From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:48399) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1ZCRKh-0004Sj-ND for qemu-devel@nongnu.org; Tue, 07 Jul 2015 07:46:41 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1ZCRKd-0003Q4-UY for qemu-devel@nongnu.org; Tue, 07 Jul 2015 07:46:39 -0400 Received: from greensocs.com ([193.104.36.180]:32781) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1ZCRKd-0003Pr-KQ for qemu-devel@nongnu.org; Tue, 07 Jul 2015 07:46:35 -0400 Message-ID: <559BBC18.50002@greensocs.com> Date: Tue, 07 Jul 2015 13:46:32 +0200 From: Frederic Konrad MIME-Version: 1.0 References: <1435330053-18733-1-git-send-email-fred.konrad@greensocs.com> <1435330053-18733-3-git-send-email-fred.konrad@greensocs.com> <87r3ok9sur.fsf@linaro.org> In-Reply-To: <87r3ok9sur.fsf@linaro.org> Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [RFC PATCH V6 02/18] replace spinlock by QemuMutex. List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: =?UTF-8?B?QWxleCBCZW5uw6ll?= Cc: mttcg@listserver.greensocs.com, peter.maydell@linaro.org, a.spyridakis@virtualopensystems.com, mark.burton@greensocs.com, agraf@suse.de, qemu-devel@nongnu.org, guillaume.delbergue@greensocs.com, pbonzini@redhat.com, alistair.francis@xilinx.com On 07/07/2015 12:15, Alex Benn=C3=A9e wrote: > fred.konrad@greensocs.com writes: > >> From: KONRAD Frederic >> >> spinlock is only used in two cases: >> * cpu-exec.c: to protect TranslationBlock >> * mem_helper.c: for lock helper in target-i386 (which seems broken)= . >> >> It's a pthread_mutex_t in user-mode so better using QemuMutex directly= in this >> case. >> It allows as well to reuse tb_lock mutex of TBContext in case of multi= thread >> TCG. >> >> Signed-off-by: KONRAD Frederic >> --- >> cpu-exec.c | 15 +++++++++++---- >> include/exec/exec-all.h | 4 ++-- >> linux-user/main.c | 6 +++--- >> target-i386/mem_helper.c | 16 +++++++++++++--- >> tcg/i386/tcg-target.c | 8 ++++++++ >> 5 files changed, 37 insertions(+), 12 deletions(-) >> >> diff --git a/cpu-exec.c b/cpu-exec.c >> index 2ffeb6e..d6336d9 100644 >> --- a/cpu-exec.c >> +++ b/cpu-exec.c >> @@ -362,7 +362,9 @@ int cpu_exec(CPUArchState *env) >> SyncClocks sc; >> =20 >> /* This must be volatile so it is not trashed by longjmp() */ >> +#if defined(CONFIG_USER_ONLY) >> volatile bool have_tb_lock =3D false; >> +#endif >> =20 >> if (cpu->halted) { >> if (!cpu_has_work(cpu)) { >> @@ -480,8 +482,10 @@ int cpu_exec(CPUArchState *env) >> cpu->exception_index =3D EXCP_INTERRUPT; >> cpu_loop_exit(cpu); >> } >> - spin_lock(&tcg_ctx.tb_ctx.tb_lock); >> +#if defined(CONFIG_USER_ONLY) >> + qemu_mutex_lock(&tcg_ctx.tb_ctx.tb_lock); >> have_tb_lock =3D true; >> +#endif > Why are the locking rules different for CONFIG_USER versus system > emulation? Looking at the final tree: > >> tb =3D tb_find_fast(env); > this eventually ends up doing a tb_lock on the find_slow path which IIR= C > is when might end up doing the actual code generation. I didn't looked at the user code. But yes we should probably end with=20 the same thing for both user mode code and system mode code. That's what Peter was suggesting before but I didn't have time to look at this right now. > >> /* Note: we do it here to avoid a gcc bug on Mac OS = X when >> doing it in tb_find_slow */ >> @@ -503,9 +507,10 @@ int cpu_exec(CPUArchState *env) >> tb_add_jump((TranslationBlock *)(next_tb & ~TB_E= XIT_MASK), >> next_tb & TB_EXIT_MASK, tb); >> } >> +#if defined(CONFIG_USER_ONLY) >> have_tb_lock =3D false; >> - spin_unlock(&tcg_ctx.tb_ctx.tb_lock); >> - >> + qemu_mutex_unlock(&tcg_ctx.tb_ctx.tb_lock); >> +#endif >> /* cpu_interrupt might be called while translating t= he >> TB, but before it is linked into a potentially >> infinite loop and becomes env->current_tb. Avoid >> @@ -572,10 +577,12 @@ int cpu_exec(CPUArchState *env) >> #ifdef TARGET_I386 >> x86_cpu =3D X86_CPU(cpu); >> #endif >> +#if defined(CONFIG_USER_ONLY) >> if (have_tb_lock) { >> - spin_unlock(&tcg_ctx.tb_ctx.tb_lock); >> + qemu_mutex_unlock(&tcg_ctx.tb_ctx.tb_lock); >> have_tb_lock =3D false; >> } >> +#endif >> } >> } /* for(;;) */ >> =20 >> diff --git a/include/exec/exec-all.h b/include/exec/exec-all.h >> index 2573e8c..44f3336 100644 >> --- a/include/exec/exec-all.h >> +++ b/include/exec/exec-all.h >> @@ -176,7 +176,7 @@ struct TranslationBlock { >> struct TranslationBlock *jmp_first; >> }; >> =20 >> -#include "exec/spinlock.h" >> +#include "qemu/thread.h" >> =20 >> typedef struct TBContext TBContext; >> =20 >> @@ -186,7 +186,7 @@ struct TBContext { >> TranslationBlock *tb_phys_hash[CODE_GEN_PHYS_HASH_SIZE]; >> int nb_tbs; >> /* any access to the tbs or the page table must use this lock */ >> - spinlock_t tb_lock; >> + QemuMutex tb_lock; >> =20 >> /* statistics */ >> int tb_flush_count; >> diff --git a/linux-user/main.c b/linux-user/main.c >> index c855bcc..bce3a98 100644 >> --- a/linux-user/main.c >> +++ b/linux-user/main.c >> @@ -107,7 +107,7 @@ static int pending_cpus; >> /* Make sure everything is in a consistent state for calling fork().= */ >> void fork_start(void) >> { >> - pthread_mutex_lock(&tcg_ctx.tb_ctx.tb_lock); >> + qemu_mutex_lock(&tcg_ctx.tb_ctx.tb_lock); >> pthread_mutex_lock(&exclusive_lock); >> mmap_fork_start(); >> } >> @@ -129,11 +129,11 @@ void fork_end(int child) >> pthread_mutex_init(&cpu_list_mutex, NULL); >> pthread_cond_init(&exclusive_cond, NULL); >> pthread_cond_init(&exclusive_resume, NULL); >> - pthread_mutex_init(&tcg_ctx.tb_ctx.tb_lock, NULL); >> + qemu_mutex_init(&tcg_ctx.tb_ctx.tb_lock); >> gdbserver_fork((CPUArchState *)thread_cpu->env_ptr); >> } else { >> pthread_mutex_unlock(&exclusive_lock); >> - pthread_mutex_unlock(&tcg_ctx.tb_ctx.tb_lock); >> + qemu_mutex_unlock(&tcg_ctx.tb_ctx.tb_lock); >> } >> } >> =20 >> diff --git a/target-i386/mem_helper.c b/target-i386/mem_helper.c >> index 1aec8a5..7106cc3 100644 >> --- a/target-i386/mem_helper.c >> +++ b/target-i386/mem_helper.c >> @@ -23,17 +23,27 @@ >> =20 >> /* broken thread support */ >> =20 >> -static spinlock_t global_cpu_lock =3D SPIN_LOCK_UNLOCKED; >> +#if defined(CONFIG_USER_ONLY) >> +QemuMutex global_cpu_lock; >> =20 >> void helper_lock(void) >> { >> - spin_lock(&global_cpu_lock); >> + qemu_mutex_lock(&global_cpu_lock); >> } >> =20 >> void helper_unlock(void) >> { >> - spin_unlock(&global_cpu_lock); >> + qemu_mutex_unlock(&global_cpu_lock); >> } >> +#else >> +void helper_lock(void) >> +{ >> +} >> + >> +void helper_unlock(void) >> +{ >> +} >> +#endif >> =20 >> void helper_cmpxchg8b(CPUX86State *env, target_ulong a0) >> { >> diff --git a/tcg/i386/tcg-target.c b/tcg/i386/tcg-target.c >> index ff4d9cf..0d7c99c 100644 >> --- a/tcg/i386/tcg-target.c >> +++ b/tcg/i386/tcg-target.c >> @@ -24,6 +24,10 @@ >> =20 >> #include "tcg-be-ldst.h" >> =20 >> +#if defined(CONFIG_USER_ONLY) >> +extern QemuMutex global_cpu_lock; >> +#endif >> + >> #ifndef NDEBUG >> static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] =3D= { >> #if TCG_TARGET_REG_BITS =3D=3D 64 >> @@ -2342,6 +2346,10 @@ static void tcg_target_init(TCGContext *s) >> tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK); >> =20 >> tcg_add_target_add_op_defs(x86_op_defs); >> + >> +#if defined(CONFIG_USER_ONLY) >> + qemu_mutex_init(global_cpu_lock); >> +#endif >> } >> =20 >> typedef struct { > I wonder if it would be better splitting the patches: > > - Convert tb spinlocks to use tb_lock > - i386: convert lock helpers to QemuMutex > > before the final > > - Remove spinlocks Yes that makes sense I think. Fred >