Skip to contentSkip to conversation

introduce QEMU_AUTO_VFREE

PATCH Superseded Vladimir Sementsov-Ogievskiy 3 participants 7 messages Jun 19, 2021 – Jun 28, 2021 Block layer

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover introduce QEMU_AUTO_VFREE

Hi all!

There is a good movement to use g_autofree macro, that helps to automatically call g_free on exit from code block.

We lack similar possibility for qemu_memalign() functions family. Let's add, it seems rather simple with help of "cleanup" attribute.

I'll update more places with a follow-up if this is accepted.

Vladimir Sementsov-Ogievskiy (2):
introduce QEMU_AUTO_VFREE
block/commit: use QEMU_AUTO_VFREE

 include/qemu/osdep.h |  2 ++
 block/commit.c       | 25 +++++++++----------------
 2 files changed, 11 insertions(+), 16 deletions(-)
Signature · 1 line
2.29.2
#1 lore raw

PATCH 1/2 introduce QEMU_AUTO_VFREE

Introduce a convenient macro, that works for qemu_memalign() like g_autofree works with g_malloc.

---
 include/qemu/osdep.h | 2 ++
 1 file changed, 2 insertions(+)
include/qemu/osdep.h +2 -0
index 236a045671..844658a764 100644
@@ -366,6 +366,8 @@ void *qemu_anon_ram_alloc(size_t size, uint64_t *align, bool shared);
 void qemu_vfree(void *ptr);
 void qemu_anon_ram_free(void *ptr, size_t size);
 
+#define QEMU_AUTO_VFREE __attribute__((cleanup(qemu_vfree)))
+
 #define QEMU_MADV_INVALID -1
 
 #if defined(CONFIG_MADVISE)
Signature · 1 line
2.29.2
#2 lore raw

PATCH 2/2 block/commit: use QEMU_AUTO_VFREE

In reply to Vladimir Sementsov-Ogievskiy
---
 block/commit.c | 25 +++++++++----------------
 1 file changed, 9 insertions(+), 16 deletions(-)
block/commit.c +9 -16
index b7f0c7c061..42792b4556 100644
@@ -119,24 +119,24 @@ static int coroutine_fn commit_run(Job *job, Error **errp)
     uint64_t delay_ns = 0;
     int ret = 0;
     int64_t n = 0; /* bytes */
-    void *buf = NULL;
+    QEMU_AUTO_VFREE void *buf = NULL;
     int64_t len, base_len;
 
-    ret = len = blk_getlength(s->top);
+    len = blk_getlength(s->top);
     if (len < 0) {
-        goto out;
+        return len;
     }
     job_progress_set_remaining(&s->common.job, len);
 
-    ret = base_len = blk_getlength(s->base);
+    base_len = blk_getlength(s->base);
     if (base_len < 0) {
-        goto out;
+        return base_len;
     }
 
     if (base_len < len) {
         ret = blk_truncate(s->base, len, false, PREALLOC_MODE_OFF, 0, NULL);
         if (ret) {
-            goto out;
+            return ret;
         }
     }
 
@@ -174,7 +174,7 @@ static int coroutine_fn commit_run(Job *job, Error **errp)
                 block_job_error_action(&s->common, s->on_error,
                                        error_in_source, -ret);
             if (action == BLOCK_ERROR_ACTION_REPORT) {
-                goto out;
+                return ret;
             } else {
                 n = 0;
                 continue;
@@ -190,12 +190,7 @@ static int coroutine_fn commit_run(Job *job, Error **errp)
         }
     }
 
-    ret = 0;
-
-out:
-    qemu_vfree(buf);
-
-    return ret;
+    return 0;
 }
 
 static const BlockJobDriver commit_job_driver = {
@@ -435,7 +430,7 @@ int bdrv_commit(BlockDriverState *bs)
     int ro;
     int64_t n;
     int ret = 0;
-    uint8_t *buf = NULL;
+    QEMU_AUTO_VFREE uint8_t *buf = NULL;
     Error *local_err = NULL;
 
     if (!drv)
@@ -556,8 +551,6 @@ int bdrv_commit(BlockDriverState *bs)
 
     ret = 0;
 ro_cleanup:
-    qemu_vfree(buf);
-
     blk_unref(backing);
     if (bdrv_cow_bs(bs) != backing_file_bs) {
         bdrv_set_backing_hd(bs, backing_file_bs, &error_abort);
Signature · 1 line
2.29.2
Reviewed-by on the v1 series
#3 lore raw
In reply to Vladimir Sementsov-Ogievskiy
Quoted 10 lines · Vladimir Sementsov-Ogievskiy

Hi all!

There is a good movement to use g_autofree macro, that helps to automatically call g_free on exit from code block.

We lack similar possibility for qemu_memalign() functions family. Let's add, it seems rather simple with help of "cleanup" attribute.

I'll update more places with a follow-up if this is accepted.

Quoted 11 lines

Vladimir Sementsov-Ogievskiy (2):
introduce QEMU_AUTO_VFREE
block/commit: use QEMU_AUTO_VFREE

 include/qemu/osdep.h |  2 ++
 block/commit.c       | 25 +++++++++----------------
 2 files changed, 11 insertions(+), 16 deletions(-)
Signature · 1 line
2.29.2
#4 lore raw
In reply to Vladimir Sementsov-Ogievskiy

Am 19.06.2021 um 16:21 hat Vladimir Sementsov-Ogievskiy geschrieben:

Quoted 9 lines

Hi all!

There is a good movement to use g_autofree macro, that helps to automatically call g_free on exit from code block.

We lack similar possibility for qemu_memalign() functions family. Let's add, it seems rather simple with help of "cleanup" attribute.

I'll update more places with a follow-up if this is accepted.

Good idea. Thanks, applied to the block branch.

Kevin

#5 lore raw

Am 23.06.2021 um 12:34 hat Kevin Wolf geschrieben:

Quoted 6 lines

Am 19.06.2021 um 16:21 hat Vladimir Sementsov-Ogievskiy geschrieben:

Quoted 5 lines

Hi all!

There is a good movement to use g_autofree macro, that helps to automatically call g_free on exit from code block.

Quoted 6 lines
Quoted 4 lines

We lack similar possibility for qemu_memalign() functions family. Let's add, it seems rather simple with help of "cleanup" attribute.

I'll update more places with a follow-up if this is accepted.

Good idea. Thanks, applied to the block branch.

I'm afraid I have to unqueue this:

Failures: 040 141 161 191 249
Failed 5 of 5 iotests

Maybe semantic merge conflicts that I didn't notice while applying or rebasing.

Kevin

#6 lore raw
Quoted 14 lines · Kevin Wolf

Am 23.06.2021 um 12:34 hat Kevin Wolf geschrieben:

Quoted 6 lines

Am 19.06.2021 um 16:21 hat Vladimir Sementsov-Ogievskiy geschrieben:

Quoted 5 lines

Hi all!

There is a good movement to use g_autofree macro, that helps to automatically call g_free on exit from code block.

Quoted 6 lines
Quoted 4 lines

We lack similar possibility for qemu_memalign() functions family. Let's add, it seems rather simple with help of "cleanup" attribute.

I'll update more places with a follow-up if this is accepted.

Good idea. Thanks, applied to the block branch.

Quoted 8 lines

I'm afraid I have to unqueue this:

Failures: 040 141 161 191 249
Failed 5 of 5 iotests

Maybe semantic merge conflicts that I didn't notice while applying or rebasing.

No, looks like I just made a wrong thing and didn't run tests (hmm, or I was "lucky" and it didn't crash previous time for some reason). I remember, I checked that qemu_vfree is called. But I didn't check the pointer. Now in gdb I see that cleanup function gets pointer to pointer. So we can't reuse qemu_vfree, but need additional function. I'll resend. Sorry for the mess :(

Signature · 2 lines
Best regards,
Vladimir