memset(0) the structure, not the pointer itself, and what ever
data happens to be after the pointer's stack location.

Size of erase block is 128 bytes when sector density (z) is zero.

branches: 1.7.2; 1.7.8; 1.7.26;
- In cfi_probe() iterate over all chip widths for every port width to find
x8/x16 configurations, or two x8 chips forming a x16 port width.
- The offset in cfi_cmd() is now always given for 64-bit port width.
It will be scaled down for the actual port width to avoid problems
with 0x2aa -> 0x554 instead of 0x555.
- Added missing cfi_reset_default() in cfi_jedec_id(). The reset is
needed, because the chip was still in CFI-Query mode.
- Removed everything dealing with cfi_opmodes. It only complicates things
and I found no reason for keeping them. The port width and chip width
is known, so I just introduced a new cfi_read_qry() function which reads
a single word using bus_space_read_N() from the QRY structure, which
can be checked for 'Q', 'R' and 'Y'. That's all we need, before
reading and unpacking the whole QRY and PRI structures.
- Added two new fields, cfi_unlock_addr1 and cfi_unlock_addr2 to
the cfi struct. The unlock offsets should be kept variable, depending
on the manufacturer or command set version.

Changes have been discussed with cliff@.

branches: 1.6.2; 1.6.6;
when unpacking CFI QRY data, byte order within a word is host order,
but order across words appears to be invariant across -el and -eb systems.

branches: 1.5.2;
- add little endian variants of cfi_opmodes tables (untested)
- try to reduce glop factor in cfi_jedec_fill()

- dump raw QRY and/or JEDEC data depending on CFI_DEBUG_QRY, CFI_DEBUG_JEDEC
- add ability to emulate CFI based on table lookup keyed by JEDEC IDs.
- warn if CFI command set is unknown

add TNF copyright

#include <sys/bus.h> not <machine/bus.h>.

Add support for CFI NOR, using MPC8536DS as initial example.
Only AMD/Fujitsu command set is suported so far.
This is still work in progress, be advised.