[Spice-devel] [RFC PATCH spice-server v4 17/22] stream-device: Limit sending queue from guest to server
Christophe de Dinechin
cdupontd at redhat.com
Tue Aug 29 10:21:23 UTC 2017
> On 25 Aug 2017, at 11:54, Frediano Ziglio <fziglio at redhat.com> wrote:
>
> Do not allow the guest to fill host memory.
Technically, it’s not the guest, is it?
> Also having a huge queue mainly cause to have a higher video
> latency.
“Also, having a huge queue increases the video latency”
>
> Signed-off-by: Frediano Ziglio <fziglio at redhat.com>
> ---
> server/stream-channel.c | 41 ++++++++++++++++++++++++++++++++++++++++-
> server/stream-channel.h | 10 ++++++++++
> server/stream-device.c | 35 ++++++++++++++++++++++++++++++++++-
> 3 files changed, 84 insertions(+), 2 deletions(-)
>
> diff --git a/server/stream-channel.c b/server/stream-channel.c
> index 53a06f91..34c8a351 100644
> --- a/server/stream-channel.c
> +++ b/server/stream-channel.c
> @@ -69,9 +69,15 @@ struct StreamChannel {
> /* size of the current video stream */
> unsigned width, height;
>
> + StreamQueueStat queue_stat;
Maybe it’s just me, but I read ‘stat’ as an abbreviation for ‘status’.
For statistics, I would rather use ‘stats’ (plural).
That being said, it looks like ‘stat’ is the common abbreviation in Spice…
> +
> /* callback to notify when a stream should be started or stopped */
> stream_channel_start_proc start_cb;
> void *start_opaque;
> +
> + /* callback to notify when queue statistics changes */
> + stream_channel_queue_stat_proc queue_cb;
> + void *queue_opaque;
> };
>
> struct StreamChannelClass {
> @@ -96,6 +102,7 @@ typedef struct StreamCreateItem {
>
> typedef struct StreamDataItem {
> RedPipeItem base;
> + StreamChannel *channel;
> // NOTE: this must be the last field in the structure
> SpiceMsgDisplayStreamData data;
> } StreamDataItem;
> @@ -453,6 +460,27 @@ stream_channel_change_format(StreamChannel *channel, const StreamMsgFormat *fmt)
> red_pipe_item_unref(&item->base);
> }
>
> +static inline void
> +stream_channel_update_queue_stat(StreamChannel *channel,
> + int32_t num_diff, int32_t size_diff)
> +{
> + channel->queue_stat.num_items += num_diff;
> + channel->queue_stat.size += size_diff;
> + if (channel->queue_cb) {
> + channel->queue_cb(channel->queue_opaque, &channel->queue_stat, channel);
> + }
> +}
> +
> +static void
> +data_item_free(RedPipeItem *base)
> +{
> + StreamDataItem *pipe_item = SPICE_UPCAST(StreamDataItem, base);
> +
> + stream_channel_update_queue_stat(pipe_item->channel, -1, -pipe_item->data.data_size);
> +
> + free(pipe_item);
> +}
> +
> void
> stream_channel_send_data(StreamChannel *channel, const void *data, size_t size, uint32_t mm_time)
> {
> @@ -466,10 +494,13 @@ stream_channel_send_data(StreamChannel *channel, const void *data, size_t size,
> RedChannel *red_channel = RED_CHANNEL(channel);
>
> StreamDataItem *item = spice_malloc(sizeof(*item) + size);
> - red_pipe_item_init(&item->base, RED_PIPE_ITEM_TYPE_STREAM_DATA);
> + red_pipe_item_init_full(&item->base, RED_PIPE_ITEM_TYPE_STREAM_DATA,
> + data_item_free);
> item->data.base.id = channel->stream_id;
> item->data.base.multi_media_time = mm_time;
> item->data.data_size = size;
> + item->channel = channel;
> + stream_channel_update_queue_stat(channel, 1, size);
> // TODO try to optimize avoiding the copy
> memcpy(item->data.data, data, size);
> red_channel_pipes_new_add(red_channel, pipe_item_new_ref, item);
> @@ -485,6 +516,14 @@ stream_channel_register_start_cb(StreamChannel *channel,
> }
>
> void
> +stream_channel_register_queue_stat_cb(StreamChannel *channel,
> + stream_channel_queue_stat_proc cb, void *opaque)
> +{
> + channel->queue_cb = cb;
> + channel->queue_opaque = opaque;
> +}
> +
> +void
> stream_channel_reset(StreamChannel *channel)
> {
> RedChannel *red_channel = RED_CHANNEL(channel);
> diff --git a/server/stream-channel.h b/server/stream-channel.h
> index bd075a95..f961d715 100644
> --- a/server/stream-channel.h
> +++ b/server/stream-channel.h
> @@ -67,6 +67,16 @@ typedef void (*stream_channel_start_proc)(void *opaque, struct StreamMsgStartSto
> void stream_channel_register_start_cb(StreamChannel *channel,
> stream_channel_start_proc cb, void *opaque);
>
> +typedef struct StreamQueueStat {
> + uint32_t num_items;
> + uint32_t size;
> +} StreamQueueStat;
> +
> +typedef void (*stream_channel_queue_stat_proc)(void *opaque, const StreamQueueStat *stats,
> + StreamChannel *channel);
> +void stream_channel_register_queue_stat_cb(StreamChannel *channel,
> + stream_channel_queue_stat_proc cb, void *opaque);
> +
> G_END_DECLS
>
> #endif /* STREAM_CHANNEL_H_ */
> diff --git a/server/stream-device.c b/server/stream-device.c
> index f72140ab..360ee60f 100644
> --- a/server/stream-device.c
> +++ b/server/stream-device.c
> @@ -44,6 +44,7 @@ struct StreamDevice {
> uint8_t hdr_pos;
> bool has_error;
> bool opened;
> + bool flow_stopped;
> StreamChannel *stream_channel;
> };
>
> @@ -67,7 +68,7 @@ stream_device_read_msg_from_dev(RedCharDevice *self, SpiceCharDeviceInstance *si
> SpiceCharDeviceInterface *sif;
> int n;
>
> - if (dev->has_error || !dev->stream_channel) {
> + if (dev->has_error || dev->flow_stopped || !dev->stream_channel) {
> return NULL;
> }
>
> @@ -165,6 +166,9 @@ handle_msg_data(StreamDevice *dev, SpiceCharDeviceInstance *sin)
> if (n <= 0) {
> break;
> }
> + // TODO collect all message ??
> + // up: we send a single frame together
> + // down: guest can cause a crash
> stream_channel_send_data(dev->stream_channel, buf, n, reds_get_mm_time());
> dev->hdr.size -= n;
> }
> @@ -218,6 +222,33 @@ stream_device_stream_start(void *opaque, StreamMsgStartStop *start,
> red_char_device_write_buffer_add(char_dev, buf);
> }
>
> +static void
> +stream_device_stream_queue_stat(void *opaque, const StreamQueueStat *stats G_GNUC_UNUSED,
> + StreamChannel *stream_channel G_GNUC_UNUSED)
> +{
> + StreamDevice *dev = (StreamDevice *) opaque;
> +
> + if (!dev->opened) {
> + return;
> + }
> +
> + // very easy control flow... if any data stop
> + // this seems a very small queue but as we use tcp
> + // there's already that queue
> + if (stats->num_items) {
I would have an explicit configuration:
if (stat->num_items >= SPICE_MAX_NUM_ITEMS) …
I wonder if we shouldn’t make this value part of the stream device,
instead of a constant?
Even after reading the comment about TCP, I find it a bit extreme
to set a max queue size of 1 :-)
> + dev->flow_stopped = true;
> + return;
> + }
> +
> + if (dev->flow_stopped) {
> + dev->flow_stopped = false;
> + // TODO resume flow...
> + // avoid recursion if we need to call get data from data handling from
> + // data handling
> + red_char_device_wakeup(&dev->parent);
> + }
> +}
> +
> RedCharDevice *
> stream_device_connect(RedsState *reds, SpiceCharDeviceInstance *sin)
> {
> @@ -262,6 +293,7 @@ allocate_channels(StreamDevice *dev)
> dev->stream_channel = stream_channel;
>
> stream_channel_register_start_cb(stream_channel, stream_device_stream_start, dev);
> + stream_channel_register_queue_stat_cb(stream_channel, stream_device_stream_queue_stat, dev);
> }
>
> static void
> @@ -280,6 +312,7 @@ stream_device_port_event(RedCharDevice *char_dev, uint8_t event)
> }
> device->hdr_pos = 0;
> device->has_error = false;
> + device->flow_stopped = false;
> red_char_device_reset(char_dev);
> if (device->stream_channel) {
> stream_channel_reset(device->stream_channel);
> --
> 2.13.5
>
> _______________________________________________
> Spice-devel mailing list
> Spice-devel at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/spice-devel
More information about the Spice-devel
mailing list