BSD Commit Log Search

   We should not be intermingling ATC/TLB and I/D cache operations, but
   for now ensure that the TBI*() functions are only made available for
   Moto and/or HP MMU configurations.

   Decorate all of the legacy pmap_bootstrap1() routines with
   __attribute__((no_instrument_function)) rather than using
   compile-with "${NOPROF_C}".

   Consolidate a bunch of sub-includes of various config definitions
   that were extremely-but-not-universally-common across the various
   m68k platforms into files.m68k.

   Apparently we have kgdb_machdep.c and kgdb_m68k.c, the latter is already
   in files.m68k, and the former everyone uses the same one, so move it to
   files.m68k, too.

   Obviously, more cleanup in this area is going to be needed.

   No need for everyone to how their own entry for vm_machdep.c

   No need for everyone to how their own entry for procfs_machdep.c

   sun2 and sun3 don't *really* need their own trimmed-down copy of
   sys_machdep.c; the common version was already perfectly suitable
   for sun3, and can be easily conditionally-trimmed-down for the
   68010 case.

   (Sharing is caring!)

   Add some missing CPU and MMU types.

   Correct typo in previous (s/1/i/) and remove gratuitous wsp (build fix)

   Add an MMU_CUSTOM mmutype value, used for bespoke MMUs that don't
   fall into one of the classical categories.

   Don't default mmutype to UNKNOWN on 68010; it is likely to be statically-
   initialized to something else already.

   Don't use mmutype as a proxy for cputype when deciding what to do
   with the cache.

   Account for the Sun3x situation in previous.

   Put all of the pmap_motorola.c vs pmap_68k.c logic into files.m68k.

   Move the do-nothing(-except-DIAGNOSTICly-panic) m68881_save/restore
   stubs to switch_subr.s so that future 68010 ports don't have to
   declare them on their own.

   switch_subr.s is now a standlone file, so the boilerplate in each
   locore.s that pulled it in is unnecessary.

   Speed up IDE device probe by up to 3 seconds when drives are absent.

   Fix pcn MAC address read on QEMU and possibly some other emulated hardware.

   Fix alpha boot-time hang on unpatched QEMU by accommodating its broken mcclock.

   Support poweroff on RISCV

   All the riscv ldscripts were marked with compat, causing them to be
   omitted if MKCOMPAT=no, but they are definitely not all compat; safest
   to just not mark them as compat until such time as we can straighten
   out which are and which aren't.

   Add alignment constraint where ASCII strings are at the end of asm functions;
   it's not clear whether this is a toolchain bug, but the combination of odd
   length strings right at the function end, the toolchain's "linker
   relaxation", and the technically variable length of encoded riscv insns
   (they can be 2 bytes, but 4 byte instructions must be 4 byte aligned) seems
   to be causing instructions in the next function to be out of alignment,
   especially with -Os.

   Our security(7) manual page has long documented the functionality
   enabled by secmodel_extensions.  It's thus not really optional, and
   declaring it so causes an odd mismatch in behavior between MODULAR and
   non-modular kernels.  It's tiny, just add it to std.

   add ld to hppa MAKEDEV; without it, virtio disks are unusable

   Update PMU reporting

   join the party and fix similar typos in few more comments.

   Fix another tyop ... in the same comment line

   s/encrytp/encrypt/ in a function name.

   Fix tyop.

   switch_subr.s is now a standlone file, so the boilerplate in each
   locore.s that pulled it in is unnecessary.