>> The answer is no with an external osc.  When programmed as an RC the PIC is
>> driving osc1.  On osc1 when the V gets to a certain level (say 4V) the pic
>> pulls it down.  When it pulls it to some low level (say 1V) it lets it go to
>> float back up, and so on.  On osc2 it is clocking out osc1/4.
>
>Right; what I was wondering about was whether, e.g., connecting a capacitor
>between an external oscillator signal and the osc1 pin would allow the PIC
>to run at the rate set by that oscillator [when the oscillator goes high,
>the PIC side of the cap would go high too until the PIC pulled it down; when
>the oscillator goes low, the cap would be discharged into the PIC's negative
>rail].  Would this be reliable for reasonable speeds? [e.g. 10Hz to 10Mhz?]
>
Probably not too reliable.  The PIC would tend to oscillate by itself using
your capacitor and whatever leakage resistance was there.  Although
specifically not recommended, maybe biasing the OSC1 pin to +2.5V or so and
then couping in your oscillator through a rather small capacitor could work.
Another possibility would be to run the RC oscillator normally, but couple
in a synchronizing signal via a voltage divider connected to the
normally-grounded end of the capacitor.  The rising edges of the external
oscillator would couple through the capacitor and trigger the PIC
prematurely, causing the RC oscillator to speed up and match the external
oscillator, provided the difference isn't too great (may not be possible to
guarantee that with PIC and component tolerances).

Figure 4.0.1 in my (regrettably still preliminary) 1994 data book
"Simplified Block Diagram of On-Chip Reset Circuit" suggests that the 1024
Tosc timer is active only for the XT or LP modes (i.e. not HS).  However,
the table posted by someone else and also shown in this book, as well as the
text, conflicts with that.  Could someone at Microchip confirm that it is
active in the HS mode as well?

With an external oscillator, at least the time from wake-up to execution of
the first instruction will be predictable (256 instruction cycles, or only
0.256 ms at 4 MHz), not a few ms + 256 instruction cycles which would occur
if the internal oscillator with a crystal were used.

-Mike