get_scratch_regs in bk 2.4 unwind.c
The bk version of unwind.c has some changes from my patch which do not look right.
static inline struct pt_regs *get_scratch_regs (struct unw_frame_info *info)
{
if (!info->pt) {
/* This should not happen with valid unwind info. */
UNW_DPRINT(0, "unwind.%s: bad unwind info: resetting info->pt\n", __FUNCTION__);
info->pt = info->sp - 16;
}
UNW_DPRINT(3, "unwind.%s: sp 0x%lx pt 0x%lx\n", __FUNCTION__, info->sp, info->pt);
return info->pt;
}__FUNCTION__ will always print get_scratch_regs which is of no use, we need the calling function. My patch passed in the calling function name as a parameter. Please revert to passing in the function name or make get_scratch_regs a #define so it gets the calling function name. Also I printed the new value of info->pt, it is useful when debugging bad unwind data.
unw_access_gr has
/* access a scratch register */
if (!info->pt) {
UNW_DPRINT(0, "unwind.%s: no pt-regs; cannot access r%d\n",
__FUNCTION__, regnum);
return -1;
}
pt = get_scratch_regs(info);Why the test for !info->pt? No other use of get_scratch_regs has that test, unw_access_[abf]r will continue with whatever data get_scratch_regs returns, using pt_regs on top of stack if info->pt is undefined. The code is inconsistent.
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Keith Owens
Keith> UNW_DPRINT(3, "unwind.%s: sp 0x%lx pt 0x%lx\n", __FUNCTION__, info->sp, info->pt);
Keith> __FUNCTION__ will always print get_scratch_regs which is of no use, we Keith> need the calling function. My patch passed in the calling function Keith> name as a parameter. Please revert to passing in the function name or Keith> make get_scratch_regs a #define so it gets the calling function name. Keith> Also I printed the new value of info->pt, it is useful when debugging Keith> bad unwind data.
We're not going to pass around arguments that are only used for debugging. If you want to, we could print __builtin_return_address(0) along with the function name (or we could use the kernel symbol table to print the symbolic name).
Keith> unw_access_gr has
Keith> /* access a scratch register */
Keith> if (!info->pt) {
Keith> UNW_DPRINT(0, "unwind.%s: no pt-regs; cannot access r%d\n",
Keith> __FUNCTION__, regnum);
Keith> return -1;
Keith> }
Keith> pt = get_scratch_regs(info); Keith> Why the test for !info->pt? No other use of get_scratch_regs Keith> has that test, unw_access_[abf]r will continue with whatever Keith> data get_scratch_regs returns, using pt_regs on top of stack Keith> if info->pt is undefined. The code is inconsistent.
I agree. Looks like something went wrong during the merge. I'll fix that.
--david