From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:42337) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1bLH2D-00042x-44 for qemu-devel@nongnu.org; Thu, 07 Jul 2016 17:40:38 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1bLH28-0006tF-Sg for qemu-devel@nongnu.org; Thu, 07 Jul 2016 17:40:36 -0400 Received: from mail-wm0-x22c.google.com ([2a00:1450:400c:c09::22c]:35139) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1bLH27-0006sp-Fa for qemu-devel@nongnu.org; Thu, 07 Jul 2016 17:40:32 -0400 Received: by mail-wm0-x22c.google.com with SMTP id z126so163839362wme.0 for ; Thu, 07 Jul 2016 14:40:31 -0700 (PDT) References: <1467735496-16256-7-git-send-email-alex.bennee@linaro.org> <1467909880-18834-1-git-send-email-sergey.fedorov@linaro.org> <1467909880-18834-4-git-send-email-sergey.fedorov@linaro.org> <877fcxmeai.fsf@linaro.org> <577EB19A.8040206@gmail.com> <577EBD45.7090003@gmail.com> From: Alex =?utf-8?Q?Benn=C3=A9e?= In-reply-to: <577EBD45.7090003@gmail.com> Date: Thu, 07 Jul 2016 22:40:34 +0100 Message-ID: <874m81m8j1.fsf@linaro.org> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 8bit Subject: Re: [Qemu-devel] [PATCH 3/3] tcg: Avoid bouncing tb_lock between tb_gen_code() and tb_add_jump() List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Sergey Fedorov Cc: Sergey Fedorov , qemu-devel@nongnu.org, patches@linaro.org, mttcg@listserver.greensocs.com, fred.konrad@greensocs.com, a.rigo@virtualopensystems.com, cota@braap.org, bobby.prani@gmail.com, rth@twiddle.net, mark.burton@greensocs.com, pbonzini@redhat.com, jan.kiszka@siemens.com, peter.maydell@linaro.org, claudio.fontana@huawei.com, Peter Crosthwaite Sergey Fedorov writes: > On 07/07/16 22:46, Sergey Fedorov wrote: >> On 07/07/16 22:36, Alex Bennée wrote: >>> Sergey Fedorov writes: >>> >>>> From: Sergey Fedorov >>>> >>>> Signed-off-by: Sergey Fedorov >>>> Signed-off-by: Sergey Fedorov >>>> --- >>>> cpu-exec.c | 15 +++++++++------ >>>> 1 file changed, 9 insertions(+), 6 deletions(-) >>>> >>>> diff --git a/cpu-exec.c b/cpu-exec.c >>>> index dd0bd5007701..54c935039592 100644 >>>> --- a/cpu-exec.c >>>> +++ b/cpu-exec.c >>>> @@ -295,7 +295,8 @@ static TranslationBlock *tb_find_slow(CPUState *cpu, >>>> >>>> /* mmap_lock is needed by tb_gen_code, and mmap_lock must be >>>> * taken outside tb_lock. As system emulation is currently >>>> - * single threaded the locks are NOPs. >>>> + * single threaded the locks are NOPs. Both locks are to be >>>> + * released at the end of tb_find_fast(). >>>> */ >>>> mmap_lock(); >>>> tb_lock(); >>>> @@ -308,9 +309,6 @@ static TranslationBlock *tb_find_slow(CPUState *cpu, >>>> /* if no translated code available, then translate it now */ >>>> tb = tb_gen_code(cpu, pc, cs_base, flags, 0); >>>> } >>>> - >>>> - tb_unlock(); >>>> - mmap_unlock(); >>> Hmm pushing these outside of tb_find_slow() makes me uncomfortable. I >>> guess tb_find_fast/slow are intimately tied together but the idea of >>> taking locks which are the responsibility of the calling function to >>> clear seems ugly to me. >> Okay, what if we also: >> >> diff --git a/cpu-exec.c b/cpu-exec.c >> index 54c935039592..ff8f92bc1dc1 100644 >> --- a/cpu-exec.c >> +++ b/cpu-exec.c >> @@ -278,45 +278,7 @@ static TranslationBlock *tb_find_physical(CPUState >> *cpu, >> return qht_lookup(&tcg_ctx.tb_ctx.htable, tb_cmp, &desc, h); >> } >> >> -static TranslationBlock *tb_find_slow(CPUState *cpu, >> - target_ulong pc, >> - target_ulong cs_base, >> - uint32_t flags) >> -{ >> - TranslationBlock *tb; >> - >> - /* Ensure that we won't find a TB in the shared hash table >> - * if it is being invalidated by some other thread. >> - * Otherwise we'd put it back to CPU's local cache. >> - * Pairs with smp_wmb() in tb_phys_invalidate(). */ >> - smp_rmb(); >> - tb = tb_find_physical(cpu, pc, cs_base, flags); >> - if (!tb) { >> - >> - /* mmap_lock is needed by tb_gen_code, and mmap_lock must be >> - * taken outside tb_lock. As system emulation is currently >> - * single threaded the locks are NOPs. Both locks are to be >> - * released at the end of tb_find_fast(). >> - */ >> - mmap_lock(); >> - tb_lock(); >> - >> - /* There's a chance that our desired tb has been translated while >> - * taking the locks so we check again inside the lock. >> - */ >> - tb = tb_find_physical(cpu, pc, cs_base, flags); >> - if (!tb) { >> - /* if no translated code available, then translate it now */ >> - tb = tb_gen_code(cpu, pc, cs_base, flags, 0); >> - } >> - } >> - >> - /* We add the TB in the virtual pc hash table for the fast lookup */ >> - atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)], tb); >> - return tb; >> -} >> - >> -static inline TranslationBlock *tb_find_fast(CPUState *cpu, >> +static inline TranslationBlock *tb_find(CPUState *cpu, >> TranslationBlock **last_tb, >> int tb_exit) >> { >> @@ -332,7 +294,34 @@ static inline TranslationBlock >> *tb_find_fast(CPUState *cpu, >> tb = atomic_read(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)]); >> if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base || >> tb->flags != flags)) { >> - tb = tb_find_slow(cpu, pc, cs_base, flags); >> + /* Ensure that we won't find a TB in the shared hash table >> + * if it is being invalidated by some other thread. >> + * Otherwise we'd put it back to CPU's local cache. >> + * Pairs with smp_wmb() in tb_phys_invalidate(). */ >> + smp_rmb(); >> + tb = tb_find_physical(cpu, pc, cs_base, flags); >> + if (!tb) { >> + >> + /* mmap_lock is needed by tb_gen_code, and mmap_lock must be >> + * taken outside tb_lock. As system emulation is currently >> + * single threaded the locks are NOPs. Both locks are to be >> + * released at the end of the function. >> + */ >> + mmap_lock(); >> + tb_lock(); >> + >> + /* There's a chance that our desired tb has been translated >> while >> + * taking the locks so we check again inside the lock. >> + */ >> + tb = tb_find_physical(cpu, pc, cs_base, flags); >> + if (!tb) { >> + /* if no translated code available, then translate it >> now */ >> + tb = tb_gen_code(cpu, pc, cs_base, flags, 0); >> + } >> + } >> + >> + /* We add the TB in the virtual pc hash table for the fast >> lookup */ >> + atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)], tb); >> } >> if (cpu->tb_flushed) { >> /* Ensure that no TB jump will be modified as the >> @@ -440,7 +429,7 @@ static inline bool cpu_handle_exception(CPUState >> *cpu, int *ret) >> && cpu->icount_decr.u16.low + cpu->icount_extra == 0) { >> /* try to cause an exception pending in the log */ >> TranslationBlock *last_tb = NULL; /* Avoid chaining TBs */ >> - cpu_exec_nocache(cpu, 1, tb_find_fast(cpu, &last_tb, 0), true); >> + cpu_exec_nocache(cpu, 1, tb_find(cpu, &last_tb, 0), true); >> *ret = -1; >> return true; >> #endif >> @@ -624,7 +613,7 @@ int cpu_exec(CPUState *cpu) >> 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); >> + tb = tb_find(cpu, &last_tb, tb_exit); >> cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit, &sc); >> /* Try to align the host and virtual clocks >> if the guest is in advance */ >> >> > > Using goto statements, the lines from tb_find_slow() which take the > locks can even be on the same indentation level as the lines resetting > them. I could prepare a patch series based on your first 3 patches and > this idea. Sure, lets see what it looks like ;-) > > Regards, > Sergey > >> >> >>>> } >>>> >>>> /* We add the TB in the virtual pc hash table for the fast lookup */ >>>> @@ -354,10 +352,15 @@ 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_lock(); >>>> + if (!tb_lock_locked()) { >>>> + tb_lock(); >>>> + } >>>> tb_add_jump(*last_tb, tb_exit, tb); >>>> - tb_unlock(); >>>> } >>>> + >>>> + tb_lock_reset(); >>>> + mmap_lock_reset(); >>>> + >>>> return tb; >>>> } >>> -- >>> Alex Bennée -- Alex Bennée