[Qemu-devel] [PATCH 3/4] ram: add support for seekable file save

[Wenchao Xia]

  It use plane layout for ram.

Signed-off-by: Wenchao Xia <address@hidden>
---
 arch_init.c |  242 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 242 insertions(+), 0 deletions(-)

diff --git a/arch_init.c b/arch_init.c
index 8daeafa..0c12095 100644
--- a/arch_init.c
+++ b/arch_init.c
@@ -657,6 +657,245 @@ static int ram_save_iterate(QEMUFile *f, void *opaque)
     return total_sent;
 }
 
+/* ram save for seekable file support */
+/* RAMBlock lay out:
+   every RAMBlock: page_number = block->length >> TARGET_PAGE_BITS
+   1st page:
+     8 bytes : offset | cont | fag
+     1 byte: strlen(block->idstr)
+     n bytes: block->idstr
+     TARGET_PAGE_SIZE bytes: page content.
+   other pages:
+     8 bytes : offset | cont | fag
+     TARGET_PAGE_SIZE bytes: page content.
+   Total:
+     8+1+strlen(block->idstr)+TARGET_PAGE_SIZE +
+     (page_number - 1) * (8 + TARGET_PAGE_SIZE) =
+     1+strlen(block->idstr)+ page_number * (8 + TARGET_PAGE_SIZE)
+   RAM_SAVE_FLAG_EOS is not included in it.
+*/
+
+static int64_t ram_file_get_block_size(RAMBlock *block)
+{
+    g_assert((block->length & (TARGET_PAGE_SIZE - 1)) == 0);
+    int64_t page_num = block->length >> TARGET_PAGE_BITS;
+    int64_t block_size = 1 + strlen(block->idstr) +
+                          page_num * (8 + TARGET_PAGE_SIZE);
+    return block_size;
+}
+
+static int64_t ram_file_get_block_total_size(void)
+{
+    int64_t total_size = 0, block_size;
+
+    RAMBlock *block = QTAILQ_FIRST(&ram_list.blocks);
+    while (block != NULL) {
+        block_size = ram_file_get_block_size(block);
+        total_size += block_size;
+        block = QTAILQ_NEXT(block, next);
+    }
+    return total_size;
+}
+
+static int64_t ram_file_get_addr_offset(RAMBlock *block, ram_addr_t offset)
+{
+    int64_t nr = offset >> TARGET_PAGE_BITS;
+    int64_t ret;
+
+    if (nr == 0) {
+        ret = 0;
+    } else {
+        ret = 1 + strlen(block->idstr) + nr * (8 + TARGET_PAGE_SIZE);
+    }
+    return ret;
+}
+
+/*
+ * ram_save_block_seekable: Writes a page of memory to a seekable f, used
+ * write the data to local images, so wirte them with address and do not
+ * compress, let block layer handler the compression.
+ *
+ * Returns:  The number of bytes written.
+ *           0 means no dirty pages
+ */
+/* assume block would not change during the live save vmstate */
+static int ram_save_block_seekable(QEMUFile *f, int64_t base, int64_t max)
+{
+    RAMBlock *block = last_seen_block;
+    ram_addr_t offset = last_offset;
+    bool complete_round = false;
+    int bytes_sent = 0;
+    MemoryRegion *mr;
+    static uint64_t block_offset;
+
+    if (!block) {
+        block = QTAILQ_FIRST(&ram_list.blocks);
+        block_offset = 0;
+    }
+
+    while (true) {
+        mr = block->mr;
+        offset = migration_bitmap_find_and_reset_dirty(mr, offset);
+        if (complete_round && block == last_seen_block &&
+            offset >= last_offset) {
+            break;
+        }
+        if (offset >= block->length) {
+            offset = 0;
+            block_offset += ram_file_get_block_size(block);
+            block = QTAILQ_NEXT(block, next);
+            if (!block) {
+                block = QTAILQ_FIRST(&ram_list.blocks);
+                complete_round = true;
+                block_offset = 0;
+            }
+        } else {
+            uint8_t *p;
+            uint64_t addr_offset, file_offset;
+            int ret;
+
+            int cont = (block == last_sent_block) ?
+                RAM_SAVE_FLAG_CONTINUE : 0;
+
+            p = memory_region_get_ram_ptr(mr) + offset;
+
+            addr_offset = ram_file_get_addr_offset(block, offset);
+            file_offset = base + block_offset + addr_offset;
+            if (file_offset > max) {
+                printf("error!:%ld, max %ld.\n", file_offset, max);
+                return -1;
+            }
+            ret = qemu_fseek(f, file_offset);
+            if (ret < 0) {
+                return ret;
+            }
+
+            /* normal page */
+            bytes_sent = save_block_hdr(f, block, offset, cont,
+                                        RAM_SAVE_FLAG_PAGE);
+            qemu_put_buffer(f, p, TARGET_PAGE_SIZE);
+            bytes_sent += TARGET_PAGE_SIZE;
+            acct_info.norm_pages++;
+
+            /* if page is unmodified, continue to the next */
+            if (bytes_sent > 0) {
+                last_sent_block = block;
+                break;
+            }
+        }
+    }
+    last_seen_block = block;
+    last_offset = offset;
+
+    return bytes_sent;
+}
+
+#define RAM_SAVE_FLAG_EOS_SIZE (8)
+static int ram_save_iterate_seekable(QEMUFile *f, int64_t base,
+                                     int64_t max, void *opaque)
+{
+    int ret;
+    int i;
+    int64_t t0;
+    int total_sent = 0;
+
+    qemu_mutex_lock_ramlist();
+
+    if (ram_list.version != last_version) {
+        reset_ram_globals();
+    }
+
+    t0 = qemu_get_clock_ns(rt_clock);
+    i = 0;
+    while ((ret = qemu_file_rate_limit(f)) == 0) {
+        int bytes_sent;
+
+        bytes_sent = ram_save_block_seekable(f, base,
+                                         max - RAM_SAVE_FLAG_EOS_SIZE);
+        /* no more blocks to sent */
+        if (bytes_sent == 0) {
+            break;
+        } else if (bytes_sent < 0) {
+            return bytes_sent;
+        }
+        total_sent += bytes_sent;
+        acct_info.iterations++;
+        /* we want to check in the 1st loop, just in case it was the 1st time
+           and we had to sync the dirty bitmap.
+           qemu_get_clock_ns() is a bit expensive, so we only check each some
+           iterations
+        */
+        if ((i & 63) == 0) {
+            uint64_t t1 = (qemu_get_clock_ns(rt_clock) - t0) / 1000000;
+            if (t1 > MAX_WAIT) {
+                DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n",
+                        t1, i);
+                break;
+            }
+        }
+        i++;
+    }
+
+    qemu_mutex_unlock_ramlist();
+
+    if (ret < 0) {
+        bytes_transferred += total_sent;
+        return ret;
+    }
+
+    bytes_transferred += total_sent;
+
+    return total_sent;
+}
+
+static int64_t ram_save_iterate_seekable_get_size(void *opaque)
+{
+    return ram_file_get_block_total_size() + RAM_SAVE_FLAG_EOS_SIZE;
+}
+
+static int ram_save_complete_seekable(QEMUFile *f, int64_t iterate_base,
+                    int64_t iterate_max, int64_t complete_base, void *opaque)
+{
+    qemu_mutex_lock_ramlist();
+    migration_bitmap_sync();
+
+    /* try transferring iterative blocks of memory */
+
+    /* flush all remaining blocks regardless of rate limiting */
+    while (true) {
+        int bytes_sent;
+
+        bytes_sent = ram_save_block_seekable(f, iterate_base,
+                                    iterate_max - RAM_SAVE_FLAG_EOS_SIZE);
+        /* no more blocks to sent */
+        if (bytes_sent == 0) {
+            break;
+        } else if (bytes_sent < 0) {
+            return bytes_sent;
+        }
+        bytes_transferred += bytes_sent;
+    }
+    /* put on flag */
+    int ret = qemu_fseek(f, ram_file_get_block_total_size() + iterate_base);
+    if (ret < 0) {
+        return ret;
+    }
+    qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
+    bytes_transferred += 8;
+
+    ret = qemu_fseek(f, complete_base);
+    if (ret < 0) {
+        return ret;
+    }
+
+    migration_end();
+
+    qemu_mutex_unlock_ramlist();
+    qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
+
+    return 0;
+}
+
 static int ram_save_complete(QEMUFile *f, void *opaque)
 {
     qemu_mutex_lock_ramlist();
@@ -880,6 +1119,9 @@ SaveVMHandlers savevm_ram_handlers = {
     .save_live_iterate = ram_save_iterate,
     .save_live_complete = ram_save_complete,
     .save_live_pending = ram_save_pending,
+    .save_live_iterate_seekable = ram_save_iterate_seekable,
+    .save_live_iterate_seekable_get_size = ram_save_iterate_seekable_get_size,
+    .save_live_complete_seekable = ram_save_complete_seekable,
     .load_state = ram_load,
     .cancel = ram_migration_cancel,
 };
-- 
1.7.1