aboutsummaryrefslogtreecommitdiffstats
path: root/target/linux/ar71xx/image/lzma-loader
Commit message (Collapse)AuthorAgeFilesLines
* ar71xx: lzma-loader: constify kernel argv arrayMatthias Schiffer2018-06-061-1/+1
| | | | | | | By making the kernel argv array const, the .data section can always be omitted from the laoder binary. Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
* ar71xx: lzma-loader: set page size to 4KBMatthias Schiffer2018-06-061-1/+1
| | | | | | | | The text section in the ELF loader is aligned to the maximum page size, which defaults to 64KB. Reduce it to the actual page size to avoid wasting flash space for this alignment. Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
* ar71xx: lzma-loader: move padding workaround to gzip stepMatthias Schiffer2018-06-062-2/+5
| | | | | | | | | | | | | | | | | | | | | | | Some devices (TP-Link TL-WR1043ND v1) don't boot reliably when the uncompressed loader is too small. This was workarounded in the loader by adding 512KB of padding to the .data section of the loader binary. This approach had two issues: - The padding was only working when .data was non-empty (otherwise the section would become NOBITS, omitting it in the binary). .data was only empty when no CMDLINE was set, leading to further workarounds like fe594bf90d09 ("ath79: fix loader-okli, lzma-loader"), and this workaround was only effective because a missing "const" led to the kernel argv being stored in .data instead of .rodata - The padding was not only added to the compressed .gz loader, but also uncompressed .bin and .elf loaders. The prevented embedding the kernel cmdline in the loader for non-gz loader types. To fix both issues, move the creation of the padding from the linker script to the gzip step. Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
* ar71xx: fix lzma-loader build with glibcJonas Gorski2017-12-131-1/+1
| | | | | | | | | | | | | | | | For an unknown reason gcc tries to link in crti.o when building with a glibc toolchain (this does not happen with other targets). Prevent this by telling gcc explicitly to not do that. Fixes the following build error: /home/jonas/git/lede/staging_dir/toolchain-mips_24kc_gcc-5.5.0_glibc/lib/gcc/mips-openwrt-linux-gnu/5.5.0/../../../../mips-openwrt-linux-gnu/lib/crti.o: In function `_init': (.init+0x18): relocation truncated to fit: R_MIPS_GOT16 against `__gmon_start__' /home/jonas/git/lede/staging_dir/toolchain-mips_24kc_gcc-5.5.0_glibc/lib/gcc/mips-openwrt-linux-gnu/5.5.0/../../../../mips-openwrt-linux-gnu/lib/crti.o: In function `_init': (.init+0x28): relocation truncated to fit: R_MIPS_CALL16 against `__gmon_start__' collect2: error: ld returned 1 exit status Signed-off-by: Jonas Gorski <jonas.gorski@gmail.com>
* ar71xx: lzma loader use LTOJulien Dusser2017-10-241-3/+5
| | | | | | | Change the Makefile to use LTO for better code optimisations. Gains are very low, only 270 bytes saved, but it's only Makefile changes. Signed-off-by: Julien Dusser <julien.dusser@free.fr>
* ar71xx: fix lzma loader performance issuesJulien Dusser2017-10-241-0/+13
| | | | | | | | | | | Some bootloaders set a cache cohenrency to a very slow mode. Use code from Linux kernel to set it to "Cacheable, noncoherent, write-back, write allocate". Perfomance impact is significant on TP-Link EAP245 board, kernel decompression time fall from 33 seconds to less than 1. Signed-off-by: Julien Dusser <julien.dusser@free.fr>
* ar71xx/lzma-loader: fix O32 ABI conformanceGabor Juhos2016-01-171-0/+3
| | | | | | | | | | | | | | | | | | | | | | According to the calling convention of the o32 ABI the caller function must reserve stack space for $a0-$a3 registers in case the callee needs to save its arguments. The assembly code of the loader does not reserve stack space for these registers thus when the 'loader_main' function needs to save its arguments, those will be stored in the 'workspace' area instead of the stack. Because the workspace area is also used by other part of the code, the saved register values gets overwritten and this often leads to failed kernel boots. Fix the code to reserve stack space for the registers to avoid this error. Signed-off-by: Gabor Juhos <juhosg@openwrt.org> SVN-Revision: 48279
* ar71xx: ensure that LOADADDR is passed to lzma-loaderFelix Fietkau2016-01-031-0/+2
| | | | | | Signed-off-by: Felix Fietkau <nbd@openwrt.org> SVN-Revision: 48101
* ar71xx: fix build of lzma-loader with binutils 2.25.1Hauke Mehrtens2015-09-121-1/+2
| | | | | | | | | | | | | | | | To make the lzma-loader build with binutils 2.25.1 it is needed to remove the .MIPS.abiflags section otherwise I get the following compile error: mips-openwrt-linux-musl-objcopy -O binary -R .reginfo -R .note -R .comment -R .mdebug -S loader loader.bin mips-openwrt-linux-musl-objcopy: Warning: Writing section `.text' to huge (ie negative) file offset 0xffffffff80a00000. mips-openwrt-linux-musl-objcopy:loader.bin[.text]: File truncated This should close #20487. Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de> SVN-Revision: 46893
* ar71xx: fix lzma-loader kernel command line passingFelix Fietkau2015-08-021-1/+2
| | | | | | | | | The generic kernel cmdline parser ignores argv[0], this caused a regression for all lzma-loader based boards with linux 4.1 Signed-off-by: Felix Fietkau <nbd@openwrt.org> SVN-Revision: 46544
* ar71xx/image: relink loader.elf image to LOADADDRGabor Juhos2012-11-202-5/+21
| | | | | | | | | This allows to load the resulting image directly from RouterBOOT. Signed-off-by: Gabor Juhos <juhosg@openwrt.org> SVN-Revision: 34271
* ar71xx: image: stop the switch from the lzma-loader on the TL-WR1043NDGabor Juhos2012-08-022-0/+745
| | | | SVN-Revision: 32946
* ar71xx: image: allow to use board specific code in the lzma-loaderGabor Juhos2012-08-022-1/+8
| | | | SVN-Revision: 32945
* ar71xx: add lzma loaderGabor Juhos2011-12-0517-0/+1941
SVN-Revision: 29443
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-06 01:57:43 +0000