Re: [PATCH v2 10/27] ftrace: Add subops logic to allow one ops to manage many

[Masami Hiramatsu (Google) <mhiramat@kernel.org>]

On Sun, 2 Jun 2024 22:06:13 -0400
Steven Rostedt <rostedt@goodmis.org> wrote:

> > > +/* Make @ops trace evenything except what all its subops do not trace */
> > > +static struct ftrace_hash *intersect_hashes(struct ftrace_ops *ops)
> > > +{
> > > +	struct ftrace_hash *new_hash = NULL;
> > > +	struct ftrace_ops *subops;
> > > +	int size_bits;
> > > +	int ret;
> > > +
> > > +	list_for_each_entry(subops, &ops->subop_list, list) {
> > > +		struct ftrace_hash *next_hash;
> > > +
> > > +		if (!new_hash) {
> > > +			size_bits = subops->func_hash->notrace_hash->size_bits;
> > > +			new_hash = alloc_and_copy_ftrace_hash(size_bits, ops->func_hash->notrace_hash);
> > > +			if (!new_hash)
> > > +				return NULL;  
> > 
> > If the first subops has EMPTY_HASH, this allocates small empty hash (!= EMPTY_HASH)
> > on `new_hash`.
> 
> Could we just change the above to be: ?
> 
> 			new_hash = ftrace_hash_empty(ops->func_hash->notrace_hash) ? EMPTY_HASH :
> 				alloc_and_copy_ftrace_hash(size_bits, ops->func_hash->notrace_hash);
> 			if (!new_hash)
> 				return NULL;  

Yeah, and if new_hash is EMPTY_HASH, we don't need looping on the rest of
the hashes, right?

> 
> 
> > 
> > > +			continue;
> > > +		}
> > > +		size_bits = new_hash->size_bits;
> > > +		next_hash = new_hash;  
> > 
> > And it is assigned to `next_hash`.
> > 
> > > +		new_hash = alloc_ftrace_hash(size_bits);
> > > +		ret = intersect_hash(&new_hash, next_hash, subops->func_hash->notrace_hash);  
> > 
> > Since the `next_hash` != EMPTY_HASH but it is empty, this keeps `new_hash`
> > empty but allocated.
> > 
> > > +		free_ftrace_hash(next_hash);
> > > +		if (ret < 0) {
> > > +			free_ftrace_hash(new_hash);
> > > +			return NULL;
> > > +		}
> > > +		/* Nothing more to do if new_hash is empty */
> > > +		if (new_hash == EMPTY_HASH)  
> > 
> > Since `new_hash` is empty but != EMPTY_HASH, this does not pass. Keep looping on.
> > 
> > > +			break;
> > > +	}
> > > +	return new_hash;  
> > 
> > And this will return empty but not EMPTY_HASH hash.
> > 
> > 
> > So, we need;
> > 
> > #define FTRACE_EMPTY_HASH_OR_NULL(hash)	(!(hash) || (hash) == EMPTY_HASH)
> > 
> > if (FTRACE_EMPTY_HASH_OR_NULL(subops->func_hash->notrace_hash)) {
> > 	free_ftrace_hash(new_hash);
> > 	new_hash = EMPTY_HASH;
> > 	break;
> > }
> > 
> > at the beginning of the loop.
> > Also, at the end of the loop,
> > 
> > if (ftrace_hash_empty(new_hash)) {
> > 	free_ftrace_hash(new_hash);
> > 	new_hash = EMPTY_HASH;
> > 	break;
> > }

And we still need this (I think this should be done in intersect_hash(), we just
need to count the number of entries.) 

> > 
> > > +}
> > > +
> > > +/* Returns 0 on equal or non-zero on non-equal */
> > > +static int compare_ops(struct ftrace_hash *A, struct ftrace_hash *B)  
> > 
> > nit: Isn't it better to be `bool hash_equal()` and return true if A == B ?
> 
> Sure. I guess I was thinking too much of strcmp() logic :-p

Yeah, it's the curse of the C programmer :( (even it is good for sorting.)

Thank you,

> 
> > 
> > Thank you,
> 
> Thanks for the review.
> 
> -- Steve
> 
> > 
> > > +{
> > > +	struct ftrace_func_entry *entry;
> > > +	int size;
> > > +	int i;
> > > +
> > > +	if (!A || A == EMPTY_HASH)
> > > +		return !(!B || B == EMPTY_HASH);
> > > +
> > > +	if (!B || B == EMPTY_HASH)
> > > +		return !(!A || A == EMPTY_HASH);
> > > +
> > > +	if (A->count != B->count)
> > > +		return 1;
> > > +
> > > +	size = 1 << A->size_bits;
> > > +	for (i = 0; i < size; i++) {
> > > +		hlist_for_each_entry(entry, &A->buckets[i], hlist) {
> > > +			if (!__ftrace_lookup_ip(B, entry->ip))
> > > +				return 1;
> > > +		}
> > > +	}
> > > +
> > > +	return 0;
> > > +}
> > > +  
> > 
> > 
> 


-- 
Masami Hiramatsu (Google) <mhiramat@kernel.org>