The unw_resume() routine resumes execution at the stack frame identified by cp. The behavior of this
routine differs slightly for local and remote unwinding.
For local unwinding, unw_resume() restores the machine state and then directly resumes execution in the
target stack frame. Thus unw_resume() does not return in this case. Restoring the machine state normally
involves restoring the ``preserved'' (callee-saved) registers. However, if execution in any of the stack
frames younger (more deeply nested) than the one identified by cp was interrupted by a signal, then
unw_resume() will restore all registers as well as the signal mask. Attempting to call unw_resume() on a
cursor which identifies the stack frame of another thread results in undefined behavior (e.g., the
program may crash).
For remote unwinding, unw_resume() installs the machine state identified by the cursor by calling the
access_reg and access_fpreg accessor callbacks as needed. Once that is accomplished, the resume accessor
callback is invoked. The unw_resume routine then returns normally (that is, unlikely for local unwinding,
unw_resume will always return for remote unwinding).
Most platforms reserve some registers to pass arguments to exception handlers (e.g., IA-64 uses r15-r18
for this purpose). These registers are normally treated like ``scratch'' registers. However, if libunwind
is used to set an exception argument register to a particular value (e.g., via unw_set_reg()), then
unw_resume() will install this value as the contents of the register. In other words, the exception
handling arguments are installed even in cases where normally only the ``preserved'' registers are
restored.
Note that unw_resume() does not invoke any unwind handlers (aka, ``personality routines''). If a program
needs this, it will have to do so on its own by obtaining the unw_proc_info_t of each unwound frame and
appropriately processing its unwind handler and language-specific data area (lsda). These steps are
generally dependent on the target-platform and are regulated by the processor-specific ABI
(application-binary interface).