Re: [OpenOCD-devel] Slow CMSIS-DAP

[Thomas S. <tho...@gm...>]

Andreas,

Thanks for the very insightful reply. 1 and 2 don't seem options for me.
For 3, any recommendations on a FTDI based adapter that works with those
changes?
For 4, how much work do you think this will be? Any guidelines on it? I am
definitely considering this route, as the CMSIS-DAP devices are becoming
more and more numerous on so many eval kits.
I am also looking at option 5, as we have some BBBs that can use the
BitBang method too.

Cheers,

- Thomas



On Mon, Jun 23, 2014 at 4:42 PM, Andreas Fritiofson <
and...@gm...> wrote:

>
>
>
> On Mon, Jun 23, 2014 at 11:46 PM, Thomas Schmid <tho...@gm...>
> wrote:
>
>> Hello,
>>
>> We are trying to switch over our part programming to use CMSIS-DAP, since
>> some of the newer parts we work with only support SWD programming. Our
>> specific setup is the new Atmel ICE (EDBG version) with the Atmel SAMR21.
>> We have it all working, but programming the flash is very slow (~1kB/s). To
>> do some tests, I chose one of our other SAM4S platforms that can do both
>> JTAG and SWD programming. Using a Segger JLink vs the Atmel ICE is like
>> light and day (with same adapter_khz settings).
>>
>> Does anybody else have this problem, and maybe a solution for it? Or does
>> anybody have an idea where the bottle neck could be? I started to look at
>> the SWD signals, and there seems to be a lot of wait time between commands
>> on the bus.
>>
>>
> Hi!
>
> The problem is that the CMSIS-DAP implementation in OpenOCD is
> synchronous, i.e. it only has a single DAP access outstanding at any time.
> USB round trips makes that unbearably slow. There are patches that are
> going to be merged very soon that changes the API so that the driver can be
> rewritten into an asynchronous style. But these patches are only to support
> the SWD implementation for the FTDI driver, the CMSIS-DAP driver has not
> been rewritten yet.
>
> So your options right now are to:
>
> 1. Live with the slow speed and know that it's going to improve some time
> in the future.
> 2. Get an ST-Link (like on the ST discovery boards) and live with the fact
> that you won't have the same level of control (for example, no vector
> catch).
> 3. Get a suitable FTDI based adapter and use the SWD patches currently in
> Gerrit [1] (or wait for them to get merged). This option is speedy, i.e.
> 700 kByte/s memory R/W. Flash programming speed of course also depends on
> the flash driver in OpenOCD and the actual flash controller.
> 4. Maybe the best option (for everyone else at least :)), rewrite the
> CMSIS-DAP driver yourself, on top of the asynchronous API [2].
> 5. There's also the possibility to fix/update the JLink, Versaloon or RPi
> SWD driver [3] that is also depending on the above API changes.
>
> Also note that the EDBG CMSIS-DAP implementation is IMHO broken [4] and
> may require quirks that harms performance, even in asynchronous mode.
>
> Regards,
> Andreas
>
> [1] The change series ending with http://openocd.zylin.com/2184
> [2] http://openocd.zylin.com/1959 which is also included in [1]
> [3] http://openocd.zylin.com/2141 http://openocd.zylin.com/1947
> http://openocd.zylin.com/2004
> [4] http://openocd.zylin.com/#/c/2088/3/src/target/cortex_m.c,cm
>