Re: [Qemu-devel] [PATCH v3 04/10] tcg: Init TB's direct jumps before making it visible

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

Sergey Fedorov <serge.fdrv@gmail.com> writes:

> On 19/04/16 13:55, Alex Bennée wrote:
>> Sergey Fedorov <sergey.fedorov@linaro.org> writes:
>>
>>> From: Sergey Fedorov <serge.fdrv@gmail.com>
>>>
>>> Initialize TB's direct jump list data fields and reset the jumps before
>>> tb_link_page() puts it into the physical hash table and the physical
>>> page list. So TB is completely initialized before it becomes visible.
>>>
>>> Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com>
>>> Signed-off-by: Sergey Fedorov <sergey.fedorov@linaro.org>
>>> ---
>>>
>>> Changes in v2:
>>>  * Tweaked a comment
>>>
>>>  translate-all.c | 27 ++++++++++++++-------------
>>>  1 file changed, 14 insertions(+), 13 deletions(-)
>>>
>>> diff --git a/translate-all.c b/translate-all.c
>>> index 7ac7916f2792..dfa7f0d64e76 100644
>>> --- a/translate-all.c
>>> +++ b/translate-all.c
>>> @@ -1133,19 +1133,6 @@ static void tb_link_page(TranslationBlock *tb, tb_page_addr_t phys_pc,
>>>          tb->page_addr[1] = -1;
>>>      }
>>>
>>> -    assert(((uintptr_t)tb & 3) == 0);
>>> -    tb->jmp_list_first = (uintptr_t)tb | 2;
>>> -    tb->jmp_list_next[0] = (uintptr_t)NULL;
>>> -    tb->jmp_list_next[1] = (uintptr_t)NULL;
>>> -
>>> -    /* init original jump addresses */
>>> -    if (tb->jmp_reset_offset[0] != TB_JMP_RESET_OFFSET_INVALID) {
>>> -        tb_reset_jump(tb, 0);
>>> -    }
>>> -    if (tb->jmp_reset_offset[1] != TB_JMP_RESET_OFFSET_INVALID) {
>>> -        tb_reset_jump(tb, 1);
>>> -    }
>>> -
>>>  #ifdef DEBUG_TB_CHECK
>>>      tb_page_check();
>>>  #endif
>>> @@ -1254,6 +1241,20 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
>>>          ROUND_UP((uintptr_t)gen_code_buf + gen_code_size + search_size,
>>>                   CODE_GEN_ALIGN);
>>>
>>> +    /* init jump list */
>>> +    assert(((uintptr_t)tb & 3) == 0);
>>> +    tb->jmp_list_first = (uintptr_t)tb | 2;
>>> +    tb->jmp_list_next[0] = (uintptr_t)NULL;
>>> +    tb->jmp_list_next[1] = (uintptr_t)NULL;
>>> +
>>> +    /* init original jump addresses wich has been set during tcg_gen_code() */
>>> +    if (tb->jmp_reset_offset[0] != TB_JMP_RESET_OFFSET_INVALID) {
>>> +        tb_reset_jump(tb, 0);
>>> +    }
>>> +    if (tb->jmp_reset_offset[1] != TB_JMP_RESET_OFFSET_INVALID) {
>>> +        tb_reset_jump(tb, 1);
>>> +    }
>>> +
>> If we are really concerned about ensuring everything is set before we
>> insert the TB into the list should we not have an explicit write barrier
>> before we call to link the page?
>
> Currently, it is synchronized by 'tb_lock', so no need in a memory
> barrier here. So this is a simple rearrangement of code to a more
> suitable place and maybe just a preparation for relaxing locking scheme
> in future. It would be ahead of time and unnecessary overhead to put a
> barrier in this patch. Do you think it's worth to mention that in the
> commit message?

Good point. Maybe it would be better to clean-up the comment in
tb_link_page() with the assumption that memory consistency for linking
the new TB is either explicit (user mode, tb_lock) or implicit in single
threaded softmmu emulation.

We can then update the comment when we add the MTTCG patches.

>
> Kind regards,
> Sergey
>
>>
>>>      /* check next page if needed */
>>>      virt_page2 = (pc + tb->size - 1) & TARGET_PAGE_MASK;
>>>      phys_page2 = -1;
>>


--
Alex Bennée