**** BEGIN LOGGING AT Thu Sep 20 02:59:58 2012
Sep 20 10:40:20 <dromede> jay7:ping
Sep 20 17:28:49 <Jay7> dromede: hi
Sep 20 17:29:50 <dromede> hi
Sep 20 17:30:04 <dromede> i mentioned before that kexecboot menu is very slow
Sep 20 17:30:11 <dromede> at least it is on my machine
Sep 20 17:30:27 <dromede> it takes literally a second to step through each menu item
Sep 20 17:30:38 <dromede> so i looked at the code a bit
Sep 20 17:30:43 <Jay7> yes, I know what caused that
Sep 20 17:30:55 <Jay7> we have bug somewhere around processing events
Sep 20 17:31:05 <dromede> well, it seems that in do_main_loop for every keypress, it would redraw the entire screen 6 times
Sep 20 17:31:08 <Jay7> about 6 redraws per event
Sep 20 17:31:11 <dromede> yes
Sep 20 17:32:08 <dromede> now, i have patch that fixes this in do_main_loop
Sep 20 17:32:42 <dromede> but i'm not sure if this is the right way to handle this
Sep 20 17:33:16 <dromede> i had something like this in mind:
Sep 20 17:33:18 <dromede> http://paste.debian.net/191285/
Sep 20 17:33:59 <dromede> with this patch, the performance is acceptable
Sep 20 17:34:42 <dromede> and if i enable the bg buffer thing,  the menu  is almost too fast :)
Sep 20 17:35:00 <dromede> the redrawing problem is due to an unhandled A_NONE event
Sep 20 17:35:11 <dromede> or due to the fact that they get created in the first place
Sep 20 17:39:59 <Jay7> ah..
Sep 20 17:40:22 <Jay7> dromede: thank you
Sep 20 17:41:06 <Jay7> I need another hackathon again to integrate all changes :)
Sep 20 17:41:25 <dromede> why does it happen? every keypress is followed by 5 A_NONE events
Sep 20 17:46:10 <Jay7> can't investigate further now
Sep 20 17:46:42 <dromede> ok
Sep 20 17:47:17 <dromede> also, i have some good news about the kernel
Sep 20 17:47:19 <Jay7> https://github.com/kexecboot/kexecboot/blob/master/evdevs.c#L280
Sep 20 17:47:27 <Jay7> enable this
Sep 20 17:47:30 <Jay7> and check log
Sep 20 17:47:56 <dromede> i finally found the cause of the suspend/resume hang when AC_IN plugged in
Sep 20 17:48:13 <Jay7> wow
Sep 20 17:48:19 <dromede> yeah
Sep 20 17:48:20 <Jay7> is it easy to fix now? :)
Sep 20 17:48:29 <dromede> it's a one liner :)
Sep 20 17:49:03 <dromede> caused by commit b6eede112673678f8a7a1680d6ad12245443829d by eric
Sep 20 17:49:18 <dromede> he himself wrote in the commit message that he should revisit it
Sep 20 17:50:01 <dromede> basically, comment out line 132 in spitz_pm.c and voila
Sep 20 17:50:24 <dromede> this line sets gpio 18 to alternate mode 1
Sep 20 17:50:35 <dromede> alternate mode 1 of gpio 18 is RDY
Sep 20 17:50:45 <dromede> i have no idea what that means or what it's used for
Sep 20 17:50:57 <dromede> i'll  report this to eric
Sep 20 17:52:09 <dromede> combine that with my offline charging patch and we finally have a kernel that has working online/offline pm code and suspend resume
Sep 20 17:53:33 <Jay7> cool! :)
Sep 20 17:54:29 <dromede> the last one that had all of that working was RP's 2.6.26
Sep 20 17:54:36 <dromede> that was _years_ ago
Sep 20 17:55:12 <Jay7> yeah
Sep 20 17:55:26 <Jay7> you are our new RP :))
Sep 20 17:56:02 <dromede> now about kexecboot, i also have some suggestions regarding the code in fb.c
Sep 20 17:56:16 <dromede> simply put, it really wasn't written with performance in mind
Sep 20 17:56:33 <dromede> and that's fine, you don't really need to it to be fast as possible
Sep 20 17:56:37 <Jay7> show me the code! :)
Sep 20 17:56:52 <dromede> well, i'll rather ask first
Sep 20 17:56:53 <dromede> so
Sep 20 17:57:29 <Jay7> I've tried to speedup fb.c during rewrite about 1-1.5 years ago
Sep 20 17:57:36 <Jay7> it was even slower before
Sep 20 17:57:53 <Jay7> just drawing pixel-by-pixel
Sep 20 17:58:09 <Jay7> I have some updated in some local branch
Sep 20 17:58:22 <Jay7> at least special func to draw rounded frame
Sep 20 17:58:50 <Jay7> because now we are drawing frames by filled rect's :)
Sep 20 17:59:09 <dromede> well, yeah
Sep 20 17:59:19 <dromede> but not really, i think :)
Sep 20 17:59:37 <Jay7> but drawing code still have place for improvements
Sep 20 17:59:53 <Jay7> 1-2-4bpp mode is just broken now
Sep 20 18:00:01 <dromede> look at this:
Sep 20 18:00:37 <dromede> http://paste.debian.net/191291/
Sep 20 18:00:46 <dromede> this is an oprofile run of kexecboot
Sep 20 18:00:55 <dromede> usage scenario: just scroll down the menu
Sep 20 18:01:02 <dromede> the problem is very obvious :)
Sep 20 18:01:28 <Jay7> https://github.com/kexecboot/kexecboot/blob/master/gui.c#L37
Sep 20 18:01:55 <Jay7> count number of overlapped fb_draw_rounded_rect() :)
Sep 20 18:03:09 <Jay7> and later in https://github.com/kexecboot/kexecboot/blob/master/gui.c#L247
Sep 20 18:03:22 <dromede> yes but fb_draw_rounded_rect and fb_draw_rect use line drawing functions
Sep 20 18:03:45 <Jay7> every line of every rect is call of fb_draw_hline_16bpp
Sep 20 18:03:56 <dromede> and line drawing routines do a lot of work _each_ time they are called
Sep 20 18:04:04 <Jay7> sure
Sep 20 18:04:05 <dromede> and some of that code should be done only once
Sep 20 18:04:13 <dromede> for instance, bgr-rgb checks
Sep 20 18:04:14 <Jay7> sure again :)
Sep 20 18:04:24 <dromede> and final pixel format composition
Sep 20 18:04:35 <Jay7> we need central ideas and todo page :)
Sep 20 18:04:47 <dromede> so i could refactor those things
Sep 20 18:04:51 <dromede> for 16bit initially
Sep 20 18:04:56 <dromede> and if it works ok
Sep 20 18:05:10 <Jay7> that would be great
Sep 20 18:05:13 <dromede> i can expand it to other format
Sep 20 18:05:15 <dromede> although
Sep 20 18:05:32 <dromede> i really don't want to mess with 1,2,4 and 8 bit formats
Sep 20 18:05:45 <Jay7> same problem here :))
Sep 20 18:06:00 <Jay7> anyway, that is rarely used now
Sep 20 18:06:06 <dromede> exactly
Sep 20 18:06:15 <Jay7> may be it's easier to drop that completely
Sep 20 18:06:24 <Jay7> or use old pixel-by-pixel drawing
Sep 20 18:06:30 <dromede> so instead of supporting it, you could just default to the text ui for anything less than 16bpp
Sep 20 18:06:50 <Jay7> good idea
Sep 20 18:08:24 <Jay7> thank you again for your contribution :)
Sep 20 18:08:39 <dromede> np
Sep 20 18:08:42 <dromede> i plan to do more
Sep 20 18:09:37 <dromede> i think kexecboot has a lot of potential in the embedded world
Sep 20 18:09:46 <dromede> maybe even wider
Sep 20 18:09:56 <Jay7> I have one man who reviewing current collections code
Sep 20 18:10:11 <dromede> collections code?
Sep 20 18:10:12 <Jay7> so waiting some contributions from him too
Sep 20 18:10:32 <Jay7> menu items/xpm lines/boot devices lists
Sep 20 18:10:44 <dromede> ah
Sep 20 18:11:03 <Jay7> we are using special code to handle every such list
Sep 20 18:11:17 <Jay7> I hope we will have something common after
Sep 20 18:12:22 <Jay7> and another crazy idea is porting qi-bootmenu UI to us :)
Sep 20 18:12:27 <Jay7> EFL-based
Sep 20 18:12:39 <dromede> now that's a _great_idea
Sep 20 18:12:47 <dromede> EVAS library was built with embedded devices in mind
Sep 20 18:13:02 <dromede> but i guess that means you have to drop klibc
Sep 20 18:13:25 <Jay7> kexecboot can be compiled against any other *libc easily
Sep 20 18:13:42 <Jay7> that's not a problem
Sep 20 18:13:52 <dromede> of course, i'm compiling my test build with eglibc
Sep 20 18:14:13 <Jay7> we just using klibc in zaurus build because of space restrictions
Sep 20 18:14:35 <dromede> hmm, i wonder how much i can reduce kexecboot binary size if i compile it as pure thumb code
Sep 20 18:14:41 <dromede> and with -Os
Sep 20 18:14:59 <Jay7> share your results with us :)
Sep 20 18:33:52 <dromede> default OE CFLAGS, dynamic, stripped:
Sep 20 18:33:53 <dromede> * 88K
Sep 20 18:33:53 <dromede> with Os, dynamic, stripped:
Sep 20 18:33:53 <dromede> * 72K
Sep 20 18:33:53 <dromede> with Os and -mthumb, dynamic, stripped:
Sep 20 18:33:53 <dromede> * 68K
Sep 20 18:33:55 <dromede> default OE CFLAGS, static, stripped:
Sep 20 18:33:57 <dromede> * 588K
Sep 20 18:34:01 <dromede> with Os, static, stripped:
Sep 20 18:34:03 <dromede> * 588K
Sep 20 18:34:06 <dromede> with Os static, -mthumb, static, stripped:
Sep 20 18:34:07 <dromede> * 564K
Sep 20 18:34:11 <dromede> so i guess glibc is a bit too much to link statically :/
Sep 20 18:34:52 <Jay7> hehe
Sep 20 18:35:33 <Jay7> let's ask Andrea to test with -mthumb -Os
Sep 20 18:35:45 <Jay7> and with klibc
Sep 20 18:36:21 <dromede> i guess it would shave off ~20 kb
**** ENDING LOGGING AT Fri Sep 21 02:59:58 2012