On 01/04/17 11:32, Alex Bennée wrote: > > Sean Bruno <sbr...@freebsd.org> writes: > >> I'm pondering where to start with getting FreeBSD's bsd-user code into >> shape so it could actually be reviewed and accepted now that its sort of >> working again (signal handling fixed finally). > > Also don't forget the parallel_cpus changes which were made for enhanced > atomic support. There was also work in fixing up exclusive_start/end > which is used for safe work like tb_flush.
Yeah, I ran head first into that wall last month. :-) We should be ok at this point as I've merged to master and am keeping up now. > >> I almost feel like the existing code should be purged, except that it >> gives a good history (and this seems lazy to me). >> >> As a first pass, I guess, I'd like to at least make i386 user run on >> x86_64. What would you folks like to see in a first pass? > > One thing that bugs me about linux-user is the amount of copy&paste > repetition that goes on for the main run loop. Perhaps with a clean-up > of the unused BSD architectures we could explore how a re-factored > run loop could look with a view to moving to common run-loop code across > both? The bsd-user code does not have that nearly as much. Hardware architectures have been moved to seperate dirs and o/s support has been broken up into seperate dirs. This has made maintenance and testing much easier, but not easy enough to get the code into your folk's tree. :-) > > Apart from syscall mapping and backend calls to the BSD ABI what > differences are there between the two user-mode implementations? > Not a substantial amount, IMO. The setup and tear down are a bit different, signal handling is a bit different, and as you mention syscall mapping/backend calls are different. >> >> sean >> >> ref: https://github.com/seanbruno/qemu-bsd-user/tree/bsd-user > > > -- > Alex Bennée >
signature.asc
Description: OpenPGP digital signature
