Thu Jul 25 19:41:31 1996 Fred Fish <fnf@cygnus.com>
for details about these changes. I was going to include the complete
entry in the checkin message but it was apparently so large it caused
cvs to dump core.
and MMALLOC_CHECK macros, and add comment indicating how host dependent
makefile fragment should modify MMALLOC_CFLAGS to not use mmalloc, or
to use it but to not do heap corruption checking.
* gdbserver/Makefile.in: Ditto.
* utils.c (init_malloc): Replace warning() use with direct call of
fprintf_unfiltered, since current_target has not yet been set and thus
we cannot use warning(). If we try to use mmcheck and it fails,
suggest that this configuration needs NO_MMCHECK or MMCHECK_FORCE
defined. Other small mmalloc related cleanups.
* config/sparc/sun4os4.mh (MMALLOC_CFLAGS): Define MMCHECK_FORCE to 1.
* config/alpha/alpha-osf2.mh (MMALLOC_CFLAGS): Set to -DNO_MMCHECK.
* config/sparc/xm-sun4os4.h (MMAP_BASE_ADDRESS, MMAP_INCREMENT):
* config/i386/xm-i386v4.h (MMAP_BASE_ADDRESS, MMAP_INCREMENT):
* config/i386/xm-linux.h (MMAP_BASE_ADDRESS, MMAP_INCREMENT):
* config/m68k/xm-hp300hpux.h (MMAP_BASE_ADDRESS, MMAP_INCREMENT):
* config/m68k/xm-m68kv4.h (MMAP_BASE_ADDRESS, MMAP_INCREMENT);
* config/m68k/xm-sun3os4.h (MMAP_BASE_ADDRESS, MMAP_INCREMENT):
* config/pa/xm-hppah.h (MMAP_BASE_ADDRESS, MMAP_INCREMENT):
* config/sparc/xm-sun4sol2.h (MMAP_BASE_ADDRESS, MMAP_INCREMENT):
Remove obsolete defines.
* config/alpha/alpha-linux.mh (MMALLOC_DISABLE):
* config/alpha/alpha-osf1.mh (MMALLOC_DISABLE):
* config/rs6000/rs6000.mh (MMALLOC_DISABLE):
* config/rs6000/aix4.mh (MMALLOC_DISABLE):
* config/powerpc/aix4.mh (MMALLOC_DISABLE):
* config/powerpc/aix.mh (MMALLOC_DISABLE):
* config/ns32k/ns32km3.mh (MMALLOC_DISABLE):
* config/mips/mipsm3.mh (MMALLOC_DISABLE):
* config/mips/decstation.mh (MMALLOC_DISABLE):
* config/m88k/cxux.mh (MMALLOC_DISABLE):
* config/i386/i386mk.mh (MMALLOC_DISABLE):
* config/i386/i386m3.mh (MMALLOC_DISABLE):
* config/i386/i386gnu.mh (MMALLOC_DISABLE):
Use MMALLOC_CFLAGS instead.
* NEWS: Add Alpha Linux as a new native configuration.
* mdebugread.c (parse_symbol): When we find a malloc() symbol with
return type VOID, assume no debugging info is available for that
object file and patch the return value into VOID *. Otherwise,
operations requiring an implicit call to malloc() will fail.
* infrun.c (wait_for_inferior): The criterion to detect entering a
sigtramp handler is now: (a) the current pc is inside a sigtramp
handler, (b) the previous pc is not in a sigtramp handler, and (c)
the current stack pointer is "inner" than the old one. Condition
(c) is new to avoid mistaking a return from a signal handler into
sigtramp as a new sigtramp invocation.
* dcache.c (struct dcache_block): Declare addr as CORE_ADDR. An
int may not be big enough to hold an address.
(dcache_hit): Ditto.
(dcache_peek_byte): Fix indentation.
* configure.in (alpha-*-linux*): Add target.
* configure: Rebuild
* config/alpha/tm-alpha.h (PROC_DESC_IS_DYN_SIGTRAMP): New macro.
(SET_PROC_DESC_IS_DYN_SIGTRAMP): Ditto.
(DYNAMIC_SIGTRAMP_OFFSET): Ditto.
(SIGCONTEXT_ADDR): Ditto.
(FRAME_PAST_SIGTRAMP_FRAME): Ditto.
* config/alpha/alpha-linux.mh: New file.
* config/alpha/alpha-linux.mt: Ditto.
* config/alpha/nm-linux.h: Ditto.
* config/alpha/tm-alphalinux.h: Ditto.
* config/alpha/xm-alphalinux.h: Ditto.
* config/alpha/xm-alphaosf.h: Renamed from xm-alpha.h.
* config/alpha/alpha-osf1.mh (XM_FILE): Change from xm-alpha.h to
xm-alphaosf.h.
* config/alpha/alpha-osf2.mh: Ditto.
* blockframe.c (find_pc_partial_function): Pass PC to
SIGTRAMP_START and SIGTRAMP_END macros for the benefit of systems
that detect sigtramp code via designated code sequences (as is the
case for Linux/Alpha, for example).
* config/i386/tm-i386bsd.h: Change SIGTRAMP_START and SIGTRAMP_END
to ignore new PC argument.
* config/m68k/tm-hp300bsd.h: Ditto.
* config/vax/tm-vax.h: Ditto.
* alpha-tdep.c (alpha_linux_sigtramp_offset): New function.
(alpha_osf_skip_sigtramp_frame): Ditto.
(push_sigtramp_desc): Ditto.
(alpha_find_saved_regs): Use SIGCONTEXT_ADDR macro to extract
sigcontext address from frame.
(alpha_saved_pc_after_call): When in sigtramp, use
alpha_frame_saved_pc() instead of read-register().
(after_prologue): When inside a dynamically generated sigtramp
function, there is no prologue, so return address of first
instruction.
(alpha_in_prologue): Fix typo in comment.
(find_proc_desc): Use macro DYNAMIC_SIGTRAMP_OFFSET to determine
whether we're inside a dynamicaly generated sigtramp function. If
so, create and push and appropriate procedure descriptor.
(alpha_frame_chain): Use macro FRAME_PAST_SIGTRAMP_FRAME to obtain
the frame past a sigtramp frame (if the current frame is indeed a
sigtramp function).
(init_extra_frame_info): Don't read next frame register off of
stack-pointer when inside a dynamiccaly generated sigtramp.
(alpha_pop_frame): Also unlink and destroy procedure descriptors
created for dynamically generated sigtramp functions.
* alpha-nat.c: When compiling under Linux, include <asm/reg.h> and
<alpha/ptrace.h> instead of <machine/reg.h>
* nlm/aio.h: Protect from multiple inclusions.
* nlm/alpha-io.S: Remove everything we don't need.
* nlm/{alpha-patch.c, alpha-patch.h, alpha-uart.c, alpha-uart.h}:
Remove, no longer needed.
* nlm/alpha.c: Merge in lots of stuff from previous files.
* nlm/alpha.h: Don't #include alpha-patch.h. Make
breakpoint_insn extern.
* Move stuff from alpha-patch.h into here.
* config/alpha/gdbserve.mt (TDEPFILES): Get rid of alpha-uart.o.
NLMCONV_FOR_TARGET. (SUBDIRS): Add nlm target.
* configure.in (powerpc-*-netware*): Automatically configure nlm
subdir.
* nlm/Makefile.in: Add {CC NLMCONV LD}_FOR_TARGET. Remove alpha
specific stuff. Make things more configurable.
* nlm/configure.in: Add powerpc-*-netware* target. Use
gdbserve.mt/cpu.c/cpu.h for target stuff. Get rid of tm/xm/nm.h
files.
* nlm/gdbserve.c: Move Alpha specific stuff into other files.
Remove lots of architecture-specific stuff.
* nlm/gdbserve.def: Add new imports.
* nlm/ppc.c, nlm/ppc.h: New files that contain PowerPC specific code.
* nlm/prelude.c: Don't include libhooks.h, get rid of call to
register library.
* nlm/prelude.o: What was this doing here?
* config/alpha/gdbserve.mt: Defs for alpha nlm stub.
* config/powerpc/gdbserve.mt: Defs for PowerPC nlm stub.
* config/powerpc/ppc-nw.mt: Defs for PowerPC target for GDB.
* config/powerpc/tm-ppc-nw.h: Ditto.
* nlmstub.def: New file, contains imports for 386 nlm stub.
* (nlm_symfile_read): Record bounds of main() so that backtrace
command will know where to stop.
* objfiles.c (objfile_relocate): Relocate entry point/func info
for backtrace as well.
* objfiles.h: Define values for invalid PCs for entry point info.
* symfile.c (init_entry_point_info): Initialize invalid values
with aforementioned macros.
* config/alpha/tm-alphanw.h: Turn on FRAME_CHAIN_VALID_ALTERNATE
to cause backtrace to stop when it gets back to main().
* config/i386/tm-i386nw.h: Ditto.
BROKEN_LARGE_ALLOCA. Do the cleanup before returning, rather than
leaving it on the chain. Reindent much of this function.
* config/sparc/{xm-sun4sol2.h,xm-sun4os4.h},
config/i386/{xm-sun386.h,xm-i386m3.h,xm-i386mach.h},
config/m68k/{sun3os4.h,xm-news.h,xm-hp300hpux.h},
config/ns32k/xm-ns32km3.h: Remove all references to
BROKEN_LARGE_ALLOCA; with the above change it is no longer needed.
* main.c, fork-child.c, many config files: Remove all
SET_STACK_LIMIT_HUGE code; with the above changes it should no
longer be needed.
* symtab.c (lookup_partial_symbol): Use if and abort, not assert.
This avoids __eprintf troubles.
* main.c (main): Surround in #ifndef MAIN_OVERRIDE. Move
initialization code which needs to be called even if we bypass the
command line stuff into gdb_init.
* utils.c (fputs_unfiltered): Surround in #ifndef
FPUTS_UNFILTERED_OVERRIDE.
* Makefile.in (libgdb.a): New target.
* utils.c: Rearrange I/O stuff a bit so that all output goes
through fputs_unfiltered. Use vasprintf; removes arbitrary limit
which made %s not work with arbitrarily large strings.
* printcmd.c (printf_command): Use printf_filtered, not
printf_unfiltered and printf, now that arbitrary limit is gone.
improve remote debugging efficiency. Also fixed problems with
doing function calls for programs with no entry points.
* infcmd.c (run_stack_dummy): Use CALL_DUMMY_ADDRESS instead of
entry_point_address.
* inferior.h (PC_IN_CALL_DUMMY): ditto.
* mdebugread.c (parse_symbol, parse_procedure, parse_external,
parse_lines): Pass section_offsets info to these routines so that
we can relocate symbol table entries upon readin.
* (psymtab_to_symtab_1): Set symtab->primary to tell
objfile_relocate to do relocations for our symbols.
* (ecoff_relocate_efi): New routine to relocate adr field of PDRs
(which hang off of the symbol table).
* Use prim_record_minimal_symbols_and_info instead of
prim_record_minimal_symbols to supply section info to make minimal
symbol relocations work.
* minsyms.c (prim_record_minimal_symbols_and_info): If section is
-1, try to deduce it from ms_type.
* objfiles.c (objfile_relocate): Use ALL_OBJFILE_SYMTABS where
appropriate. Handle relocation of MIPS_EFI symbols special. Also,
add code to relocate objfile->sections data structure.
* remote.c (get_offsets): Use new protocol message to acquire
section offsets from the target.
* (remote_wait): Get rid of relocation stuff. That's all handled
by objfile_relocate now.
* config/alpha/alpha-nw.mt (TM_FILE): Use tm-alphanw.h.
* config/alpha/tm-alpha.h: Define CALL_DUMMY_ADDRESS, and
VM_MIN_ADDRESS.
* config/alpha/tm-alphanw.h: DECR_PC_AFTER_BREAK=0, VM_MIN_ADDRESS=0.
The old way (using #undef in tm.h) was ugly and asking for
trouble, because it makes it possible for some file to use the
wrong definition. Move definition of HOST_CHAR_BIT after definition
of TARGET_CHAR_BIT.
* config/alpha/tm-alpha.h, config/h8300/tm-h8300.h,
config/h8500/tm-h8500.h, config/z8k/tm-z8k.h: Don't undef TARGET_*_BIT
before defining them.
calling vprintf, just make the appropriate calls to printf.
* printcmd.c, config/pa/xm-pa.h, config/mips/xm-makeva.h,
config/alpha/xm-alpha.h, config/m88k/xm-m88k.h: Remove all
traces of makeva*. My apologies to everyone (including me!)
who spent so much time getting it to work on various machines,
but look at the bright side, at least you won't have to do it
again in the future.
* printcmd.c (printf_command): Make a cleanup for val_args (fixes
a memory leak).
host.
* sh-tdep.c (frame_find_saved_regs): Use NUM_REGS rather than hard
wired (and wrong) constant.
* values.c (unpack_long): Add case to unpack when target object is
sizeof(int).
* config/sh/tm-sh.h (REGISTER_NAMES): Know about the news ones the
simulator defines.
Stan Shebs