https://sourceware.org/bugzilla/show_bug.cgi?id=31921
Bug ID: 31921
Summary: [ARM][2.36] Linker produces bad executable
Product: binutils
Version: 2.36
Status: UNCONFIRMED
Severity: normal
Priority: P2
Component: ld
Assignee: unassigned at sourceware dot org
Reporter: christoph.franck at arcor dot de
Target Milestone: ---
Created attachment 15591
--> https://sourceware.org/bugzilla/attachment.cgi?id=15591&action=edit
Code that reproduces the issue
ld appears to produce a bad executable when the section .arm.exidx had
different LMA and VMA.
The attached file contains two linker control files,
STM32H563AIIXQ_FLASH2RAM_GOOD.ld and STM32H563AIIXQ_FLASH2RAM_BAD.ld. The
difference between the two files is that in _GOOD, .arm.exidx is in an output
section with identical LMA and VMA (>FLASH) while in _BAD, .arm.exidx is in an
output section with different LMA and VMA (>RAM AT> FLASH).
Building with the two linker control file yields /debug/linker_bug_GOOD.elf and
linker_bug_BAD.elf respectively.
In linker_bug_GOOD.elf, the _init and _fini functions are functional as
expected:
arm-none-eabi-objdump Linker_Bug_GOOD.elf -D --start-address=0x2000049C
--stop-address=0x200004b4
Linker_Bug_GOOD.elf: file format elf32-littlearm
Disassembly of section .data:
2000049c <_init>:
2000049c: b5f8 push {r3, r4, r5, r6, r7, lr}
2000049e: bf00 nop
200004a0: bcf8 pop {r3, r4, r5, r6, r7}
200004a2: bc08 pop {r3}
200004a4: 469e mov lr, r3
200004a6: 4770 bx lr
200004a8 <_fini>:
200004a8: b5f8 push {r3, r4, r5, r6, r7, lr}
200004aa: bf00 nop
200004ac: bcf8 pop {r3, r4, r5, r6, r7}
200004ae: bc08 pop {r3}
200004b0: 469e mov lr, r3
200004b2: 4770 bx lr
In linker_bug_BAD.elf, _init appears to be incomplete and program execution
falls through into _fini. The stack is left in an inconsistent state when the
code execution "returns" from a call to _init. The CPU produces a hard fault
shortly after the return.
arm-none-eabi-objdump Linker_Bug_BAD.elf -D --start-address=0x200004A4
--stop-address=0x200004b4
Linker_Bug_BAD.elf: file format elf32-littlearm
Disassembly of section .data:
200004a4 <_init>:
200004a4: b5f8 push {r3, r4, r5, r6, r7, lr}
200004a6: bf00 nop
200004a8 <_fini>:
200004a8: b5f8 push {r3, r4, r5, r6, r7, lr}
200004aa: bf00 nop
200004ac: bcf8 pop {r3, r4, r5, r6, r7}
200004ae: bc08 pop {r3}
200004b0: 469e mov lr, r3
200004b2: 4770 bx lr
Depending on the rest of the code, it looks like the issue can also affect
other functions than _init/_fini.
ld version:
arm-none-eabi-ld --version
GNU ld (GNU Tools for STM32 10.3-2021.10.20230912-2000) 2.36.1.20210621
This issue does not occur with more recent versions of binutils, but I did not
find a corresponding bug report for ld version 2.36.
--
You are receiving this mail because:
You are on the CC list for the bug.
