diff --git a/block/qcow2-cache.c b/block/qcow2-cache.c
index 340a6f2b26..710d4b1828 100644
--- a/block/qcow2-cache.c
+++ b/block/qcow2-cache.c
@@ -25,6 +25,7 @@
 #include "block_int.h"
 #include "qemu-common.h"
 #include "qcow2.h"
+#include "trace.h"
 
 typedef struct Qcow2CachedTable {
     void*   table;
@@ -100,6 +101,9 @@ static int qcow2_cache_entry_flush(BlockDriverState *bs, Qcow2Cache *c, int i)
         return 0;
     }
 
+    trace_qcow2_cache_entry_flush(qemu_coroutine_self(),
+                                  c == s->l2_table_cache, i);
+
     if (c->depends) {
         ret = qcow2_cache_flush_dependency(bs, c);
     } else if (c->depends_on_flush) {
@@ -132,10 +136,13 @@ static int qcow2_cache_entry_flush(BlockDriverState *bs, Qcow2Cache *c, int i)
 
 int qcow2_cache_flush(BlockDriverState *bs, Qcow2Cache *c)
 {
+    BDRVQcowState *s = bs->opaque;
     int result = 0;
     int ret;
     int i;
 
+    trace_qcow2_cache_flush(qemu_coroutine_self(), c == s->l2_table_cache);
+
     for (i = 0; i < c->size; i++) {
         ret = qcow2_cache_entry_flush(bs, c, i);
         if (ret < 0 && result != -ENOSPC) {
@@ -218,6 +225,9 @@ static int qcow2_cache_do_get(BlockDriverState *bs, Qcow2Cache *c,
     int i;
     int ret;
 
+    trace_qcow2_cache_get(qemu_coroutine_self(), c == s->l2_table_cache,
+                          offset, read_from_disk);
+
     /* Check if the table is already cached */
     for (i = 0; i < c->size; i++) {
         if (c->entries[i].offset == offset) {
@@ -227,6 +237,8 @@ static int qcow2_cache_do_get(BlockDriverState *bs, Qcow2Cache *c,
 
     /* If not, write a table back and replace it */
     i = qcow2_cache_find_entry_to_replace(c);
+    trace_qcow2_cache_get_replace_entry(qemu_coroutine_self(),
+                                        c == s->l2_table_cache, i);
     if (i < 0) {
         return i;
     }
@@ -236,6 +248,8 @@ static int qcow2_cache_do_get(BlockDriverState *bs, Qcow2Cache *c,
         return ret;
     }
 
+    trace_qcow2_cache_get_read(qemu_coroutine_self(),
+                               c == s->l2_table_cache, i);
     c->entries[i].offset = 0;
     if (read_from_disk) {
         if (c == s->l2_table_cache) {
@@ -258,6 +272,10 @@ found:
     c->entries[i].cache_hits++;
     c->entries[i].ref++;
     *table = c->entries[i].table;
+
+    trace_qcow2_cache_get_done(qemu_coroutine_self(),
+                               c == s->l2_table_cache, i);
+
     return 0;
 }
 
diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c
index 07a2e936fd..a791bbe594 100644
--- a/block/qcow2-cluster.c
+++ b/block/qcow2-cluster.c
@@ -27,6 +27,7 @@
 #include "qemu-common.h"
 #include "block_int.h"
 #include "block/qcow2.h"
+#include "trace.h"
 
 int qcow2_grow_l1_table(BlockDriverState *bs, int min_size, bool exact_size)
 {
@@ -170,6 +171,8 @@ static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
 
     old_l2_offset = s->l1_table[l1_index];
 
+    trace_qcow2_l2_allocate(bs, l1_index);
+
     /* allocate a new l2 entry */
 
     l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
@@ -184,6 +187,7 @@ static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
 
     /* allocate a new entry in the l2 cache */
 
+    trace_qcow2_l2_allocate_get_empty(bs, l1_index);
     ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
     if (ret < 0) {
         return ret;
@@ -216,6 +220,7 @@ static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
     /* write the l2 table to the file */
     BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
 
+    trace_qcow2_l2_allocate_write_l2(bs, l1_index);
     qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
     ret = qcow2_cache_flush(bs, s->l2_table_cache);
     if (ret < 0) {
@@ -223,6 +228,7 @@ static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
     }
 
     /* update the L1 entry */
+    trace_qcow2_l2_allocate_write_l1(bs, l1_index);
     s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
     ret = write_l1_entry(bs, l1_index);
     if (ret < 0) {
@@ -230,9 +236,11 @@ static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
     }
 
     *table = l2_table;
+    trace_qcow2_l2_allocate_done(bs, l1_index, 0);
     return 0;
 
 fail:
+    trace_qcow2_l2_allocate_done(bs, l1_index, ret);
     qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
     s->l1_table[l1_index] = old_l2_offset;
     return ret;
@@ -584,6 +592,8 @@ int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
     uint64_t cluster_offset = m->cluster_offset;
     bool cow = false;
 
+    trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
+
     if (m->nb_clusters == 0)
         return 0;
 
@@ -695,6 +705,9 @@ int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
     unsigned int nb_clusters, i = 0;
     QCowL2Meta *old_alloc;
 
+    trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset,
+                                      n_start, n_end);
+
     ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
     if (ret < 0) {
         return ret;
@@ -793,7 +806,7 @@ again:
     QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
 
     /* allocate a new cluster */
-
+    trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
     cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size);
     if (cluster_offset < 0) {
         ret = cluster_offset;
diff --git a/block/qcow2.c b/block/qcow2.c
index eb5ea485d9..6f53ec6b7d 100644
--- a/block/qcow2.c
+++ b/block/qcow2.c
@@ -29,6 +29,7 @@
 #include "block/qcow2.h"
 #include "qemu-error.h"
 #include "qerror.h"
+#include "trace.h"
 
 /*
   Differences with QCOW:
@@ -569,6 +570,9 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
         .nb_clusters = 0,
     };
 
+    trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
+                                 remaining_sectors);
+
     qemu_co_queue_init(&l2meta.dependent_requests);
 
     qemu_iovec_init(&hd_qiov, qiov->niov);
@@ -579,6 +583,7 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
 
     while (remaining_sectors != 0) {
 
+        trace_qcow2_writev_start_part(qemu_coroutine_self());
         index_in_cluster = sector_num & (s->cluster_sectors - 1);
         n_end = index_in_cluster + remaining_sectors;
         if (s->crypt_method &&
@@ -619,6 +624,8 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
 
         BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
         qemu_co_mutex_unlock(&s->lock);
+        trace_qcow2_writev_data(qemu_coroutine_self(),
+                                (cluster_offset >> 9) + index_in_cluster);
         ret = bdrv_co_writev(bs->file,
                              (cluster_offset >> 9) + index_in_cluster,
                              cur_nr_sectors, &hd_qiov);
@@ -637,6 +644,7 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
         remaining_sectors -= cur_nr_sectors;
         sector_num += cur_nr_sectors;
         bytes_done += cur_nr_sectors * 512;
+        trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_nr_sectors);
     }
     ret = 0;
 
@@ -647,6 +655,7 @@ fail:
 
     qemu_iovec_destroy(&hd_qiov);
     qemu_vfree(cluster_data);
+    trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
 
     return ret;
 }
diff --git a/trace-events b/trace-events
index c5d0f0f547..d818ff1395 100644
--- a/trace-events
+++ b/trace-events
@@ -312,6 +312,30 @@ scsi_request_sense(int target, int lun, int tag) "target %d lun %d tag %d"
 # vl.c
 vm_state_notify(int running, int reason) "running %d reason %d"
 
+# block/qcow2.c
+qcow2_writev_start_req(void *co, int64_t sector, int nb_sectors) "co %p sector %" PRIx64 " nb_sectors %d"
+qcow2_writev_done_req(void *co, int ret) "co %p ret %d"
+qcow2_writev_start_part(void *co) "co %p"
+qcow2_writev_done_part(void *co, int cur_nr_sectors) "co %p cur_nr_sectors %d"
+qcow2_writev_data(void *co, uint64_t offset) "co %p offset %" PRIx64
+
+qcow2_alloc_clusters_offset(void *co, uint64_t offset, int n_start, int n_end) "co %p offet %" PRIx64 " n_start %d n_end %d"
+qcow2_cluster_alloc_phys(void *co) "co %p"
+qcow2_cluster_link_l2(void *co, int nb_clusters) "co %p nb_clusters %d"
+
+qcow2_l2_allocate(void *bs, int l1_index) "bs %p l1_index %d"
+qcow2_l2_allocate_get_empty(void *bs, int l1_index) "bs %p l1_index %d"
+qcow2_l2_allocate_write_l2(void *bs, int l1_index) "bs %p l1_index %d"
+qcow2_l2_allocate_write_l1(void *bs, int l1_index) "bs %p l1_index %d"
+qcow2_l2_allocate_done(void *bs, int l1_index, int ret) "bs %p l1_index %d ret %d"
+
+qcow2_cache_get(void *co, int c, uint64_t offset, bool read_from_disk) "co %p is_l2_cache %d offset %" PRIx64 " read_from_disk %d"
+qcow2_cache_get_replace_entry(void *co, int c, int i) "co %p is_l2_cache %d index %d"
+qcow2_cache_get_read(void *co, int c, int i) "co %p is_l2_cache %d index %d"
+qcow2_cache_get_done(void *co, int c, int i) "co %p is_l2_cache %d index %d"
+qcow2_cache_flush(void *co, int c) "co %p is_l2_cache %d"
+qcow2_cache_entry_flush(void *co, int c, int i) "co %p is_l2_cache %d index %d"
+
 # block/qed-l2-cache.c
 qed_alloc_l2_cache_entry(void *l2_cache, void *entry) "l2_cache %p entry %p"
 qed_unref_l2_cache_entry(void *entry, int ref) "entry %p ref %d"