[Bug ld/32377] New: [RISCV]Variables within the range of gp ± 2KB have not been indexed

[mumuxi_ll at outlook dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=32377

            Bug ID: 32377
           Summary: [RISCV]Variables within the range of gp ± 2KB have not
                    been indexed
           Product: binutils
           Version: unspecified
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: ld
          Assignee: unassigned at sourceware dot org
          Reporter: mumuxi_ll at outlook dot com
  Target Milestone: ---

Hi all,
Here is the test case and ld file：

test.c:
================
int i = 1;
int main(){
  return i;
}
================

gp_reloc.ld:
================
MEMORY
{  
  ROM :  ORIGIN = 0x000030000    LENGTH = 0x10000
}


SECTIONS{
    .text : {
        *(.text)
        *(.rodata *.rodata.*)
    } >ROM

    .data : {
      *(.data)
      PROVIDE( __global_pointer$ = . + 0x400 );
       *(.sdata .sdata.* .sdata*)
    *(.gnu.linkonce.s.*)
    /* readonly data placed in RAM for access speed */
    . = ALIGN(8);
    *(.srodata.cst16)
    *(.srodata.cst8)
    *(.srodata.cst4)
    *(.srodata.cst2)
    *(.srodata .srodata.*)
     *(.sbss*)
    } >ROM

  PROVIDE( __bss_start = . );
    .bss (NOLOAD)   : ALIGN(8)
  {
    *(.sbss*)
    *(.gnu.linkonce.sb.*)
    *(.bss .bss.*)
    *(.gnu.linkonce.b.*)
    *(COMMON)
    . = ALIGN(4);
  } >ROM
  PROVIDE( _end = . );

}
================

Perform simple compilation and linking using the following command:
$riscv64-unknown-elf-gcc -march=rv32imac -mabi=ilp32 -c test.c -o test.o  
$riscv64-unknown-elf-gcc -march=rv32imac -mabi=ilp32 -o test test.o -T
gp_reloc.ld -Wl,-Map=test.map  
$riscv64-unknown-elf-objdump -d test> test.dasm

The disassembly of the main function is as follows, and it can be seen that the
variable i is indexed through the GP register
================
00030082 <main>:
   30082:       1141                    addi    sp,sp,-16
   30084:       c606                    sw      ra,12(sp)
   30086:       c422                    sw      s0,8(sp)
   30088:       0800                    addi    s0,sp,16
   3008a:       c041a783                lw      a5,-1020(gp) # 31728 <i>
   3008e:       853e                    mv      a0,a5
   30090:       40b2                    lw      ra,12(sp)
   30092:       4422                    lw      s0,8(sp)
   30094:       0141                    addi    sp,sp,16
   30096:       8082                    ret
================

When I change the value of `__globalis_pointer` to . + 0x800, the corresponding
disassembly is as follows:
================
0003008a <main>:
   3008a:       1141                    addi    sp,sp,-16
   3008c:       c606                    sw      ra,12(sp)
   3008e:       c422                    sw      s0,8(sp)
   30090:       0800                    addi    s0,sp,16
   30092:       000317b7                lui     a5,0x31
   30096:       7507a783                lw      a5,1872(a5) # 31750 <i>
   3009a:       853e                    mv      a0,a5
   3009c:       40b2                    lw      ra,12(sp)
   3009e:       4422                    lw      s0,8(sp)
   300a0:       0141                    addi    sp,sp,16
   300a2:       8082                    ret
================

At this point, it becomes indexing the variable i through the `lui+lw`,
however, upon reviewing the map file, it is clear that i falls within the range
of gp ± 2KB
================
                0x00031f4c       PROVIDE (__global_pointer$ = (. + 0x800))
 *(.sdata .sdata.* .sdata*)
 .sdata         0x0003174c        0x4 ......
                0x0003174c            .........
 .sdata         0x00031750        0x4 test.o
                0x00031750                i
================

So I want to know if this is a bug in ld or if there are additional
restrictions on gp.

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