FROMLIST: iommu/iova: Add a best-fit algorithm

Using the best-fit algorithm, instead of the first-fit
algorithm, may reduce fragmentation when allocating
IOVAs.

Bug: 149544392
Link: https://lore.kernel.org/lkml/7239ddd532e94a4371289f3be23c66a3@codeaurora.org/
Change-Id: Icfbac0cc7be972a092915335508cbc73c47471cf
Signed-off-by: Isaac J. Manjarres <isaacm@codeaurora.org>

drivers/iommu/dma-iommu.c

@@ -380,6 +380,24 @@ int iommu_dma_reserve_iova(struct device *dev, dma_addr_t base,
 }
 EXPORT_SYMBOL(iommu_dma_reserve_iova);
 
+/*
+ * Should be called prior to using dma-apis.
+ */
+int iommu_dma_enable_best_fit_algo(struct device *dev)
+{
+	struct iommu_domain *domain;
+	struct iova_domain *iovad;
+
+	domain = iommu_get_domain_for_dev(dev);
+	if (!domain || !domain->iova_cookie)
+		return -EINVAL;
+
+	iovad = &((struct iommu_dma_cookie *)domain->iova_cookie)->iovad;
+	iovad->best_fit = true;
+	return 0;
+}
+EXPORT_SYMBOL(iommu_dma_enable_best_fit_algo);
+
 /**
  * dma_info_to_prot - Translate DMA API directions and attributes to IOMMU API
  *                    page flags.

drivers/iommu/iova.c

@@ -50,6 +50,7 @@ init_iova_domain(struct iova_domain *iovad, unsigned long granule,
 	iovad->anchor.pfn_lo = iovad->anchor.pfn_hi = IOVA_ANCHOR;
 	rb_link_node(&iovad->anchor.node, NULL, &iovad->rbroot.rb_node);
 	rb_insert_color(&iovad->anchor.node, &iovad->rbroot);
+	iovad->best_fit = false;
 	init_iova_rcaches(iovad);
 }
 EXPORT_SYMBOL_GPL(init_iova_domain);
@@ -227,6 +228,70 @@ iova32_full:
 	return -ENOMEM;
 }
 
+static int __alloc_and_insert_iova_best_fit(struct iova_domain *iovad,
+					    unsigned long size,
+					    unsigned long limit_pfn,
+					    struct iova *new, bool size_aligned)
+{
+	struct rb_node *curr, *prev;
+	struct iova *curr_iova, *prev_iova;
+	unsigned long flags;
+	unsigned long align_mask = ~0UL;
+	struct rb_node *candidate_rb_parent;
+	unsigned long new_pfn, candidate_pfn = ~0UL;
+	unsigned long gap, candidate_gap = ~0UL;
+
+	if (size_aligned)
+		align_mask <<= limit_align(iovad, fls_long(size - 1));
+
+	/* Walk the tree backwards */
+	spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);
+	curr = &iovad->anchor.node;
+	prev = rb_prev(curr);
+	for (; prev; curr = prev, prev = rb_prev(curr)) {
+		curr_iova = rb_entry(curr, struct iova, node);
+		prev_iova = rb_entry(prev, struct iova, node);
+
+		limit_pfn = min(limit_pfn, curr_iova->pfn_lo);
+		new_pfn = (limit_pfn - size) & align_mask;
+		gap = curr_iova->pfn_lo - prev_iova->pfn_hi - 1;
+		if ((limit_pfn >= size) && (new_pfn > prev_iova->pfn_hi)
+				&& (gap < candidate_gap)) {
+			candidate_gap = gap;
+			candidate_pfn = new_pfn;
+			candidate_rb_parent = curr;
+			if (gap == size)
+				goto insert;
+		}
+	}
+
+	curr_iova = rb_entry(curr, struct iova, node);
+	limit_pfn = min(limit_pfn, curr_iova->pfn_lo);
+	new_pfn = (limit_pfn - size) & align_mask;
+	gap = curr_iova->pfn_lo - iovad->start_pfn;
+	if (limit_pfn >= size && new_pfn >= iovad->start_pfn &&
+			gap < candidate_gap) {
+		candidate_gap = gap;
+		candidate_pfn = new_pfn;
+		candidate_rb_parent = curr;
+	}
+
+insert:
+	if (candidate_pfn == ~0UL) {
+		spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
+		return -ENOMEM;
+	}
+
+	/* pfn_lo will point to size aligned address if size_aligned is set */
+	new->pfn_lo = candidate_pfn;
+	new->pfn_hi = new->pfn_lo + size - 1;
+
+	/* If we have 'prev', it's a valid place to start the insertion. */
+	iova_insert_rbtree(&iovad->rbroot, new, candidate_rb_parent);
+	spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
+	return 0;
+}
+
 static struct kmem_cache *iova_cache;
 static unsigned int iova_cache_users;
 static DEFINE_MUTEX(iova_cache_mutex);
@@ -302,8 +367,13 @@ alloc_iova(struct iova_domain *iovad, unsigned long size,
 	if (!new_iova)
 		return NULL;
 
-	ret = __alloc_and_insert_iova_range(iovad, size, limit_pfn + 1,
-			new_iova, size_aligned);
+	if (iovad->best_fit) {
+		ret = __alloc_and_insert_iova_range(iovad, size,
+				limit_pfn + 1, new_iova, size_aligned);
+	} else {
+		ret = __alloc_and_insert_iova_range(iovad, size, limit_pfn + 1,
+				new_iova, size_aligned);
+	}
 
 	if (ret) {
 		free_iova_mem(new_iova);

include/linux/dma-iommu.h

@@ -40,6 +40,8 @@ void iommu_dma_get_resv_regions(struct device *dev, struct list_head *list);
 int iommu_dma_reserve_iova(struct device *dev, dma_addr_t base,
 			   u64 size);
 
+int iommu_dma_enable_best_fit_algo(struct device *dev);
+
 #else /* CONFIG_IOMMU_DMA */
 
 struct iommu_domain;
@@ -87,5 +89,10 @@ static inline int iommu_dma_reserve_iova(struct device *dev, dma_addr_t base,
 	return -ENODEV;
 }
 
+static inline int iommu_dma_enable_best_fit_algo(struct device *dev)
+{
+	return -ENODEV;
+}
+
 #endif	/* CONFIG_IOMMU_DMA */
 #endif	/* __DMA_IOMMU_H */

include/linux/iova.h

@@ -95,6 +95,7 @@ struct iova_domain {
 						   flush-queues */
 	atomic_t fq_timer_on;			/* 1 when timer is active, 0
 						   when not */
+	bool best_fit;
 };
 
 static inline unsigned long iova_size(struct iova *iova)