Skip to contentSkip to conversation

memcg: replace memcg ID idr with xarray

PATCH Applied Shakeel Butt 5 participants 6 messages Aug 15, 2024 – Aug 16, 2024

Subsystems (3)
lore
Share & tools
lore

Shakeel Butt

author maintainer
#0 lore raw

PATCH v2 1/1 memcg: replace memcg ID idr with xarray

At the moment memcg IDs are managed through IDR which requires external synchronization mechanisms and makes the allocation code a bit awkward. Let's switch to xarray and make the code simpler.

---

Changes since v1:
- Fix error path in mem_cgroup_alloc (Dan Carpenter)

 mm/memcontrol.c | 39 ++++++++++-----------------------------
 1 file changed, 10 insertions(+), 29 deletions(-)
mm/memcontrol.c +10 -29
index df84683a0e1c..e8e03a5e1e5e 100644
@@ -3408,29 +3408,12 @@ static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
  */
 
 #define MEM_CGROUP_ID_MAX	((1UL << MEM_CGROUP_ID_SHIFT) - 1)
-static DEFINE_IDR(mem_cgroup_idr);
-static DEFINE_SPINLOCK(memcg_idr_lock);
-
-static int mem_cgroup_alloc_id(void)
-{
-	int ret;
-
-	idr_preload(GFP_KERNEL);
-	spin_lock(&memcg_idr_lock);
-	ret = idr_alloc(&mem_cgroup_idr, NULL, 1, MEM_CGROUP_ID_MAX + 1,
-			GFP_NOWAIT);
-	spin_unlock(&memcg_idr_lock);
-	idr_preload_end();
-	return ret;
-}
+static DEFINE_XARRAY_ALLOC1(mem_cgroup_ids);
 
 static void mem_cgroup_id_remove(struct mem_cgroup *memcg)
 {
 	if (memcg->id.id > 0) {
-		spin_lock(&memcg_idr_lock);
-		idr_remove(&mem_cgroup_idr, memcg->id.id);
-		spin_unlock(&memcg_idr_lock);
-
+		xa_erase(&mem_cgroup_ids, memcg->id.id);
 		memcg->id.id = 0;
 	}
 }
@@ -3465,7 +3448,7 @@ static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
 struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
 {
 	WARN_ON_ONCE(!rcu_read_lock_held());
-	return idr_find(&mem_cgroup_idr, id);
+	return xa_load(&mem_cgroup_ids, id);
 }
 
 #ifdef CONFIG_SHRINKER_DEBUG
@@ -3558,17 +3541,17 @@ static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent)
 	struct mem_cgroup *memcg;
 	int node, cpu;
 	int __maybe_unused i;
-	long error = -ENOMEM;
+	long error;
 
 	memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL);
 	if (!memcg)
-		return ERR_PTR(error);
+		return ERR_PTR(-ENOMEM);
 
-	memcg->id.id = mem_cgroup_alloc_id();
-	if (memcg->id.id < 0) {
-		error = memcg->id.id;
+	error = xa_alloc(&mem_cgroup_ids, &memcg->id.id, NULL,
+			 XA_LIMIT(1, MEM_CGROUP_ID_MAX), GFP_KERNEL);
+	if (error)
 		goto fail;
-	}
+	error = -ENOMEM;
 
 	memcg->vmstats = kzalloc(sizeof(struct memcg_vmstats),
 				 GFP_KERNEL_ACCOUNT);
@@ -3709,9 +3692,7 @@ static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
 	 * publish it here at the end of onlining. This matches the
 	 * regular ID destruction during offlining.
 	 */
-	spin_lock(&memcg_idr_lock);
-	idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
-	spin_unlock(&memcg_idr_lock);
+	xa_store(&mem_cgroup_ids, memcg->id.id, memcg, GFP_KERNEL);
 
 	return 0;
 offline_kmem:
Signature · 1 line
2.43.5
maintainer
#1 lore raw
Quoted 76 lines · Shakeel Butt

At the moment memcg IDs are managed through IDR which requires external synchronization mechanisms and makes the allocation code a bit awkward. Let's switch to xarray and make the code simpler.

---

Changes since v1:
- Fix error path in mem_cgroup_alloc (Dan Carpenter)

 mm/memcontrol.c | 39 ++++++++++-----------------------------
 1 file changed, 10 insertions(+), 29 deletions(-)
mm/memcontrol.c +5 -22
index df84683a0e1c..e8e03a5e1e5e 100644
@@ -3408,29 +3408,12 @@ static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
  */
 
 #define MEM_CGROUP_ID_MAX	((1UL << MEM_CGROUP_ID_SHIFT) - 1)
-static DEFINE_IDR(mem_cgroup_idr);
-static DEFINE_SPINLOCK(memcg_idr_lock);
-
-static int mem_cgroup_alloc_id(void)
-{
-	int ret;
-
-	idr_preload(GFP_KERNEL);
-	spin_lock(&memcg_idr_lock);
-	ret = idr_alloc(&mem_cgroup_idr, NULL, 1, MEM_CGROUP_ID_MAX + 1,
-			GFP_NOWAIT);
-	spin_unlock(&memcg_idr_lock);
-	idr_preload_end();
-	return ret;
-}
+static DEFINE_XARRAY_ALLOC1(mem_cgroup_ids);
 
 static void mem_cgroup_id_remove(struct mem_cgroup *memcg)
 {
 	if (memcg->id.id > 0) {
-		spin_lock(&memcg_idr_lock);
-		idr_remove(&mem_cgroup_idr, memcg->id.id);
-		spin_unlock(&memcg_idr_lock);
-
+		xa_erase(&mem_cgroup_ids, memcg->id.id);
 		memcg->id.id = 0;
 	}
 }
@@ -3465,7 +3448,7 @@ static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
 struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
 {
 	WARN_ON_ONCE(!rcu_read_lock_held());
-	return idr_find(&mem_cgroup_idr, id);
+	return xa_load(&mem_cgroup_ids, id);
 }
 
 #ifdef CONFIG_SHRINKER_DEBUG
@@ -3558,17 +3541,17 @@ static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent)
 	struct mem_cgroup *memcg;
 	int node, cpu;
 	int __maybe_unused i;
-	long error = -ENOMEM;
+	long error;
 
 	memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL);
 	if (!memcg)
-		return ERR_PTR(error);
+		return ERR_PTR(-ENOMEM);

mm/memcontrol.c static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent) · line 3554

Quoted 10 lines
mm/memcontrol.c +4 -4
 
-	memcg->id.id = mem_cgroup_alloc_id();
-	if (memcg->id.id < 0) {
-		error = memcg->id.id;
+	error = xa_alloc(&mem_cgroup_ids, &memcg->id.id, NULL,
+			 XA_LIMIT(1, MEM_CGROUP_ID_MAX), GFP_KERNEL);
+	if (error)
 		goto fail;
-	}
+	error = -ENOMEM;

There is another subtle change here: xa_alloc() returns -EBUSY in the case of the address space exhaustion, while the old code returned -ENOSPC. It's unlikely a big practical problem.

#2 lore raw
Quoted 3 lines · Roman Gushchin

There is another subtle change here: xa_alloc() returns -EBUSY in the case of the address space exhaustion, while the old code returned -ENOSPC. It's unlikely a big practical problem.

I decided that EBUSY was the right errno for this situation;

#define EBUSY           16      /* Device or resource busy */
#define ENOSPC          28      /* No space left on device */

ENOSPC seemed wrong; the device isn't out of space.

#3 lore raw
Quoted 11 lines · Matthew Wilcox
Quoted 3 lines · Roman Gushchin

There is another subtle change here: xa_alloc() returns -EBUSY in the case of the address space exhaustion, while the old code returned -ENOSPC. It's unlikely a big practical problem.

I decided that EBUSY was the right errno for this situation;

#define EBUSY           16      /* Device or resource busy */
#define ENOSPC          28      /* No space left on device */

ENOSPC seemed wrong; the device isn't out of space.

The thing is that this is observable by userspace - mkdir would return a different and potentially unexpected errno. We can try and see whether anybody complains or just translate the error.

Signature · 2 lines
Michal Hocko
SUSE Labs
maintainer
#4 lore raw
Quoted 8 lines · On Fri, 16 Aug 2024 09:19:58 +0200 Michal Hocko <[email protected]> wrote:
Quoted 7 lines · Matthew Wilcox
Quoted 3 lines · Roman Gushchin

There is another subtle change here: xa_alloc() returns -EBUSY in the case of the address space exhaustion, while the old code returned -ENOSPC. It's unlikely a big practical problem.

I decided that EBUSY was the right errno for this situation;

Quoted 8 lines
Quoted 4 lines
#define EBUSY           16      /* Device or resource busy */
#define ENOSPC          28      /* No space left on device */

ENOSPC seemed wrong; the device isn't out of space.

The thing is that this is observable by userspace - mkdir would return a different and potentially unexpected errno. We can try and see whether anybody complains or just translate the error.

The mkdir(2) manpage doesn't list EBUSY. Maybe ENOMEM is close enough.

#5 lore raw
Quoted 13 lines · Andrew Morton
Quoted 11 lines · On Fri, 16 Aug 2024 09:19:58 +0200 Michal Hocko <[email protected]> wrote:
Quoted 10 lines · Matthew Wilcox
Quoted 3 lines · Roman Gushchin

There is another subtle change here: xa_alloc() returns -EBUSY in the case of the address space exhaustion, while the old code returned -ENOSPC. It's unlikely a big practical problem.

I decided that EBUSY was the right errno for this situation;

#define EBUSY           16      /* Device or resource busy */
#define ENOSPC          28      /* No space left on device */
Quoted 7 lines
Quoted 5 lines
Quoted 1 line

ENOSPC seemed wrong; the device isn't out of space.

The thing is that this is observable by userspace - mkdir would return a different and potentially unexpected errno. We can try and see whether anybody complains or just translate the error.

The mkdir(2) manpage doesn't list EBUSY. Maybe ENOMEM is close enough.

it used to report ENOSPC with xarray.

Signature · 2 lines
Michal Hocko
SUSE Labs