#include "qemu/osdep.h"
#include "qemu-common.h"
#include "block/block.h"
-#include "qemu/queue.h"
+#include "qemu/rcu_queue.h"
#include "qemu/sockets.h"
#include "qemu/cutils.h"
#include "trace.h"
AioHandler *node;
struct epoll_event event;
- QLIST_FOREACH(node, &ctx->aio_handlers, node) {
+ QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
int r;
if (node->deleted || !node->pfd.events) {
continue;
bool is_new = false;
bool deleted = false;
+ qemu_lockcnt_lock(&ctx->list_lock);
+
node = find_aio_handler(ctx, fd);
/* Are we deleting the fd handler? */
if (!io_read && !io_write && !io_poll) {
if (node == NULL) {
+ qemu_lockcnt_unlock(&ctx->list_lock);
return;
}
g_source_remove_poll(&ctx->source, &node->pfd);
/* If the lock is held, just mark the node as deleted */
- if (ctx->walking_handlers) {
+ if (qemu_lockcnt_count(&ctx->list_lock)) {
node->deleted = 1;
node->pfd.revents = 0;
} else {
/* Otherwise, delete it for real. We can't just mark it as
- * deleted because deleted nodes are only cleaned up after
- * releasing the walking_handlers lock.
+ * deleted because deleted nodes are only cleaned up while
+ * no one is walking the handlers list.
*/
QLIST_REMOVE(node, node);
deleted = true;
/* Alloc and insert if it's not already there */
node = g_new0(AioHandler, 1);
node->pfd.fd = fd;
- QLIST_INSERT_HEAD(&ctx->aio_handlers, node, node);
+ QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node);
g_source_add_poll(&ctx->source, &node->pfd);
is_new = true;
}
aio_epoll_update(ctx, node, is_new);
+ qemu_lockcnt_unlock(&ctx->list_lock);
aio_notify(ctx);
if (deleted) {
ctx->poll_started = started;
- ctx->walking_handlers++;
- QLIST_FOREACH(node, &ctx->aio_handlers, node) {
+ qemu_lockcnt_inc(&ctx->list_lock);
+ QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
IOHandler *fn;
if (node->deleted) {
fn(node->opaque);
}
}
- ctx->walking_handlers--;
+ qemu_lockcnt_dec(&ctx->list_lock);
}
bool aio_pending(AioContext *ctx)
{
AioHandler *node;
+ bool result = false;
- QLIST_FOREACH(node, &ctx->aio_handlers, node) {
+ /*
+ * We have to walk very carefully in case aio_set_fd_handler is
+ * called while we're walking.
+ */
+ qemu_lockcnt_inc(&ctx->list_lock);
+
+ QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
int revents;
revents = node->pfd.revents & node->pfd.events;
if (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR) && node->io_read &&
aio_node_check(ctx, node->is_external)) {
- return true;
+ result = true;
+ break;
}
if (revents & (G_IO_OUT | G_IO_ERR) && node->io_write &&
aio_node_check(ctx, node->is_external)) {
- return true;
+ result = true;
+ break;
}
}
+ qemu_lockcnt_dec(&ctx->list_lock);
- return false;
+ return result;
}
-/*
- * Note that dispatch_fds == false has the side-effect of post-poning the
- * freeing of deleted handlers.
- */
-bool aio_dispatch(AioContext *ctx, bool dispatch_fds)
+static bool aio_dispatch_handlers(AioContext *ctx)
{
- AioHandler *node = NULL;
+ AioHandler *node, *tmp;
bool progress = false;
- /*
- * If there are callbacks left that have been queued, we need to call them.
- * Do not call select in this case, because it is possible that the caller
- * does not need a complete flush (as is the case for aio_poll loops).
- */
- if (aio_bh_poll(ctx)) {
- progress = true;
- }
-
/*
* We have to walk very carefully in case aio_set_fd_handler is
* called while we're walking.
*/
- if (dispatch_fds) {
- node = QLIST_FIRST(&ctx->aio_handlers);
- }
- while (node) {
- AioHandler *tmp;
- int revents;
+ qemu_lockcnt_inc(&ctx->list_lock);
- ctx->walking_handlers++;
+ QLIST_FOREACH_SAFE_RCU(node, &ctx->aio_handlers, node, tmp) {
+ int revents;
revents = node->pfd.revents & node->pfd.events;
node->pfd.revents = 0;
progress = true;
}
- tmp = node;
- node = QLIST_NEXT(node, node);
+ if (node->deleted) {
+ if (qemu_lockcnt_dec_if_lock(&ctx->list_lock)) {
+ QLIST_REMOVE(node, node);
+ g_free(node);
+ qemu_lockcnt_inc_and_unlock(&ctx->list_lock);
+ }
+ }
+ }
+
+ qemu_lockcnt_dec(&ctx->list_lock);
+ return progress;
+}
- ctx->walking_handlers--;
+/*
+ * Note that dispatch_fds == false has the side-effect of post-poning the
+ * freeing of deleted handlers.
+ */
+bool aio_dispatch(AioContext *ctx, bool dispatch_fds)
+{
+ bool progress;
- if (!ctx->walking_handlers && tmp->deleted) {
- QLIST_REMOVE(tmp, node);
- g_free(tmp);
- }
+ /*
+ * If there are callbacks left that have been queued, we need to call them.
+ * Do not call select in this case, because it is possible that the caller
+ * does not need a complete flush (as is the case for aio_poll loops).
+ */
+ progress = aio_bh_poll(ctx);
+
+ if (dispatch_fds) {
+ progress |= aio_dispatch_handlers(ctx);
}
/* Run our timers */
bool progress = false;
AioHandler *node;
- QLIST_FOREACH(node, &ctx->aio_handlers, node) {
+ QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
if (!node->deleted && node->io_poll &&
node->io_poll(node->opaque)) {
progress = true;
* Note that ctx->notify_me must be non-zero so this function can detect
* aio_notify().
*
- * Note that the caller must have incremented ctx->walking_handlers.
+ * Note that the caller must have incremented ctx->list_lock.
*
* Returns: true if progress was made, false otherwise
*/
int64_t end_time;
assert(ctx->notify_me);
- assert(ctx->walking_handlers > 0);
+ assert(qemu_lockcnt_count(&ctx->list_lock) > 0);
assert(ctx->poll_disable_cnt == 0);
trace_run_poll_handlers_begin(ctx, max_ns);
*
* ctx->notify_me must be non-zero so this function can detect aio_notify().
*
- * Note that the caller must have incremented ctx->walking_handlers.
+ * Note that the caller must have incremented ctx->list_lock.
*
* Returns: true if progress was made, false otherwise
*/
if (blocking && ctx->poll_max_ns && ctx->poll_disable_cnt == 0) {
/* See qemu_soonest_timeout() uint64_t hack */
int64_t max_ns = MIN((uint64_t)aio_compute_timeout(ctx),
- (uint64_t)ctx->poll_max_ns);
+ (uint64_t)ctx->poll_ns);
if (max_ns) {
poll_set_started(ctx, true);
int ret = 0;
bool progress;
int64_t timeout;
+ int64_t start = 0;
aio_context_acquire(ctx);
progress = false;
atomic_add(&ctx->notify_me, 2);
}
- ctx->walking_handlers++;
+ qemu_lockcnt_inc(&ctx->list_lock);
+
+ if (ctx->poll_max_ns) {
+ start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
+ }
if (try_poll_mode(ctx, blocking)) {
progress = true;
/* fill pollfds */
if (!aio_epoll_enabled(ctx)) {
- QLIST_FOREACH(node, &ctx->aio_handlers, node) {
+ QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
if (!node->deleted && node->pfd.events
&& aio_node_check(ctx, node->is_external)) {
add_pollfd(node);
atomic_sub(&ctx->notify_me, 2);
}
+ /* Adjust polling time */
+ if (ctx->poll_max_ns) {
+ int64_t block_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start;
+
+ if (block_ns <= ctx->poll_ns) {
+ /* This is the sweet spot, no adjustment needed */
+ } else if (block_ns > ctx->poll_max_ns) {
+ /* We'd have to poll for too long, poll less */
+ int64_t old = ctx->poll_ns;
+
+ if (ctx->poll_shrink) {
+ ctx->poll_ns /= ctx->poll_shrink;
+ } else {
+ ctx->poll_ns = 0;
+ }
+
+ trace_poll_shrink(ctx, old, ctx->poll_ns);
+ } else if (ctx->poll_ns < ctx->poll_max_ns &&
+ block_ns < ctx->poll_max_ns) {
+ /* There is room to grow, poll longer */
+ int64_t old = ctx->poll_ns;
+ int64_t grow = ctx->poll_grow;
+
+ if (grow == 0) {
+ grow = 2;
+ }
+
+ if (ctx->poll_ns) {
+ ctx->poll_ns *= grow;
+ } else {
+ ctx->poll_ns = 4000; /* start polling at 4 microseconds */
+ }
+
+ if (ctx->poll_ns > ctx->poll_max_ns) {
+ ctx->poll_ns = ctx->poll_max_ns;
+ }
+
+ trace_poll_grow(ctx, old, ctx->poll_ns);
+ }
+ }
+
aio_notify_accept(ctx);
/* if we have any readable fds, dispatch event */
}
npfd = 0;
- ctx->walking_handlers--;
+ qemu_lockcnt_dec(&ctx->list_lock);
/* Run dispatch even if there were no readable fds to run timers */
if (aio_dispatch(ctx, ret > 0)) {
#endif
}
-void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns, Error **errp)
+void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
+ int64_t grow, int64_t shrink, Error **errp)
{
- /* No thread synchronization here, it doesn't matter if an incorrect poll
- * timeout is used once.
+ /* No thread synchronization here, it doesn't matter if an incorrect value
+ * is used once.
*/
ctx->poll_max_ns = max_ns;
+ ctx->poll_ns = 0;
+ ctx->poll_grow = grow;
+ ctx->poll_shrink = shrink;
aio_notify(ctx);
}