**** BEGIN LOGGING AT Tue Dec 14 02:59:57 2010 Dec 14 04:29:39 google gstreamer-openmax Dec 14 08:57:07 morning Dec 14 10:11:09 Dan__: I used to; last cycle Ubuntu released OMAP4 images but the DSP-enabling stuff was mostly in a PPA https://launchpad.net/~tiomap-dev/+archive/release Dec 14 10:11:42 some of it might work on OMAP3, but I don't know how much; the official OMAP3 stuff should be downloaded from TI Dec 14 10:40:52 omap4 dsp is different then omap3 one Dec 14 13:10:54 lool: hi. what was the question regarding the OMAP4 PPA? Dec 14 13:17:04 ndec: 03:00 < Dan__> hey, does anyone know anything about getting DSP working in Ubuntu on a Beagleboard? Dec 14 13:17:07 03:00 < Dan__> specifically, dsp-bridge Dec 14 13:22:20 lool: thx... Dec 14 15:08:32 hi, I have a pandaboard now running ubuntu downloaded from http://omappedia.org/wiki/Ubuntu_Pre-built_Binaries_Guide Dec 14 15:08:45 I'd like to switch it to server mode (no X), any idea on how to do that? Dec 14 15:12:15 guerby: 2 options for maverick, removing the x related packages or creating a minimal rootfs with rootstock Dec 14 15:12:36 for natty we'll also have a minimal image available to install Dec 14 15:12:42 rsalveti, apt-get remove xorg ? Dec 14 15:14:08 xserver-xorg will remove most xorg related package Dec 14 15:14:13 *packages Dec 14 15:14:35 rsalveti, ok will try that. I don't see where in /etc it selects what to run at boot Dec 14 15:14:54 guerby: upstart, check at /etc/init/ Dec 14 15:15:32 rsalveti, thx Dec 14 15:33:05 guerby: how about adding 'text' in the bootargs. you just need to edit /boot/boot.script, add 'text', run sudo flash-kernel and reboot Dec 14 15:33:16 rsalveti: that should work, right, ^^^^ Dec 14 15:33:41 ndec, thx for the tip Dec 14 15:33:48 ndec: sure, but the graphics packages will be at the rootfs still Dec 14 15:34:08 guerby: : if it's not a problem for you, that will work fine Dec 14 15:34:14 rsalveti, having the packages around on rootfs is not an issue for me (some of my users do require them) Dec 14 15:34:22 rsalveti, ndec thx Dec 14 15:34:31 guerby: that should do everything except start gdm, which is likely what you want. not removing xorg-* will give you the ability to use is as a desktop later if needed. Dec 14 15:34:42 yup Dec 14 15:35:01 rsalveti, out of curiosity how text as kernel arg will impact upstart ? Dec 14 15:35:04 guerby: it's definitely recommended to use external USB drive too. i saw your comment on pandaboard room. Dec 14 15:35:22 guerby: the kernel is not going to use the argument, but it'll be used by ubuntu later on Dec 14 15:35:35 guerby: grep text /etc/init/* ;-) Dec 14 15:35:40 ndec, /home will probably be over NFS, but I could add an USB disk Dec 14 15:36:18 guerby: the USB disk will *drastically* improve overall performance. we have lots of perf issues when SD card is being read and written in //. Dec 14 15:36:39 guerby: see http://omappedia.org/wiki/Ubuntu_on_OMAP_FAQ#I_want_to_install_Ubuntu_on_external_USB_hard_disk_instead_of_sluggish_SD_card Dec 14 15:36:41 ndec, I had bad experiences with USB disks and sheevaplug, but may be pandaboard does USB handling and power better Dec 14 15:36:52 ndec, thx for the link Dec 14 15:37:09 guerby: I am using exclusively USB drive, as many others around here! Dec 14 15:38:18 ndec, ok I saw what gdm.conf is doing with /proc/cmdline :) Dec 14 15:38:35 guerby: yep ;-) Dec 14 15:39:17 ndec, other question: I saw a note that 1GB was not properly supported hence the 420+256, are newer kernels able to cope with 1GB ? Dec 14 15:40:12 guerby: not yet. note the memory hole is only if you plan to use TI multimedia firmware (so called ducati). e.g. for 1080p. if you don't use it, you can use the memory and don't need the hole Dec 14 15:41:03 jayabharath, hi, I got the power supply and the pandaboard is now running ubuntu :) soon it will be announced Dec 14 15:41:28 guerby: nice! Dec 14 15:42:18 ndec, I won't do any multimedia with my pandaboard, so I just have to remove "mem=460M@0x80000000 mem=256M@0xA0000000" from /boot/boot.script? Dec 14 15:42:58 guerby: yes, and replace it by mem=768M. until we fix the 1Gb issue. Dec 14 15:43:44 ndec: hi.. any ideas on when the 1Gb issue might be resolved.. dint hear much noise about it ... Dec 14 15:44:24 ndec, ok. this will not screw up "bootm 0x80000000 0x81600000" with just mem=768M ? Dec 14 15:44:27 jayabharath: nope. likely not on .35 kernel. Dec 14 15:44:44 ndec: ok Dec 14 15:45:14 guerby: no. this are 2 different things. the bootm is to load the uImage and uInitrd into DDR. Dec 14 15:45:23 ndec, ok thx Dec 14 15:45:41 ndec, I did play with that stuff a while ago (was on sheevaplug IIRC) Dec 14 15:45:46 ndec, but I wanted to be sure :) Dec 14 15:46:19 guerby: np. Dec 14 15:47:11 ndec, are you in France BTW? Dec 14 15:47:48 guerby: yes Dec 14 15:47:55 guerby: Nice, France Dec 14 15:47:58 guerby: you? Dec 14 15:48:16 ndec, I live in Castres but I often go to Toulouse and Paris (and I'll be at FOSDEM in Bruxelles :) Dec 14 15:48:33 ndec, the pandaboard is for http://gcc.gnu.org/wiki/CompileFarm Dec 14 15:48:54 guerby: who are you working for? Dec 14 15:49:17 guerby: oh, gcc build farm. cool! Dec 14 15:49:24 ndec, I have my own company now, I do consulting in finance & software Dec 14 15:49:35 guerby: thx Dec 14 15:49:46 ndec, but I do spend quite some time on not-for-profit stuff including the farm Dec 14 15:50:28 cool Dec 14 15:52:34 ndec, rsalveti the GCC name is historical, the farm is open to all free software (GPL, BSD, MIT ...) Dec 14 15:52:54 that's nice Dec 14 15:53:08 and yes, if you want to build on panda a usb-drive will help a *lot* Dec 14 15:53:50 guerby: how about this: https://blueprints.launchpad.net/ubuntu/+spec/other-arm-n-public-panda-ppa-build-cluster ;-) Dec 14 15:53:59 rsalveti, I think I'll try /home over NFS first it's a bit less cable :) Dec 14 15:54:15 :-) Dec 14 15:54:53 ndec, most distro already have their farm for various arch, I know some are using the GCC farm to build packages though Dec 14 15:56:01 I have to test nfsroot for my panda one day Dec 14 15:56:13 should be comparable to usbhdd Dec 14 15:56:20 when it comes to speed Dec 14 15:57:24 hrw, I had one sheevaplug on NFS root, it did work fine (the NFS server had 8GB of RAM that helped :) Dec 14 15:57:46 94Mbps network speed (iperf) Dec 14 15:57:49 here I'll just put /home that's where most I/O will be done (may be /tmp too) Dec 14 15:58:04 guerby: my sheevaplug has system in nand Dec 14 15:58:09 hrw, sheeva is gigabit IIRC Dec 14 15:58:15 guerby: it is Dec 14 15:58:23 guerby: and gets ~800Mbps Dec 14 15:58:39 I didn't check pandaboard, looks like usb eth so likely 100 Mbps Dec 14 15:59:05 guerby: I said: 94Mbps Dec 14 15:59:29 hrw, ok :) Dec 14 15:59:41 guerby: usb hdd got 12.5MB/s only Dec 14 16:03:19 bbl Dec 14 16:19:28 hrw: check out http://omappedia.org/wiki/Panda_Test_Data -- looks like USB 'flash' drives can expect about 20Mb/s write and 41Mbs read speed.. Dec 14 16:22:31 jayabharath: I hope that it is 20MB/s not 20Mb/s Dec 14 16:22:39 :D Dec 14 16:22:49 hrw: right Dec 14 16:26:29 write test assumes that card is mounted with sync Dec 14 16:28:26 19.3MB/s read, 3.6MB/s write my class10 card has Dec 14 16:30:03 harddrive: 16.7MB/s write, 20.3MB/s read Dec 14 17:32:21 rsalveti, got http://pastebin.com/qTTDuYDu at the end of pt-get upgrade Dec 14 17:32:34 rsalveti, is this a known issue? Dec 14 17:37:16 guerby: unfortunately yes, this should be ok once you update your kernel and reboot Dec 14 17:37:23 then the alsa-utils package will be upgraded normally Dec 14 17:39:28 rsalveti, ok I'll reboot :) Dec 14 18:24:48 rsalveti, ok upgrade did work after reboot :) Dec 14 18:25:00 rsalveti, something funny it looks like the MAC address has changed ... Dec 14 18:25:00 cool :-) Dec 14 18:27:04 rsalveti, well yes MAC changed from 32:57:F8:93:E1:CD to d2:92:9c:66:81:be Dec 14 18:27:09 not good for DHCP Dec 14 18:27:41 guerby: yeah, we don't have a valid eeprom at ths usb chip Dec 14 18:27:58 so it'll be automatically generated on every boot Dec 14 18:28:15 unless you set your own at the network config or at the kernel cmdline Dec 14 18:28:36 rsalveti, what's the file to edit to fix the MAC? Dec 14 18:29:52 rsalveti, adding text to cmdline did work to get rid of graphical stuff Dec 14 18:31:04 guerby: /etc/network/interface Dec 14 18:31:12 check bug 673504 Dec 14 18:31:14 Launchpad bug 673504 in linux-ti-omap4 (Ubuntu Natty) (and 2 other projects) "Pandaboard chooses a new IP address on each boot (affects: 2) (heat: 20)" [High,Fix committed] https://launchpad.net/bugs/673504 Dec 14 18:31:25 https://bugs.launchpad.net/ubuntu/+source/linux-ti-omap4/+bug/673504 Dec 14 18:31:29 rsalveti, :) Dec 14 18:31:29 Launchpad bug 673504 in linux-ti-omap4 (Ubuntu Natty) (and 2 other projects) "Pandaboard chooses a new IP address on each boot (affects: 2) (heat: 20)" [High,Fix committed] Dec 14 18:33:01 "This bug was fixed in the package linux-ti-omap4 - 2.6.35-903.19" Dec 14 18:33:09 # cat /proc/version_signature Dec 14 18:33:09 Ubuntu 2.6.35-903.19-omap4 2.6.35.3 Dec 14 18:33:21 well let's see another reboot Dec 14 18:33:54 ah no that's not the same issue Dec 14 18:34:00 I'll use the kernel param then Dec 14 18:34:47 guerby: not the same issue, but shows how you can set up your own mac Dec 14 18:35:02 you'll still get another mac on every reboot, unless you set it up manually Dec 14 18:35:27 this bug was related with the mac regeneration on every ifup/ifdown Dec 14 18:35:30 even worse :-) Dec 14 18:37:42 rsalveti, cmdline worked :) Dec 14 18:39:52 now checking mem=786M Dec 14 18:40:29 Mem: 734676k total, Dec 14 18:40:47 seems to work too :) now NFS then GCC bootstrap Dec 14 18:45:19 apt-get install nfs-kernel-server Dec 14 18:45:43 cool Dec 14 19:07:27 Someone uses BeagleBoard? Dec 14 19:16:18 NFS /home working Dec 14 19:16:28 150MB rsync running Dec 14 19:53:22 Hmm, anyone have recommendations for a good arm board for use as a nas? Dec 14 19:55:01 I'd want: gigabit Ethernet (perhaps dual), more than one SATA (esata is okay) port, and armv7l (so no Kirkwood)? Dec 14 19:55:36 Er, that should ne a statement, not a question. Dec 14 20:03:46 rsalveti: do you still have a panda board on your desk ? Dec 14 20:03:54 mpoirier: yes Dec 14 20:04:04 rsalveti: powered and running ? Dec 14 20:04:10 yes Dec 14 20:04:16 cool Dec 14 20:04:52 rsalveti: I got a question from nhg, he need to find the cpu speed in MHZ . Dec 14 20:05:11 I thought it would be under /proc/cpuinfo but apparently not. Dec 14 20:05:32 it shows only the bogomips Dec 14 20:05:56 yeah...how to find the actual Ghz numbers? Dec 14 20:06:13 rsalveti: meet nhg Dec 14 20:06:37 "cat /proc/cpuinfo | grep MHz" ?? Dec 14 20:06:48 the uboot I'm using also doesn't seem to print it Dec 14 20:07:05 rsalveti: could you think of a place where the info could be found ? Dec 14 20:07:08 How about just finding the model number and google away? Dec 14 20:07:33 cat /proc/cpuinfo gives only bogompis Dec 14 20:07:50 mpoirier: nhg: also don't know about it, try asking at #pandaboard Dec 14 20:08:26 cpufreq-info Dec 14 20:09:14 From "cpufrequtils" Dec 14 20:10:30 comradekingu1: yes that seems to be a valid option. Dec 14 20:10:44 That didnt work on my machine becase configuration moved to /etc/modprobe.d/, but your milage may vary. Could be fixed though Dec 14 20:11:03 nhg: you should try ^^^^ Dec 14 20:11:06 not mine either Dec 14 20:11:31 nhg: do you have a network connection ? Dec 14 20:11:48 on the panda board that is... Dec 14 20:12:21 WLAN Dec 14 20:12:30 yes Dec 14 20:12:43 no ethernet Dec 14 20:12:43 then: sudo apt-get install cpufrequtils Dec 14 20:12:52 doesn't work on panda Dec 14 20:12:58 Say, does panda now support cpufreq? Dec 14 20:13:08 it doesn't Dec 14 20:13:23 at least not at ubuntu Dec 14 20:13:36 nhg: go to #pandaboard then... Dec 14 20:14:12 asked the question on pandaboard Dec 14 20:19:33 Argh, ubuntu has supported Dec 14 20:20:16 marvell dove boards for like 1 or 2 years, but you still can't buy any anywhere! Dec 14 20:24:14 rsalveti, hmmm not good: http://pastebin.com/enX6nAbZ Dec 14 20:24:50 rsalveti, may be mem=768M is the culprit? Dec 14 20:25:50 guerby: this is not a kernel panic, it's just that you don't have more memory left Dec 14 20:26:14 rsalveti, no swap happened Dec 14 20:26:22 Mem: 734676k total, 679100k used, 55576k free, 12708k buffers Dec 14 20:26:22 Swap: 524284k total, 0k used, 524284k free, 491024k cached Dec 14 20:26:57 rsalveti, rather surpising for ENOMEM situation, 491MB in cache Dec 14 20:27:38 hm, true Dec 14 20:28:14 rsalveti, I will reboot with the original mem= to see if it changes something Dec 14 20:28:31 don't think it will help much Dec 14 20:29:06 with 1gb the instability we got was basically a random kernel panic Dec 14 20:29:06 rsalveti, this was just a time svn co -q svn://gcc.gnu.org/svn/gcc/trunk Dec 14 20:29:48 (with /home over NFS) Dec 14 20:32:51 hmm I keep getting kswapd0: page allocation failure even with svn being 30MB Dec 14 20:33:34 may be I'll just deactivate swap Dec 14 20:34:17 swapoff -a done Dec 14 20:35:56 meanwhile I'm reading to try to understand why you could get it Dec 14 20:42:19 at least it's harmless Dec 14 20:43:28 rsalveti, the first svn failed so it had some impact. No more errors since swapoff, the second svn continues Dec 14 20:44:34 guerby: it failed? but the kernel will probably try to alloc again after failing to find more memory Dec 14 20:44:56 weird is that the swap itself was unused Dec 14 20:47:11 rsalveti, my full /var/log/messages is here: wget http://guerby.org/ftp/messages-pandaboard Dec 14 20:48:16 rsalveti, I got plenty of failures I didn't notice before my svn Dec 14 20:48:58 Hmm, anything interesting for ARM coming out early 2011? Dec 14 20:49:05 rsalveti, I got new ones even with swapoff Dec 14 20:49:26 Most of the new stuff talked about this year seems to not actually be available. Dec 14 20:49:36 Like, Marvell Armada. Dec 14 20:49:37 ok, so not related with swap Dec 14 20:50:01 Available as in not-ungodly-expensive boards. Dec 14 20:59:43 seems that the usb driver is the hungry one Dec 14 21:00:09 the ethernet chip and the usb host is the same chip Dec 14 21:00:13 connected by usb Dec 14 21:00:57 rsalveti, hmmm Dec 14 21:02:56 rsalveti, I just rebooted with no swap and original mem= Dec 14 21:03:15 don't think it'll help, but it's a good try Dec 14 21:03:26 rsalveti, this time I'm monitoring /var/log/messages Dec 14 21:03:55 rsalveti, only thing specific is that I have two openvpn running so tun driver is there too (if it helps) Dec 14 21:21:28 rsalveti, I launched a GCC bootstrap, it failed quickly with a tool reporting "out of memory allocating 4072 bytes after a total of 199240912 bytes" Dec 14 21:22:06 guerby: can you try that with 768M and swap? Dec 14 21:22:30 rsalveti, I'm removing a ulimit from my script, then I'll try with 768M and swap Dec 14 21:22:50 one thing is the userspace application having problems to allocate memory, another one is the driver itself failing to find the suitable page Dec 14 21:23:05 in our case I believe that the driver is the hungry one Dec 14 21:25:39 rsalveti, nothing in /var/log/messages so far Dec 14 21:26:32 but alloc of 200MB shouldn't fail, at least it doesn't fail on a sheevaplug with 512MB RAM and ulimit 500MB Dec 14 21:27:21 guerby: but were you using swap this time? Dec 14 21:27:53 rsalveti, swapoff when it happened (I commented out the swap line in fstab before reboot) Dec 14 21:28:03 ok Dec 14 21:28:38 guerby: have to leave now for dinner, otherwise the restaurant will close, can you please open one bug for it? Dec 14 21:29:10 while I believe it should not harm you, it'd be good to understand why ehci is requesting so much memory Dec 14 21:29:45 and if this is expected, it should at least set __GFP_NOWARN, like we have with some other drivers Dec 14 21:31:39 rsalveti, ok good dinner :) Dec 14 21:31:40 Mem: 681844k total, 368152k used, 313692k free, 6076k buffers Dec 14 21:31:41 Swap: 0k total, 0k used, 0k free, 240456k cached Dec 14 21:31:54 it happened again and I was looking at top ^ Dec 14 21:32:02 guerby: open a bug against https://launchpad.net/ubuntu/maverick/+source/linux-ti-omap4 Dec 14 21:32:10 no way a 200MB program can fail malloc here Dec 14 21:32:14 rsalveti, will do Dec 14 21:32:23 guerby: the process failed again? Dec 14 21:32:30 rsalveti, yep Dec 14 21:32:33 same place Dec 14 21:32:42 this time I just removed the ulimit Dec 14 21:32:55 interesting Dec 14 21:33:07 well, be back later Dec 14 21:38:50 rsalveti, https://bugs.launchpad.net/ubuntu/+source/linux-ti-omap4/+bug/690370 Dec 14 21:38:53 Launchpad bug 690370 in linux-ti-omap4 (Ubuntu) "Strange out of memory on pandaboard (affects: 1) (heat: 6)" [Undecided,New] Dec 14 21:40:31 ndec, hi again :) looks like I'm pushing a bit too much my pandaboard :) Dec 14 21:41:00 guerby: what happened? Dec 14 21:41:24 ndec, I put some info in launchpad Dec 14 21:41:35 ^ Dec 14 21:47:06 guerby: so there is no swap, right? have you tried with swap? Dec 14 21:48:04 ndec, I will try but I strongly doubt it's the issue, there was more than 400MB free when the alloc failed Dec 14 21:48:22 ndec, and the same bootstrap works on a sheevaplug 512MB RAM and ulimit 500MB Dec 14 21:48:51 guerby: i don't remember if highmem is enabled in the kernel. is it? Dec 14 21:49:22 ndec, first try was with orig mem=x mem=y, I'm now trying with mem=768M Dec 14 21:49:28 ndec, how do I know highmem ? Dec 14 21:49:42 guerby: well, i think it is. grep HIGHMEM /boot/config-XXXX Dec 14 21:50:04 guerby: i am running a custom kernel, but I think i still have the 10.10 kernel, and it seems to be enabled. Dec 14 21:50:09 ndec, CONFIG_HIGHMEM=y Dec 14 21:50:24 ndec, running Ubuntu 2.6.35-903.19-omap4 2.6.35.3 Dec 14 21:50:31 guerby: we started to see such random mem issues when we enabled highmem. Dec 14 21:50:37 rsalveti: ^^^ Dec 14 21:50:53 ndec, rsalveti is away for dinner Dec 14 21:51:12 guerby: thx Dec 14 21:52:17 guerby: do you know how to recompile a kernel a-la ubuntu? Dec 14 21:52:27 guerby: it would be good to try your test without highmeme Dec 14 21:52:31 highmem Dec 14 21:53:14 ndec, I'll learn :) issue is that I probably won't have time to play more before next week Dec 14 21:53:30 guerby: np. i think that should help: http://omappedia.org/wiki/Ubuntu_kernel_for_OMAP4 Dec 14 21:53:37 guerby: i will be out next 2 weeks ;-) Dec 14 21:54:35 ndec, interesting I got past the failure point with mem=768M Dec 14 21:54:44 ndec, may be memory with hole is the issue Dec 14 22:22:26 hmm make bootstrap failed a bit later on "virtual memory exhausted: Cannot allocate memory", nothing in /var/log/messages Dec 14 22:22:33 let's try again with swap on Dec 14 22:29:43 ndec, with mem=768M and no swap make bootstrap failed a bit later on "virtual memory exhausted: Cannot allocate memory", nothing in /var/log/messages Dec 14 22:29:50 ndec, I'm trying with swap now Dec 14 22:51:15 gn Dec 14 22:52:27 guerby: If you have networking available on that system, could you run "apport-bug -p linux-ti-omap -u 690370" when you get a chance? This way I can mark the bug triaged and the kernel team won't autoclose it as invalid. Dec 14 22:54:39 GrueMaster, I get stuck in the launchpad login service, it doesn't seem to work in text mode Dec 14 22:55:13 You don't have keyboard/monitor/X? Dec 14 22:55:22 GrueMaster, no I deactivated all that stuff Dec 14 22:56:15 GrueMaster, there's no way to do it CLI ? Dec 14 22:56:18 Ah. I think you can run apport-cli and it will create a .crash file that can be copied to another system. Dec 14 22:56:22 Let me check. Dec 14 22:57:13 Use --save= to save the bug report. It can then be transfered to another system and uploaded from there. Dec 14 22:57:43 Send me the .crash file and I can patch it into the bug report if you like. Dec 14 22:58:58 GrueMaster, # apport-bug --save=toto Dec 14 22:58:58 No pending crash reports. Try --help for more information. Dec 14 23:00:38 Oh. Dec 14 23:00:48 GrueMaster, rsalveti will look into it, I need to go to bed now :) Dec 14 23:00:58 apport-cli --save= -p linux-ti-omap Dec 14 23:01:11 He's on my team. I'm the QA guy. Dec 14 23:01:26 No problem, we can get the data tomorrow. Dec 14 23:14:22 I'm currently working on some manual recompilation of Qt with SGX enabled, and looking at this rather odd source package from TI, I wondered if there are any efforts from the community. I see rsalveti's name pop up in respect of devmem2 on lauchpad Dec 14 23:16:29 A side question: Can somehow apt-get source a armel pkg. from my (non armel) host? Dec 14 23:16:35 I wonder: would something higher than 768, but lower than 1024, work? Dec 14 23:22:17 sveinse: The source for armel packages *should* be the same as every other arch (i.e, identical). Dec 14 23:23:23 I believe all source packages are at archive.ubuntu.com and only armel (and other less-popular hardware) have binaries on ports.ubuntu.com. Dec 14 23:24:28 I need to take a look at some specific sources, so the binaries doesnt interest me as much right now Dec 14 23:25:05 right. Just saying that all binaries come from the same source. Dec 14 23:25:07 Since I'm lazy and apt-get n00b, I just chrooted to my armel target and ran apt-get sources from there... Inefficient, but it gets the job done Dec 14 23:25:16 regardles of arch. Dec 14 23:26:06 Doubt this package (powervr-omap3) has any interest outside of omap3/armel though Dec 14 23:26:26 But I get your point that all sources come from the pool Dec 14 23:26:32 Probably not. If it is in the ppa, that's a different matter. Dec 14 23:27:24 apt-cache show will give you a full link to the source as well. Dec 14 23:28:26 ah. Excellent Dec 14 23:28:28 thanks Dec 14 23:28:59 rsalveti: you around? Dec 14 23:29:54 I think he's feeding. Dec 14 23:43:12 well it's night for me, so I'll catch up with him tomorrow Dec 15 00:04:34 * rsalveti back Dec 15 00:06:24 sveinse: you can just recompile qt with mesa, using gles and then use the sgx libraries Dec 15 00:07:04 the current sgx package doesn't provide the headers needed to build packages, because of the lack of a proper soname Dec 15 00:07:37 so if you plan to build and use anything with gles, just do it with the mesa libraries and then install the sgx when you want to test with it Dec 15 00:15:12 sooo annoying though :) Dec 15 00:15:35 guess crosscompiling would help though Dec 15 00:18:10 * sveinse back Dec 15 00:19:07 OK. here I am building both the SGX package and qt from sources, both cross compiled Dec 15 00:20:45 rsalveti: what is mesa, and how does it fit with Qt. I mean is it any different from compiling Qt and SGX together. (I have very little knowledge of what is inbetween these two sets of software) Dec 15 00:21:12 sveinse: why are you cross compiling the sgx package? Dec 15 00:21:16 it's just binaries Dec 15 00:21:33 Partly history Dec 15 00:21:37 the only thing to compile there is the kernel modules, and that doesn't help your qt build Dec 15 00:21:58 sveinse: mesa is the software implementation of gl and gles Dec 15 00:22:22 sgx should just replace it fine, but we don't have a proper soname at the sgx libraries Dec 15 00:22:32 We're building a target system of Ubuntu Maverick. Since we're this early in the development, we haven't had time to setup proper .deb package generation Dec 15 00:23:17 So we have some software being compiled, custom kernel (for custom board), custom Qt (since we're not using X11, but want GLES), and so on Dec 15 00:23:27 oh, ok Dec 15 00:23:41 And. we want to cross compile, since not able to do so make our build server useless Dec 15 00:24:34 Plan is though, that we migrate over to standard debs later on (since we want to be able to apt-get update our system). Dec 15 00:24:55 ok, now I get it Dec 15 00:24:58 Using the powervr-omap3-dkms is something I want to do Dec 15 00:26:01 FYI: I have been able to make a system for cross building apps into a ubuntu target system without too many worries. Dec 15 00:26:14 Thanks to the armel-cross package! Dec 15 00:26:27 cool Dec 15 00:26:49 Have to say I'm a little proud of it, but noone here really understands that ;) Dec 15 00:27:43 Coming back to the issue here: I'm looking at compiling SGX (or repackageing it, since its mostly bins) and compiling Qt with support for it Dec 15 00:27:43 :-) Dec 15 00:28:04 You say I can compile Qt with generic mesa support, you say? Dec 15 00:28:11 ok, so the only thing currently missing are the khronos headers Dec 15 00:28:39 you can build with the sgx libraries, if you grab the correct khronos headers, that you can already find at the mesa-dev packages Dec 15 00:28:57 I believe it'll be easier for you now if you just build agains the sgx libraries Dec 15 00:29:06 as you're not worried about proper packaging atm Dec 15 00:29:31 atm, yes. But it sounds like this will backfire later... Dec 15 00:30:01 the only problem is that the lack of a proper soname makes it incompatible with the mesa packages Dec 15 00:30:34 we, as a distro, cares about that because we can't be sure that the user will install the drivers, and that the board will contain a sgx chip Dec 15 00:30:44 I remember something about some headers while compiling Qt some while ago during the concept test... Caused by the missing Khronos headers, right Dec 15 00:30:51 so we need the generic implementation to work Dec 15 00:31:07 Ah, since they are hardcoded in the SGX package Dec 15 00:31:35 yes, so currently on ubuntu if we build something against mesa, it works with the sgx package Dec 15 00:31:43 but not the opposite Dec 15 00:32:07 so if you only care about using your software on omap with sgx installed, it's fine for you to just build agains the sgx packages Dec 15 00:32:23 yes, I do, thanks Dec 15 00:32:25 the only thing you'll need atm is the proper header files, as we currently don't deploy it Dec 15 00:33:22 Is the missing headers a bug/glitch, or is it an NDA thing? Dec 15 00:34:38 it's just because we don't want people building packages with the sgx one at ubuntu in general Dec 15 00:34:52 to avoid bugs with the lack of a proper soname Dec 15 00:35:08 but you can find the headers at the mesa package Dec 15 00:35:12 let me look it for you Dec 15 00:35:39 I admit I have to check google what mesa is... Dec 15 00:36:17 sveinse: it's an opensource implementation of gl and gles Dec 15 00:36:29 and you can run gl/gles by software if you need Dec 15 00:36:38 even on your desktop you can now run gles applications with mesa Dec 15 00:36:52 you can also have drivers to accelerate it Dec 15 00:38:05 Interesting thing though. The first word in OpenGL is open.... Dec 15 00:39:25 sveinse: but an open implementation :-) Dec 15 00:39:28 sveinse: check http://ports.ubuntu.com/ubuntu-ports/pool/main/m/mesa/ Dec 15 00:39:41 packages: libegl1-mesa-dev libgles1-mesa-dev libgles2-mesa-dev Dec 15 00:39:54 thanks Dec 15 00:39:56 there you'll find all the needed headers to build for gles Dec 15 00:40:11 or just grab the mesa source and look for the files Dec 15 00:40:25 sveinse: http://paste.ubuntu.com/543851/ Dec 15 00:41:14 * sveinse is still exited to see what effect GL has on the speed and resposiveness of his Qt app... Dec 15 00:43:13 If only Ubuntu had Qt compiled for qws without X11, I wouldn't have to recompile Qt Dec 15 00:43:47 soon we'll have arm public ppas and you'll be able to create a ppa for that :-) Dec 15 00:44:27 sorry, ppa? Dec 15 00:45:14 sveinse: If only Qt upstream provided a supported way to do that without having to maintain two complete copies of Qt. Dec 15 00:46:51 Yeah, I see that. To my knowledge it is only libQtGui.so which is bound to the gfx-system, and the rest isnt. But you still need to make that distinction while configuring Dec 15 00:47:08 sveinse: ppa is one way to have a public repository at the launchpad, so you can submit your packages, launchpad will build and make them available for you Dec 15 00:47:18 then you can use the repository entry at your sources.list Dec 15 00:48:04 ScottK: that would be nice to see, but don't think they will put resources on it Dec 15 00:49:19 rsalveti: That's the reason there's no qt-qws in the archive. It's a lot of maintenance work/buildd time to double it. Dec 15 00:49:21 Having worked with setting up a proper build system for cross compiled Qt apps, I have a handful of comment about the Qt build system.... Dec 15 00:49:37 Probably none of the printable in public .... Dec 15 00:51:00 Well (without starting another fight) its similar to Mac's: If you fit inside the box Apple and/or Nokia has prepared for you it's great! If you dont: Well then the box become Pandora's box... Dec 15 00:53:51 rsalveti: FYI: You know that TI has released a 4.00.00.01 version of their SGX? Dec 15 00:54:58 sveinse: yes, but afaik the only thing they changed was the support for one additional chip Dec 15 00:55:10 that's why we didn't update the package Dec 15 00:55:18 great, thought you should know Dec 15 00:55:25 thanks anyway :-) Dec 15 00:56:42 I think I will use the powervr dkms driver for the kernel part, and my custom Qt+SGX, since I need to custom build Qt anyways Dec 15 00:57:00 Did you come to a solution in regards of the devmem2 thing? Dec 15 00:58:43 Seems like I have to make something in upstart. The SGX driver wants to do some magic during bootup Dec 15 01:00:08 Magic seems to be pvrsrvinit, of which is a precompiled binary thing Dec 15 01:01:47 Well, I'm really going to bed this time Dec 15 01:01:52 rsalveti: Thanks a lot for you assistance Dec 15 01:01:56 Appreciate it! Dec 15 01:02:26 sveinse: yes, check the upstart file from our sgx package Dec 15 01:02:33 you'll see the hack heh :-) Dec 15 01:02:54 you first need to identify the omap 3 version and then use the proper pvrsrvinit Dec 15 01:03:12 np, ping me later if you have issues with it Dec 15 01:03:24 The powervr-omap3 package? Dec 15 01:04:19 sveinse: opengles-sgx-omap3 Dec 15 01:05:37 Hm.. Not available from my source (http://ports.ubuntu.com/ubuntu-ports maverick main universe) Dec 15 01:06:42 try multiverse Dec 15 01:06:57 pool/multiverse/o/opengles-sgx-omap3/opengles-sgx-omap3_3.01.00.07-0ubuntu2_armel.deb Dec 15 01:07:38 got it Dec 15 01:09:28 Well, thanks. Going to look at it in the morning. Night Dec 15 01:09:36 later **** ENDING LOGGING AT Wed Dec 15 02:59:57 2010