Re: loadbsd memory segment (Was: NetBSD Amiga - Memory restrictions)
David Brownlee <[email protected]> Sat, 25 Mar 2023 18:38:07 +0000
| Newsgroups | gmane.os.netbsd.ports.amiga |
|---|---|
| Message-ID | <CAGN_6pZob7_ODKC_aA0LQS-Pahk2tigEOQU9fGvTb4bLm6JQiQ@mail.gmail.com> |
On Sat, 25 Mar 2023 at 17:44, Karoly Balogh <[email protected]> wrote: > > Hi, > > On Sat, 25 Mar 2023, Frank Wille wrote: > > > >> > An other thing is loadbsd, is it possible to load the Kernel in the > > >> > TK-Memory 0$800000 instaed of the biggest available block, which > > >> > could e slow Z3 RAM? > > >> > > >> Option -p should work, which selects the memory segment with the > > >> highest priority instead of largest size. > > > > > > Maybe the default heuristics could be adjusted - load the kernel into > > > the highest priority memory segment of at least X MB (and make X an > > > optional command line option). > > > > Agreed. Selecting the memory segment by priority is mostly the better > > option. > > > > But other NetBSD developers already must have had these throughts decades > > ago, so I'm always afraid we are missing something... ;) > > Maybe, but decades ago the Amiga hardware landscape looked very different. > The 128MB (or 256MB memory) memory that was theoretically available as > "localfast" seemed huge, and was almost always the largest block. It's > still true for most A1200 installations of course. But meanwhile 112MB > mods for the big-box mainboards became a thing, huge ZIII expansions > became a thing, etc. > > I think the kernel should be always loaded to the CPU card local fast ram, > if possible, for performance reasons, which works just fine with the > memory-priority system, as that chunk is always the highest prio. That would be ideal in almost every case - but the edge case of a machine with a small amount of fast RAM and a larger amount of slower memory would then fail. Having a 'minimum amount of RAM' value, adjustable by a param, combined with defaulting to -p should make the default work in ~every case, and be optimal in almost all Continuing my self appointed role as "loud voice that encourages others to do the actual work" - is anyone interested? :) David