[PATCH] dfu-util: Don't sleep unnecessarily

Tormod Volden lists.tormod at gmail.com
Fri Mar 11 23:37:42 CET 2011

On Wed, Mar 9, 2011 at 10:14 AM, Bernard Blackham wrote:
>> On Wed, 2011-03-09 at 16:49, Bernard Blackham wrote:
>> > The DFU spec says that bwPollTimeout is the amount of time that
>> > should pass between DFU_GETSTATUS requests. If the status is
>> > already good, we do not need to sleep as we will not poll again.
>> >
>> > This speeds up DFU downloads directly into memory.

The patch looks fine to me. If I understand the 1.1 spec correctly,
the bwPollTimeout only needs to be honoured by the host in the
dfuDNBUSY case. On the other hand, I would expect an ideal device to
adjust the bwPollTimeout according to what kind of operation it is
busy with, thus set a small value (or zero) when writing to RAM, and
likewise, set it to zero if the status is not dfuDNBUSY. This is the
case for the devices I tested it on, so no speed-up was seen here. But
your patch will certainly help those less ideal devices out there :)

> One point to note - like the OpenMoko version (which it was based
> upon), the current u-boot patch does not populate bwPollTimeout
> correctly.  I will be posting a fixed patch shortly, so it will not
> need a quirk. If people ask, please direct them to fix their u-boot
> instead :)

Great. Please consider adjusting the bwPollTimeout as explained above,
looking at your maple patch it seems like it is currently hardcoded to
5 ms in all cases.

BTW, I found this comment in your code:
/* FIXME: only accept getstatus if bwPollTimeout
 * has elapsed */
IMO, this should be respected by the host, so that you don't need to
take care of it in the device.

Also wondering about your USB_REQ_DFU_DETACH extension. Can you not
use a zero-length DNLOAD to enter dfuMANIFEST-WAIT-RESET?


More information about the devel mailing list