Re: [Qemu-devel] [PATCH 0/2] target/sh4: add missing tcg_temp_free()

["Alex Bennée" <alex.bennee@linaro.org>]

Philippe Mathieu-Daudé <f4bug@amsat.org> writes:

> Hi,
>
> After reading Alex commenting on IRC "java --version failing on sh4" I remember
> this series staged for 2.12.
>
> This might help for:
>
>   root@6e10336e48ac:/etc/apt# java --version
>   qemu-sh4: /home/alex/lsrc/qemu/qemu.git/tcg/tcg.h:703: temp_idx: Assertion `n >= 0 && n < tcg_ctx->nb_temps' failed.
>   qemu: uncaught target signal 11 (Segmentation fault) - core dumped

Sadly it's decode_gusa which trips for me:

  #0  0x00007ffff6941428 in __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:54
  #1  0x00007ffff694302a in __GI_abort () at abort.c:89
  #2  0x00007ffff6939bd7 in __assert_fail_base (fmt=<optimised out>, assertion=assertion@entry=0x555555678fc8 "n >= 0 && n < tcg_ctx->nb_temps", file=file@entry=0x555555678fa0 "/home/alex/lsrc/qemu/qemu.git/tcg/tcg.h", line=line@entry=703, function=function@entry=0x55555567aa38 <__PRETTY_FUNCTION__.23740> "temp_idx") at assert.c:92
  #3  0x00007ffff6939c82 in __GI___assert_fail (assertion=0x555555678fc8 "n >= 0 && n < tcg_ctx->nb_temps", file=0x555555678fa0 "/home/alex/lsrc/qemu/qemu.git/tcg/tcg.h", line=703, function=0x55555567aa38 <__PRETTY_FUNCTION__.23740> "temp_idx") at assert.c:101
  #4  0x0000555555585e03 in temp_idx (ts=0x555555902f20 <tcg_init_ctx>) at /home/alex/lsrc/qemu/qemu.git/tcg/tcg.h:703
  #5  0x0000555555585e5b in tcgv_i32_temp (v=0x0) at /home/alex/lsrc/qemu/qemu.git/tcg/tcg.h:724
  #6  0x000055555558bc2f in tcg_temp_free_i32 (arg=0x0) at /home/alex/lsrc/qemu/qemu.git/tcg/tcg.c:1053
  #7  0x00005555555ff077 in decode_gusa (ctx=0x7ffff7f67c00, env=0x5555579adf60, pmax_insns=0x7ffff7f67bec) at /home/alex/lsrc/qemu/qemu.git/target/sh4/translate.c:2193
  #8  0x00005555555ff303 in gen_intermediate_code (cs=0x5555579a5cc0, tb=0x55555593c6c0 <static_code_gen_buffer+78976>) at /home/alex/lsrc/qemu/qemu.git/target/sh4/translate.c:2268
  #9  0x00005555555bc656 in tb_gen_code (cpu=0x5555579a5cc0, pc=2134168040, cs_base=2134168044, flags=528320, cflags=524288) at /home/alex/lsrc/qemu/qemu.git/accel/tcg/translate-all.c:1292
  #10 0x00005555555b9ff9 in tb_find (cpu=0x5555579a5cc0, last_tb=0x0, tb_exit=0, cf_mask=524288) at /home/alex/lsrc/qemu/qemu.git/accel/tcg/cpu-exec.c:402
  #11 0x00005555555ba77d in cpu_exec (cpu=0x5555579a5cc0) at /home/alex/lsrc/qemu/qemu.git/accel/tcg/cpu-exec.c:735
  #12 0x00005555555c0ed2 in cpu_loop (env=0x5555579adf60) at /home/alex/lsrc/qemu/qemu.git/linux-user/main.c:2684
  #13 0x00005555555d001c in clone_func (arg=0x7fffffffc990) at /home/alex/lsrc/qemu/qemu.git/linux-user/syscall.c:6264
  #14 0x00007ffff6cdd6ba in start_thread (arg=0x7ffff7f68700) at pthread_create.c:333
  #15 0x00007ffff6a133dd in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
  #7  0x00005555555ff077 in decode_gusa (ctx=0x7ffff7f67c00, env=0x5555579adf60, pmax_insns=0x7ffff7f67bec) at /home/alex/lsrc/qemu/qemu.git/target/sh4/translate.c:2193
  2193	        tcg_temp_free_i32(op_arg);

The line:

    /* If op_src is not a valid register, then op_arg was a constant.  */
    if (op_src < 0) {
        tcg_temp_free_i32(op_arg);
    }

Looks pretty sketchy to me, why not check is op_arg is allocated
directly? Constants still need to be freed over a block. I think TCG
still keeps them around if it re-uses them.

>
> Regards,
>
> Philippe.
>
> Philippe Mathieu-Daudé (2):
>   target/sh4: add missing tcg_temp_free() in gen_conditional_jump()
>   target/sh4: add missing tcg_temp_free() in _decode_opc()
>
>  target/sh4/translate.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)


--
Alex Bennée