When I first started adding 64bit support, I created corresponding 64bit variants of read/write functions while still maintaining the old 32bit interfaces. It was done that way to minimise breakage to old 32bit targets.
you can see some comments in http://openocd.zylin.com/#/c/2523/ Andreas Fritiofson mentions there was already a patch in Gerrit that changes the target addresses throughout OpenOCD to a 64-bit type. http://openocd.zylin.com/1200 This patch was a much bigger change which affects all targets but arguably would allow support for all 64bit targets and not just Arm Aarch64. I went ahead and reworked http://openocd.zylin.com/1200 and converted all my code ontop of that. So that's a very brief history why it was done this way. Now the question of Aarch32 and Aarch64 handling. I see that those are different enough that a new aarch64 was created, just like cortex_m as to cortex_a. Since both cortex_m and cortex_a are riddled with armv7 assumption and aarch64 is different is quite a few areas like crosstrigger, register and instruction usage. My intention at the time was to reuse (or rename) cortex_a.c to be the Aarch32 target, but perhaps that might be too optimistic since Aarch32 debug is also different enough - it uses thumb2 instructions etc. I've not been too active with openocd for awhile as you've notice I haven't pushed any new patches. I haven't look closely enough at Liu's series, but I believe its not based off http://openocd.zylin.com/1200. David ________________________________________ From: Matthias Welwarsky <matth...@welwarsky.de> Sent: Thursday, September 8, 2016 2:14 AM To: David Ung Cc: openocd-devel@lists.sourceforge.net Subject: Re: OpenOCD Aarch64 support On Wednesday 07 September 2016 21:13:43 you wrote: > sure > > if you require a more general discussion it might be best to cc > openocd-devel, since I see quite a few people are interested on 64bit > support. Some is just into details of the implementation, some is about broader topics, so I'll just Cc the devel mailing list from here on. Let me fill you in on my current state: I've started by rebasing your patch series up to here: http://openocd.zylin.com/#/c/2747/4 onto the current master branch, to incorporate the DAP related refactoring that meanwhile went into master. Then I looked into merging changes that Pierre Kuo added in between in this patch: http://openocd.zylin.com/#/c/2523/, which is apparently what also Peter Griffin of Linaro used to get debugging working on the Hikey board. However, I think that 2523 goes too far in certain places, for example by eliminating all code that discerns between AARCH32 and AARCH64 execution state of a PE, e.g. when reading registers from the core. I think while this patch fixes some important aspects of AARCH64 debugging, it breaks AARCH32 debugging in the process. There's also the patch series from Alamy Liu, which also adds on top of your series, but it's introducing some far ranging changes in the architecture which I'm quite sure I won't be following. Did you ever look closer into 2523 or the Liu patches? What do you think of them? On a broader scope, was there a specific technical reason why you created a new aarch64 target instread of integrating with cortex_a ? Of course there are some notable differences, but there is also now a lot of duplicated code. > > David BR, Matthias ----------------------------------------------------------------------------------- This email message is for the sole use of the intended recipient(s) and may contain confidential information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply email and destroy all copies of the original message. ----------------------------------------------------------------------------------- ------------------------------------------------------------------------------ _______________________________________________ OpenOCD-devel mailing list OpenOCD-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/openocd-devel
