Ran out of hardware fifo

[gmikitiuk]

Hi,

Is there a way in tinyusb to extend usb hardware fifo with software anyhow?
I've ran out of available space - dfifo_alloc asserts. I'm currently running USB HS with: 2x CDC, 1xAudio, 1xDFU, 1xMSC. Audio is taking quite a lot of the fifo due to multichannel configuration.

[hathach]

can you uncomment this line https://github.com/hathach/tinyusb/blob/master/src/portable/synopsys/dwc2/dcd_dwc2.c#L208 have LOG=2 and post the allocation detail.

[HiFiPhile]

I think we can add 2 tunables to set Tx/Rx FIFO to 2*MaxPacketSize or not.
Actually Rx FIFO is 2*size and Tx FIFO is not.
Testing #2966 I found set Tx FIFO to 2*size sometimes can improve throughput a lot depending the host.

[gmikitiuk]

I do not use TU_LOG but I've re-reouted the call to my logging library:

TX FIFO 0: allocated 16 words at offset 1008
TX FIFO 2: allocated 228 words at offset 780
TX FIFO 1: allocated 1 words at offset 779
TX FIFO 3: allocated 2 words at offset 777
TX FIFO 4: allocated 128 words at offset 649
TX FIFO 5: allocated 2 words at offset 647
TX FIFO 6: allocated 128 words at offset 519

After these, assert is triggered.

[hathach]

Forgot to ask, what is your configuration descriptor, you can just list the endpoint type and max packet size in order. Are you using both iso in and out or only 1 direction ?

[gmikitiuk]

I do use iso in and out

CFG_TUD_AUDIO_FUNC_1_EP_OUT_SZ_MAX 912U
CFG_TUD_AUDIO_FUNC_1_EP_IN_SZ_MAX 912U

CDC - 512
CDC2 - 512
DFU - 512
MSC - 512

• some things like feedback endpoint, ctrl etc...

I'm quite sure I'm on the limit of the hardware, the real question is if I can do anything about or I have to discard something to make space.

[hathach]

yeah for now, you need to reduce your endpoint usage since your sum exceed 4K buffer. Though there is a way to go around this, we can implement share/dynamic tx fifo for bulk endpoint (they all use fixed 512 byte) used by cdc/cdc2/msc (dfu use control only so it does not consume additional fifo). Especially cdc since they are not always has something to send. The approach would be

• take note of submitted transfer as pending buffer + size
• round robin through pending transfer in tx complete isr by dynamic assigned it to pending endpoint

it is easier said than done, but is still doable if you are willing to put the effort.

[gmikitiuk]

The main problem is - they're all needed and I'm wondering if there is anything I could do to work around it.

Initially I though I could remove DFU descriptor and prepare separate USB descriptor which would be presented whenever there is some kind of flag set on bootup(ie. pressed button). This is not really working well due to Windows not supporting DFU natively, if I present DFU-only descriptor and replace drivers with Zadig, then when I go back to "normal" descriptor everything is wrong because of the replaced driver.

I'm also quite sure audio endpoints must be that big to support 32bit, 32ch configuration (7B * 32ch * 4B) = 896B.

[hathach]

The main problem is - they're all needed and I'm wondering if there is anything I could do to work around it.

The work around is mentioned above, if you could submit the PR I will be happy to review it.