[Bug ld/29803] New: Relax failed between different output section

[lifang_xia at linux dot alibaba.com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29803

            Bug ID: 29803
           Summary: Relax failed between different output section
           Product: binutils
           Version: 2.40 (HEAD)
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: ld
          Assignee: unassigned at sourceware dot org
          Reporter: lifang_xia at linux dot alibaba.com
  Target Milestone: ---

Hi all,
I have met a problem in risc-v relaxation. But I think it would be generic bug,
not only in risc-v.
Here is a case for riscv:
```
section .text.main
.global main
main:
        call foo
        call foo
        call foo
        call foo
        call foo
        call foo
        call foo
        .size main, . - main

.section .text.foo
.global foo
foo:
        .option norvc
        .rept 504
        nop
        .endr
        call bar
        .option rvc

        ret
        .size foo, . - foo

.section .bar.text
.global bar
bar:
        ret
        .size bar, . - bar
```

build with as:
```
./gas/as-new -o 1.o 1.s -march=rv32gc -mabi=ilp32d -mrelax
```

link file:
```
MEMORY
{
        RAM1 : ORIGIN = 0x0, LENGTH = 0x1000
        RAM2 : ORIGIN = 0x1000, LENGTH = 0x400
}

SECTIONS
{
        .text :
        {
                *(.text)
                *(.text.*)
        } > RAM1

        .bar.text :
        {
                *(.bar.text)
        } > RAM2
}
```

link command:
```
./ld/ld-new -o 1 1.o  -T1.ld -m elf32lriscv
```

It will get an error: 
(.text.foo+0x7e0): relocation truncated to fit: R_RISCV_RVC_JUMP against symbol
`bar' defined in .bar.text section in 1.o

Because of the .foo.text's output_offset(base address) doest not update after
the relaxtion in main in time. (ldlang.c -> lang_size_sections_1 -> case
lang_input_section_enum)
```
 case lang_input_section_enum:
   {
     asection *i;

     i = s->input_section.section;
     if (relax)
       {
         bool again;

         if (!bfd_relax_section (i->owner, i, &link_info, &again))
           einfo (_("%F%P: can't relax section: %E\n"));
         if (again)
           *relax = true;
       }

     dot = size_input_section (prev, output_section_statement,
                               fill, &removed, dot);
   }
   break;

```
the size_input_section only update the size of current input section. But it
doest not update the next input section's offset(base address).

How about add lang_size_sections_1 after bfd_relax_section?

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