This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
[PING][PATCH] Power: Correct little-endian e500v2 GPR frame offsets
- From: "Maciej W. Rozycki" <macro at codesourcery dot com>
- To: <gdb-patches at sourceware dot org>
- Cc: Andrew Cagney <cagney at gnu dot org>, Kevin Buettner <kevinb at redhat dot com>
- Date: Mon, 10 Mar 2014 18:44:59 +0000
- Subject: [PING][PATCH] Power: Correct little-endian e500v2 GPR frame offsets
- Authentication-results: sourceware.org; auth=none
- References: <alpine dot DEB dot 1 dot 10 dot 1402282355070 dot 18199 at tp dot orcam dot me dot uk>
Ping!
On Sat, 1 Mar 2014, Maciej W. Rozycki wrote:
> Hi,
>
> This change corrects GPR frame offset calculation for the e500v2
> processor. On this target, featuring the SPE APU, GPRs are 64-bit and are
> held in stack frames whole with the use of `evstdd' and `evldd'
> instructions. Their integer 32-bit part occupies the low-order word and
> therefore its offset varies between the two endiannesses possible. Code
> in rs6000_frame_cache however assumes the big endianness and
> unconditionally uses an offset of 4 from the address of the whole 64-bit
> quantity in memory. This causes a failure in little-endian testing:
>
> (gdb) PASS: gdb.base/return-nodebug.exp: double: return from function with no debug info without a cast
> return (double) -1
> Make selected stack frame return now? (y or n) y
> #0 0x100002dc in main () at .../gdb/testsuite/gdb.base/return-nodebug.c:30
> 30 t = func ();
> (gdb) PASS: gdb.base/return-nodebug.exp: double: return from function with
> no debug info with a cast
> advance marker
> marker () at .../gdb/testsuite/gdb.base/return-nodebug.c:23
> 23 }
> (gdb) PASS: gdb.base/return-nodebug.exp: double: advance to marker
> print /d t
> $1 = 0
> (gdb) FAIL: gdb.base/return-nodebug.exp: double: full width of the returned result
>
> The change below addresses the problem by assuming an offset of 0 for
> little-endian targets.
>
> Regression-tested for both endiannesses with the powerpc-eabi target
> (`-mcpu=8548 -mfloat-gprs=double -mspe=yes -mabi=spe' GCC options) and no
> changes in results other than fixing the little-endian failure described
> above.
>
> OK to apply?
>
> 2014-02-28 Maciej W. Rozycki <macro@codesourcery.com>
>
> gdb/
> * rs6000-tdep.c (rs6000_frame_cache): Correct little-endian
> GPR offset into SPE pseudo registers.
>
> Maciej
>
> gdb-rs6000-ev-gpr-le.diff
> Index: gdb-fsf-trunk-quilt/gdb/rs6000-tdep.c
> ===================================================================
> --- gdb-fsf-trunk-quilt.orig/gdb/rs6000-tdep.c 2014-02-28 23:14:50.298928402 +0000
> +++ gdb-fsf-trunk-quilt/gdb/rs6000-tdep.c 2014-02-28 23:15:04.298928235 +0000
> @@ -3257,12 +3257,13 @@ rs6000_frame_cache (struct frame_info *t
> {
> int i;
> CORE_ADDR ev_addr = cache->base + fdata.ev_offset;
> + CORE_ADDR off = byte_order == BFD_ENDIAN_BIG ? 4 : 0;
> for (i = fdata.saved_ev; i < ppc_num_gprs; i++)
> {
> cache->saved_regs[tdep->ppc_ev0_regnum + i].addr = ev_addr;
> - cache->saved_regs[tdep->ppc_gp0_regnum + i].addr = ev_addr + 4;
> + cache->saved_regs[tdep->ppc_gp0_regnum + i].addr = ev_addr + off;
> ev_addr += register_size (gdbarch, tdep->ppc_ev0_regnum);
> - }
> + }
> }
> }
>
>