From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by master.gitmailbox.com (Postfix) with ESMTP id D8799423BB for ; Mon, 17 Jan 2022 09:07:37 +0000 (UTC) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 7620F68AF75; Mon, 17 Jan 2022 11:07:34 +0200 (EET) Received: from mail0.khirnov.net (red.khirnov.net [176.97.15.12]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id A643B68AF34 for ; Mon, 17 Jan 2022 11:07:28 +0200 (EET) Received: from localhost (localhost [IPv6:::1]) by mail0.khirnov.net (Postfix) with ESMTP id C530124017C for ; Mon, 17 Jan 2022 10:07:27 +0100 (CET) Received: from mail0.khirnov.net ([IPv6:::1]) by localhost (mail0.khirnov.net [IPv6:::1]) (amavisd-new, port 10024) with ESMTP id x5fj8LccDl_d for ; Mon, 17 Jan 2022 10:07:27 +0100 (CET) Received: from lain.red.khirnov.net (lain.red.khirnov.net [IPv6:2001:67c:1138:4306::3]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256 client-signature RSA-PSS (2048 bits) client-digest SHA256) (Client CN "lain.red.khirnov.net", Issuer "smtp.khirnov.net SMTP CA" (verified OK)) by mail0.khirnov.net (Postfix) with ESMTPS id 31436240179 for ; Mon, 17 Jan 2022 10:07:27 +0100 (CET) Received: by lain.red.khirnov.net (Postfix, from userid 1000) id 1F4A016008E; Mon, 17 Jan 2022 10:07:27 +0100 (CET) From: Anton Khirnov To: FFmpeg development discussions and patches In-Reply-To: References: <20220103003318.442892-1-aicommander@gmail.com> <164172029148.2375.17275383230465998544@lain.red.khirnov.net> Mail-Followup-To: FFmpeg development discussions and patches Date: Mon, 17 Jan 2022 10:07:26 +0100 Message-ID: <164241044699.23111.6270013532570362968@lain.red.khirnov.net> User-Agent: alot/0.8.1 MIME-Version: 1.0 Subject: Re: [FFmpeg-devel] [PATCH] lavu/videotoolbox: add support for memory mapping frames X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" Archived-At: List-Archive: List-Post: Quoting Cameron Gutman (2022-01-10 09:17:37) > > > On Jan 9, 2022, at 3:24 AM, Anton Khirnov wrote: > > > > Quoting Cameron Gutman (2022-01-03 01:33:19) > >> Signed-off-by: Cameron Gutman > >> --- > >> libavutil/hwcontext_videotoolbox.c | 25 +++++++++++++++++++++++++ > >> 1 file changed, 25 insertions(+) > >> > >> diff --git a/libavutil/hwcontext_videotoolbox.c b/libavutil/hwcontext_videotoolbox.c > >> index 0a8dbe9f33..026127d412 100644 > >> --- a/libavutil/hwcontext_videotoolbox.c > >> +++ b/libavutil/hwcontext_videotoolbox.c > >> @@ -711,6 +711,30 @@ fail: > >> return err; > >> } > >> > >> +static int vt_map_from(AVHWFramesContext *hwfc, AVFrame *dst, > >> + const AVFrame *src, int flags) > >> +{ > >> + int err; > >> + > >> + if (dst->format == AV_PIX_FMT_NONE) > >> + dst->format = hwfc->sw_format; > >> + else if (dst->format != hwfc->sw_format) > >> + return AVERROR(ENOSYS); > >> + > >> + err = vt_map_frame(hwfc, dst, src, flags); > >> + if (err) > >> + return err; > >> + > >> + dst->width = src->width; > >> + dst->height = src->height; > >> + > >> + err = av_frame_copy_props(dst, src); > >> + if (err) > >> + return err; > > > > Don't you need to unmap the frame in this error path? > > > > Maybe? It's complicated... > > The mapping is still written to dst and it will be unmapped when > av_frame_unref() is called on dst. However, if the caller wants to try again > with same dst frame for some reason, then it looks like it will leak the first > "failed" mapping. AFAICT, another call to ff_hwframe_map_create() will > overwrite dst->buf[0] without unrefing it first, but that makes sense given > that the docs say "dst should be blank" (arguably that "should" ought to be a > "must" in this case). However, this isn't the full story. > > None of the existing map_from() implementations (VAAPI, DRM, DXVA2) actually > unmap when av_frame_copy_props() fails. The only ones that don't have this bug > are OpenCL and Vulkan, and that's only because they forgot to call > av_frame_copy_props() entirely! > > Ideally, you'd have nothing modified in dst when av_hwframe_map() fails, > but that isn't the case in practice. Much of the mapping process involves > irreversible changes to the dst frame, including overwriting dst->format, > dst->width, dst->height, dst->data, dst->linesize, even partially copied > frame properties prior to av_frame_copy_props() failing. > > Given these limitations, it seems like the only sane thing to do with a dst > frame after failure of av_hwframe_map() (other than ENOSYS) is to unref/free. > The frame is basically in an undefined state after such a failure. If that's > the case, then we don't actually have a leak here since av_frame_unref() > will do the right thing and free the old mapping. Right, but who will call this av_frame_unref(). The doxy for av_hwframe_map() does not specify what precisely happens on failure, but I would expect it to clean dst. -- Anton Khirnov _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".