Hi Oleksij, ----- On 6 Jun, 2020, at 12:11, Oleksij Rempel linux@xxxxxxxxxxxxxxxx wrote: > Hi Clement, > > This changes are needed to make your patch partially work on my system: > https://github.com/olerem/barebox/commit/90ad4f3cd8cbcfd361ff780bcfce1573ae73053d Thanks for investigating and sorry for the remainign errors. Some xzalloc checks can probably be omitted or I think I should replace the existing xzalloc by calloc to allow failing on elf loading (which seems more sane). > > I found following issues: > - the elf->list was not initiated My bad, I missed it while moving elf init to a specific function... I started to wonder what I had tested side but It worked because the list is only used in elf_release_regions (which wasn't called). > - my system has only 64MiB RAM and only 8MiB is available for barebox > alloc. Since you patch is reading complete file in to alloc, i was not > able to start kernel which was already copied to alloc by: > cp /mnt/tftp/kernel . Indeed, strangely, mips code used to do a read_file before loading the elf file in the mips bootm code. There is probably something different but I can't quite get it. Anyway, it will be way better by not reading the whole file. > This is still not fixed, but with some sanity checks the user is > notified about it. > - lseek(fd, 0, SEEK_SET); is not working on tftp mount point. This was > the reason why elf header actually passed check, but had a trash in > elf->buf later. Arg, and I did not checked lseek return value... I will use open/close as you did. > > On systems with limited amount of RAM, it would be better to read the > header and request RAM regions on first stand, and then directly read > them from file in to the requested regions. Ok, I will modify the elf loader to do so. Apparently, tftp supports forward lseek so it should be possible. > > And there is one more thing, by using "dryrun" function (boot(m) -d), i > would expect to be able to see reserved regions with "iomem" command. > Currently it is not the case. I can modify that but I guess this is the default behavior no ? In the bootm command: ret = handler->bootm(data); if (data->dryrun) printf("Dryrun. Aborted\n"); err_out: if (data->os_res) release_sdram_region(data->os_res); if (data->initrd_res) release_sdram_region(data->initrd_res); So if the dryrun argument was given, the complete err_out path is taken and lead to releasing all regions. But I can do it if needed. Thanks again for testing and sorry for letting such things pass in the patches. Clément > > Am 10.05.20 um 13:31 schrieb Clément Leger: >> Hi Oleksij, >> >> ----- On 10 May, 2020, at 06:31, Oleksij Rempel o.rempel@xxxxxxxxxxxxxx wrote: >> >>> Hi Clement, >>> >>> just in case it makes a difference. I load this file over tftp (boot >>> net) >> >> I manage to load your elf file on kvx using bootm (I do not have net boot) >> and here is what I get: >> >> -------------------------------------------- >> Board: KONIC 200 (K200) >> malloc space: 0x110306050 -> 0x200000000 (size 3.7 GiB) >> export: No such file or directory >> >> Hit any to stop autoboot: 3 >> barebox:/ bootm env/ore-linux-dpt-module >> Entry Point: 80980000 >> >> Loading ELF 'env/ore-linux-dpt-module' >> Loading phdr to 0x0000000080980000 (2740416 bytes) >> ... >> -------------------------------------------- >> >> So one segment is loaded which is correct according to readelf. >> elf file is correct and can be loaded using bootm. >> >> Could you try loading it using bootm directly ? >> This will confirm if it's related to net boot or not (although I don't >> know what are the underlying differences). >> >> Thanks, >> >> Clément >> >>> >>> On Sat, May 09, 2020 at 09:24:55PM +0200, Clément Leger wrote: >>>> Hi Oleksij, >>>> >>>> ----- On 9 May, 2020, at 18:51, Oleksij Rempel o.rempel@xxxxxxxxxxxxxx wrote: >>>> >>>>> On Sat, May 09, 2020 at 04:51:42PM +0200, Oleksij Rempel wrote: >>>>>> Hi Clement, >>>>>> >>>>>> suddenly it is still exploding. I'll try to investigate the reason. >>>>> >>>>> common/elf.c: >>>>> load_elf_image_phdr() >>>>> elf_hdr_e_phnum() <-- returns 0 >>>>> for (i = 0; i < elf_hdr_e_phnum(elf, buf) ; ++i) { >>>>> ... so we will never get here. >>>>> >>>>> Probably we should add a sanity check here and be more verbose in this >>>>> case. >>>> >>>> There is something odd because your elf is correct and has 2 program >>>> headers. So this should not be a problem (but a sanity check makes >>>> sense !). >>>> >>>> I suspect that the elf size computation is wrong with your elf file >>>> and leads to incompletely loaded elf file but I tried to compute >>>> it using the sequence of code used in barebox and everything seems ok. >>>> Maybe there is something with the endianess/elfclass going wrong. >>>> I tried again on kvx to be sure I did not made a mistake but it works >>>> and the elf entry address is correctly used. >>>> I'll try to add handling for different endianess and load your elf >>>> file on kvx to debug that. >>>> >>>>> >>>>> here is my elf image: >>>>> https://github.com/olerem/barebox/blob/new-elf/ore-linux-dpt-module >>>> >>>> I succeeded in building Distrokit images but I can't get it under barebox. >>>> There is only 4M oof RAM and each time I tries to modify it, it won't boot >>>> anymore. >>>> I tried integrating the kernel (and your eore-linux-dpt-module elf) in >>>> the environment but it also won't boot. Do you have any idea on how to >>>> do it ? >>>> Even if the kernel won't start under qemu , at least I'll be able to >>>> understand what is wrong. >>>> >>>> Thanks for testing, >>>> >>>> Clément >>>> >>>>> >>>>>> On Fri, May 08, 2020 at 07:04:04PM +0200, Clement Leger wrote: >>>>>>> Currently, when booting an elf file using "bootm /dev/mtdx", bootm will >>>>>>> simply pass the file to the bootm and the read done on it will read the >>>>>>> entire flash partition. This series starts by some cleanup and then add an >>>>>>> elf_open function to load the elf file size only based on the elf header. >>>>>>> A special handling for the elf file is also added in bootm data to allow >>>>>>> using directly the elf file structure. Finally the mips bootm is modified >>>>>>> to use bootm_load_os directly instead of manual elf loading. >>>>>>> >>>>>>> Compilation for both mips and arm has been tested but run on qemu-malta was not >>>>>>> possible. Changing the MALLOC_SIZE to allow loading a kernel always lead to a >>>>>>> non-bootable system. Changes have been tested on kvx architecture for which >>>>>>> bootm support has been added and will be submitted. >>>>>>> >>>>>>> Changes v3 -> v4 >>>>>>> - Fix init of elf entry address to be used by bootm_load_elf >>>>>>> >>>>>>> Changes v2 -> v3 >>>>>>> - Integrate elf loading in bootm_load_os >>>>>>> - Add patch to remove now unused elf_load_image/elf_release_image >>>>>>> - Use malloc instead of xmalloc and check return value >>>>>>> >>>>>>> Changes v1 -> v2 >>>>>>> - Add BOOTM_ELF config to select elf support and add checks in code >>>>>>> - Add an elf_get_mem_size function to avoid computing elf size in bootm.c >>>>>>> - Use xmalloc and read_full in elf_open instead of xzalloc/read >>>>>>> - Fix data->elf NULL reset >>>>>>> - Remove elf struct entirely from mips bootm code >>>>>>> >>>>>>> Clement Leger (7): >>>>>>> common: elf: add computation of elf boundaries >>>>>>> common: elf: fix warning on 32 bits architectures >>>>>>> common: elf: split init to be reused from other function >>>>>>> common: elf: add elf_open, elf_close and elf_load >>>>>>> common: bootm: add support for elf file loading >>>>>>> mips: lib: bootm: use bootm elf loading capabilities >>>>>>> common: elf: remove elf_load_image/elf_release_image >>>>>>> >>>>>>> arch/mips/lib/bootm.c | 31 +++++------- >>>>>>> common/Kconfig | 8 +++ >>>>>>> common/bootm.c | 33 +++++++++++++ >>>>>>> common/elf.c | 111 +++++++++++++++++++++++++++++++++++------- >>>>>>> include/bootm.h | 3 ++ >>>>>>> include/elf.h | 16 +++++- >>>>>>> 6 files changed, 163 insertions(+), 39 deletions(-) >>>>>>> >>>>>>> -- >>>>>>> 2.17.1 >>>>>>> >>>>>>> >>>>>>> _______________________________________________ >>>>>>> barebox mailing list >>>>>>> barebox@xxxxxxxxxxxxxxxxxxx >>>>>>> http://lists.infradead.org/mailman/listinfo/barebox >>>>>>> >>>>>> >>>>>> -- >>>>>> Pengutronix e.K. | | >>>>>> Steuerwalder Str. 21 | http://www.pengutronix.de/ | >>>>>> 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | >>>>>> Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 | >>>>> >>>>> >>>>> >>>>> -- >>>>> Pengutronix e.K. | | >>>>> Steuerwalder Str. 21 | http://www.pengutronix.de/ | >>>>> 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | >>>>> Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 | >>>> >>>> _______________________________________________ >>>> barebox mailing list >>>> barebox@xxxxxxxxxxxxxxxxxxx >>>> http://lists.infradead.org/mailman/listinfo/barebox >>> >>> -- >>> Pengutronix e.K. | | >>> Steuerwalder Str. 21 | http://www.pengutronix.de/ | >>> 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | >>> Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 | >> >> _______________________________________________ >> barebox mailing list >> barebox@xxxxxxxxxxxxxxxxxxx >> http://lists.infradead.org/mailman/listinfo/barebox >> > > > -- > Regards, > Oleksij _______________________________________________ barebox mailing list barebox@xxxxxxxxxxxxxxxxxxx http://lists.infradead.org/mailman/listinfo/barebox
