**** BEGIN LOGGING AT Sun Apr 18 02:59:57 2010 Apr 18 04:15:28 nice... the kernel on my beagleboard just segfaulted. Apr 18 04:17:37 * DanaG wishes there were a board that had a superset of "openrd-base" and "openrd-client". Apr 18 04:17:49 Base doesn't have many USB ports, but client doesn't have PCIe slot. Apr 18 04:18:45 What we need is some good way to share open board layouts, and a contract assembly plant with quick turnaround. Apr 18 04:21:44 The openrd-client would be perfect if not for that stupid onboard XGI. Apr 18 04:22:36 With a PCIe slot, instead, we could stick an ATI card in there and have arm eyefinity. =þ Apr 18 04:27:54 Or just add a USB hub to openrd-base Apr 18 04:36:45 Does the client use a hub, or 4 real separate ports? I'm not quite sure how that works. Apr 18 04:37:10 Even on an ordinary x86 system... is there any difference between 4 ports off the root, versus 4 ports on a hub off one root port? Apr 18 04:38:49 Base also doesn't have audio, rs485 (not that I'd know what to do with it), or the second GbE connector. Apr 18 04:40:02 Audio is trivial: use USB. Apr 18 04:40:25 I'm not sure of the implementation, but I suspect it's an internal hub, rather than real separate ports. Apr 18 04:40:40 Except for drives, this doesn't typically matter all that much. Apr 18 04:41:25 hmm, I wonder which would have better quality... onboard, or USB? I doubt many SOCs have 24/96 abilities. =þ Apr 18 04:41:29 s/many/any/ Apr 18 04:44:39 hmm, looking at the block diagram, it does look like merely an onboard hub. Apr 18 04:45:21 http://www.globalscaletechnologies.com/t-openrdbdetails.aspx#hw_block Apr 18 04:45:21 http://www.globalscaletechnologies.com/t-openrdcdetails.aspx#hw_block Apr 18 04:46:11 That's common. It's a cheap component, and adds perceived value to a board if there's leftover space. Apr 18 04:46:39 also weird... base block-diagram shows only esata... yet manual shows an onboard sata as well. Apr 18 04:46:41 Real ports typically have to be in the SoC directly, so it's rare to not get the same number on every board for a given SoC. Apr 18 04:47:02 (well, some boards trim them, but usually at least provide headers when they do) Apr 18 04:47:04 The real bummer is the loss of that second GbE. Apr 18 04:47:23 "onboard sata" might be a USB-SATA interface on the board. Apr 18 04:47:56 * persia has heard of that on a few boards now Apr 18 04:48:03 nope, the block diagram in the openrd-base manual (pdf) shows sata-#1 internal and sata-#2 esata. Apr 18 04:49:17 That's just branding. The difference between "SATA" and "eSATA" is the plug. Apr 18 04:49:27 heh, every time I see "OpenOCD", I think "Open Obsessive-Compulsive Disorder" Apr 18 04:49:55 Does it have a header, if it doesn't have a plug? Apr 18 04:50:10 openrd-client block diagram on web site: 2 sata connectors (one internal, one external). openrd-base on web page: 1 sata. OpenRD-Base PDF: 2 sata. Apr 18 04:50:51 Hurrah for communication between the tech writers and the marketing dept.! Apr 18 04:50:56 Oh, and a handy hint for win7: there are built-in "RNDIS-Compatible" drivers that work perfectly... and are signed. Apr 18 04:51:14 Just have to uncheck "show only compatible drivers". Apr 18 04:52:11 It's likely they revised the product at some point, and forgot to update the web site. Apr 18 04:58:29 Revision log in the PDF shows only one version... so I wonder which is the current state: one port, or two? Apr 18 05:16:23 NetworkManager: wireless_get_range(): (wlan0): couldn't get driver range information (95). Apr 18 05:16:24 NetworkManager: constructor(): (wlan0): Device unsupported, ignoring. Apr 18 05:20:30 grr, stupid networkmanager, Apr 18 05:20:31 . Apr 18 05:35:42 wlan0 Interface doesn't support scanning. Apr 18 05:35:46 nice job, kernel. Apr 18 05:42:18 /boot/config-2.6.34-rc4-l0:# CONFIG_CFG80211_WEXT is not set Apr 18 05:42:23 /boot/config-2.6.33-500-omap:# CONFIG_CFG80211_WEXT is not set Apr 18 05:52:32 hmm, the omap kernels seem all screwed up to various degrees. **** BEGIN LOGGING AT Sun Apr 18 07:53:18 2010 Apr 18 12:27:44 hi all. I'm thinking of using ubuntu on a Moxa box - is this likely to function well? Apr 18 12:40:37 Which Moxa box? (model #) Apr 18 12:41:29 erm - don't remember - I just found out ubuntu supports arm... it'll be a new model. don't they all use arm7? Apr 18 12:43:01 nope.. always look closely, it looks like (moxa.com) is an xscale house which is old armv5 hardware, which would mean Jaunty only. (bring your own kernel off course) Apr 18 13:28:17 asac: ping? Apr 18 16:37:08 hi have a arm related but non ubuntu related question Apr 18 16:37:24 can someone confirm that "ldr , = " is a valid opcode for gas ? Apr 18 18:10:42 DanaG, i just enabled CONFIG_CFG80211_WEXT it'll be on my 2.6.33.2-dl7 uploads.. Apr 18 18:10:53 Cool. Apr 18 18:11:05 Oh yeah, which time-zone are you in? Apr 18 18:11:28 us central.. Apr 18 18:11:49 ah. Apr 18 18:31:01 Cool, thanks. Apr 18 18:31:39 Do you have a kernel somewhere that's compiled with musb_hdrc built-in and yet all the g_ether and such built as modules? Apr 18 18:38:23 kinda... I have too many Bx boards, so i don't build that way: but the config diff is located here: http://bazaar.launchpad.net/~beagleboard-kernel/%2Bjunk/2.6-stable/annotate/head:/patches/rcn/002-defconfig-disable_gadget_stuff.diff just use that to change patches/lucid-defconfig in my 2.6-stable repo Apr 18 18:41:28 andruk, it seems to work: /dev/spidev3.0 /dev/spidev3.1 /dev/spidev4.0 Apr 18 19:09:42 * XorA|gone wonders why so many segfaults :-( Apr 18 19:16:17 rcn-ee: is there any kernel config stuff you can think of that would make apt/dpkg segfault a lot? Apr 18 19:16:33 rcn-ee: Angstrong is rock solid on this kernel Apr 18 19:16:59 yeap.. that sounds like the thumb erata.. Apr 18 19:17:24 you guys are using thumb? Apr 18 19:17:39 XorA|gone: To test, does karmic run clean, and just lucid segfault (karmic was -marm) Apr 18 19:17:50 make sure: CONFIG_ARM_ERRATA_430973=y is set... lucid is thumb2... Apr 18 19:18:12 all r1pX cores such as the omap35x need it.. Apr 18 19:18:49 rcn-ee: ah cool, Angstrom ditched thumb as it was too unreliable with a lot of gcc/binutils mixes Apr 18 19:19:28 One of the luxuries of being able to specify the use of the native internal toolchain :) Apr 18 19:19:32 yeap, i know what you mean.. i started with angstrom too.. the ubuntu guys wanted to try it. ;) Apr 18 19:19:59 It's (mostly) working for us, after we ported a heap of apps. Apr 18 19:20:06 persia: that doesnt help, we found many many glaring bugs in binutils Apr 18 19:20:29 XorA|gone: Do you know if they have been fixed? I know we carry some patches. Apr 18 19:20:35 I think they finally got fedup of us bitching though, recent ones work a whole lot better Apr 18 19:20:36 If not, that might be unfortunate. Apr 18 19:20:54 I suspect that's the case (and thanks for the complaints :) ) Apr 18 19:21:27 we use Phil Blundel to do our nagging :-) Apr 18 19:22:24 anyway I am off to check my defconfig Apr 18 19:22:54 lucid working nice on zoom2 apart from segfaults though Apr 18 19:24:33 that sounds like the config... other then segfaulting most things will still run... Apr 18 19:25:18 i ran into it in teh early alpha1-2 days since i was runing deb-pkg on top of fakeroot, and fakeroot was segfaulting.. Apr 18 19:29:03 rcn-ee: do you have all 3 erratas turned on? Apr 18 19:29:54 no, i haven't ran into needing 458693 or 460075 yet Apr 18 19:30:02 rcn-ee: just found your defconfig Apr 18 19:30:20 CONFIG_ARM_ERRATA_430973: ARM errata: Stale prediction on replaced interworking branch Apr 18 19:30:56 basicly if you have thumb binary and arm binary running, something bad could happen.. Apr 18 19:31:26 yeah I remember the discussion on linux-omap now Apr 18 19:31:38 I just forgot as we dont use thumb Apr 18 19:32:27 yeap, it only came into play with lucid.. but i had to go back and enable the option for all since my builders run squeeze and build all distro's in chroots... Apr 18 19:35:16 * XorA|gone feels lucky this compile isnt being done on the zoom2 :) Apr 18 19:35:41 it only takes 5ish hours with all modules.. ;) Apr 18 19:37:47 * XorA|gone curses and hunts who broke OE Apr 18 19:38:29 ah my bad, I acked the patch :-) Apr 18 19:49:18 * XorA|gone notices ubuntu needs a rule in udev to link /dev/fb0 to /dev/fb Apr 18 19:51:27 What unfixably depends on /dev/fb ? Apr 18 19:52:02 persia: well you could patch xserver-video-omapfb (and omap3 version) to look properly Apr 18 19:52:43 That sounds less invasive (only affects the one bit that has issues). Apr 18 19:52:47 Does a patch exist? Apr 18 19:53:03 persia: no, as we just create the link Apr 18 19:53:39 persia: I will get tasked with that eventually by TI more than likely Apr 18 19:53:49 heh, OK. Apr 18 19:54:13 The alternative would be to have the xserver-xorg-video-omapfb package install a udev rule, but that feels hacky. Apr 18 19:54:21 (on the other hand, it's probably faster) Apr 18 19:54:33 dh_installudev ought give you the hints to do that easily. Apr 18 19:56:17 But we should avoid patching udev, as folks on other platforms don't need/want the link. Apr 18 20:01:27 persia: yeah, I see that Apr 18 20:01:54 I think the issue is the meanings on the fb's is kind of hardcoded into omap3 Apr 18 20:02:05 fb1 fb2 are the hardware overlays for video Apr 18 20:06:34 rcn-ee: thanks, looking a lot better now Apr 18 20:07:19 good to hear XorA|gone Apr 18 20:08:14 just two bugs to figure out on zoom2 and its usable Apr 18 20:08:40 which two are left? Apr 18 20:08:53 touchscreen is uncalibrated and keyboard is not working Apr 18 20:09:08 usb keyboard is working, so it must be something simple Apr 18 20:30:12 Does the driver provide a /dev/input/event interface? Apr 18 20:39:34 persia: I think so, Im pretty sure its some minor issue, its working in poky and angstrom with xorg 1.8 with no patches Apr 18 20:40:55 I'm just wondering if it's a side-effect of Ubuntu's attempts to not use xorg.conf Apr 18 20:41:22 Lots of stuff that works just fine needs tweaking to work in Ubuntu due to needing to fit into autodetection Apr 18 20:41:27 ah crap, it could be a rule missing from hal/libudev so it doesnt ID it as a keyboard Apr 18 20:41:38 we use autodetection as well Apr 18 20:41:54 Right. I thought I remembered ndec saying something about the zoom2 keyboard being a bit funny. Apr 18 20:42:17 I dont remeber us having a custom rule, but its possible we do Apr 18 20:42:35 * XorA|gone adds it to the list Apr 18 20:42:51 Oh, and in case you didn't notice, hal has been deeply stripped in Ubuntu as part of the efforts to make it go away entirely, so if you're depending on HAL behaviour, you mayfind it differs. Apr 18 20:43:57 persia: there is a dev branch of Angstrom that is nearly totally hal gone, so we should be about the same level Apr 18 20:44:06 persia: it will be some minor tweak Im sure Apr 18 20:44:33 Probably. Glad to hear that Angstrom is dropping HAL also: it always feels better to be part of the crowd :) Apr 18 20:44:38 Ill be sure to post patches when I actually start coding on this Apr 18 20:46:08 BTW is there any plans to make Lucid initialise the OTG port on boot? Apr 18 20:46:20 I doubt it. Lucid is in final-freeze. Apr 18 20:46:31 ok Apr 18 20:46:36 So only RC stuff is getting pushed at this point. Apr 18 20:46:50 * XorA|gone is still a little upset at the EHCI port on his beagle burning out Apr 18 20:47:12 Understandably. Apr 18 20:48:15 Hmm, so will it end up still having that kernel oops (and segfault on reboot)? Apr 18 20:48:23 That's what I get with the Lucid kernel on my beagle. Apr 18 20:48:31 DanaG: Which bug #? Apr 18 20:48:52 erm, lemme' see if there is one. Apr 18 20:49:16 I promise that if there's not a bug, there's no chance it will be changed for lucid :) Apr 18 20:51:19 What's the plan for officially installing the ubuntu kernel? Will it have an equivalent of update-grub? Apr 18 20:51:40 No. Apr 18 20:52:00 But that's mostly because update-grub does a very different thing than is currently suitable Apr 18 20:52:07 (someone should port grub and fix this) Apr 18 20:52:19 Hmm, bummer. it'd be nice to have it auto-copy uimage to the fat16 partition. Apr 18 20:52:25 that's what I meant. Apr 18 20:52:49 yeah, I guess that is way different from grub, after all. Apr 18 20:52:56 DanaG: Well, flash-kernel can do some of that: it's designed to copy the kernel from /boot to somewhere else for devices that don't have decent bootloaders. Apr 18 20:53:12 hmm, zoom2 keyboard decides to work again Apr 18 20:53:19 And I believe the kernel calls flash-kernel on postinst. Apr 18 20:53:50 XorA|gone: heisenbug? Apr 18 20:54:00 I also figured out something interesting with my asix ethernet: Apr 18 20:54:05 persia: maybe the driver went wrong due to that thumb issue Apr 18 20:54:17 The thing that was making it not work on boot... was that I had "ethtool -s eth0 wol g" in rc.local. Apr 18 20:54:18 Ah, that might be it. Apr 18 20:54:23 Apparently enabling WOL kills asix. Apr 18 20:54:27 so just touchscreen Apr 18 20:54:38 and the patch Angstrom has for that is not suitable upstream Apr 18 20:54:52 What's the issue with touchscreen? Apr 18 20:55:24 I beleive it thinks its a touchpad, not a touchscreen Apr 18 20:55:42 but its how to tell libudev how to properly identify them Apr 18 20:55:55 Oh, yeah. The Netwalker has the same issue. Apr 18 20:56:26 The (hacky) solution in the netwalker was to just hint the synaptics driver when the device was discovered in HAL. Apr 18 20:56:30 But that's 9.04. Apr 18 20:56:41 For lucid, you might get away with the same trick in udev. Apr 18 20:56:45 koen did a hack where if driver returns pressure use tslib Apr 18 20:56:52 (but better to fix properly, if one can) Apr 18 20:57:17 * XorA|gone will talk to hrw he had similar issues with bug which he might have fixed Apr 18 20:57:28 That's a plan :) Apr 18 20:58:51 not as fast as that omap4 board, but speed isnt bad on the UI :) Apr 18 21:02:01 Everyone fusses about speed, but really, unless you're doing something especially complex (like rendering multi-input graphic interfaces in realtime: e.g. web browsing), most stuff doens't really need a high-spec processor. Apr 18 21:08:58 wlan0 Interface doesn't support scanning. Apr 18 21:11:11 ah. here's my oops: https://bugs.launchpad.net/ubuntu/+source/linux-ti-omap/+bug/563650 Apr 18 21:11:15 Launchpad bug 563650 in linux-ti-omap (Ubuntu Lucid) (and 1 other project) "DSS2 oops when shutting down while DPMS is active (affects: 1) (heat: 8)" [Medium,Confirmed] Apr 18 21:12:36 DanaG: OK. That's been determined to not be release-critical: it's scheduled for a post-release update. Apr 18 21:12:44 (milestoned against lucid-updates) Apr 18 21:13:05 Ah, the only other big thing is "iwlist s" not working -- probably due to WEXT being disabled. Apr 18 21:13:05 /boot/config-2.6.33-500-omap:# CONFIG_CFG80211_WEXT is not set Apr 18 21:13:13 DanaG: same ooops present on zoom2 Apr 18 21:15:17 NetworkManager: wireless_get_range(): (wlan0): couldn't get driver range information (95). Apr 18 21:15:17 NetworkManager: constructor(): (wlan0): Device unsupported, ignoring. Apr 18 21:18:28 well thats me happy for tonight, thanks for the help guys Apr 18 21:19:05 https://bugs.launchpad.net/ubuntu/+source/linux-ti-omap/+bug/566238 Apr 18 21:19:07 Launchpad bug 566238 in linux-ti-omap (Ubuntu) "wlan0 "Interface doesn't support scanning." -- CONFIG_CFG80211_WEXT is not set (affects: 1)" [Undecided,New] Apr 19 01:18:31 * Martyn swears loudly at the tegra2 board Apr 19 01:18:42 this chip has the oddest frigging bottlenecks in performance! Apr 19 01:20:31 So, I've got this source that hardcodes "armv4l" in ./configure, and fails to build for any other `uname -m`. Anyone have a good suggestion on the safest way to patch that? Apr 19 01:20:56 right now, I'm trying to improve streams performance .. and do you know where I found the bottleneck? Apr 19 01:21:10 I can't make memcpy() fast. Apr 19 01:21:36 DMA is fast as all getgo Apr 19 01:21:43 mem copy? Slow. Slow as dog Apr 19 01:22:49 What! Apr 19 01:23:20 * persia can't imagine how to implement something that would have that particular combination of performance metrics. Apr 19 01:24:20 __aeabi_memcpy4 Apr 19 01:24:27 is what is being used .. aligned memcpy ... Apr 19 01:24:32 it _should_ be fast. Apr 19 01:25:15 I'm going to try to force it to use __aeabi_memcpy instead Apr 19 01:25:21 persia : Any ideas? Apr 19 01:25:29 None, sorry. Apr 19 01:25:53 I don't really understand that low-level stuff. Apr 19 01:26:22 There is no NEON on the tegra2, so I know the GNU optimizer won't try for that Apr 19 01:26:30 (saving a potential 50 cycle penalty) Apr 19 01:29:29 Well .. what I think I'll do is use PLD, ldm/stm about 32bytes to match the cache line size, and keep things aligned Apr 19 01:29:39 but it SUCKS that I need to do that. Apr 19 01:29:44 write my own memcpy() Apr 19 02:56:23 hmm, say, those Marvell 1.2GHz ARM CPUs... how do they compare in desktop usage (i.e. firefox page-load, and such) to a 1.6GHz Atom, or to a 1.6GHz Athlon XP? **** ENDING LOGGING AT Mon Apr 19 02:59:56 2010