On Wed, Nov 10, 2010 at 8:35 PM, Jaroslav Kysela <perex@xxxxxxxx> wrote:
> On Wed, 10 Nov 2010, Manu Abraham wrote:
>
>> On Wed, Nov 10, 2010 at 5:00 PM, Jaroslav Kysela <perex@xxxxxxxx> wrote:
>>>
>>> On Wed, 10 Nov 2010, Manu Abraham wrote:
>>>
>>>> Ok, so period should be the interval at which an IRQ is expected.
>>>> ie, if i set a buffer size for 40mS, the period should be 40, I guess.
>>>>
>>>> That said, what should be a recommended value for period ? If hardware
>>>> needs to generate an IRQ at the end of each period ?
>>>
>>> It's good to allow the lowledevel driver to pass all possible period
>>> sizes
>>> to the user space. The application does decision how many wakeups
>>> expects.
>>
>>
>> Ok, I adapted it to look thus,
>>
>> static struct snd_pcm_hardware saa7231_capture_info = {
>> .info = (SNDRV_PCM_INFO_MMAP |
>> SNDRV_PCM_INFO_INTERLEAVED |
>> SNDRV_PCM_INFO_BLOCK_TRANSFER |
>> SNDRV_PCM_INFO_MMAP_VALID),
>>
>> .formats = (SNDRV_PCM_FMTBIT_S16_LE |
>> SNDRV_PCM_FMTBIT_S16_BE |
>> SNDRV_PCM_FMTBIT_S8 |
>> SNDRV_PCM_FMTBIT_U8 |
>> SNDRV_PCM_FMTBIT_U16_LE |
>> SNDRV_PCM_FMTBIT_U16_BE),
>>
>> .rates = SNDRV_PCM_RATE_32000,
>>
>> .rate_min = 32000,
>> .rate_max = 48000,
>>
>> .channels_min = 2,
>> .channels_max = 2,
>>
>> .buffer_bytes_max = 512 * 4096,
>> #if 0
>> .period_bytes_min = 192,
>> .period_bytes_max = 1536,
>> .periods_min = 2,
>> .periods_max = 1024,
>> #endif
>> .period_bytes_min = 1920, /* 10mS @ 48khz */
>> .period_bytes_max = 768000, /* 4S @ 48kHz */
>>
>> .periods_min = 10, /* 10mS */
>> .periods_max = 4000000, /* 4S */
>
> ??? 10 (period_bytes_min) * 10 (periods_min) = 100mS . Note that
> period_bytes_* limits periods size, but periods_* count of periods per the
> whole audio DMA ring buffer. Define smallest number of periods (it's usually
> value one or two for periods_min).
Ok.
>> };
>>
>> /*
>> * 10mS Buffer lengths
>> * @48kHz 1920 bytes
>> * @44.1kHz 1764 bytes
>> * @32kHz 1280 bytes
>> *
>> *
>> * period min = 10mS @48k: 1920 bytes @44.1k: 1764 bytes @32k: 1280
>> bytes
>> * period max = 1S @48k:192000 bytes @44.1k:176400 bytes @32k:128000
>> bytes
>> * period = 4S @48k:768000 bytes @44.1k:705600 bytes @32k:512000
>> bytes
>> */
>> static int saa7231_capture_open(struct snd_pcm_substream *pcm)
>> {
>> struct saa7231_audio *audio = snd_pcm_substream_chip(pcm);
>> struct snd_pcm_runtime *rt = pcm->runtime;
>> struct saa7231_dev *saa7231 = audio->saa7231;
>> int err;
>>
>> dprintk(SAA7231_DEBUG, 1, "()");
>> rt->hw = saa7231_capture_info;
>> snd_pcm_hw_constraint_list(rt, 0, SNDRV_PCM_HW_PARAM_RATE,
>> &hw_constraint_rates);
>> err = snd_pcm_hw_constraint_integer(rt,
>> SNDRV_PCM_HW_PARAM_PERIODS);
>> if (err < 0) {
>> dprintk(SAA7231_ERROR, 1, "snd_pcm_hw_constraint_integer()
>> failed.
>> ret=%d", err);
>> return err;
>> }
>
>> err = snd_pcm_hw_constraint_minmax(rt,
>> SNDRV_PCM_HW_PARAM_PERIOD_SIZE, 10, 4000000);
>> if (err < 0) {
>> dprintk(SAA7231_ERROR, 1, "snd_pcm_hw_constraint_minmax()
>> failed.
>> ret=%d", err);
>> return err;
>> }
>> err = snd_pcm_hw_constraint_minmax(rt,
>> SNDRV_PCM_HW_PARAM_PERIOD_BYTES, 1920, 768000);
>> if (err < 0) {
>> dprintk(SAA7231_ERROR, 1, "snd_pcm_hw_constraint_minmax()
>> failed.
>> ret=%d", err);
>> return err;
>> }
>
> Remove these minmax contrains. You already do limiting in hw->ops.
Ok.
>> return 0;
>> }
>>
>> /*
>> * saa7231_hw_params()
>> *
>> * This callback is called when the hardware parameter (hw_params) is setup
>> * by the application, ie., once when the buffer size, the period size,
>> * the format etc are defined for the PCM substream.
>> */
>> static int saa7231_hw_params(struct snd_pcm_substream *pcm, struct
>> snd_pcm_hw_params *params)
>> {
>> struct saa7231_audio *audio = snd_pcm_substream_chip(pcm);
>> struct saa7231_dev *saa7231 = audio->saa7231;
>> struct snd_pcm_runtime *rt = pcm->runtime;
>>
>> struct saa7231_dmabuf *buffer;
>> struct saa7231_stream *stream;
>> struct saa7231_dmabuf *dmabuf;
>>
>> struct page **ptable;
>>
>> void *mem;
>> void *dma_area;
>>
>> int i, j, pages, bytes, periods, bufsiz, pt_size;
>>
>> dprintk(SAA7231_DEBUG, 1, "DEBUG: ()");
>> periods = params_periods(params);
>> bytes = params_period_bytes(params);
>>
>> bufsiz = periods * bytes;
>> dprintk(SAA7231_DEBUG, 1, "bufsiz=%d periods=%d bytes=%d",
>> bufsiz,
>> periods,
>> bytes);
>>
>> pages = bufsiz / PAGE_SIZE;
>
> This is wrong. It should be 'pages = (bufsize + PAGE_SIZE - 1) / PAGE_SIZE'.
Ok.
>> /* enable stream */
>> /* initializing 8 buffer with "pages" pages each .. */
>> stream = saa7231_stream_init(saa7231, AUDIO_CAPTURE, ADAPTER_INT,
>> 0, pages);
>> if (!stream) {
>> dprintk(SAA7231_ERROR, 1, "ERROR: Registering stream");
>> return -ENOMEM;
>> }
>> audio->stream = stream;
>> buffer = stream->dmabuf;
>> saa7231_add_irqevent(saa7231, 43, SAA7231_EDGE_RISING,
>> saa7231_audio_evhandler, "AS2D_AVIS");
>> dprintk(SAA7231_DEBUG, 1, "Mapping %d buffers with %d pages each",
>> XS2D_BUFFERS, pages);
>
> Unfortunately, I don't understand the role of XS2D_BUFFERS. The ALSA bufsize
> is the whole DMA area (you should use params_buffer_bytes() to get this
> value instead of calculating this using periods * period_size).
Ok. The hardware has 8 SG buffers each. each of the buffer heads are
setup on to the MMU.
each of these buffers XS2D_BUFFERS can handle a maximum of 512 pages,
ie page table
for the SG buffer can be a maximum of 4096 bytes long.
> It means: Just allocate number of pages required for buffer_bytes. Don't
> play with periods (except the interrupts).
>
>> /*
>> * For a single DMA buffer:
>> * each page takes a u64 size in a page table
>> * number of entries a single page can hold = PAGE_SIZE / entry
>> size
>> * ie entries.max = PAGE_SIZE / 8 => 4096/8 = 512
>> * Now, we have entries.req = 7 ("pages") per buffer
>> *
>> * For all in XS2D_BUFFERS:
>> * page table should be large enough to hold all the pages in each
>> DMA buffer
>> * total number of pages = pages * XS2D_BUFFERS
>> * max buffers that we need to consider = 512 * 8, this needs 8
>> pages
>> for a page table
>> *
>> * On a general note, we can calculate pages for page table as
>> * page_table_size = total_pages / 512, with a minimum of a single
>> page
>> */
>>
>> #define MAX_ENTRIES_PER_PAGE (PAGE_SIZE / 8)
>>
>> pt_size = (pages * XS2D_BUFFERS) / MAX_ENTRIES_PER_PAGE;
>> /* minimum 1 page required for the table */
>> if (pt_size < 1)
>> pt_size = 1;
>>
>> dprintk(SAA7231_DEBUG, 1, "Page Table array size=%d", pt_size);
>> ptable = kzalloc((sizeof (struct page) * pt_size), GFP_KERNEL);
>> if (!ptable) {
>> dprintk(SAA7231_ERROR, 1, "ERROR: No memory to allocate
>> virtual map");
>> return -ENOMEM;
>> }
>> audio->ptable = ptable;
>>
>> for (i = 0; i < XS2D_BUFFERS; i ++) {
>> dmabuf = &buffer[i];
>> mem = dmabuf->vmalloc;
>> for (j = 0; j < pages; j++)
>> ptable[j] = virt_to_page(mem);
>> }
>
> This looks broken. ptable is overwritten for each XS2D_BUFFERS iteration.
True.
>> #if 0
>> dma_area = vmap(ptable, (pages * XS2D_BUFFERS), VM_MAP,
>> PAGE_KERNEL);
>> rt->dma_area = dma_area;
>> rt->dma_bytes = XS2D_BUFFERS * pages * SAA7231_PAGE_SIZE;
>> rt->dma_addr = 0;
>> #endif
>> return 0;
>> }
>
> I think that the second argument is number of linux mem pages. Your value
> seems strange.
Uh, oh. I had assumed that each of the 8 XS2D_Buffers would contain "pages"
each and hence..
Regards,
Manu
_______________________________________________
Alsa-devel mailing list
Alsa-devel@xxxxxxxxxxxxxxxx
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
