While the ABI (application binary interface) recommends that the
frame pointer register (e.g., %ebp on i386) contain the address of
the current stack frame, it is not mandatory. A called routine is
only required to save and restore the register. Otherwise, it is free
to use the register for other purposes. Since there are so few
registers on Intel processors, it is very tempting to use the frame
pointer register for other things.
The stack frame is mostly needed only by meta-tools: debuggers,
performance samplers, and so on. It might be needed for some things
inside a program, such as exception handling. However, a called
routine that wants to throw an exception could restore the register
and then throw the exception. Debuggers, samplers, and similar tools
may try to use the frame pointer register to unwind the stack, but
they should be prepared to accept failure if it is not valid.
These messages are telling you the sampler encountered a frame
pointer register being used for something else, so it cannot gather
further information about that particular sample. The sampler should
at least collect the program counter where this occurred, from which
the executing routine can be determined.
GCC usually maintains the frame pointer register as a frame pointer
but can be told not to with the -fomit-frame-pointer and -momit-leaf-
frame-pointer options. High-performance routines may be compiled this
way, since making the extra register available can reduce register
spill. Routines in the Accelerate library commonly use the register
for other purposes.
