Optimise 64-bit IOVA allocations

[rm: rewrote commit message to clarify]

---

v4: No change

 drivers/iommu/iova.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

drivers/iommu/iova.c +3 -6

index 33edfa794ae9..f129ff4f5c89 100644
@@ -342,15 +342,12 @@ private_find_iova(struct iova_domain *iovad, unsigned long pfn)
 	while (node) {
 		struct iova *iova = rb_entry(node, struct iova, node);
 
-		/* If pfn falls within iova's range, return iova */
-		if ((pfn >= iova->pfn_lo) && (pfn <= iova->pfn_hi)) {
-			return iova;
-		}
-
 		if (pfn < iova->pfn_lo)
 			node = node->rb_left;
-		else if (pfn > iova->pfn_lo)
+		else if (pfn > iova->pfn_hi)
 			node = node->rb_right;
+		else
+			return iova;	/* pfn falls within iova's range */
 	}
 
 	return NULL;

Signature · 1 line

2.13.4.dirty

Robin Murphy

author maintainer

2017-09-19 16:31 UTC #2 lore raw

PATCH v4 2/6 iommu/iova: Optimise the padding calculation

From: Zhen Lei <[email protected]>

The mask for calculating the padding size doesn't change, so there's no need to recalculate it every loop iteration. Furthermore, Once we've done that, it becomes clear that we don't actually need to calculate a padding size at all - by flipping the arithmetic around, we can just combine the upper limit, size, and mask directly to check against the lower limit.

For an arm64 build, this alone knocks 20% off the object code size of the entire alloc_iova() function!

Signed-off-by Zhen Lei <[email protected]>

[rm: simplified more of the arithmetic, rewrote commit message]

---

v4:
- Round align_mask up instead of down (oops!)
- Remove redundant !curr check
- Introduce new_pfn variable here to reduce churn in later patches

 drivers/iommu/iova.c | 42 +++++++++++++++---------------------------
 1 file changed, 15 insertions(+), 27 deletions(-)

drivers/iommu/iova.c +15 ~~-27~~

index f129ff4f5c89..20be9a8b3188 100644
@@ -182,24 +182,17 @@ iova_insert_rbtree(struct rb_root *root, struct iova *iova,
 	rb_insert_color(&iova->node, root);
 }
 
-/*
- * Computes the padding size required, to make the start address
- * naturally aligned on the power-of-two order of its size
- */
-static unsigned int
-iova_get_pad_size(unsigned int size, unsigned int limit_pfn)
-{
-	return (limit_pfn - size) & (__roundup_pow_of_two(size) - 1);
-}
-
 static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 		unsigned long size, unsigned long limit_pfn,
 			struct iova *new, bool size_aligned)
 {
 	struct rb_node *prev, *curr = NULL;
 	unsigned long flags;
-	unsigned long saved_pfn;
-	unsigned int pad_size = 0;
+	unsigned long saved_pfn, new_pfn;
+	unsigned long align_mask = ~0UL;
+
+	if (size_aligned)
+		align_mask <<= fls_long(size - 1);
 
 	/* Walk the tree backwards */
 	spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);

@@ -209,31 +202,26 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 	while (curr) {
 		struct iova *curr_iova = rb_entry(curr, struct iova, node);
 
-		if (limit_pfn <= curr_iova->pfn_lo) {
+		if (limit_pfn <= curr_iova->pfn_lo)
 			goto move_left;
-		} else if (limit_pfn > curr_iova->pfn_hi) {
-			if (size_aligned)
-				pad_size = iova_get_pad_size(size, limit_pfn);
-			if ((curr_iova->pfn_hi + size + pad_size) < limit_pfn)
-				break;	/* found a free slot */
-		}
+
+		if (((limit_pfn - size) & align_mask) > curr_iova->pfn_hi)
+			break;	/* found a free slot */
+
 		limit_pfn = curr_iova->pfn_lo;
 move_left:
 		prev = curr;
 		curr = rb_prev(curr);
 	}
 
-	if (!curr) {
-		if (size_aligned)
-			pad_size = iova_get_pad_size(size, limit_pfn);
-		if ((iovad->start_pfn + size + pad_size) > limit_pfn) {
-			spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
-			return -ENOMEM;
-		}
+	new_pfn = (limit_pfn - size) & align_mask;
+	if (limit_pfn < size || new_pfn < iovad->start_pfn) {
+		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 = limit_pfn - (size + pad_size);
+	new->pfn_lo = new_pfn;
 	new->pfn_hi = new->pfn_lo + size - 1;
 
 	/* If we have 'prev', it's a valid place to start the insertion. */

Signature · 1 line

2.13.4.dirty

Robin Murphy

author maintainer

2017-09-19 16:31 UTC #3 lore raw

PATCH v4 3/6 iommu/iova: Add rbtree anchor node

Add a permanent dummy IOVA reservation to the rbtree, such that we can always access the top of the address space instantly. The immediate benefit is that we remove the overhead of the rb_last() traversal when not using the cached node, but it also paves the way for further simplifications.

---

v4: New

 drivers/iommu/iova.c | 15 +++++++++++++--
 include/linux/iova.h |  1 +
 2 files changed, 14 insertions(+), 2 deletions(-)

drivers/iommu/iova.c +13 -2

index 20be9a8b3188..03b677afb109 100644
@@ -24,6 +24,9 @@
 #include <linux/bitops.h>
 #include <linux/cpu.h>
 
+/* The anchor node sits above the top of the usable address space */
+#define IOVA_ANCHOR	~0UL
+
 static bool iova_rcache_insert(struct iova_domain *iovad,
 			       unsigned long pfn,
 			       unsigned long size);

@@ -54,6 +57,9 @@ init_iova_domain(struct iova_domain *iovad, unsigned long granule,
 	iovad->dma_32bit_pfn = pfn_32bit + 1;
 	iovad->flush_cb = NULL;
 	iovad->fq = NULL;
+	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);
 	init_iova_rcaches(iovad);
 }
 EXPORT_SYMBOL_GPL(init_iova_domain);

@@ -112,7 +118,7 @@ __get_cached_rbnode(struct iova_domain *iovad, unsigned long *limit_pfn)
 {
 	if ((*limit_pfn > iovad->dma_32bit_pfn) ||
 		(iovad->cached32_node == NULL))
-		return rb_last(&iovad->rbroot);
+		return rb_prev(&iovad->anchor.node);
 	else {
 		struct rb_node *prev_node = rb_prev(iovad->cached32_node);
 		struct iova *curr_iova =

@@ -246,7 +252,8 @@ EXPORT_SYMBOL(alloc_iova_mem);
 
 void free_iova_mem(struct iova *iova)
 {
-	kmem_cache_free(iova_cache, iova);
+	if (iova->pfn_lo != IOVA_ANCHOR)
+		kmem_cache_free(iova_cache, iova);
 }
 EXPORT_SYMBOL(free_iova_mem);

@@ -680,6 +687,10 @@ reserve_iova(struct iova_domain *iovad,
 	struct iova *iova;
 	unsigned int overlap = 0;
 
+	/* Don't allow nonsensical pfns */
+	if (WARN_ON((pfn_hi | pfn_lo) > (ULLONG_MAX >> iova_shift(iovad))))
+		return NULL;
+
 	spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);
 	for (node = rb_first(&iovad->rbroot); node; node = rb_next(node)) {
 		if (__is_range_overlap(node, pfn_lo, pfn_hi)) {

include/linux/iova.h +1 -0

index d179b9bf7814..22dc30a28387 100644
@@ -74,6 +74,7 @@ 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;
+	struct iova	anchor;		/* rbtree lookup anchor */
 	struct iova_rcache rcaches[IOVA_RANGE_CACHE_MAX_SIZE];	/* IOVA range caches */
 
 	iova_flush_cb	flush_cb;	/* Call-Back function to flush IOMMU

Signature · 1 line

2.13.4.dirty

Robin Murphy

author maintainer

2017-09-19 16:31 UTC #4 lore raw

PATCH v4 4/6 iommu/iova: Simplify cached node logic

The logic of __get_cached_rbnode() is a little obtuse, but then __get_prev_node_of_cached_rbnode_or_last_node_and_update_limit_pfn() wouldn't exactly roll off the tongue...

Now that we have the invariant that there is always a valid node to start searching downwards from, everything gets a bit easier to follow if we simplify that function to do what it says on the tin and return the cached node (or anchor node as appropriate) directly. In turn, we can then deduplicate the rb_prev() and limit_pfn logic into the main loop itself, further reduce the amount of code under the lock, and generally make the inner workings a bit less subtle.

---

v4: New

 drivers/iommu/iova.c | 42 ++++++++++++++----------------------------
 1 file changed, 14 insertions(+), 28 deletions(-)

drivers/iommu/iova.c +14 ~~-28~~

index 03b677afb109..c93a6c46bcb1 100644
@@ -114,18 +114,12 @@ int init_iova_flush_queue(struct iova_domain *iovad,
 EXPORT_SYMBOL_GPL(init_iova_flush_queue);
 
 static struct rb_node *
-__get_cached_rbnode(struct iova_domain *iovad, unsigned long *limit_pfn)
+__get_cached_rbnode(struct iova_domain *iovad, unsigned long limit_pfn)
 {
-	if ((*limit_pfn > iovad->dma_32bit_pfn) ||
-		(iovad->cached32_node == NULL))
-		return rb_prev(&iovad->anchor.node);
-	else {
-		struct rb_node *prev_node = rb_prev(iovad->cached32_node);
-		struct iova *curr_iova =
-			rb_entry(iovad->cached32_node, struct iova, node);
-		*limit_pfn = curr_iova->pfn_lo;
-		return prev_node;
-	}
+	if (limit_pfn <= iovad->dma_32bit_pfn && iovad->cached32_node)
+		return iovad->cached32_node;
+
+	return &iovad->anchor.node;
 }
 
 static void

@@ -192,7 +186,8 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 		unsigned long size, unsigned long limit_pfn,
 			struct iova *new, bool size_aligned)
 {
-	struct rb_node *prev, *curr = NULL;
+	struct rb_node *curr, *prev;
+	struct iova *curr_iova;
 	unsigned long flags;
 	unsigned long saved_pfn, new_pfn;
 	unsigned long align_mask = ~0UL;

@@ -203,29 +198,20 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 	/* Walk the tree backwards */
 	spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);
 	saved_pfn = limit_pfn;
-	curr = __get_cached_rbnode(iovad, &limit_pfn);
-	prev = curr;
-	while (curr) {
-		struct iova *curr_iova = rb_entry(curr, struct iova, node);
-
-		if (limit_pfn <= curr_iova->pfn_lo)
-			goto move_left;
-
-		if (((limit_pfn - size) & align_mask) > curr_iova->pfn_hi)
-			break;	/* found a free slot */
-
-		limit_pfn = curr_iova->pfn_lo;
-move_left:
+	curr = __get_cached_rbnode(iovad, limit_pfn);
+	curr_iova = rb_entry(curr, struct iova, node);
+	do {
+		limit_pfn = min(limit_pfn, curr_iova->pfn_lo);
+		new_pfn = (limit_pfn - size) & align_mask;
 		prev = curr;
 		curr = rb_prev(curr);
-	}
+		curr_iova = rb_entry(curr, struct iova, node);
+	} while (curr && new_pfn <= curr_iova->pfn_hi);
 
-	new_pfn = (limit_pfn - size) & align_mask;
 	if (limit_pfn < size || new_pfn < iovad->start_pfn) {
 		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 = new_pfn;
 	new->pfn_hi = new->pfn_lo + size - 1;

Signature · 1 line

2.13.4.dirty

Robin Murphy

author maintainer

2017-09-19 16:31 UTC #5 lore raw

PATCH v4 5/6 iommu/iova: Extend rbtree node caching

The cached node mechanism provides a significant performance benefit for allocations using a 32-bit DMA mask, but in the case of non-PCI devices or where the 32-bit space is full, the loss of this benefit can be significant - on large systems there can be many thousands of entries in the tree, such that walking all the way down to find free space every time becomes increasingly awful.

Maintain a similar cached node for the whole IOVA space as a superset of the 32-bit space so that performance can remain much more consistent.

Inspired by work by Zhen Lei <[email protected]>.

---

v4:
- Adjust to simplified __get_cached_rbnode() behaviour
- Cosmetic tweaks

 drivers/iommu/iova.c | 43 +++++++++++++++++++++----------------------
 include/linux/iova.h |  3 ++-
 2 files changed, 23 insertions(+), 23 deletions(-)

drivers/iommu/iova.c +21 ~~-22~~

index c93a6c46bcb1..a125a5786dbf 100644
@@ -51,6 +51,7 @@ init_iova_domain(struct iova_domain *iovad, unsigned long granule,
 
 	spin_lock_init(&iovad->iova_rbtree_lock);
 	iovad->rbroot = RB_ROOT;
+	iovad->cached_node = NULL;
 	iovad->cached32_node = NULL;
 	iovad->granule = granule;
 	iovad->start_pfn = start_pfn;

@@ -119,39 +120,38 @@ __get_cached_rbnode(struct iova_domain *iovad, unsigned long limit_pfn)
 	if (limit_pfn <= iovad->dma_32bit_pfn && iovad->cached32_node)
 		return iovad->cached32_node;
 
+	if (iovad->cached_node)
+		return iovad->cached_node;
+
 	return &iovad->anchor.node;
 }
 
 static void
-__cached_rbnode_insert_update(struct iova_domain *iovad,
-	unsigned long limit_pfn, struct iova *new)
+__cached_rbnode_insert_update(struct iova_domain *iovad, struct iova *new)
 {
-	if (limit_pfn != iovad->dma_32bit_pfn)
-		return;
-	iovad->cached32_node = &new->node;
+	if (new->pfn_hi < iovad->dma_32bit_pfn)
+		iovad->cached32_node = &new->node;
+	else
+		iovad->cached_node = &new->node;
 }
 
 static void
 __cached_rbnode_delete_update(struct iova_domain *iovad, struct iova *free)
 {
 	struct iova *cached_iova;
-	struct rb_node *curr;
+	struct rb_node **curr;
 
-	if (!iovad->cached32_node)
+	if (free->pfn_hi < iovad->dma_32bit_pfn)
+		curr = &iovad->cached32_node;
+	else
+		curr = &iovad->cached_node;
+
+	if (!*curr)
 		return;
-	curr = iovad->cached32_node;
-	cached_iova = rb_entry(curr, struct iova, node);
 
-	if (free->pfn_lo >= cached_iova->pfn_lo) {
-		struct rb_node *node = rb_next(&free->node);
-		struct iova *iova = rb_entry(node, struct iova, node);
-
-		/* only cache if it's below 32bit pfn */
-		if (node && iova->pfn_lo < iovad->dma_32bit_pfn)
-			iovad->cached32_node = node;
-		else
-			iovad->cached32_node = NULL;
-	}
+	cached_iova = rb_entry(*curr, struct iova, node);
+	if (free->pfn_lo >= cached_iova->pfn_lo)
+		*curr = rb_next(&free->node);
 }
 
 /* Insert the iova into domain rbtree by holding writer lock */

@@ -189,7 +189,7 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 	struct rb_node *curr, *prev;
 	struct iova *curr_iova;
 	unsigned long flags;
-	unsigned long saved_pfn, new_pfn;
+	unsigned long new_pfn;
 	unsigned long align_mask = ~0UL;
 
 	if (size_aligned)

@@ -197,7 +197,6 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 
 	/* Walk the tree backwards */
 	spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);
-	saved_pfn = limit_pfn;
 	curr = __get_cached_rbnode(iovad, limit_pfn);
 	curr_iova = rb_entry(curr, struct iova, node);
 	do {

@@ -218,7 +217,7 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 
 	/* If we have 'prev', it's a valid place to start the insertion. */
 	iova_insert_rbtree(&iovad->rbroot, new, prev);
-	__cached_rbnode_insert_update(iovad, saved_pfn, new);
+	__cached_rbnode_insert_update(iovad, new);
 
 	spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);

include/linux/iova.h +2 -1

index 22dc30a28387..5eaedf77b152 100644
@@ -70,7 +70,8 @@ struct iova_fq {
 struct iova_domain {
 	spinlock_t	iova_rbtree_lock; /* Lock to protect update of rbtree */
 	struct rb_root	rbroot;		/* iova domain rbtree root */
-	struct rb_node	*cached32_node; /* Save last alloced node */
+	struct rb_node	*cached_node;	/* Save last alloced node */
+	struct rb_node	*cached32_node; /* Save last 32-bit alloced node */
 	unsigned long	granule;	/* pfn granularity for this domain */
 	unsigned long	start_pfn;	/* Lower limit for this domain */
 	unsigned long	dma_32bit_pfn;

Signature · 1 line

2.13.4.dirty

Tomasz Nowicki

2017-09-20 12:59 UTC #6 lore raw

Re: [PATCH v4 5/6] iommu/iova: Extend rbtree node caching

Hi Robin,

Quoted 83 lines · Robin Murphy

Maintain a similar cached node for the whole IOVA space as a superset of the 32-bit space so that performance can remain much more consistent.

Inspired by work by Zhen Lei <[email protected]>.

---

v4:
- Adjust to simplified __get_cached_rbnode() behaviour
- Cosmetic tweaks

  drivers/iommu/iova.c | 43 +++++++++++++++++++++----------------------
  include/linux/iova.h |  3 ++-
  2 files changed, 23 insertions(+), 23 deletions(-)

drivers/iommu/iova.c +16 ~~-12~~

index c93a6c46bcb1..a125a5786dbf 100644
@@ -51,6 +51,7 @@ init_iova_domain(struct iova_domain *iovad, unsigned long granule,
 
  	spin_lock_init(&iovad->iova_rbtree_lock);
  	iovad->rbroot = RB_ROOT;
+	iovad->cached_node = NULL;
  	iovad->cached32_node = NULL;
  	iovad->granule = granule;
  	iovad->start_pfn = start_pfn;

@@ -119,39 +120,38 @@ __get_cached_rbnode(struct iova_domain *iovad, unsigned long limit_pfn)
  	if (limit_pfn <= iovad->dma_32bit_pfn && iovad->cached32_node)
  		return iovad->cached32_node;
 
+	if (iovad->cached_node)
+		return iovad->cached_node;
+
  	return &iovad->anchor.node;
  }
 
  static void
-__cached_rbnode_insert_update(struct iova_domain *iovad,
-	unsigned long limit_pfn, struct iova *new)
+__cached_rbnode_insert_update(struct iova_domain *iovad, struct iova *new)
  {
-	if (limit_pfn != iovad->dma_32bit_pfn)
-		return;
-	iovad->cached32_node = &new->node;
+	if (new->pfn_hi < iovad->dma_32bit_pfn)
+		iovad->cached32_node = &new->node;
+	else
+		iovad->cached_node = &new->node;
  }
 
  static void
  __cached_rbnode_delete_update(struct iova_domain *iovad, struct iova *free)
  {
  	struct iova *cached_iova;
-	struct rb_node *curr;
+	struct rb_node **curr;
 
-	if (!iovad->cached32_node)
+	if (free->pfn_hi < iovad->dma_32bit_pfn)
+		curr = &iovad->cached32_node;
+	else
+		curr = &iovad->cached_node;
+
+	if (!*curr)
  		return;
-	curr = iovad->cached32_node;
-	cached_iova = rb_entry(curr, struct iova, node);
 
-	if (free->pfn_lo >= cached_iova->pfn_lo) {
-		struct rb_node *node = rb_next(&free->node);
-		struct iova *iova = rb_entry(node, struct iova, node);

drivers/iommu/iova.c __get_cached_rbnode(struct iova_domain *iovad, unsigned long limit_pfn) · line 154

Quoted 10 lines

drivers/iommu/iova.c +3 -7

-
-		/* only cache if it's below 32bit pfn */
-		if (node && iova->pfn_lo < iovad->dma_32bit_pfn)
-			iovad->cached32_node = node;
-		else
-			iovad->cached32_node = NULL;
-	}
+	cached_iova = rb_entry(*curr, struct iova, node);
+	if (free->pfn_lo >= cached_iova->pfn_lo)
+		*curr = rb_next(&free->node);

IMO, we may incorrectly update iovad->cached32_node here.

32-bit boundary
-------- -------- | -------- --------
| | | | | | | | |
---- IOVA0 ----- IOVA1 ----------------- IOVA3 ----- anchor |

    |        |     |        |       |         |        |     |        |
     --------       --------        |          --------       --------

If we free last IOVA from 32-bit space (IOVA1) we will update iovad->cached32_node to IOVA3.

Thanks,
Tomasz

Robin Murphy

author maintainer

2017-09-20 17:27 UTC #7 lore raw

Re: [PATCH v4 5/6] iommu/iova: Extend rbtree node caching

Quoted 99 lines · Tomasz Nowicki

Hi Robin,
Quoted 83 lines · Robin Murphy
The cached node mechanism provides a significant performance benefit for allocations using a 32-bit DMA mask, but in the case of non-PCI devices or where the 32-bit space is full, the loss of this benefit can be significant - on large systems there can be many thousands of entries in the tree, such that walking all the way down to find free space every time becomes increasingly awful.
Maintain a similar cached node for the whole IOVA space as a superset of the 32-bit space so that performance can remain much more consistent.
Inspired by work by Zhen Lei <[email protected]>.
Tested-by Ard Biesheuvel <[email protected]>
Tested-by Zhen Lei <[email protected]>
Tested-by Nate Watterson <[email protected]>
Signed-off-by Robin Murphy <[email protected]>
---
v4:
- Adjust to simplified __get_cached_rbnode() behaviour
- Cosmetic tweaks
drivers/iommu/iova.c | 43 +++++++++++++++++++++---------------------- include/linux/iova.h | 3 ++-
  2 files changed, 23 insertions(+), 23 deletions(-)
drivers/iommu/iova.c +3 -0
index c93a6c46bcb1..a125a5786dbf 100644
@@ -51,6 +51,7 @@ init_iova_domain(struct iova_domain *iovad, unsigned
 long granule,
         spin_lock_init(&iovad->iova_rbtree_lock);
       iovad->rbroot = RB_ROOT;
+    iovad->cached_node = NULL;
       iovad->cached32_node = NULL;
       iovad->granule = granule;
       iovad->start_pfn = start_pfn;
@@ -119,39 +120,38 @@ __get_cached_rbnode(struct iova_domain *iovad,
 unsigned long limit_pfn)
       if (limit_pfn <= iovad->dma_32bit_pfn && iovad->cached32_node)
           return iovad->cached32_node;
   +    if (iovad->cached_node)
+        return iovad->cached_node;
+
       return &iovad->anchor.node;
   }
static void
drivers/iommu/iova.c +1 -2
-__cached_rbnode_insert_update(struct iova_domain *iovad,
-    unsigned long limit_pfn, struct iova *new)
+__cached_rbnode_insert_update(struct iova_domain *iovad, struct iova
*new)
{
-    if (limit_pfn != iovad->dma_32bit_pfn)
-        return;
-    iovad->cached32_node = &new->node;
+    if (new->pfn_hi < iovad->dma_32bit_pfn)
+        iovad->cached32_node = &new->node;
+    else
+        iovad->cached_node = &new->node;
}
static void
__cached_rbnode_delete_update(struct iova_domain *iovad, struct iova *free)
{
      struct iova *cached_iova;
-    struct rb_node *curr;
+    struct rb_node **curr;
-    if (!iovad->cached32_node)
+    if (free->pfn_hi < iovad->dma_32bit_pfn)
+        curr = &iovad->cached32_node;
+    else
+        curr = &iovad->cached_node;
+
+    if (!*curr)
          return;
-    curr = iovad->cached32_node;
-    cached_iova = rb_entry(curr, struct iova, node);
-    if (free->pfn_lo >= cached_iova->pfn_lo) {
-        struct rb_node *node = rb_next(&free->node);
-        struct iova *iova = rb_entry(node, struct iova, node); -
-        /* only cache if it's below 32bit pfn */
Quoted 8 lines
-        if (node && iova->pfn_lo < iovad->dma_32bit_pfn) -            iovad->cached32_node = node;
-        else
-            iovad->cached32_node = NULL;
-    }
+    cached_iova = rb_entry(*curr, struct iova, node); +    if (free->pfn_lo >= cached_iova->pfn_lo)
+        *curr = rb_next(&free->node);
IMO, we may incorrectly update iovad->cached32_node here.
                              32-bit boundary
     --------       --------        |          --------       --------

Quoted 7 lines

    |        |     |        |       |         |        |     |        |
----  IOVA0   -----  IOVA1   -----------------  IOVA3   -----  anchor |
    |        |     |        |       |         |        |     |        |
     --------       --------        |          --------       --------

If we free last IOVA from 32-bit space (IOVA1) we will update iovad->cached32_node to IOVA3.

That in itself is intentional and correct - the cached node is where we *start* searching for free space, so if the topmost 32-bit pfn is free, we want to start from whichever node represents the next-highest pfn. Even if more 64-bit IOVAs were allocated below IOVA3 before the next 32-bit allocation, walking through those is better that walking all the way down from the anchor.

However, you have made me realise that if IOVA3 were freed before the next 32-bit allocation things we'd only update cached_node and leave cached32_node with a stale pointer...

I'll get a v5 out tomorrow (and keep the new stuff at the end of the series this time)

Thanks,
Robin.

Robin Murphy

author maintainer

2017-09-19 16:31 UTC #8 lore raw

PATCH v4 6/6 iommu/iova: Make dma_32bit_pfn implicit

From: Zhen Lei <[email protected]>

Now that the cached node optimisation can apply to all allocations, the couple of users which were playing tricks with dma_32bit_pfn in order to benefit from it can stop doing so. Conversely, there is also no need for all the other users to explicitly calculate a 'real' 32-bit PFN, when init_iova_domain() can happily do that itself from the page granularity.

CC Thierry Reding <[email protected]>

CC Jonathan Hunter <[email protected]>

CC David Airlie <[email protected]>

CC Sudeep Dutt <[email protected]>

CC Ashutosh Dixit <[email protected]>

Signed-off-by Zhen Lei <[email protected]>

[rm: use iova_shift(), rewrote commit message]

---

v4: No change

 drivers/gpu/drm/tegra/drm.c      |  3 +--
 drivers/gpu/host1x/dev.c         |  3 +--
 drivers/iommu/amd_iommu.c        |  7 ++-----
 drivers/iommu/dma-iommu.c        | 18 +-----------------
 drivers/iommu/intel-iommu.c      | 11 +++--------
 drivers/iommu/iova.c             |  4 ++--
 drivers/misc/mic/scif/scif_rma.c |  3 +--
 include/linux/iova.h             |  5 ++---
 8 files changed, 13 insertions(+), 41 deletions(-)

drivers/gpu/drm/tegra/drm.c +1 -2

index 597d563d636a..b822e484b7e5 100644
@@ -155,8 +155,7 @@ static int tegra_drm_load(struct drm_device *drm, unsigned long flags)
 
 		order = __ffs(tegra->domain->pgsize_bitmap);
 		init_iova_domain(&tegra->carveout.domain, 1UL << order,
-				 carveout_start >> order,
-				 carveout_end >> order);
+				 carveout_start >> order);
 
 		tegra->carveout.shift = iova_shift(&tegra->carveout.domain);
 		tegra->carveout.limit = carveout_end >> tegra->carveout.shift;

drivers/gpu/host1x/dev.c +1 -2

index 7f22c5c37660..5267c62e8896 100644
@@ -198,8 +198,7 @@ static int host1x_probe(struct platform_device *pdev)
 
 		order = __ffs(host->domain->pgsize_bitmap);
 		init_iova_domain(&host->iova, 1UL << order,
-				 geometry->aperture_start >> order,
-				 geometry->aperture_end >> order);
+				 geometry->aperture_start >> order);
 		host->iova_end = geometry->aperture_end;
 	}

drivers/iommu/amd_iommu.c +2 -5

index 51f8215877f5..647ab7691aee 100644
@@ -63,7 +63,6 @@
 /* IO virtual address start page frame number */
 #define IOVA_START_PFN		(1)
 #define IOVA_PFN(addr)		((addr) >> PAGE_SHIFT)
-#define DMA_32BIT_PFN		IOVA_PFN(DMA_BIT_MASK(32))
 
 /* Reserved IOVA ranges */
 #define MSI_RANGE_START		(0xfee00000)

@@ -1788,8 +1787,7 @@ static struct dma_ops_domain *dma_ops_domain_alloc(void)
 	if (!dma_dom->domain.pt_root)
 		goto free_dma_dom;
 
-	init_iova_domain(&dma_dom->iovad, PAGE_SIZE,
-			 IOVA_START_PFN, DMA_32BIT_PFN);
+	init_iova_domain(&dma_dom->iovad, PAGE_SIZE, IOVA_START_PFN);
 
 	if (init_iova_flush_queue(&dma_dom->iovad, iova_domain_flush_tlb, NULL))
 		goto free_dma_dom;

@@ -2696,8 +2694,7 @@ static int init_reserved_iova_ranges(void)
 	struct pci_dev *pdev = NULL;
 	struct iova *val;
 
-	init_iova_domain(&reserved_iova_ranges, PAGE_SIZE,
-			 IOVA_START_PFN, DMA_32BIT_PFN);
+	init_iova_domain(&reserved_iova_ranges, PAGE_SIZE, IOVA_START_PFN);
 
 	lockdep_set_class(&reserved_iova_ranges.iova_rbtree_lock,
 			  &reserved_rbtree_key);

drivers/iommu/dma-iommu.c +1 ~~-17~~

index 9d1cebe7f6cb..191be9c80a8a 100644
@@ -292,18 +292,7 @@ int iommu_dma_init_domain(struct iommu_domain *domain, dma_addr_t base,
 		/* ...then finally give it a kicking to make sure it fits */
 		base_pfn = max_t(unsigned long, base_pfn,
 				domain->geometry.aperture_start >> order);
-		end_pfn = min_t(unsigned long, end_pfn,
-				domain->geometry.aperture_end >> order);
 	}
-	/*
-	 * PCI devices may have larger DMA masks, but still prefer allocating
-	 * within a 32-bit mask to avoid DAC addressing. Such limitations don't
-	 * apply to the typical platform device, so for those we may as well
-	 * leave the cache limit at the top of their range to save an rb_last()
-	 * traversal on every allocation.
-	 */
-	if (dev && dev_is_pci(dev))
-		end_pfn &= DMA_BIT_MASK(32) >> order;
 
 	/* start_pfn is always nonzero for an already-initialised domain */
 	if (iovad->start_pfn) {

@@ -312,16 +301,11 @@ int iommu_dma_init_domain(struct iommu_domain *domain, dma_addr_t base,
 			pr_warn("Incompatible range for DMA domain\n");
 			return -EFAULT;
 		}
-		/*
-		 * If we have devices with different DMA masks, move the free
-		 * area cache limit down for the benefit of the smaller one.
-		 */
-		iovad->dma_32bit_pfn = min(end_pfn + 1, iovad->dma_32bit_pfn);
 
 		return 0;
 	}
 
-	init_iova_domain(iovad, 1UL << order, base_pfn, end_pfn);
+	init_iova_domain(iovad, 1UL << order, base_pfn);
 	if (!dev)
 		return 0;

drivers/iommu/intel-iommu.c +3 -8

index 6784a05dd6b2..ebb48353dd39 100644
@@ -82,8 +82,6 @@
 #define IOVA_START_PFN		(1)
 
 #define IOVA_PFN(addr)		((addr) >> PAGE_SHIFT)
-#define DMA_32BIT_PFN		IOVA_PFN(DMA_BIT_MASK(32))
-#define DMA_64BIT_PFN		IOVA_PFN(DMA_BIT_MASK(64))
 
 /* page table handling */
 #define LEVEL_STRIDE		(9)

@@ -1878,8 +1876,7 @@ static int dmar_init_reserved_ranges(void)
 	struct iova *iova;
 	int i;
 
-	init_iova_domain(&reserved_iova_list, VTD_PAGE_SIZE, IOVA_START_PFN,
-			DMA_32BIT_PFN);
+	init_iova_domain(&reserved_iova_list, VTD_PAGE_SIZE, IOVA_START_PFN);
 
 	lockdep_set_class(&reserved_iova_list.iova_rbtree_lock,
 		&reserved_rbtree_key);

@@ -1938,8 +1935,7 @@ static int domain_init(struct dmar_domain *domain, struct intel_iommu *iommu,
 	unsigned long sagaw;
 	int err;
 
-	init_iova_domain(&domain->iovad, VTD_PAGE_SIZE, IOVA_START_PFN,
-			DMA_32BIT_PFN);
+	init_iova_domain(&domain->iovad, VTD_PAGE_SIZE, IOVA_START_PFN);
 
 	err = init_iova_flush_queue(&domain->iovad,
 				    iommu_flush_iova, iova_entry_free);

@@ -4897,8 +4893,7 @@ static int md_domain_init(struct dmar_domain *domain, int guest_width)
 {
 	int adjust_width;
 
-	init_iova_domain(&domain->iovad, VTD_PAGE_SIZE, IOVA_START_PFN,
-			DMA_32BIT_PFN);
+	init_iova_domain(&domain->iovad, VTD_PAGE_SIZE, IOVA_START_PFN);
 	domain_reserve_special_ranges(domain);
 
 	/* calculate AGAW */

drivers/iommu/iova.c +2 -2

index a125a5786dbf..dbe26067250e 100644
@@ -40,7 +40,7 @@ static void fq_flush_timeout(unsigned long data);
 
 void
 init_iova_domain(struct iova_domain *iovad, unsigned long granule,
-	unsigned long start_pfn, unsigned long pfn_32bit)
+	unsigned long start_pfn)
 {
 	/*
 	 * IOVA granularity will normally be equal to the smallest

@@ -55,7 +55,7 @@ init_iova_domain(struct iova_domain *iovad, unsigned long granule,
 	iovad->cached32_node = NULL;
 	iovad->granule = granule;
 	iovad->start_pfn = start_pfn;
-	iovad->dma_32bit_pfn = pfn_32bit + 1;
+	iovad->dma_32bit_pfn = 1UL << (32 - iova_shift(iovad));
 	iovad->flush_cb = NULL;
 	iovad->fq = NULL;
 	iovad->anchor.pfn_lo = iovad->anchor.pfn_hi = IOVA_ANCHOR;

drivers/misc/mic/scif/scif_rma.c +1 -2

index 329727e00e97..c824329f7012 100644
@@ -39,8 +39,7 @@ void scif_rma_ep_init(struct scif_endpt *ep)
 	struct scif_endpt_rma_info *rma = &ep->rma_info;
 
 	mutex_init(&rma->rma_lock);
-	init_iova_domain(&rma->iovad, PAGE_SIZE, SCIF_IOVA_START_PFN,
-			 SCIF_DMA_64BIT_PFN);
+	init_iova_domain(&rma->iovad, PAGE_SIZE, SCIF_IOVA_START_PFN);
 	spin_lock_init(&rma->tc_lock);
 	mutex_init(&rma->mmn_lock);
 	INIT_LIST_HEAD(&rma->reg_list);

include/linux/iova.h +2 -3

index 5eaedf77b152..c696ee81054e 100644
@@ -155,7 +155,7 @@ struct iova *reserve_iova(struct iova_domain *iovad, unsigned long pfn_lo,
 	unsigned long pfn_hi);
 void copy_reserved_iova(struct iova_domain *from, struct iova_domain *to);
 void init_iova_domain(struct iova_domain *iovad, unsigned long granule,
-	unsigned long start_pfn, unsigned long pfn_32bit);
+	unsigned long start_pfn);
 int init_iova_flush_queue(struct iova_domain *iovad,
 			  iova_flush_cb flush_cb, iova_entry_dtor entry_dtor);
 struct iova *find_iova(struct iova_domain *iovad, unsigned long pfn);

@@ -231,8 +231,7 @@ static inline void copy_reserved_iova(struct iova_domain *from,
 
 static inline void init_iova_domain(struct iova_domain *iovad,
 				    unsigned long granule,
-				    unsigned long start_pfn,
-				    unsigned long pfn_32bit)
+				    unsigned long start_pfn)
 {
 }

Signature · 1 line

2.13.4.dirty

Robin Murphy

author maintainer

2017-09-20 11:02 UTC #9 lore raw

[PATCH v4 7/6] iommu/iova: Make cached_node always valid

With the anchor node at the top of the rbtree, there is always a valid node for rb_next() to return, such that cached_node is only ever NULL until the first allocation. Initialising it to point at the anchor node gets rid of that window and makes the NULL checking entirely redundant.