Is ESP32 S2/S3 support realistic?

[tobozo]

Hi and million thanks for your awesome library 👍

An idea came up while I was playing with an ESP32-C3 and esp32-arduino-lib-builder to produce a package with memory protection disabled, and accidentally built for ESP32 S2/S3 too.

Although S2/S3 already have builtin USB support, this support is limited to either host or device role, but what if esp32_usb_soft_host is used along with tinyUSB acting as a device?

I chose to focus on ESP32S2/esp32_usb_soft_host alone, I don't have a broken out ESP32-S3 and tinyUSB considerations can wait.

Adapting the code was mostly adding || defined CONFIG_IDF_TARGET_ESP32S2 to #if CONFIG_IDF_TARGET_* conditional blocks and assigning pins.

However checkPins() detection fails, and now the doubt is there 🤔

Some thoughts:

• not sure what value to assign to BLINK_GPIO, tried the builtin led pin, scl pin, random pins and got not change
• not sure if cpu_ll_enable_cycle_count(); is needed in function usb_process(), enabling/disabling produces no change
• using the makeOpcodes() from ESP32 works fine, maybe ESP32S2 wants different opcodes

(In the event the situation is realistic)
Possible use cases of an ESP32S2 using esp32_usb_soft_host/tinyUSB and acting both as host+device:

• Gaming: DIY dongle for joypad/joystick auto fire + programming
• Security research: HID MITM keylogger, mouselogger, injector, etc
• ...

So here's my question:

Is it a xxx limitation prevents yyy feature situation or is it worth researching further?

Thanks again for creating this project, it brought hours of fun ❤️

[tobozo]

well this implementation is really dirty and probably wrong on many levels but after some fiddling I had a keyboard successfully detected on an ESP32-S2

https://github.com/tobozo/ESP32-USB-Soft-Host/releases/tag/v0.1.2

I didn't succeed at using opcodes with the S2, memory management is different and disabling memory protection wasn't enough to prevent it from crashing.

The ugly fix for the ESP32-S2 is to use a substitute for cpuDelay():

    #define NOP() asm volatile ("nop")
    void IRAM_ATTR cpuDelay(uint32_t cycles_count )
    {
      if( cycles_count > 0 ) {
        while( cycles_count-->1 ) NOP();
      }
    }

I'm quite sure the maths don't add up though, maybe this works on a misunderstanding?