On Wed, Jan 19, 2011 at 02:16:13PM -0800, Guenter Roeck wrote:
> On Wed, 2011-01-19 at 16:14 -0500, Ben Dooks wrote:
> > On Wed, Jan 19, 2011 at 01:26:37PM -0800, Guenter Roeck wrote:
> > > Signed-off-by: Guenter Roeck <guenter.roeck@xxxxxxxxxxxx>
> [ ... ]
>
> > > +/* Structure to hold all of our device specific stuff */
> > > +struct i2c_diolan_u2c {
> > > + struct usb_device *usb_dev; /* the usb device for this device */
> > > + struct usb_interface *interface;/* the interface for this device */
> > > + struct i2c_adapter adapter; /* i2c related things */
> > > + int olen; /* Output buffer length */
> > > + int ocount; /* Number of enqueued messages */
> > > + u8 obuffer[DIOLAN_OUTBUF_LEN]; /* output buffer */
> > > + u8 ibuffer[DIOLAN_INBUF_LEN]; /* input buffer */
> >
> > you should cache-line align the buffers to avoid any problems with
> > dma-vs-cache, etc.
> >
> There are several other drivers which don't cache-align the buffers used
> with usb_bulk_msg(). Do those all have potential problems, or is this
> only relevant for optimization ?
>
> If it is just for optimization, I'd argue that the speed here is so slow
> that it won't make much of a difference, if any, if the buffers were
> aligned.
It is more to do with the way some architectures/cpus deal with cache
coherency with resepect to DMA... in some cases it is possible for hardware
to be dealing with a buffer that the cpu may still be using if they are
not appropriately aligned. I forget the exact case, but losing a few bytes
in a buffer is preferable to the other case.
I also believe that X86 style architectures do not have this problem.
--
Ben Dooks, ben@xxxxxxxxx, http://www.fluff.org/ben/
Large Hadron Colada: A large Pina Colada that makes the universe disappear.
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
