into wailua.hpl.hp.com:/home/davidm/src/unwind/libunwind
2004/08/27 09:33:38+02:00 ibm.com!masbock
[PATCH] x86-64: fix unw_get_proc_info()
The unw_get_proc_info() function from my x86_64 patch is bogus. It
doesn't update the passed in proc info. Here is a fix for it.
Signed-off-by: Max Asbock <masbock@us.ibm.com>
Signed-off-by: David Mosberger <davidm@hpl.hp.com>
BKrev: 41348449Xbf4YtlvXOTgwFap8u4eug
execution in a function with a floating stacked-register frame
(i.e., a function whose stacked registers are _not_ backed by
valid memory).
BKrev: 4134842er3hI9a53izod1_4UXvJycg
For ia64, if the longjmp is crossing a signal-frame, install
the new signal mask by updating the sc_mask member in the
oldest signal-frame.
(Logical change 1.259)
(rbs_cover_and_flush): Take additional arguments "dirty_partition",
"dirty_rnat", and "bspstore". When it is necessary to copy
backing-store words, copy them to "dirty_partition" instead of
the target rbs, which may be full already.
(Logical change 1.259)
remote_lookup() may return the wrong table entry or an invalid entry
combining the start address from one entry with the fde_offset from
another. The similar search in lookup() finds the correct entry, but
may perform unnecessary comparisons to get there. Also fix the
prototype of remote_lookup().
Signed-off-by: Mark F Young <marky@model.com>
Signed-off-by: David Mosberger <davidm@hpl.hp.com>
BKrev: 412af9f4qbo3tgB2jDrMsp8-ODj9iA
in the right order. This should fix a problem where the wrong library
was picked up if an old libunwind was installed already. Reported
by HJ Lu.
BKrev: 4125dff9XwKbhk-7KOb0sPYHpZHbAA
from lib_LTLIBRARIES_cdep to here.
(libunwind_la_SOURCES_x86_64_common): Mention x86_64/ucontext_i.h so
it gets included in the tar-ball, too.
(lib_LTLIBRARIES): List $(lib_LTLIBRARIES_cdep_setjmp) after the
other libraries.
(Logical change 1.254)
to work. Also, bt works now correctly, except that it still
crashes when attempting to unwind across a signal-handler on
an alternate signal-stack.
BKrev: 4125df2334Is8GIamgpMJ-630S9EBw
printing them. Avoids compiler-warning for 64-bit targets.
(apply_reg_state): Handle the special case where the stack-pointer
plays the role of the CFA and the stack-pointer hasn't
been saved. This is based on a patch by Max Asbock.
(Logical change 1.253)
In theory, this may not be needed. In practice, I find that
Red Hat Enterprise Linux AS release 3, the _start() routine has
no unwind-info, but we need to be able to unwind into this
routine to find the end-of-frame-chain marker (RBP == 0).
(Logical change 1.253)
