[Qemu-devel] [PATCH v2 3/3] CPUPhysMemoryClient: Batch contiguous addresses when playing catchup

[Alex Williamson <alex.williamson@redhat.com>]

When a phys memory client registers and we play catchup by walking
the page tables, we can make a huge improvement in the number of
times the set_memory callback is called by batching contiguous
pages together.  With a 4G guest, this reduces the number of callbacks
at registration from 1048866 to 296.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
---

 exec.c |   38 ++++++++++++++++++++++++++++++++------
 1 files changed, 32 insertions(+), 6 deletions(-)

diff --git a/exec.c b/exec.c
index bbd5c86..a0678a4 100644
--- a/exec.c
+++ b/exec.c
@@ -1741,14 +1741,21 @@ static int cpu_notify_migration_log(int enable)
     return 0;
 }
 
+struct last_map {
+    target_phys_addr_t start_addr;
+    ram_addr_t size;
+    ram_addr_t phys_offset;
+};
+
 /* The l1_phys_map provides the upper P_L1_BITs of the guest physical
  * address.  Each intermediate table provides the next L2_BITs of guest
  * physical address space.  The number of levels vary based on host and
  * guest configuration, making it efficient to build the final guest
  * physical address by seeding the L1 offset and shifting and adding in
  * each L2 offset as we recurse through them. */
-static void phys_page_for_each_1(CPUPhysMemoryClient *client,
-                                 int level, void **lp, target_phys_addr_t addr)
+static void phys_page_for_each_1(CPUPhysMemoryClient *client, int level,
+                                 void **lp, target_phys_addr_t addr,
+                                 struct last_map *map)
 {
     int i;
 
@@ -1760,15 +1767,29 @@ static void phys_page_for_each_1(CPUPhysMemoryClient *client,
         addr <<= L2_BITS + TARGET_PAGE_BITS;
         for (i = 0; i < L2_SIZE; ++i) {
             if (pd[i].phys_offset != IO_MEM_UNASSIGNED) {
-                client->set_memory(client, addr | i << TARGET_PAGE_BITS,
-                                   TARGET_PAGE_SIZE, pd[i].phys_offset);
+                target_phys_addr_t start_addr = addr | i << TARGET_PAGE_BITS;
+
+                if (map->size &&
+                    start_addr == map->start_addr + map->size &&
+                    pd[i].phys_offset == map->phys_offset + map->size) {
+
+                    map->size += TARGET_PAGE_SIZE;
+                    continue;
+                } else if (map->size) {
+                    client->set_memory(client, map->start_addr,
+                                       map->size, map->phys_offset);
+                }
+
+                map->start_addr = start_addr;
+                map->size = TARGET_PAGE_SIZE;
+                map->phys_offset = pd[i].phys_offset;
             }
         }
     } else {
         void **pp = *lp;
         for (i = 0; i < L2_SIZE; ++i) {
             phys_page_for_each_1(client, level - 1, pp + i,
-                                 (addr << L2_BITS) | i);
+                                 (addr << L2_BITS) | i, map);
         }
     }
 }
@@ -1776,9 +1797,14 @@ static void phys_page_for_each_1(CPUPhysMemoryClient *client,
 static void phys_page_for_each(CPUPhysMemoryClient *client)
 {
     int i;
+    struct last_map map = { 0 };
+
     for (i = 0; i < P_L1_SIZE; ++i) {
         phys_page_for_each_1(client, P_L1_SHIFT / L2_BITS - 1,
-                             l1_phys_map + i, i);
+                             l1_phys_map + i, i, &map);
+    }
+    if (map.size) {
+        client->set_memory(client, map.start_addr, map.size, map.phys_offset);
     }
 }