Blame SOURCES/kvm-block-Use-a-single-global-AioWait.patch

383d26
From 54d30086f066f1094871c4886f3a6dee51263d76 Mon Sep 17 00:00:00 2001
383d26
From: Kevin Wolf <kwolf@redhat.com>
383d26
Date: Fri, 21 Sep 2018 12:46:29 +0200
383d26
Subject: [PATCH 2/3] block: Use a single global AioWait
383d26
383d26
RH-Author: Kevin Wolf <kwolf@redhat.com>
383d26
Message-id: <20180921124630.29036-3-kwolf@redhat.com>
383d26
Patchwork-id: 82231
383d26
O-Subject: [RHV-7.6 qemu-kvm-rhev PATCH 2/3] block: Use a single global AioWait
383d26
Bugzilla: 1618584
383d26
RH-Acked-by: Jeffrey Cody <jcody@redhat.com>
383d26
RH-Acked-by: Paolo Bonzini <pbonzini@redhat.com>
383d26
RH-Acked-by: John Snow <jsnow@redhat.com>
383d26
RH-Acked-by: Eric Blake <eblake@redhat.com>
383d26
383d26
When draining a block node, we recurse to its parent and for subtree
383d26
drains also to its children. A single AIO_WAIT_WHILE() is then used to
383d26
wait for bdrv_drain_poll() to become true, which depends on all of the
383d26
nodes we recursed to. However, if the respective child or parent becomes
383d26
quiescent and calls bdrv_wakeup(), only the AioWait of the child/parent
383d26
is checked, while AIO_WAIT_WHILE() depends on the AioWait of the
383d26
original node.
383d26
383d26
Fix this by using a single AioWait for all callers of AIO_WAIT_WHILE().
383d26
383d26
This may mean that the draining thread gets a few more unnecessary
383d26
wakeups because an unrelated operation got completed, but we already
383d26
wake it up when something _could_ have changed rather than only if it
383d26
has certainly changed.
383d26
383d26
Apart from that, drain is a slow path anyway. In theory it would be
383d26
possible to use wakeups more selectively and still correctly, but the
383d26
gains are likely not worth the additional complexity. In fact, this
383d26
patch is a nice simplification for some places in the code.
383d26
383d26
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
383d26
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
383d26
---
383d26
 block.c                   |  5 -----
383d26
 block/block-backend.c     | 11 ++++-------
383d26
 block/io.c                |  7 ++-----
383d26
 blockjob.c                | 13 +------------
383d26
 include/block/aio-wait.h  | 11 +++++------
383d26
 include/block/block.h     |  6 +-----
383d26
 include/block/block_int.h |  3 ---
383d26
 include/block/blockjob.h  | 10 ----------
383d26
 job.c                     |  3 +--
383d26
 util/aio-wait.c           | 11 ++++++-----
383d26
 10 files changed, 20 insertions(+), 60 deletions(-)
383d26
383d26
diff --git a/block.c b/block.c
383d26
index e89b5e3..fbd569c 100644
383d26
--- a/block.c
383d26
+++ b/block.c
383d26
@@ -4847,11 +4847,6 @@ AioContext *bdrv_get_aio_context(BlockDriverState *bs)
383d26
     return bs ? bs->aio_context : qemu_get_aio_context();
383d26
 }
383d26
 
383d26
-AioWait *bdrv_get_aio_wait(BlockDriverState *bs)
383d26
-{
383d26
-    return bs ? &bs->wait : NULL;
383d26
-}
383d26
-
383d26
 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
383d26
 {
383d26
     aio_co_enter(bdrv_get_aio_context(bs), co);
383d26
diff --git a/block/block-backend.c b/block/block-backend.c
383d26
index 466bc27..36922d1 100644
383d26
--- a/block/block-backend.c
383d26
+++ b/block/block-backend.c
383d26
@@ -88,7 +88,6 @@ struct BlockBackend {
383d26
      * Accessed with atomic ops.
383d26
      */
383d26
     unsigned int in_flight;
383d26
-    AioWait wait;
383d26
 };
383d26
 
383d26
 typedef struct BlockBackendAIOCB {
383d26
@@ -1300,7 +1299,7 @@ static void blk_inc_in_flight(BlockBackend *blk)
383d26
 static void blk_dec_in_flight(BlockBackend *blk)
383d26
 {
383d26
     atomic_dec(&blk->in_flight);
383d26
-    aio_wait_kick(&blk->wait);
383d26
+    aio_wait_kick();
383d26
 }
383d26
 
383d26
 static void error_callback_bh(void *opaque)
383d26
@@ -1609,9 +1608,8 @@ void blk_drain(BlockBackend *blk)
383d26
     }
383d26
 
383d26
     /* We may have -ENOMEDIUM completions in flight */
383d26
-    AIO_WAIT_WHILE(&blk->wait,
383d26
-            blk_get_aio_context(blk),
383d26
-            atomic_mb_read(&blk->in_flight) > 0);
383d26
+    AIO_WAIT_WHILE(blk_get_aio_context(blk),
383d26
+                   atomic_mb_read(&blk->in_flight) > 0);
383d26
 
383d26
     if (bs) {
383d26
         bdrv_drained_end(bs);
383d26
@@ -1630,8 +1628,7 @@ void blk_drain_all(void)
383d26
         aio_context_acquire(ctx);
383d26
 
383d26
         /* We may have -ENOMEDIUM completions in flight */
383d26
-        AIO_WAIT_WHILE(&blk->wait, ctx,
383d26
-                atomic_mb_read(&blk->in_flight) > 0);
383d26
+        AIO_WAIT_WHILE(ctx, atomic_mb_read(&blk->in_flight) > 0);
383d26
 
383d26
         aio_context_release(ctx);
383d26
     }
383d26
diff --git a/block/io.c b/block/io.c
383d26
index 3313958..7a99f7b 100644
383d26
--- a/block/io.c
383d26
+++ b/block/io.c
383d26
@@ -38,8 +38,6 @@
383d26
 /* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */
383d26
 #define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
383d26
 
383d26
-static AioWait drain_all_aio_wait;
383d26
-
383d26
 static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
383d26
     int64_t offset, int bytes, BdrvRequestFlags flags);
383d26
 
383d26
@@ -555,7 +553,7 @@ void bdrv_drain_all_begin(void)
383d26
     }
383d26
 
383d26
     /* Now poll the in-flight requests */
383d26
-    AIO_WAIT_WHILE(&drain_all_aio_wait, NULL, bdrv_drain_all_poll());
383d26
+    AIO_WAIT_WHILE(NULL, bdrv_drain_all_poll());
383d26
 
383d26
     while ((bs = bdrv_next_all_states(bs))) {
383d26
         bdrv_drain_assert_idle(bs);
383d26
@@ -709,8 +707,7 @@ void bdrv_inc_in_flight(BlockDriverState *bs)
383d26
 
383d26
 void bdrv_wakeup(BlockDriverState *bs)
383d26
 {
383d26
-    aio_wait_kick(bdrv_get_aio_wait(bs));
383d26
-    aio_wait_kick(&drain_all_aio_wait);
383d26
+    aio_wait_kick();
383d26
 }
383d26
 
383d26
 void bdrv_dec_in_flight(BlockDriverState *bs)
383d26
diff --git a/blockjob.c b/blockjob.c
383d26
index 617d86f..06f2429 100644
383d26
--- a/blockjob.c
383d26
+++ b/blockjob.c
383d26
@@ -221,20 +221,9 @@ int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
383d26
     return 0;
383d26
 }
383d26
 
383d26
-void block_job_wakeup_all_bdrv(BlockJob *job)
383d26
-{
383d26
-    GSList *l;
383d26
-
383d26
-    for (l = job->nodes; l; l = l->next) {
383d26
-        BdrvChild *c = l->data;
383d26
-        bdrv_wakeup(c->bs);
383d26
-    }
383d26
-}
383d26
-
383d26
 static void block_job_on_idle(Notifier *n, void *opaque)
383d26
 {
383d26
-    BlockJob *job = opaque;
383d26
-    block_job_wakeup_all_bdrv(job);
383d26
+    aio_wait_kick();
383d26
 }
383d26
 
383d26
 bool block_job_is_internal(BlockJob *job)
383d26
diff --git a/include/block/aio-wait.h b/include/block/aio-wait.h
383d26
index 600fad1..46f86f9 100644
383d26
--- a/include/block/aio-wait.h
383d26
+++ b/include/block/aio-wait.h
383d26
@@ -54,9 +54,10 @@ typedef struct {
383d26
     unsigned num_waiters;
383d26
 } AioWait;
383d26
 
383d26
+extern AioWait global_aio_wait;
383d26
+
383d26
 /**
383d26
  * AIO_WAIT_WHILE:
383d26
- * @wait: the aio wait object
383d26
  * @ctx: the aio context, or NULL if multiple aio contexts (for which the
383d26
  *       caller does not hold a lock) are involved in the polling condition.
383d26
  * @cond: wait while this conditional expression is true
383d26
@@ -72,9 +73,9 @@ typedef struct {
383d26
  * wait on conditions between two IOThreads since that could lead to deadlock,
383d26
  * go via the main loop instead.
383d26
  */
383d26
-#define AIO_WAIT_WHILE(wait, ctx, cond) ({                         \
383d26
+#define AIO_WAIT_WHILE(ctx, cond) ({                               \
383d26
     bool waited_ = false;                                          \
383d26
-    AioWait *wait_ = (wait);                                       \
383d26
+    AioWait *wait_ = &global_aio_wait;                             \
383d26
     AioContext *ctx_ = (ctx);                                      \
383d26
     /* Increment wait_->num_waiters before evaluating cond. */     \
383d26
     atomic_inc(&wait_->num_waiters);                               \
383d26
@@ -102,14 +103,12 @@ typedef struct {
383d26
 
383d26
 /**
383d26
  * aio_wait_kick:
383d26
- * @wait: the aio wait object that should re-evaluate its condition
383d26
- *
383d26
  * Wake up the main thread if it is waiting on AIO_WAIT_WHILE().  During
383d26
  * synchronous operations performed in an IOThread, the main thread lets the
383d26
  * IOThread's event loop run, waiting for the operation to complete.  A
383d26
  * aio_wait_kick() call will wake up the main thread.
383d26
  */
383d26
-void aio_wait_kick(AioWait *wait);
383d26
+void aio_wait_kick(void);
383d26
 
383d26
 /**
383d26
  * aio_wait_bh_oneshot:
383d26
diff --git a/include/block/block.h b/include/block/block.h
383d26
index 356712c..8e78daf 100644
383d26
--- a/include/block/block.h
383d26
+++ b/include/block/block.h
383d26
@@ -406,13 +406,9 @@ void bdrv_drain_all_begin(void);
383d26
 void bdrv_drain_all_end(void);
383d26
 void bdrv_drain_all(void);
383d26
 
383d26
-/* Returns NULL when bs == NULL */
383d26
-AioWait *bdrv_get_aio_wait(BlockDriverState *bs);
383d26
-
383d26
 #define BDRV_POLL_WHILE(bs, cond) ({                       \
383d26
     BlockDriverState *bs_ = (bs);                          \
383d26
-    AIO_WAIT_WHILE(bdrv_get_aio_wait(bs_),                 \
383d26
-                   bdrv_get_aio_context(bs_),              \
383d26
+    AIO_WAIT_WHILE(bdrv_get_aio_context(bs_),              \
383d26
                    cond); })
383d26
 
383d26
 int bdrv_pdiscard(BlockDriverState *bs, int64_t offset, int bytes);
383d26
diff --git a/include/block/block_int.h b/include/block/block_int.h
383d26
index b7806e3..ff923b7 100644
383d26
--- a/include/block/block_int.h
383d26
+++ b/include/block/block_int.h
383d26
@@ -782,9 +782,6 @@ struct BlockDriverState {
383d26
     unsigned int in_flight;
383d26
     unsigned int serialising_in_flight;
383d26
 
383d26
-    /* Kicked to signal main loop when a request completes. */
383d26
-    AioWait wait;
383d26
-
383d26
     /* counter for nested bdrv_io_plug.
383d26
      * Accessed with atomic ops.
383d26
     */
383d26
diff --git a/include/block/blockjob.h b/include/block/blockjob.h
383d26
index 2290bbb..ede0bd8 100644
383d26
--- a/include/block/blockjob.h
383d26
+++ b/include/block/blockjob.h
383d26
@@ -122,16 +122,6 @@ int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
383d26
 void block_job_remove_all_bdrv(BlockJob *job);
383d26
 
383d26
 /**
383d26
- * block_job_wakeup_all_bdrv:
383d26
- * @job: The block job
383d26
- *
383d26
- * Calls bdrv_wakeup() for all BlockDriverStates that have been added to the
383d26
- * job. This function is to be called whenever child_job_drained_poll() would
383d26
- * go from true to false to notify waiting drain requests.
383d26
- */
383d26
-void block_job_wakeup_all_bdrv(BlockJob *job);
383d26
-
383d26
-/**
383d26
  * block_job_set_speed:
383d26
  * @job: The job to set the speed for.
383d26
  * @speed: The new value
383d26
diff --git a/job.c b/job.c
383d26
index 5b53e43..3a7db59 100644
383d26
--- a/job.c
383d26
+++ b/job.c
383d26
@@ -973,7 +973,6 @@ void job_complete(Job *job, Error **errp)
383d26
 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
383d26
 {
383d26
     Error *local_err = NULL;
383d26
-    AioWait dummy_wait = {};
383d26
     int ret;
383d26
 
383d26
     job_ref(job);
383d26
@@ -987,7 +986,7 @@ int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
383d26
         return -EBUSY;
383d26
     }
383d26
 
383d26
-    AIO_WAIT_WHILE(&dummy_wait, job->aio_context,
383d26
+    AIO_WAIT_WHILE(job->aio_context,
383d26
                    (job_drain(job), !job_is_completed(job)));
383d26
 
383d26
     ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;
383d26
diff --git a/util/aio-wait.c b/util/aio-wait.c
383d26
index b8a8f86..b487749 100644
383d26
--- a/util/aio-wait.c
383d26
+++ b/util/aio-wait.c
383d26
@@ -26,21 +26,22 @@
383d26
 #include "qemu/main-loop.h"
383d26
 #include "block/aio-wait.h"
383d26
 
383d26
+AioWait global_aio_wait;
383d26
+
383d26
 static void dummy_bh_cb(void *opaque)
383d26
 {
383d26
     /* The point is to make AIO_WAIT_WHILE()'s aio_poll() return */
383d26
 }
383d26
 
383d26
-void aio_wait_kick(AioWait *wait)
383d26
+void aio_wait_kick(void)
383d26
 {
383d26
     /* The barrier (or an atomic op) is in the caller.  */
383d26
-    if (atomic_read(&wait->num_waiters)) {
383d26
+    if (atomic_read(&global_aio_wait.num_waiters)) {
383d26
         aio_bh_schedule_oneshot(qemu_get_aio_context(), dummy_bh_cb, NULL);
383d26
     }
383d26
 }
383d26
 
383d26
 typedef struct {
383d26
-    AioWait wait;
383d26
     bool done;
383d26
     QEMUBHFunc *cb;
383d26
     void *opaque;
383d26
@@ -54,7 +55,7 @@ static void aio_wait_bh(void *opaque)
383d26
     data->cb(data->opaque);
383d26
 
383d26
     data->done = true;
383d26
-    aio_wait_kick(&data->wait);
383d26
+    aio_wait_kick();
383d26
 }
383d26
 
383d26
 void aio_wait_bh_oneshot(AioContext *ctx, QEMUBHFunc *cb, void *opaque)
383d26
@@ -67,5 +68,5 @@ void aio_wait_bh_oneshot(AioContext *ctx, QEMUBHFunc *cb, void *opaque)
383d26
     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
383d26
 
383d26
     aio_bh_schedule_oneshot(ctx, aio_wait_bh, &data);
383d26
-    AIO_WAIT_WHILE(&data.wait, ctx, !data.done);
383d26
+    AIO_WAIT_WHILE(ctx, !data.done);
383d26
 }
383d26
-- 
383d26
1.8.3.1
383d26