On Mon, 11 Dec 2000 17:09:53 -0800,
"John Hawkes" <hawkes@xxxxxxxxxxxx> wrote:
>From: "Keith Owens" <kaos@xxxxxxxxxxxxxxxxx>
>...
>> For consistency I will add frame pointers to the routines
>> that do not have it but recommend that frame pointers not be used.
>
>As it turns out, kernprof's mcount functionality claims to need frame
>pointers (to the point where the high-level kernel Makefile turns on
>CONFIG_FRAME_POINTER if CONFIG_MCOUNT is set), yet the current version
>0.9 of the kernprof patch doesn't mimic what the kdb patch does in
>semaphore.c and insert the frame pointer instructions into the entry and
>exit of those selected asm routines. Unless someone tells me I'm wrong,
>I'll add those CONFIG_FRAME_POINTER instructions into the kernprof
>patch.
There have been claims in the past that mcount() gets the wrong return
address on ix86 if frame pointers are not available. I tend to trust
the people who made those claims but I could never reproduce it myself.
|