From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner+willy=40w.ods.org-S1751031AbWDXR5W@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
	id S1751031AbWDXR5W (ORCPT <rfc822;willy@w.ods.org>);
	Mon, 24 Apr 2006 13:57:22 -0400
Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751034AbWDXR5W
	(ORCPT <rfc822;linux-kernel-outgoing>);
	Mon, 24 Apr 2006 13:57:22 -0400
Received: from mail.suse.de ([195.135.220.2]:64189 "EHLO mx1.suse.de")
	by vger.kernel.org with ESMTP id S1751026AbWDXR5W (ORCPT
	<rfc822;linux-kernel@vger.kernel.org>);
	Mon, 24 Apr 2006 13:57:22 -0400
Date: Mon, 24 Apr 2006 19:57:15 +0200
Message-ID: <s5hejzm26x0.wl%tiwai@suse.de>
From: Takashi Iwai <tiwai@suse.de>
To: Adrian McMenamin <adrian@mcmen.demon.co.uk>
Cc: Alsa-devel <alsa-devel@lists.sourceforge.net>,
       linux-sh <linuxsh-dev@lists.sourceforge.net>,
       LKML <linux-kernel@vger.kernel.org>, Lee Revell <rlrevell@joe-job.com>,
       Paul Mundt <lethal@linux-sh.org>
Subject: Re: [linuxsh-dev] Re: [Alsa-devel] [PATCH] Add Dreamcast AICA	driver to alsa-driver
In-Reply-To: <1145900695.9243.4.camel@localhost.localdomain>
References: <1145831786.9242.38.camel@localhost.localdomain>
	<1145832383.9242.41.camel@localhost.localdomain>
	<s5hacab44bw.wl%tiwai@suse.de>
	<1145900695.9243.4.camel@localhost.localdomain>
User-Agent: Wanderlust/2.12.0 (Your Wildest Dreams) SEMI/1.14.6 (Maruoka)
 FLIM/1.14.7 (=?ISO-8859-4?Q?Sanj=F2?=) APEL/10.6 MULE XEmacs/21.5 (beta25)
 (eggplant) (+CVS-20060326) (i386-suse-linux)
MIME-Version: 1.0 (generated by SEMI 1.14.6 - "Maruoka")
Content-Type: text/plain; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
X-Mailing-List: linux-kernel@vger.kernel.org

At Mon, 24 Apr 2006 18:44:55 +0100,
Adrian McMenamin wrote:
> 
> > 
> > > +static int stereo_buffer_transfer(struct snd_pcm_substream
> > > +				  *substream, int buffer_size, int period)
> > > +{
> > 
> > I feel this transfer-and-wait could be done more efficiently using
> > workq than doing it in timer callback.  The trigger(start) and
> > aica_period_elapsed() calls queue_work() at each time.
> > 
> > The most work of spu_begin_dma() should be put in the workq, too.
> > 
> 
> heh. If you remember, a couple of months ago I had this in a kernel
> thread - ie much the same - and you or Lee said I should absolutely not
> use that mechanism :)

Well, a source code tell you better than hundreds words :)


> > You can write a single function for both mono and two-channel
> > streams.
> > 
> 
> How can I do that given the dma has to be serialised?

The second dma_xfer() can be in if (channels > 1) block (in addition
to different transfer bytes).  Something like below:

static int buffer_transfer(struct snd_pcm_substream*substream,
		int buffer_size, int period)
{
	int transferred;
	int dma_countout;
	struct snd_pcm_runtime *runtime;
	int period_offset;
	long dma_flags;

	period_offset = period;
	period_offset %= (AICA_PERIOD_NUMBER / runtime->channels);
	runtime = substream->runtime;
	/* transfer left and then right */
	dma_flags = claim_dma_lock();
	dma_countout = 0;
	dma_xfer(0,
		 runtime->dma_area + (AICA_PERIOD_SIZE * period_offset),
		 AICA_CHANNEL0_OFFSET +
		 (AICA_PERIOD_SIZE * period_offset), buffer_size / 2, 5);
	/* wait for completion */
	do {
		udelay(5);
		transferred = get_dma_residue(0);
		dma_countout++;
		if (dma_countout > 0x10000)
			break;	/* Approx 1/3 sec timeout in case of hardware failure */
	}
	while (transferred < buffer_size / 2);
	if (runtime->channels > 1) {
		dma_xfer(0,
			 AICA_BUFFER_SIZE / 2 + runtime->dma_area +
			 (AICA_PERIOD_SIZE * period_offset),
			 AICA_CHANNEL1_OFFSET +
			 (AICA_PERIOD_SIZE * period_offset), buffer_size / 2, 5);
		/* have to wait again */
		dma_countout = 0;
		do {
			udelay(5);
			transferred = get_dma_residue(0);
			dma_countout++;
			if (dma_countout > 0x10000)
				break;
		}
		while (transferred < buffer_size / 2);
	}
	release_dma_lock(dma_flags);
	return 0;
}


Takashi