I think it's better to give USB submit_xfer() an extra parameter to specify the actual amount of data in the buffer.

[XIVN1987]

Change the code to look like the following:

static mp_obj_t usb_device_submit_xfer(size_t n_args, const mp_obj_t *args) {
    ...

    uint len = (n_args == 3) ? buf_info.len : mp_obj_get_int(args[3]);
    result = usbd_edpt_xfer(USBD_RHPORT, ep_addr, buf_info.buf, len);

    ...
}
static MP_DEFINE_CONST_FUN_OBJ_VAR_BETWEEN(usb_device_submit_xfer_obj, 3, 4, usb_device_submit_xfer);

This avoids calling submit_xfer as follows

self.usbd.submit_xfer(self.EP_IN, self.in_buf[:xfer_bytes])

Instead, it can be written as:

self.usbd.submit_xfer(self.EP_IN, self.in_buf, xfer_bytes)

The former cannot be used in ISR because it causes memory allocation, and allocating in ISR causes MemoryError.

[GitHubsSilverBullet]

This avoids calling submit_xfer as follows
self.usbd.submit_xfer(self.EP_IN, self.in_buf[:xfer_bytes])
The former cannot be used in ISR because it causes memory allocation, and allocating in ISR causes MemoryError.

A memoryview would serve the same purpose (not allocating slices).

[XIVN1987]

Slice operation on memoryview causes allocation