> Do we just not use these values and go with safe (i.e. slow) defaults?

After some digging around, this seems to be the case.

Just for the archives, here are the timings I've come up with:

The adapter I'm working on doesn't seem to support 3.3V cards, so all attribute
access is with 300ns timing acording to the PC Card standard. I calculate that
value from the sbus frequency now.

The standard furthermore has some tables listing the various access times.
For the delay value there are minimum values of 10ns (for 100ns access) and
20ns for all others. Which nicely maps to 1 or two cycle counts.

Since our framework doesn't propagate timining requirements to socket drivers,
I'm for now using the attribute memory access timings everywhere.

We should consider adding an optional function

 void socket_set_timing(pcmcia_chipset_handle_t,
	 int mem_ns, int io_ns, int use_wait);

function to struct pcmcia_chip_functions.

Or is this not needed nowadays since all pcmcia cards used in practice are
16bit and able to use WAIT# properly?

Which would mean I should use 300ns timinig for attribute memory, but 100ns
timing otherwise - and WAIT# will handle everything else for me.


Martin