avcodec/h264dec: Fix data race when updating decode_error_flags

When using multi-threaded decoding, every decoding thread
has its own DBP consisting of H264Pictures and each of these
points to its own AVFrames. They are synced during
update_thread_context via av_frame_ref() and therefore
the threads actually decoding (as well as all the others)
must not modify any field that is copied by av_frame_ref()
after ff_thread_finish_setup().

Yet this is exactly what happens when an error occurs
during decoding and the AVFrame's decode_error_flags are updated.
Given that these errors only become apparent during decoding,
this can't be set before ff_thread_finish_setup() without
defeating the point of frame-threading; in practice,
this meant that the decoder did not set these flags correctly
in case frame-threading was in use. (This means that e.g.
the ffmpeg cli tool fails to output its "corrupt decoded frame"
message in a nondeterministic fashion.)

This commit fixes this by adding a new H264Picture field
that is actually propagated across threads; the field
is an AVBufferRef* whose data is an atomic_int; it is
atomic in order to allow multiple threads to update it
concurrently and not to provide synchronization
between the threads setting the field and the thread
ultimately returning the AVFrame.

This unfortunately has the overhead of one allocation
per H264Picture (both the original one as well as
creating a reference to an existing one), even in case
of no errors. In order to mitigate this, an AVBufferPool
has been used and only if frame-threading is actually
in use. This expense will be removed as soon as
a proper API for refcounted objects (not based upon
AVBuffer) is in place.

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>

libavcodec/h264_picture.c

@@ -54,6 +54,7 @@ void ff_h264_unref_picture(H264Context *h, H264Picture *pic)
         av_buffer_unref(&pic->motion_val_buf[i]);
         av_buffer_unref(&pic->ref_index_buf[i]);
     }
+    av_buffer_unref(&pic->decode_error_flags);
 
     memset((uint8_t*)pic + off, 0, sizeof(*pic) - off);
 }
@@ -136,6 +137,10 @@ int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src)
         dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
     }
 
+    ret = av_buffer_replace(&dst->decode_error_flags, src->decode_error_flags);
+    if (ret < 0)
+        goto fail;
+
     h264_copy_picture_params(dst, src);
 
     return 0;
@@ -186,6 +191,10 @@ int ff_h264_replace_picture(H264Context *h, H264Picture *dst, const H264Picture
 
     dst->hwaccel_picture_private = src->hwaccel_picture_private;
 
+    ret = av_buffer_replace(&dst->decode_error_flags, src->decode_error_flags);
+    if (ret < 0)
+        goto fail;
+
     h264_copy_picture_params(dst, src);
 
     return 0;

libavcodec/h264_slice.c

@@ -210,6 +210,13 @@ static int alloc_picture(H264Context *h, H264Picture *pic)
     if (ret < 0)
         goto fail;
 
+    if (h->decode_error_flags_pool) {
+        pic->decode_error_flags = av_buffer_pool_get(h->decode_error_flags_pool);
+        if (!pic->decode_error_flags)
+            goto fail;
+        atomic_init((atomic_int*)pic->decode_error_flags->data, 0);
+    }
+
     if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & AV_CODEC_FLAG_GRAY && pic->f->data[2]) {
         int h_chroma_shift, v_chroma_shift;
         av_pix_fmt_get_chroma_sub_sample(pic->f->format,

libavcodec/h264dec.c

@@ -307,6 +307,12 @@ static int h264_init_context(AVCodecContext *avctx, H264Context *h)
 
     ff_h264_sei_uninit(&h->sei);
 
+    if (avctx->active_thread_type & FF_THREAD_FRAME) {
+        h->decode_error_flags_pool = av_buffer_pool_init(sizeof(atomic_int), NULL);
+        if (!h->decode_error_flags_pool)
+            return AVERROR(ENOMEM);
+    }
+
     h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? avctx->thread_count : 1;
     h->slice_ctx = av_calloc(h->nb_slice_ctx, sizeof(*h->slice_ctx));
     if (!h->slice_ctx) {
@@ -353,6 +359,8 @@ static av_cold int h264_decode_end(AVCodecContext *avctx)
 
     h->cur_pic_ptr = NULL;
 
+    av_buffer_pool_uninit(&h->decode_error_flags_pool);
+
     av_freep(&h->slice_ctx);
     h->nb_slice_ctx = 0;
 
@@ -739,7 +747,16 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size)
 
     // set decode_error_flags to allow users to detect concealed decoding errors
     if ((ret < 0 || h->er.error_occurred) && h->cur_pic_ptr) {
-        h->cur_pic_ptr->f->decode_error_flags |= FF_DECODE_ERROR_DECODE_SLICES;
+        if (h->cur_pic_ptr->decode_error_flags) {
+            /* Frame-threading in use */
+            atomic_int *decode_error = (atomic_int*)h->cur_pic_ptr->decode_error_flags->data;
+            /* Using atomics here is not supposed to provide syncronisation;
+             * they are merely used to allow to set decode_error from both
+             * decoding threads in case of coded slices. */
+            atomic_fetch_or_explicit(decode_error, FF_DECODE_ERROR_DECODE_SLICES,
+                                     memory_order_relaxed);
+        } else
+            h->cur_pic_ptr->f->decode_error_flags |= FF_DECODE_ERROR_DECODE_SLICES;
     }
 
     ret = 0;
@@ -762,6 +779,7 @@ end:
 
         H264SliceContext *sl = h->slice_ctx;
         int use_last_pic = h->last_pic_for_ec.f->buf[0] && !sl->ref_count[0];
+        int decode_error_flags = 0;
 
         ff_h264_set_erpic(&h->er.cur_pic, h->cur_pic_ptr);
 
@@ -779,7 +797,15 @@ end:
         if (sl->ref_count[1])
             ff_h264_set_erpic(&h->er.next_pic, sl->ref_list[1][0].parent);
 
-        ff_er_frame_end(&h->er, NULL);
+        ff_er_frame_end(&h->er, &decode_error_flags);
+        if (decode_error_flags) {
+            if (h->cur_pic_ptr->decode_error_flags) {
+                atomic_int *decode_error = (atomic_int*)h->cur_pic_ptr->decode_error_flags->data;
+                atomic_fetch_or_explicit(decode_error, decode_error_flags,
+                                         memory_order_relaxed);
+            } else
+                h->cur_pic_ptr->f->decode_error_flags |= decode_error_flags;
+        }
         if (use_last_pic)
             memset(&sl->ref_list[0][0], 0, sizeof(sl->ref_list[0][0]));
     }
@@ -851,6 +877,14 @@ static int output_frame(H264Context *h, AVFrame *dst, H264Picture *srcp)
     if (srcp->needs_fg && (ret = av_frame_copy_props(dst, srcp->f)) < 0)
         return ret;
 
+    if (srcp->decode_error_flags) {
+        atomic_int *decode_error = (atomic_int*)srcp->decode_error_flags->data;
+        /* The following is not supposed to provide synchronisation at all:
+         * given that srcp has already finished decoding, decode_error
+         * has already been set to its final value. */
+        dst->decode_error_flags |= atomic_load_explicit(decode_error, memory_order_relaxed);
+    }
+
     av_dict_set(&dst->metadata, "stereo_mode", ff_h264_sei_stereo_mode(&h->sei.common.frame_packing), 0);
 
     if (srcp->sei_recovery_frame_cnt == 0)

libavcodec/h264dec.h

@@ -152,6 +152,9 @@ typedef struct H264Picture {
 
     int mb_width, mb_height;
     int mb_stride;
+
+    /* data points to an atomic_int */
+    AVBufferRef *decode_error_flags;
 } H264Picture;
 
 typedef struct H264Ref {
@@ -549,6 +552,7 @@ typedef struct H264Context {
     AVBufferPool *mb_type_pool;
     AVBufferPool *motion_val_pool;
     AVBufferPool *ref_index_pool;
+    AVBufferPool *decode_error_flags_pool;
     int ref2frm[MAX_SLICES][2][64];     ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
 } H264Context;