[PATCH 1/1] usb: gadget: f_uac2: finalize wMaxPacketSize according to bandwidth

[PATCH 1/1] usb: gadget: f_uac2: finalize wMaxPacketSize according to bandwidth

[Peter Chen]

According to USB Audio Device 2.0 Spec, Ch4.10.1.1:
wMaxPacketSize is defined as follows:
Maximum packet size this endpoint is capable of sending or receiving
when this configuration is selected.
This is determined by the audio bandwidth constraints of the endpoint.

In current code, the wMaxPacketSize is defined as the maximum packet size
for ISO endpoint, and it will let the host reserve much more space than
it really needs, so that we can't let more endpoints work together at
one frame.

We find this issue when we try to let 4 f_uac2 gadgets work together [1].

[1]http://www.spinics.net/lists/linux-usb/msg123478.html

Cc: <stable@vger.kernel.org> #v3.18+
Cc: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Peter Chen <peter.chen@freescale.com>
---
 drivers/usb/gadget/function/f_uac2.c | 24 ++++++++++++++++++++++++
 1 file changed, 24 insertions(+)

diff --git a/drivers/usb/gadget/function/f_uac2.c b/drivers/usb/gadget/function/f_uac2.c
index 6d3eb8b..0ed6f0e 100644
--- a/drivers/usb/gadget/function/f_uac2.c
+++ b/drivers/usb/gadget/function/f_uac2.c
@@ -987,6 +987,7 @@ afunc_bind(struct usb_configuration *cfg, struct usb_function *fn)
 	struct f_uac2_opts *uac2_opts;
 	struct usb_string *us;
 	int ret;
+	u16 c_max_packet_size, p_max_packet_size;
 
 	uac2_opts = container_of(fn->fi, struct f_uac2_opts, func_inst);
 
@@ -1070,6 +1071,29 @@ afunc_bind(struct usb_configuration *cfg, struct usb_function *fn)
 	uac2->p_prm.uac2 = uac2;
 	uac2->c_prm.uac2 = uac2;
 
+	/* Calculate wMaxPacketSize according to audio bandwidth */
+	c_max_packet_size = uac2_opts->c_chmask * uac2_opts->c_ssize
+		* uac2_opts->c_srate / 1000;
+	p_max_packet_size = uac2_opts->p_chmask * uac2_opts->p_ssize
+		* uac2_opts->p_srate / 1000;
+	if ((c_max_packet_size > fs_epout_desc.wMaxPacketSize) ||
+		(p_max_packet_size > fs_epin_desc.wMaxPacketSize)) {
+		dev_err(dev, "parameters are incorrect\n");
+		goto err;
+	}
+	/*
+	 * Keep max packet size is multiplier of 4, and
+	 * a little larger than bandwidth.
+	 * Eg, for frame rate 44100, 1 channel, and 16 bits data
+	 * we need to reserve more than 44 * 2 bytes.
+	 */
+	c_max_packet_size +=  (4 - (c_max_packet_size % 4));
+	fs_epin_desc.wMaxPacketSize = min(cpu_to_le16(c_max_packet_size),
+		fs_epin_desc.wMaxPacketSize);
+	p_max_packet_size +=  (4 - (p_max_packet_size % 4));
+	fs_epout_desc.wMaxPacketSize = min(cpu_to_le16(p_max_packet_size),
+		fs_epout_desc.wMaxPacketSize);
+
 	hs_epout_desc.bEndpointAddress = fs_epout_desc.bEndpointAddress;
 	hs_epout_desc.wMaxPacketSize = fs_epout_desc.wMaxPacketSize;
 	hs_epin_desc.bEndpointAddress = fs_epin_desc.bEndpointAddress;
-- 
1.9.1

Re: [PATCH 1/1] usb: gadget: f_uac2: finalize wMaxPacketSize according to bandwidth

[Daniel Mack]

Hi,

On 07/14/2015 04:29 AM, Peter Chen wrote:
> According to USB Audio Device 2.0 Spec, Ch4.10.1.1:
> wMaxPacketSize is defined as follows:
> Maximum packet size this endpoint is capable of sending or receiving
> when this configuration is selected.
> This is determined by the audio bandwidth constraints of the endpoint.
> 
> In current code, the wMaxPacketSize is defined as the maximum packet size
> for ISO endpoint, and it will let the host reserve much more space than
> it really needs, so that we can't let more endpoints work together at
> one frame.

...

> diff --git a/drivers/usb/gadget/function/f_uac2.c b/drivers/usb/gadget/function/f_uac2.c
> index 6d3eb8b..0ed6f0e 100644
> --- a/drivers/usb/gadget/function/f_uac2.c
> +++ b/drivers/usb/gadget/function/f_uac2.c
> @@ -987,6 +987,7 @@ afunc_bind(struct usb_configuration *cfg, struct usb_function *fn)
>  	struct f_uac2_opts *uac2_opts;
>  	struct usb_string *us;
>  	int ret;
> +	u16 c_max_packet_size, p_max_packet_size;
>  
>  	uac2_opts = container_of(fn->fi, struct f_uac2_opts, func_inst);
>  
> @@ -1070,6 +1071,29 @@ afunc_bind(struct usb_configuration *cfg, struct usb_function *fn)
>  	uac2->p_prm.uac2 = uac2;
>  	uac2->c_prm.uac2 = uac2;
>  
> +	/* Calculate wMaxPacketSize according to audio bandwidth */
> +	c_max_packet_size = uac2_opts->c_chmask * uac2_opts->c_ssize
> +		* uac2_opts->c_srate / 1000;
> +	p_max_packet_size = uac2_opts->p_chmask * uac2_opts->p_ssize
> +		* uac2_opts->p_srate / 1000;

For high-speed endpoints, we in fact have 8 microframes per frame, and
the factor also depends on the endpoint's polling interval. See what
afunc_set_alt() does here.

Did you test this in HS or FS setups or both?

> +	if ((c_max_packet_size > fs_epout_desc.wMaxPacketSize) ||
> +		(p_max_packet_size > fs_epin_desc.wMaxPacketSize)) {
> +		dev_err(dev, "parameters are incorrect\n");
> +		goto err;
> +	}
> +	/*
> +	 * Keep max packet size is multiplier of 4, and
> +	 * a little larger than bandwidth.
> +	 * Eg, for frame rate 44100, 1 channel, and 16 bits data
> +	 * we need to reserve more than 44 * 2 bytes.
> +	 */
> +	c_max_packet_size +=  (4 - (c_max_packet_size % 4));

I guess DIV_ROUND_UP() allows you to account for the packets that are
sent when the devision residue accumulator overflows. See
agdev_iso_complete().

> +	fs_epin_desc.wMaxPacketSize = min(cpu_to_le16(c_max_packet_size),
> +		fs_epin_desc.wMaxPacketSize);
> +	p_max_packet_size +=  (4 - (p_max_packet_size % 4));
> +	fs_epout_desc.wMaxPacketSize = min(cpu_to_le16(p_max_packet_size),
> +		fs_epout_desc.wMaxPacketSize);
> +
>  	hs_epout_desc.bEndpointAddress = fs_epout_desc.bEndpointAddress;
>  	hs_epout_desc.wMaxPacketSize = fs_epout_desc.wMaxPacketSize;

We probably have to pass different values for the two settings, as they
can't be the same once they contain pre-calculated values.


Thanks,
Daniel

Re: [PATCH 1/1] usb: gadget: f_uac2: finalize wMaxPacketSize according to bandwidth

[Peter Chen]

On Tue, Jul 14, 2015 at 04:04:23PM -0400, Daniel Mack wrote:
> Hi,
> 
> On 07/14/2015 04:29 AM, Peter Chen wrote:
> > According to USB Audio Device 2.0 Spec, Ch4.10.1.1:
> > wMaxPacketSize is defined as follows:
> > Maximum packet size this endpoint is capable of sending or receiving
> > when this configuration is selected.
> > This is determined by the audio bandwidth constraints of the endpoint.
> > 
> > In current code, the wMaxPacketSize is defined as the maximum packet size
> > for ISO endpoint, and it will let the host reserve much more space than
> > it really needs, so that we can't let more endpoints work together at
> > one frame.
> 
> ...
> 
> > diff --git a/drivers/usb/gadget/function/f_uac2.c b/drivers/usb/gadget/function/f_uac2.c
> > index 6d3eb8b..0ed6f0e 100644
> > --- a/drivers/usb/gadget/function/f_uac2.c
> > +++ b/drivers/usb/gadget/function/f_uac2.c
> > @@ -987,6 +987,7 @@ afunc_bind(struct usb_configuration *cfg, struct usb_function *fn)
> >  	struct f_uac2_opts *uac2_opts;
> >  	struct usb_string *us;
> >  	int ret;
> > +	u16 c_max_packet_size, p_max_packet_size;
> >  
> >  	uac2_opts = container_of(fn->fi, struct f_uac2_opts, func_inst);
> >  
> > @@ -1070,6 +1071,29 @@ afunc_bind(struct usb_configuration *cfg, struct usb_function *fn)
> >  	uac2->p_prm.uac2 = uac2;
> >  	uac2->c_prm.uac2 = uac2;
> >  
> > +	/* Calculate wMaxPacketSize according to audio bandwidth */
> > +	c_max_packet_size = uac2_opts->c_chmask * uac2_opts->c_ssize
> > +		* uac2_opts->c_srate / 1000;
> > +	p_max_packet_size = uac2_opts->p_chmask * uac2_opts->p_ssize
> > +		* uac2_opts->p_srate / 1000;
> 
> For high-speed endpoints, we in fact have 8 microframes per frame, and
> the factor also depends on the endpoint's polling interval. See what
> afunc_set_alt() does here.
> 
> Did you test this in HS or FS setups or both?

It only has problem at FS setup.

> 
> > +	if ((c_max_packet_size > fs_epout_desc.wMaxPacketSize) ||
> > +		(p_max_packet_size > fs_epin_desc.wMaxPacketSize)) {
> > +		dev_err(dev, "parameters are incorrect\n");
> > +		goto err;
> > +	}
> > +	/*
> > +	 * Keep max packet size is multiplier of 4, and
> > +	 * a little larger than bandwidth.
> > +	 * Eg, for frame rate 44100, 1 channel, and 16 bits data
> > +	 * we need to reserve more than 44 * 2 bytes.
> > +	 */
> > +	c_max_packet_size +=  (4 - (c_max_packet_size % 4));
> 
> I guess DIV_ROUND_UP() allows you to account for the packets that are
> sent when the devision residue accumulator overflows. See
> agdev_iso_complete().

Thanks, I will check it.

> 
> > +	fs_epin_desc.wMaxPacketSize = min(cpu_to_le16(c_max_packet_size),
> > +		fs_epin_desc.wMaxPacketSize);
> > +	p_max_packet_size +=  (4 - (p_max_packet_size % 4));
> > +	fs_epout_desc.wMaxPacketSize = min(cpu_to_le16(p_max_packet_size),
> > +		fs_epout_desc.wMaxPacketSize);
> > +
> >  	hs_epout_desc.bEndpointAddress = fs_epout_desc.bEndpointAddress;
> >  	hs_epout_desc.wMaxPacketSize = fs_epout_desc.wMaxPacketSize;
> 
> We probably have to pass different values for the two settings, as they
> can't be the same once they contain pre-calculated values.
> 
> 

You mean we may have different wMaxPacketSize value for FS and HS?
Current f_uac2 does not support it, it only one group parameter
for packet size.

-- 

Best Regards,
Peter Chen