HTM issues from Anton

[Joel Stanley]

• pdbg core trace should check if the tracing core is in ST mode
• Single issue mode is terribly slow. --start should setup HID0 and –stop should undo it
• --dump vs --analyse
• Mikey suggests a –record which works like perf
• --analyze overwrites an existing file without prompting, but doesn't truncate what is there! A more sane sequence would be to refuse to overwrite unless an -f/–force option is specified, and then we truncate the existing file.
• htm option seems to be missing help text
• htm is hiding behind the mysterious --expert option
• We hardwire the buffers to 1GB at the moment. An option –size argument would be useful, and ideally it would take standard units (–size 4G)

[apopple]

htm is hiding behind the mysterious --expert option

It didn't take @mikey long to convince me that --expert was at best ambiguous. It stems from my own ideas around debugging and pdbg were that it shouldn't ever change system state in "unexpected" ways.

So for example getgpr shouldn't stop instructions executing, but instead tell the user what needs to be done to stop instructions executing if they haven't already been. Reason is that generally you don't want your debugging tools screwing with your system in unexpected ways, but they should help a user discover what to do (whether or not we've succeeded with that is certainly debatable).

So the current thinking is we will make user-friendly versions of commands like getgpr and htm which will put the system in the right state if it isn't already and generally just "do what you want" (for certain values of "you" and "want") and a --safe mode which will only "do what you say".