**** BEGIN LOGGING AT Mon Sep 14 02:59:57 2009 Sep 14 06:36:06 lennert, rabeeh: hi, armin76 said to say he sent me :-) i've got the mv78xx0 that destined to be a buildbox for Gentoo Sep 14 06:36:24 with weird behavior during early boot Sep 14 06:36:30 involving machineIDs Sep 14 06:37:37 define weird behavior Sep 14 06:38:32 so initial used a kernel with mv78xx0_defconfig Sep 14 06:38:35 no changes to the config Sep 14 06:38:58 but when I ran bootm, I get the following from uboot: Sep 14 06:38:59 Error: unrecognized/unsupported machine ID (r1 = 0x00000831). Sep 14 06:39:04 Available machine support: Sep 14 06:39:08 00000693 Marvell DB-78x00-BP Development Board Sep 14 06:39:12 00000857 Marvell RD-78x00-MASA Development Board Sep 14 06:39:30 0x831 is the machineid for the reference sheevaplug, so i'm not sure why the board is claiming to be it Sep 14 06:39:47 robbat2: hi Sep 14 06:39:58 when the board is very definetly the DB-78100-A-BP Sep 14 06:40:01 robbat2: you need to set 'mainlinuxLinux' environment variable to yes Sep 14 06:40:06 it is set Sep 14 06:40:35 one sec, i'll reboot back to uboot to grab the current printenv output Sep 14 06:40:42 robbat2: then set arcNumber to 2135 (which is 0x857 in decimal) Sep 14 06:40:55 robbat2|na: you probably set arcNumber to that of the sheevaplug Sep 14 06:41:15 ah, now I was asking a question earlier as to how to change how the board identified Sep 14 06:41:18 rabeeh: and i don't think he has a mASA, or does he? Sep 14 06:41:23 and nobody could answer me Sep 14 06:41:32 robbat2|na: the board is identified to the kernel by the arcNumber uboot variable Sep 14 06:41:41 robbat2|na: do you have a metal box or an atx-ish mainboard? Sep 14 06:41:42 i grep the u-boot 1.1.4 source for it Sep 14 06:41:47 but it seems to have been heavily modified Sep 14 06:41:56 hes has DB Sep 14 06:41:57 and I can't find the matching source in violation of the GPL I think Sep 14 06:42:01 it's BTX Sep 14 06:42:02 78100 rev A0 Sep 14 06:42:08 full-length BTX specifically Sep 14 06:42:21 robbat2|na: then you need to set arcNumber to 1683 Sep 14 06:42:35 product rev 1.7, serial number 256 :-) Sep 14 06:43:15 robbat2: oh. you got the 256 then :) Sep 14 06:44:37 where's the matching source for this u-boot btw? Sep 14 06:46:07 robbat2|na: on the extranet if you have an extranet account, or rabeeh can email it to you Sep 14 06:46:45 nope, i don't have an extranet account, but I can sign up for one if needs be Sep 14 06:47:02 robbat2: no need for that. i'll send sources. Sep 14 06:47:11 * robbat2|na goes back to unpatch his MACHINE_START changes Sep 14 06:51:01 that's looking better Sep 14 06:51:07 rtc-ds1307: probe of 0-0068 failed with error -5 Sep 14 06:51:29 that's the only error in the boot sequence now Sep 14 06:54:57 rabeeh, actually, from the uboot prompt, it's A1, not A0 Sep 14 06:55:08 ok Sep 14 07:18:01 one last question on the hardware for this evening, before I go to bed Sep 14 07:18:26 the specs said 4GiB RAM support, and i've got 2x2GiB DIMMs in there, but uboot only picks up 3GiB Sep 14 07:18:44 SysClock = 333Mhz , TClock = 200Mhz Sep 14 07:18:44 Enable 2T DRAM CAS Latency = 5 tRP = 5 tRAS = 15 tRCD=5 Sep 14 07:18:44 DRAM CS[0] base 0x00000000 size 1GB Sep 14 07:18:44 DRAM CS[1] base 0x40000000 size 1GB Sep 14 07:18:44 DRAM CS[2] base 0x80000000 size 1GB Sep 14 07:18:44 DRAM Total size 3GB 64bit width Sep 14 07:20:02 robbat2|na: that's cause it needs some physical memory space for things like internal registers Sep 14 07:20:10 robbat2|na: since the physical address space is only 4G Sep 14 07:20:29 what's the ARM equivilent of the PC remapping? Sep 14 07:20:51 PC remapping? Sep 14 07:21:00 assuming that it's like the PCI address space that's usually 256MiB @ 0xA0000000 Sep 14 07:21:28 where x86 has PAE, ARM has supersections (to map things between 4G and 64G) Sep 14 07:21:35 but supersections are kind of broken, since they map in 16M chunks only Sep 14 07:21:51 PAE, that's the acronym Sep 14 07:21:57 sorry, been up for 17 hours Sep 14 07:22:09 so you can't use supersections for RAM Sep 14 07:22:18 after a week of ~19 hour days Sep 14 07:22:48 we do have highmem on arm now, but that still won't help for >=4G mappings Sep 14 07:23:24 registers take up an entire 1GiB of mapping, or just that it doesn't handle discontigious mappings of having the registers in the middle? Sep 14 07:23:50 registers take 1M, PCI mappings take a couple hundred megs Sep 14 07:23:59 in theory you could map 512M of the last 1G in the upper 1G or so i guess Sep 14 07:24:07 i'd have to check if the dram controller allows that Sep 14 07:24:38 * robbat2|na has a lot to pack into RAM for faster builds Sep 14 07:25:20 howabout writing a block device driver for the last GB and using it as swap? :) Sep 14 07:25:39 suihkulokki, that would still depend on the DRAM controller Sep 14 07:25:47 and said driver exists already Sep 14 07:25:50 phram/slram etc Sep 14 07:26:32 suihkulokki: well, you _could_ use supersections for the access to the upper gig of dram, except mv78xx0 doesn't do supersections or 36bit phys addr space :) Sep 14 07:27:43 MTRR, that was the other thing I was thinking of earlier Sep 14 07:28:01 arm doesn't have mtrrs, all that info is expressed in the page table mappings directly Sep 14 07:28:34 so it's going to come down to those allow a discontigious mapping Sep 14 07:29:06 if you can make the ram controller allow it and clear the c0000000-dfffffff phys region, then i guess you could try it Sep 14 07:29:14 you'll still need highmem though Sep 14 07:29:25 and no single process can use more than 3g of virtual address space Sep 14 07:29:34 yup, standard i386 limitations Sep 14 07:29:36 so going beyond 3g would mainly be useful if you have multiple processes Sep 14 07:29:40 that each consume lots of ram Sep 14 07:29:48 standard 32bit address space limitations :) Sep 14 07:29:50 tmpfs and related tricks Sep 14 07:30:45 along with my favourite of tmpfs-backed loop-device, reiserfs formatted, tail-packed :-) Sep 14 07:31:11 for really packed data without inode overhead Sep 14 07:31:29 squashfs would be a contender, but it's not R/W Sep 14 07:42:14 robbat2|na: ZFS! Sep 14 07:42:16 * arachnist hides Sep 14 07:42:30 arachnist, if ZFS had tail-packing, i'd consider it Sep 14 07:42:38 but to date, reiserfs3 is the only filesystem that has it Sep 14 07:42:47 i use it for a very specific purpose Sep 14 07:42:53 of data i'm willing to lose in a reboot Sep 14 07:43:19 but it makes a notable performance difference with the rsync.gentoo.org mirrors Sep 14 07:43:56 if rsync could serve a packed squashfs or something similar without unpacking it, that would be even better Sep 14 07:43:59 but until then Sep 14 07:44:28 reiserfs on top of tmpfs is the best way to pack a a lot of tiny files into a tree for distribution Sep 14 14:38:55 hello Sep 14 14:39:05 anyone here already running 2.6.31 on their plug? Sep 14 14:39:45 for whatever reason i cannot get it to boot, u-boot starts loading the kernel and ends up with an error, something about invalid instruction iirc (dont have the plug handy right now) **** ENDING LOGGING AT Tue Sep 15 02:59:57 2009