[U-Boot] [PATCH] fastboot: Fix OUT transaction length alignment

[Roger Quadros]

Hi,

On 13/04/16 15:01, Semen Protsenko wrote:
> From: Sam Protsenko <semen.protsenko at linaro.org>
> 
> Some UDC controllers may require buffer size to be aligned to
> wMaxPacketSize. It's indicated by gadget->quirk_ep_out_aligned_size
> field being set to "true" (in UDC driver code). In that case
> rx_bytes_expected must be aligned to wMaxPacket size, otherwise stuck on
> transaction will happen. For example, it's required by DWC3 controller
> data manual:
> 
>     section 8.2.3.3 Buffer Size Rules and Zero-Length Packets:
> 
>     For OUT endpoints, the following rules apply:
>     - The BUFSIZ field must be ≥ 1 byte.
>     - The total size of a Buffer Descriptor must be a multiple of
>       MaxPacketSize
>     - A received zero-length packet still requires a MaxPacketSize buffer.
>       Therefore, if the expected amount of data to be received is a multiple
>       of MaxPacketSize, software should add MaxPacketSize bytes to the buffer
>       to sink a possible zero-length packet at the end of the transfer.
> 
> But other UDC controllers don't need such alignment, so mentioned field
> is set to "false". If buffer size is aligned to wMaxPacketSize, those
> controllers may stuck on transaction. The example is DWC2.
> 
> This patch checks gadget->quirk_ep_out_aligned_size field and aligns
> rx_bytes_expected to wMaxPacketSize only when it's needed.
> 
> Signed-off-by: Sam Protsenko <semen.protsenko at linaro.org>
> ---
>  drivers/usb/gadget/f_fastboot.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
> 
> diff --git a/drivers/usb/gadget/f_fastboot.c b/drivers/usb/gadget/f_fastboot.c
> index 2e87fee..54dcce0 100644
> --- a/drivers/usb/gadget/f_fastboot.c
> +++ b/drivers/usb/gadget/f_fastboot.c
> @@ -58,6 +58,7 @@ static unsigned int fastboot_flash_session_id;
>  static unsigned int download_size;
>  static unsigned int download_bytes;
>  static bool is_high_speed;
> +static bool quirk_ep_out_aligned_size;
>  
>  static struct usb_endpoint_descriptor fs_ep_in = {
>  	.bLength            = USB_DT_ENDPOINT_SIZE,
> @@ -240,6 +241,8 @@ static int fastboot_set_alt(struct usb_function *f,
>  	debug("%s: func: %s intf: %d alt: %d\n",
>  	      __func__, f->name, interface, alt);
>  
> +	quirk_ep_out_aligned_size = gadget->quirk_ep_out_aligned_size;
> +
>  	/* make sure we don't enable the ep twice */
>  	if (gadget->speed == USB_SPEED_HIGH) {
>  		ret = usb_ep_enable(f_fb->out_ep, &hs_ep_out);
> @@ -435,12 +438,18 @@ static unsigned int rx_bytes_expected(unsigned int maxpacket)
>  		return 0;
>  	if (rx_remain > EP_BUFFER_SIZE)
>  		return EP_BUFFER_SIZE;
> +
> +	if (!quirk_ep_out_aligned_size)
> +		goto out;
> +
>  	if (rx_remain < maxpacket) {
>  		rx_remain = maxpacket;
>  	} else if (rx_remain % maxpacket != 0) {
>  		rem = rx_remain % maxpacket;
>  		rx_remain = rx_remain + (maxpacket - rem);
>  	}
> +
> +out:
>  	return rx_remain;
>  }
>  
> 

Why do we need a special flag for this driver if other drivers e.g. mass storage
are doing perfectly fine without it?

This patch is just covering up the real problem, by bypassing the faulty code
with a flag.

The buffer size is EP_BUFFER_SIZE and is already aligned to wMaxPacketSize so
the problem shouldn't have happened in the first place. But it is happening
indicating something else is wrong.

Why is the code using wMaxPacketSize for the check. why can't it just use usb_ep->maxpacket?

cheers,
-roger