Skip to contentSkip to conversation

iommu/iova: Retry from last rb tree node if iova search fails

PATCH Superseded [email protected] 3 participants 10 messages Aug 20, 2020 – Sep 30, 2020

Subsystems (2)

v2 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/2 iommu/iova: Retry from last rb tree node if iova search fails

From: Vijayanand Jitta <[email protected]>

When ever a new iova alloc request comes iova is always searched from the cached node and the nodes which are previous to cached node. So, even if there is free iova space available in the nodes which are next to the cached node iova allocation can still fail because of this approach.

Consider the following sequence of iova alloc and frees on 1GB of iova space

1) alloc - 500MB
2) alloc - 12MB
3) alloc - 499MB
4) free - 12MB which was allocated in step 2
5) alloc - 13MB

After the above sequence we will have 12MB of free iova space and cached node will be pointing to the iova pfn of last alloc of 13MB which will be the lowest iova pfn of that iova space. Now if we get an alloc request of 2MB we just search from cached node and then look for lower iova pfn's for free iova and as they aren't any, iova alloc fails though there is 12MB of free iova space.

To avoid such iova search failures do a retry from the last rb tree node when iova search fails, this will search the entire tree and get an iova if its available.

---
 drivers/iommu/iova.c | 23 +++++++++++++++++------
 1 file changed, 17 insertions(+), 6 deletions(-)
drivers/iommu/iova.c +17 -6
index 49fc01f..4e77116 100644
@@ -184,8 +184,9 @@ 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 new_pfn;
+	unsigned long new_pfn, low_pfn_new;
 	unsigned long align_mask = ~0UL;
+	unsigned long high_pfn = limit_pfn, low_pfn = iovad->start_pfn;
 
 	if (size_aligned)
 		align_mask <<= fls_long(size - 1);
@@ -198,15 +199,25 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 
 	curr = __get_cached_rbnode(iovad, limit_pfn);
 	curr_iova = rb_entry(curr, struct iova, node);
+	low_pfn_new = curr_iova->pfn_hi + 1;
+
+retry:
 	do {
-		limit_pfn = min(limit_pfn, curr_iova->pfn_lo);
-		new_pfn = (limit_pfn - size) & align_mask;
+		high_pfn = min(high_pfn, curr_iova->pfn_lo);
+		new_pfn = (high_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);
-
-	if (limit_pfn < size || new_pfn < iovad->start_pfn) {
+	} while (curr && new_pfn <= curr_iova->pfn_hi && new_pfn >= low_pfn);
+
+	if (high_pfn < size || new_pfn < low_pfn) {
+		if (low_pfn == iovad->start_pfn && low_pfn_new < limit_pfn) {
+			high_pfn = limit_pfn;
+			low_pfn = low_pfn_new;
+			curr = &iovad->anchor.node;
+			curr_iova = rb_entry(curr, struct iova, node);
+			goto retry;
+		}
 		iovad->max32_alloc_size = size;
 		goto iova32_full;
 	}
Signature · 7 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation
1.9.1

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
#1 lore raw

PATCH v2 2/2 iommu/iova: Free global iova rcache on iova alloc failure

From: Vijayanand Jitta <[email protected]>

When ever an iova alloc request fails we free the iova ranges present in the percpu iova rcaches and then retry but the global iova rcache is not freed as a result we could still see iova alloc failure even after retry as global rcache is holding the iova's which can cause fragmentation. So, free the global iova rcache as well and then go for the retry.

---
 drivers/iommu/iova.c | 23 +++++++++++++++++++++++
 include/linux/iova.h |  6 ++++++
 2 files changed, 29 insertions(+)
drivers/iommu/iova.c +23 -0
index 4e77116..5836c87 100644
@@ -442,6 +442,7 @@ struct iova *find_iova(struct iova_domain *iovad, unsigned long pfn)
 		flush_rcache = false;
 		for_each_online_cpu(cpu)
 			free_cpu_cached_iovas(cpu, iovad);
+		free_global_cached_iovas(iovad);
 		goto retry;
 	}
 
@@ -1055,5 +1056,27 @@ void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad)
 	}
 }
 
+/*
+ * free all the IOVA ranges of global cache
+ */
+void free_global_cached_iovas(struct iova_domain *iovad)
+{
+	struct iova_rcache *rcache;
+	unsigned long flags;
+	int i, j;
+
+	for (i = 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) {
+		rcache = &iovad->rcaches[i];
+		spin_lock_irqsave(&rcache->lock, flags);
+		for (j = 0; j < rcache->depot_size; ++j) {
+			iova_magazine_free_pfns(rcache->depot[j], iovad);
+			iova_magazine_free(rcache->depot[j]);
+			rcache->depot[j] = NULL;
+		}
+		rcache->depot_size = 0;
+		spin_unlock_irqrestore(&rcache->lock, flags);
+	}
+}
+
 MODULE_AUTHOR("Anil S Keshavamurthy <[email protected]>");
 MODULE_LICENSE("GPL");
include/linux/iova.h +6 -0
index a0637ab..a905726 100644
@@ -163,6 +163,7 @@ int init_iova_flush_queue(struct iova_domain *iovad,
 struct iova *split_and_remove_iova(struct iova_domain *iovad,
 	struct iova *iova, unsigned long pfn_lo, unsigned long pfn_hi);
 void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad);
+void free_global_cached_iovas(struct iova_domain *iovad);
 #else
 static inline int iova_cache_get(void)
 {
@@ -270,6 +271,11 @@ static inline void free_cpu_cached_iovas(unsigned int cpu,
 					 struct iova_domain *iovad)
 {
 }
+
+static inline void free_global_cached_iovas(struct iova_domain *iovad)
+{
+}
+
 #endif
 
 #endif
Signature · 7 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation
1.9.1

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
#5 lore raw
In reply to [email protected]
Quoted 70 lines · On 8/20/2020 6:19 PM, [email protected] wrote:

From: Vijayanand Jitta <[email protected]>

When ever a new iova alloc request comes iova is always searched from the cached node and the nodes which are previous to cached node. So, even if there is free iova space available in the nodes which are next to the cached node iova allocation can still fail because of this approach.

Consider the following sequence of iova alloc and frees on 1GB of iova space

1) alloc - 500MB
2) alloc - 12MB
3) alloc - 499MB
4) free - 12MB which was allocated in step 2
5) alloc - 13MB

After the above sequence we will have 12MB of free iova space and cached node will be pointing to the iova pfn of last alloc of 13MB which will be the lowest iova pfn of that iova space. Now if we get an alloc request of 2MB we just search from cached node and then look for lower iova pfn's for free iova and as they aren't any, iova alloc fails though there is 12MB of free iova space.

To avoid such iova search failures do a retry from the last rb tree node when iova search fails, this will search the entire tree and get an iova if its available.

---
 drivers/iommu/iova.c | 23 +++++++++++++++++------
 1 file changed, 17 insertions(+), 6 deletions(-)
drivers/iommu/iova.c +10 -6
index 49fc01f..4e77116 100644
@@ -184,8 +184,9 @@ 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 new_pfn;
+	unsigned long new_pfn, low_pfn_new;
 	unsigned long align_mask = ~0UL;
+	unsigned long high_pfn = limit_pfn, low_pfn = iovad->start_pfn;
 
 	if (size_aligned)
 		align_mask <<= fls_long(size - 1);
@@ -198,15 +199,25 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 
 	curr = __get_cached_rbnode(iovad, limit_pfn);
 	curr_iova = rb_entry(curr, struct iova, node);
+	low_pfn_new = curr_iova->pfn_hi + 1;
+
+retry:
 	do {
-		limit_pfn = min(limit_pfn, curr_iova->pfn_lo);
-		new_pfn = (limit_pfn - size) & align_mask;
+		high_pfn = min(high_pfn, curr_iova->pfn_lo);
+		new_pfn = (high_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);
-
-	if (limit_pfn < size || new_pfn < iovad->start_pfn) {
+	} while (curr && new_pfn <= curr_iova->pfn_hi && new_pfn >= low_pfn);
+
+	if (high_pfn < size || new_pfn < low_pfn) {

drivers/iommu/iova.c static int __alloc_and_insert_iova_range(struct iova_domain *iovad, · line 223

Quoted 10 lines
drivers/iommu/iova.c +7 -0
+		if (low_pfn == iovad->start_pfn && low_pfn_new < limit_pfn) {
+			high_pfn = limit_pfn;
+			low_pfn = low_pfn_new;
+			curr = &iovad->anchor.node;
+			curr_iova = rb_entry(curr, struct iova, node);
+			goto retry;
+		}
 		iovad->max32_alloc_size = size;
 		goto iova32_full;
 	}

ping ?

Thanks,
Vijay

Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of Code Aurora Forum, hosted by The Linux Foundation
#6 lore raw
Quoted 86 lines · Vijayanand Jitta
Quoted 70 lines · On 8/20/2020 6:19 PM, [email protected] wrote:

From: Vijayanand Jitta <[email protected]>

When ever a new iova alloc request comes iova is always searched from the cached node and the nodes which are previous to cached node. So, even if there is free iova space available in the nodes which are next to the cached node iova allocation can still fail because of this approach.

Consider the following sequence of iova alloc and frees on 1GB of iova space

1) alloc - 500MB
2) alloc - 12MB
3) alloc - 499MB
4) free - 12MB which was allocated in step 2
5) alloc - 13MB

After the above sequence we will have 12MB of free iova space and cached node will be pointing to the iova pfn of last alloc of 13MB which will be the lowest iova pfn of that iova space. Now if we get an alloc request of 2MB we just search from cached node and then look for lower iova pfn's for free iova and as they aren't any, iova alloc fails though there is 12MB of free iova space.

To avoid such iova search failures do a retry from the last rb tree node when iova search fails, this will search the entire tree and get an iova if its available.

---
 drivers/iommu/iova.c | 23 +++++++++++++++++------
 1 file changed, 17 insertions(+), 6 deletions(-)
drivers/iommu/iova.c +10 -6
index 49fc01f..4e77116 100644
@@ -184,8 +184,9 @@ 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 new_pfn;
+	unsigned long new_pfn, low_pfn_new;
 	unsigned long align_mask = ~0UL;
+	unsigned long high_pfn = limit_pfn, low_pfn = iovad->start_pfn;
 
 	if (size_aligned)
 		align_mask <<= fls_long(size - 1);
@@ -198,15 +199,25 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 
 	curr = __get_cached_rbnode(iovad, limit_pfn);
 	curr_iova = rb_entry(curr, struct iova, node);
+	low_pfn_new = curr_iova->pfn_hi + 1;
+
+retry:
 	do {
-		limit_pfn = min(limit_pfn, curr_iova->pfn_lo);
-		new_pfn = (limit_pfn - size) & align_mask;
+		high_pfn = min(high_pfn, curr_iova->pfn_lo);
+		new_pfn = (high_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);
-
-	if (limit_pfn < size || new_pfn < iovad->start_pfn) {
+	} while (curr && new_pfn <= curr_iova->pfn_hi && new_pfn >= low_pfn);
+
+	if (high_pfn < size || new_pfn < low_pfn) {

drivers/iommu/iova.c static int __alloc_and_insert_iova_range(struct iova_domain *iovad, · line 223

Quoted 10 lines
drivers/iommu/iova.c +7 -0
+		if (low_pfn == iovad->start_pfn && low_pfn_new < limit_pfn) {
+			high_pfn = limit_pfn;
+			low_pfn = low_pfn_new;
+			curr = &iovad->anchor.node;
+			curr_iova = rb_entry(curr, struct iova, node);
+			goto retry;
+		}
 		iovad->max32_alloc_size = size;
 		goto iova32_full;
 	}

ping ?

Quoted 3 lines

Thanks,
Vijay

ping ?

Thanks,
Vijay

Signature · 6 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of Code Aurora Forum, hosted by The Linux Foundation
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
maintainer
#7 lore raw
Quoted 1 line · Vijayanand Jitta

ping ?

Robin needs to have a look first.

Regards,

Joerg

maintainer Reviewed-by on v2 1/2
#8 lore raw
In reply to [email protected]
Quoted 49 lines · On 2020-08-20 13:49, [email protected] wrote:

From: Vijayanand Jitta <[email protected]>

When ever a new iova alloc request comes iova is always searched from the cached node and the nodes which are previous to cached node. So, even if there is free iova space available in the nodes which are next to the cached node iova allocation can still fail because of this approach.

Consider the following sequence of iova alloc and frees on 1GB of iova space

1) alloc - 500MB
2) alloc - 12MB
3) alloc - 499MB
4) free - 12MB which was allocated in step 2
5) alloc - 13MB

After the above sequence we will have 12MB of free iova space and cached node will be pointing to the iova pfn of last alloc of 13MB which will be the lowest iova pfn of that iova space. Now if we get an alloc request of 2MB we just search from cached node and then look for lower iova pfn's for free iova and as they aren't any, iova alloc fails though there is 12MB of free iova space.

To avoid such iova search failures do a retry from the last rb tree node when iova search fails, this will search the entire tree and get an iova if its available.

---
  drivers/iommu/iova.c | 23 +++++++++++++++++------
  1 file changed, 17 insertions(+), 6 deletions(-)
drivers/iommu/iova.c +2 -1
index 49fc01f..4e77116 100644
@@ -184,8 +184,9 @@ 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 new_pfn;
+	unsigned long new_pfn, low_pfn_new;
  	unsigned long align_mask = ~0UL;
+	unsigned long high_pfn = limit_pfn, low_pfn = iovad->start_pfn;
 
  	if (size_aligned)
  		align_mask <<= fls_long(size - 1);

drivers/iommu/iova.c static int __alloc_and_insert_iova_range(struct iova_domain *iovad, · line 202

Quoted 4 lines
drivers/iommu/iova.c +1 -0
@@ -198,15 +199,25 @@ static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
 
  	curr = __get_cached_rbnode(iovad, limit_pfn);
  	curr_iova = rb_entry(curr, struct iova, node);
+	low_pfn_new = curr_iova->pfn_hi + 1;

Could we call "low_pfn_new" something like "retry_pfn" instead? This code already has unavoidable readability struggles with so many different "pfn"s in play, so having two different meanings of "new" really doesn't help.

Other than that, I think this looks OK (IIRC it's basically what I originally suggested), so with the naming tweaked,

Quoted 27 lines
drivers/iommu/iova.c +14 -5
+
+retry:
  	do {
-		limit_pfn = min(limit_pfn, curr_iova->pfn_lo);
-		new_pfn = (limit_pfn - size) & align_mask;
+		high_pfn = min(high_pfn, curr_iova->pfn_lo);
+		new_pfn = (high_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);
-
-	if (limit_pfn < size || new_pfn < iovad->start_pfn) {
+	} while (curr && new_pfn <= curr_iova->pfn_hi && new_pfn >= low_pfn);
+
+	if (high_pfn < size || new_pfn < low_pfn) {
+		if (low_pfn == iovad->start_pfn && low_pfn_new < limit_pfn) {
+			high_pfn = limit_pfn;
+			low_pfn = low_pfn_new;
+			curr = &iovad->anchor.node;
+			curr_iova = rb_entry(curr, struct iova, node);
+			goto retry;
+		}
  		iovad->max32_alloc_size = size;
  		goto iova32_full;
  	}
maintainer
#2 lore raw

Re: [PATCH v2 2/2] iommu/iova: Free global iova rcache on iova alloc failure

In reply to [email protected]
Quoted 29 lines · On 2020-08-20 13:49, [email protected] wrote:

From: Vijayanand Jitta <[email protected]>

When ever an iova alloc request fails we free the iova ranges present in the percpu iova rcaches and then retry but the global iova rcache is not freed as a result we could still see iova alloc failure even after retry as global rcache is holding the iova's which can cause fragmentation. So, free the global iova rcache as well and then go for the retry.

---
  drivers/iommu/iova.c | 23 +++++++++++++++++++++++
  include/linux/iova.h |  6 ++++++
  2 files changed, 29 insertions(+)
drivers/iommu/iova.c +1 -0
index 4e77116..5836c87 100644
@@ -442,6 +442,7 @@ struct iova *find_iova(struct iova_domain *iovad, unsigned long pfn)
  		flush_rcache = false;
  		for_each_online_cpu(cpu)
  			free_cpu_cached_iovas(cpu, iovad);
+		free_global_cached_iovas(iovad);
  		goto retry;
  	}
 

drivers/iommu/iova.c void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad) · line 1062

Quoted 7 lines
drivers/iommu/iova.c +4 -0
@@ -1055,5 +1056,27 @@ void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad)
  	}
  }
 
+/*
+ * free all the IOVA ranges of global cache
+ */
+void free_global_cached_iovas(struct iova_domain *iovad)

As John pointed out last time, this should be static and the header changes dropped.

(TBH we should probably register our own hotplug notifier instance for a flush queue, so that external code has no need to poke at the per-CPU caches either)

Robin.

Quoted 45 lines
drivers/iommu/iova.c +18 -0
+{
+	struct iova_rcache *rcache;
+	unsigned long flags;
+	int i, j;
+
+	for (i = 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) {
+		rcache = &iovad->rcaches[i];
+		spin_lock_irqsave(&rcache->lock, flags);
+		for (j = 0; j < rcache->depot_size; ++j) {
+			iova_magazine_free_pfns(rcache->depot[j], iovad);
+			iova_magazine_free(rcache->depot[j]);
+			rcache->depot[j] = NULL;
+		}
+		rcache->depot_size = 0;
+		spin_unlock_irqrestore(&rcache->lock, flags);
+	}
+}
+
  MODULE_AUTHOR("Anil S Keshavamurthy <[email protected]>");
  MODULE_LICENSE("GPL");
include/linux/iova.h +6 -0
index a0637ab..a905726 100644
@@ -163,6 +163,7 @@ int init_iova_flush_queue(struct iova_domain *iovad,
  struct iova *split_and_remove_iova(struct iova_domain *iovad,
  	struct iova *iova, unsigned long pfn_lo, unsigned long pfn_hi);
  void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad);
+void free_global_cached_iovas(struct iova_domain *iovad);
  #else
  static inline int iova_cache_get(void)
  {
@@ -270,6 +271,11 @@ static inline void free_cpu_cached_iovas(unsigned int cpu,
  					 struct iova_domain *iovad)
  {
  }
+
+static inline void free_global_cached_iovas(struct iova_domain *iovad)
+{
+}
+
  #endif
 
  #endif
#3 lore raw

Re: [PATCH v2 2/2] iommu/iova: Free global iova rcache on iova alloc failure

Quoted 41 lines · Robin Murphy
Quoted 28 lines · On 2020-08-20 13:49, [email protected] wrote:

From: Vijayanand Jitta <[email protected]>

When ever an iova alloc request fails we free the iova ranges present in the percpu iova rcaches and then retry but the global iova rcache is not freed as a result we could still see iova alloc failure even after retry as global rcache is holding the iova's which can cause fragmentation. So, free the global iova rcache as well and then go for the retry.

---
  drivers/iommu/iova.c | 23 +++++++++++++++++++++++ include/linux/iova.h |  6 ++++++

  2 files changed, 29 insertions(+)
drivers/iommu/iova.c +1 -0
index 4e77116..5836c87 100644
@@ -442,6 +442,7 @@ struct iova *find_iova(struct iova_domain *iovad,
 unsigned long pfn)
           flush_rcache = false;
           for_each_online_cpu(cpu)
               free_cpu_cached_iovas(cpu, iovad);
+        free_global_cached_iovas(iovad);
           goto retry;
       }
Quoted 8 lines

  @@ -1055,5 +1056,27 @@ void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad)
      }
  }
  +/*
+ * free all the IOVA ranges of global cache
+ */
+void free_global_cached_iovas(struct iova_domain *iovad)

As John pointed out last time, this should be static and the header changes dropped.

Quoted 6 lines

(TBH we should probably register our own hotplug notifier instance for a flush queue, so that external code has no need to poke at the per-CPU caches either)

Robin.

Right, I have made it static and dropped header changes in v3. can you please review that.

Thanks,
Vijay

Quoted 46 lines
Quoted 46 lines
drivers/iommu/iova.c +18 -0
+{
+    struct iova_rcache *rcache;
+    unsigned long flags;
+    int i, j;
+
+    for (i = 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) {
+        rcache = &iovad->rcaches[i];
+        spin_lock_irqsave(&rcache->lock, flags);
+        for (j = 0; j < rcache->depot_size; ++j) {
+            iova_magazine_free_pfns(rcache->depot[j], iovad);
+            iova_magazine_free(rcache->depot[j]);
+            rcache->depot[j] = NULL;
+        }
+        rcache->depot_size = 0;
+        spin_unlock_irqrestore(&rcache->lock, flags);
+    }
+}
+

  MODULE_AUTHOR("Anil S Keshavamurthy <[email protected]>"); MODULE_LICENSE("GPL");

include/linux/iova.h +0 -0
index a0637ab..a905726 100644
@@ -163,6 +163,7 @@ int init_iova_flush_queue(struct iova_domain *iovad,
   struct iova *split_and_remove_iova(struct iova_domain *iovad,
       struct iova *iova, unsigned long pfn_lo, unsigned long pfn_hi);

  void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad);
+void free_global_cached_iovas(struct iova_domain *iovad); #else
  static inline int iova_cache_get(void)
  {

include/linux/iova.h +5 -0
@@ -270,6 +271,11 @@ static inline void free_cpu_cached_iovas(unsigned
 int cpu,
                        struct iova_domain *iovad)
   {
   }
+
+static inline void free_global_cached_iovas(struct iova_domain *iovad)
+{
+}
+
   #endif
     #endif
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of Code Aurora Forum, hosted by The Linux Foundation
#9 lore raw
In reply to Robin Murphy
Quoted 60 lines · Robin Murphy
Quoted 49 lines · On 2020-08-20 13:49, [email protected] wrote:

From: Vijayanand Jitta <[email protected]>

When ever a new iova alloc request comes iova is always searched from the cached node and the nodes which are previous to cached node. So, even if there is free iova space available in the nodes which are next to the cached node iova allocation can still fail because of this approach.

Consider the following sequence of iova alloc and frees on 1GB of iova space

1) alloc - 500MB
2) alloc - 12MB
3) alloc - 499MB
4) free -  12MB which was allocated in step 2
5) alloc - 13MB

After the above sequence we will have 12MB of free iova space and cached node will be pointing to the iova pfn of last alloc of 13MB which will be the lowest iova pfn of that iova space. Now if we get an alloc request of 2MB we just search from cached node and then look for lower iova pfn's for free iova and as they aren't any, iova alloc fails though there is 12MB of free iova space.

To avoid such iova search failures do a retry from the last rb tree node when iova search fails, this will search the entire tree and get an iova if its available.

---
  drivers/iommu/iova.c | 23 +++++++++++++++++------

  1 file changed, 17 insertions(+), 6 deletions(-)
drivers/iommu/iova.c +2 -1
index 49fc01f..4e77116 100644
@@ -184,8 +184,9 @@ 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 new_pfn;
+    unsigned long new_pfn, low_pfn_new;
       unsigned long align_mask = ~0UL;
+    unsigned long high_pfn = limit_pfn, low_pfn = iovad->start_pfn;
         if (size_aligned)
           align_mask <<= fls_long(size - 1);

drivers/iommu/iova.c static int __alloc_and_insert_iova_range(struct · line 202

Quoted 4 lines
drivers/iommu/iova.c +1 -0
@@ -198,15 +199,25 @@ static int __alloc_and_insert_iova_range(struct
 iova_domain *iovad,
         curr = __get_cached_rbnode(iovad, limit_pfn);
       curr_iova = rb_entry(curr, struct iova, node);
+    low_pfn_new = curr_iova->pfn_hi + 1;

Could we call "low_pfn_new" something like "retry_pfn" instead? This code already has unavoidable readability struggles with so many different "pfn"s in play, so having two different meanings of "new" really doesn't help.

Quoted 5 lines

Other than that, I think this looks OK (IIRC it's basically what I originally suggested), so with the naming tweaked,

Thanks for review, I have renamed it to retry_pfn in v4.

Thanks,
Vijay

Quoted 28 lines
Quoted 28 lines
drivers/iommu/iova.c +14 -5
+
+retry:
       do {
-        limit_pfn = min(limit_pfn, curr_iova->pfn_lo);
-        new_pfn = (limit_pfn - size) & align_mask;
+        high_pfn = min(high_pfn, curr_iova->pfn_lo);
+        new_pfn = (high_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);
-
-    if (limit_pfn < size || new_pfn < iovad->start_pfn) {
+    } while (curr && new_pfn <= curr_iova->pfn_hi && new_pfn >=
 low_pfn);
+
+    if (high_pfn < size || new_pfn < low_pfn) {
+        if (low_pfn == iovad->start_pfn && low_pfn_new < limit_pfn) {
+            high_pfn = limit_pfn;
+            low_pfn = low_pfn_new;
+            curr = &iovad->anchor.node;
+            curr_iova = rb_entry(curr, struct iova, node);
+            goto retry;
+        }
           iovad->max32_alloc_size = size;
           goto iova32_full;

      }

Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of Code Aurora Forum, hosted by The Linux Foundation
#4 lore raw

Re: [PATCH v2 2/2] iommu/iova: Free global iova rcache on iova alloc failure

In reply to Vijayanand Jitta
Quoted 48 lines · Vijayanand Jitta
Quoted 45 lines · Robin Murphy
Quoted 28 lines · On 2020-08-20 13:49, [email protected] wrote:

From: Vijayanand Jitta <[email protected]>

When ever an iova alloc request fails we free the iova ranges present in the percpu iova rcaches and then retry but the global iova rcache is not freed as a result we could still see iova alloc failure even after retry as global rcache is holding the iova's which can cause fragmentation. So, free the global iova rcache as well and then go for the retry.

---
  drivers/iommu/iova.c | 23 +++++++++++++++++++++++ include/linux/iova.h |  6 ++++++

  2 files changed, 29 insertions(+)
drivers/iommu/iova.c +1 -0
index 4e77116..5836c87 100644
@@ -442,6 +442,7 @@ struct iova *find_iova(struct iova_domain *iovad,
 unsigned long pfn)
           flush_rcache = false;
           for_each_online_cpu(cpu)
               free_cpu_cached_iovas(cpu, iovad);
+        free_global_cached_iovas(iovad);
           goto retry;
       }
Quoted 8 lines

  @@ -1055,5 +1056,27 @@ void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad)
      }
  }
  +/*
+ * free all the IOVA ranges of global cache
+ */
+void free_global_cached_iovas(struct iova_domain *iovad)

As John pointed out last time, this should be static and the header changes dropped.

(TBH we should probably register our own hotplug notifier instance for a flush queue, so that external code has no need to poke at the per-CPU caches either)

Quoted 8 lines
Quoted 2 lines

Robin.

Right, I have made it static and dropped header changes in v3. can you please review that.

Thanks,
Vijay

Please review v4 instead of v3, I have updated other patch as well in v4.

Thanks,
Vijay

Quoted 47 lines
Quoted 46 lines
Quoted 46 lines
drivers/iommu/iova.c +18 -0
+{
+    struct iova_rcache *rcache;
+    unsigned long flags;
+    int i, j;
+
+    for (i = 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) {
+        rcache = &iovad->rcaches[i];
+        spin_lock_irqsave(&rcache->lock, flags);
+        for (j = 0; j < rcache->depot_size; ++j) {
+            iova_magazine_free_pfns(rcache->depot[j], iovad);
+            iova_magazine_free(rcache->depot[j]);
+            rcache->depot[j] = NULL;
+        }
+        rcache->depot_size = 0;
+        spin_unlock_irqrestore(&rcache->lock, flags);
+    }
+}
+

  MODULE_AUTHOR("Anil S Keshavamurthy <[email protected]>"); MODULE_LICENSE("GPL");

include/linux/iova.h +0 -0
index a0637ab..a905726 100644
@@ -163,6 +163,7 @@ int init_iova_flush_queue(struct iova_domain *iovad,
   struct iova *split_and_remove_iova(struct iova_domain *iovad,
       struct iova *iova, unsigned long pfn_lo, unsigned long pfn_hi);

  void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad);
+void free_global_cached_iovas(struct iova_domain *iovad); #else
  static inline int iova_cache_get(void)
  {

include/linux/iova.h +5 -0
@@ -270,6 +271,11 @@ static inline void free_cpu_cached_iovas(unsigned
 int cpu,
                        struct iova_domain *iovad)
   {
   }
+
+static inline void free_global_cached_iovas(struct iova_domain *iovad)
+{
+}
+
   #endif
     #endif
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of Code Aurora Forum, hosted by The Linux Foundation