Hi@warmonkey
Sorry for the delay. 😢
I've dig into HAL to see why it was uint16_t and found the reason, even if changed to uint32_t it will be always truncated to uint16_t when accessing the PMA (R/W).

So I will not merge this PR, sorry.

Hi@warmonkey Sorry for the delay. 😢 I've dig into HAL to see why it was uint16_t and found the reason, even if changed to uint32_t it will be always truncated to uint16_t when accessing the PMA (R/W).

Arduino_Core_STM32/system/Drivers/STM32F1xx_HAL_Driver/Src/stm32f1xx_ll_usb.c

Line 2817 in345138a

voidUSB_WritePMA(USB_TypeDefconst*USBx,uint8_t*pbUsrBuf,uint16_twPMABufAddr,uint16_twNBytes)

So I will not merge this PR, sorry.

On STM32F4 USB_WritePMA will not be called. The actual function to send data over USB is USBD_LL_Transmit() -> ... -> USB_EPStartXfer()

There two paths can trigger the USBD_LL_Transmit(): USBSerial::write() -> CDC_continue_transmit(), USBD_CDC_TransmitCplt() -> CDC_continue_transmit() -> USBD_CDC_TransmitPacket() -> USBD_LL_Transmit() -> HAL_PCD_EP_Transmit() -> USB_EPStartXfer()

In CDC_continue_transmit() file: usbd_cdc_if.c line 345, it's callingbuffer = CDC_TransmitQueue_ReadBlock(&TransmitQueue, &size); which returns the size to send.
In CDC_TransmitQueue_ReadBlock() file: cdc_queue.c line 88, the size value is limited to 65472 in my commit

as stated in stm32f4xx_ll_usb.c function USB_EPStartXfer line 782:

Program the transfer size and packet count as follows: xfersize = N * maxpacket + short_packet pktcnt = N + (short_packet_exist ? 1 : 0)...pktcnt = (uint16_t)((ep->xfer_len + ep->maxpacket - 1U) / ep->maxpacket);USBx_INEP(epnum)->DIEPTSIZ |= (USB_OTG_DIEPTSIZ_PKTCNT & (pktcnt << 19));...USBx_INEP(epnum)->DIEPTSIZ |= (USB_OTG_DIEPTSIZ_XFRSIZ & ep->xfer_len);

from the code above we know that the size to send from CDC_TransmitQueue_ReadBlock() previously, is assigned to ep->xfer_len (at stm32f4xx_hal_pcd.c line 1940 for stm32f4), then write into OTG_FS_DIEPTSIZx or OTG_HS_DIEPTSIZx register as:

PKTCNT = (int)(xfer_len/MAX_PACKET_SIZE)     //64 for FS, 512 for HSXFRSIZ = xfer_len

from RM0008 RM0090 RM0432 we know the PKTCNT has 10bits, max value 0x3ff (1023), XFRSIZ has 19bits, max value 0x7ffff (524287). 1023 * 512 = 523776 < 524287

but, it's not easy to know the USB is running in HS or FS mode.

In conclusion, we should limit the maximum size returned by CDC_TransmitQueue_ReadBlock() to OTG_MAX_PKTCNTUSB_FS_MAX_PACKET_SIZE which is 102364 (for now).

I will readpdev->ep_in[CDCInEpAdd & 0xFU].maxpacket instead in the next commit.

The CDC queue is a software buffer, it's not relevant to the PMA area. The PMA is located in0x4000 6000(1) - 0x4000 63FF Shared USB/CAN SRAM 512 bytes (RM0008 Page 52) it has only 512 bytes shared with multiple endpoints. User should copy data from/to PMA manually.

Before I submit this PR I checked multiple times, compiled and deployed on actual products. I ran the test for at least 24hrs.

Im working on an improvement. Will usemaxpacket in USBD_EndpointTypeDef instead of use a fixed value 64. Just finished but i need more real world test.