Hi Geert,
Am 18.06.2021 um 20:13 schrieb Geert Uytterhoeven:
How does re-probing after a card change for a builtin driver work?
Changing the permission bits is a minor issue.
Oh right, this driver predates the driver framework, and doesn't support
PCMCIA hotplug. So auto-unregister on removal doesn't work.
Even using unbind/bind in sysfs won't work.
So rmmod/modprobe is the only thing that has a chance to work...
The comment there says isa_type must be set as early as possible, so I'd
rather leave that alone, and add an 'else' clause here.
This of course raise the question whether we ought to move the entire
isa_type handling into arch code instead - make it a generic
amiga_pcmcia_16bit option settable via sysfs. There may be other 16 bit
cards that require the same treatment, and duplicating PCMCIA mode
switching all over the place could be avoided. Opinions?
Indeed.
The only downside I can see is that setting isa_type needs to be done
ahead of modprobe, through sysfs. That might be a little error prone.
Still, can we autodetect in the driver?
Guess we'll have to find out how the 16 bit cards behave if first poked
in 8 bit mode, attempting to force a reset of the 8390 chip, and
switching to 16 bit mode if this fails. That's normally done in
apne_probe1() which runs after init_pcmcia(), so we can't rely on the
result of a 8390 reset autoprobe to do the PCMCIA software reset there.
The 8390 reset part does not rely on anything else in apne_probe1(), so
that code can be lifted out of apne_probe1() and run early in
apne_probe() (after the check for an inserted PCMCIA card). I'll try and
prepare a patch for Alex to test that method.
I just realized that might not work - ínit_pcmcia() enables the PCMCIA
interface for us, so the early 8390 reset may not go through at all...
The module parameter may have to stay as a fallback option at least.
I'm wondering how this is handled on PCs with PCMCIA, or if there
really is something special about Amiga PCMCIA hardware...
What's special about Amiga PCMCIA hardware is that the card reset isn't
connected for those 16 bit cards, so pcmcia_reset() does not work.
I was mostly thinking about the difference between 8-bit and 16-bit
accesses.
No idea. I've never even seen an 8 bit PCMCIA card - I have a few old
16/32 bit ones around that were great for crashing my PowerBook, nothing
else...
And I'd really like to get rid of the CONFIG_APNE100MBIT option,
i.e. always include the support, if possible.
I can't see why that wouldn't be possible - the only downside is that we
force MULTI_ISA=1 always for Amiga, and lose the optimizations done for
MUTLI_ISA=0 in io_mm.h. Unless we autoprobe, we can use isa_type to
guard against running a software reset on 8 bit cards ...
The latter sounds like a neat trick...
Yes, we can at least get rid of the APNE100MBIT option that way. I'll
have to think about the autoprobe a bit more.
Cheers,
Michael
Gr{oetje,eeting}s,
Geert
