[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH v3 04/10] tcg: Init TB's direct jumps before mak
From: |
Sergey Fedorov |
Subject: |
Re: [Qemu-devel] [PATCH v3 04/10] tcg: Init TB's direct jumps before making it visible |
Date: |
Tue, 19 Apr 2016 15:42:02 +0300 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.6.0 |
On 19/04/16 13:55, Alex Bennée wrote:
> Sergey Fedorov <address@hidden> writes:
>
>> From: Sergey Fedorov <address@hidden>
>>
>> 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 <address@hidden>
>> Signed-off-by: Sergey Fedorov <address@hidden>
>> ---
>>
>> 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?
Kind regards,
Sergey
>
>> /* check next page if needed */
>> virt_page2 = (pc + tb->size - 1) & TARGET_PAGE_MASK;
>> phys_page2 = -1;
>
[Qemu-devel] [PATCH v3 10/10] tcg: Moderate direct block chaining safety checks in user mode, Sergey Fedorov, 2016/04/10