Fixes #313.

Fixes #317.

Fixes #312.

This resolves problems when matching against instructions that aren't
MXLEN bits wide, and allows matches against data accesses of specific
sizes.

* datacount cannot be 0

It must be at least 1 so GPRs can be accessed.

* Clarify minimum \Fdatacount for RVxx

* Move \Fdatacount details to commentary.

* Clarifications requested by Jeremy Bennett

* Targets don't have to implement access to all regs

* Clarify that trigger registers are WARL.

* Clarify what a trigger chain is and when it fires.

* Incorporate feedback.

Previously it was confusing that especially Table 3.8 would end up in
the dmstatus Section. #300

* Completing progbuf exec is I/O for fence insts.

We need something like this so a debugger can ensure a memory change is
propagated to all harts in a system.

* Completing progbuf is just output

Also explain how this might work in an execution-based implementation.

* Add uncached to abstract access memory.

Requested by Thomas Wicki, and seems straightforward enough.

* \Funcached is optional.

* Slight rephrasing.

* It's OK to implement "partial commands."

* Replace no-program-buffer example with no-csr

* Make "uncached" bit simply "target-specific."

* Only write busy to \Fcmderr if \Fcmderr is 0.

* Be a bit more precise.

* Only look at \Fcmderr when \Fbusy is 0

Also suggest how to deal with failures after sending a burst of
commands.

* Incorporate feedback.

It's been a while. I haven't been committing the PDF with every PR
because the conflicts are a pain.

This reverts commit 7c66968e.

* Add Access Memory abstract command

Source provided by Thomas Wicki

* Add autoincrement to abstract memory access

Also added examples in the debugger implementation section.

* Thomas requested Robert get credit for this idea.

* Incorporate review back, add aamvirtual bit

Removed transfer and postexecute bits from abstract memory access.
Renamed a few fields to avoid confusion.
Add aamvirtual bit to abstract memory access.

* Fix memory references to include abstract command

* Fix typo.

* Specify an Exception Trigger

I tried to get all the sources from the Syntacore proposal in, but did
not add the various timing options. We do get interrupt sources now as
well as the exception sources.

* Redo etrigger.

Now we support every source that's explicitly mentioned in the mcause
definition in the privspec.

Also make a separate table for \Faction instead of repeating its
definition over and over.

* Allow more interrupts/causes to be triggered on.

Clarified use of \Faction a bit as well.

* Incorporate review feedback.

* Attempt to clarify u/s/m

* Break exception trigger into two triggers.

This avoids using tdata3, which helps with certain implementations where
trigger types are writable.

* Rebuild pdf.

* Make trigger types writable.

This gives a hardware more options when implementing triggers. Instead
of requiring separate triggers for separate types, they could opt to
reuse the configuration storage.

* Unimplemented tinfo traps, doesn't read 0.

* Update detection algorithm for previous commit

* Explain priority assignment rationale.

Issue #276.

* Make cause priorities suggested.

This fixes a corner case when chaining triggers, #275.

* Reading tselect doesn't guarantee a valid trigger.

* Clarify algorithm to enumerate triggers.

* Reword debuggers reading back tselect.

resethaltreq: Proposal for forcing a hart into debug mode out of reset

resethaltreq: add a bit to dmstatus to indicate whether this functionality is present

resethaltreq: Modify the proposal to have set/clear bits

resethaltreq: correct some usages of debugger->Debug Module

Issue #264.

mprven: Add a bit to enable MPRV to take effect in debug mode

MPRVEN: minor text edits

Appendix: correct & clarify what debugger/DM do

Remove merge commits from the changelog

* Debuggers must not write sbcs while sbbusy is set

Issue #258

* Grammar.