[Bug ld/30791] New: Do not merge sections with same name and SHF_LINK_ORDER, but different sh_link value

[zimao at microsoft dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=30791

            Bug ID: 30791
           Summary: Do not merge sections with same name and
                    SHF_LINK_ORDER, but different sh_link value
           Product: binutils
           Version: unspecified
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: ld
          Assignee: unassigned at sourceware dot org
          Reporter: zimao at microsoft dot com
  Target Milestone: ---

When a section has both SHF_LINK_ORDER and sh_link set, it means this section
is kind of a child section of the linked to section.

When merging two sections with the same name, both have SHF_LINK_ORDER set, but
different sh_link value, one of the sh_link information will be absent in the
merged section.

In a real world example, both GCC (13.1) and clang (14.0) utilize this behavior
for the "-ffunction-sections -fpatchable-function-entry=N". The compiler will
generate multiple sections all named as "__patchable_function_entries", with
different sh_link. Each section will be linked its own .text section.

But when running code like `ld -r -o merged.o a.o b.o`, the current ld will
merge all these "__patchable_function_entries" section into a single section.

On the other hand, on the LLVM side, lld will keep these "pfe" sections
separately after the linking.
ref link: https://reviews.llvm.org/D68094

I think ld should not merge these sections as well, just like the behavior in
lld.


Example code:

$ cat a.c
int a(){return 1;}
int b(){return 2;}

$ cat b.c
int c(){return 3;}
int d(){return 4;}

$ gcc -c -ffunction-sections -fpatchable-function-entry=2 a.c b.c
$ readelf -S a.o

  [Nr] Name              Type             Address           Offset
       Size              EntSize          Flags  Link  Info  Align
  [ 3] .text.a           PROGBITS         0000000000000000  00000040
       000000000000000d  0000000000000000  AX       0     0     16
  [ 4] __patchable_[...] PROGBITS         0000000000000000  00000050
       0000000000000008  0000000000000000 WAL       3     0     8
  [ 5] .rela__patch[...] RELA             0000000000000000  00000188
       0000000000000018  0000000000000018   I      14     4     8
  [ 6] .text.b           PROGBITS         0000000000000000  00000060
       000000000000000d  0000000000000000  AX       0     0     16
  [ 7] __patchable_[...] PROGBITS         0000000000000000  00000070
       0000000000000008  0000000000000000 WAL       6     0     8
  [ 8] .rela__patch[...] RELA             0000000000000000  000001a0
       0000000000000018  0000000000000018   I      14     7     8

(some content removed, same result for the b.o)

$ ld -r -o merged.o a.o b.o
readelf -S merged.o | grep -E "(text)|(patchable)" -A1
  [ 1] .text             PROGBITS         0000000000000000  00000040
       0000000000000000  0000000000000000  AX       0     0     4
  [ 2] .text.a           PROGBITS         0000000000000000  00000040
       000000000000000d  0000000000000000  AX       0     0     16
  [ 3] .text.b           PROGBITS         0000000000000000  00000050
       000000000000000d  0000000000000000  AX       0     0     16
  [ 4] .text.c           PROGBITS         0000000000000000  00000060
       000000000000000d  0000000000000000  AX       0     0     16
  [ 5] .text.d           PROGBITS         0000000000000000  00000070
       000000000000000d  0000000000000000  AX       0     0     16
--
  [ 8] __patchable_[...] PROGBITS         0000000000000000  00000130
       0000000000000020  0000000000000000 WAL       2     0     8

$ ld.lld -r -o merged-2.o a.o b.o
$ readelf -S merged-2.o | grep -E "(text)|(patchable)" -A1
  [ 1] .text             PROGBITS         0000000000000000  00000040
       0000000000000000  0000000000000000  AX       0     0     4
  [ 2] .text.a           PROGBITS         0000000000000000  00000040
       000000000000000d  0000000000000000  AX       0     0     16
  [ 3] __patchable_[...] PROGBITS         0000000000000000  00000050
       0000000000000008  0000000000000000 WAL       2     0     8
--
  [ 5] .text.b           PROGBITS         0000000000000000  00000060
       000000000000000d  0000000000000000  AX       0     0     16
  [ 6] __patchable_[...] PROGBITS         0000000000000000  00000070
       0000000000000008  0000000000000000 WAL       5     0     8
--
  [12] .text.c           PROGBITS         0000000000000000  00000130
       000000000000000d  0000000000000000  AX       0     0     16
  [13] __patchable_[...] PROGBITS         0000000000000000  00000140
       0000000000000008  0000000000000000 WAL      12     0     8
--
  [15] .text.d           PROGBITS         0000000000000000  00000150
       000000000000000d  0000000000000000  AX       0     0     16
  [16] __patchable_[...] PROGBITS         0000000000000000  00000160
       0000000000000008  0000000000000000 WAL      15     0     8

-- 
You are receiving this mail because:
You are on the CC list for the bug.