Skip to contentSkip to conversation

bpf: Reduce memory usage for bpf_global_percpu_ma

PATCH Applied Yonghong Song 4 participants 16 messages Dec 22, 2023 – Jan 4, 2024

Subsystems (4)
lore
Share & tools
lore

Yonghong Song

author reviewer
#0 lore raw

v6 cover bpf: Reduce memory usage for bpf_global_percpu_ma

Currently when a bpf program intends to allocate memory for percpu kptr, the verifier will call bpf_mem_alloc_init() to prefill all supported unit sizes and this caused memory consumption very big for large number of cpus. For example, for 128-cpu system, the total memory consumption with initial prefill is ~175MB. Things will become worse for systems with even more cpus.

Patch 1 avoids unnecessary extra percpu memory allocation. Patch 2 adds objcg to bpf_mem_alloc at init stage so objcg can be associated with root cgroup and objcg can be passed to later bpf_mem_alloc_percpu_unit_init().
Patch 3 addresses memory consumption issue by avoiding to prefill with all unit sizes, i.e. only prefilling with user specified size. Patch 4 further reduces memory consumption by limiting the number of prefill entries for percpu memory allocation. Patch 5 has much smaller low/high watermarks for percpu allocation to reduce memory consumption.
Patch 6 rejects percpu memory allocation with bpf_global_percpu_ma when allocation size is greater than 512 bytes.
Patch 7 fixed test_bpf_ma test due to Patch 5.
Patch 8 added one test to show the verification failure log message.

Changelogs:
v5 -> v6:
. Change bpf_mem_alloc_percpu_init() to add objcg as one of parameters.
For bpf_global_percpu_ma, the objcg is NULL, corresponding root memcg.
v4 -> v5:
. Do not do bpf_global_percpu_ma initialization at init stage, instead
doing initialization when the verifier knows it is going to be used by bpf prog.
. Using much smaller low/high watermarks for percpu allocation.
v3 -> v4:
. Add objcg to bpf_mem_alloc during init stage.
. Initialize objcg at init stage but use it in bpf_mem_alloc_percpu_unit_init(). . Remove check_obj_size() in bpf_mem_alloc_percpu_unit_init().
v2 -> v3:
. Clear the bpf_mem_cache if prefill fails.
. Change test_bpf_ma percpu allocation tests to use bucket_size
as allocation size instead of bucket_size - 8.
. Remove __GFP_ZERO flag from __alloc_percpu_gfp() call.
v1 -> v2:
. Avoid unnecessary extra percpu memory allocation.
. Add a separate function to do bpf_global_percpu_ma initialization . promote.
. Promote function static 'sizes' array to file static.
. Add comments to explain to refill only one item for percpu alloc.

Yonghong Song (8):
bpf: Avoid unnecessary extra percpu memory allocation
bpf: Add objcg to bpf_mem_alloc
bpf: Allow per unit prefill for non-fix-size percpu memory allocator
bpf: Refill only one percpu element in memalloc
bpf: Use smaller low/high marks for percpu allocation
bpf: Limit up to 512 bytes for bpf_global_percpu_ma allocation selftests/bpf: Cope with 512 bytes limit with bpf_global_percpu_ma selftests/bpf: Add a selftest with > 512-byte percpu allocation size

 include/linux/bpf_mem_alloc.h                 |  8 ++
 kernel/bpf/memalloc.c                         | 93 ++++++++++++++++---
 kernel/bpf/verifier.c                         | 45 ++++++---
 .../selftests/bpf/prog_tests/test_bpf_ma.c    | 20 ++--
 .../selftests/bpf/progs/percpu_alloc_fail.c   | 18 ++++
 .../testing/selftests/bpf/progs/test_bpf_ma.c | 66 ++++++-------
 6 files changed, 184 insertions(+), 66 deletions(-)
Signature · 1 line
2.34.1
author reviewer
#1 lore raw

PATCH v6 1/8 bpf: Avoid unnecessary extra percpu memory allocation

Currently, for percpu memory allocation, say if the user requests allocation size to be 32 bytes, the actually calculated size will be 40 bytes and it further rounds to 64 bytes, and eventually 64 bytes are allocated, wasting 32-byte memory.

Change bpf_mem_alloc() to calculate the cache index based on the user-provided allocation size so unnecessary extra memory can be avoided.

---
 kernel/bpf/memalloc.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
kernel/bpf/memalloc.c +3 -1
index aa0fbf000a12..288ec4a967d0 100644
@@ -833,7 +833,9 @@ void notrace *bpf_mem_alloc(struct bpf_mem_alloc *ma, size_t size)
 	if (!size)
 		return NULL;
 
-	idx = bpf_mem_cache_idx(size + LLIST_NODE_SZ);
+	if (!ma->percpu)
+		size += LLIST_NODE_SZ;
+	idx = bpf_mem_cache_idx(size);
 	if (idx < 0)
 		return NULL;
 
Signature · 1 line
2.34.1
author reviewer
#2 lore raw

PATCH v6 2/8 bpf: Add objcg to bpf_mem_alloc

In reply to Yonghong Song

The objcg is a bpf_mem_alloc level property since all bpf_mem_cache's are with the same objcg. This patch made such a property explicit. The next patch will use this property to save and restore objcg for percpu unit allocator.

---
 include/linux/bpf_mem_alloc.h |  1 +
 kernel/bpf/memalloc.c         | 11 ++++++-----
 2 files changed, 7 insertions(+), 5 deletions(-)
include/linux/bpf_mem_alloc.h +1 -0
index bb1223b21308..acef8c808599 100644
@@ -11,6 +11,7 @@ struct bpf_mem_caches;
 struct bpf_mem_alloc {
 	struct bpf_mem_caches __percpu *caches;
 	struct bpf_mem_cache __percpu *cache;
+	struct obj_cgroup *objcg;
 	bool percpu;
 	struct work_struct work;
 };
kernel/bpf/memalloc.c +6 -5
index 288ec4a967d0..4a21050f0359 100644
@@ -523,6 +523,7 @@ int bpf_mem_alloc_init(struct bpf_mem_alloc *ma, int size, bool percpu)
 		if (memcg_bpf_enabled())
 			objcg = get_obj_cgroup_from_current();
 #endif
+		ma->objcg = objcg;
 		for_each_possible_cpu(cpu) {
 			c = per_cpu_ptr(pc, cpu);
 			c->unit_size = unit_size;
@@ -542,6 +543,7 @@ int bpf_mem_alloc_init(struct bpf_mem_alloc *ma, int size, bool percpu)
 #ifdef CONFIG_MEMCG_KMEM
 	objcg = get_obj_cgroup_from_current();
 #endif
+	ma->objcg = objcg;
 	for_each_possible_cpu(cpu) {
 		cc = per_cpu_ptr(pcc, cpu);
 		for (i = 0; i < NUM_CACHES; i++) {
@@ -691,9 +693,8 @@ void bpf_mem_alloc_destroy(struct bpf_mem_alloc *ma)
 			rcu_in_progress += atomic_read(&c->call_rcu_ttrace_in_progress);
 			rcu_in_progress += atomic_read(&c->call_rcu_in_progress);
 		}
-		/* objcg is the same across cpus */
-		if (c->objcg)
-			obj_cgroup_put(c->objcg);
+		if (ma->objcg)
+			obj_cgroup_put(ma->objcg);
 		destroy_mem_alloc(ma, rcu_in_progress);
 	}
 	if (ma->caches) {
@@ -709,8 +710,8 @@ void bpf_mem_alloc_destroy(struct bpf_mem_alloc *ma)
 				rcu_in_progress += atomic_read(&c->call_rcu_in_progress);
 			}
 		}
-		if (c->objcg)
-			obj_cgroup_put(c->objcg);
+		if (ma->objcg)
+			obj_cgroup_put(ma->objcg);
 		destroy_mem_alloc(ma, rcu_in_progress);
 	}
 }
Signature · 1 line
2.34.1
author reviewer
#3 lore raw

PATCH v6 3/8 bpf: Allow per unit prefill for non-fix-size percpu memory allocator

In reply to Yonghong Song

Commit 41a5db8d8161 ("Add support for non-fix-size percpu mem allocation") added support for non-fix-size percpu memory allocation.
Such allocation will allocate percpu memory for all buckets on all cpus and the memory consumption is in the order to quadratic. For example, let us say, 4 cpus, unit size 16 bytes, so each cpu has 16 * 4 = 64 bytes, with 4 cpus, total will be 64 * 4 = 256 bytes. Then let us say, 8 cpus with the same unit size, each cpu
has 16 * 8 = 128 bytes, with 8 cpus, total will be 128 * 8 = 1024 bytes. So if the number of cpus doubles, the number of memory consumption will be 4 times. So for a system with large number of cpus, the memory consumption goes up quickly with quadratic order.
For example, for 4KB percpu allocation, 128 cpus. The total memory consumption will 4KB * 128 * 128 = 64MB. Things will become worse if the number of cpus is bigger (e.g., 512, 1024, etc.)

In Commit 41a5db8d8161, the non-fix-size percpu memory allocation is done in boot time, so for system with large number of cpus, the initial percpu memory consumption is very visible. For example, for 128 cpu system, the total percpu memory allocation will be at least (16 + 32 + 64 + 96 + 128 + 196 + 256 + 512 + 1024 + 2048 + 4096)
* 128 * 128 = ~138MB.
which is pretty big. It will be even bigger for larger number of cpus.

Note that the current prefill also allocates 4 entries if the unit size is less than 256. So on top of 138MB memory consumption, this will add more consumption with
3 * (16 + 32 + 64 + 96 + 128 + 196 + 256) * 128 * 128 = ~38MB. Next patch will try to reduce this memory consumption.

Later on, Commit 1fda5bb66ad8 ("bpf: Do not allocate percpu memory at init stage") moved the non-fix-size percpu memory allocation to bpf verificaiton stage. Once a particular bpf_percpu_obj_new() is called by bpf program, the memory allocator will try to fill in the cache with all sizes, causing the same amount of percpu memory consumption as in the boot stage.

To reduce the initial percpu memory consumption for non-fix-size percpu memory allocation, instead of filling the cache with all supported allocation sizes, this patch intends to fill the cache only for the requested size. As typically users will not use large percpu data structure, this can save memory significantly. For example, the allocation size is 64 bytes with 128 cpus. Then total percpu memory amount will be 64 * 128 * 128 = 1MB, much less than previous 138MB.

---
 include/linux/bpf_mem_alloc.h |  7 +++++
 kernel/bpf/memalloc.c         | 57 ++++++++++++++++++++++++++++++++++-
 kernel/bpf/verifier.c         | 37 ++++++++++++++---------
 3 files changed, 86 insertions(+), 15 deletions(-)
include/linux/bpf_mem_alloc.h +7 -0
index acef8c808599..aaf004d94322 100644
@@ -22,8 +22,15 @@ struct bpf_mem_alloc {
  * 'size = 0' is for bpf_mem_alloc which manages many fixed-size objects.
  * Alloc and free are done with bpf_mem_{alloc,free}() and the size of
  * the returned object is given by the size argument of bpf_mem_alloc().
+ * If percpu equals true, error will be returned in order to avoid
+ * large memory consumption and the below bpf_mem_alloc_percpu_unit_init()
+ * should be used to do on-demand per-cpu allocation for each size.
  */
 int bpf_mem_alloc_init(struct bpf_mem_alloc *ma, int size, bool percpu);
+/* Initialize a non-fix-size percpu memory allocator */
+int bpf_mem_alloc_percpu_init(struct bpf_mem_alloc *ma, struct obj_cgroup *objcg);
+/* The percpu allocation with a specific unit size. */
+int bpf_mem_alloc_percpu_unit_init(struct bpf_mem_alloc *ma, int size);
 void bpf_mem_alloc_destroy(struct bpf_mem_alloc *ma);
 
 /* kmalloc/kfree equivalent: */
kernel/bpf/memalloc.c +56 -1

86 lines not on this page · raw message

kernel/bpf/verifier.c +23 -14
index f13008d27f35..5b2bfa2cbb7b 100644
@@ -12141,20 +12141,6 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
 				if (meta.func_id == special_kfunc_list[KF_bpf_obj_new_impl] && !bpf_global_ma_set)
 					return -ENOMEM;
 
-				if (meta.func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl]) {
-					if (!bpf_global_percpu_ma_set) {
-						mutex_lock(&bpf_percpu_ma_lock);
-						if (!bpf_global_percpu_ma_set) {
-							err = bpf_mem_alloc_init(&bpf_global_percpu_ma, 0, true);
-							if (!err)
-								bpf_global_percpu_ma_set = true;
-						}
-						mutex_unlock(&bpf_percpu_ma_lock);
-						if (err)
-							return err;
-					}
-				}
-
 				if (((u64)(u32)meta.arg_constant.value) != meta.arg_constant.value) {
 					verbose(env, "local type ID argument must be in range [0, U32_MAX]\n");
 					return -EINVAL;
@@ -12175,6 +12161,29 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
 					return -EINVAL;
 				}
 
+				if (meta.func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl]) {
+					if (!bpf_global_percpu_ma_set) {
+						mutex_lock(&bpf_percpu_ma_lock);
+						if (!bpf_global_percpu_ma_set) {
+							/* Charge memory allocated with bpf_global_percpu_ma to
+							 * root memcg. The obj_cgroup for root memcg is NULL.
+							 */
+							err = bpf_mem_alloc_percpu_init(&bpf_global_percpu_ma, NULL);
+							if (!err)
+								bpf_global_percpu_ma_set = true;
+						}
+						mutex_unlock(&bpf_percpu_ma_lock);
+						if (err)
+							return err;
+					}
+
+					mutex_lock(&bpf_percpu_ma_lock);
+					err = bpf_mem_alloc_percpu_unit_init(&bpf_global_percpu_ma, ret_t->size);
+					mutex_unlock(&bpf_percpu_ma_lock);
+					if (err)
+						return err;
+				}
+
 				struct_meta = btf_find_struct_meta(ret_btf, ret_btf_id);
 				if (meta.func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl]) {
 					if (!__btf_type_is_scalar_struct(env, ret_btf, ret_t, 0)) {
Signature · 1 line
2.34.1
Acked-by on v6 3/8
#4 lore raw

Re: [PATCH bpf-next v6 3/8] bpf: Allow per unit prefill for non-fix-size percpu memory allocator

Quoted 45 lines · Yonghong Song

Commit 41a5db8d8161 ("Add support for non-fix-size percpu mem allocation") added support for non-fix-size percpu memory allocation.
Such allocation will allocate percpu memory for all buckets on all cpus and the memory consumption is in the order to quadratic. For example, let us say, 4 cpus, unit size 16 bytes, so each cpu has 16 * 4 = 64 bytes, with 4 cpus, total will be 64 * 4 = 256 bytes. Then let us say, 8 cpus with the same unit size, each cpu
has 16 * 8 = 128 bytes, with 8 cpus, total will be 128 * 8 = 1024 bytes. So if the number of cpus doubles, the number of memory consumption will be 4 times. So for a system with large number of cpus, the memory consumption goes up quickly with quadratic order.
For example, for 4KB percpu allocation, 128 cpus. The total memory consumption will 4KB * 128 * 128 = 64MB. Things will become worse if the number of cpus is bigger (e.g., 512, 1024, etc.)

In Commit 41a5db8d8161, the non-fix-size percpu memory allocation is done in boot time, so for system with large number of cpus, the initial percpu memory consumption is very visible. For example, for 128 cpu system, the total percpu memory allocation will be at least (16 + 32 + 64 + 96 + 128 + 196 + 256 + 512 + 1024 + 2048 + 4096)
* 128 * 128 = ~138MB.
which is pretty big. It will be even bigger for larger number of cpus.

Note that the current prefill also allocates 4 entries if the unit size is less than 256. So on top of 138MB memory consumption, this will add more consumption with
3 * (16 + 32 + 64 + 96 + 128 + 196 + 256) * 128 * 128 = ~38MB. Next patch will try to reduce this memory consumption.

Later on, Commit 1fda5bb66ad8 ("bpf: Do not allocate percpu memory at init stage") moved the non-fix-size percpu memory allocation to bpf verificaiton stage. Once a particular bpf_percpu_obj_new() is called by bpf program, the memory allocator will try to fill in the cache with all sizes, causing the same amount of percpu memory consumption as in the boot stage.

To reduce the initial percpu memory consumption for non-fix-size percpu memory allocation, instead of filling the cache with all supported allocation sizes, this patch intends to fill the cache only for the requested size. As typically users will not use large percpu data structure, this can save memory significantly. For example, the allocation size is 64 bytes with 128 cpus. Then total percpu memory amount will be 64 * 128 * 128 = 1MB, much less than previous 138MB.

Quoted 1 line
#5 lore raw

Re: [PATCH bpf-next v6 3/8] bpf: Allow per unit prefill for non-fix-size percpu memory allocator

In reply to Yonghong Song
Quoted 5 lines · Yonghong Song

Later on, Commit 1fda5bb66ad8 ("bpf: Do not allocate percpu memory at init stage") moved the non-fix-size percpu memory allocation to bpf verificaiton stage.

Only noticed after already applying and didn't want to force push to fix it up.
But in the future please do not break commit one-liner into multiple lines. Commit should be lower case as well.

spellcheck is a good idea as well.

kernel/bpf/memalloc.c int bpf_mem_alloc_init(struct bpf_mem_alloc *ma, int size, bool percpu) · line 606

Quoted 6 lines
kernel/bpf/memalloc.c +6 -0
+
+       for_each_possible_cpu(cpu) {
+               cc = per_cpu_ptr(pcc, cpu);
+               c = &cc->cache[i];
+               if (cpu == 0 && c->unit_size)
+                       break;

I think this part ensures that repeated bpf_percpu_obj_new() in a bpf prog don't keep prefilling, right?
I think it works, but cpu == 0 part is confusing. It will work with just: if (c->unit_size) break; right?

Quoted 12 lines
kernel/bpf/memalloc.c +12 -0
+
+               c->unit_size = unit_size;
+               c->objcg = objcg;
+               c->percpu_size = percpu_size;
+               c->tgt = c;
+
+               init_refill_work(c);
+               prefill_mem_cache(c, cpu);
+       }
+
+       return 0;
+}
author reviewer
#6 lore raw

Re: [PATCH bpf-next v6 3/8] bpf: Allow per unit prefill for non-fix-size percpu memory allocator

Quoted 9 lines · Alexei Starovoitov
Quoted 4 lines · Yonghong Song

Later on, Commit 1fda5bb66ad8 ("bpf: Do not allocate percpu memory at init stage") moved the non-fix-size percpu memory allocation to bpf verificaiton stage.

Only noticed after already applying and didn't want to force push to fix it up.
But in the future please do not break commit one-liner into multiple lines. Commit should be lower case as well.

Sorry about the format issue. Will pay attention to this next time and as you suggested, will run spellcheck as well.

Quoted 9 lines

spellcheck is a good idea as well.

kernel/bpf/memalloc.c int bpf_mem_alloc_init(struct bpf_mem_alloc *ma, int size, bool percpu) · line 606

Quoted 6 lines
kernel/bpf/memalloc.c +6 -0
+
+       for_each_possible_cpu(cpu) {
+               cc = per_cpu_ptr(pcc, cpu);
+               c = &cc->cache[i];
+               if (cpu == 0 && c->unit_size)
+                       break;
Quoted 6 lines

I think this part ensures that repeated bpf_percpu_obj_new() in a bpf prog don't keep prefilling, right?
I think it works, but cpu == 0 part is confusing. It will work with just: if (c->unit_size) break; right?

You are right. cpu == 0 is a leftover from one of previous revisions and should be removed. Will send a followup to remove it.

Quoted 13 lines
Quoted 13 lines
kernel/bpf/memalloc.c +12 -0
+
+               c->unit_size = unit_size;
+               c->objcg = objcg;
+               c->percpu_size = percpu_size;
+               c->tgt = c;
+
+               init_refill_work(c);
+               prefill_mem_cache(c, cpu);
+       }
+
+       return 0;
+}
author reviewer
#7 lore raw

PATCH v6 4/8 bpf: Refill only one percpu element in memalloc

In reply to Yonghong Song

Typically for percpu map element or data structure, once allocated, most operations are lookup or in-place update. Deletion are really rare. Currently, for percpu data strcture, 4 elements will be refilled if the size is <= 256. Let us just do with one element for percpu data. For example, for size 256 and 128 cpus, the potential saving will be 3 * 256 * 128 * 128 = 12MB.

---
 kernel/bpf/memalloc.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
kernel/bpf/memalloc.c +9 -4
index f71da07eb8a0..a8ee6fb8401c 100644
@@ -485,11 +485,16 @@ static void init_refill_work(struct bpf_mem_cache *c)
 
 static void prefill_mem_cache(struct bpf_mem_cache *c, int cpu)
 {
-	/* To avoid consuming memory assume that 1st run of bpf
-	 * prog won't be doing more than 4 map_update_elem from
-	 * irq disabled region
+	int cnt = 1;
+
+	/* To avoid consuming memory, for non-percpu allocation, assume that
+	 * 1st run of bpf prog won't be doing more than 4 map_update_elem from
+	 * irq disabled region if unit size is less than or equal to 256.
+	 * For all other cases, let us just do one allocation.
 	 */
-	alloc_bulk(c, c->unit_size <= 256 ? 4 : 1, cpu_to_node(cpu), false);
+	if (!c->percpu_size && c->unit_size <= 256)
+		cnt = 4;
+	alloc_bulk(c, cnt, cpu_to_node(cpu), false);
 }
 
 /* When size != 0 bpf_mem_cache for each cpu.
Signature · 1 line
2.34.1
author reviewer
#8 lore raw

PATCH v6 5/8 bpf: Use smaller low/high marks for percpu allocation

In reply to Yonghong Song

Currently, refill low/high marks are set with the assumption of normal non-percpu memory allocation. For example, for an allocation size 256, for non-percpu memory allocation, low mark is 32 and high mark is 96, resulting in the batch allocation of 48 elements and the allocated memory will be 48 * 256 = 12KB for this particular cpu. Assuming an 128-cpu system, the total memory consumption across all cpus will be 12K * 128 = 1.5MB memory.

This might be okay for non-percpu allocation, but may not be good for percpu allocation, which will consume 1.5MB * 128 = 192MB memory in the worst case if every cpu has a chance of memory allocation.

In practice, percpu allocation is very rare compared to non-percpu allocation. So let us have smaller low/high marks which can avoid unnecessary memory consumption.

---
 kernel/bpf/memalloc.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
kernel/bpf/memalloc.c +7 -1
index a8ee6fb8401c..460c8f38fed6 100644
@@ -464,11 +464,17 @@ static void notrace irq_work_raise(struct bpf_mem_cache *c)
  * consume ~ 11 Kbyte per cpu.
  * Typical case will be between 11K and 116K closer to 11K.
  * bpf progs can and should share bpf_mem_cache when possible.
+ *
+ * Percpu allocation is typically rare. To avoid potential unnecessary large
+ * memory consumption, set low_mark = 1 and high_mark = 3, resulting in c->batch = 1.
  */
 static void init_refill_work(struct bpf_mem_cache *c)
 {
 	init_irq_work(&c->refill_work, bpf_mem_refill);
-	if (c->unit_size <= 256) {
+	if (c->percpu_size) {
+		c->low_watermark = 1;
+		c->high_watermark = 3;
+	} else if (c->unit_size <= 256) {
 		c->low_watermark = 32;
 		c->high_watermark = 96;
 	} else {
Signature · 1 line
2.34.1
Acked-by on v6 5/8
#9 lore raw

Re: [PATCH bpf-next v6 5/8] bpf: Use smaller low/high marks for percpu allocation

Quoted 14 lines · Yonghong Song

Currently, refill low/high marks are set with the assumption of normal non-percpu memory allocation. For example, for an allocation size 256, for non-percpu memory allocation, low mark is 32 and high mark is 96, resulting in the batch allocation of 48 elements and the allocated memory will be 48 * 256 = 12KB for this particular cpu. Assuming an 128-cpu system, the total memory consumption across all cpus will be 12K * 128 = 1.5MB memory.

This might be okay for non-percpu allocation, but may not be good for percpu allocation, which will consume 1.5MB * 128 = 192MB memory in the worst case if every cpu has a chance of memory allocation.

Quoted 5 lines

In practice, percpu allocation is very rare compared to non-percpu allocation. So let us have smaller low/high marks which can avoid unnecessary memory consumption.

author reviewer
#10 lore raw

PATCH v6 6/8 bpf: Limit up to 512 bytes for bpf_global_percpu_ma allocation

In reply to Yonghong Song

For percpu data structure allocation with bpf_global_percpu_ma, the maximum data size is 4K. But for a system with large number of cpus, bigger data size (e.g., 2K, 4K) might consume a lot of memory. For example, the percpu memory consumption with unit size 2K and 1024 cpus will be 2K * 1K * 1k = 2GB memory.

We should discourage such usage. Let us limit the maximum data size to be 512 for bpf_global_percpu_ma allocation.

---
 kernel/bpf/verifier.c | 8 ++++++++
 1 file changed, 8 insertions(+)
kernel/bpf/verifier.c +8 -0
index 5b2bfa2cbb7b..6c02905c8700 100644
@@ -195,6 +195,8 @@ struct bpf_verifier_stack_elem {
 					  POISON_POINTER_DELTA))
 #define BPF_MAP_PTR(X)		((struct bpf_map *)((X) & ~BPF_MAP_PTR_UNPRIV))
 
+#define BPF_GLOBAL_PERCPU_MA_MAX_SIZE  512
+
 static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx);
 static int release_reference(struct bpf_verifier_env *env, int ref_obj_id);
 static void invalidate_non_owning_refs(struct bpf_verifier_env *env);
@@ -12162,6 +12164,12 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
 				}
 
 				if (meta.func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl]) {
+					if (ret_t->size > BPF_GLOBAL_PERCPU_MA_MAX_SIZE) {
+						verbose(env, "bpf_percpu_obj_new type size (%d) is greater than %d\n",
+							ret_t->size, BPF_GLOBAL_PERCPU_MA_MAX_SIZE);
+						return -EINVAL;
+					}
+
 					if (!bpf_global_percpu_ma_set) {
 						mutex_lock(&bpf_percpu_ma_lock);
 						if (!bpf_global_percpu_ma_set) {
Signature · 1 line
2.34.1
author reviewer
#11 lore raw

PATCH v6 7/8 selftests/bpf: Cope with 512 bytes limit with bpf_global_percpu_ma

In reply to Yonghong Song

In the previous patch, the maximum data size for bpf_global_percpu_ma is 512 bytes. This breaks selftest test_bpf_ma. The test is adjusted in two aspects:
- Since the maximum allowed data size for bpf_global_percpu_ma is

    512, remove all tests beyond that, names sizes 1024, 2048 and 4096.

- Previously the percpu data size is bucket_size - 8 in order to avoid percpu allocation into the next bucket. This patch removed such data size adjustment thanks to Patch 1.

Also, a better way to generate BTF type is used than adding a member to the value struct.

---
 .../selftests/bpf/prog_tests/test_bpf_ma.c    | 20 ++++--
 .../testing/selftests/bpf/progs/test_bpf_ma.c | 66 +++++++++----------
 2 files changed, 46 insertions(+), 40 deletions(-)
tools/testing/selftests/bpf/prog_tests/test_bpf_ma.c +13 -7
index d3491a84b3b9..ccae0b31ac6c 100644
@@ -14,7 +14,8 @@ static void do_bpf_ma_test(const char *name)
 	struct test_bpf_ma *skel;
 	struct bpf_program *prog;
 	struct btf *btf;
-	int i, err;
+	int i, err, id;
+	char tname[32];
 
 	skel = test_bpf_ma__open();
 	if (!ASSERT_OK_PTR(skel, "open"))
@@ -25,16 +26,21 @@ static void do_bpf_ma_test(const char *name)
 		goto out;
 
 	for (i = 0; i < ARRAY_SIZE(skel->rodata->data_sizes); i++) {
-		char name[32];
-		int id;
-
-		snprintf(name, sizeof(name), "bin_data_%u", skel->rodata->data_sizes[i]);
-		id = btf__find_by_name_kind(btf, name, BTF_KIND_STRUCT);
-		if (!ASSERT_GT(id, 0, "bin_data"))
+		snprintf(tname, sizeof(tname), "bin_data_%u", skel->rodata->data_sizes[i]);
+		id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
+		if (!ASSERT_GT(id, 0, tname))
 			goto out;
 		skel->rodata->data_btf_ids[i] = id;
 	}
 
+	for (i = 0; i < ARRAY_SIZE(skel->rodata->percpu_data_sizes); i++) {
+		snprintf(tname, sizeof(tname), "percpu_bin_data_%u", skel->rodata->percpu_data_sizes[i]);
+		id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
+		if (!ASSERT_GT(id, 0, tname))
+			goto out;
+		skel->rodata->percpu_data_btf_ids[i] = id;
+	}
+
 	prog = bpf_object__find_program_by_name(skel->obj, name);
 	if (!ASSERT_OK_PTR(prog, "invalid prog name"))
 		goto out;
tools/testing/selftests/bpf/progs/test_bpf_ma.c +33 -33

120 lines not on this page · raw message

Signature · 1 line
2.34.1
author reviewer
#12 lore raw

PATCH v6 8/8 selftests/bpf: Add a selftest with > 512-byte percpu allocation size

In reply to Yonghong Song

Add a selftest to capture the verification failure when the allocation size is greater than 512.

---
 .../selftests/bpf/progs/percpu_alloc_fail.c    | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)
tools/testing/selftests/bpf/progs/percpu_alloc_fail.c +18 -0
index 1a891d30f1fe..f2b8eb2ff76f 100644
@@ -17,6 +17,10 @@ struct val_with_rb_root_t {
 	struct bpf_spin_lock lock;
 };
 
+struct val_600b_t {
+	char b[600];
+};
+
 struct elem {
 	long sum;
 	struct val_t __percpu_kptr *pc;
@@ -161,4 +165,18 @@ int BPF_PROG(test_array_map_7)
 	return 0;
 }
 
+SEC("?fentry.s/bpf_fentry_test1")
+__failure __msg("bpf_percpu_obj_new type size (600) is greater than 512")
+int BPF_PROG(test_array_map_8)
+{
+	struct val_600b_t __percpu_kptr *p;
+
+	p = bpf_percpu_obj_new(struct val_600b_t);
+	if (!p)
+		return 0;
+
+	bpf_percpu_obj_drop(p);
+	return 0;
+}
+
 char _license[] SEC("license") = "GPL";
Signature · 1 line
2.34.1
#13 lore raw
In reply to Yonghong Song

Hi,

Quoted 7 lines · Yonghong Song

Currently when a bpf program intends to allocate memory for percpu kptr, the verifier will call bpf_mem_alloc_init() to prefill all supported unit sizes and this caused memory consumption very big for large number of cpus. For example, for 128-cpu system, the total memory consumption with initial prefill is ~175MB. Things will become worse for systems with even more cpus.

Quoted 14 lines

Patch 1 avoids unnecessary extra percpu memory allocation. Patch 2 adds objcg to bpf_mem_alloc at init stage so objcg can be associated with root cgroup and objcg can be passed to later bpf_mem_alloc_percpu_unit_init().
Patch 3 addresses memory consumption issue by avoiding to prefill with all unit sizes, i.e. only prefilling with user specified size. Patch 4 further reduces memory consumption by limiting the number of prefill entries for percpu memory allocation. Patch 5 has much smaller low/high watermarks for percpu allocation to reduce memory consumption.
Patch 6 rejects percpu memory allocation with bpf_global_percpu_ma when allocation size is greater than 512 bytes.
Patch 7 fixed test_bpf_ma test due to Patch 5.
Patch 8 added one test to show the verification failure log message.

FYI. After applying the patch set, the memory consumption in bpf memory benchmark [1] on 8-CPU VM decreases a lot:

Before the patch set:

$ for i in 1 4 8; do ./bench -w3 -d10 bpf_ma -p${i} -a --percpu; done |
grep Summary
Summary: per-prod alloc   14.16 ± 0.59M/s free   36.18 ± 0.39M/s, total
memory usage  183.71 ± 10.38MiB
Summary: per-prod alloc   12.35 ± 1.10M/s free   35.79 ± 0.51M/s, total
memory usage  744.52 ± 11.64MiB
Summary: per-prod alloc   11.15 ± 0.20M/s free   35.72 ± 0.27M/s, total
memory usage 2545.98 ± 537.57MiB

After the patch set:

$ for i in 1 4 8; do ./bench -w3 -d10 bpf_ma -p${i} -a --percpu; done |
grep Summary
Summary: per-prod alloc    0.86 ± 0.00M/s free   37.29 ± 0.11M/s, total
memory usage    0.00 ± 0.00MiB
Summary: per-prod alloc    0.85 ± 0.00M/s free   36.70 ± 0.24M/s, total
memory usage    0.00 ± 0.00MiB
Summary: per-prod alloc    0.84 ± 0.00M/s free   37.21 ± 0.17M/s, total
memory usage    0.00 ± 0.00MiB

However the allocation performance also degrades a lot. It seems it is due to patch 5 (bpf: Use smaller low/high marks for percpu allocation), because c->batch is 1 now, so each allocation needs one run of irq_work.

[1]:
https://lore.kernel.org/bpf/[email protected]/

Quoted 43 lines

Changelogs:
v5 -> v6:
. Change bpf_mem_alloc_percpu_init() to add objcg as one of parameters.
For bpf_global_percpu_ma, the objcg is NULL, corresponding root memcg.
v4 -> v5:
. Do not do bpf_global_percpu_ma initialization at init stage, instead
doing initialization when the verifier knows it is going to be used by bpf prog.
. Using much smaller low/high watermarks for percpu allocation.
v3 -> v4:
. Add objcg to bpf_mem_alloc during init stage.
. Initialize objcg at init stage but use it in bpf_mem_alloc_percpu_unit_init(). . Remove check_obj_size() in bpf_mem_alloc_percpu_unit_init().
v2 -> v3:
. Clear the bpf_mem_cache if prefill fails.
. Change test_bpf_ma percpu allocation tests to use bucket_size
as allocation size instead of bucket_size - 8.
. Remove __GFP_ZERO flag from __alloc_percpu_gfp() call.
v1 -> v2:
. Avoid unnecessary extra percpu memory allocation.
. Add a separate function to do bpf_global_percpu_ma initialization . promote.
. Promote function static 'sizes' array to file static.
. Add comments to explain to refill only one item for percpu alloc.

Yonghong Song (8):
bpf: Avoid unnecessary extra percpu memory allocation
bpf: Add objcg to bpf_mem_alloc
bpf: Allow per unit prefill for non-fix-size percpu memory allocator
bpf: Refill only one percpu element in memalloc
bpf: Use smaller low/high marks for percpu allocation
bpf: Limit up to 512 bytes for bpf_global_percpu_ma allocation selftests/bpf: Cope with 512 bytes limit with bpf_global_percpu_ma selftests/bpf: Add a selftest with > 512-byte percpu allocation size

 include/linux/bpf_mem_alloc.h                 |  8 ++
 kernel/bpf/memalloc.c                         | 93 ++++++++++++++++---
 kernel/bpf/verifier.c                         | 45 ++++++---
 .../selftests/bpf/prog_tests/test_bpf_ma.c    | 20 ++--
 .../selftests/bpf/progs/percpu_alloc_fail.c   | 18 ++++
 .../testing/selftests/bpf/progs/test_bpf_ma.c | 66 ++++++-------
 6 files changed, 184 insertions(+), 66 deletions(-)
author reviewer
#14 lore raw
Quoted 49 lines · Hou Tao

Hi,

Quoted 7 lines · Yonghong Song

Currently when a bpf program intends to allocate memory for percpu kptr, the verifier will call bpf_mem_alloc_init() to prefill all supported unit sizes and this caused memory consumption very big for large number of cpus. For example, for 128-cpu system, the total memory consumption with initial prefill is ~175MB. Things will become worse for systems with even more cpus.

Quoted 14 lines

Patch 1 avoids unnecessary extra percpu memory allocation. Patch 2 adds objcg to bpf_mem_alloc at init stage so objcg can be associated with root cgroup and objcg can be passed to later bpf_mem_alloc_percpu_unit_init().
Patch 3 addresses memory consumption issue by avoiding to prefill with all unit sizes, i.e. only prefilling with user specified size. Patch 4 further reduces memory consumption by limiting the number of prefill entries for percpu memory allocation. Patch 5 has much smaller low/high watermarks for percpu allocation to reduce memory consumption.
Patch 6 rejects percpu memory allocation with bpf_global_percpu_ma when allocation size is greater than 512 bytes.
Patch 7 fixed test_bpf_ma test due to Patch 5.
Patch 8 added one test to show the verification failure log message.

FYI. After applying the patch set, the memory consumption in bpf memory benchmark [1] on 8-CPU VM decreases a lot:

Before the patch set:

$ for i in 1 4 8; do ./bench -w3 -d10 bpf_ma -p${i} -a --percpu; done |
grep Summary
Summary: per-prod alloc   14.16 ± 0.59M/s free   36.18 ± 0.39M/s, total
memory usage  183.71 ± 10.38MiB
Summary: per-prod alloc   12.35 ± 1.10M/s free   35.79 ± 0.51M/s, total
memory usage  744.52 ± 11.64MiB
Summary: per-prod alloc   11.15 ± 0.20M/s free   35.72 ± 0.27M/s, total
memory usage 2545.98 ± 537.57MiB

After the patch set:

$ for i in 1 4 8; do ./bench -w3 -d10 bpf_ma -p${i} -a --percpu; done |
grep Summary
Summary: per-prod alloc    0.86 ± 0.00M/s free   37.29 ± 0.11M/s, total
memory usage    0.00 ± 0.00MiB
Summary: per-prod alloc    0.85 ± 0.00M/s free   36.70 ± 0.24M/s, total
memory usage    0.00 ± 0.00MiB
Summary: per-prod alloc    0.84 ± 0.00M/s free   37.21 ± 0.17M/s, total
memory usage    0.00 ± 0.00MiB
Quoted 3 lines

However the allocation performance also degrades a lot. It seems it is due to patch 5 (bpf: Use smaller low/high marks for percpu allocation), because c->batch is 1 now, so each allocation needs one run of irq_work.

Thanks for benchmarking! With low watermark to be 1 and c->batch to 1 as well, there will be more overhead due to irq_work. In practice, I expect we should not see a lot of such percpu map element updates or percpu kptr allocaitons.

Quoted 46 lines

[1]:
https://lore.kernel.org/bpf/[email protected]/

Quoted 43 lines

Changelogs:
v5 -> v6:
. Change bpf_mem_alloc_percpu_init() to add objcg as one of parameters.
For bpf_global_percpu_ma, the objcg is NULL, corresponding root memcg.
v4 -> v5:
. Do not do bpf_global_percpu_ma initialization at init stage, instead
doing initialization when the verifier knows it is going to be used by bpf prog.
. Using much smaller low/high watermarks for percpu allocation.
v3 -> v4:
. Add objcg to bpf_mem_alloc during init stage.
. Initialize objcg at init stage but use it in bpf_mem_alloc_percpu_unit_init(). . Remove check_obj_size() in bpf_mem_alloc_percpu_unit_init().
v2 -> v3:
. Clear the bpf_mem_cache if prefill fails.
. Change test_bpf_ma percpu allocation tests to use bucket_size
as allocation size instead of bucket_size - 8.
. Remove __GFP_ZERO flag from __alloc_percpu_gfp() call.
v1 -> v2:
. Avoid unnecessary extra percpu memory allocation.
. Add a separate function to do bpf_global_percpu_ma initialization . promote.
. Promote function static 'sizes' array to file static.
. Add comments to explain to refill only one item for percpu alloc.

Yonghong Song (8):
bpf: Avoid unnecessary extra percpu memory allocation
bpf: Add objcg to bpf_mem_alloc
bpf: Allow per unit prefill for non-fix-size percpu memory allocator
bpf: Refill only one percpu element in memalloc
bpf: Use smaller low/high marks for percpu allocation
bpf: Limit up to 512 bytes for bpf_global_percpu_ma allocation selftests/bpf: Cope with 512 bytes limit with bpf_global_percpu_ma selftests/bpf: Add a selftest with > 512-byte percpu allocation size

  include/linux/bpf_mem_alloc.h                 |  8 ++
  kernel/bpf/memalloc.c                         | 93 ++++++++++++++++---
  kernel/bpf/verifier.c                         | 45 ++++++---
  .../selftests/bpf/prog_tests/test_bpf_ma.c    | 20 ++--
  .../selftests/bpf/progs/percpu_alloc_fail.c   | 18 ++++
  .../testing/selftests/bpf/progs/test_bpf_ma.c | 66 ++++++-------
  6 files changed, 184 insertions(+), 66 deletions(-)
#15 lore raw
In reply to Yonghong Song
This series was applied to bpf/bpf-next.git (master) by Alexei Starovoitov <[email protected]>:

Hello:

This series was applied to bpf/bpf-next.git (master) by Alexei Starovoitov <[email protected]>:

Quoted 8 lines · On Thu, 21 Dec 2023 19:17:29 -0800 you wrote:

Currently when a bpf program intends to allocate memory for percpu kptr, the verifier will call bpf_mem_alloc_init() to prefill all supported unit sizes and this caused memory consumption very big for large number of cpus. For example, for 128-cpu system, the total memory consumption with initial prefill is ~175MB. Things will become worse for systems with even more cpus.

[...]

Here is the summary with links:
- [bpf-next,v6,1/8] bpf: Avoid unnecessary extra percpu memory allocation

    https://git.kernel.org/bpf/bpf-next/c/9beda16c257d
  - [bpf-next,v6,2/8] bpf: Add objcg to bpf_mem_alloc
    https://git.kernel.org/bpf/bpf-next/c/9fc8e802048a
  - [bpf-next,v6,3/8] bpf: Allow per unit prefill for non-fix-size percpu memory allocator
    https://git.kernel.org/bpf/bpf-next/c/c39aa3b289e9
  - [bpf-next,v6,4/8] bpf: Refill only one percpu element in memalloc
    https://git.kernel.org/bpf/bpf-next/c/5b95e638f134
  - [bpf-next,v6,5/8] bpf: Use smaller low/high marks for percpu allocation
    https://git.kernel.org/bpf/bpf-next/c/0e2ba9f96f9b
  - [bpf-next,v6,6/8] bpf: Limit up to 512 bytes for bpf_global_percpu_ma allocation
    https://git.kernel.org/bpf/bpf-next/c/5c1a37653260
  - [bpf-next,v6,7/8] selftests/bpf: Cope with 512 bytes limit with bpf_global_percpu_ma
    https://git.kernel.org/bpf/bpf-next/c/21f5a801c171
  - [bpf-next,v6,8/8] selftests/bpf: Add a selftest with > 512-byte percpu allocation size
    https://git.kernel.org/bpf/bpf-next/c/adc8c4549d9e

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html