| Message ID | 50BE0CDB.5000704@pengutronix.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Marc, On Tue, 2012-12-04 at 14:46 +0000, Marc Kleine-Budde wrote: > we've just discovered a linker problem with newer toolchains, that > support --build-id and xip kernels. The resulting vmlinux is about 13 > MiB big, but the xipimage ~2GiB. > > A look at the vmlinux's objdump shows that the .notes section is around > 2 GiB and it's placed by the linker script into the RAM. With the ROM > starting at 0x0 this results into a ~2 GiB big image. > > > [16] .data PROGBITS 88008000 0a0000 012aa0 00 WA 0 0 256 > > [17] .notes NOTE 8801aaa0 0b2aa0 000024 00 AX 0 0 4 > > [18] .bss NOBITS 8801aae0 0b2ac4 005414 00 WA 0 0 32 > > After discarding the .notes section in the objcopy (see patch below) the > resulting binary images has a proper size again. Could you please send me the configuration you're using to get the XIP kernel? I'll try to get my head around this problem and check how other architectures are dealing with the problem. Thanks! Pawe?
On 12/11/2012 02:01 PM, Pawel Moll wrote: > On Tue, 2012-12-04 at 14:46 +0000, Marc Kleine-Budde wrote: >> we've just discovered a linker problem with newer toolchains, that >> support --build-id and xip kernels. The resulting vmlinux is about 13 >> MiB big, but the xipimage ~2GiB. >> >> A look at the vmlinux's objdump shows that the .notes section is around >> 2 GiB and it's placed by the linker script into the RAM. With the ROM >> starting at 0x0 this results into a ~2 GiB big image. >> >>> [16] .data PROGBITS 88008000 0a0000 012aa0 00 WA 0 0 256 >>> [17] .notes NOTE 8801aaa0 0b2aa0 000024 00 AX 0 0 4 >>> [18] .bss NOBITS 8801aae0 0b2ac4 005414 00 WA 0 0 32 >> >> After discarding the .notes section in the objcopy (see patch below) the >> resulting binary images has a proper size again. > > Could you please send me the configuration you're using to get the XIP > kernel? I'll try to get my head around this problem and check how other > architectures are dealing with the problem. What about putting the notes section into the read only segment, i.e. the ROM? .config is attached. It's against 3.6.10 + some patches. The problem does not occur with our uclibc based toolchain, which doesn't crate a .notes section, but it still happens with a newlib based one. > ${CROSS_COMPILE}gcc -v > Using built-in specs. > COLLECT_GCC=/home/mol/vcs/OSELAS.Toolchain-2011.02/inst.keep/opt/OSELAS.Toolchain-2011.11.0/arm-cortexm3-eabi/gcc-4.7.2-newlib-1.19.0-binutils-2.22/bin/arm-cortexm3-eabi-gcc > COLLECT_LTO_WRAPPER=/home/mol/vcs/OSELAS.Toolchain-2011.02/inst.keep/opt/OSELAS.Toolchain-2011.11.0/arm-cortexm3-eabi/gcc-4.7.2-newlib-1.19.0-binutils-2.22/bin/../libexec/gcc/arm-cortexm3-eabi/4.7.2/lto-wrapper > Target: arm-cortexm3-eabi > Configured with: /home/mol/vcs/OSELAS.Toolchain-2011.02/platform-arm-cortexm3-eabi-gcc-4.7.2-newlib-1.19.0-binutils-2.22/build-cross/gcc-4.7.2/configure --build=x86_64-host-linux-gnu --host=x86_64-host-linux-gnu --target=arm-cortexm3-eabi --enable-multilib --with-float=soft --with-fpu=vfp --with-cpu=cortex-m3 --with-mode=thumb --with-newlib --disable-__cxa_atexit --enable-sjlj-exceptions --disable-nls --disable-decimal-float --disable-fixed-point --disable-win32-registry --enable-symvers=gnu --with-pkgversion=OSELAS.Toolchain-2011.11.0 --enable-threads=single --with-system-zlib --with-gmp=/home/mol/vcs/OSELAS.Toolchain-2011.02/platform-arm-cortexm3-eabi-gcc-4.7.2-newlib-1.19.0-binutils-2.22/sysroot-host --with-mpfr=/home/mol/vcs/OSELAS.Toolchain-2011.02/platform-arm-cortexm3-eabi-gcc-4.7.2-newlib-1.19.0-binutils-2.22/sysroot-host --prefix=/home/mol/vcs/OSELAS.Toolchain-2011.02/inst/opt/OSELAS.Toolchain-2011.11.0/arm-cortexm3-eabi/gcc-4.7.2-newlib-1.19.0-binutils-2.22 --enable-langua ges=c --enable-c99 --enable-long-long --enable-libstdcxx-debug --enable-profile --disable-shared --disable-libssp --enable-checking=release > Thread model: single > gcc version 4.7.2 (OSELAS.Toolchain-2011.11.0) Here the readelf output: ROM starts at 0x0 RAM starts at 88008000 > Section Headers: > [Nr] Name Type Addr Off Size ES Flg Lk Inf Al > [ 0] NULL 00000000 000000 000000 00 0 0 0 > [ 1] .bootloader PROGBITS 00000000 008000 000300 00 AX 0 0 8 > [ 2] .head.text PROGBITS 00000300 008300 00004c 00 AX 0 0 4 > [ 3] .text PROGBITS 00000360 008360 077844 00 AX 0 0 32 > [ 4] .text.init PROGBITS 00077ba4 07fba4 00009c 00 AX 0 0 4 > [ 5] .rodata PROGBITS 00078000 080000 00d9c8 00 A 0 0 32 > [ 6] __param PROGBITS 000859c8 08d9c8 0000a0 00 A 0 0 4 > [ 7] __modver PROGBITS 00085a68 08da68 000598 00 A 0 0 4 > [ 8] .ARM.unwind_idx ARM_EXIDX 00086000 08e000 000040 00 AL 3 0 4 > [ 9] .init.text PROGBITS 00086040 08e040 006b28 00 AX 0 0 8 > [10] .exit.text PROGBITS 0008cb68 094b68 0001f0 00 AX 0 0 4 > [11] .init.proc.info PROGBITS 0008cd58 094d58 000054 00 AX 0 0 2 > [12] .init.arch.info PROGBITS 0008cdac 094dac 000044 00 A 0 0 4 > [13] .init.tagtable PROGBITS 0008cdf0 094df0 000040 00 A 0 0 4 > [14] .init.data PROGBITS 0008ce30 094e30 000388 00 WA 0 0 4 > [15] .data PROGBITS 88008000 098000 012aa0 00 WA 0 0 256 > [16] .notes NOTE 8801aaa0 0aaaa0 000024 00 A 0 0 4 > [17] .bss NOBITS 8801aae0 0aaac4 005414 00 WA 0 0 32 > [18] .comment PROGBITS 00000000 0aaac4 000028 01 MS 0 0 1 > [19] .ARM.attributes ARM_ATTRIBUTES 00000000 0aaaec 00002b 00 0 0 1 > [20] .debug_line PROGBITS 00000000 0aab17 098048 00 0 0 1 > [21] .debug_info PROGBITS 00000000 142b5f 88b2f3 00 0 0 1 > [22] .debug_abbrev PROGBITS 00000000 9cde52 06a5dc 00 0 0 1 > [23] .debug_aranges PROGBITS 00000000 a38430 003b20 00 0 0 8 > [24] .debug_ranges PROGBITS 00000000 a3bf50 0296e0 00 0 0 8 > [25] .debug_frame PROGBITS 00000000 a65630 025388 00 0 0 4 > [26] .debug_str PROGBITS 00000000 a8a9b8 03ef39 01 MS 0 0 1 > [27] .debug_loc PROGBITS 00000000 ac98f1 169767 00 0 0 1 > [28] .shstrtab STRTAB 00000000 c33058 000141 00 0 0 1 > [29] .symtab SYMTAB 00000000 c33674 03bf70 10 30 11065 4 > [30] .strtab STRTAB 00000000 c6f5e4 0205da 00 0 0 1 I can put the kernel patches and the toolchain on our ftp for download. Marc
On Tue, 2012-12-11 at 13:43 +0000, Marc Kleine-Budde wrote: > What about putting the notes section into the read only segment, i.e. > the ROM? Yes, this is probably a good idea. I'll see to it, thanks! Pawe?
--- a/arch/arm/Makefile +++ b/arch/arm/Makefile @@ -18,7 +18,7 @@ ifeq ($(CONFIG_CPU_ENDIAN_BE8),y) LDFLAGS_vmlinux += --be8 endif -OBJCOPYFLAGS :=-O binary -R .comment -S +OBJCOPYFLAGS :=-O binary -R .comment -R .notes -S GZFLAGS :=-9 #KBUILD_CFLAGS +=-pipe