Re: [PATCH] mm: vmalloc: annotate find_vmap_area_exceed_addr_lock() for lockdep

[Uladzislau Rezki <urezki@gmail.com>]

Hello, Jens, Omar!

> On Tue, Mar 26, 2024 at 04:24:01PM -0600, Jens Axboe wrote:
> > On 3/26/24 3:25 PM, Jens Axboe wrote:
> > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> > > index 22aa63f4ef63..26a69fa6809c 100644
> > > --- a/mm/vmalloc.c
> > > +++ b/mm/vmalloc.c
> > > @@ -1032,7 +1032,7 @@ find_vmap_area_exceed_addr_lock(unsigned long addr, struct vmap_area **va)
> > >  	for (i = 0; i < nr_vmap_nodes; i++) {
> > >  		vn = &vmap_nodes[i];
> > >  
> > > -		spin_lock(&vn->busy.lock);
> > > +		spin_lock_nested(&vn->busy.lock, i);
> > >  		va_lowest = __find_vmap_area_exceed_addr(addr, &vn->busy.root);
> > >  		if (va_lowest) {
> > >  			if (!va_node || va_lowest->va_start < (*va)->va_start) {
> > 
> > Omar said he tested this and ran into lockdep complaining as it only
> > supports 8 subclasses. So this patch can't work, but that still leaves
> > the current kernel code buggy...
> > 	
> It is a bit tricky. Let me rewrite it so a lockdep does not complain.
> 
> Thank you for your report.
> 

Could you please check and test below? It is based on latest 6.9-rc1 tip.
I have reworked it a bit and now it does not hold two locks so the lockdep
should not complain.

<snip>
diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 22aa63f4ef63..9b1a41e12d70 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -989,6 +989,27 @@ unsigned long vmalloc_nr_pages(void)
 	return atomic_long_read(&nr_vmalloc_pages);
 }
 
+static struct vmap_area *__find_vmap_area(unsigned long addr, struct rb_root *root)
+{
+	struct rb_node *n = root->rb_node;
+
+	addr = (unsigned long)kasan_reset_tag((void *)addr);
+
+	while (n) {
+		struct vmap_area *va;
+
+		va = rb_entry(n, struct vmap_area, rb_node);
+		if (addr < va->va_start)
+			n = n->rb_left;
+		else if (addr >= va->va_end)
+			n = n->rb_right;
+		else
+			return va;
+	}
+
+	return NULL;
+}
+
 /* Look up the first VA which satisfies addr < va_end, NULL if none. */
 static struct vmap_area *
 __find_vmap_area_exceed_addr(unsigned long addr, struct rb_root *root)
@@ -1025,47 +1046,40 @@ __find_vmap_area_exceed_addr(unsigned long addr, struct rb_root *root)
 static struct vmap_node *
 find_vmap_area_exceed_addr_lock(unsigned long addr, struct vmap_area **va)
 {
-	struct vmap_node *vn, *va_node = NULL;
-	struct vmap_area *va_lowest;
+	unsigned long va_start_lowest;
+	struct vmap_node *vn;
 	int i;
 
-	for (i = 0; i < nr_vmap_nodes; i++) {
+repeat:
+	for (i = 0, va_start_lowest = 0; i < nr_vmap_nodes; i++) {
 		vn = &vmap_nodes[i];
 
 		spin_lock(&vn->busy.lock);
-		va_lowest = __find_vmap_area_exceed_addr(addr, &vn->busy.root);
-		if (va_lowest) {
-			if (!va_node || va_lowest->va_start < (*va)->va_start) {
-				if (va_node)
-					spin_unlock(&va_node->busy.lock);
-
-				*va = va_lowest;
-				va_node = vn;
-				continue;
-			}
-		}
+		*va = __find_vmap_area_exceed_addr(addr, &vn->busy.root);
+
+		if (*va)
+			if (!va_start_lowest || (*va)->va_start < va_start_lowest)
+				va_start_lowest = (*va)->va_start;
 		spin_unlock(&vn->busy.lock);
 	}
 
-	return va_node;
-}
-
-static struct vmap_area *__find_vmap_area(unsigned long addr, struct rb_root *root)
-{
-	struct rb_node *n = root->rb_node;
+	/*
+	 * Check if found VA exists, it might it is gone away.
+	 * In this case we repeat the search because a VA has
+	 * been removed concurrently thus we need to proceed
+	 * with next one what is a rare case.
+	 */
+	if (va_start_lowest) {
+		vn = addr_to_node(va_start_lowest);
 
-	addr = (unsigned long)kasan_reset_tag((void *)addr);
+		spin_lock(&vn->busy.lock);
+		*va = __find_vmap_area(va_start_lowest, &vn->busy.root);
 
-	while (n) {
-		struct vmap_area *va;
+		if (*va)
+			return vn;
 
-		va = rb_entry(n, struct vmap_area, rb_node);
-		if (addr < va->va_start)
-			n = n->rb_left;
-		else if (addr >= va->va_end)
-			n = n->rb_right;
-		else
-			return va;
+		spin_unlock(&vn->busy.lock);
+		goto repeat;
 	}
 
 	return NULL;
<snip>

Thank you!

--
Uladzislau Rezki