...@gmail.com on 27 Oct 2010 at 9:09

[GoogleCodeExporter]

My understanding is.. 2nd Init able to allow us to get our own kernel (thats 
why we have CM6 now). The only one we couldn't modify now is boot.img, am i 
right?

Original comment by ComFla...@gmail.com on 27 Oct 2010 at 9:23

[GoogleCodeExporter]

Kernel is a part of the signed boot.img.
2nd-init allows us to use custom init and init.rc scripts in a clean way.
2nd-boot would allow us to use a custom kernel, but the ultimate modem panic 
during kernel restart haven't been resolved yet.
Therefore we are still bound to use only the kernels compiled and signed by 
Motorola.

Original comment by kabal...@gmail.com on 27 Oct 2010 at 10:46

• Changed title: Request Memory SWAP support
• Added labels: Type-Enhancement
• Removed labels: Type-Defect

[GoogleCodeExporter]

Thanks nadlabak for xplaination. Hope 2nd boot become reality, so we could have 
better custom kernel

Original comment by ComFla...@gmail.com on 27 Oct 2010 at 11:37

[GoogleCodeExporter]

Original comment by kabal...@gmail.com on 5 Nov 2010 at 11:42

• Changed state: Kernel_issue

[GoogleCodeExporter]

Hey Pavel!

We got some discussion going on regarding this topic here:
http://www.android-hilfe.de/root-hacking-modding-fuer-motorola-milestone/53575-d
iskussion-256-mb-ram-zu-wenig-swap.html

Question: As the kernel does not support swapping, is there something as a 
"swap.ko", that could be ripped somewhere (Droid2, X, Nexus, whereever) and 
injected into the milestone? As done for openVPN, governors, overclocking, etc.?

As a background: SD-card nearly reads at double data rate than internal flash 
memory. So to increase I/O Performance one could copy as much as possible on 
the SD-card and remount/relink it there (e.g. /data/... /system/... 
/cache/...). But this will cause a lot of problems, e.g. during mounting the SD 
card over USB as a mass storage device. Also the data would have to be copied 
on every boot -> sloooow. Nice and direct workaround to make use of the 
increased data rate would be a swap memory of e.g. 128 MB or similar. I bet it 
would be faster than loading everything from the slow internal NAND flash.

As a comparison (using "J Bench Mark"): On my milestone with 8 kb buffer:
Internal Flash: 0,5 MB/sec write 3,5 MB/sec read
SD card: 3,5 MB/sec write 6,5 MB/sec read
(it's a class 6 SD card)

BR!

PS: I WISH my programming skills would be better and I could search for a 
solution on my own. I always feel awkward asking such stuff to "nadlabak 
again", who got his baby, family, CM6 etc. going on in parallel.

Original comment by doc.pa...@gmail.com on 11 Nov 2010 at 11:27

[GoogleCodeExporter]

Some parts of the kernel source are ready to be compiled either as a direct 
part of the kernel or as a loadable module. So, compiling the modules for 
kernel is quite easy (although sometimes the sources need to be patched to work 
correctly with the particular kernel). Btw., the modules are not being ripped 
from other phones, they are simply compiled from sources.

But some parts of the kernel source are only to be compiled as a direct part of 
the kernel - eg. the swap support. Remaking of such parts to a module is not an 
easy task. That is because a direct part of a kernel has access to all kernel 
symbols (functions, data structures), while a module has access only to those 
symbols that have been specifically exported to be available for use from 
modules.

Although there are ways to make the unexported symbols available in modules by 
finding the direct addresses of them in /proc/kallsyms - that is btw. used in 
the overclock and interactive governor modules (the addresses supplied to the 
modules as parameters are the addresses of the very symbols that need to be 
used but have not been exported for modules by kernel).

The swap code integration in the kernel is very complex. Separating it to a 
module would need excessive amount of work, while I'm not sure if it is at all 
possible, given the fact that the module would not be present in the kernel 
from the start, but only later when the first insmod command can be executed...

Original comment by kabal...@gmail.com on 11 Nov 2010 at 12:06

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

-> :P :P :P

Thnx for that explanation!

Original comment by doc.pa...@gmail.com on 11 Nov 2010 at 12:42

[GoogleCodeExporter]

Thanks kabaldan for explanations.
I understand that it would be hard to make such a module but it would be great 
if a skilled dev can give it a try. 

More and more memory intensive apps are coming to the market (especially games) 
and i feel sooo sad to not be able to use them because of 256 MiB limitation 
(example Dungeon Defenders runs great on Droid that have a custom kernel using 
swap feature. The app just won't load on milestone). Even for the whole system 
speed, this feature would be soo great ... 

Well keep up the good work on CyanogenMod port guys !

Original comment by aurelien...@gmail.com on 10 Jan 2011 at 10:23

[GoogleCodeExporter]

swap and compcache would be great ^^
i've tried in ubuntu, compcache can compres up to 80% memory

Original comment by superbiji on 11 Jan 2011 at 3:02

[GoogleCodeExporter]

Hmm...

Adding swap will be something huge, yes.
But there are a few other things to think about:
-Put more pressure to the developer to reduce the memory-usage from their 
applications. Many yust eat way too much in my oppinion.
-increse the usable ram. Why is the inactive ram so much? I don't think that 
cache should be higher priority than not-killing programs...
-letting the widgets into the same VM as the homescreen. I don't think that 
its's good that every widget need its own VM. Or: build some more basic infos 
like IP, CPU-Freq, RAM, temperature and the like directly into the homescreen.
-Let the vm compress the addresspace from apps and safe it to the sd-card, 
instead of killing them. This should be quite fast - much faster than 
restarting apps all over the time. A fast compression should be quite enough.
-improve the restarting-behavior from the most important appliations like the 
browser. Why do they need to reload the whole page? I think that they should 
save the complete data before being killed, so that even if there are killed, 
the restarting is almost not noticable.
-deactivate the kernel-OOM alltogether and let an android-process controll it. 
This way we could archive much fine control which tasks are being killed - and 
could do things like automatic saving their states, send "save memory" signals 
to the processes and the like.
-instead of building swap in the kernel, build it in dalvik. This should be 
much more easy. This should cover most processes, because on a "normal" andrid 
system there aren't such many traditional linuxtasks.

Original comment by mifritsc...@gmail.com on 19 Sep 2011 at 10:33

[GoogleCodeExporter]

Yust another way: ask Motorola to build a kernel with swap ;) I think that this 
is way more probably that unlocking the bootloader, at least in the next time...

Original comment by mifritsc...@gmail.com on 20 Sep 2011 at 6:02

[GoogleCodeExporter]

-stop the crappy apps2sd to mount every single app during start, but do it if 
needed (for version 1.0, umounting isn't really needed I think, because after 
even 5 days of usage, in most cases only a tiny part of apps are used since the 
start)
-kill the oom-killer :D I've seen situation where the oom got invoked even 
besides 44 of free RAM+ 60MB inactive! I think it would be worth the experiment 
how far the memory managment can get if it hasn't the "tool" oom-killer... Yes, 
I know, at some point it will procuce  a big fat BAM, but I assume not in 
normal usage situations and and atleast not in the first 5 minutes. Then we 
could build our own userspace oom-killer which does kill processes only if 
really needed.

Original comment by mifritsc...@gmail.com on 20 Sep 2011 at 4:04

[GoogleCodeExporter]

Issue 563 has been merged into this issue.

Original comment by kabal...@gmail.com on 27 Nov 2011 at 12:26

[GoogleCodeExporter]

wouldn't it be great if somebody like nadlaback ask motorola to get a key to 
sign cyanogenmod for feature development for the milestone, maybe he will get 
it, that in the future cm could get some important features.

Original comment by itgch...@googlemail.com on 22 Feb 2012 at 10:09

[GoogleCodeExporter]

Custom kernel with working gsm/umts is now possible on Milestone, see

http://forum.xda-developers.com/showthread.php?p=32512060#post32512060

and the original breakthrough thread

http://forum.xda-developers.com/showthread.php?t=1908687

There's still a lot of work to be done, but new era has been already started.

Original comment by kabal...@gmail.com on 13 Oct 2012 at 11:53

• Changed state: Started

[GoogleCodeExporter]

Im trying 7.2.4 with swap enabled now.. thanks kabaldan. its working great now. 
64MB swap size with compcache disabled is my sweet spot. No more reboot, FC and 
minimal slowdown.

Additional Setting:
/sys/block/mmcblk0/queue/read_ahead_kb = 2048kb

VM Max Heap Size = 48MB

SDcard speed = 16GB Class 10

Original comment by ComFla...@gmail.com on 24 Oct 2012 at 7:22

[GoogleCodeExporter]

Original comment by kabal...@gmail.com on 5 Nov 2012 at 11:30

• Changed state: Fixed