From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Alexander E. Patrakov" Subject: Re: [PATCH 0/9] Misc fixes related to rewinds Date: Sun, 14 Sep 2014 01:31:11 +0600 Message-ID: <54149B7F.6000002@gmail.com> References: <1410633021-20395-1-git-send-email-patrakov@gmail.com> <54149789.2060102@perex.cz> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; Format="flowed" Content-Transfer-Encoding: 7bit Return-path: Received: from mail-lb0-f181.google.com (mail-lb0-f181.google.com [209.85.217.181]) by alsa0.perex.cz (Postfix) with ESMTP id DD6D0261B02 for ; Sat, 13 Sep 2014 21:31:13 +0200 (CEST) Received: by mail-lb0-f181.google.com with SMTP id z11so2535477lbi.26 for ; Sat, 13 Sep 2014 12:31:13 -0700 (PDT) In-Reply-To: <54149789.2060102@perex.cz> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: alsa-devel-bounces@alsa-project.org Sender: alsa-devel-bounces@alsa-project.org To: Jaroslav Kysela , alsa-devel@alsa-project.org Cc: tiwai@suse.de, clemens@ladisch.de, David Henningsson List-Id: alsa-devel@alsa-project.org 14.09.2014 01:14, Jaroslav Kysela wrote: > Date 13.9.2014 20:30, Alexander E. Patrakov wrote: >> The idea of the series is to fix the two issues that I found [1] for the > > I applied all your patches to alsa-lib's repo, but... > >> hw plugin. snd_pcm_rewindable() sometimes returned negative values that >> are actually negative amounts of samples and not error codes. Also, it >> bases its calculations on stale hardware position pointer, which is not >> what PulseAudio wants (alternatively, we can document the need to call >> snd_pcm_avail() before snd_pcm_rewindable(), but I don't like it). > > The hw sync is expensive and the application might do this sync multiple > times when woken up. I think that it must be clear that: > > 1) only snd_pcm_avail(), snd_pcm_delay(), snd_pcm_avail_delay() > does the real hw sync > 2) snd_pcm_avail(), snd_pcm_delay(), snd_pcm_avail_delay(), > snd_pcm_rewindable() and snd_pcm_forwardable() does > hw sync (and change all plugins to respect this) > > I don't like the situation "be somewhere between because it's good for > one purpose"... I understand the concern. I have specifically not added the call to hwsync directly to snd_pcm_rewindable implementation (although it would have resulted in a smaller patch), because that would indeed cause double-hwsync and the resulting inefficiency. I made sure that all plugins either make the hwsync thing themselves or rely on the slave to do that for them, but not both. If you find an error and/or spot a case of a double-hwsync in a plugin chain, please complain. One known case of double-hwsync is the following pattern: an application calls snd_pcm_rewindable(), thinks about it, and then calls snd_pcm_rewind(). Which, due to PATCH 2/9, calls the rewindable callback again, resulting in the second hwsync. I don't know which way out is best: either ignore, or revert the intention of PATCH 2/9, or revert the whole PATCH 8/9 and replace it with a documentation change. OTOH, I made a mistake of not adding David Henningsson to the CC list during the initial submission. If PulseAudio would need to synchronize hardware pointers even after conversion to snd_pcm_rewindable() for some other reason, then the need for PATCH 8/9 is not that obvious, and maybe it should be reverted and replaced with a documentation fix. -- Alexander E. Patrakov