From: Blue Swirl <blauwirbel@gmail.com>
To: Aurelien Jarno <aurelien@aurel32.net>
Cc: qemu-devel <qemu-devel@nongnu.org>
Subject: Re: [Qemu-devel] [PATCH 06/11] TCG/x86: use stack for TCG temps
Date: Tue, 17 May 2011 23:08:03 +0300 [thread overview]
Message-ID: <BANLkTimhvEAurAz0i=RefinbsMzLxa1ziQ@mail.gmail.com> (raw)
In-Reply-To: <20110517184636.GW30615@hall.aurel32.net>
On Tue, May 17, 2011 at 9:46 PM, Aurelien Jarno <aurelien@aurel32.net> wrote:
> On Sat, May 14, 2011 at 10:38:40PM +0300, Blue Swirl wrote:
>> Use stack instead of temp_buf array in CPUState for TCG
>> temps.
>>
>> Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
>> ---
>> tcg/i386/tcg-target.c | 19 ++++++++++---------
>> 1 files changed, 10 insertions(+), 9 deletions(-)
>>
>> diff --git a/tcg/i386/tcg-target.c b/tcg/i386/tcg-target.c
>> index 01747f3..0e168ea 100644
>> --- a/tcg/i386/tcg-target.c
>> +++ b/tcg/i386/tcg-target.c
>> @@ -1918,19 +1918,22 @@ static void tcg_target_qemu_prologue(TCGContext *s)
>>
>> /* TB prologue */
>>
>> - /* Save all callee saved registers. */
>> - for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
>> - tcg_out_push(s, tcg_target_callee_save_regs[i]);
>> - }
>> -
>> - /* Reserve some stack space. */
>> + /* Reserve some stack space, also for TCG temps. */
>> push_size = 1 + ARRAY_SIZE(tcg_target_callee_save_regs);
>> push_size *= TCG_TARGET_REG_BITS / 8;
>>
>> - frame_size = push_size + TCG_STATIC_CALL_ARGS_SIZE;
>> + frame_size = push_size + TCG_STATIC_CALL_ARGS_SIZE +
>> + CPU_TEMP_BUF_NLONGS * sizeof(long);
>> frame_size = (frame_size + TCG_TARGET_STACK_ALIGN - 1) &
>> ~(TCG_TARGET_STACK_ALIGN - 1);
>> stack_addend = frame_size - push_size;
>> + tcg_set_frame(s, TCG_REG_ESP, 0, CPU_TEMP_BUF_NLONGS * sizeof(long));
>
> You should probably use TCG_REG_CALL_STACK instead of hardcoading the
> register.
OK.
>> +
>> + /* Save all callee saved registers. */
>> + for (i = 0; i < ARRAY_SIZE(tcg_target_callee_save_regs); i++) {
>> + tcg_out_push(s, tcg_target_callee_save_regs[i]);
>> + }
>> +
>> tcg_out_addi(s, TCG_REG_ESP, -stack_addend);
>>
>> /* jmp *tb. */
>> @@ -1979,6 +1982,4 @@ static void tcg_target_init(TCGContext *s)
>> tcg_regset_set_reg(s->reserved_regs, TCG_REG_ESP);
>>
>> tcg_add_target_add_op_defs(x86_op_defs);
>> - tcg_set_frame(s, TCG_AREG0, offsetof(CPUState, temp_buf),
>> - CPU_TEMP_BUF_NLONGS * sizeof(long));
>> }
>
> Note that this patch is likely to break calls to helpers which need
> parameters on the stack, by judging at the current code (I haven't
> tested it in practice):
>
> | if (allocate_args) {
> | tcg_out_addi(s, TCG_REG_CALL_STACK, -STACK_DIR(call_stack_size));
> | }
>
> The stack register (esp) is decreased.
>
> | stack_offset = TCG_TARGET_CALL_STACK_OFFSET;
> | for(i = nb_regs; i < nb_params; i++) {
> | arg = args[nb_oargs + i];
> | #ifdef TCG_TARGET_STACK_GROWSUP
> | stack_offset -= sizeof(tcg_target_long);
> | #endif
> | if (arg != TCG_CALL_DUMMY_ARG) {
> | ts = &s->temps[arg];
> | if (ts->val_type == TEMP_VAL_REG) {
> | tcg_out_st(s, ts->type, ts->reg, TCG_REG_CALL_STACK, stack_offset);
> | } else if (ts->val_type == TEMP_VAL_MEM) {
> | reg = tcg_reg_alloc(s, tcg_target_available_regs[ts->type],
> | s->reserved_regs);
>
> tcg_reg_alloc() may spill some register, and save them in the allocated
> frame. If it is referenced by the stack pointer, given it has been
> changed, it won't be save at the write place.
>
>
> | /* XXX: not correct if reading values from the stack */
> | tcg_out_ld(s, ts->type, reg, ts->mem_reg, ts->mem_offset);
>
> As the comment said, if ts->mem_reg is the stack register (like with
> this patch), given it has been increased above, the wrong value will be
> read.
>
> | tcg_out_st(s, ts->type, reg, TCG_REG_CALL_STACK, stack_offset);
> | } else if (ts->val_type == TEMP_VAL_CONST) {
> | reg = tcg_reg_alloc(s, tcg_target_available_regs[ts->type],
> | s->reserved_regs);
> | /* XXX: sign extend may be needed on some targets */
> | tcg_out_movi(s, ts->type, reg, ts->val);
> | tcg_out_st(s, ts->type, reg, TCG_REG_CALL_STACK, stack_offset);
> | } else {
> | tcg_abort();
> | }
Maybe the frame pointer register could be set up and used instead, but
that would cost one extra register.
Alternatively it may be possible to avoid changing stack pointer. We
know in advance all possible helpers and the number of their
parameters, so the stack could be preallocated as suggested by a
comment.
next prev parent reply other threads:[~2011-05-17 20:08 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-05-14 19:38 [Qemu-devel] [PATCH 06/11] TCG/x86: use stack for TCG temps Blue Swirl
2011-05-17 14:57 ` Richard Henderson
2011-05-17 18:46 ` Aurelien Jarno
2011-05-17 20:08 ` Blue Swirl [this message]
2011-05-18 16:09 ` Richard Henderson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='BANLkTimhvEAurAz0i=RefinbsMzLxa1ziQ@mail.gmail.com' \
--to=blauwirbel@gmail.com \
--cc=aurelien@aurel32.net \
--cc=qemu-devel@nongnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).