From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:40429) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1eEWPV-0008Mt-8S for qemu-devel@nongnu.org; Tue, 14 Nov 2017 03:17:34 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1eEWPT-0006Ql-UP for qemu-devel@nongnu.org; Tue, 14 Nov 2017 03:17:33 -0500 Received: from mail.ispras.ru ([83.149.199.45]:52660) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1eEWPT-0006Q4-Fq for qemu-devel@nongnu.org; Tue, 14 Nov 2017 03:17:31 -0500 From: Pavel Dovgalyuk Date: Tue, 14 Nov 2017 11:17:39 +0300 Message-ID: <20171114081739.27640.85769.stgit@pasha-VirtualBox> In-Reply-To: <20171114081630.27640.53933.stgit@pasha-VirtualBox> References: <20171114081630.27640.53933.stgit@pasha-VirtualBox> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable Subject: [Qemu-devel] [RFC PATCH v2 12/26] cpus: push BQL lock to qemu_*_wait_io_event List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: qemu-devel@nongnu.org Cc: kwolf@redhat.com, peter.maydell@linaro.org, boost.lists@gmail.com, quintela@redhat.com, jasowang@redhat.com, mst@redhat.com, zuban32s@gmail.com, maria.klimushenkova@ispras.ru, dovgaluk@ispras.ru, kraxel@redhat.com, pavel.dovgaluk@ispras.ru, pbonzini@redhat.com, alex.bennee@linaro.org From: Alex Benn=C3=A9e We only really need to grab the lock for initial setup (so we don't race with the thread-spawning thread). After that we can drop the lock for the whole main loop and only grab it for waiting for IO events. There is a slight wrinkle for the round-robin TCG thread as we also expire timers which needs to be done under BQL as they are in the main-loop. This is stage one of reducing the lock impact as we can drop the requirement of implicit BQL for async work and only grab the lock when we need to sleep on the cpu->halt_cond. Signed-off-by: Alex Benn=C3=A9e Tested-by: Pavel Dovgalyuk --- accel/kvm/kvm-all.c | 4 ---- cpus.c | 27 ++++++++++++++++++++------- target/i386/hax-all.c | 2 -- 3 files changed, 20 insertions(+), 13 deletions(-) diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c index f290f48..8d1d2c4 100644 --- a/accel/kvm/kvm-all.c +++ b/accel/kvm/kvm-all.c @@ -1857,9 +1857,7 @@ int kvm_cpu_exec(CPUState *cpu) return EXCP_HLT; } =20 - qemu_mutex_unlock_iothread(); cpu_exec_start(cpu); - do { MemTxAttrs attrs; =20 @@ -1989,8 +1987,6 @@ int kvm_cpu_exec(CPUState *cpu) } while (ret =3D=3D 0); =20 cpu_exec_end(cpu); - qemu_mutex_lock_iothread(); - if (ret < 0) { cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE); vm_stop(RUN_STATE_INTERNAL_ERROR); diff --git a/cpus.c b/cpus.c index 7a3abea..3aeba5d 100644 --- a/cpus.c +++ b/cpus.c @@ -1127,6 +1127,8 @@ static bool qemu_tcg_should_sleep(CPUState *cpu) =20 static void qemu_tcg_wait_io_event(CPUState *cpu) { + qemu_mutex_lock_iothread(); + while (qemu_tcg_should_sleep(cpu)) { stop_tcg_kick_timer(); qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); @@ -1135,15 +1137,21 @@ static void qemu_tcg_wait_io_event(CPUState *cpu) start_tcg_kick_timer(); =20 qemu_wait_io_event_common(cpu); + + qemu_mutex_unlock_iothread(); } =20 static void qemu_kvm_wait_io_event(CPUState *cpu) { + qemu_mutex_lock_iothread(); + while (cpu_thread_is_idle(cpu)) { qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); } =20 qemu_wait_io_event_common(cpu); + + qemu_mutex_unlock_iothread(); } =20 static void *qemu_kvm_cpu_thread_fn(void *arg) @@ -1169,6 +1177,8 @@ static void *qemu_kvm_cpu_thread_fn(void *arg) =20 /* signal CPU creation */ cpu->created =3D true; + qemu_mutex_unlock_iothread(); + qemu_cond_signal(&qemu_cpu_cond); =20 do { @@ -1211,10 +1221,10 @@ static void *qemu_dummy_cpu_thread_fn(void *arg) =20 /* signal CPU creation */ cpu->created =3D true; + qemu_mutex_unlock_iothread(); qemu_cond_signal(&qemu_cpu_cond); =20 while (1) { - qemu_mutex_unlock_iothread(); do { int sig; r =3D sigwait(&waitset, &sig); @@ -1225,6 +1235,7 @@ static void *qemu_dummy_cpu_thread_fn(void *arg) } qemu_mutex_lock_iothread(); qemu_wait_io_event_common(cpu); + qemu_mutex_unlock_iothread(); } =20 return NULL; @@ -1313,11 +1324,9 @@ static int tcg_cpu_exec(CPUState *cpu) #ifdef CONFIG_PROFILER ti =3D profile_getclock(); #endif - qemu_mutex_unlock_iothread(); cpu_exec_start(cpu); ret =3D cpu_exec(cpu); cpu_exec_end(cpu); - qemu_mutex_lock_iothread(); #ifdef CONFIG_PROFILER tcg_time +=3D profile_getclock() - ti; #endif @@ -1377,6 +1386,7 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg) qemu_wait_io_event_common(cpu); } } + qemu_mutex_unlock_iothread(); =20 start_tcg_kick_timer(); =20 @@ -1386,6 +1396,9 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg) cpu->exit_request =3D 1; =20 while (1) { + + qemu_mutex_lock_iothread(); + /* Account partial waits to QEMU_CLOCK_VIRTUAL. */ qemu_account_warp_timer(); =20 @@ -1394,6 +1407,8 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg) */ handle_icount_deadline(); =20 + qemu_mutex_unlock_iothread(); + if (!cpu) { cpu =3D first_cpu; } @@ -1419,9 +1434,7 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg) cpu_handle_guest_debug(cpu); break; } else if (r =3D=3D EXCP_ATOMIC) { - qemu_mutex_unlock_iothread(); cpu_exec_step_atomic(cpu); - qemu_mutex_lock_iothread(); break; } } else if (cpu->stop) { @@ -1462,6 +1475,7 @@ static void *qemu_hax_cpu_thread_fn(void *arg) current_cpu =3D cpu; =20 hax_init_vcpu(cpu); + qemu_mutex_unlock_iothread(); qemu_cond_signal(&qemu_cpu_cond); =20 while (1) { @@ -1512,6 +1526,7 @@ static void *qemu_tcg_cpu_thread_fn(void *arg) cpu->created =3D true; cpu->can_do_io =3D 1; current_cpu =3D cpu; + qemu_mutex_unlock_iothread(); qemu_cond_signal(&qemu_cpu_cond); =20 /* process any pending work */ @@ -1536,9 +1551,7 @@ static void *qemu_tcg_cpu_thread_fn(void *arg) g_assert(cpu->halted); break; case EXCP_ATOMIC: - qemu_mutex_unlock_iothread(); cpu_exec_step_atomic(cpu); - qemu_mutex_lock_iothread(); default: /* Ignore everything else? */ break; diff --git a/target/i386/hax-all.c b/target/i386/hax-all.c index 3ce6950..9fd60d9 100644 --- a/target/i386/hax-all.c +++ b/target/i386/hax-all.c @@ -513,11 +513,9 @@ static int hax_vcpu_hax_exec(CPUArchState *env) =20 hax_vcpu_interrupt(env); =20 - qemu_mutex_unlock_iothread(); cpu_exec_start(cpu); hax_ret =3D hax_vcpu_run(vcpu); cpu_exec_end(cpu); - qemu_mutex_lock_iothread(); =20 /* Simply continue the vcpu_run if system call interrupted */ if (hax_ret =3D=3D -EINTR || hax_ret =3D=3D -EAGAIN) {