BSD Commit Log Search

   lint: remove temporary code to identify floating point bug

   On sparc64, SoftFloat 2a invoked undefined behavior, returning wrong
   results. The code added here didn't trigger, as the ATF_SH environment
   variable was not set in the real test run.

   See tests/lib/libc/t_long_double.c for more specialized tests to
   pinpoint the exact problem.

   tests/libc: clean up test for long double

   Previously, the test had invoked undefined behavior when converting long
   double to uint64_t, by using input values outside the portable range.
   Explicitly mark this part of the test and make it easy to disable.

   x68k_init() -> machine_init()

   Use machine_init_common().

   Use machine_init_common().

   Use machine_init_common().

   Use pmap_kenter_pa() to map the boot info structure.

   Use machine_init_common().

   machine_init_common() now does the job of loading phys mem into
   the VM manage page pool.

   Add a machine_init_common() routine that will be the holding tank
   for common tasks across m68k machine_init() routines.  For now, it
   only initializes the message buffer, and only on the seglist.h-using
   platforms (mvme68k, next68k, virt68k).

   Note that there will be some minor code duplication (as dead code)
   as the various platforms make the transition to using this new routine,
   but that seems preferable to large unweildy diffs across multiple platforms.

   Centralize the declaration of msgbufpa.  Not all m68k platforms use this,
   but it's part of forthcoming unification changes, and does no harm in
   the meantime.

   As part of this, a hook for Sun platforms (__HAVE_M68K_PRIVATE_MSGSBUF)
   is provided to handle the scenario somewhat unique to Sun2, but where
   the pattern is also present in Sun3/Sun3x -- On Sun2, PROM mappings of
   4 (2K) pages are used for the message buffer, and there is technically
   no guarantee that those pages will be physically contiguous.  Instead,
   the existing PageMap entries are fixed up during VM bootstrap to ensure
   that the permissions are correct, but the existing VA->PA mappings
   remain.

   This PageMap treatment is also done on Sun3, although since the Sun3
   page size is 8K, only one page is used.  Similarly, on Sun3x, we just
   use the first physical (8K) page and create our own mapping for it.
   However, it's easiest just to treat all the Sun platforms the same
   and embrace the lovable little oddballs that they are.

   Centralize the declaration of msgbufaddr for m68k platforms.

   More catch-up with other m68k platforms:
   - pmap_bootstrap() -> pmap_bootstrap1()
   - Pass the updated "nextpa" to machine_init(), and use it to calculate
     avail_start and avail_end there, rather than in pmap_bootstrap1().

   PR/60150: Izumi Tsutsui: Increase alignment so that libuv can use the bottom
   3 bits.

   fic_init() -> machine_init()

   More catch-up needs to be done here.

   Get rid of MSGBUFOFF; it's always defined as 0, and thus doesn't
   actually do anything.  Initialize msgbufaddr when the VAs are
   actually allocated.

   fix various typos in comments.

   Wrap/Un-wrap UVMHIST_LOG lines. NFCI.

   Be consistent around pte_nv_entry definition and usage.

   NFCI.
    CVS: ----------------------------------------------------------------------

   cut.c: minor KNF and other formatting tweaks (NFCI)

   s/liner/linear/ in comment.

   Remove a global that is not, in fact, used in init_main.c, despite what
   the comment claims.

   Consolidate the declaration of phys_map.

   libc: fix undefined behavior in int64_to_float128

   When negating a signed integer, the integer must not be LLONG_MIN, as
   negating that value would not change the sign. GCC's -ftrapv mode
   detects cases like this reliably.

   In this case, on sparc64 with -O2 (but not -O0 or -O1),
   countLeadingZeros64(LLONG_MIN) returned 48 instead of the desired 0, and
   int64_to_float128(LLONG_MIN) returned -0x1p+14 instead of the desired
   -0x1p+63.

   The same pattern is used in several other conversion functions. These
   will be fixed in a follow-up commit once the lint tests for msg_380 and
   msg_381 on sparc64 work again.

   cut.1: add cross-reference to colrm(1)

   Consolidate the declaration of some globals.  More work to be done here.

   Align loading of physical memory with virt68k, and stop keeping two
   copies of this information (one in the phys_seg_list[] and one in
   mem_clusters[]; the latter is now gone).

   Use the data structure from <m68k/seglist.h> to describe the memory
   configuration extracted from linux bootinfo.  Align virt68k's memory
   loading with mvme68k (which will also get some changes in a follow-on
   commit).  This will be used as a template to eventually merge this
   code for all m68k platforms.

   PR 60154: do not print arbitrary control characters when printing
   file names while checking hashes with the -c option.

   Move seglist.h into a common m68k location.