On 03/29/2017 01:25 AM, Masami Hiramatsu wrote:

On Wed, 29 Mar 2017 08:30:05 +0200
Ingo Molnar [off-list ref] wrote:

quoted

* Masami Hiramatsu [off-list ref] wrote:

quoted

@@ -1824,6 +1823,30 @@ void unregister_jprobes(struct jprobe **jps, int num)
 EXPORT_SYMBOL_GPL(unregister_jprobes);
 
 #ifdef CONFIG_KRETPROBES
+
+/* Try to use free instance first, if failed, try to allocate new instance */
+struct kretprobe_instance *kretprobe_alloc_instance(struct kretprobe *rp)
+{
+	struct kretprobe_instance *ri = NULL;
+	unsigned long flags = 0;
+
+	raw_spin_lock_irqsave(&rp->lock, flags);
+	if (!hlist_empty(&rp->free_instances)) {
+		ri = hlist_entry(rp->free_instances.first,
+				struct kretprobe_instance, hlist);
+		hlist_del(&ri->hlist);
+	}
+	raw_spin_unlock_irqrestore(&rp->lock, flags);
+
+	/* Populate max active instance if possible */
+	if (!ri && rp->maxactive < KRETPROBE_MAXACTIVE_ALLOC) {
+		ri = kmalloc(sizeof(*ri) + rp->data_size, GFP_ATOMIC);
+		if (ri)
+			rp->maxactive++;
+	}
+
+	return ri;
+}
 /*
  * This kprobe pre_handler is registered with every kretprobe. When probe
  * hits it will set up the return probe.

@@ -1846,14 +1869,8 @@ static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
 	}
 
 	/* TODO: consider to only swap the RA after the last pre_handler fired */
-	hash = hash_ptr(current, KPROBE_HASH_BITS);
-	raw_spin_lock_irqsave(&rp->lock, flags);
-	if (!hlist_empty(&rp->free_instances)) {
-		ri = hlist_entry(rp->free_instances.first,
-				struct kretprobe_instance, hlist);
-		hlist_del(&ri->hlist);
-		raw_spin_unlock_irqrestore(&rp->lock, flags);
-
+	ri = kretprobe_alloc_instance(rp);
+	if (ri) {
 		ri->rp = rp;
 		ri->task = current;
 

@@ -1868,13 +1885,13 @@ static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
 
 		/* XXX(hch): why is there no hlist_move_head? */
 		INIT_HLIST_NODE(&ri->hlist);
+		hash = hash_ptr(current, KPROBE_HASH_BITS);
 		kretprobe_table_lock(hash, &flags);
 		hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
 		kretprobe_table_unlock(hash, &flags);
-	} else {
+	} else
 		rp->nmissed++;
-		raw_spin_unlock_irqrestore(&rp->lock, flags);
-	}
+
 	return 0;
 }
 NOKPROBE_SYMBOL(pre_handler_kretprobe);

So this is something I missed while the original code was merged, but the concept 
looks a bit weird: why do we do any "allocation" while a handler is executing?

That's fundamentally fragile. What's the maximum number of parallel 
'kretprobe_instance' required per kretprobe - one per CPU?

It depends on the place where we put the probe. If the probed function will be
blocked (yield to other tasks), then we need a same number of threads on
the system which can invoke the function. So, ultimately, it is same
as function_graph tracer, we need it for each thread.

Isn't it also possible that the function may be reentrant?  Whether by
plain recursion or an interrupt call, this leads to multiple live
instances even for a given thread.