| Felipe Balbi | 4dc64e5 | 2011-08-19 18:10:58 +0300 | [diff] [blame^] | 1 |  | 
 | 2 |  TODO | 
 | 3 | ~~~~~~ | 
 | 4 | Please pick something while reading :) | 
 | 5 |  | 
 | 6 | - Implement streaming support for BULK endpoints | 
 | 7 |   Tatyana's patch "usb: Add streams support to the gadget framework" | 
 | 8 |   introduces streaming support for the gadget driver. | 
 | 9 |   Every usb_request has new field called stream_id which holds its id. | 
 | 10 |   Every usb_ep has a field num_supported_strms which describes the max | 
 | 11 |   number of streams supported (for this ep). | 
 | 12 |   UAS is AFAIK the only gadget with streaming support. | 
 | 13 |  | 
 | 14 | - Convert interrupt handler to per-ep-thread-irq | 
 | 15 |  | 
 | 16 |   As it turns out some DWC3-commands ~1ms to complete. Currently we spin | 
 | 17 |   until the command completes which is bad. | 
 | 18 |  | 
 | 19 |   Implementation idea: | 
 | 20 |   - dwc core implements a demultiplexing irq chip for interrupts per | 
 | 21 |     endpoint. The interrupt numbers are allocated during probe and belong | 
 | 22 |     to the device. If MSI provides per-endpoint interrupt this dummy | 
 | 23 |     interrupt chip can be replaced with "real" interrupts. | 
 | 24 |   - interrupts are requested / allocated on usb_ep_enable() and removed on | 
 | 25 |     usb_ep_disable(). Worst case are 32 interrupts, the lower limit is two | 
 | 26 |     for ep0/1. | 
 | 27 |   - dwc3_send_gadget_ep_cmd() will sleep in wait_for_completion_timeout() | 
 | 28 |     until the command completes. | 
 | 29 |   - the interrupt handler is split into the following pieces: | 
 | 30 |     - primary handler of the device | 
 | 31 |       goes through every event and calls generic_handle_irq() for event | 
 | 32 |       it. On return from generic_handle_irq() in acknowledges the event | 
 | 33 |       counter so interrupt goes away (eventually). | 
 | 34 |  | 
 | 35 |     - threaded handler of the device | 
 | 36 |       none | 
 | 37 |  | 
 | 38 |     - primary handler of the EP-interrupt | 
 | 39 |       reads the event and tries to process it. Everything that requries | 
 | 40 |       sleeping is handed over to the Thread. The event is saved in an | 
 | 41 |       per-endpoint data-structure. | 
 | 42 |       We probably have to pay attention not to process events once we | 
 | 43 |       handed something to thread so we don't process event X prio Y | 
 | 44 |       where X > Y. | 
 | 45 |  | 
 | 46 |     - threaded handler of the EP-interrupt | 
 | 47 |       handles the remaining EP work which might sleep such as waiting | 
 | 48 |       for command completion. | 
 | 49 |  | 
 | 50 |   Latency: | 
 | 51 |    There should be no increase in latency since the interrupt-thread has a | 
 | 52 |    high priority and will be run before an average task in user land | 
 | 53 |    (except the user changed priorities). |