From mboxrd@z Thu Jan 1 00:00:00 1970 References: <87r1h57kx5.fsf@xenomai.org> From: Philippe Gerum Subject: Re: [PATCH 00/17] Dovetail integration - the finals In-reply-to: Date: Mon, 14 Jun 2021 08:24:21 +0200 Message-ID: <87k0mx6k16.fsf@xenomai.org> MIME-Version: 1.0 Content-Type: text/plain List-Id: Discussions about the Xenomai project List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Jan Kiszka Cc: xenomai@xenomai.org, Hongzhan Chen Jan Kiszka writes: > On 13.06.21 19:07, Philippe Gerum wrote: >> >> Jan Kiszka writes: >> >>> On top, I would also still like to understand - and ideally address >>> prior to releasing a new major version - if we cannot align the ARM >>> syscall ABI of Xenomai to the kernel /wrt argument passing. Now is a >>> good chance to do that. >> >> That would be fairly simple implementation-wise, problem though: this >> would require to add this change to the next I-pipe/arm releases, >> invalidating all previous I-pipe releases for 3.2 in the same move. In >> theory, all existing I-pipe/arm patches starting from kernel 3.10 can >> run the -next branch (although it may be better to stick with 4.14+ in >> practice). >> > > 4.14 is long dead, 4.19 could be the lowest target. Yes, that might > involve some some complications. Maybe we could add a feature flag for > the new ABI, start with dovetail and then look into lifting 4.19 and 5.4 > I-pipe. > Dovetail already follows the ARM EABI, only using a marker to distinguish in-band syscalls from foreign ones in r7, on all archs. This said, I plan to work on another option for conveying oob syscalls based on prctl() asap, this would be saner than invalidating a whole range of syscall numbers for the regular kernel. In addition, this might help enabling tools like valgrind at little/no cost. The earlier the better to investigate this, I'd rather not have to maintain multiple ABIs in Dovetail. -- Philippe.