[Spice-devel] [PATCH 4/4] replay: Handle cursor commands

[Jonathon Jongsma]

On Fri, 2016-06-03 at 10:59 +0100, Frediano Ziglio wrote:
> Signed-off-by: Frediano Ziglio <fziglio at redhat.com>
> ---
>  server/tests/replay.c | 50 ++++++++++++++++++++++++++++++++------------------
>  1 file changed, 32 insertions(+), 18 deletions(-)
> 
> diff --git a/server/tests/replay.c b/server/tests/replay.c
> index d552327..528609b 100644
> --- a/server/tests/replay.c
> +++ b/server/tests/replay.c
> @@ -52,7 +52,8 @@ static gboolean print_count = FALSE;
>  static guint ncommands = 0;
>  static pid_t client_pid;
>  static GMainLoop *loop = NULL;
> -static GAsyncQueue *aqueue = NULL;
> +static GAsyncQueue *cmd_queue = NULL;
> +static GAsyncQueue *cursor_queue = NULL;
>  static long total_size;
>  
>  static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
> @@ -113,10 +114,12 @@ static gboolean fill_queue_idle(gpointer user_data)
>      gboolean keep = FALSE;
>      gboolean wakeup = FALSE;
>  
> -    while (g_async_queue_length(aqueue) < 50) {
> +    while ((g_async_queue_length(cmd_queue) +
> +            g_async_queue_length(cursor_queue)) < 50) {
>          QXLCommandExt *cmd = spice_replay_next_cmd(replay, qxl_worker);
>          if (!cmd) {
> -            g_async_queue_push(aqueue, GINT_TO_POINTER(-1));
> +            g_async_queue_push(cmd_queue, GINT_TO_POINTER(-1));
> +            g_async_queue_push(cursor_queue, GINT_TO_POINTER(-1));
>              goto end;
>          }
>  
> @@ -127,7 +130,11 @@ static gboolean fill_queue_idle(gpointer user_data)
>          }
>  
>          wakeup = TRUE;
> -        g_async_queue_push(aqueue, cmd);
> +        if (cmd->cmd.type == QXL_CMD_CURSOR) {
> +            g_async_queue_push(cursor_queue, cmd);
> +        } else {
> +            g_async_queue_push(cmd_queue, cmd);
> +        }
>      }
>  
>  end:
> @@ -166,17 +173,21 @@ end:
>  
>  
>  // called from spice_server thread (i.e. red_worker thread)
> -static int get_command(QXLInstance *qin, QXLCommandExt *ext)
> +static int get_command_from(QXLInstance *qin, QXLCommandExt *ext, GAsyncQueue
> *queue)
>  {
>      QXLCommandExt *cmd;
>  
> -    if (g_async_queue_length(aqueue) == 0) {
> +    if (g_async_queue_length(cmd_queue) == 0 &&
> +        g_async_queue_length(cursor_queue) == 0) {

why not just check "g_async_queue_length(queue) == 0"? if we're trying to get a
command from the cursor_queue and it's empty, should the fact that the other
queue is not empty prevent us from trying to fill the queue? It's probably fine
this way, just curious.

>          /* could use a gcondition ? */
>          fill_queue();
>          return FALSE;
>      }
>  
> -    cmd = g_async_queue_try_pop(aqueue);
> +    cmd = g_async_queue_try_pop(queue);
> +    if (cmd == NULL) {
> +        return FALSE;
> +    }
>      if (GPOINTER_TO_INT(cmd) == -1) {
>          g_main_loop_quit(loop);
>          return FALSE;
> @@ -187,8 +198,14 @@ static int get_command(QXLInstance *qin, QXLCommandExt
> *ext)
>      return TRUE;
>  }
>  
> -static int req_cmd_notification(QXLInstance *qin)
> +static int get_command(QXLInstance *qin, QXLCommandExt *ext)
> +{
> +    return get_command_from(qin, ext, cmd_queue);
> +}
> +
> +static int req_notification(QXLInstance *qin)
>  {
> +    /* we don't have currently message pending */
>      return TRUE;
>  }
>  
> @@ -214,12 +231,7 @@ static void release_resource(QXLInstance *qin, struct
> QXLReleaseInfoExt release_
>  
>  static int get_cursor_command(QXLInstance *qin, struct QXLCommandExt *ext)
>  {
> -    return FALSE;
> -}
> -
> -static int req_cursor_notification(QXLInstance *qin)
> -{
> -    return TRUE;
> +    return get_command_from(qin, ext, cursor_queue);
>  }
>  
>  static void notify_update(QXLInstance *qin, uint32_t update_id)
> @@ -243,10 +255,10 @@ static QXLInterface display_sif = {
>      .set_mm_time = set_mm_time,
>      .get_init_info = get_init_info,
>      .get_command = get_command,
> -    .req_cmd_notification = req_cmd_notification,
> +    .req_cmd_notification = req_notification,
>      .release_resource = release_resource,
>      .get_cursor_command = get_cursor_command,
> -    .req_cursor_notification = req_cursor_notification,
> +    .req_cursor_notification = req_notification,
>      .notify_update = notify_update,
>      .flush_resources = flush_resources,
>  };
> @@ -379,7 +391,8 @@ int main(int argc, char **argv)
>          exit(1);
>      }
>  
> -    aqueue = g_async_queue_new();
> +    cmd_queue = g_async_queue_new();
> +    cursor_queue = g_async_queue_new();
>      core = basic_event_loop_init();
>      core->channel_event = replay_channel_event;
>  
> @@ -414,7 +427,8 @@ int main(int argc, char **argv)
>          g_print("Counted %d commands\n", ncommands);
>  
>      end_replay();
> -    g_async_queue_unref(aqueue);
> +    g_async_queue_unref(cmd_queue);
> +    g_async_queue_unref(cursor_queue);
>  
>      /* FIXME: there should be a way to join server threads before:
>       * g_main_loop_unref(loop);

aside from the minor question above, looks fine.

Acked-by: Jonathon Jongsma <jjongsma at redhat.com>