Hi Heiko, On Mon, Feb 19, 2024 at 12:32:48PM +0100, Heiko Carstens wrote: > On Fri, Feb 16, 2024 at 12:55:53PM -0700, Nathan Chancellor wrote: > > When attempting to boot a kernel compiled with OBJCOPY=llvm-objcopy, > > there is a crash right at boot: > > > > Out of memory allocating 6d7800 bytes 8 aligned in range 0:20000000 > > Reserved memory ranges: > > 0000000000000000 a394c3c30d90cdaf DECOMPRESSOR > > Usable online memory ranges (info source: sclp read info [3]): > > 0000000000000000 0000000020000000 > > Usable online memory total: 20000000 Reserved: a394c3c30d90cdaf Free: 0 > > Call Trace: > > (sp:0000000000033e90 [<0000000000012fbc>] physmem_alloc_top_down+0x5c/0x104) > > sp:0000000000033f00 [<0000000000011d56>] startup_kernel+0x3a6/0x77c > > sp:0000000000033f60 [<00000000000100f4>] startup_normal+0xd4/0xd4 > > > > GNU objcopy does not have any issues. Looking at differences between the > > object files in each build reveals info.bin does not get properly > > populated with llvm-objcopy, which results in an empty .vmlinux.info > > section. > ... > > Closes: https://github.com/ClangBuiltLinux/linux/issues/1996 > > Link: https://github.com/llvm/llvm-project/commit/3c02cb7492fc78fb678264cebf57ff88e478e14f > > Suggested-by: Ulrich Weigand <ulrich.weigand@xxxxxxxxxx> > > Signed-off-by: Nathan Chancellor <nathan@xxxxxxxxxx> > > Thanks a lot, applied! > > However when building the kernel with "LLVM=1" I can see Thanks for testing, we are so close! > something else which looks like an llvm-objdump bug to me: > > $make LLVM=1 bzImage > ... > SECTCMP .boot.data > llvm-objdump: warning: section '.boot.data' mentioned in a -j/--section option, but not found in any input file > > This works without warning with GNU objcopy, and actually the output > is also different: > > $ objdump -v > GNU objdump (GNU Binutils for Ubuntu) 2.41 > > $ objdump -t -j .boot.data arch/s390/boot/vmlinux > > arch/s390/boot/vmlinux: file format elf64-s390 > > SYMBOL TABLE: > 0000000000020000 l O .boot.data 0000000000003000 sclp_info_sccb > 00000000000240e0 l O .boot.data 0000000000000001 sclp_info_sccb_valid > 0000000000023010 g O .boot.data 0000000000000008 ident_map_size > 0000000000023018 g O .boot.data 00000000000010c8 physmem_info > 00000000000250e2 g .boot.data 0000000000000000 __boot_data_end > 0000000000020000 g .boot.data 0000000000000000 __boot_data_start > 00000000000240e2 g O .boot.data 0000000000001000 early_command_line > 0000000000023008 g O .boot.data 0000000000000008 early_ipl_comp_list_size > 0000000000023000 g O .boot.data 0000000000000008 early_ipl_comp_list_addr > > While with llvm-copy: > > $ llvm-objdump --version > LLVM (http://llvm.org/): > LLVM version 19.0.0git > > $ llvm-objdump -t -j .boot.data arch/s390/boot/vmlinux > > arch/s390/boot/vmlinux: file format elf64-s390 > > SYMBOL TABLE: > 0000000000000200 l .head.text 0000000000000000 ipl_start > 0000000000010020 l .head.text 00000000000000d4 startup_normal > 00000000000101b0 l .head.text 00000000000000b2 startup_kdump > 0000000000010280 l .head.text 000000000000005a startup_pgm_check_handler > 000000000001025c l .head.text 0000000000000000 startup_kdump_relocated > 0000000000000000 l df *ABS* 0000000000000000 als.c > 000000000001e040 l O .rodata 0000000000000018 als > 000000000001f6f0 l O .data 0000000000000050 print_missing_facilities.als_str > 0000000000011800 l F .text 00000000000000e2 print_machine_type > ... > 0000000000020000 l O .boot.data 0000000000003000 sclp_info_sccb > 00000000000240e0 l O .boot.data 0000000000000001 sclp_info_sccb_valid > ... and so on (everything is dumped) > llvm-objdump: warning: section '.boot.data' mentioned in a -j/--section option, but not found in any input file > > So somehow llvmdump's "-j/--section" option doesn not seem to work. Re-reading Jordan's response to my initial report about this a couple of years ago at https://github.com/ClangBuiltLinux/linux/issues/859, I think I understand the issue now. '-j' / '--section' does work but not for '-t'. It works for '-h' and a couple of other flags: https://github.com/llvm/llvm-project/blob/61a96e5afadd034e7d13126f0e43731bbad7ad89/llvm/test/tools/llvm-objdump/section-filter.test and because '-t' does not work with '-j', there is a warning because the '-j' argument values went unhandled: https://github.com/llvm/llvm-project/blob/61a96e5afadd034e7d13126f0e43731bbad7ad89/llvm/tools/llvm-objdump/llvm-objdump.cpp#L485 https://github.com/llvm/llvm-project/blob/61a96e5afadd034e7d13126f0e43731bbad7ad89/llvm/tools/llvm-objdump/llvm-objdump.cpp#L3672 which is what Jordan's upstream report (https://llvm.org/pr50085) was about fixing. However, it may not be too hard to fix '-t' to work with '-j', I intend to take a look at it at some point this week. However, since I am more of a kernel hacker than I am an LLVM one, I came up with a potential solution in arch/s390/boot/Makefile, which is basically just filtering the symbol table manually with grep... it appears to produce stable results based on a small test Makefile I have to make sure the output looks sane before running through sha256sum. I'd be happy to send this as a formal patch if you'd accept it for full LLVM=1 compatibility with LLVM 19.0.0+ and Linux 6.9+. Cheers, Nathan diff --git a/arch/s390/boot/Makefile b/arch/s390/boot/Makefile index d40135efdec4..be3655825b4c 100644 --- a/arch/s390/boot/Makefile +++ b/arch/s390/boot/Makefile @@ -56,9 +56,9 @@ clean-files += vmlinux.map quiet_cmd_section_cmp = SECTCMP $* define cmd_section_cmp - s1=`$(OBJDUMP) -t -j "$*" "$<" | sort | \ + s1=`$(OBJDUMP) -t "$<" | grep "\s$*\s\+" | sort | \ sed -n "/0000000000000000/! s/.*\s$*\s\+//p" | sha256sum`; \ - s2=`$(OBJDUMP) -t -j "$*" "$(word 2,$^)" | sort | \ + s2=`$(OBJDUMP) -t "$(word 2,$^)" | grep "\s$*\s\+" | sort | \ sed -n "/0000000000000000/! s/.*\s$*\s\+//p" | sha256sum`; \ if [ "$$s1" != "$$s2" ]; then \ echo "error: section $* differs between $< and $(word 2,$^)" >&2; \
