From 06831001ac8949b0801e0d20c347d97339769a20 Mon Sep 17 00:00:00 2001 From: Paolo Bonzini Date: Fri, 3 Mar 2023 14:37:51 +0100 Subject: [PATCH] atomics: eliminate mb_read/mb_set qatomic_mb_read and qatomic_mb_set were the very first atomic primitives introduced for QEMU; their semantics are unclear and they provide a false sense of safety. The last use of qatomic_mb_read() has been removed, so delete it. qatomic_mb_set() instead can survive as an optimized qatomic_set()+smp_mb(), similar to Linux's smp_store_mb(), but rename it to qatomic_set_mb() to match the order of the two operations. Reviewed-by: Richard Henderson Signed-off-by: Paolo Bonzini --- accel/tcg/cpu-exec.c | 2 +- accel/tcg/tcg-accel-ops-mttcg.c | 2 +- accel/tcg/tcg-accel-ops-rr.c | 4 ++-- docs/devel/atomics.rst | 27 ++++----------------------- include/qemu/atomic.h | 17 +++++------------ monitor/qmp.c | 2 +- softmmu/cpus.c | 2 +- softmmu/physmem.c | 2 +- target/arm/hvf/hvf.c | 2 +- tests/unit/test-aio-multithread.c | 2 +- util/qemu-coroutine-lock.c | 4 ++-- 11 files changed, 20 insertions(+), 46 deletions(-) diff --git a/accel/tcg/cpu-exec.c b/accel/tcg/cpu-exec.c index 1cf4f1fa22..42086525d7 100644 --- a/accel/tcg/cpu-exec.c +++ b/accel/tcg/cpu-exec.c @@ -774,7 +774,7 @@ static inline bool cpu_handle_interrupt(CPUState *cpu, * Ensure zeroing happens before reading cpu->exit_request or * cpu->interrupt_request (see also smp_wmb in cpu_exit()) */ - qatomic_mb_set(&cpu_neg(cpu)->icount_decr.u16.high, 0); + qatomic_set_mb(&cpu_neg(cpu)->icount_decr.u16.high, 0); if (unlikely(qatomic_read(&cpu->interrupt_request))) { int interrupt_request; diff --git a/accel/tcg/tcg-accel-ops-mttcg.c b/accel/tcg/tcg-accel-ops-mttcg.c index 5d72c9b1bd..b320ff0037 100644 --- a/accel/tcg/tcg-accel-ops-mttcg.c +++ b/accel/tcg/tcg-accel-ops-mttcg.c @@ -119,7 +119,7 @@ static void *mttcg_cpu_thread_fn(void *arg) } } - qatomic_mb_set(&cpu->exit_request, 0); + qatomic_set_mb(&cpu->exit_request, 0); qemu_wait_io_event(cpu); } while (!cpu->unplug || cpu_can_run(cpu)); diff --git a/accel/tcg/tcg-accel-ops-rr.c b/accel/tcg/tcg-accel-ops-rr.c index 70b9b89073..23e4d0f452 100644 --- a/accel/tcg/tcg-accel-ops-rr.c +++ b/accel/tcg/tcg-accel-ops-rr.c @@ -244,7 +244,7 @@ static void *rr_cpu_thread_fn(void *arg) while (cpu && cpu_work_list_empty(cpu) && !cpu->exit_request) { /* Store rr_current_cpu before evaluating cpu_can_run(). */ - qatomic_mb_set(&rr_current_cpu, cpu); + qatomic_set_mb(&rr_current_cpu, cpu); current_cpu = cpu; @@ -287,7 +287,7 @@ static void *rr_cpu_thread_fn(void *arg) qatomic_set(&rr_current_cpu, NULL); if (cpu && cpu->exit_request) { - qatomic_mb_set(&cpu->exit_request, 0); + qatomic_set_mb(&cpu->exit_request, 0); } if (icount_enabled() && all_cpu_threads_idle()) { diff --git a/docs/devel/atomics.rst b/docs/devel/atomics.rst index 248076375b..ff9b5ee30c 100644 --- a/docs/devel/atomics.rst +++ b/docs/devel/atomics.rst @@ -102,28 +102,10 @@ Similar operations return the new value of ``*ptr``:: typeof(*ptr) qatomic_or_fetch(ptr, val) typeof(*ptr) qatomic_xor_fetch(ptr, val) -``qemu/atomic.h`` also provides loads and stores that cannot be reordered -with each other:: +``qemu/atomic.h`` also provides an optimized shortcut for +``qatomic_set`` followed by ``smp_mb``:: - typeof(*ptr) qatomic_mb_read(ptr) - void qatomic_mb_set(ptr, val) - -However these do not provide sequential consistency and, in particular, -they do not participate in the total ordering enforced by -sequentially-consistent operations. For this reason they are deprecated. -They should instead be replaced with any of the following (ordered from -easiest to hardest): - -- accesses inside a mutex or spinlock - -- lightweight synchronization primitives such as ``QemuEvent`` - -- RCU operations (``qatomic_rcu_read``, ``qatomic_rcu_set``) when publishing - or accessing a new version of a data structure - -- other atomic accesses: ``qatomic_read`` and ``qatomic_load_acquire`` for - loads, ``qatomic_set`` and ``qatomic_store_release`` for stores, ``smp_mb`` - to forbid reordering subsequent loads before a store. + void qatomic_set_mb(ptr, val) Weak atomic access and manual memory barriers @@ -523,8 +505,7 @@ and memory barriers, and the equivalents in QEMU: | :: | | | | a = qatomic_read(&x); | - | qatomic_set(&x, a + 2); | - | smp_mb(); | + | qatomic_set_mb(&x, a + 2); | | b = qatomic_read(&y); | +--------------------------------+ diff --git a/include/qemu/atomic.h b/include/qemu/atomic.h index f85834ee8b..d95612f7a0 100644 --- a/include/qemu/atomic.h +++ b/include/qemu/atomic.h @@ -259,24 +259,17 @@ # define smp_mb__after_rmw() smp_mb() #endif -/* qatomic_mb_read/set semantics map Java volatile variables. They are - * less expensive on some platforms (notably POWER) than fully - * sequentially consistent operations. - * - * As long as they are used as paired operations they are safe to - * use. See docs/devel/atomics.rst for more discussion. +/* + * On some architectures, qatomic_set_mb is more efficient than a store + * plus a fence. */ -#define qatomic_mb_read(ptr) \ - qatomic_load_acquire(ptr) - #if !defined(QEMU_SANITIZE_THREAD) && \ (defined(__i386__) || defined(__x86_64__) || defined(__s390x__)) -/* This is more efficient than a store plus a fence. */ -# define qatomic_mb_set(ptr, i) \ +# define qatomic_set_mb(ptr, i) \ ({ (void)qatomic_xchg(ptr, i); smp_mb__after_rmw(); }) #else -# define qatomic_mb_set(ptr, i) \ +# define qatomic_set_mb(ptr, i) \ ({ qatomic_store_release(ptr, i); smp_mb(); }) #endif diff --git a/monitor/qmp.c b/monitor/qmp.c index c8e0156974..6eee450fe4 100644 --- a/monitor/qmp.c +++ b/monitor/qmp.c @@ -246,7 +246,7 @@ static QMPRequest *monitor_qmp_dispatcher_pop_any(void) * * Clear qmp_dispatcher_co_busy before reading request. */ - qatomic_mb_set(&qmp_dispatcher_co_busy, false); + qatomic_set_mb(&qmp_dispatcher_co_busy, false); WITH_QEMU_LOCK_GUARD(&monitor_lock) { QMPRequest *req_obj; diff --git a/softmmu/cpus.c b/softmmu/cpus.c index 9cbc8172b5..fed20ffb5d 100644 --- a/softmmu/cpus.c +++ b/softmmu/cpus.c @@ -405,7 +405,7 @@ static void qemu_cpu_stop(CPUState *cpu, bool exit) void qemu_wait_io_event_common(CPUState *cpu) { - qatomic_mb_set(&cpu->thread_kicked, false); + qatomic_set_mb(&cpu->thread_kicked, false); if (cpu->stop) { qemu_cpu_stop(cpu, false); } diff --git a/softmmu/physmem.c b/softmmu/physmem.c index 9d7e172260..588d0d166b 100644 --- a/softmmu/physmem.c +++ b/softmmu/physmem.c @@ -3132,7 +3132,7 @@ void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len, bounce.buffer = NULL; memory_region_unref(bounce.mr); /* Clear in_use before reading map_client_list. */ - qatomic_mb_set(&bounce.in_use, false); + qatomic_set_mb(&bounce.in_use, false); cpu_notify_map_clients(); } diff --git a/target/arm/hvf/hvf.c b/target/arm/hvf/hvf.c index ad65603445..5900dc788f 100644 --- a/target/arm/hvf/hvf.c +++ b/target/arm/hvf/hvf.c @@ -1229,7 +1229,7 @@ static void hvf_wait_for_ipi(CPUState *cpu, struct timespec *ts) * Use pselect to sleep so that other threads can IPI us while we're * sleeping. */ - qatomic_mb_set(&cpu->thread_kicked, false); + qatomic_set_mb(&cpu->thread_kicked, false); qemu_mutex_unlock_iothread(); pselect(0, 0, 0, 0, ts, &cpu->hvf->unblock_ipi_mask); qemu_mutex_lock_iothread(); diff --git a/tests/unit/test-aio-multithread.c b/tests/unit/test-aio-multithread.c index 80c5d4e2e6..08d4570ccb 100644 --- a/tests/unit/test-aio-multithread.c +++ b/tests/unit/test-aio-multithread.c @@ -154,7 +154,7 @@ static coroutine_fn void test_multi_co_schedule_entry(void *opaque) n = g_test_rand_int_range(0, NUM_CONTEXTS); schedule_next(n); - qatomic_mb_set(&to_schedule[id], qemu_coroutine_self()); + qatomic_set_mb(&to_schedule[id], qemu_coroutine_self()); /* finish_cb can run here. */ qemu_coroutine_yield(); g_assert(to_schedule[id] == NULL); diff --git a/util/qemu-coroutine-lock.c b/util/qemu-coroutine-lock.c index 84a50a9e91..2534435388 100644 --- a/util/qemu-coroutine-lock.c +++ b/util/qemu-coroutine-lock.c @@ -202,7 +202,7 @@ static void coroutine_fn qemu_co_mutex_lock_slowpath(AioContext *ctx, push_waiter(mutex, &w); /* - * Add waiter before reading mutex->handoff. Pairs with qatomic_mb_set + * Add waiter before reading mutex->handoff. Pairs with qatomic_set_mb * in qemu_co_mutex_unlock. */ smp_mb__after_rmw(); @@ -310,7 +310,7 @@ void coroutine_fn qemu_co_mutex_unlock(CoMutex *mutex) our_handoff = mutex->sequence; /* Set handoff before checking for waiters. */ - qatomic_mb_set(&mutex->handoff, our_handoff); + qatomic_set_mb(&mutex->handoff, our_handoff); if (!has_waiters(mutex)) { /* The concurrent lock has not added itself yet, so it * will be able to pick our handoff. -- 2.11.0