[dwc2] bInterval time for interrupt endpoint not respected

[maximevince]

Operating System

Linux

Board

STM32F7

Firmware

tinyUSB master, USB host on STM32F7 w/ DWC USB IP Core

What happened ?

When connecting a HS device with an interrupt endpoint with bInterval = 1 (1 microframe, 125us), it is being polled only at 2 microframe interval (250us), in case of a NAK.

How to reproduce ?

Connected a device with bInterval = 1 on it's interrupt endpoint

Debug Log as txt file (LOG/CFG_TUSB_DEBUG=2)

Debug log not relevant, I think.

The issue seems to be in handle_channel_in_slave -> channel_xfer_in_retry.

Screenshots

I have checked existing issues, dicussion and documentation

• I confirm I have checked existing issues, dicussion and documentation.

[maximevince]

What I am seeing is this alternating interval of:

• the channel IRQ finding a NAK (+HALTED), then deferring issuing a new IN token to the SOF handler. (every 250us)
• then 1 microframe later: the SOF IRQ re-scheduling the IN token, (so also every 250us, but with 125us offset to the channel IRQ)

It seems to me that the channel IRQ handler should immediately re-issue a new IN token (submit URB) after receiving the NAK, such that on the next microframe we can get the ACK / NACK response.

[maximevince]

I've found a potential solution by calling edpt_xfer_kickoff() from the channel_xfer_in_retry() handler, for hcint & HCINT_HALTED.
This is essentially the same thing as the SOF handler would do, but then right after receiving the NACK, instead of wasting a microframe waiting for the next SOF to come along.

Any comments on what The Right Way to do this, is? @HiFiPhile @roma-jam

[HiFiPhile]

I need to do some tests to understand better periodic transfer... With 1st look it seems not only bInterval = 1 but all bInterval have an extra SOF period.