Op do 24 okt. 2019 om 22:56 schreef Andreas 'count' Kotes <count-vger.kernel.org+linux-
What needs to happen for stable, reliable support for existing PCMCIA network cards people already have? How can we unblock those possible Linux users?
That would involve quite a few things. I did most of this back in 2.6 times, and had the standard pcnet_cs, 3c589_cs and serial_cs drivers working on my Amiga 1200. Some of it was a bit hackish though, changes to <asm/io.h> were discussed but never implemented. The need to be able to build a multi-machine kernel (with Atari, Q40 and Amiga support in a single kernel) was one of the things that interfered (it could probably be solved by adding an extra level of indirection ;-)). Then Real Life happened and I didn't touch my Amiga for 10+ years... 1) A proper driver for the PCMCIA slot, including hot plugging and proper controller reset. This involves rewriting part of the IRQ handling to make Gayle a separate IRQ controller instead of the hard coded hack that is used now. That shouldn't be too hard. Support for setting up different data path widths (IO_DATA_PATH_WIDTH_16 etc.) will require changes to <asm/io.h>, see point 2. 2) Rework <asm/io.h> so inb()/inw()/inl(), outb()/outw()/outl(), insb()/insw()/insl() and outsb()/outsw()/outsl() "just work" for PCMCIA address space without having to create ugly hacks in the generic drivers. This is not trivial because of the strange memory mapping of the Amiga PCMCIA slot. 3) ... 4) Profit! Oh, the 3c589_cs driver needed an ugly hack because some of the registers just didn't read out properly on the Amiga with 16-bit access. That issue is also described in the source code for the AmigaOS driver. Kind regards, Kars.