Re: [Qemu-devel] [PATCH COLO-Frame v10 16/38] COLO: Flush PVM's cached RAM into SVM's memory

[Dr. David Alan Gilbert]

* zhanghailiang (address@hidden) wrote:
> During the time of VM's running, PVM may dirty some pages, we will transfer
> PVM's dirty pages to SVM and store them into SVM's RAM cache at next 
> checkpoint
> time. So, the content of SVM's RAM cache will always be some with PVM's memory
> after checkpoint.
> 
> Instead of flushing all content of PVM's RAM cache into SVM's MEMORY,
> we do this in a more efficient way:
> Only flush any page that dirtied by PVM since last checkpoint.
> In this way, we can ensure SVM's memory same with PVM's.
> 
> Besides, we must ensure flush RAM cache before load device state.

Yes, just a couple of minor comments below; mostly OK.

> Signed-off-by: zhanghailiang <address@hidden>
> Signed-off-by: Li Zhijian <address@hidden>
> Signed-off-by: Gonglei <address@hidden>
> ---
> v10: trace the number of dirty pages that be received.
> ---
>  include/migration/colo.h |  1 +
>  migration/colo.c         |  2 --
>  migration/ram.c          | 40 ++++++++++++++++++++++++++++++++++++++++
>  trace-events             |  1 +
>  4 files changed, 42 insertions(+), 2 deletions(-)
> 
> diff --git a/include/migration/colo.h b/include/migration/colo.h
> index 8edd5f1..be2890b 100644
> --- a/include/migration/colo.h
> +++ b/include/migration/colo.h
> @@ -32,4 +32,5 @@ bool migration_incoming_in_colo_state(void);
>  /* ram cache */
>  int colo_init_ram_cache(void);
>  void colo_release_ram_cache(void);
> +void colo_flush_ram_cache(void);
>  #endif
> diff --git a/migration/colo.c b/migration/colo.c
> index 1339774..0efab21 100644
> --- a/migration/colo.c
> +++ b/migration/colo.c
> @@ -381,8 +381,6 @@ void *colo_process_incoming_thread(void *opaque)
>          }
>          qemu_mutex_unlock_iothread();
>  
> -        /* TODO: flush vm state */
> -
>          ret = colo_ctl_put(mis->to_src_file, COLO_COMMAND_VMSTATE_LOADED, 0);
>          if (ret < 0) {
>              goto out;
> diff --git a/migration/ram.c b/migration/ram.c
> index 70879bd..d7e0e37 100644
> --- a/migration/ram.c
> +++ b/migration/ram.c
> @@ -1601,6 +1601,7 @@ static int ram_load(QEMUFile *f, void *opaque, int 
> version_id)
>      int flags = 0, ret = 0;
>      static uint64_t seq_iter;
>      int len = 0;
> +    bool need_flush = false;
>  
>      seq_iter++;
>  
> @@ -1669,6 +1670,8 @@ static int ram_load(QEMUFile *f, void *opaque, int 
> version_id)
>                  ret = -EINVAL;
>                  break;
>              }
> +
> +            need_flush = true;
>              ch = qemu_get_byte(f);
>              ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
>              break;
> @@ -1679,6 +1682,8 @@ static int ram_load(QEMUFile *f, void *opaque, int 
> version_id)
>                  ret = -EINVAL;
>                  break;
>              }
> +
> +            need_flush = true;
>              qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
>              break;
>          case RAM_SAVE_FLAG_COMPRESS_PAGE:
> @@ -1711,6 +1716,7 @@ static int ram_load(QEMUFile *f, void *opaque, int 
> version_id)
>                  ret = -EINVAL;
>                  break;
>              }
> +            need_flush = true;

You can probably move the 'need_flush' to the big if near the top of the loop 
in the
current version.

>              break;
>          case RAM_SAVE_FLAG_EOS:
>              /* normal exit */
> @@ -1730,6 +1736,11 @@ static int ram_load(QEMUFile *f, void *opaque, int 
> version_id)
>      }
>  
>      rcu_read_unlock();
> +
> +    if (!ret  && ram_cache_enable && need_flush) {
> +        DPRINTF("Flush ram_cache\n");

trace_

> +        colo_flush_ram_cache();
> +    }
>      DPRINTF("Completed load of VM with exit code %d seq iteration "
>              "%" PRIu64 "\n", ret, seq_iter);
>      return ret;
> @@ -1799,6 +1810,35 @@ void colo_release_ram_cache(void)
>      rcu_read_unlock();
>  }
>  
> +/*
> + * Flush content of RAM cache into SVM's memory.
> + * Only flush the pages that be dirtied by PVM or SVM or both.
> + */
> +void colo_flush_ram_cache(void)
> +{
> +    RAMBlock *block = NULL;
> +    void *dst_host;
> +    void *src_host;
> +    ram_addr_t  offset = 0;
> +
> +    trace_colo_flush_ram_cache(migration_dirty_pages);
> +    rcu_read_lock();
> +    block = QLIST_FIRST_RCU(&ram_list.blocks);
> +    while (block) {
> +        offset = migration_bitmap_find_and_reset_dirty(block, offset);

You'll need to rework that a little (I split that into
migration_bitmap_find_dirty and migration_bitmap_clear_dirty)

> +        if (offset >= block->used_length) {
> +            offset = 0;
> +            block = QLIST_NEXT_RCU(block, next);
> +        } else {
> +            dst_host = block->host + offset;
> +            src_host = block->host_cache + offset;
> +            memcpy(dst_host, src_host, TARGET_PAGE_SIZE);
> +        }
> +    }
> +    rcu_read_unlock();
> +    assert(migration_dirty_pages == 0);
> +}
> +
>  static SaveVMHandlers savevm_ram_handlers = {
>      .save_live_setup = ram_save_setup,
>      .save_live_iterate = ram_save_iterate,
> diff --git a/trace-events b/trace-events
> index ee4679c..c98bc13 100644
> --- a/trace-events
> +++ b/trace-events
> @@ -1232,6 +1232,7 @@ qemu_file_fclose(void) ""
>  migration_bitmap_sync_start(void) ""
>  migration_bitmap_sync_end(uint64_t dirty_pages) "dirty_pages %" PRIu64""
>  migration_throttle(void) ""
> +colo_flush_ram_cache(uint64_t dirty_pages) "dirty_pages %" PRIu64""
>  
>  # hw/display/qxl.c
>  disable qxl_interface_set_mm_time(int qid, uint32_t mm_time) "%d %d"
> -- 
> 1.8.3.1
> 
> 
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK