linux-aio: share one LinuxAioState within an AioContext

This has better performance because it executes fewer system calls
and does not use a bottom half per disk.

Originally proposed by Ming Lei.

[Changed #include "raw-aio.h" to "block/raw-aio.h" in win32-aio.c to fix
build error as reported by Peter Maydell <peter.maydell@linaro.org>.
--Stefan]

Acked-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1467650000-51385-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

squash! linux-aio: share one LinuxAioState within an AioContext

7 files changed, 58 insertions, 111 deletions

diff --git a/async.c b/async.c
index b4bf205a0c..6caa98cd92 100644
--- a/async.c
+++ b/async.c
@@ -29,6 +29,7 @@
 #include "block/thread-pool.h"
 #include "qemu/main-loop.h"
 #include "qemu/atomic.h"
+#include "block/raw-aio.h"
 
 /***********************************************************/
 /* bottom halves (can be seen as timers which expire ASAP) */
@@ -242,6 +243,14 @@ aio_ctx_finalize(GSource     *source)
     qemu_bh_delete(ctx->notify_dummy_bh);
     thread_pool_free(ctx->thread_pool);
 
+#ifdef CONFIG_LINUX_AIO
+    if (ctx->linux_aio) {
+        laio_detach_aio_context(ctx->linux_aio, ctx);
+        laio_cleanup(ctx->linux_aio);
+        ctx->linux_aio = NULL;
+    }
+#endif
+
     qemu_mutex_lock(&ctx->bh_lock);
     while (ctx->first_bh) {
         QEMUBH *next = ctx->first_bh->next;
@@ -282,6 +291,17 @@ ThreadPool *aio_get_thread_pool(AioContext *ctx)
     return ctx->thread_pool;
 }
 
+#ifdef CONFIG_LINUX_AIO
+LinuxAioState *aio_get_linux_aio(AioContext *ctx)
+{
+    if (!ctx->linux_aio) {
+        ctx->linux_aio = laio_init();
+        laio_attach_aio_context(ctx->linux_aio, ctx);
+    }
+    return ctx->linux_aio;
+}
+#endif
+
 void aio_notify(AioContext *ctx)
 {
     /* Write e.g. bh->scheduled before reading ctx->notify_me.  Pairs
@@ -345,6 +365,9 @@ AioContext *aio_context_new(Error **errp)
                            false,
                            (EventNotifierHandler *)
                            event_notifier_dummy_cb);
+#ifdef CONFIG_LINUX_AIO
+    ctx->linux_aio = NULL;
+#endif
     ctx->thread_pool = NULL;
     qemu_mutex_init(&ctx->bh_lock);
     rfifolock_init(&ctx->lock, aio_rfifolock_cb, ctx);
diff --git a/block/linux-aio.c b/block/linux-aio.c
index 5c104bd3cd..1d702a5fc6 100644
--- a/block/linux-aio.c
+++ b/block/linux-aio.c
@@ -50,6 +50,8 @@ typedef struct {
 } LaioQueue;
 
 struct LinuxAioState {
+    AioContext *aio_context;
+
     io_context_t ctx;
     EventNotifier e;
 
@@ -227,15 +229,14 @@ static void ioq_submit(LinuxAioState *s)
 
 void laio_io_plug(BlockDriverState *bs, LinuxAioState *s)
 {
-    assert(!s->io_q.plugged);
-    s->io_q.plugged = 1;
+    s->io_q.plugged++;
 }
 
 void laio_io_unplug(BlockDriverState *bs, LinuxAioState *s)
 {
     assert(s->io_q.plugged);
-    s->io_q.plugged = 0;
-    if (!s->io_q.blocked && !QSIMPLEQ_EMPTY(&s->io_q.pending)) {
+    if (--s->io_q.plugged == 0 &&
+        !s->io_q.blocked && !QSIMPLEQ_EMPTY(&s->io_q.pending)) {
         ioq_submit(s);
     }
 }
@@ -325,6 +326,7 @@ void laio_detach_aio_context(LinuxAioState *s, AioContext *old_context)
 
 void laio_attach_aio_context(LinuxAioState *s, AioContext *new_context)
 {
+    s->aio_context = new_context;
     s->completion_bh = aio_bh_new(new_context, qemu_laio_completion_bh, s);
     aio_set_event_notifier(new_context, &s->e, false,
                            qemu_laio_completion_cb);
diff --git a/block/raw-posix.c b/block/raw-posix.c
index d1c3bd8e47..20f4d7aa8d 100644
--- a/block/raw-posix.c
+++ b/block/raw-posix.c
@@ -32,7 +32,7 @@
 #include "trace.h"
 #include "block/thread-pool.h"
 #include "qemu/iov.h"
-#include "raw-aio.h"
+#include "block/raw-aio.h"
 #include "qapi/util.h"
 #include "qapi/qmp/qstring.h"
 
@@ -137,10 +137,6 @@ typedef struct BDRVRawState {
     int open_flags;
     size_t buf_align;
 
-#ifdef CONFIG_LINUX_AIO
-    int use_aio;
-    LinuxAioState *aio_ctx;
-#endif
 #ifdef CONFIG_XFS
     bool is_xfs:1;
 #endif
@@ -154,9 +150,6 @@ typedef struct BDRVRawState {
 typedef struct BDRVRawReopenState {
     int fd;
     int open_flags;
-#ifdef CONFIG_LINUX_AIO
-    int use_aio;
-#endif
 } BDRVRawReopenState;
 
 static int fd_open(BlockDriverState *bs);
@@ -374,58 +367,15 @@ static void raw_parse_flags(int bdrv_flags, int *open_flags)
     }
 }
 
-static void raw_detach_aio_context(BlockDriverState *bs)
-{
 #ifdef CONFIG_LINUX_AIO
-    BDRVRawState *s = bs->opaque;
-
-    if (s->use_aio) {
-        laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs));
-    }
-#endif
-}
-
-static void raw_attach_aio_context(BlockDriverState *bs,
-                                   AioContext *new_context)
+static bool raw_use_aio(int bdrv_flags)
 {
-#ifdef CONFIG_LINUX_AIO
-    BDRVRawState *s = bs->opaque;
-
-    if (s->use_aio) {
-        laio_attach_aio_context(s->aio_ctx, new_context);
-    }
-#endif
-}
-
-#ifdef CONFIG_LINUX_AIO
-static int raw_set_aio(LinuxAioState **aio_ctx, int *use_aio, int bdrv_flags)
-{
-    int ret = -1;
-    assert(aio_ctx != NULL);
-    assert(use_aio != NULL);
     /*
      * Currently Linux do AIO only for files opened with O_DIRECT
      * specified so check NOCACHE flag too
      */
-    if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
-                      (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
-
-        /* if non-NULL, laio_init() has already been run */
-        if (*aio_ctx == NULL) {
-            *aio_ctx = laio_init();
-            if (!*aio_ctx) {
-                goto error;
-            }
-        }
-        *use_aio = 1;
-    } else {
-        *use_aio = 0;
-    }
-
-    ret = 0;
-
-error:
-    return ret;
+    return (bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
+                         (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO);
 }
 #endif
 
@@ -494,13 +444,7 @@ static int raw_open_common(BlockDriverState *bs, QDict *options,
     s->fd = fd;
 
 #ifdef CONFIG_LINUX_AIO
-    if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
-        qemu_close(fd);
-        ret = -errno;
-        error_setg_errno(errp, -ret, "Could not set AIO state");
-        goto fail;
-    }
-    if (!s->use_aio && (bdrv_flags & BDRV_O_NATIVE_AIO)) {
+    if (!raw_use_aio(bdrv_flags) && (bdrv_flags & BDRV_O_NATIVE_AIO)) {
         error_setg(errp, "aio=native was specified, but it requires "
                          "cache.direct=on, which was not specified.");
         ret = -EINVAL;
@@ -567,8 +511,6 @@ static int raw_open_common(BlockDriverState *bs, QDict *options,
     }
 #endif
 
-    raw_attach_aio_context(bs, bdrv_get_aio_context(bs));
-
     ret = 0;
 fail:
     if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
@@ -603,18 +545,6 @@ static int raw_reopen_prepare(BDRVReopenState *state,
     state->opaque = g_new0(BDRVRawReopenState, 1);
     raw_s = state->opaque;
 
-#ifdef CONFIG_LINUX_AIO
-    raw_s->use_aio = s->use_aio;
-
-    /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
-     * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
-     * won't override aio_ctx if aio_ctx is non-NULL */
-    if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
-        error_setg(errp, "Could not set AIO state");
-        return -1;
-    }
-#endif
-
     if (s->type == FTYPE_CD) {
         raw_s->open_flags |= O_NONBLOCK;
     }
@@ -689,9 +619,6 @@ static void raw_reopen_commit(BDRVReopenState *state)
 
     qemu_close(s->fd);
     s->fd = raw_s->fd;
-#ifdef CONFIG_LINUX_AIO
-    s->use_aio = raw_s->use_aio;
-#endif
 
     g_free(state->opaque);
     state->opaque = NULL;
@@ -1329,9 +1256,10 @@ static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
         if (!bdrv_qiov_is_aligned(bs, qiov)) {
             type |= QEMU_AIO_MISALIGNED;
 #ifdef CONFIG_LINUX_AIO
-        } else if (s->use_aio) {
+        } else if (bs->open_flags & BDRV_O_NATIVE_AIO) {
+            LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
             assert(qiov->size == bytes);
-            return laio_co_submit(bs, s->aio_ctx, s->fd, offset, qiov, type);
+            return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
 #endif
         }
     }
@@ -1357,9 +1285,9 @@ static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
 static void raw_aio_plug(BlockDriverState *bs)
 {
 #ifdef CONFIG_LINUX_AIO
-    BDRVRawState *s = bs->opaque;
-    if (s->use_aio) {
-        laio_io_plug(bs, s->aio_ctx);
+    if (bs->open_flags & BDRV_O_NATIVE_AIO) {
+        LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
+        laio_io_plug(bs, aio);
     }
 #endif
 }
@@ -1367,9 +1295,9 @@ static void raw_aio_plug(BlockDriverState *bs)
 static void raw_aio_unplug(BlockDriverState *bs)
 {
 #ifdef CONFIG_LINUX_AIO
-    BDRVRawState *s = bs->opaque;
-    if (s->use_aio) {
-        laio_io_unplug(bs, s->aio_ctx);
+    if (bs->open_flags & BDRV_O_NATIVE_AIO) {
+        LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
+        laio_io_unplug(bs, aio);
     }
 #endif
 }
@@ -1389,13 +1317,6 @@ static void raw_close(BlockDriverState *bs)
 {
     BDRVRawState *s = bs->opaque;
 
-    raw_detach_aio_context(bs);
-
-#ifdef CONFIG_LINUX_AIO
-    if (s->use_aio) {
-        laio_cleanup(s->aio_ctx);
-    }
-#endif
     if (s->fd >= 0) {
         qemu_close(s->fd);
         s->fd = -1;
@@ -1954,9 +1875,6 @@ BlockDriver bdrv_file = {
     .bdrv_get_allocated_file_size
                         = raw_get_allocated_file_size,
 
-    .bdrv_detach_aio_context = raw_detach_aio_context,
-    .bdrv_attach_aio_context = raw_attach_aio_context,
-
     .create_opts = &raw_create_opts,
 };
 
@@ -2402,9 +2320,6 @@ static BlockDriver bdrv_host_device = {
     .bdrv_probe_blocksizes = hdev_probe_blocksizes,
     .bdrv_probe_geometry = hdev_probe_geometry,
 
-    .bdrv_detach_aio_context = raw_detach_aio_context,
-    .bdrv_attach_aio_context = raw_attach_aio_context,
-
     /* generic scsi device */
 #ifdef __linux__
     .bdrv_aio_ioctl     = hdev_aio_ioctl,
@@ -2524,9 +2439,6 @@ static BlockDriver bdrv_host_cdrom = {
     .bdrv_get_allocated_file_size
                         = raw_get_allocated_file_size,
 
-    .bdrv_detach_aio_context = raw_detach_aio_context,
-    .bdrv_attach_aio_context = raw_attach_aio_context,
-
     /* removable device support */
     .bdrv_is_inserted   = cdrom_is_inserted,
     .bdrv_eject         = cdrom_eject,
@@ -2657,9 +2569,6 @@ static BlockDriver bdrv_host_cdrom = {
     .bdrv_get_allocated_file_size
                         = raw_get_allocated_file_size,
 
-    .bdrv_detach_aio_context = raw_detach_aio_context,
-    .bdrv_attach_aio_context = raw_attach_aio_context,
-
     /* removable device support */
     .bdrv_is_inserted   = cdrom_is_inserted,
     .bdrv_eject         = cdrom_eject,
diff --git a/block/raw-win32.c b/block/raw-win32.c
index 62edb1a6cc..9b813d99ae 100644
--- a/block/raw-win32.c
+++ b/block/raw-win32.c
@@ -27,7 +27,7 @@
 #include "qemu/timer.h"
 #include "block/block_int.h"
 #include "qemu/module.h"
-#include "raw-aio.h"
+#include "block/raw-aio.h"
 #include "trace.h"
 #include "block/thread-pool.h"
 #include "qemu/iov.h"
diff --git a/block/win32-aio.c b/block/win32-aio.c
index 2d509a9a7b..95e3ab1541 100644
--- a/block/win32-aio.c
+++ b/block/win32-aio.c
@@ -27,7 +27,7 @@
 #include "block/block_int.h"
 #include "qemu/module.h"
 #include "block/aio.h"
-#include "raw-aio.h"
+#include "block/raw-aio.h"
 #include "qemu/event_notifier.h"
 #include "qemu/iov.h"
 #include <windows.h>
diff --git a/include/block/aio.h b/include/block/aio.h
index 88a64eeb3c..afd72a7516 100644
--- a/include/block/aio.h
+++ b/include/block/aio.h
@@ -47,6 +47,9 @@ typedef struct AioHandler AioHandler;
 typedef void QEMUBHFunc(void *opaque);
 typedef void IOHandler(void *opaque);
 
+struct ThreadPool;
+struct LinuxAioState;
+
 struct AioContext {
     GSource source;
 
@@ -119,6 +122,13 @@ struct AioContext {
     /* Thread pool for performing work and receiving completion callbacks */
     struct ThreadPool *thread_pool;
 
+#ifdef CONFIG_LINUX_AIO
+    /* State for native Linux AIO.  Uses aio_context_acquire/release for
+     * locking.
+     */
+    struct LinuxAioState *linux_aio;
+#endif
+
     /* TimerLists for calling timers - one per clock type */
     QEMUTimerListGroup tlg;
 
@@ -335,6 +345,9 @@ GSource *aio_get_g_source(AioContext *ctx);
 /* Return the ThreadPool bound to this AioContext */
 struct ThreadPool *aio_get_thread_pool(AioContext *ctx);
 
+/* Return the LinuxAioState bound to this AioContext */
+struct LinuxAioState *aio_get_linux_aio(AioContext *ctx);
+
 /**
  * aio_timer_new:
  * @ctx: the aio context
diff --git a/block/raw-aio.h b/include/block/raw-aio.h
index a4cdbbf1b7..a4cdbbf1b7 100644
--- a/block/raw-aio.h
+++ b/include/block/raw-aio.h