[PATCH v7 0/3] Eliminate multifd flush

[PATCH v7 0/3] Eliminate multifd flush

[Juan Quintela]

Hi

In this v7:
- Rebased to last upstream
- Rename the capability to a property.  So we move all the problems
  that we have on last review dissaper because it is not a capability.

So now, it is works as expected.  Enabled for old machine types,
disabled for new machine types.  Users will only found it if they go through the migration properties.

Please review.

In this v6:
- Rename multifd-sync-after-each-section to
         multifd-flush-after-each-section
- Redo comments (thanks Markus)
- Redo how to comment capabilities that are enabled/disabled during
  development. (thanks Markus)

Please, review.

In this v5:
- Remove RAM Flags documentation (already on PULL request)
- rebase on top of PULL request.

Please review.

Based-on: <20230213025150.71537-1-quintela@redhat.com>
          Migration 20230213 patches

In this v4:
- Rebased on top of migration-20230209 PULL request
- Integrate two patches in that pull request
- Rebase
- Address Eric reviews.

Please review.

In this v3:
- update to latest upstream.
- fix checkpatch errors.

Please, review.

In this v2:
- update to latest upstream
- change 0, 1, 2 values to defines
- Add documentation for SAVE_VM_FLAGS
- Add missing qemu_fflush(), it made random hangs for migration test
  (only for tls, no clue why).

Please, review.

[v1]
Upstream multifd code synchronize all threads after each RAM section.  This is suboptimal.
Change it to only flush after we go trough all ram.

Preserve all semantics for old machine types.

Juan Quintela (3):
  multifd: Create property multifd-flush-after-each-section
  multifd: Protect multifd_send_sync_main() calls
  multifd: Only flush once each full round of memory

 migration/migration.h | 12 ++++++++++++
 hw/core/machine.c     |  1 +
 migration/migration.c |  9 +++++++++
 migration/ram.c       | 44 +++++++++++++++++++++++++++++++++++++------
 4 files changed, 60 insertions(+), 6 deletions(-)

-- 
2.39.2

[PATCH v7 1/3] multifd: Create property multifd-flush-after-each-section

[Juan Quintela]

We used to flush all channels at the end of each RAM section
sent.  That is not needed, so preparing to only flush after a full
iteration through all the RAM.

Default value of the property is false.  But we return "true" in
migrate_multifd_flush_after_each_section() until we implement the code
in following patches.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

---

Rename each-iteration to after-each-section
Rename multifd-sync-after-each-section to
       multifd-flush-after-each-section
---
 migration/migration.h | 13 +++++++++++++
 hw/core/machine.c     |  1 +
 migration/migration.c | 13 +++++++++++++
 3 files changed, 27 insertions(+)

diff --git a/migration/migration.h b/migration/migration.h
index 2da2f8a164..d0d89411e2 100644
--- a/migration/migration.h
+++ b/migration/migration.h
@@ -365,6 +365,18 @@ struct MigrationState {
      */
     bool decompress_error_check;
 
+    /*
+     * flush every channel after each section sent.
+     *
+     * This assures that we can't mix pages from one iteration through
+     * ram pages with pages for the following iteration.  We really
+     * only need to do this flush after we have go through all the
+     * dirty pages.  For historical reasons, we do that after each
+     * section.  This is suboptimal (we flush too many times).
+     * Default value is false.  Setting this property has no effect
+     * until the patch that removes this comment.  (since 8.1)
+     */
+    bool multifd_flush_after_each_section;
     /*
      * This decides the size of guest memory chunk that will be used
      * to track dirty bitmap clearing.  The size of memory chunk will
@@ -424,6 +436,7 @@ int migrate_multifd_channels(void);
 MultiFDCompression migrate_multifd_compression(void);
 int migrate_multifd_zlib_level(void);
 int migrate_multifd_zstd_level(void);
+bool migrate_multifd_flush_after_each_section(void);
 
 #ifdef CONFIG_LINUX
 bool migrate_use_zero_copy_send(void);
diff --git a/hw/core/machine.c b/hw/core/machine.c
index 45e3d24fdc..3bd92a25b5 100644
--- a/hw/core/machine.c
+++ b/hw/core/machine.c
@@ -56,6 +56,7 @@ const size_t hw_compat_7_1_len = G_N_ELEMENTS(hw_compat_7_1);
 GlobalProperty hw_compat_7_0[] = {
     { "arm-gicv3-common", "force-8-bit-prio", "on" },
     { "nvme-ns", "eui64-default", "on"},
+    { "migration", "multifd-flush-after-each-section", "on"},
 };
 const size_t hw_compat_7_0_len = G_N_ELEMENTS(hw_compat_7_0);
 
diff --git a/migration/migration.c b/migration/migration.c
index ae2025d9d8..ca773d00fc 100644
--- a/migration/migration.c
+++ b/migration/migration.c
@@ -2701,6 +2701,17 @@ bool migrate_use_multifd(void)
     return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
 }
 
+bool migrate_multifd_flush_after_each_section(void)
+{
+    MigrationState *s = migrate_get_current();
+
+    /*
+     * Until the patch that remove this comment, we always return that
+     * the property is enabled.
+     */
+    return true || s->multifd_flush_after_each_section;
+}
+
 bool migrate_pause_before_switchover(void)
 {
     MigrationState *s;
@@ -4441,6 +4452,8 @@ static Property migration_properties[] = {
                      send_section_footer, true),
     DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
                       decompress_error_check, true),
+    DEFINE_PROP_BOOL("multifd-flush-after-each-section", MigrationState,
+                      multifd_flush_after_each_section, true),
     DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
                       clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
 
-- 
2.39.2

[PATCH v7 2/3] multifd: Protect multifd_send_sync_main() calls

[Juan Quintela]

We only need to do that on the ram_save_iterate() call on sending and
on destination when we get a RAM_SAVE_FLAG_EOS.

In setup() and complete() we need to synch in both new and old cases,
so don't add a check there.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

---

Remove the wrappers that we take out on patch 5.
---
 migration/ram.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/migration/ram.c b/migration/ram.c
index 96e8a19a58..f019b77808 100644
--- a/migration/ram.c
+++ b/migration/ram.c
@@ -3408,9 +3408,11 @@ static int ram_save_iterate(QEMUFile *f, void *opaque)
 out:
     if (ret >= 0
         && migration_is_setup_or_active(migrate_get_current()->state)) {
-        ret = multifd_send_sync_main(rs->pss[RAM_CHANNEL_PRECOPY].pss_channel);
-        if (ret < 0) {
-            return ret;
+        if (migrate_multifd_flush_after_each_section()) {
+            ret = multifd_send_sync_main(rs->pss[RAM_CHANNEL_PRECOPY].pss_channel);
+            if (ret < 0) {
+                return ret;
+            }
         }
 
         qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
@@ -4166,7 +4168,9 @@ int ram_load_postcopy(QEMUFile *f, int channel)
 
         case RAM_SAVE_FLAG_EOS:
             /* normal exit */
-            multifd_recv_sync_main();
+            if (migrate_multifd_flush_after_each_section()) {
+                multifd_recv_sync_main();
+            }
             break;
         default:
             error_report("Unknown combination of migration flags: 0x%x"
@@ -4437,7 +4441,9 @@ static int ram_load_precopy(QEMUFile *f)
             break;
         case RAM_SAVE_FLAG_EOS:
             /* normal exit */
-            multifd_recv_sync_main();
+            if (migrate_multifd_flush_after_each_section()) {
+                multifd_recv_sync_main();
+            }
             break;
         default:
             if (flags & RAM_SAVE_FLAG_HOOK) {
-- 
2.39.2

[PATCH v7 3/3] multifd: Only flush once each full round of memory

[Juan Quintela]

We need to add a new flag to mean to flush at that point.
Notice that we still flush at the end of setup and at the end of
complete stages.

Signed-off-by: Juan Quintela <quintela@redhat.com>

---

Add missing qemu_fflush(), now it passes all tests always.
---
 migration/migration.h |  3 +--
 migration/migration.c |  6 +-----
 migration/ram.c       | 28 +++++++++++++++++++++++++++-
 3 files changed, 29 insertions(+), 8 deletions(-)

diff --git a/migration/migration.h b/migration/migration.h
index d0d89411e2..816814a24e 100644
--- a/migration/migration.h
+++ b/migration/migration.h
@@ -373,8 +373,7 @@ struct MigrationState {
      * only need to do this flush after we have go through all the
      * dirty pages.  For historical reasons, we do that after each
      * section.  This is suboptimal (we flush too many times).
-     * Default value is false.  Setting this property has no effect
-     * until the patch that removes this comment.  (since 8.1)
+     * Default value is false. (since 8.1)
      */
     bool multifd_flush_after_each_section;
     /*
diff --git a/migration/migration.c b/migration/migration.c
index ca773d00fc..353359b0d2 100644
--- a/migration/migration.c
+++ b/migration/migration.c
@@ -2705,11 +2705,7 @@ bool migrate_multifd_flush_after_each_section(void)
 {
     MigrationState *s = migrate_get_current();
 
-    /*
-     * Until the patch that remove this comment, we always return that
-     * the property is enabled.
-     */
-    return true || s->multifd_flush_after_each_section;
+    return s->multifd_flush_after_each_section;
 }
 
 bool migrate_pause_before_switchover(void)
diff --git a/migration/ram.c b/migration/ram.c
index f019b77808..977ab25351 100644
--- a/migration/ram.c
+++ b/migration/ram.c
@@ -85,6 +85,7 @@
 #define RAM_SAVE_FLAG_XBZRLE   0x40
 /* 0x80 is reserved in qemu-file.h for RAM_SAVE_FLAG_HOOK */
 #define RAM_SAVE_FLAG_COMPRESS_PAGE    0x100
+#define RAM_SAVE_FLAG_MULTIFD_FLUSH    0x200
 /* We can't use any flag that is bigger than 0x200 */
 
 int (*xbzrle_encode_buffer_func)(uint8_t *, uint8_t *, int,
@@ -1595,6 +1596,7 @@ retry:
  * associated with the search process.
  *
  * Returns:
+ *         <0: An error happened
  *         PAGE_ALL_CLEAN: no dirty page found, give up
  *         PAGE_TRY_AGAIN: no dirty page found, retry for next block
  *         PAGE_DIRTY_FOUND: dirty page found
@@ -1622,6 +1624,15 @@ static int find_dirty_block(RAMState *rs, PageSearchStatus *pss)
         pss->page = 0;
         pss->block = QLIST_NEXT_RCU(pss->block, next);
         if (!pss->block) {
+            if (!migrate_multifd_flush_after_each_section()) {
+                QEMUFile *f = rs->pss[RAM_CHANNEL_PRECOPY].pss_channel;
+                int ret = multifd_send_sync_main(f);
+                if (ret < 0) {
+                    return ret;
+                }
+                qemu_put_be64(f, RAM_SAVE_FLAG_MULTIFD_FLUSH);
+                qemu_fflush(f);
+            }
             /*
              * If memory migration starts over, we will meet a dirtied page
              * which may still exists in compression threads's ring, so we
@@ -2614,6 +2625,9 @@ static int ram_find_and_save_block(RAMState *rs)
                     break;
                 } else if (res == PAGE_TRY_AGAIN) {
                     continue;
+                } else if (res < 0) {
+                    pages = res;
+                    break;
                 }
             }
         }
@@ -3300,6 +3314,10 @@ static int ram_save_setup(QEMUFile *f, void *opaque)
         return ret;
     }
 
+    if (!migrate_multifd_flush_after_each_section()) {
+        qemu_put_be64(f, RAM_SAVE_FLAG_MULTIFD_FLUSH);
+    }
+
     qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
     qemu_fflush(f);
 
@@ -3485,6 +3503,9 @@ static int ram_save_complete(QEMUFile *f, void *opaque)
         return ret;
     }
 
+    if (!migrate_multifd_flush_after_each_section()) {
+        qemu_put_be64(f, RAM_SAVE_FLAG_MULTIFD_FLUSH);
+    }
     qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
     qemu_fflush(f);
 
@@ -4165,7 +4186,9 @@ int ram_load_postcopy(QEMUFile *f, int channel)
             }
             decompress_data_with_multi_threads(f, page_buffer, len);
             break;
-
+        case RAM_SAVE_FLAG_MULTIFD_FLUSH:
+            multifd_recv_sync_main();
+            break;
         case RAM_SAVE_FLAG_EOS:
             /* normal exit */
             if (migrate_multifd_flush_after_each_section()) {
@@ -4439,6 +4462,9 @@ static int ram_load_precopy(QEMUFile *f)
                 break;
             }
             break;
+        case RAM_SAVE_FLAG_MULTIFD_FLUSH:
+            multifd_recv_sync_main();
+            break;
         case RAM_SAVE_FLAG_EOS:
             /* normal exit */
             if (migrate_multifd_flush_after_each_section()) {
-- 
2.39.2