2c241bd35e6f ("perf symbols: Make sym->end be the first address after the symbol range")

2c241bd35e6f ("perf symbols: Make sym->end be the first address after the symbol range")

[Borislav Petkov]

Hi guys,

so I'm looking at

2c241bd35e6f ("perf symbols: Make sym->end be the first address after the symbol range")

Do I see it correctly that with it, we cannot find zero-sized symbols in
symbols__find() or was it never meant to do so?

symbols__find():

	...

        while (n) {
                struct symbol *s = rb_entry(n, struct symbol, rb_node);

                if (ip < s->start)
                        n = n->rb_left;
                else if (ip >= s->end)
                        n = n->rb_right;
                else
                        return s;
        }

Imagine ip == s->start == s->end:

...
	symbols__find: looking at 0xffffffff81000000, 0xffffffff81000000, st_size: 0
...

I'm playing with this in my own project and this way of doing the
comparison above won't find ELF objects like the one above in vmlinux,
for example.

Btw, there are more than one objects on that address:

$ readelf -a vmlinux | grep ffffffff81000000
     1: ffffffff81000000     0 SECTION LOCAL  DEFAULT    1
 46550: ffffffff81000000     0 NOTYPE  GLOBAL DEFAULT    1 _text
 67061: ffffffff81000000     0 NOTYPE  GLOBAL DEFAULT    1 startup_64

And I think you'd probably want to have some way of restarting the
traversal through the rbtree to find all and have the higher level
callers filter out the symbols depending on what they're looking for,
i.e. NOTYPE, FUNC, OBJECT, range or exact match, etc, etc...

I've done this here locally and it seems to work so far, need to hammer
on it longer. But basically I'm passing through @cur the element I've
just found so that it can give me the next one at that @ip.

Hmmm...

/*
 * @cur is the current symbol from which to start searching
 */
static struct symbol *sym_find(struct rb_root *root, u64 ip, struct symbol *cur)
{
        struct rb_node *n;

        if (!root)
                return NULL;

        n = root->rb_node;

        while (n) {
                u64 sym_start, sym_end;
                struct symbol *s = rb_entry(n, struct symbol, rb_node);

                if (cur && s == cur) {
                        n = rb_next(n);
                        continue;
                }

                sym_start = SYM_START(s);
                sym_end   = SYM_END(s);

                if (ip < sym_start)
                        n = n->rb_left;
                else if (ip > sym_end)
                        n = n->rb_right;
                else
                        return s;
        }
        return NULL;
}

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--