[compile] Relocate memcpy

[compile] Relocate memcpy

[Yao Qi]

I triage the fails in gdb.compile on arm-linux,

(gdb) compile code -- ;
(gdb) compile print vararray

Program received signal SIGILL, Illegal instruction.
0xf77eb01c in _gdb_expr (__regs=0xf77ea000, __gdb_out_param=0xf77e9000) at gdb command line:3
3	gdb command line: No such file or directory.
The program being debugged was signaled while in a function called from GDB.
GDB remains in the frame where the signal was received.
To change this behavior use "set unwindonsignal on".
Evaluation of the expression containing the function
(_gdb_expr) will be abandoned.

the code in the inferior is,

   0xf77eb016 <+22>:	f9 68	ldr	r1, [r7, #12]
   0xf77eb018 <+24>:	38 68	ldr	r0, [r7, #0]
   0xf77eb01a <+26>:	52 dd	ble.n	0xf77eb0c2
=> 0xf77eb01c <+28>:	f4 ff 00 bf			; <UNDEFINED> instruction: 0xfff4bf00
   0xf77eb020 <+32>:	10 37	adds	r7, #16
   0xf77eb022 <+34>:	bd 46	mov	sp, r7
   0xf77eb024 <+36>:	80 bd	pop	{r7, pc}

and the code in object file is (output from objdump)

  16:	68f9      	ldr	r1, [r7, #12]
  18:	6838      	ldr	r0, [r7, #0]
  1a:	f7ff fffe 	bl	0 <memcpy>
  1e:	bf00      	nop
  20:	3710      	adds	r7, #16
  22:	46bd      	mov	sp, r7
  24:	bd80      	pop	{r7, pc}

Looks the instruction "bl memcpy" is patched incorrectly.  Can anyone
help me to find which part I should dig it deeper?

I do "set debug compile 1", and see the log,

allocated 0x26 bytes at 0xf77eb000 prot 5
ELF mst_text symbol "memcpy" relocated to 0xf76b7ac0
allocated 0x1 bytes at 0xf77ea000 for registers
allocated 0x14 bytes at 0xf77e9000 for printed value

(gdb) p memcpy
$1 = {<text variable, no debug info>} 0xf76b7ac0 <memcpy>

My understanding is some thing patches "bl memcpy" with the address told
by gdb.  I don't know the patching is done by libcc1 or something else.

Note that I post my question here instead of gcc@gcc.gnu.org because
libcc1 is used by gdb, and was done by gdb people.

-- 
Yao (齐尧)

Re: [compile] Relocate memcpy

[Jan Kratochvil]

On Wed, 24 Jan 2018 13:23:14 +0100, Yao Qi wrote:
> My understanding is some thing patches "bl memcpy" with the address told
> by gdb.  I don't know the patching is done by libcc1 or something else.

gdb/compile/compile-object-load.c contains something like ld.so - it loads
object (.o) files and processes their relocations (rather bfd/ does that) to
copy them to mmap()ed areas.  I was debugging it only on i686+x86_64.


Jan