22 Jan
2017
22 Jan
'17
1:04 p.m.
On Sun, 22 Jan 2017, Tony Duell wrote:
On most PCs the RTC chip is mapped as I/O ports, not
memory, so you couldn't
execute code from its RAM anyway. I suppose on (say) a 68000 based machine
you could (there is no separate I/O port address space on that processor) but
I have never seen it done.
It's not the I/O vs memory mapping that is the problem with treating the
data space of the MC146818 and its derivatives as executable code; most
processor architectures do not have a dedicated I/O address space anyway.
The issue with these chips is indirect addressing, that is that ordinarily
they only present a pair of 8-bit registers on the bus they are attached
to, one being an address register to select the intended location within
the embedded memory and the other being a data window register which only
gives access to a single location at a time.
Direct linear access to data in these chips can be achieved with a small
amount of glue logic, but this is seldom done. I only know of a single
line of computers, namely the MIPS processor based DECstation machines,
which had such logic so that internal locations of their DS1287 chip
appeared linearly in its dedicated MMIO space slot, providing for atomic
RMW accesses from the CPU for example.
But even there the locations were only mapped onto the least significant
byte lane of the 32-bit internal I/O bus, so direct execution would not be
possible due to the unpredictable (or at least uncontrollable, e.g. if
pull-ups or pull-downs were provided) nature of bit patterns presented in
the upper 24 bits of each 32-bit memory word. You'd have to copy data
first from the chip's internal RAM to main RAM to execute it, just as it
was done by DECstation firmware with some option board code which was
contained in 8-bit ROM chips.
Maciej