| 00:20.36 | *** part/#arm-netbook gordan (~gordan@78-86-195-86.zone2.bethere.co.uk) |
| 10:23.40 | *** join/#arm-netbook gordan (~gordan@78-86-195-86.zone2.bethere.co.uk) |
| 16:42.09 | A2Sheds | gordan: is there anything to "funky" in the Red Sleeve kernels to not play well with RTAI or Xenomai? |
| 16:45.11 | gordan | Well, the RS kernels are largely going to be whatever works for each device, reduced down to the minimum number of kernel sources possible. |
| 16:45.35 | gordan | At the moment I'm fighting the RH kernel (2.6.32.-220) onto a SheevaPlug. |
| 16:45.55 | gordan | That isn't being plain sailing. |
| 16:46.08 | gordan | And that's for Marvell Kirkwood, which was supported in mainline in 2.6.32 |
| 16:46.32 | gordan | It could be a uboot issue, but I am seeing errors like "undefined instruction" when booting the kernel. |
| 16:46.47 | gordan | I'll try a newer uboot, but the last attempt misflashed so I have to unbrick it first. |
| 16:47.32 | gordan | All other devices are likely to have their own kernels built from their own sources, unless patches for multiple devices apply cleanly and result in a single source that works for them. |
| 16:47.50 | gordan | Again, things like Kirkwood that have been supported for ages will likely work on most later kernels out of the box. |
| 16:48.57 | gordan | For Tegra2 based devices I will probably use 2.6.38, and that kernel should work for Kirkwoods, too, if the RH kernel ends up being too problematic (their thousands of patches have a tendency to break most configurations that aren't their default configuration). |
| 16:49.11 | gordan | For everything else, we'll have to see. |
| 16:49.56 | gordan | Or to answer your question directly - each supported device will have, ideally, a single kernel for it's SoC. |
| 16:50.13 | gordan | i.e. one for Tegra2, one for Marvell Kirkwood, one for Marvell Armada, etc. |
| 16:50.51 | gordan | It is almost certain that most of not all kernels will NOT be based on the RHEL kernels. |
| 16:51.04 | gordan | The only exception is the Marvell Kirkwood, and the jury is still out on that one. |
| 16:57.29 | A2Sheds | http://code.google.com/p/miniemc2/ he has xenomai on arm9 mini2440 2.6.35.9 kernel just FYI |
| 16:58.39 | gordan | Fair enough. Since there is already a working kernel for that device worst case, it is that kernel that will be rpm-ed up for it. |
| 16:59.02 | A2Sheds | I see |
| 16:59.30 | gordan | In the ideal case, any patches required for that SoC will apply cleanly against another kernel used for a different device. In that instance the same source will be used to build the binary kernels for both devices. |
| 16:59.47 | gordan | The problem is that SoCs are all incompatible on boot level. |
| 17:00.00 | gordan | And you can't build one kernel that will boot on multiple SoCs. |
| 17:00.04 | A2Sheds | somebody here is working on an all open toolchain for building with no mystery/magic openembedded or similar |
| 17:00.22 | A2Sheds | yeah, the fun of ARM |
| 17:00.25 | gordan | The toolchain will be the RHEL one. |
| 17:00.40 | gordan | Only the kernel will be SoC specific, the userspace is common. |
| 17:00.49 | A2Sheds | ok, good |
| 17:01.09 | gordan | But even though you need different kernel binaries, I'm hoping that the number of source trees will be << than the number of SoCs. |
| 17:01.21 | gordan | i.e. same source tree used for building the kernels for different SoCs. |
| 17:02.03 | A2Sheds | I haven't checked in a few months but even Russel King was getting fed up with all the uncooperative kernel patches |
| 17:02.30 | gordan | e.g. if I use the 2.6.38 tree for Tegra2 (based on ChromeOS kernel), that _should_ also work with Marvell Kirkwood and Marvell Armada SoCs since both of the mentioned Marvells have been in mainline since < 2.6.33, Kirkwood since before 2.6.32. |
| 17:02.47 | gordan | The problem is that there is no common boot mechanism in ARM kernels. |
| 17:02.58 | gordan | The boostraps are all different, hence why you need a different kernel for each one. |
| 17:04.09 | gordan | Linaro are working on persuading all the member vendors to unify their designs so that a single boot loader and a single kernel can boot on all of their SoCs. |
| 17:04.24 | A2Sheds | it was frustrating to watch all the ARM tablets released with closed bootloaders and all the lockdown |
| 17:04.27 | gordan | But it'll be years before that actually comes to fruition. |
| 17:04.56 | gordan | You think that's bad? Similar thing is going to be happening to x86 PCs. |
| 17:05.08 | gordan | BIOS lock down so that they won't run a boot loader that isn't signed by an authorised key. |
| 17:05.19 | A2Sheds | yeah, I've been working with coreboot for year |
| 17:05.23 | A2Sheds | years |
| 17:05.37 | gordan | And there was talk of them requiring PC vendors to lock down their BIOS so they won't boot anything other than Windows. |
| 17:05.44 | gordan | Which is _nuts_. |
| 17:05.53 | gordan | There was some backtracking on that, but it is concerning. |
| 17:06.18 | A2Sheds | well AMD moved to being much more open |
| 17:06.28 | A2Sheds | Intel has become more closed |
| 17:06.33 | gordan | It's getting to a point where we're going to have to start soldering JTAGs to our x86 PCs to flash the BIOS with coreboot and escape the lock-in. |
| 17:06.49 | gordan | Yeah, but AMD hardware is pretty shit these days. |
| 17:07.04 | A2Sheds | if you wish |
| 17:07.22 | gordan | That's my perception anyway. |
| 17:07.44 | gordan | At least comparing the Core2/Phenom2 generation. That is the most recent one I have in my machines. |
| 17:07.58 | gordan | AMD's margins for error on their CPUs are too tight for comfort. |
| 17:08.02 | A2Sheds | we were discussing what Linux distro to try and work with on ARM |
| 17:08.26 | gordan | Who's "we"? |
| 17:08.40 | A2Sheds | my company |
| 17:08.54 | gordan | And I thought the world was all moving to Ubuntu, with only a few of us hardcore non-conformists using something else. :) |
| 17:09.17 | gordan | Because other distros don't require people who know what they are doing to use them, or something like that. :) |
| 17:09.26 | gordan | Sorry, I got that backwards. |
| 17:09.32 | A2Sheds | no Ubuntu even for x86 |
| 17:09.35 | gordan | Because other distros DO require -"- |
| 17:10.14 | gordan | But if you are interested in RedSleeve, drop me an email on the mailing list and I can let you know as soon as it's ready. |
| 17:10.22 | A2Sheds | we looked at ARM Slack as well |
| 17:10.23 | gordan | What ARM hardware will you be running it on? |
| 17:10.34 | gordan | What SoC? |
| 17:10.39 | A2Sheds | the A10 for one |
| 17:10.54 | gordan | Sure. Any others? |
| 17:11.01 | gordan | Just asking to make sure I have kernels ready for them. |
| 17:11.24 | A2Sheds | possibly various OMAP |
| 17:11.54 | gordan | OK. I haven't got a Pandaboard or a Beagleboard yet. Will acquire some in the near future and see what I can cook up. |
| 17:12.10 | A2Sheds | I have had a panda for a while |
| 17:12.22 | gordan | I can only provide kernels for the hardware that I can get my hands on. |
| 17:12.22 | A2Sheds | no way to get silicon |
| 17:12.28 | gordan | For obvious reasons. |
| 17:12.33 | A2Sheds | yeah |
| 17:13.29 | gordan | At the moment, I can only promise that Marvell Kirkwood and Nvidia Tegra2 based devices will be supported from the first public beta release, since that is the hardware I have at the moment (SheevaPlug, GuruPlug, DreamPlug for Kirkwood, Toshiba AC100 for Tegra2). |
| 17:13.31 | A2Sheds | I was tempted to reverse and unlock some of the Archos omap tablets , but in the end I decided not to funnel any $$ their way |
| 17:14.03 | gordan | Yeah, it's an awful lot of work to put in if you're not going to earn your effort's worth of profit on the back of it. |
| 17:14.23 | gordan | As somebody once said, if your business hasn't got a profitable plan, it's not a business, it's a hobby. :) |
| 17:14.39 | A2Sheds | then the A10 popped up and decided to see what I can do to make this thing happen |
| 17:14.46 | gordan | I have Marvell Armada hardware in the post, and will be getting a D2Plug (also Armada) when they are available to cross-test. |
| 17:14.53 | gordan | Indeed. |
| 17:15.05 | A2Sheds | board design, fabrication. whatever |
| 17:15.41 | A2Sheds | I tried to work with Marvell |
| 17:15.57 | gordan | I'll try getting a kernel working as soon as I can get my hands on a dev kit board (if I can get it for a reasonably amount, there is a limit on how much I am prepared to spend on a device just to put it all the work to make it work with the distro just so I could say that my distro runs on it). |
| 17:16.02 | A2Sheds | it never went anywhere, plus they went more closed on the docs they said they would open |
| 17:16.10 | gordan | Marvell are usually quite good. What was the problem? |
| 17:16.26 | gordan | What's closed? |
| 17:16.37 | gordan | Kirkwood and Armada have full support in the mainline kernel. |
| 17:16.44 | A2Sheds | problems with silicon, pricing |
| 17:17.06 | A2Sheds | nobody in China really wanted to work with them |
| 17:17.19 | gordan | Sure, I can understand silicon issues and pricing, but closedness? |
| 17:17.20 | A2Sheds | no low cost boards or tablets on the market |
| 17:17.43 | gordan | Very few SoC manufacturers can say they have full mainlike kernel support. |
| 17:17.48 | gordan | And Marvell are one of them. |
| 17:17.51 | gordan | And they keep up to date. |
| 17:17.57 | gordan | Armada has been in mainline since 2.6.33 |
| 17:17.59 | gordan | IIRC |
| 17:18.37 | A2Sheds | we wanted to make Kirkwood the first coreboot ARM device |
| 17:18.59 | gordan | What stopped you? |
| 17:19.14 | A2Sheds | we even had a marvell dev to lead the effort, they wouldn't open the docs for the bootloader section of the hardware |
| 17:19.30 | gordan | They use uboot by default, and the source is available. |
| 17:19.35 | gordan | What was the problem? |
| 17:19.47 | A2Sheds | no public docs |
| 17:19.49 | gordan | Could you not just see what uboot does, skip uboot and get the kernel to boot directly? |
| 17:20.05 | gordan | source code is the <s>most</s> ONLY reliable documentation you will ever get. |
| 17:20.29 | A2Sheds | I'm trying to recall, but I don't think the uboot source has comments for every register setting, jst magic |
| 17:20.33 | gordan | A10 is the same - from what Luke said, useful things like SATA aren't even documented on the spec sheets, yet the code is implemented for it in the kernel. |
| 17:20.51 | A2Sheds | yeah, but $5 makes it interesting |
| 17:21.02 | gordan | IMO, you can't really argue the closedness when you have working source. :) |
| 17:21.05 | A2Sheds | cut them a little slack |
| 17:21.28 | gordan | Sure, it's not as easy as having docs in english for a clean re-implementation, but it's a hell of a lot better than nothing. |
| 17:23.13 | A2Sheds | magic settings for init/boot was the issue |
| 17:24.26 | A2Sheds | i wish i could find a Kirkwood something-itx for $60 |
| 17:25.57 | A2Sheds | I'd even live with a tegra2 or omap44xx for $70 board |
| 17:27.07 | A2Sheds | I'd be happy to provide schematics, gerbers and a BOM if TI or nvidia would cooperate |
| 17:32.02 | gordan | LOL! There are only 2 boards that I am aware for that have ARM chips and are *TX format. |
| 17:32.13 | gordan | But $60 is pushing it. |
| 17:32.54 | A2Sheds | whatever small format |
| 17:32.58 | gordan | Atmel did a board ages ago, but it's limited to something like 128MB of RAM and has a 266MHz chip on it, not really big enough for non-embedded use, and for embedded use you don't want *TX form factor. |
| 17:33.36 | gordan | Compulab do an Armada board that is u-ATX, but they are $350, prettymuch what you'll pay for an industrial Atom N450 motherboard. |
| 17:33.37 | A2Sheds | small, even in the shape of a heart would be ok |
| 17:34.04 | gordan | I have a couple of insdustrial Atom N450 motherboards, and they suck, though. |
| 17:34.11 | A2Sheds | yeah, I've spoken with them and the other SOC-on-module vendors |
| 17:34.14 | gordan | Don't buy Advancetech. |
| 17:34.40 | gordan | They didn't implement clock/voltage scaling, and when questioned about this, they claimed it was an "overclocking feature". |
| 17:35.12 | A2Sheds | I've been designing x86 and risc board for decades, I use off-the-shelf as well |
| 17:35.18 | gordan | It was also undlear how they managed to get a motherboard where the CPU+MH package +SB that are supposed to have a combined TDP of < 10W to suck > 30W of power. |
| 17:35.35 | gordan | s/undlear/unclear/ |
| 17:35.57 | A2Sheds | speel checking bot? |
| 17:36.17 | gordan | Seems the server runs a bot that understands when you put in a regex. :) |
| 17:36.20 | A2Sheds | sp speel/spell |
| 17:36.23 | gordan | s/foo/bar/ |
| 17:36.45 | gordan | e.g. foo bar baz |
| 17:36.50 | gordan | s/bar/bar2/ |
| 17:37.14 | gordan | Skype does similar. |
| 17:38.16 | A2Sheds | well Red Sleeves on A10 is of interest |
| 17:38.51 | gordan | I'll get it working as soon as I have an A10 dev board and working kernel sources. |
| 17:38.51 | A2Sheds | I'm trying to avoid rolling our own again |
| 17:39.07 | gordan | RSEL6 is 99% identical to RHEL6. |
| 17:39.20 | gordan | Same sources, plus patches required to build+function on ARM. |
| 17:39.30 | A2Sheds | and personally I'm trying to stay on the hardware end of things on the A10 |
| 17:39.50 | gordan | In all about 100 patched packages, IIRC, I can check the number of patched packages later if you really want to know. |
| 17:39.56 | A2Sheds | make sure we all have board |
| 17:40.00 | A2Sheds | +s |
| 17:40.11 | gordan | Sure |
| 17:40.31 | A2Sheds | I've been on too many failed projects like this |
| 17:40.40 | gordan | I can imagine. |
| 17:40.47 | A2Sheds | a flurry of activity and then silence |
| 17:40.56 | A2Sheds | like the current factory in China |
| 17:41.13 | gordan | Bah. For some reason the kernels I'm building just aren't playing ball, and I can't figure out if it's a uboot issue, mkimage (uboot-tools) issue or just a kernel that is getting miscompiled... |
| 17:41.39 | A2Sheds | I just think they are stumped and don't know the next step and still make $$ |
| 17:41.49 | gordan | LOL! |
| 17:41.56 | gordan | Ooo, hang on - different error message! |
| 17:42.31 | gordan | It's no longer saying undefined instruction, it's now saying "software interrupt"... But that could just be the same error with different print(). |
| 17:42.47 | A2Sheds | a tablet design they can copy, but having to completely layout a new board is challenging to the beginner |
| 17:42.53 | A2Sheds | heh |
| 17:43.08 | gordan | Hmm, methinks I'll have to try a vanilla kernel next. |
| 17:43.11 | gordan | See if that works. |
| 17:43.34 | gordan | If it does, I know it's not uboot-tools (I just flashed uboot to the one that works on a different device). |
| 17:43.48 | gordan | Gotta go. |
| 17:43.50 | gordan | aways |
| 17:43.51 | A2Sheds | http://www.mini-box.com/pico-SAM9G45-X we started working with this right before the A10 came along |
| 17:44.31 | gordan | A bit too small for me. |
| 17:44.55 | gordan | And considering the Raspberry Pi will be out at half that price with near identical spec, why would you bother with that board? |
| 17:45.06 | gordan | Anyway, gotta go. Will be back in a few hours. |
| 17:45.11 | gordan | _really_ aways |