IRC log of #cubox of Wed 15 May 2013. All times are in CEST < Back to index

08:42 rabeeh frilled: ping
09:47 frilled rabeeh, pong
09:52 rabeeh frilled: hey
09:52 rabeeh what's that pixels issue?
09:58 Marmotte rabeeh: some pixels are missing on top of the screen for me, and on top and right for frilled
09:58 rabeeh which resolution?
09:59 Marmotte 1920x1080-32@60 for both
10:00 frilled [10:00:02] [20:58:31] rabeeh, ever seen the image on the console fb offset to the top right via HDMI? My consoles have like 1 pixel on the rightmost column and maybe 2 or 3 pixels on the top line shaved off. That's on 1920x1080x32@60 with alarm/xf86-video-dove 0.3.4-9
10:00 frilled [10:00:02] [20:58:52] how was your vacation BTW ;)
10:00 rabeeh frilled: vacation was great :)
10:01 rabeeh thanks.
10:01 rabeeh which kernel is that?
10:01 frilled Good to hear :-)
10:01 frilled oof, lemme check the repo, I have no connection to the cubox right now
10:01 frilled it's the current alarm v7 kernel anyway
10:02 frilled http://archlinuxarm.org/packages?arch=&search=cubox -> 3.5.7-12
10:03 Marmotte I picked the first kernel and modules here : http://www.solid-run.com/phpbb/viewtopic.php?f=9&t=897
10:03 frilled as far as I remember X11 displays correctly, need to re-check that, though, once I get home
10:03 frilled (VPN not currently working ;P)
10:03 Marmotte frilled: your problem doesn't appear on X11 ?
10:04 Marmotte I saw missing pixels under awesome wm (Debian Wheezy armhf)
10:04 frilled Marmotte, I *think* not, but I need to re-check that.
10:04 frilled So, no comment right now.
10:04 Marmotte I didn't check on TTY :)
10:04 frilled I usually don't X on that box :)
10:06 frilled I ordered a second one, though ^_^
10:07 Marmotte I saw on the wiki that the Logitech K400 was tested, I have the K400r, which has more shortcuts, is it useful to add a precision ?
10:08 Marmotte (all shortcuts works without any custom configuration :D)
10:10 Marmotte (I didn't expect the pinch-to-zoom to work on Iceweasel so easily)
12:36 Marmotte frilled, rabeeh: I just checked, some pixels are missing in TTY too for mee
12:37 rabeeh Marmotte: how do you check that on tty?
12:37 rabeeh can you draw a border ?
12:37 frilled Uhm, the text is just cut off, easy to check ;P
12:37 frilled top lines of characters are missing
12:38 frilled like "h" almost looks like "n"
12:38 Marmotte the top of the prompt is not displayed :)
12:56 dv_ rabeeh: ping
13:20 rabeeh dv_: i'm looking into your email now
13:23 rabeeh dv_: just sent you the logs
13:27 dv_ alright, I restarted a fresh build using your settings. this will take a while.
13:28 rabeeh :)
13:28 rabeeh a while?
13:28 rabeeh a long while i would say :)
13:29 rabeeh oh; more questions to all
13:29 rabeeh i'm rewriting some scripts to install ubuntu from scratch
13:29 rabeeh i can have a rootfs.tar.xz for users to download through the installer; or i can apt-get install from the CuBox installer
13:29 rabeeh which one is better?
13:30 rabeeh the second one takes longer time to execute; but you get freshest debs from the repositories
13:49 dbsx Rabeeh: use a rootfs.tar.xz, which you update say every 6 months. Tell users they should apt-get update.
13:49 dbsx That way you have an install that works, it may not work after the update.
13:49 rabeeh ok.
13:53 dbsx rabeeh, by chance you wouldn't happen to have a set of instructions for a minimal debian/awesome setup?
13:56 rabeeh nop
13:58 rabeeh dbsx: the good thing about installing from the web directly is that in the installer you can install base; then through a submenu in the installer you can install openbox, xfce, awesome or mpd etc...
13:58 rabeeh or zoneminder
13:58 rabeeh all controlled part of the CuBox installer?
13:58 dv_ rabeeh: I noticed that apparently you did not enable parallel builds
13:59 rabeeh oh
13:59 dv_ i thought you have an i7? and how much RAM?
13:59 rabeeh i7
13:59 rabeeh 8GB
14:00 dv_ hmm then uncommenting BB_NUMBER_THREADS = "4" and PARALLEL_MAKE = "-j 4" should be OK
14:00 dv_ builds happen _much_ faster then
14:01 dv_ also, I have been using the poky tarball, not git. this difference may be significant, since the tarball is months older
14:01 dv_ anyway, lets see what comes out. I need to leave now though. later
14:02 rabeeh ok. i'll try switching to the tarball
14:02 rabeeh but i'll do that tonight
14:04 Marmotte rabeeh: for our missing pixels issue, I tested on GeeXboX, and it seems to be OK
14:05 rabeeh yes
14:05 rabeeh i tested on ubuntu; and i'm sort of having 1 pixel missing too
14:05 rabeeh i'm not sure why
14:05 rabeeh geexbox are using 3.5.7 kernel
14:05 Marmotte on my Wheezy, I am using 3.5.7 too (http://www.solid-run.com/phpbb/viewtopic.php?f=9&t=897)
14:06 rabeeh what is your xorg.conf?
14:07 Marmotte Debian's default, or none if there is no xorg.conf by default, I didn't look :D
14:08 Marmotte (I have my cubox since last Friday :P)
14:08 rabeeh probably none
14:08 rabeeh which means you are using 16 bpp color depth
14:08 rabeeh where geexbox uses 32bpp
14:08 rabeeh RGBA with alpha blending
14:09 Marmotte I have put the video= argument on the kernel line in my uboot script
14:09 Marmotte (without that, I didn't have any X display)
14:10 Marmotte (I don't know if it's necessary or if I did a mistake :P)
14:14 dbsx Rabeeh: the most common questions that I have been asked are:
14:15 dbsx 1. why cant we just have tar.xz of the keenel of distro?
14:15 dbsx 2. I can't get the installer to work, can you help?
14:15 dbsx kee - ker
14:16 dbsx and that should have been kernel and distro
14:16 Marmotte dbsx: did you get an answer for awesome ? (I was disconnected after your question :D)
14:17 dbsx Marmotte: no.
14:17 dbsx 3. How do I build the video drivers for xxx?
14:18 Marmotte I made some tests on Debian Wheezy, and I have awesome running after a clean debootstrap, then "apt-get install xinit xserver-xorg awesome", and I put "awesome" in my .xinitrc
14:18 Marmotte (I always debootstrap with minimal profile)
14:19 dbsx Marmotte: I have the same on jessie. I am also a minmalist
14:19 rabeeh how big is that?
14:19 rabeeh (in .xz)
14:20 dbsx minimal debian is < 50M
14:20 dbsx or was that not the question
14:21 rabeeh i mean with awesome
14:22 rabeeh which one would you pick between xfce4, lxde, awesome, openbox, matchbox
14:22 rabeeh ?
14:22 rabeeh oh; and blackbox?
14:22 rabeeh just installed awesome on ubuntu
14:22 Marmotte I use Xfce since 2008, and I am switching to awesome now :)
14:22 rabeeh why?
14:23 dbsx awesome fits a usage case, and the Lua hooks are a winner
14:23 Marmotte too much custom bindings, Xfce becomes unappropriated for me :)
14:23 Marmotte and I hate the mouse :)
14:24 dbsx Rabeeh: I benchmarked Lua and Luajit on CuBox. It is pretty quick.
14:25 Marmotte I never really used openbox, so I can't say if it's good or not ^^
14:25 Marmotte (LXDE is openbox based, if I remember correctly)
14:26 dbsx xfce4 has its fans (probably the most popular of the list (guess))
14:46 rabeeh dbsx: the only kernel i can fully rely today is my tree 3.6.9
14:46 rabeeh beyond that the DT based are not complete
14:46 rabeeh that's referring back to the 'most common questions' you previously noted
14:47 rabeeh it's missing some patches about CEC, and Audio (which i doubt they are fully working)
14:48 rabeeh those are mainly coming from geexbox devs
14:48 dbsx rudi's patches?
14:49 dbsx So I can reduce my 4 different kernel gits to one?
14:49 dbsx Thank you.
14:51 Marmotte rabeeh: I just made a debootstrap armhf with "--include=rsyslog,netbase,net-tools,ifupdown,isc-dhcp-client,bind9-host,openssh-server,xserver-xorg,xinit,awesome"
14:51 Marmotte after "dpkg --configure -a" and "apt-get autoclean", du says 176 Mo
14:54 Marmotte (for Debian Wheezy)
14:59 dbsx Marmotte: find /usr/share/doc/ -name '*' -type f -exec rm {} \; > /dev/null 2>&1
14:59 dbsx find /usr/share/man/ -name '*' -type f -exec rm {} \; > /dev/null 2>&1
14:59 dbsx find /usr/share/doc-base/ -name '*' -type f -exec rm {} \; > /dev/null 2>&1
14:59 dbsx find /var/cache/apt/ -name '*' -type f -exec rm {} \; > /dev/null 2>&1
14:59 dbsx Will trim about 30M
15:00 Marmotte probably :)
15:00 Marmotte I did that for my minimal ramdisk :)
15:02 Marmotte (a simple Debian system in a ramdisk that I load from tftp)
15:02 dbsx cool
15:03 Marmotte I made that first when I bought my Sheevaplug, and I reuse for the cubox :)
15:03 Marmotte (I use debirf to build the ramdisk)
15:06 rabeeh Marmotte: i can add debootstrap in CuBox installer
15:06 rabeeh or you can start from a netboot image
15:06 rabeeh (initrd netboot)
15:07 rabeeh personally i hate initrd
15:07 rabeeh i use initramfs since it bundles both the kernel and the filesystem.
15:07 rabeeh actually i have an 18MB xorg + xfreerdp + kernel and other goodies for tiny thin-client
15:08 rabeeh i'm looking now for tftp server through the internet
15:08 rabeeh so that you don't even need the microsd
15:09 dbsx 1. I hate netboot. I can send you a wheezy debootstrap script (tested on lots of cuboxes)
15:09 dbsx 2. Booting uImage from flash is hard to beat.
15:09 rabeeh dbsx, Marmotte: here is my ubuntu CuBox installer script -
15:09 rabeeh http://pastebin.com/BB6tCaEG
15:09 rabeeh i'm re-testing it back again.
15:10 rabeeh the final stage would be a chroot to apt-get install xubuntu-desktop
15:10 rabeeh dbsx: the only missing thing is that i'm using an oldie kernel from ubuntu-core (3.5.7)
15:11 rabeeh so probably i can have my 3.6.9 kernel with a bunch of kernel modules extracted on the rootfs after installation is done.
15:20 dbsx only comment on the script: add --overwrite to the tar
15:22 rabeeh why?
15:22 rabeeh i'm formatting the microsd
15:22 rabeeh remove_all_partitions() function
15:22 rabeeh and then
15:23 rabeeh create_one_partition ext4
15:23 dbsx paranoia, and habit
15:23 rabeeh ok. i will
15:34 dbsx rabeeh: I have watched 20 hours of video since 20130509 GeexBox. With and without passthru audio. If nothing else rudi;s patches are stable
15:34 rabeeh which build are you using?
15:35 dbsx The devel of 20130509 I can look up the link if you like
15:35 rabeeh no need
15:35 rabeeh performance wise; CEC is enabled by default which does tons of i2c polling
15:35 rabeeh just disable that and see how performance becomes even better
15:36 rabeeh i'm not saying that they are buggy; i'm saying no other developer reviewed them
15:36 rabeeh rudi has been doing great job supporting CuBox.
15:38 dbsx Not rapt about CEC. No lags in video or audio (except for 1 junk avi). It has CPU to spare.
15:38 dbsx Can one disable CEC without a rebuild? Do you know?
15:41 rabeeh yes
15:42 rabeeh setting
15:42 dbsx ok
15:42 rabeeh system --> input devices --> peripherals
15:43 rabeeh are you familiar with the 'graphics scaling' too?
15:43 rabeeh oh; yesterday i'v added rutorrent as default for every build
15:43 rabeeh i sent a patch for that (thanks to tomlohave for including rutorrent in openbricks)
15:44 rabeeh rutorrent is great since it searches, downloads links and easy to use web i.f. for rtorrent
15:44 rabeeh i hooked a usb drive, configured rtorrent (via the web) to download everything to that drive
15:44 dbsx I will test it
15:59 _rmk_ oh my, the marvell i2c driver in mainline seems rather broken.
15:59 _rmk_ the si clock driver uses smbus messages, which provokes this behaviour...
16:00 _rmk_ - start, write address + write, send a byte, restart, write address + write, restart, write address + read, *bus hang*
16:03 rabeeh i2c controller inside dove isn't something to be proud of
16:03 dbsx gnite all
16:04 frilled nite dbsx
16:04 rabeeh good night dbsx
16:12 _rmk_ rabeeh: neither is the driver
16:12 _rmk_ its racy as fsck
16:13 _rmk_ I've been noticing things going wrong on the cubox as I change stuff, and simply adding another printk shifts the code around enough to peturb the race
16:15 _rmk_ that i2c driver is already crappy enough that it uses an interruptible wait, which will abort any i2c transaction initiated from a process which uses signals
16:16 rabeeh _rmk_: there is some race condition internally that you need to usleep(1) from time to time
16:16 dv_ ouch
16:16 dv_ hello again btw
16:16 rabeeh i forgot where exactly
16:17 rabeeh that was long time ago
16:17 _rmk_ I'm just sending a mail to Mark Greer about his driver :)
16:18 dv_ oh _rmk_, in your opinion, does going from softfp to hardfp generally yield a significant gain?
16:19 dv_ I have heard conflicting opinions about this one. of course there are big gains in some special cases, but I mean overall, for example when building an entire distro either with soft- or hardfp
16:21 _rmk_ are you referring to the difference between armel and armhf?
16:21 rabeeh _rmk_: i think he is referring to -mfloat-abi=hardfp
16:22 rabeeh armel and armhf differs a lot (armel is armv5 and armhf is armv7; besides the floating point abi difference)
16:26 _rmk_ the big difference between armel and armhf is how the fp arguments get passed. in armel, it is through integer registers. armhf is through the vfp registers instead.
16:26 _rmk_ and some armv5 has vfp too so I don't buy your explanation
16:27 rabeeh _rmk_: first Orion chips from Marvell has armv5+vfp (i worked for that team)
16:27 rabeeh as i recall 88E5082?
16:28 rabeeh 88f5182
16:30 _rmk_ okay, the marvell i2c driver is devoid of maintainer now
16:30 _rmk_ no entry in MAINTAINERS and the address in the file doesn't work
16:33 _rmk 16:33 * _rmk_ sends the interruptible wait patch off to the i2c maintainers
16:33 _rmk_ ... and will continue to try and fix this race
16:46 _rmk 16:46 * _rmk_ ponders how to fix this... having a state machine running in interrupt context, and falling back to process context for the next bit of the message is insane and actually the cause of this problem - especially because the FSM gets in front of the process context when sending the restart+address
17:15 _rmk_ that fixed it :)
17:15 rabeeh :)
17:16 _rmk_ now I need to work out how to make it a little more bullet proof given this horrid FSM Mark has in here
17:39 _rmk 17:39 * _rmk_ tries this again; I think the TDA9981 had random registers scribbled to as a result of that bug
17:39 _rmk_ which made the colours all wrong
17:39 _rmk_ err TDA19988
17:39 _rmk_ either that or I've introduced a bug
18:39 _rmk_ fsck, I think that i2c driver crap has messed up my TDA19988
18:39 _rmk 18:39 * _rmk_ tries going back to the NXP driver for it
18:49 _rmk_ hmm, that solved it.
18:49 _rmk_ I wonder if its some register which the DRM TDA19988 driver isn't writing then
19:03 _rmk_ well, it _does_ work :)
19:04 _rmk_ I guess I just need to add "debug the drm tda19988 driver some more" to the list of things to look at :(
19:30 rabeeh what is the drm tda19988 thing?
19:30 rabeeh how it's related to drm?
20:09 _rmk_ rabeeh: yes, it's a driver for use only by DRM based drivers, because it's implemented as a drm slave-encoder driver
21:33 _rmk_ erm, can I ask a rather fundamental question about kirkwood-i2s.c
21:33 _rmk_ this fundamental question is...
21:33 _rmk_ what if your record and playback rates being asked of the driver are different?
21:33 _rmk_ it seems to me that the last one requested gets to set the rate on the entire i2s controller
21:34 _rmk_ (yes, I know the cubox doesn't support record tho.)
21:45 _rmk_ oh, hang on, the capture side is only ever clocked externally
21:45 jnettlet _rmk_, that is a bug we found in our mmp i2s controller. So it probably exists for the kirkwood also.
21:47 _rmk_ or is it... the func spec is being totally unclear on this
21:47 jnettlet we are currently forcing the rates at the hardware level to be the same and are letting software resample as needed.
21:48 rabeeh jnettlet: i'm not sure if it's the same i2s controller
21:49 _rmk_ ah yes, its the same clock, so what's it mumbling on about the record and playback rates being different and needing the dco ppm tweaked
21:50 jnettlet rabeeh, the i2s controller is not exactly the same, however the drivers are written in a very similar manner.
21:50 _rmk_ the LRCLK is shared for god sake, so there's utterly no way that the rates can't be different
21:50 rabeeh _rmk_: true
21:51 rabeeh i'm not sure about SPDIF in
21:51 _rmk_ so yes, jnettlet, this suffers from the same problem :)
21:51 _rmk_ it doesn't have spdif in
21:51 _rmk_ lets try an experiment...
21:52 _rmk_ oh, can't on the cubox because we thankfully don't advertise any capture device
21:52 _rmk_ so this bug remains hidden
21:54 _rmk_ jnettlet: did you push the fix upstream?
21:54 rabeeh right
21:54 rabeeh no spdif-in in this version
21:54 _rmk_ jnettlet: if so, which kernel, and which file?
21:54 jnettlet _rmk_, the fix was in a different but similar driver.
21:54 jnettlet let me dig up the git commit
21:56 rabeeh _rmk_: for i2s in it requires i2slrclk, i2sdi and i2sbclk signals (page 242 in the spec)
21:56 rabeeh so the clock for i2s in comes externally; you should be needing the dco for that
21:57 jnettlet _rmk_, hmmm. the fix that got pushed was in the codec. I guess they didn't take my fix. http://dev.laptop.org/git/olpc-kernel/commit/?h=arm-3.5&id=df89b8bd72dca70706253a2510ee9676200d64da
21:57 jnettlet I don't have my git repos handy as everything is packed away for an impending move.
21:58 _rmk_ I wonder how many suffer this bug actually
22:07 rabeeh _rmk_: you are right
22:07 _rmk_ oh, oh oh oh, I'm sooo going to have to fix that friggin tda19988 driver.
22:07 rabeeh the spec says that too - 'The I2S input and I2S output can only operate at the same sample rate'
22:07 rabeeh :)
22:08 _rmk_ I can't put up with blue being pink, and orange being green :(
22:08 _rmk_ rabeeh: yea, I'll add it to something to look at.
22:08 _rmk_ I'm soo going to have to push out some of these fixes
22:12 jnettlet _rmk_, don't know if you carried it over to your drm driver, but the framebuffer driver has a rbswap property to handle various LCD's
22:13 _rmk_ jnettlet: it doesn't quite look like red/blue swap, though I can try that by tweaking the muxes on the tda19988 right now
22:14 jnettlet _rmk_, it is a register setting on LCD_SPU_DMA_CTRL0
22:16 _rmk_ no, they're all swapped over
22:17 _rmk_ after the NXP driver has touched the TDA19988, I need to program the tda19988 muxes as reg 00:20 -> 01, 00:21 -> 23, 00:22 -> 45 (each nibble controls how a 4-bit lane of the RGB88 is mapped)
22:17 _rmk_ that's always worked before I had the "accident" with the i2c driver
22:18 _rmk_ now, it wants after power cycling... 00:20 -> 23, 00:21 -> 45, 00:22 -> 01
22:20 _rmk_ (what I can't remember is which way the mapping goes, whether the data decides on the RGB888 input bits or whether the register offset does)
22:22 _rmk_ right, each register nibble defines the input pins used
22:28 _rmk_ ok. time to make my head explode through reading the NXP driver...
22:31 _rmk_ *ARGH*
22:32 _rmk_ rabeeh: so, could you please tell me definitively how the TDA19988 is connected to the Armada 510 please, particularly how VP[ABC][0:7] are connected to LCD_D[23:0] ?
23:32 _rmk_ right, bank 00 reg 27 is the problem case
23:48 _rmk_ and that's... another mux in the video path
23:48 _rmk_ defaulting at power on to no translation (VIP output 23:0 maps to the next stage 23:0 on powerup)