Subject: [RFC PATCH v2] migration: calculate remaining pages accurately during the bulk stage

* Subject: [RFC PATCH v2] migration: calculate remaining pages accurately during the bulk stage
@ 2018-09-05 14:17 ` Quan Xu
  0 siblings, 0 replies; 4+ messages in thread
From: Quan Xu @ 2018-09-05 14:17 UTC (permalink / raw)
  To: kvm, qemu-devel; +Cc: Dr. David Alan Gilbert, quintela

 From 7de4cc7c944bfccde0ef10992a7ec882fdcf0508 Mon Sep 17 00:00:00 2001
From: Quan Xu <quan.xu0@gmail.com>
Date: Wed, 5 Sep 2018 22:06:58 +0800
Subject: [RFC PATCH v2] migration: calculate remaining pages accurately 
during the bulk stage

Since the bulk stage assumes in (migration_bitmap_find_dirty) that every
page is dirty, initialize the number of dirty pages at the beggining of
the iteration and then decrease it for each processed page.

Signed-off-by: Quan Xu <quan.xu0@gmail.com>
---
  migration/ram.c | 7 ++++++-
  1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/migration/ram.c b/migration/ram.c
index 79c8942..1a11436 100644
--- a/migration/ram.c
+++ b/migration/ram.c
@@ -290,6 +290,8 @@ struct RAMState {
      uint32_t last_version;
      /* We are in the first round */
      bool ram_bulk_stage;
+    /* Remaining bytes in the first round */
+    uint64_t ram_bulk_bytes;
      /* How many times we have dirty too many pages */
      int dirty_rate_high_cnt;
      /* these variables are used for bitmap sync */
@@ -1540,6 +1542,7 @@ unsigned long migration_bitmap_find_dirty(RAMState 
*rs, RAMBlock *rb,

      if (rs->ram_bulk_stage && start > 0) {
          next = start + 1;
+        rs->ram_bulk_bytes -= TARGET_PAGE_SIZE;
      } else {
          next = find_next_bit(bitmap, size, start);
      }
@@ -2001,6 +2004,7 @@ static bool find_dirty_block(RAMState *rs, 
PageSearchStatus *pss, bool *again)
              /* Flag that we've looped */
              pss->complete_round = true;
              rs->ram_bulk_stage = false;
+            rs->ram_bulk_bytes = 0;
              if (migrate_use_xbzrle()) {
                  /* If xbzrle is on, stop using the data compression at 
this
                   * point. In theory, xbzrle can do better than 
compression.
@@ -2513,6 +2517,7 @@ static void ram_state_reset(RAMState *rs)
      rs->last_page = 0;
      rs->last_version = ram_list.version;
      rs->ram_bulk_stage = true;
+    rs->ram_bulk_bytes = ram_bytes_total();
  }

  #define MAX_WAIT 50 /* ms, half buffered_file limit */
@@ -3308,7 +3313,7 @@ static void ram_save_pending(QEMUFile *f, void 
*opaque, uint64_t max_size,
          /* We can do postcopy, and all the data is postcopiable */
          *res_compatible += remaining_size;
      } else {
-        *res_precopy_only += remaining_size;
+        *res_precopy_only += remaining_size + rs->ram_bulk_bytes;
      }
  }

--
1.8.3.1

^ permalink raw reply related	[flat|nested] 4+ messages in thread