iommu/iova: Keep count of 32-bit PFNs

Optimise for the case where the 32-bit space is full, so that
allocation attempts fail fast to reduce lock contention.
Signed-off-by: Robin Murphy <robin.murphy@arm.com>

drivers/iommu/iova.c

@@ -56,6 +56,7 @@ init_iova_domain(struct iova_domain *iovad, unsigned long granule,
 	iovad->granule = granule;
 	iovad->start_pfn = start_pfn;
 	iovad->dma_32bit_pfn = 1UL << (32 - iova_shift(iovad));
+	iovad->free_32bit_pfns = iovad->dma_32bit_pfn - start_pfn;
 	iovad->flush_cb = NULL;
 	iovad->fq = NULL;
 	iovad->anchor.pfn_lo = iovad->anchor.pfn_hi = IOVA_ANCHOR;
@@ -114,6 +115,24 @@ int init_iova_flush_queue(struct iova_domain *iovad,
 }
 EXPORT_SYMBOL_GPL(init_iova_flush_queue);
 
+static void insert_update_32bit_pfn(struct iova_domain *iovad, struct iova *new)
+{
+	unsigned long pfn_lo = new->pfn_lo;
+	unsigned long pfn_hi = min(new->pfn_hi, iovad->dma_32bit_pfn);
+
+	if (pfn_lo >= iovad->dma_32bit_pfn)
+		iovad->free_32bit_pfns -= pfn_hi - pfn_lo + 1;
+}
+
+static void delete_update_32bit_pfn(struct iova_domain *iovad, struct iova *free)
+{
+	unsigned long pfn_lo = free->pfn_lo;
+	unsigned long pfn_hi = min(free->pfn_hi, iovad->dma_32bit_pfn);
+
+	if (pfn_lo >= iovad->dma_32bit_pfn)
+		iovad->free_32bit_pfns += pfn_hi - pfn_lo + 1;
+}
+
 static struct rb_node *
 __get_cached_rbnode(struct iova_domain *iovad, unsigned long limit_pfn)
 {
@@ -130,6 +149,8 @@ __cached_rbnode_insert_update(struct iova_domain *iovad, struct iova *new)
 		iovad->cached32_node = &new->node;
 	else
 		iovad->cached_node = &new->node;
+
+	insert_update_32bit_pfn(iovad, new);
 }
 
 static void
@@ -145,6 +166,8 @@ __cached_rbnode_delete_update(struct iova_domain *iovad, struct iova *free)
 	cached_iova = rb_entry(iovad->cached_node, struct iova, node);
 	if (free->pfn_lo >= cached_iova->pfn_lo)
 		iovad->cached_node = rb_next(&free->node);
+
+	delete_update_32bit_pfn(iovad, free);
 }
 
 /* Insert the iova into domain rbtree by holding writer lock */
@@ -188,8 +211,13 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 	if (size_aligned)
 		align_mask <<= fls_long(size - 1);
 
-	/* Walk the tree backwards */
 	spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);
+
+	/* Fail fast if a 32-bit allocation cannot possibly succeed */
+	if (limit_pfn <= iovad->dma_32bit_pfn && iovad->free_32bit_pfns < size)
+		goto no_mem;
+
+	/* Walk the tree backwards */
 	curr = __get_cached_rbnode(iovad, limit_pfn);
 	curr_iova = rb_entry(curr, struct iova, node);
 	do {
@@ -200,10 +228,8 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 		curr_iova = rb_entry(curr, struct iova, node);
 	} while (curr && new_pfn <= curr_iova->pfn_hi);
 
-	if (limit_pfn < size || new_pfn < iovad->start_pfn) {
-		spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
-		return -ENOMEM;
-	}
+	if (limit_pfn < size || new_pfn < iovad->start_pfn)
+		goto no_mem;
 
 	/* pfn_lo will point to size aligned address if size_aligned is set */
 	new->pfn_lo = new_pfn;
@@ -214,9 +240,11 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 	__cached_rbnode_insert_update(iovad, new);
 
 	spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
-
-
 	return 0;
+
+no_mem:
+	spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
+	return -ENOMEM;
 }
 
 static struct kmem_cache *iova_cache;
@@ -678,6 +706,8 @@ reserve_iova(struct iova_domain *iovad,
 	 * or need to insert remaining non overlap addr range
 	 */
 	iova = __insert_new_range(iovad, pfn_lo, pfn_hi);
+	if (iova)
+		insert_update_32bit_pfn(iovad, iova);
 finish:
 
 	spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);

include/linux/iova.h

@@ -77,6 +77,7 @@ struct iova_domain {
 	unsigned long	granule;	/* pfn granularity for this domain */
 	unsigned long	start_pfn;	/* Lower limit for this domain */
 	unsigned long	dma_32bit_pfn;
+	unsigned int	free_32bit_pfns;
 	struct iova	anchor;		/* rbtree lookup anchor */
 	struct iova_rcache rcaches[IOVA_RANGE_CACHE_MAX_SIZE];	/* IOVA range caches */