Re: [Qemu-devel] [PATCH v3 04/11] tcg: Prepare safe access to tb_flushed out of tb_lock

[Alex Bennée]

Sergey Fedorov <address@hidden> writes:

> On 14/07/16 15:45, Alex Bennée wrote:
>> Sergey Fedorov <address@hidden> writes:
>>
>>> From: Sergey Fedorov <address@hidden>
>>>
>>> Ensure atomicity of CPU's 'tb_flushed' access for future translation
>>> block lookup out of 'tb_lock'.
>>>
>>> This field can only be touched from another thread by tb_flush() in user
>>> mode emulation. So the only access to be atomic is:
>>>  * a single write in tb_flush();
>>>  * reads/writes out of 'tb_lock'.
>> It might worth mentioning the barrier here.
>
> Do you mean atomic_set() vs. atomic_mb_set()?

Yes.

>
>>
>>> In future, before enabling MTTCG in system mode, tb_flush() must be safe
>>> and this field becomes unnecessary.
>>>
>>> Signed-off-by: Sergey Fedorov <address@hidden>
>>> Signed-off-by: Sergey Fedorov <address@hidden>
>>> ---
>>>  cpu-exec.c      | 16 +++++++---------
>>>  translate-all.c |  4 ++--
>>>  2 files changed, 9 insertions(+), 11 deletions(-)
>>>
>>> diff --git a/cpu-exec.c b/cpu-exec.c
>>> index d6178eab71d4..c973e3b85922 100644
>>> --- a/cpu-exec.c
>>> +++ b/cpu-exec.c
>>> @@ -338,13 +338,6 @@ static inline TranslationBlock *tb_find_fast(CPUState 
>>> *cpu,
>>>                   tb->flags != flags)) {
>>>          tb = tb_find_slow(cpu, pc, cs_base, flags);
>>>      }
>>> -    if (cpu->tb_flushed) {
>>> -        /* Ensure that no TB jump will be modified as the
>>> -         * translation buffer has been flushed.
>>> -         */
>>> -        *last_tb = NULL;
>>> -        cpu->tb_flushed = false;
>>> -    }
>>>  #ifndef CONFIG_USER_ONLY
>>>      /* We don't take care of direct jumps when address mapping changes in
>>>       * system emulation. So it's not safe to make a direct jump to a TB
>>> @@ -356,7 +349,12 @@ static inline TranslationBlock *tb_find_fast(CPUState 
>>> *cpu,
>>>  #endif
>>>      /* See if we can patch the calling TB. */
>>>      if (last_tb && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
>>> -        tb_add_jump(last_tb, tb_exit, tb);
>>> +        /* Check if translation buffer has been flushed */
>>> +        if (cpu->tb_flushed) {
>>> +            cpu->tb_flushed = false;
>>> +        } else {
>>> +            tb_add_jump(last_tb, tb_exit, tb);
>>> +        }
>>>      }
>>>      tb_unlock();
>>>      return tb;
>>> @@ -618,7 +616,7 @@ int cpu_exec(CPUState *cpu)
>>>              }
>>>
>>>              last_tb = NULL; /* forget the last executed TB after exception 
>>> */
>>> -            cpu->tb_flushed = false; /* reset before first TB lookup */
>>> +            atomic_mb_set(&cpu->tb_flushed, false); /* reset before first 
>>> TB lookup */
>>>              for(;;) {
>>>                  cpu_handle_interrupt(cpu, &last_tb);
>>>                  tb = tb_find_fast(cpu, last_tb, tb_exit);
>>> diff --git a/translate-all.c b/translate-all.c
>>> index fdf520a86d68..788fed1e0765 100644
>>> --- a/translate-all.c
>>> +++ b/translate-all.c
>>> @@ -845,7 +845,6 @@ void tb_flush(CPUState *cpu)
>>>          > tcg_ctx.code_gen_buffer_size) {
>>>          cpu_abort(cpu, "Internal error: code buffer overflow\n");
>>>      }
>>> -    tcg_ctx.tb_ctx.nb_tbs = 0;
>>>
>>>      CPU_FOREACH(cpu) {
>>>          int i;
>>> @@ -853,9 +852,10 @@ void tb_flush(CPUState *cpu)
>>>          for (i = 0; i < TB_JMP_CACHE_SIZE; ++i) {
>>>              atomic_set(&cpu->tb_jmp_cache[i], NULL);
>>>          }
>>> -        cpu->tb_flushed = true;
>>> +        atomic_mb_set(&cpu->tb_flushed, true);
>>>      }
>>>
>>> +    tcg_ctx.tb_ctx.nb_tbs = 0;
>>>      qht_reset_size(&tcg_ctx.tb_ctx.htable, CODE_GEN_HTABLE_SIZE);
>> I can see the sense of moving the setting of nb_tbs but is it strictly
>> required as part of this patch?
>
> Yes, otherwise tb_alloc() may start allocation TBs from the beginning of
> the translation buffer before 'tb_flushed' is updated.

Ahh yes I see. Thanks

Reviewed-by: Alex Bennée <address@hidden>


>
> Kind regards,
> Sergey
>
>>
>>>      page_flush_tb();
>> Otherwise:
>>
>> Reviewed-by: Alex Bennée <address@hidden>
>>
>> --
>> Alex Bennée


--
Alex Bennée