[RFC] dwc2: Don't assume URB transfer_buffer are dword-aligned

Message ID 5c313be6d2f3c786d1f159a74758263a2ae0ad66.1489709097.git.mchehab@s-opensource.com (mailing list archive)
State Under Review, archived
Delegated to: Mauro Carvalho Chehab
Headers

Commit Message

Mauro Carvalho Chehab March 17, 2017, 12:08 a.m. UTC
  The dwc2 hardware doesn't like to do DMA transfers without
aligning data in DWORD. The driver also assumes that, even
when there's no DMA, at dwc2_read_packet().

That cause buffer overflows, preventing some drivers to work.

In the specific case of uvc_driver, it uses an array where
it caches the content of video controls, passing it to the
USB stack via usb_control_msg(). Typical controls use 1 or 2
bytes. The net result is that other values of the buffer
gets overriden when this function is called.

Fix it by changing the logic at dwc2_alloc_dma_aligned_buffer()
to ensure that the buffer used for DMA will be DWORD-aligned.

Detected with uvc driver.

Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
---

On Raspberry Pi 3, I was unable to test  dwc2_read_packet(), as this was
never called there. However, the change at the second hunk, e. g. at
dwc2_alloc_dma_aligned_buffer() made UVC to answer the same way
as on x86, while reading the values for the device controls.

 drivers/usb/dwc2/hcd.c | 29 ++++++++++++++++++-----------
 1 file changed, 18 insertions(+), 11 deletions(-)
  

Comments

Mauro Carvalho Chehab March 29, 2017, 6:49 p.m. UTC | #1
Em Wed, 29 Mar 2017 11:57:22 +0200
Greg Kroah-Hartman <gregkh@linuxfoundation.org> escreveu:

> On Tue, Mar 28, 2017 at 06:48:02AM -0300, Mauro Carvalho Chehab wrote:
> > Em Fri, 17 Mar 2017 10:24:15 +0900
> > Greg Kroah-Hartman <gregkh@linuxfoundation.org> escreveu:
> >   
> > > On Thu, Mar 16, 2017 at 09:08:40PM -0300, Mauro Carvalho Chehab wrote:  
> > > > The dwc2 hardware doesn't like to do DMA transfers without
> > > > aligning data in DWORD. The driver also assumes that, even
> > > > when there's no DMA, at dwc2_read_packet().
> > > > 
> > > > That cause buffer overflows, preventing some drivers to work.    
> > > 
> > > Why aren't the drivers being fixed?  This is a well-known (hopefully)
> > > restriction on USB data buffers, can't the uvc_driver be fixed?  
> > 
> > I talked to Laurent about on IRC. He told that he is willing to consider
> > that option, if the USB API explicitly states that all buffers must be
> > dword-aligned (and/or buffer sizes).
> > 
> > IMHO, he has a point: if this is a restriction of for usb transfer
> > buffers, it should be documented somewhere.
> > 
> > Yet, a quick check with:
> > 	$ git grep -i dword Documentation/usb/
> > 	$ git grep -i align Documentation/usb/
> > 
> > Didn't hit anything related to it.   
> 
> Hm, most of the USB documentation is in the kerneldoc in the USB code
> itself, and is built that way.  I'm pretty sure this is documented
> somewhere, but I can't seem to find it at the moment either :(
> 
> Care to write a patch for that?  :)

Sure. Btw, I noticed that not all USB documents were converted
yet to ReST, so I took the time to convert the core documents to ReST
too.

I kept the driver-specific documentation at Documentation/usb.
The final result is at:
	http://www.infradead.org/~mchehab/kernel_docs/driver-api/usb/index.html

I'll be sending the documentation patches in a few.

Thanks,
Mauro
  

Patch

diff --git a/drivers/usb/dwc2/hcd.c b/drivers/usb/dwc2/hcd.c
index a73722e27d07..c53d3c24d5b5 100644
--- a/drivers/usb/dwc2/hcd.c
+++ b/drivers/usb/dwc2/hcd.c
@@ -572,20 +572,26 @@  u32 dwc2_calc_frame_interval(struct dwc2_hsotg *hsotg)
 void dwc2_read_packet(struct dwc2_hsotg *hsotg, u8 *dest, u16 bytes)
 {
 	u32 __iomem *fifo = hsotg->regs + HCFIFO(0);
-	u32 *data_buf = (u32 *)dest;
-	int word_count = (bytes + 3) / 4;
-	int i;
-
-	/*
-	 * Todo: Account for the case where dest is not dword aligned. This
-	 * requires reading data from the FIFO into a u32 temp buffer, then
-	 * moving it into the data buffer.
-	 */
+	u32 *data_buf = (u32 *)dest, tmp;
+	int word_count = bytes >> 2;
+	int i, j;
 
 	dev_vdbg(hsotg->dev, "%s(%p,%p,%d)\n", __func__, hsotg, dest, bytes);
 
 	for (i = 0; i < word_count; i++, data_buf++)
 		*data_buf = dwc2_readl(fifo);
+
+	/* Handle the case where the buffer is not dword-aligned */
+	if (bytes & 0x3) {
+		u8 *buf = (u8 *)data_buf;
+
+		tmp = dwc2_readl(fifo);
+
+		i <<= 2;
+		for (j = 0; i < bytes; j++, i++, dest++)
+			*buf = tmp >> (8 * j);
+	}
+
 }
 
 /**
@@ -2594,8 +2600,9 @@  static int dwc2_alloc_dma_aligned_buffer(struct urb *urb, gfp_t mem_flags)
 	size_t kmalloc_size;
 
 	if (urb->num_sgs || urb->sg ||
-	    urb->transfer_buffer_length == 0 ||
-	    !((uintptr_t)urb->transfer_buffer & (DWC2_USB_DMA_ALIGN - 1)))
+	    urb->transfer_buffer_length == 0 || (
+	    !((uintptr_t)urb->transfer_buffer & (DWC2_USB_DMA_ALIGN - 1)) &&
+	    !((uintptr_t)(urb->transfer_buffer + urb->transfer_buffer_length) & (DWC2_USB_DMA_ALIGN - 1))))
 		return 0;
 
 	/* Allocate a buffer with enough padding for alignment */