(remote_lookup): New function.
(_Uia64_search_unwind_table): Handle UNW_INFO_FORMAT_REMOTE_TABLE.
(_Uia64_find_dyn_list): Take unw_dyn_info_t pointer argument instead of its
members and handle UNW_INFO_FORMAT_REMOTE_TABLE.
(Logical change 1.126)
not needed because we default to "no caching" and if the application
turns on caching via unw_set_caching_policy(), the cache will get
initialized at that time.
(Logical change 1.124)
(get_script_cache): Return NULL if caching is disabled. If
HAVE_ATOMIC_OPS_H is defined, set the busy flag and if
the script cache is busy already, return NULL.
(put_script_cache): Don't allow it to be called for UNW_CACHE_NONE case.
If HAVE_ATOMIC_OPS_H is defined, clear the "busy" flag via
AO_CLEAR.
(ia64_get_cached_proc_info): If we're unable to acquire the cache,
return -UNW_ENOINFO.
(script_init): New function.
(script_new): Replace open code with call to script_init().
(uncached_find_save_locs): New function.
(ia64_find_save_locs): Move code for uncached case to
uncached_find_save_locs(). If get_script_cache() is unable to
lock the script-cache, fall back to uncached_find_save_locs().
(ia64_script_cache_init): Delete.
(Logical change 1.124)
early. Always flush cache after changing the policy.
This is to ensure that the per-thread cache gets initialized
and no stale info remains.
(Logical change 1.118)
(ia64_init): Use unw.lock to protect against races during
initialization. Disable signal delivery during initialization
to ensure thread-safety (e.g., unw_init_local() may call
ia64_init() and the former must be thread-safe).
(Logical change 1.111)
personality-routine and LSDA address and dont' forget to
dereference the personality-routine function-pointer, so we
really get the address of the function-descriptor of the
personality routine.
(callback): Set the segment-base based on the segment that contains
the unwind-table. Normally, that the text segment but in
special cases, such as the kernel's gate DSO, it will be
in a different segment.
(Logical change 1.108)
hash-table, collision-chains, and hash-table in addition
to clearing the script's IP value. The latter wasn't nearly
sufficient because it could cause loops on the collision
chain...
(get_script_cache): Also lock the cache if global caching is in
effect. Take additional "saved_sigmaskp" argument.
After flushing the cache, update the cache generation
here.
(put_script_cache): New function.
(ia64_get_cached_proc_info): Adjust for get_script_cache() and
put_script_cache() changes.
(ia64_find_save_locs): Ditto.
(ia64_script_cache_init): Call flush_script_cache() instead of
opening-coding the equivalent code.
(Logical change 1.96)
state-record.
(ia64_fetch_proc_info): Rename from get_proc_info().
(create_state_record_for): Assert that c->pi_valid is TRUE at the
time of entry. Remove call to get_proc_info().
(Logical change 1.95)
a value that is potentially > 31. Casting it to (unsigned long)
is not enough on 32-bit platforms.
(create_state_record_for): Cast 0xf to (unw_word_t) instead of
(unsigned long) to avoid losing bits on 32-bit platforms.
(Logical change 1.92)
Linux kernel "struct pt_regs" layout.
(update_frame_state): Handle ABI_MARKER_OLD_LINUX_SIGTRAMP and
ABI_MARKRE_OLD_LINUX_INTERRUPT. Cast values of type unw_word_t
to (unsigned long) before printing them.
(Logical change 1.91)
(linux_scratch_loc): Add support for new pt_regs layout.
(access_nat): Check last_abi_marker instead of sigcontext_addr and
address-space ABI.
(Logical change 1.91)
(update_frame_state): Handle ABI_MARKER_LINUX_INTERRUPT by calling linux_interrupt().
If c->abi_marker is non-zero, copy it to c->last_abi_marker.
(Logical change 1.89)
(access_reg) [HAVE_SYS_UC_ACCESS_H]: Sanitize and simplify this routine. Turns out
uc_access(3) works pretty much the way we need it, the only problem is that
the unwind info for getcontext() seems to be missing as of HP-UX 11.22.
(ia64_uc_access_reg): New function.
(ia64_uc_access_fpreg): Ditto.
(ia64_local_addr_space_init): Initialize "abi" member of local address space based
on the platform we're compiling on.
(Logical change 1.87)
have an empty file here, as otherwise we won't get a valid program.
2003/04/24 12:58:16-07:00 mostang.com!davidm
Rename: src/ia64/Lmk_cursor_i-ia64.c -> src/ia64/mk_Lcursor_i-ia64.c
(Logical change 1.87)
(remote_install_cursor): Enable the short-cut for Linux only. Since HP-UX uses
an opaque data-structure for ucontext_t, we can't play the same trick
there. Fix two typos: "> 0" -> ">= 0".
(unw_resume): Handle non-UNW_LOCAL_ONLY case simp;u bu calling
remote_install_cursor().
(Logical change 1.87)
support for HP-UX.
(access_nat): Check for uc-location and pass it on to ia64_put()/ia64_get() if
it is. Handle NaT-bit-saved-in-scratch-register only for Linux.
(ia64_access_reg): Change dprintf() into debug() statement.
(ia64_access_fpreg): Move fph-partition initialization code to ia64_scratch_loc().
Change dprintf() into debug() statement.
(Logical change 1.87)
end up not using it.
(UNWIND_TBL_32BIT): New macro.
(struct unwind_header): Rename "unknown1" member to "header_version".
(tdep_find_proc_info) [HAVE_DLMODINFO]: Don't forget to initialize di.gp.
Check unwind-header version and format.
(Logical change 1.87)
Rename Gmk_cursor to mk_Gcursor, Lmk_cursor to mk_Lcursor so to avoid
scripts/make-L-files overwriting the (handcrafted) local version.
(EXTRA_DIST): Mention libunwind_la_SOURCES_os_hpux.
(Logical change 1.87)
Update code to work both for UNW_LOCAL_ONLY-version of the cursor and the
generic version of the cursor.
(ia64_install_cursor): New macro.
}(Logical change 1.85)
Update code to work both for UNW_LOCAL_ONLY-version of the cursor and the
generic version of the cursor.
(ia64_install_cursor): New macro.
2003/04/23 11:25:31-07:00 mostang.com!davidm
Rename: src/ia64/_Uia64_install_cursor.S -> src/ia64/Ginstall_cursor.S
(Logical change 1.85)
and Linstall_cursor.S.
(Gmk_cursor_i_ia64_SOURCES): Rename from mk_cursor_i_ia64_SOURCES.
(Lmk_cursor_i_ia64_SOURCES): New mcro.
(Gcursor_i.h): Rename rule from cursor_i.h.
(Lcursor_i.h): New rule.
(noinst_PROGRAMS): Rename mk_cursor_i-ia64 to Gmk_cursor_i-ia64 and add
Lmk_cursor_i-ia64.
(BUILT_SOURCES): Ditto.
(Logical change 1.85)
(struct_offset): Delete.
(IA64_LOC_REG): New macro (replaces IA64_LOC() for registers).
(IA64_LOC_ADDR): New macro (replaces IA64_LOC() for addresses).
(IA64_LOC_UC_ADDR): New macro.
(IA64_NULL_LOC): Ditto.
(IA64_IS_NULL_LOC): Ditto.
(IA64_IS_UC_LOC): Ditto.
(struct ia64_state_record): Replace is_signal_frame bit with abi_marker.
(ia64_strloc): New declaration.
(rbs_switch): Change from UNW_ARCH_OBJ() to UNW_OBJ().
(rbs_find_stacked): Ditto.
(rbs_cover_and_flush): Ditto.
(rbs_get_rnat_loc): New inline function.
(rbs_loc): Ditto.
(Logical change 1.84)
Adjust for renaming of Linux sigcontext offset macros.
(check_rbs_switch): Handle Linux-case only if abi_marker==ABI_MARKER_LINUX_SIGTRAMP.
(update_frame_state): For cursor's with a non-zero abi_marker, do special
Linux or HP-UX sigtramp handling as appropriate.
Clear abi_marker instead of is_signal_frame.
(Logical change 1.84)
(emit_nat_info): For registers saved in a general-register, use IA64_INSN_SET_REG
instead of IA64_INSN_SET.
Set insn.dst to register number, instead of unw.preg_index[i].
(compile_reg): Set val directly to register number, instead of unw.preg_index[i].
Use IA64_INSN_INC_PSP instead of IA64_INSN_ADD to pop a fixed stack frame.
(build_script): Handle signal-frames by directly copying abi_marker, rather than
emitting an IA64_INSN_SET instruction.
(run_script): Copy abi_marker into cursor. Rather than updating a word, update
a save-location. Only exceptions are IA64_INSN_INC_PSP and
IA64_INSN_LOAD_PSP.
(Logical change 1.84)
Mask out only the least-significant 2 bits (rather than the least significant
3 bits).
2003/04/22 22:15:48-07:00 mostang.com!davidm
Rename: src/ia64/_Uia64_install_context.S -> src/ia64/_Uia64_install_cursor.S
(Logical change 1.84)
(REASON_SYSCALL): Ditto.
(access_reg): Use __uc_get_reason() to determine how ucontext_t was created.
Handle UNW_IA64_AR_BSP differently depending on whether or not
ucontext_t was created by a syscall. Ditto for UNW_IA64_IP and
UNW_IA64_CFM. Add debug statement.
(Logical change 1.77)
is not defined.
(tdep_find_proc_info) [HAVE_DLMODINFO): Declare new type "struct unwind_header"
to accommodate HP-UX unwind header. Fetch unwind table from this
header.
(Logical change 1.75)
(_UPT_find_proc_info): On ia64, when returning info about the kernel's unwind
table, copy the unwind info into a malloc'd buffer and use
unw_local_addr_space to search the table, since the table is stored
in local memory.
(Logical change 1.71)
Test for HAVE_ASM_PTRACE_OFFSETS_H instead of UNW_TARGET_IA64.
Issue warning for platforms using ttrace() instead of ptrace() (HP-UX
is the only platform that does that).
(Logical change 1.63)
Include <inttypes.h> instead of <stdint.h>. HP-UX doesn't have the latter.
Use "inline" instead of "__inline__" (the former gets properly #define'd by
config.h).
(Logical change 1.63)
(_Uia64_search_unwind_table): Be sure to clear "proc_info" members as documented
by unw_get_proc_info(3).
Include <dlfcn.h> if we have dlmodinfo() (HP-UX only).
(tdep_find_proc_info): Use dl_iterate_phdr() only if we have it (glibc-based
systems). Use dlmodinfo() if we have it (HP-UX).
(Logical change 1.63)
Return -UNW_ENOMEM if string buffer is too small. This makes the routine
compatible with the definition of the unw_get_proc_name(3).
(Logical change 1.63)
(unwi_dyn_remote_find_proc_info): Maintain as->dyn_info_list_addr as a cache of
the most recently obtained dyn-info-list address.
(Logical change 1.61)
(_longjmp): Ensure that we have at least 4 exception-handling args. (This code
will need updating to make it work on x86, where only 2 exception handling
args are available).
(longjmp): If we are compiling with GCC, use an alias-attribute to alias it with
_longjmp(). This is more efficient and works around a gcc-3.2/ia64 bug
which causes bad unwind info when a noreturn function is a last call.
(Logical change 1.59)
(_longjmp): Ensure that we have at least 4 exception-handling args. (This code
will need updating to make it work on x86, where only 2 exception handling
args are available).
(Logical change 1.59)
(SOSLOCK): Remove
(SOSUNLOCK): Ditto.
(LOCK): Ditto
(UNLOCK): Ditto.
(sos_alloc): If HAVE_CMPXCHG is defined, use cmpxchg_ptr() to atomically allocate
memory in a lock-free manner. Otherwise, fall back on disabling signal
delivery and holding the sos_lock during memory allocation.
(sos_free): Remove.
(mempool_init): Initialize pool lock. Replace LOCK/UNLOCK with appropriate
signal-masking & mutex_lock/unlock calls.
(mempool_free): Replace LOCK/UNLOC with appropriate signal-masking & mutex_lock/unlock
calls.
(Logical change 1.59)
(ia64_get_cached_proc_info): Rename from ia64_script_lookup() and modify it so
it directly updates the proc-info if the necessary script is in the cache.
(script_new): Remove locking comments and adjust formatting.
(ia64_find_save_locs): If global caching is enabled, mask signal delivery and
hold the cache lock while accessing the cache.
(ia64_script_cache_init): Initialize the cache lock.
(Logical change 1.59)
(parse_dynamic): Fix it so it actually works.
(create_state_record_for): Initialize return_link_reg to zero.
Don't print "no unwind info" unconditionally when UNW_DEBUG is
in effect.
(Logical change 1.56)
(intern_dyn_info): Skip over padding rather than fetching it.
Start intern_regions() at address of first region, not just at the next
sequential address.
(unwi_dyn_remote_find_proc_info): Allocate "di" dynamically (it needs to persist
until put_unwind_info() is called).
(Logical change 1.56)
Skip over "prev_addr". Read start/end-IP directly into the "di" structure,
not into temporary variables. Skip padding by directly adjusting the
address (don't bother fetching padding values; that only slows things down).
Return 0 only if we really did find dynamic unwind info.
(Logical change 1.55)
(.info): Use @gprel() reloc instead of @segrel() reloc for _U_dyn_info_list.
The latter doesn't work because the unwind-info is in a different segment.
(ia64_init): Clear unw.r0.
(Logical change 1.55)
(desc_prologue): Fix handling of region_start so it doesn't get corrupted by
state-record pushing/popping.
(parse_dynamic): Allow the last region to contain a negative insn_count, indicating
that the region covers the last portion of a procedure.
(Logical change 1.54)
we now switch to the new backing-store immediately.
(rbs_underflow): Remove.
(rbs_find_stacked): No longer assume that c->bsp is contained in the current
rbs. If it is not, first find the backing-store that's backing c->bsp.
(rbs_cover_and_flush): Use rbs_contains() instead of (broken) open code.
Handle the case where the initial bsp points to an RNaT slot.
(Logical change 1.52)
