[PATCH v9 4/5] xhci: mediatek: support MTK xHCI host controller
chunfeng.yun at mediatek.com
Mon Oct 19 23:29:14 PDT 2015
On Mon, 2015-10-19 at 14:25 +0300, Mathias Nyman wrote:
> >> So basically we are trying to use as many microframes as possible with as few packets
> >> per microframe as possible.
> >> Did I understand this correctly?
> > Yes, you are right.
> >> How will devices react if they expect to get 16 packets every 16th microframe,
> >> but they get one packet every microframe instead?
> > I think that the synchronous endpoint must specify its period by
> > bInterval, but can't specify how data should be transfered during the
> > period by the host, and it just only receives data passively. So the
> > device can receive data correctly in the case(bInterval is 5).
> > quote from usb3_r1.0 section4.4.8 Isochronous Transfers:
> > "The host can request data from the device or send data to the device at
> > any time during the service interval for a particular endpoint on that
> > device"
> As I understand the 4.4.8 section it just means the device can't assume a fixed
> time interval between transfers, meaning that the host can use the last microframe
> in one esit and the first microframe in the next esit, but still only use 1 microframe
> per esit.
> Section 22.214.171.124 describes how a 11 packet isoc transfer is allowed to be split
> to 1 burst of 11 packets, 2 burst (8 + 3), 3 burst (4+4+3) 6 bursts (2+2+2+2+2+1) or
> 11 bursts of 1. These are however all within the same microframe. Splitting the
> transfer into several microframes in a esit kind of makes the whole interval concept pointless.
It doesn't say that the packets should be transfered within the same
microframe (bus interval), as I understand it means service interval;
The direct prove resides in figure 8-56/8-57.
1. BI, bus interval, a 125 us period that establishes the internal
boundary of service interval, aka uframe;
2. SSI, Support Smart Isochronous;
3. DBI, Data in this Bus Interval is done;
4. NBI, Numbers of Bus Interval;
As the figure shows, the service interval = 8 BI, that host distribute 2
packets @1st uframe, keep U1/U2 state for the next 3uframe, then
transmit 4 packets @4th uframe, and the remaining 3 packet in the last
Please notice that this just is an example illustrated by spec, but we
can derive the conclusion that the distribution of packet in a service
interval is completely decided by host, and can split isoc transfers
across multiple uframes.
PS: as you can see, MTK implementation of schedule algorithms is an
implementation of Smart Isochronous of which the smart side resides in
More information about the linux-arm-kernel