https://sourceware.org/bugzilla/show_bug.cgi?id=32896
--- Comment #8 from Matthew Malcomson <mmalcomson at nvidia dot com> --- (In reply to H.J. Lu from comment #6) > (In reply to Matthew Malcomson from comment #5) > > I think that this proposed patch could allow putting the PT_TLS segment > > offset past the end of the file (that I think was a problem in > > https://sourceware.org/bugzilla/show_bug.cgi?id=12921 ). > > > > This because it doesn't adjust the `off` that indicates position into the > > file. > > The updated patch is at: > > https://sourceware.org/pipermail/binutils/2025-April/140755.html I had a look at this patch, and it does seem to ensure that the PT_TLS segment has the correct alignment requirements. I noticed that with this patch one can end up with a very large PT_TLS filesize if you re-order the .tbss and .tdata sections. Not sure whether this would be problematic in some way. I did have to use a testcase that was triggering a linker warning so it may be invalid. vshcmd: > cat tbss-makes-huge-segment.s # Put enough data into the file to get to the next `maxpagesize` .section .rodata,"a",@progbits .align 4 .zero 0x10000 .section .tbss,"awT",@nobits .align 4 .type abc, @object .size abc, 4 abc: .zero 4 # Have .tdata and .tbss, so that the # `assign_file_positions_for_non_load_sections` function takes the file # position from the two separate sections. .section .tdata,"awT",@progbits .align 4 .type val, @object .size val, 4 val: .word 13 .text .global _start _start: tls-segfault [11:13:17] $ vshcmd: > # Only difference between default linker script and the one I'm using vshcmd: > # here is the location of the .tbss section (now before the .tdata vshcmd: > # section). vshcmd: > diff segment-too-large.ld \ vshcmd: > <(../build-aarch64-none-linux-gnu/install/bin/aarch64-none-linux-gnu-ld -verbose | sed -n '/=====/,/=====/p') > 0a1,10 > ================================================== > /* Script for -z combreloc */ > /* Copyright (C) 2014-2025 Free Software Foundation, Inc. > Copying and distribution of this script, with or without modification, > are permitted in any medium without royalty provided the copyright > notice and this notice are preserved. */ > OUTPUT_FORMAT("elf64-littleaarch64", "elf64-bigaarch64", > "elf64-littleaarch64") > OUTPUT_ARCH(aarch64) > ENTRY(_start) > SEARCH_DIR("=/aarch64-none-linux-gnu/lib64"); > SEARCH_DIR("=/usr/local/lib64"); SEARCH_DIR("=/lib64"); > SEARCH_DIR("=/usr/lib64"); SEARCH_DIR("=/aarch64-none-linux-gnu/lib"); > SEARCH_DIR("=/usr/local/lib"); SEARCH_DIR("=/lib"); SEARCH_DIR("=/usr/lib"); 98d107 < .tbss : { *(.tbss .tbss.* .gnu.linkonce.tb.*) *(.tcommon) } 103a113 > .tbss : { *(.tbss .tbss.* .gnu.linkonce.tb.*) *(.tcommon) } 245a256 > ================================================== tls-segfault [11:13:20] $ vshcmd: > ../build-aarch64-none-linux-gnu/install/bin/aarch64-none-linux-gnu-as \ vshcmd: > tbss-makes-huge-segment.s -o tbss-makes-huge-segment.o vshcmd: > ../build-aarch64-none-linux-gnu/install/bin/aarch64-none-linux-gnu-ld \ vshcmd: > -T segment-too-large.ld \ vshcmd: > tbss-makes-huge-segment.o -o tbss-makes-huge-segment vshcmd: > ../build-aarch64-none-linux-gnu/install/bin/aarch64-none-linux-gnu-readelf --segments --wide tbss-makes-huge-segment > tls-segfault [11:13:22] $ > > > ../build-aarch64-none-linux-gnu/install/bin/aarch64-none-linux-gnu-ld: > tbss-makes-huge-segment: section .tdata lma 0x42ffe0 adjusted to 0x42ffe4 tls-segfault [11:13:22] $ Elf file type is EXEC (Executable file) Entry point 0x400120 There are 4 program headers, starting at offset 64 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align LOAD 0x000000 0x0000000000400000 0x0000000000400000 0x010120 0x010120 R 0x10000 LOAD 0x01ffe0 0x000000000042ffe0 0x000000000042ffe0 0x000004 0x000004 RW 0x10000 TLS 0x000010 0x000000000042ffe0 0x000000000042ffe0 0x01ffd4 0x000008 R 0x10 GNU_RELRO 0x000010 0x000000000042ffe0 0x000000000042ffe0 0x000004 0x000020 R 0x1 Section to Segment mapping: Segment Sections... 00 .rodata 01 .tdata 02 .tbss .tdata 03 tls-segfault [11:13:22] $ -- You are receiving this mail because: You are on the CC list for the bug.
