[Bug target/63842] New: x86-64 large PIC model may load the GOT base into the wrong register

[Bug target/63842] New: x86-64 large PIC model may load the GOT base into the wrong register

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63842

            Bug ID: 63842
           Summary: x86-64 large PIC model may load the GOT base into the
                    wrong register
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hjl.tools at gmail dot com
                CC: ubizjak at gmail dot com
            Target: x86-64

According to x86-64 psABI, when PLT is more 2GB from GOT, linker must use
large model PLT layout:

PLT0: pushq 8(%r15)   # GOT[1]
      jmpq  *16(%r15) # GOT[2]

That is fpr large PIC mode,  if a function calls another function via PLT,
it must load GOT base into %r15 before making function call. However,
GCC generates:

[hjl@gnu-6 pr17592]$ cat foo.c
#include <stdio.h>

void
foo (void)
{
  printf ("hello\n");
}
[hjl@gnu-6 pr17592]$ make foo.s
gcc -O2  -mcmodel=large -fpic -S foo.c
[hjl@gnu-6 pr17592]$ cat foo.s
    .file    "foo.c"
    .section    .rodata.str1.1,"aMS",@progbits,1
.LC0:
    .string    "hello"
    .text
    .p2align 4,,15
    .globl    foo
    .type    foo, @function
foo:
.LFB11:
    .cfi_startproc
.L2:
    leaq    .L2(%rip), %rcx
    movabsq    $_GLOBAL_OFFSET_TABLE_-.L2, %r11
    movabsq    $.LC0@GOTOFF, %rax
    addq    %r11, %rcx
    leaq    (%rcx,%rax), %rdi
    movabsq    $puts@PLTOFF, %rax
    addq    %rcx, %rax
    jmp    *%rax
    .cfi_endproc
.LFE11:
    .size    foo, .-foo
    .ident    "GCC: (GNU) 4.8.3 20140911 (Red Hat 4.8.3-7)"
    .section    .note.GNU-stack,"",@progbits
[hjl@gnu-6 pr17592]$ 

It works fine as long as large PLT layout isn't used.  But it fails
when large PLT layout is used.  Since puts is called via PLT, GCC
must load GOT base into %r15, instead of %r11.

[Bug target/63842] x86-64 large PIC model may load the GOT base into the wrong register

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63842

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> ---
The proper code sequence should be

pushq %r15
load GOT base into %r15
load address of puts PLT entry into %rdx
call *(%rdx)
popq %r15

That is similar to 32-bit.

[Bug target/63842] x86-64 large PIC model may load the GOT base into the wrong register

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63842

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-11-27
     Ever confirmed|0                           |1

--- Comment #2 from H.J. Lu <hjl.tools at gmail dot com> ---
r218132 generates:

main:
.LFB11:
    .cfi_startproc
.L2:
    leaq    .L2(%rip), %rax
    movabsq    $_GLOBAL_OFFSET_TABLE_-.L2, %r11
    movabsq    $.LC0@GOTOFF, %rdx
    pushq    %r15
    .cfi_def_cfa_offset 16
    .cfi_offset 15, -16
    addq    %r11, %rax
    leaq    (%rax,%rdx), %rdi
    movabsq    $puts@PLTOFF, %rdx
    movq    %rax, %r15
    addq    %rax, %rdx
    call    *%rdx
    xorl    %eax, %eax
    popq    %r15
    .cfi_def_cfa_offset 8
    ret
    .cfi_endproc
.LFE11:

It looks OK.  But it is hard to test without a linker supporting large
model PLT.  I will leave it open until I finish large model PLT support
in binutils so that I can verify it.

[Bug target/63842] x86-64 large PIC model may load the GOT base into the wrong register

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63842

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
   Target Milestone|---                         |5.0
             Status|NEW                         |RESOLVED

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed.

[Bug target/63842] x86-64 large PIC model may load the GOT base into the wrong register

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63842

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|FIXED                       |DUPLICATE

--- Comment #4 from H.J. Lu <hjl.tools at gmail dot com> ---
Dup.

*** This bug has been marked as a duplicate of bug 63833 ***