[Qemu-devel] [RFC 03/13] block: add BlockDevOps->drain_threads_cb()

[Stefan Hajnoczi <stefanha@redhat.com>]

There are times when the QEMU main loop wishes to drain I/O requests.
Up until now bdrv_drain_all() meant that no new guest I/O will be
processed until the next event loop iteration.

This is no longer true with dataplane since it runs outside the QEMU
global mutex.  The BlockDevOps->drain_threads_cb() interface allows the
device model to drain and stop threads.  Once draining completes, the
QEMU main loop can be sure that no further I/O will take place until
next main loop iteration.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 block.c               | 6 ++++++
 include/block/block.h | 9 +++++++++
 2 files changed, 15 insertions(+)

diff --git a/block.c b/block.c
index 1c9df27..a3323b2 100644
--- a/block.c
+++ b/block.c
@@ -1430,6 +1430,12 @@ void bdrv_drain_all(void)
     BlockDriverState *bs;
     bool busy;
 
+    QTAILQ_FOREACH(bs, &bdrv_states, list) {
+        if (bs->dev_ops && bs->dev_ops->drain_threads_cb) {
+            bs->dev_ops->drain_threads_cb(bs->dev_opaque);
+        }
+    }
+
     do {
         busy = qemu_aio_wait();
 
diff --git a/include/block/block.h b/include/block/block.h
index 2307f67..85147bb 100644
--- a/include/block/block.h
+++ b/include/block/block.h
@@ -60,6 +60,15 @@ typedef struct BlockDevOps {
      * Runs when the size changed (e.g. monitor command block_resize)
      */
     void (*resize_cb)(void *opaque);
+    /*
+     * Notifies the device model to drain any emulation threads
+     *
+     * Upon return, there must be no new or pending requests outside the QEMU
+     * main loop.  The device model may restart emulation threads after this
+     * main loop iteration, for example in a hardware register handler
+     * function.
+     */
+    void (*drain_threads_cb)(void *opaque);
 } BlockDevOps;
 
 #define BDRV_O_RDWR        0x0002
-- 
1.8.1.4