I noticed two things:

- gdb reads using PT_READ_I, but writes using PT_WRITE_D. This is
  inconsistent. It should either write using PT_WRITE_I, or read using
  PT_READ_D, and have a seperate call to use for installing and removing
  breakpoints that uses the _I calls (either is fine with NetBSD).

- even if the above is fixed, we need to do something about our ptrace()
  call. It occurs to me that a working and not-too-wrong solution would be to:

* make PT_WRITE_I call the PT_WRITE_D function, then call a macro like this:

#include <machine/pmap.h>

#ifndef PMAP_SYNC_PROC_CACHES
#define PMAP_SYNC_PROC_CACHES(p, addr, len)
#endif

...

	case PT_WRITE_I:
		/* PT_WRITE_D code, then */
		PMAP_SYNC_PROC_CACHES(p, addr, len);
		break;

PMAP_SYNC_PROC_CACHES would call appropriate functions where necessary.

The potential VAC machine performance problem Charles that mentioned is not
handled by this.

Regards,
	Ignatios