diff --git a/block/nvme.c b/block/nvme.c
index 7ed5f570bc..b8262ebfd9 100644
--- a/block/nvme.c
+++ b/block/nvme.c
@@ -1176,25 +1176,35 @@ fail:
 
 typedef struct {
     Coroutine *co;
+    bool skip_yield;
     int ret;
-    AioContext *ctx;
 } NVMeCoData;
 
-static void nvme_rw_cb_bh(void *opaque)
-{
-    NVMeCoData *data = opaque;
-    qemu_coroutine_enter(data->co);
-}
-
 static void nvme_rw_cb(void *opaque, int ret)
 {
     NVMeCoData *data = opaque;
+
     data->ret = ret;
-    if (!data->co) {
-        /* The rw coroutine hasn't yielded, don't try to enter. */
-        return;
+
+    if (data->co == qemu_coroutine_self()) {
+        /*
+         * Fast path: We are inside of the request coroutine (through
+         * nvme_submit_command, nvme_deferred_fn, nvme_process_completion).
+         * We can set data->skip_yield here to keep the coroutine from
+         * yielding, and then we don't need to schedule a BH to wake it.
+         */
+        data->skip_yield = true;
+    } else {
+        /*
+         * Safe to call: The case where we run in the request coroutine is
+         * handled above, so we must be independent of it; and without
+         * skip_yield set, the coroutine will yield.
+         * No need to release NVMeQueuePair.lock (we are called without it
+         * held).  (Note: If we enter the coroutine here, @data will
+         * probably be dangling once aio_co_wake() returns.)
+         */
+        aio_co_wake(data->co);
     }
-    replay_bh_schedule_oneshot_event(data->ctx, nvme_rw_cb_bh, data);
 }
 
 static coroutine_fn int nvme_co_prw_aligned(BlockDriverState *bs,
@@ -1218,7 +1228,7 @@ static coroutine_fn int nvme_co_prw_aligned(BlockDriverState *bs,
         .cdw12 = cpu_to_le32(cdw12),
     };
     NVMeCoData data = {
-        .ctx = bdrv_get_aio_context(bs),
+        .co = qemu_coroutine_self(),
         .ret = -EINPROGRESS,
     };
 
@@ -1235,9 +1245,7 @@ static coroutine_fn int nvme_co_prw_aligned(BlockDriverState *bs,
         return r;
     }
     nvme_submit_command(ioq, req, &cmd, nvme_rw_cb, &data);
-
-    data.co = qemu_coroutine_self();
-    while (data.ret == -EINPROGRESS) {
+    if (!data.skip_yield) {
         qemu_coroutine_yield();
     }
 
@@ -1333,7 +1341,7 @@ static coroutine_fn int nvme_co_flush(BlockDriverState *bs)
         .nsid = cpu_to_le32(s->nsid),
     };
     NVMeCoData data = {
-        .ctx = bdrv_get_aio_context(bs),
+        .co = qemu_coroutine_self(),
         .ret = -EINPROGRESS,
     };
 
@@ -1341,9 +1349,7 @@ static coroutine_fn int nvme_co_flush(BlockDriverState *bs)
     req = nvme_get_free_req(ioq);
     assert(req);
     nvme_submit_command(ioq, req, &cmd, nvme_rw_cb, &data);
-
-    data.co = qemu_coroutine_self();
-    if (data.ret == -EINPROGRESS) {
+    if (!data.skip_yield) {
         qemu_coroutine_yield();
     }
 
@@ -1384,7 +1390,7 @@ static coroutine_fn int nvme_co_pwrite_zeroes(BlockDriverState *bs,
     };
 
     NVMeCoData data = {
-        .ctx = bdrv_get_aio_context(bs),
+        .co = qemu_coroutine_self(),
         .ret = -EINPROGRESS,
     };
 
@@ -1404,9 +1410,7 @@ static coroutine_fn int nvme_co_pwrite_zeroes(BlockDriverState *bs,
     assert(req);
 
     nvme_submit_command(ioq, req, &cmd, nvme_rw_cb, &data);
-
-    data.co = qemu_coroutine_self();
-    while (data.ret == -EINPROGRESS) {
+    if (!data.skip_yield) {
         qemu_coroutine_yield();
     }
 
@@ -1434,7 +1438,7 @@ static int coroutine_fn nvme_co_pdiscard(BlockDriverState *bs,
     };
 
     NVMeCoData data = {
-        .ctx = bdrv_get_aio_context(bs),
+        .co = qemu_coroutine_self(),
         .ret = -EINPROGRESS,
     };
 
@@ -1479,9 +1483,7 @@ static int coroutine_fn nvme_co_pdiscard(BlockDriverState *bs,
     trace_nvme_dsm(s, offset, bytes);
 
     nvme_submit_command(ioq, req, &cmd, nvme_rw_cb, &data);
-
-    data.co = qemu_coroutine_self();
-    while (data.ret == -EINPROGRESS) {
+    if (!data.skip_yield) {
         qemu_coroutine_yield();
     }