[wdmaudiodev] Re: Asynchronous sink feedback question

  • From: Chris Fryer <chris@xxxxxxxxxxxxxxxxxxxxxxxxxxx>
  • To: wdmaudiodev@xxxxxxxxxxxxx
  • Date: Wed, 21 Apr 2021 17:04:41 +0100

No problem, I appreciate the co-operation.

bInterval is 4 for the feedback and data EPs.

Yes I found the implicit FB attributes from this Apple tech note. It made no difference to Windows operation.
https://developer.apple.com/library/archive/technotes/tn2274/_index.html#//apple_ref/doc/uid/DTS40010116-CH1-TNTAG4_3 <https://developer.apple.com/library/archive/technotes/tn2274/_index.html#//apple_ref/doc/uid/DTS40010116-CH1-TNTAG4_3>

On 21/04/2021 16:48, Wade Dawson wrote:

Well, Chris, I did start off telling you I was confused by that… Sorry for the incorrect info, looks like you’re right about it.  Also, based on this:

https://docs.microsoft.com/en-us/windows-hardware/drivers/audio/usb-2-0-audio-drivers <https://docs.microsoft.com/en-us/windows-hardware/drivers/audio/usb-2-0-audio-drivers>

the MS UAC2 driver doesn’t support implicit FB. (Though our devices still “work” with the class driver and have no FB EPs)


  * What are you setting bInterval to on the feedback EP?  I would
    guess 0 (1uFrame)?  Have you tried setting it to 3 (8uFrames)?
      The driver may just be doing 3.  (Windows doesn’t allow iso
    transfers of  < 8 uFrames currently so updating the fraction every
    uFrame is pointless since you can’t respond with a granularity of
    < 8 uFrames.)
  * When you removed the EP, did you also set bmAttributes 5..4 =
    b’10’ to indicate implicit FB?

FWIW, we’ve long stopped implementing iso feedback EPs in our devices and rely solely on implicit sync or async, though we do have separate in and out interfaces.

*From: *wdmaudiodev-bounce@xxxxxxxxxxxxx <wdmaudiodev-bounce@xxxxxxxxxxxxx>
*Date: *Wednesday, April 21, 2021 at 11:03 AM
*To: *wdmaudiodev@xxxxxxxxxxxxx <wdmaudiodev@xxxxxxxxxxxxx>
*Subject: *[wdmaudiodev] Re: Asynchronous sink feedback question

I don't know yet but I've just seen that the linux kernel is about to implement the same interpretation as I have for feedback in HS mode.

+static void u_audio_set_fback_frequency(enum usb_device_speed speed,
+                               unsigned int freq, void *buf)
+ u32 ff = 0;
+ if (speed == USB_SPEED_FULL) {
+        /*
+        * Full-speed feedback endpoints report frequency
+        * in samples/microframe
+        * Format is encoded in Q10.10 left-justified in the 24 bits,
+        * so that it has a Q10.14 format.
+        */
+        ff = DIV_ROUND_UP((freq << 14), 1000);
+ } else {
+        /*
+        * High-speed feedback endpoints report frequency
+        * in samples/microframe.
+        * Format is encoded in Q12.13 fitted into four bytes so that
+        * the binary point is located between the second and the third
+        * byte fromat (that is Q16.16)
+        *
+        * Prevent integer overflow by calculating in Q12.13 fromat and
+        * then shifting to Q16.16
+        */
+        ff = DIV_ROUND_UP((freq << 13), (8*1000)) << 3;
+ }
+ *(__le32 *)buf = cpu_to_le32(ff);


On 21/04/2021 15:51, Wade Dawson wrote:

    Hi Chris.  What does MacOs think about that change?

    *From: *wdmaudiodev-bounce@xxxxxxxxxxxxx
    *Date: *Wednesday, April 21, 2021 at 5:23 AM
    *To: *wdmaudiodev@xxxxxxxxxxxxx <mailto:wdmaudiodev@xxxxxxxxxxxxx>
    <wdmaudiodev@xxxxxxxxxxxxx> <mailto:wdmaudiodev@xxxxxxxxxxxxx>
    *Subject: *[wdmaudiodev] Re: Asynchronous sink feedback question

    Thanks Wade but I must be missing something because 5.12 seems
    fairly clear to me that Ff on a high speed endpoint should be
    samples per microframe?

    The standard says (5.12 Special Considerations for Isochronous
    /"Note: The examples in this section describe USB for an example
    involving full-speed endpoints. The
    general example details are also appropriate for high-speed
    endpoints when corresponding changes are
    made; for example, frame replaced with microframe, 1 ms replaced
    with 125 μ s, rate adjustments made
    between full-speed and high-speed, etc."

    /Then every reference to frame in the Feedback section is written
    as (micro)frame
    /"F//f is expressed in number of samples per (micro)frame. The
    F//f value consists of an integer part that
    represents the (integer) number of samples per (micro)frame and a
    fractional part that represents the
    “fraction” of a sample that would be needed to match the sampling
    frequency F//s to a resolution of 1 Hz or

    I read the section on implicit feedback for asynchronous sinks
    ( <> Implicit Feedback):
    /"One or more asynchronous sink endpoints are accompanied by an
    asynchronous source endpoint. The
    data rate on the source endpoint can be used as implicit feedback
    information to adjust the data rate on
    the sink endpoint(s)."/

    I have an asynchronous sink endpoint and an asynchronous source
    endpoint in the same interface so I  simply removed the explicit
    feedback endpoint:
    [1]0004.0108::04/21/2021-08:35:06.229 [USBAudio2](0x0101): AS
    interface with bInterfaceNumber=0x01 bAlternateSetting=0x01 uses
    asynch synchronization but does not implement a feedback endpoint
    [1]0004.0108::04/21/2021-08:35:06.229 [USBAudio2](0x01): Create
    failed NTSTATUS=C0440025

    That suggests usbaudio2.sys does not look to implement implicit
    feedback. Is that right?


    On 20/04/2021 17:42, Wade Dawson wrote:

        I’ve been confused by that as well.  The fraction is relative
        to frame (1ms) not microframe (125uS), so 48 is correct.  I’m
        fairly certain the Windows UAC2 driver will do implicit
        feedback, no?  I.e., no feedback EP?  The osx driver does for


        *From: *wdmaudiodev-bounce@xxxxxxxxxxxxx
        *Date: *Tuesday, April 20, 2021 at 10:09 AM
        *To: *wdmaudiodev@xxxxxxxxxxxxx
        <mailto:wdmaudiodev@xxxxxxxxxxxxx> <wdmaudiodev@xxxxxxxxxxxxx>
        *Subject: *[wdmaudiodev] Asynchronous sink feedback question

        My understanding of the USB 2.0 spec (section
        <http://sdpha2.ucsd.edu/Lab_Equip_Manuals/usb_20.pdf> for
        asynchronous sinks was that the sink in high speed mode should
        provide feedback of Ff formatted as a 16.16 number where Ff is
        the number of samples per _micro_ frame (125us period).

        A 48kHz sink will then return values close to 6.0 (0x00060000
        in 16.16 format).

        We implemented this and tested some time ago on Windows 10
        1903 (or similar). Testing with Windows 10 20 2H we're seeing
        audio glitches.

        Event logs tell us that the driver is ignoring the feedback
        values because they're out of range:
        [0]0004.01C4::04/20/2021-13:36:16.321 [USBAudio2]feedback
        value 0x00062000 is out of range [0x002f0000,0x00310000],
        ignoring value
        [0]0004.01C4::04/20/2021-13:36:16.321 [USBAudio2]feedback
        value 0x0005e000 is out of range [0x002f0000,0x00310000],
        ignoring value
        [0]0004.01C4::04/20/2021-13:36:16.321 [USBAudio2]feedback
        value 0x00062000 is out of range [0x002f0000,0x00310000],
        ignoring value
        [0]0004.01C4::04/20/2021-13:36:16.321 [USBAudio2]feedback
        value 0x00062000 is out of range [0x002f0000,0x00310000],
        ignoring value

        It appears the driver wants a value close to 48.0 which would
        be samples per frame instead of samples per micro frame.

        Have I misinterpreted the spec?

        I didn't investigate to this detail on the previous release of
        Windows. I can see this is a new release of usbaudio2.sys. Is
        this a new check?

        I have tried reporting samples per frame and the glitches stop
        on Windows 10 but MacOS then glitches.

        Any hints gratefully received!

        ChrisHi Chris.

Other related posts: