qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Sebastian Macke <sebastian@macke.de>
To: Max Filippov <jcmvbkbc@gmail.com>
Cc: openrisc@lists.openrisc.net, openrisc@lists.opencores.org,
	qemu-devel <qemu-devel@nongnu.org>, Ethan Hunt <proljc@gmail.com>
Subject: Re: [Qemu-devel] [PATCH 03/13] target-openrisc: Separate of load/store instructions
Date: Tue, 29 Oct 2013 14:36:50 -0700	[thread overview]
Message-ID: <52702A72.8040705@macke.de> (raw)
In-Reply-To: <CAMo8Bf+reRZiEY_QX-033FU=CpTLi4vMJzigAmCu8h4s_Ab1pg@mail.gmail.com>

On 29/10/2013 1:05 PM, Max Filippov wrote:
> On Tue, Oct 29, 2013 at 11:04 PM, Sebastian Macke <sebastian@macke.de> wrote:
>> This patch separates the load and store instruction to a
>> separate function.
>> The repetition of the source code can be reduced and further
>> optimizations can be implemented.
>> In this case it checks for a zero offset and optimizes it.
>>
>> Additional this patch solves a severe bug for the softmmu emulation.
>> The pc has to be saved as these instructions can fail and lead
>> to a tlb miss exception.
> In case of an exception we re-translate the TB to find the PC where
> the exception happened, see cpu_restore_state call from the tlb_fill
> function. Also this applies to both user and system emulation, but
> you only handle the system emulation case.


The problem is the epcr register in the interrupt routine in which the 
current pc must be saved.
Of course in the user emulation case the interrupt handler is never 
executed.

When is the pc of the fault determined? Before or after the interrupt 
handler?
Finding this problem gave me a long headache. But it would be nice if 
there is a better solution.


>
>> Signed-off-by: Sebastian Macke <sebastian@macke.de>
>> ---
>>   target-openrisc/translate.c | 130 ++++++++++++++++++++++++++------------------
>>   1 file changed, 76 insertions(+), 54 deletions(-)
>>
>> diff --git a/target-openrisc/translate.c b/target-openrisc/translate.c
>> index 31f8717..1bb686c 100644
>> --- a/target-openrisc/translate.c
>> +++ b/target-openrisc/translate.c
>> @@ -692,6 +692,73 @@ static void dec_calc(DisasContext *dc, uint32_t insn)
>>       }
>>   }
>>
>> +static void gen_loadstore(DisasContext *dc, uint32 op0,
>> +                          uint32_t ra, uint32_t rb, uint32_t rd,
>> +                          uint32_t offset)
>> +{
>> +
>> +/* The load and store instructions can fail and lead to a
>> + *  tlb miss exception. The correct pc has to be stored for
>> + *  this case.
>> + */
>> +#if !defined(CONFIG_USER_ONLY)
>> +    tcg_gen_movi_tl(cpu_pc, dc->pc);
>> +#endif
>> +
>> +    TCGv t0 = cpu_R[ra];
>> +    if (offset != 0) {
>> +        t0 = tcg_temp_new();
>> +        tcg_gen_addi_tl(t0, cpu_R[ra], sign_extend(offset, 16));
>> +    }
>> +
>> +    switch (op0) {
>> +    case 0x21:    /* l.lwz */
>> +        tcg_gen_qemu_ld32u(cpu_R[rd], t0, dc->mem_idx);
>> +        break;
>> +
>> +    case 0x22:    /* l.lws */
>> +        tcg_gen_qemu_ld32s(cpu_R[rd], t0, dc->mem_idx);
>> +        break;
>> +
>> +    case 0x23:    /* l.lbz */
>> +        tcg_gen_qemu_ld8u(cpu_R[rd], t0, dc->mem_idx);
>> +        break;
>> +
>> +    case 0x24:    /* l.lbs */
>> +        tcg_gen_qemu_ld8s(cpu_R[rd], t0, dc->mem_idx);
>> +        break;
>> +
>> +    case 0x25:    /* l.lhz */
>> +        tcg_gen_qemu_ld16u(cpu_R[rd], t0, dc->mem_idx);
>> +        break;
>> +
>> +    case 0x26:    /* l.lhs */
>> +        tcg_gen_qemu_ld16s(cpu_R[rd], t0, dc->mem_idx);
>> +        break;
>> +
>> +    case 0x35:    /* l.sw */
>> +        tcg_gen_qemu_st32(cpu_R[rb], t0, dc->mem_idx);
>> +        break;
>> +
>> +    case 0x36:    /* l.sb */
>> +        tcg_gen_qemu_st8(cpu_R[rb], t0, dc->mem_idx);
>> +        break;
>> +
>> +    case 0x37:    /* l.sh */
>> +        tcg_gen_qemu_st16(cpu_R[rb], t0, dc->mem_idx);
>> +        break;
>> +
>> +    default:
>> +    break;
> Broken indentation.
>
>> +    }
>> +
>> +    if (offset != 0) {
>> +        tcg_temp_free(t0);
>> +    }
>> +
>> +}
>> +
>> +
>>   static void dec_misc(DisasContext *dc, uint32_t insn)
>>   {
>>       uint32_t op0, op1;
>> @@ -843,62 +910,32 @@ static void dec_misc(DisasContext *dc, uint32_t insn)
>>
>>       case 0x21:    /* l.lwz */
>>           LOG_DIS("l.lwz r%d, r%d, %d\n", rd, ra, I16);
>> -        {
>> -            TCGv t0 = tcg_temp_new();
>> -            tcg_gen_addi_tl(t0, cpu_R[ra], sign_extend(I16, 16));
>> -            tcg_gen_qemu_ld32u(cpu_R[rd], t0, dc->mem_idx);
>> -            tcg_temp_free(t0);
>> -        }
>> +        gen_loadstore(dc, op0, ra, rb, rd, I16);
>>           break;
>>
>>       case 0x22:    /* l.lws */
>>           LOG_DIS("l.lws r%d, r%d, %d\n", rd, ra, I16);
>> -        {
>> -            TCGv t0 = tcg_temp_new();
>> -            tcg_gen_addi_tl(t0, cpu_R[ra], sign_extend(I16, 16));
>> -            tcg_gen_qemu_ld32s(cpu_R[rd], t0, dc->mem_idx);
>> -            tcg_temp_free(t0);
>> -        }
>> +        gen_loadstore(dc, op0, ra, rb, rd, I16);
>>           break;
>>
>>       case 0x23:    /* l.lbz */
>>           LOG_DIS("l.lbz r%d, r%d, %d\n", rd, ra, I16);
>> -        {
>> -            TCGv t0 = tcg_temp_new();
>> -            tcg_gen_addi_tl(t0, cpu_R[ra], sign_extend(I16, 16));
>> -            tcg_gen_qemu_ld8u(cpu_R[rd], t0, dc->mem_idx);
>> -            tcg_temp_free(t0);
>> -        }
>> +        gen_loadstore(dc, op0, ra, rb, rd, I16);
>>           break;
>>
>>       case 0x24:    /* l.lbs */
>>           LOG_DIS("l.lbs r%d, r%d, %d\n", rd, ra, I16);
>> -        {
>> -            TCGv t0 = tcg_temp_new();
>> -            tcg_gen_addi_tl(t0, cpu_R[ra], sign_extend(I16, 16));
>> -            tcg_gen_qemu_ld8s(cpu_R[rd], t0, dc->mem_idx);
>> -            tcg_temp_free(t0);
>> -        }
>> +        gen_loadstore(dc, op0, ra, rb, rd, I16);
>>           break;
>>
>>       case 0x25:    /* l.lhz */
>>           LOG_DIS("l.lhz r%d, r%d, %d\n", rd, ra, I16);
>> -        {
>> -            TCGv t0 = tcg_temp_new();
>> -            tcg_gen_addi_tl(t0, cpu_R[ra], sign_extend(I16, 16));
>> -            tcg_gen_qemu_ld16u(cpu_R[rd], t0, dc->mem_idx);
>> -            tcg_temp_free(t0);
>> -        }
>> +        gen_loadstore(dc, op0, ra, rb, rd, I16);
>>           break;
>>
>>       case 0x26:    /* l.lhs */
>>           LOG_DIS("l.lhs r%d, r%d, %d\n", rd, ra, I16);
>> -        {
>> -            TCGv t0 = tcg_temp_new();
>> -            tcg_gen_addi_tl(t0, cpu_R[ra], sign_extend(I16, 16));
>> -            tcg_gen_qemu_ld16s(cpu_R[rd], t0, dc->mem_idx);
>> -            tcg_temp_free(t0);
>> -        }
>> +        gen_loadstore(dc, op0, ra, rb, rd, I16);
>>           break;
>>
>>       case 0x27:    /* l.addi */
>> @@ -1047,32 +1084,17 @@ static void dec_misc(DisasContext *dc, uint32_t insn)
>>
>>       case 0x35:    /* l.sw */
>>           LOG_DIS("l.sw %d, r%d, r%d, %d\n", I5, ra, rb, I11);
>> -        {
>> -            TCGv t0 = tcg_temp_new();
>> -            tcg_gen_addi_tl(t0, cpu_R[ra], sign_extend(tmp, 16));
>> -            tcg_gen_qemu_st32(cpu_R[rb], t0, dc->mem_idx);
>> -            tcg_temp_free(t0);
>> -        }
>> +        gen_loadstore(dc, op0, ra, rb, rd, tmp);
>>           break;
>>
>>       case 0x36:    /* l.sb */
>>           LOG_DIS("l.sb %d, r%d, r%d, %d\n", I5, ra, rb, I11);
>> -        {
>> -            TCGv t0 = tcg_temp_new();
>> -            tcg_gen_addi_tl(t0, cpu_R[ra], sign_extend(tmp, 16));
>> -            tcg_gen_qemu_st8(cpu_R[rb], t0, dc->mem_idx);
>> -            tcg_temp_free(t0);
>> -        }
>> +        gen_loadstore(dc, op0, ra, rb, rd, tmp);
>>           break;
>>
>>       case 0x37:    /* l.sh */
>> +        gen_loadstore(dc, op0, ra, rb, rd, tmp);
>>           LOG_DIS("l.sh %d, r%d, r%d, %d\n", I5, ra, rb, I11);
> In other cases you do it in the reverse order.
> Looks like all these cases can be further consolidated into
> a pair of LOG_DIS(); gen_loadstore(); with a small table for
> LOG_DIS format string each.

You are right. This is not optimal. But before it was worse. I will try 
to I optimize it in the V2 patchset.


>
>> -        {
>> -            TCGv t0 = tcg_temp_new();
>> -            tcg_gen_addi_tl(t0, cpu_R[ra], sign_extend(tmp, 16));
>> -            tcg_gen_qemu_st16(cpu_R[rb], t0, dc->mem_idx);
>> -            tcg_temp_free(t0);
>> -        }
>>           break;
>>
>>       default:

  reply	other threads:[~2013-10-29 21:37 UTC|newest]

Thread overview: 48+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2013-10-29 19:04 [Qemu-devel] [PATCH 00/13] target-openrisc: More optimizations and corrections Sebastian Macke
2013-10-29 19:04 ` [Qemu-devel] [PATCH 01/13] target-openrisc: Implement translation block chaining Sebastian Macke
2013-10-29 19:04 ` [Qemu-devel] [PATCH 02/13] target-openrisc: Separate Delayed slot handling from main loop Sebastian Macke
2013-10-29 19:04 ` [Qemu-devel] [PATCH 03/13] target-openrisc: Separate of load/store instructions Sebastian Macke
2013-10-29 20:05   ` Max Filippov
2013-10-29 21:36     ` Sebastian Macke [this message]
2013-10-29 21:49       ` Richard Henderson
2013-10-29 22:55       ` Max Filippov
2013-10-29 23:37         ` Sebastian Macke
2013-10-29 19:04 ` [Qemu-devel] [PATCH 04/13] target-openrisc: sync flags only when necessary Sebastian Macke
2013-10-29 21:51   ` Richard Henderson
2013-10-29 19:04 ` [Qemu-devel] [PATCH 05/13] target-openrisc: Remove TLB flush on exception Sebastian Macke
2013-10-29 19:47   ` Peter Maydell
2013-10-29 22:41     ` Sebastian Macke
2013-11-01 18:58       ` Peter Maydell
2013-11-02  1:21         ` Richard Henderson
2013-11-06 22:59           ` [Qemu-devel] [Openrisc] " Edgar E. Iglesias
2013-11-02  1:29       ` [Qemu-devel] " Richard Henderson
2013-10-29 19:04 ` [Qemu-devel] [PATCH 06/13] target-openrisc: Remove TLB flush from l.rfe instruction Sebastian Macke
2013-10-29 21:01   ` Max Filippov
2013-10-29 21:53     ` Sebastian Macke
2013-10-29 22:20       ` Max Filippov
2013-10-29 23:14         ` Sebastian Macke
2013-10-29 19:04 ` [Qemu-devel] [PATCH 07/13] target-openrisc: Correct l.cmov conditional check Sebastian Macke
2013-10-29 21:15   ` Max Filippov
2013-10-29 21:23     ` Sebastian Macke
2013-10-29 19:04 ` [Qemu-devel] [PATCH 08/13] target-openrisc: Test for Overflow exception statically Sebastian Macke
2013-10-29 21:25   ` Max Filippov
2013-10-29 22:06     ` Sebastian Macke
2013-10-29 19:04 ` [Qemu-devel] [PATCH 09/13] target-openrisc: Add CPU which neglects Carry and Overflow Flag Sebastian Macke
2013-10-30 18:14   ` Richard Henderson
2013-10-30 19:22     ` Sebastian Macke
2013-10-30 19:31       ` Richard Henderson
2013-10-29 19:04 ` [Qemu-devel] [PATCH 10/13] target-openrisc: Correct target number for 64 bit llseek Sebastian Macke
2013-10-29 19:04 ` [Qemu-devel] [PATCH 11/13] target-openrisc: use jmp_pc as flag variable for branches Sebastian Macke
2013-10-30 18:33   ` Richard Henderson
2013-10-30 19:07     ` Sebastian Macke
2013-10-30 19:32       ` Richard Henderson
2013-10-30 19:47       ` Richard Henderson
2013-10-30 21:08         ` Sebastian Macke
2013-10-30 22:02           ` Richard Henderson
2013-10-31  0:29             ` Sebastian Macke
2013-10-29 19:04 ` [Qemu-devel] [PATCH 12/13] target-openrisc: Add correct gdb information for the pc value Sebastian Macke
2013-10-29 19:04 ` [Qemu-devel] [PATCH 13/13] target-openrisc: Add In-circuit emulator support Sebastian Macke
2013-10-29 19:53 ` [Qemu-devel] [PATCH 00/13] target-openrisc: More optimizations and corrections Peter Maydell
2013-10-29 21:15 ` Max Filippov
2013-10-29 21:22   ` Sebastian Macke
2013-10-31 11:47     ` Jia Liu

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=52702A72.8040705@macke.de \
    --to=sebastian@macke.de \
    --cc=jcmvbkbc@gmail.com \
    --cc=openrisc@lists.opencores.org \
    --cc=openrisc@lists.openrisc.net \
    --cc=proljc@gmail.com \
    --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).