On Thu, 10 May 2012, Clemens Ladisch wrote:
> Alan Stern wrote:
> > Can you explain briefly how the snd-usb-audio driver decides how many
> > URBs to submit and how many isochronous packets to use in each URB?
>
> The default is to have 8 ms per URB (which can be adjusted with the
> nrpacks module parameter) and to have enough URBs for a total queue
> length of 24 ms.
>
> These values are adjusted down if the application has configured the
> period size (which is the application-visible interrupt interval) to be
> smaller.
Looking through the code in data_ep_set_params(), I'm a little puzzled.
Here's the confusing part, with questions intermingled:
/* decide how many packets to be used */
if (is_playback && !snd_usb_endpoint_implicit_feedback_sink(ep)) {
Why is this case (audio-out with an explicit feedback endpoint) treated
differently? Just because the packet size can shrink?
unsigned int minsize, maxpacks;
/* determine how small a packet can be */
minsize = (ep->freqn >> (16 - ep->datainterval))
* (frame_bits >> 3);
/* with sync from device, assume it can be 12% lower */
if (sync_ep)
minsize -= minsize >> 3;
minsize = max(minsize, 1u);
Instead of 1u, shouldn't this be (frame_bits >> 3)? I assume a packet
can't contain a fraction of a sample.
total_packs = (period_bytes + minsize - 1) / minsize;
Here total_packs ends up being the number of packets that will fit into
period_bytes (rounded up).
/* we need at least two URBs for queueing */
if (total_packs < 2) {
total_packs = 2;
} else {
/* and we don't want too long a queue either */
maxpacks = max(MAX_QUEUE * packs_per_ms, urb_packs * 2);
total_packs = min(total_packs, maxpacks);
}
} else {
while (urb_packs > 1 && urb_packs * maxsize >= period_bytes)
urb_packs >>= 1;
Is there some reason for doing a right shift each iteration? Why not
just compute temp = max(1, (period_bytes - 1) / maxsize) and then set
urb_packs to min(urb_packs, temp), with no loop?
total_packs = MAX_URBS * urb_packs;
Here total_packs in the number of packets used by all URBs, where each
URB must fit into period_bytes.
}
Why is total_packs so different in the two branches? You mentioned
above that period_bytes reflects the application-visible interrupt
interval, which would correspond to the size of a single URB.
Therefore, in the first branch of the "if" statement, it looks like the
value you compute really should be assigned to urb_packs, not
total_packs.
Maybe I have misunderstood what period_bytes is supposed to be. Still,
shouldn't it mean the same thing in both branches of the "if"
statement?
Alan Stern
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
