**** BEGIN LOGGING AT Tue Apr 03 02:59:56 2007 Apr 03 07:41:34 kaloz * r6845 /trunk/target/linux/ixp4xx-2.6/patches/300-wg302v2.patch: fix typo Apr 03 07:42:14 kaloz * r6846 /trunk/target/linux/ixp4xx-2.6/config/default: enable OTP support for IXP4xx Apr 03 07:47:31 Kaloz: what is OTP support? Apr 03 07:48:05 One Time Programmable blocks Apr 03 07:48:19 ah, which ixp4xx device has those? Apr 03 07:48:32 the Pronghorn Metro stored the mac on those blocks, and have some OTB blocks free as well Apr 03 07:48:32 (well, which board with an ixp4xx and otp flash on it) Apr 03 07:48:55 no idea how to access it from linux (it shows up in dmesg, but not in /proc/mtd) Apr 03 07:48:57 interesting. what are you planning to use them for? Apr 03 07:49:21 well, for now i want to read the mac address from them :) Apr 03 07:49:47 ah, I thought you might be going to use them for the secret rootkit support that couldn't be removed ... Apr 03 07:51:24 the space wouldn't be enough for that :) Apr 03 07:51:51 anyways, any idea on how to access it? it shows up as a different flash region with different erase size and stuff Apr 03 07:52:29 btw, squashfs works for me, so i guess the bootup problem has to be related to the rtc drive ron the nslu Apr 03 08:01:44 rwhitby: btw, http://www.adiengineering.com/products/data/Pronghorn_Metro_Board_Users_Manual_R110.pdf Apr 03 08:02:12 rwhitby: and "3.10" there Apr 03 08:17:46 rwhitby: found code for it :) Apr 03 08:34:39 is it normal for kmalloc() to fail on a request of 704k? Apr 03 08:39:17 sn9__: On i386, kmalloc fails on more than 128k. Apr 03 08:39:40 what if i call it twelve times? Apr 03 08:39:50 sn9__: IIRC, kmalloc is for physically contiguous pages, while vmalloc is for a block of RAM. Apr 03 08:40:05 oh, maybe that can work Apr 03 08:40:26 http://lwn.net/images/pdf/LDD3/ch08.pdf is worth a read. Apr 03 08:40:53 i used kmalloc because that's what brcm47xx-2.6 used Apr 03 08:41:07 Do you need physically contiguous pages? Apr 03 08:41:11 no Apr 03 08:41:20 just logically contiguous Apr 03 08:41:36 vmalloc is plenty good enough then. Apr 03 08:41:55 if it can do a meg or more, great Apr 03 08:42:16 Or code it up with get_zeroed_page and friends. Apr 03 08:42:31 is the inverse vfree()? Apr 03 08:42:34 Read LDD3 chapter 8 (the PDF I linked); in general, allocating that much memory in kernel context is seen as bad. Apr 03 08:42:42 Yes. Apr 03 08:42:50 it's freed a few lines later Apr 03 08:42:57 What's it used for? Apr 03 08:43:03 io buffer Apr 03 08:43:41 That's an impressively large I/O buffer; hardware incapable of SG DMA? Apr 03 08:44:21 You may also find the DMA interface worth looking at (http://lwn.net/images/pdf/LDD3/ch15.pdf). And don't forget request_firmware if you're loading a firmware image. Apr 03 08:45:26 And there's get_user_pages to consider if you're doing I/O for userspace Apr 03 08:46:22 the hardware is flash (silly, i know) Apr 03 08:47:39 i'm trying to duplicate the brcm47xx-2.6 method of replacing the crc32 on boot by recalculating it in-kernel Apr 03 08:48:04 Can you not just map the existing flash, and recalculate the CRC in a register? Apr 03 08:48:21 but i also need to reflash, too Apr 03 08:48:49 You should only need to reflah one sector at a time, right? Apr 03 08:49:37 one sector, period Apr 03 08:50:59 Should be able to remap the existing image, calculate the CRC, then rewrite just one sector (64 or 128K). Apr 03 08:51:46 ok, i'm an idiot Apr 03 08:51:49 Can you reuse the MTD layer to do the rewrite? Apr 03 08:51:57 Make them handle the allocation, you see. Apr 03 08:51:58 i was Apr 03 08:53:36 oh, now i remember why i did it this way Apr 03 08:53:55 i need to write to the buffer Apr 03 08:54:35 If you're early enough in boot, vmalloc/do something/vfree should work OK. It's just frowned upon by the LKML mob if there's another way to do it. Apr 03 08:55:05 it would make no sense to just reflash the same contents with no changes. i can't edit what i ioremap(), right? Apr 03 08:55:57 No, as it's just remapping the flash address space. I'm just surprised that you need to reflash all 704k, not just 4 bytes for a checksum. Apr 03 08:56:38 i don't Apr 03 08:57:16 but i need to read it all contiguously with the part that gets overwritten, AFTER the overwrite Apr 03 08:57:59 but before the flash Apr 03 08:58:11 vmalloc *should* work, but I'd be inclined to find a way to avoid needing such a large alloc if I could. Apr 03 09:04:22 vmalloc did the trick, but the crc32 is giving the wrong result for some reason Apr 03 09:50:22 florian * r6847 /trunk/ (3 files in 3 dirs): Remove useless mksyshdr utility, replaced by mkcsysimg, Edimax images are no longer broken, and have been tested Apr 03 11:56:14 mbm * r6849 /trunk/ (Makefile include/package-ipkg.mk scripts/feeds.sh): Using find's -exec to remove directories causes 'not found' errors when find tries to search the removed directory. Hopefully fixed now. Apr 03 13:59:11 mbm * r6850 /trunk/target/ (imagebuilder/Makefile sdk/Makefile): more [6849] Apr 03 15:55:45 mbm * r6851 /trunk/include/shell.sh: bug #1547 Apr 03 15:56:09 [florian]: big.patch and rdc-2.4.tar.bz2 updated, again Apr 03 19:45:15 <[florian]> sn9__: can you please remind me of the url ? Apr 03 19:54:48 florian * r6852 /trunk/target/linux/adm5120-2.6/files/arch/mips/adm5120/prom.c: Use CONFIG_CMDLINE instead of hardcoded kernel command line Apr 03 19:55:12 florian * r6853 /trunk/tools/firmware-utils/src/mkcsysimg.c: mkcsysimg typos Apr 03 21:02:07 florian * r6854 /branches/whiterussian/openwrt/package/base-files/default/etc/init.d/S05nvram: Add missing ending brackets (#1545) Apr 03 21:31:14 {Nico}: The dropbear update has now I would say been pretty well tested - did you get the message from before when I said it was ok ? Apr 03 21:50:52 <[florian]> h3sp4wn: which package provide ldconfig ? Apr 03 21:51:56 uClibc-0.9.28.2/utils (you just need to run make there) Apr 03 21:52:21 <[florian]> ok Apr 03 21:53:33 <[mbm]> cool, I can rewrite a whole page of the Makefile with a single rule Apr 03 21:53:36 <[mbm]> %-prereq %-download %-clean %-compile %-install: $(MAKE) -C $* $(patsubst $*-%,%,$@) Apr 03 21:54:56 <[florian]> that is intensive use of make :) Apr 03 21:58:29 <[mbm]> actually it's much cleaner than what it replaces Apr 03 21:59:25 <[florian]> you could certainly do the same for the clean target Apr 03 22:40:21 florian * r6855 /trunk/target/linux/adm5120-2.6/files/include/asm-mips/mach-adm5120/adm5120_info.h: Add the infrastructure to support different flash sizes/address Apr 03 22:50:29 florian * r6856 /trunk/ (3 files in 3 dirs): Add ldd and ldconfig support, template packages derived from gcc and uclibc Apr 03 22:54:11 <[florian]> h3sp4wn: you have your ldconfig now Apr 03 23:04:01 mbm * r6857 /trunk/ (Config.in Makefile include/scan.mk package/Makefile rules.mk): Tidy up Makefiles Apr 03 23:04:42 <[mbm]> could I get a few people to check that last commit? seems to work here but it changes so many things Apr 03 23:05:39 <[florian]> sure Apr 03 23:06:34 <[mbm]> one of the most obvious improvements is that you shouldn't see the 'Collecting info ...' stuff unless the sources actually change Apr 03 23:13:47 <[florian]> seems to be ok here Apr 03 23:14:37 also seems to be ok here Apr 03 23:14:49 <[mbm]> :) Apr 03 23:15:19 <[florian]> I should probably test on a gentoo ;) Apr 03 23:15:26 <[mbm]> suppose that means that if I did break anything it isn't obvious Apr 03 23:15:32 <[mbm]> which is good Apr 03 23:21:43 <[mbm]> *mumble* broke a few things Apr 03 23:23:43 <[mbm]> btw why was ldd added? Apr 03 23:23:56 <[mbm]> kind of pointless Apr 03 23:26:04 <[mbm]> (all ldd does is set the LD_TRACE_LOADED_OBJECTS variable) Apr 03 23:29:07 mbm * r6858 /trunk/target/ (imagebuilder/Makefile sdk/files/Makefile): Accidentially broke sdk and imagebuilder in [6857] Apr 03 23:29:31 * thepeople_mobile restarts build Apr 03 23:30:21 <[mbm]> did a "full" build of 6858, worked fine Apr 03 23:30:39 <[mbm]> admittedly everything was already built and in build_mipsel Apr 03 23:31:04 <[mbm]> but I had everything except python and rdiff selected Apr 03 23:31:04 I am doing x86 and everything is already built here also Apr 03 23:31:21 python compiles for me, python24 doesn't Apr 03 23:31:43 <[mbm]> yeah I think this was the python24 version Apr 03 23:31:58 <[mbm]> tons of warchings about the arch because it was using the wrong include files Apr 03 23:32:45 I gave my warnings and errors to lst a while back Apr 03 23:32:54 <[mbm]> :) Apr 03 23:33:07 <[mbm]> I'm thinking of just tagging the package as broken Apr 03 23:33:58 I think he is going straight to python 2.5 and dumping the 2.4 package Apr 03 23:36:24 [florian]: sure: http://homepage.mac.com/danielg4/big.patch and http://homepage.mac.com/danielg4/rdc-2.4.tar.bz2 Apr 03 23:37:56 <[mbm]> umm .. that python makefile is a little odd Apr 04 00:12:34 the last fix you did mbm worked for me also Apr 04 00:14:08 <[mbm]> hmm wtf, both python and python24 compile the same version of python Apr 04 00:14:19 <[mbm]> so we have two identical packages Apr 04 00:18:14 mbm * r6859 /packages/lang/python24/: remove duplicate python package Apr 04 00:18:18 <[mbm]> :P Apr 04 00:18:34 :-) Apr 04 02:31:19 anybody awake? in the current iteration of the flash mapping driver for the rdc3210, an attempt is made to rewrite the flash crc32 if the rootfs is jffs2 so as not to panic the bootloader after the rootfs changes, รก la brcm47xx-2.6, and the first flash block is constructed in memory correctly, but when it is flashed back, one stray byte (or word, dunno) gets inexplicably zeroed, thus changing what crc32 would be correct, and defeatin Apr 04 02:31:19 g the purpose of that part of the code. any ideas as to why this is happening? Apr 04 02:32:41 this is outside the svn tree for the moment, btw Apr 04 02:47:07 mbm: python 2.4 was there still for legacy Apr 04 02:47:22 python 2.5 was being brought in, but it was broken and hasnt been fixed yet Apr 04 02:47:34 either way, both of them were a little hokey. Apr 04 02:48:10 travis gave me some compiler errors for compiling on 64bit archs, but i didnt have that same issue Apr 04 02:48:18 locale was the main issue. Apr 04 02:48:52 anyway, it is jmj that is working on it, not me.. i've been busy with some other stuff. Apr 04 02:49:05 i'll mention it to him in the morning if he doesnt see this Apr 04 02:50:31 lst: let me know if you want me to test the new package Apr 04 02:54:22 the current one compiles, but it shouldnt actually work right Apr 04 02:54:31 from what jmj told me Apr 04 02:54:41 uclibc is missing locale support Apr 04 02:54:48 and python indirectly relies on it Apr 04 02:55:10 i dont know the full details, but i will bug him about it tomorrow Apr 04 02:55:29 ok Apr 04 02:55:43 i guarantee it will be working soon though, as i need it :) Apr 04 02:56:12 :-) Apr 04 02:56:22 which reminds me, i have 5 or 6 packages i've updated that i havent committed too Apr 04 02:56:26 i need to get on that :) **** ENDING LOGGING AT Wed Apr 04 02:59:57 2007