[PATCH bpf-next v8 0/7] bpf: Introduce BPF_F_CPU and BPF_F_ALL_CPUS flags for percpu maps

[PATCH bpf-next v8 0/7] bpf: Introduce BPF_F_CPU and BPF_F_ALL_CPUS flags for percpu maps

[Leon Hwang]

This patch set introduces the BPF_F_CPU and BPF_F_ALL_CPUS flags for
percpu maps, as the requirement of BPF_F_ALL_CPUS flag for percpu_array
maps was discussed in the thread of
"[PATCH bpf-next v3 0/4] bpf: Introduce global percpu data"[1].

The goal of BPF_F_ALL_CPUS flag is to reduce data caching overhead in light
skeletons by allowing a single value to be reused to update values across all
CPUs. This avoids the M:N problem where M cached values are used to update a
map on N CPUs kernel.

The BPF_F_CPU flag is accompanied by *flags*-embedded cpu info, which
specifies the target CPU for the operation:

* For lookup operations: the flag field alongside cpu info enable querying
  a value on the specified CPU.
* For update operations: the flag field alongside cpu info enable
  updating value for specified CPU.

Links:
[1] https://lore.kernel.org/bpf/20250526162146.24429-1-leon.hwang@linux.dev/

Changes:
v7 -> v8:
* Address comments from Andrii:
  * Check BPF_F_LOCK when update percpu_array, percpu_hash and
    lru_percpu_hash maps.
  * Refactor flags check in __htab_map_lookup_and_delete_batch().
  * Keep value_size unaligned and copy value using copy_map_value() in
    __htab_map_lookup_and_delete_batch() when BPF_F_CPU is specified.
  * Update warn message in libbpf's validate_map_op().
  * Update comment of libbpf's bpf_map__lookup_elem().

v6 -> v7:
* Get correct value size for percpu_hash and lru_percpu_hash in
  update_batch API.
* Set 'count' as 'max_entries' in test cases for lookup_batch API.
* Address comment from Alexei:
  * Move cpu flags check into bpf_map_check_op_flags().

v5 -> v6:
* Move bpf_map_check_op_flags() from 'bpf.h' to 'syscall.c'.
* Address comments from Alexei:
  * Drop the refactoring code of data copying logic for percpu maps.
  * Drop bpf_map_check_op_flags() wrappers.

v4 -> v5:
* Address comments from Andrii:
  * Refactor data copying logic for all percpu maps.
  * Drop this_cpu_ptr() micro-optimization.
  * Drop cpu check in libbpf's validate_map_op().
  * Enhance bpf_map_check_op_flags() using *allowed flags* instead of
    'extra_flags_mask'.

v3 -> v4:
* Address comments from Andrii:
  * Remove unnecessary map_type check in bpf_map_value_size().
  * Reduce code churn.
  * Remove unnecessary do_delete check in
    __htab_map_lookup_and_delete_batch().
  * Introduce bpf_percpu_copy_to_user() and bpf_percpu_copy_from_user().
  * Rename check_map_flags() to bpf_map_check_op_flags() with
    extra_flags_mask.
  * Add human-readable pr_warn() explanations in validate_map_op().
  * Use flags in bpf_map__delete_elem() and
    bpf_map__lookup_and_delete_elem().
  * Drop "for alignment reasons".
v3 link: https://lore.kernel.org/bpf/20250821160817.70285-1-leon.hwang@linux.dev/

v2 -> v3:
* Address comments from Alexei:
  * Use BPF_F_ALL_CPUS instead of BPF_ALL_CPUS magic.
  * Introduce these two cpu flags for all percpu maps.
* Address comments from Jiri:
  * Reduce some unnecessary u32 cast.
  * Refactor more generic map flags check function.
  * A code style issue.
v2 link: https://lore.kernel.org/bpf/20250805163017.17015-1-leon.hwang@linux.dev/

v1 -> v2:
* Address comments from Andrii:
  * Embed cpu info as high 32 bits of *flags* totally.
  * Use ERANGE instead of E2BIG.
  * Few format issues.

Leon Hwang (7):
  bpf: Introduce internal bpf_map_check_op_flags helper function
  bpf: Introduce BPF_F_CPU and BPF_F_ALL_CPUS flags
  bpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu_array
    maps
  bpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu_hash
    and lru_percpu_hash maps
  bpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for
    percpu_cgroup_storage maps
  libbpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu maps
  selftests/bpf: Add cases to test BPF_F_CPU and BPF_F_ALL_CPUS flags

 include/linux/bpf-cgroup.h                    |   4 +-
 include/linux/bpf.h                           |  44 +++-
 include/uapi/linux/bpf.h                      |   2 +
 kernel/bpf/arraymap.c                         |  24 +-
 kernel/bpf/hashtab.c                          |  79 ++++--
 kernel/bpf/local_storage.c                    |  22 +-
 kernel/bpf/syscall.c                          |  65 +++--
 tools/include/uapi/linux/bpf.h                |   2 +
 tools/lib/bpf/bpf.h                           |   8 +
 tools/lib/bpf/libbpf.c                        |  26 +-
 tools/lib/bpf/libbpf.h                        |  21 +-
 .../selftests/bpf/prog_tests/percpu_alloc.c   | 234 ++++++++++++++++++
 .../selftests/bpf/progs/percpu_alloc_array.c  |  32 +++
 13 files changed, 473 insertions(+), 90 deletions(-)

--
2.50.1

[PATCH bpf-next v8 1/7] bpf: Introduce internal bpf_map_check_op_flags helper function

[Leon Hwang]

It is to unify map flags checking for lookup_elem, update_elem,
lookup_batch and update_batch APIs.

Therefore, it will be convenient to check BPF_F_CPU and BPF_F_ALL_CPUS
flags in it for these APIs in next patch.

Acked-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
---
 include/linux/bpf.h  | 11 +++++++++++
 kernel/bpf/syscall.c | 34 +++++++++++-----------------------
 2 files changed, 22 insertions(+), 23 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index ea2ed6771cc60..536a640246fec 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -3768,4 +3768,15 @@ int bpf_prog_get_file_line(struct bpf_prog *prog, unsigned long ip, const char *
 			   const char **linep, int *nump);
 struct bpf_prog *bpf_prog_find_from_stack(void);
 
+static inline int bpf_map_check_op_flags(struct bpf_map *map, u64 flags, u64 allowed_flags)
+{
+	if (flags & ~allowed_flags)
+		return -EINVAL;
+
+	if ((flags & BPF_F_LOCK) && !btf_record_has_field(map->record, BPF_SPIN_LOCK))
+		return -EINVAL;
+
+	return 0;
+}
+
 #endif /* _LINUX_BPF_H */
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index adb05d235011f..84261a0211c51 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -1708,9 +1708,6 @@ static int map_lookup_elem(union bpf_attr *attr)
 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 		return -EINVAL;
 
-	if (attr->flags & ~BPF_F_LOCK)
-		return -EINVAL;
-
 	CLASS(fd, f)(attr->map_fd);
 	map = __bpf_map_get(f);
 	if (IS_ERR(map))
@@ -1718,9 +1715,9 @@ static int map_lookup_elem(union bpf_attr *attr)
 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
 		return -EPERM;
 
-	if ((attr->flags & BPF_F_LOCK) &&
-	    !btf_record_has_field(map->record, BPF_SPIN_LOCK))
-		return -EINVAL;
+	err = bpf_map_check_op_flags(map, attr->flags, BPF_F_LOCK);
+	if (err)
+		return err;
 
 	key = __bpf_copy_key(ukey, map->key_size);
 	if (IS_ERR(key))
@@ -1783,11 +1780,9 @@ static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
 		goto err_put;
 	}
 
-	if ((attr->flags & BPF_F_LOCK) &&
-	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
-		err = -EINVAL;
+	err = bpf_map_check_op_flags(map, attr->flags, ~0);
+	if (err)
 		goto err_put;
-	}
 
 	key = ___bpf_copy_key(ukey, map->key_size);
 	if (IS_ERR(key)) {
@@ -1991,13 +1986,9 @@ int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
 	void *key, *value;
 	int err = 0;
 
-	if (attr->batch.elem_flags & ~BPF_F_LOCK)
-		return -EINVAL;
-
-	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
-	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
-		return -EINVAL;
-	}
+	err = bpf_map_check_op_flags(map, attr->batch.elem_flags, BPF_F_LOCK);
+	if (err)
+		return err;
 
 	value_size = bpf_map_value_size(map);
 
@@ -2054,12 +2045,9 @@ int generic_map_lookup_batch(struct bpf_map *map,
 	u32 value_size, cp, max_count;
 	int err;
 
-	if (attr->batch.elem_flags & ~BPF_F_LOCK)
-		return -EINVAL;
-
-	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
-	    !btf_record_has_field(map->record, BPF_SPIN_LOCK))
-		return -EINVAL;
+	err = bpf_map_check_op_flags(map, attr->batch.elem_flags, BPF_F_LOCK);
+	if (err)
+		return err;
 
 	value_size = bpf_map_value_size(map);
 
-- 
2.50.1

[PATCH bpf-next v8 2/7] bpf: Introduce BPF_F_CPU and BPF_F_ALL_CPUS flags

[Leon Hwang]

Introduce BPF_F_CPU and BPF_F_ALL_CPUS flags and check them for
following APIs:

* 'map_lookup_elem()'
* 'map_update_elem()'
* 'generic_map_lookup_batch()'
* 'generic_map_update_batch()'

And, get the correct value size for these APIs.

Acked-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
---
 include/linux/bpf.h            | 23 ++++++++++++++++++++++-
 include/uapi/linux/bpf.h       |  2 ++
 kernel/bpf/syscall.c           | 31 +++++++++++++++++--------------
 tools/include/uapi/linux/bpf.h |  2 ++
 4 files changed, 43 insertions(+), 15 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 536a640246fec..80719765d40e6 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -3768,14 +3768,35 @@ int bpf_prog_get_file_line(struct bpf_prog *prog, unsigned long ip, const char *
 			   const char **linep, int *nump);
 struct bpf_prog *bpf_prog_find_from_stack(void);
 
+static inline bool bpf_map_supports_cpu_flags(enum bpf_map_type map_type)
+{
+	return false;
+}
+
 static inline int bpf_map_check_op_flags(struct bpf_map *map, u64 flags, u64 allowed_flags)
 {
-	if (flags & ~allowed_flags)
+	u32 cpu;
+
+	if ((u32)flags & ~allowed_flags)
 		return -EINVAL;
 
 	if ((flags & BPF_F_LOCK) && !btf_record_has_field(map->record, BPF_SPIN_LOCK))
 		return -EINVAL;
 
+	if (!(flags & BPF_F_CPU) && flags >> 32)
+		return -EINVAL;
+
+	if (flags & (BPF_F_CPU | BPF_F_ALL_CPUS)) {
+		if (!bpf_map_supports_cpu_flags(map->map_type))
+			return -EINVAL;
+		if ((flags & BPF_F_CPU) && (flags & BPF_F_ALL_CPUS))
+			return -EINVAL;
+
+		cpu = flags >> 32;
+		if ((flags & BPF_F_CPU) && cpu >= num_possible_cpus())
+			return -ERANGE;
+	}
+
 	return 0;
 }
 
diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index ae83d8649ef1c..f827353424ec5 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -1372,6 +1372,8 @@ enum {
 	BPF_NOEXIST	= 1, /* create new element if it didn't exist */
 	BPF_EXIST	= 2, /* update existing element */
 	BPF_F_LOCK	= 4, /* spin_lock-ed map_lookup/map_update */
+	BPF_F_CPU	= 8, /* cpu flag for percpu maps, upper 32-bit of flags is a cpu number */
+	BPF_F_ALL_CPUS	= 16, /* update value across all CPUs for percpu maps */
 };
 
 /* flags for BPF_MAP_CREATE command */
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index 84261a0211c51..d1bbc309b62ff 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -133,12 +133,14 @@ bool bpf_map_write_active(const struct bpf_map *map)
 	return atomic64_read(&map->writecnt) != 0;
 }
 
-static u32 bpf_map_value_size(const struct bpf_map *map)
-{
-	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
-	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
-	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
-	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
+static u32 bpf_map_value_size(const struct bpf_map *map, u64 flags)
+{
+	if (flags & (BPF_F_CPU | BPF_F_ALL_CPUS))
+		return round_up(map->value_size, 8);
+	else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
+		 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
+		 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
+		 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 		return round_up(map->value_size, 8) * num_possible_cpus();
 	else if (IS_FD_MAP(map))
 		return sizeof(u32);
@@ -1715,7 +1717,7 @@ static int map_lookup_elem(union bpf_attr *attr)
 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
 		return -EPERM;
 
-	err = bpf_map_check_op_flags(map, attr->flags, BPF_F_LOCK);
+	err = bpf_map_check_op_flags(map, attr->flags, BPF_F_LOCK | BPF_F_CPU);
 	if (err)
 		return err;
 
@@ -1723,7 +1725,7 @@ static int map_lookup_elem(union bpf_attr *attr)
 	if (IS_ERR(key))
 		return PTR_ERR(key);
 
-	value_size = bpf_map_value_size(map);
+	value_size = bpf_map_value_size(map, attr->flags);
 
 	err = -ENOMEM;
 	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
@@ -1790,7 +1792,7 @@ static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
 		goto err_put;
 	}
 
-	value_size = bpf_map_value_size(map);
+	value_size = bpf_map_value_size(map, attr->flags);
 	value = kvmemdup_bpfptr(uvalue, value_size);
 	if (IS_ERR(value)) {
 		err = PTR_ERR(value);
@@ -1986,11 +1988,12 @@ int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
 	void *key, *value;
 	int err = 0;
 
-	err = bpf_map_check_op_flags(map, attr->batch.elem_flags, BPF_F_LOCK);
+	err = bpf_map_check_op_flags(map, attr->batch.elem_flags,
+				     BPF_F_LOCK | BPF_F_CPU | BPF_F_ALL_CPUS);
 	if (err)
 		return err;
 
-	value_size = bpf_map_value_size(map);
+	value_size = bpf_map_value_size(map, attr->batch.elem_flags);
 
 	max_count = attr->batch.count;
 	if (!max_count)
@@ -2045,11 +2048,11 @@ int generic_map_lookup_batch(struct bpf_map *map,
 	u32 value_size, cp, max_count;
 	int err;
 
-	err = bpf_map_check_op_flags(map, attr->batch.elem_flags, BPF_F_LOCK);
+	err = bpf_map_check_op_flags(map, attr->batch.elem_flags, BPF_F_LOCK | BPF_F_CPU);
 	if (err)
 		return err;
 
-	value_size = bpf_map_value_size(map);
+	value_size = bpf_map_value_size(map, attr->batch.elem_flags);
 
 	max_count = attr->batch.count;
 	if (!max_count)
@@ -2171,7 +2174,7 @@ static int map_lookup_and_delete_elem(union bpf_attr *attr)
 		goto err_put;
 	}
 
-	value_size = bpf_map_value_size(map);
+	value_size = bpf_map_value_size(map, 0);
 
 	err = -ENOMEM;
 	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index ae83d8649ef1c..f827353424ec5 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h
@@ -1372,6 +1372,8 @@ enum {
 	BPF_NOEXIST	= 1, /* create new element if it didn't exist */
 	BPF_EXIST	= 2, /* update existing element */
 	BPF_F_LOCK	= 4, /* spin_lock-ed map_lookup/map_update */
+	BPF_F_CPU	= 8, /* cpu flag for percpu maps, upper 32-bit of flags is a cpu number */
+	BPF_F_ALL_CPUS	= 16, /* update value across all CPUs for percpu maps */
 };
 
 /* flags for BPF_MAP_CREATE command */
-- 
2.50.1

[PATCH bpf-next v8 3/7] bpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu_array maps

[Leon Hwang]

Introduce support for the BPF_F_ALL_CPUS flag in percpu_array maps to
allow updating values for all CPUs with a single value for both
update_elem and update_batch APIs.

Introduce support for the BPF_F_CPU flag in percpu_array maps to allow:

* update value for specified CPU for both update_elem and update_batch
APIs.
* lookup value for specified CPU for both lookup_elem and lookup_batch
APIs.

The BPF_F_CPU flag is passed via:

* map_flags of lookup_elem and update_elem APIs along with embedded cpu
info.
* elem_flags of lookup_batch and update_batch APIs along with embedded
cpu info.

Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
---
 include/linux/bpf.h   |  9 +++++++--
 kernel/bpf/arraymap.c | 24 +++++++++++++++++++++---
 kernel/bpf/syscall.c  |  2 +-
 3 files changed, 29 insertions(+), 6 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 80719765d40e6..9c4101f537c86 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -2718,7 +2718,7 @@ int map_set_for_each_callback_args(struct bpf_verifier_env *env,
 				   struct bpf_func_state *callee);
 
 int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value);
-int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value);
+int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value, u64 flags);
 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
 			   u64 flags);
 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
@@ -3770,7 +3770,12 @@ struct bpf_prog *bpf_prog_find_from_stack(void);
 
 static inline bool bpf_map_supports_cpu_flags(enum bpf_map_type map_type)
 {
-	return false;
+	switch (map_type) {
+	case BPF_MAP_TYPE_PERCPU_ARRAY:
+		return true;
+	default:
+		return false;
+	}
 }
 
 static inline int bpf_map_check_op_flags(struct bpf_map *map, u64 flags, u64 allowed_flags)
diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c
index 80b1765a31596..a8bf1f99ad548 100644
--- a/kernel/bpf/arraymap.c
+++ b/kernel/bpf/arraymap.c
@@ -307,7 +307,7 @@ static void *percpu_array_map_lookup_percpu_elem(struct bpf_map *map, void *key,
 	return per_cpu_ptr(array->pptrs[index & array->index_mask], cpu);
 }
 
-int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
+int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value, u64 map_flags)
 {
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	u32 index = *(u32 *)key;
@@ -325,11 +325,18 @@ int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
 	size = array->elem_size;
 	rcu_read_lock();
 	pptr = array->pptrs[index & array->index_mask];
+	if (map_flags & BPF_F_CPU) {
+		cpu = map_flags >> 32;
+		copy_map_value_long(map, value, per_cpu_ptr(pptr, cpu));
+		check_and_init_map_value(map, value);
+		goto unlock;
+	}
 	for_each_possible_cpu(cpu) {
 		copy_map_value_long(map, value + off, per_cpu_ptr(pptr, cpu));
 		check_and_init_map_value(map, value + off);
 		off += size;
 	}
+unlock:
 	rcu_read_unlock();
 	return 0;
 }
@@ -402,7 +409,7 @@ int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
 	int cpu, off = 0;
 	u32 size;
 
-	if (unlikely(map_flags > BPF_EXIST))
+	if (unlikely((map_flags & BPF_F_LOCK) || (u32)map_flags > BPF_F_ALL_CPUS))
 		/* unknown flags */
 		return -EINVAL;
 
@@ -423,11 +430,22 @@ int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
 	size = array->elem_size;
 	rcu_read_lock();
 	pptr = array->pptrs[index & array->index_mask];
+	if (map_flags & BPF_F_CPU) {
+		cpu = map_flags >> 32;
+		copy_map_value_long(map, per_cpu_ptr(pptr, cpu), value);
+		bpf_obj_free_fields(array->map.record, per_cpu_ptr(pptr, cpu));
+		goto unlock;
+	}
 	for_each_possible_cpu(cpu) {
 		copy_map_value_long(map, per_cpu_ptr(pptr, cpu), value + off);
 		bpf_obj_free_fields(array->map.record, per_cpu_ptr(pptr, cpu));
-		off += size;
+		/* same user-provided value is used if BPF_F_ALL_CPUS is
+		 * specified, otherwise value is an array of per-CPU values.
+		 */
+		if (!(map_flags & BPF_F_ALL_CPUS))
+			off += size;
 	}
+unlock:
 	rcu_read_unlock();
 	return 0;
 }
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index d1bbc309b62ff..91868fded80e2 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -318,7 +318,7 @@ static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 		err = bpf_percpu_hash_copy(map, key, value);
 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
-		err = bpf_percpu_array_copy(map, key, value);
+		err = bpf_percpu_array_copy(map, key, value, flags);
 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
-- 
2.50.1

[PATCH bpf-next v8 4/7] bpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu_hash and lru_percpu_hash maps

[Leon Hwang]

Introduce BPF_F_ALL_CPUS flag support for percpu_hash and lru_percpu_hash
maps to allow updating values for all CPUs with a single value for both
update_elem and update_batch APIs.

Introduce BPF_F_CPU flag support for percpu_hash and lru_percpu_hash
maps to allow:

* update value for specified CPU for both update_elem and update_batch
APIs.
* lookup value for specified CPU for both lookup_elem and lookup_batch
APIs.

The BPF_F_CPU flag is passed via:

* map_flags along with embedded cpu info.
* elem_flags along with embedded cpu info.

Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
---
 include/linux/bpf.h  |  4 ++-
 kernel/bpf/hashtab.c | 79 ++++++++++++++++++++++++++++++--------------
 kernel/bpf/syscall.c |  2 +-
 3 files changed, 59 insertions(+), 26 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 9c4101f537c86..04041dd35d42f 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -2717,7 +2717,7 @@ int map_set_for_each_callback_args(struct bpf_verifier_env *env,
 				   struct bpf_func_state *caller,
 				   struct bpf_func_state *callee);
 
-int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value);
+int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value, u64 flags);
 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value, u64 flags);
 int bpf_percpu_hash_update(struct bpf_map *map, void *key, void *value,
 			   u64 flags);
@@ -3772,6 +3772,8 @@ static inline bool bpf_map_supports_cpu_flags(enum bpf_map_type map_type)
 {
 	switch (map_type) {
 	case BPF_MAP_TYPE_PERCPU_ARRAY:
+	case BPF_MAP_TYPE_PERCPU_HASH:
+	case BPF_MAP_TYPE_LRU_PERCPU_HASH:
 		return true;
 	default:
 		return false;
diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
index c2fcd0cd51e51..38aa8745cfbee 100644
--- a/kernel/bpf/hashtab.c
+++ b/kernel/bpf/hashtab.c
@@ -945,7 +945,7 @@ static void free_htab_elem(struct bpf_htab *htab, struct htab_elem *l)
 }
 
 static void pcpu_copy_value(struct bpf_htab *htab, void __percpu *pptr,
-			    void *value, bool onallcpus)
+			    void *value, bool onallcpus, u64 map_flags)
 {
 	if (!onallcpus) {
 		/* copy true value_size bytes */
@@ -954,15 +954,26 @@ static void pcpu_copy_value(struct bpf_htab *htab, void __percpu *pptr,
 		u32 size = round_up(htab->map.value_size, 8);
 		int off = 0, cpu;
 
+		if (map_flags & BPF_F_CPU) {
+			cpu = map_flags >> 32;
+			copy_map_value_long(&htab->map, per_cpu_ptr(pptr, cpu), value);
+			return;
+		}
+
 		for_each_possible_cpu(cpu) {
 			copy_map_value_long(&htab->map, per_cpu_ptr(pptr, cpu), value + off);
-			off += size;
+			/* same user-provided value is used if BPF_F_ALL_CPUS
+			 * is specified, otherwise value is an array of per-CPU
+			 * values.
+			 */
+			if (!(map_flags & BPF_F_ALL_CPUS))
+				off += size;
 		}
 	}
 }
 
 static void pcpu_init_value(struct bpf_htab *htab, void __percpu *pptr,
-			    void *value, bool onallcpus)
+			    void *value, bool onallcpus, u64 map_flags)
 {
 	/* When not setting the initial value on all cpus, zero-fill element
 	 * values for other cpus. Otherwise, bpf program has no way to ensure
@@ -980,7 +991,7 @@ static void pcpu_init_value(struct bpf_htab *htab, void __percpu *pptr,
 				zero_map_value(&htab->map, per_cpu_ptr(pptr, cpu));
 		}
 	} else {
-		pcpu_copy_value(htab, pptr, value, onallcpus);
+		pcpu_copy_value(htab, pptr, value, onallcpus, map_flags);
 	}
 }
 
@@ -992,7 +1003,7 @@ static bool fd_htab_map_needs_adjust(const struct bpf_htab *htab)
 static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
 					 void *value, u32 key_size, u32 hash,
 					 bool percpu, bool onallcpus,
-					 struct htab_elem *old_elem)
+					 struct htab_elem *old_elem, u64 map_flags)
 {
 	u32 size = htab->map.value_size;
 	bool prealloc = htab_is_prealloc(htab);
@@ -1050,7 +1061,7 @@ static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
 			pptr = *(void __percpu **)ptr;
 		}
 
-		pcpu_init_value(htab, pptr, value, onallcpus);
+		pcpu_init_value(htab, pptr, value, onallcpus, map_flags);
 
 		if (!prealloc)
 			htab_elem_set_ptr(l_new, key_size, pptr);
@@ -1155,7 +1166,7 @@ static long htab_map_update_elem(struct bpf_map *map, void *key, void *value,
 	}
 
 	l_new = alloc_htab_elem(htab, key, value, key_size, hash, false, false,
-				l_old);
+				l_old, map_flags);
 	if (IS_ERR(l_new)) {
 		/* all pre-allocated elements are in use or memory exhausted */
 		ret = PTR_ERR(l_new);
@@ -1271,9 +1282,11 @@ static long htab_map_update_elem_in_place(struct bpf_map *map, void *key,
 	u32 key_size, hash;
 	int ret;
 
-	if (unlikely(map_flags > BPF_EXIST))
+	if (unlikely(!onallcpus && map_flags > BPF_EXIST))
 		/* unknown flags */
 		return -EINVAL;
+	if (unlikely(onallcpus && ((map_flags & BPF_F_LOCK) || (u32)map_flags > BPF_F_ALL_CPUS)))
+		return -EINVAL;
 
 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
 		     !rcu_read_lock_bh_held());
@@ -1299,7 +1312,7 @@ static long htab_map_update_elem_in_place(struct bpf_map *map, void *key,
 		/* Update value in-place */
 		if (percpu) {
 			pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
-					value, onallcpus);
+					value, onallcpus, map_flags);
 		} else {
 			void **inner_map_pptr = htab_elem_value(l_old, key_size);
 
@@ -1308,7 +1321,7 @@ static long htab_map_update_elem_in_place(struct bpf_map *map, void *key,
 		}
 	} else {
 		l_new = alloc_htab_elem(htab, key, value, key_size,
-					hash, percpu, onallcpus, NULL);
+					hash, percpu, onallcpus, NULL, map_flags);
 		if (IS_ERR(l_new)) {
 			ret = PTR_ERR(l_new);
 			goto err;
@@ -1334,9 +1347,11 @@ static long __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
 	u32 key_size, hash;
 	int ret;
 
-	if (unlikely(map_flags > BPF_EXIST))
+	if (unlikely(!onallcpus && map_flags > BPF_EXIST))
 		/* unknown flags */
 		return -EINVAL;
+	if (unlikely(onallcpus && ((map_flags & BPF_F_LOCK) || (u32)map_flags > BPF_F_ALL_CPUS)))
+		return -EINVAL;
 
 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
 		     !rcu_read_lock_bh_held());
@@ -1374,10 +1389,10 @@ static long __htab_lru_percpu_map_update_elem(struct bpf_map *map, void *key,
 
 		/* per-cpu hash map can update value in-place */
 		pcpu_copy_value(htab, htab_elem_get_ptr(l_old, key_size),
-				value, onallcpus);
+				value, onallcpus, map_flags);
 	} else {
 		pcpu_init_value(htab, htab_elem_get_ptr(l_new, key_size),
-				value, onallcpus);
+				value, onallcpus, map_flags);
 		hlist_nulls_add_head_rcu(&l_new->hash_node, head);
 		l_new = NULL;
 	}
@@ -1689,9 +1704,9 @@ __htab_map_lookup_and_delete_batch(struct bpf_map *map,
 	void __user *ukeys = u64_to_user_ptr(attr->batch.keys);
 	void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
 	u32 batch, max_count, size, bucket_size, map_id;
+	u64 elem_map_flags, map_flags, allowed_flags;
 	u32 bucket_cnt, total, key_size, value_size;
 	struct htab_elem *node_to_free = NULL;
-	u64 elem_map_flags, map_flags;
 	struct hlist_nulls_head *head;
 	struct hlist_nulls_node *n;
 	unsigned long flags = 0;
@@ -1701,9 +1716,12 @@ __htab_map_lookup_and_delete_batch(struct bpf_map *map,
 	int ret = 0;
 
 	elem_map_flags = attr->batch.elem_flags;
-	if ((elem_map_flags & ~BPF_F_LOCK) ||
-	    ((elem_map_flags & BPF_F_LOCK) && !btf_record_has_field(map->record, BPF_SPIN_LOCK)))
-		return -EINVAL;
+	allowed_flags = BPF_F_LOCK;
+	if (!do_delete && is_percpu)
+		allowed_flags |= BPF_F_CPU;
+	ret = bpf_map_check_op_flags(map, elem_map_flags, allowed_flags);
+	if (ret)
+		return ret;
 
 	map_flags = attr->batch.flags;
 	if (map_flags)
@@ -1726,7 +1744,7 @@ __htab_map_lookup_and_delete_batch(struct bpf_map *map,
 	key_size = htab->map.key_size;
 	value_size = htab->map.value_size;
 	size = round_up(value_size, 8);
-	if (is_percpu)
+	if (is_percpu && !(elem_map_flags & BPF_F_CPU))
 		value_size = size * num_possible_cpus();
 	total = 0;
 	/* while experimenting with hash tables with sizes ranging from 10 to
@@ -1809,10 +1827,17 @@ __htab_map_lookup_and_delete_batch(struct bpf_map *map,
 			void __percpu *pptr;
 
 			pptr = htab_elem_get_ptr(l, map->key_size);
-			for_each_possible_cpu(cpu) {
-				copy_map_value_long(&htab->map, dst_val + off, per_cpu_ptr(pptr, cpu));
-				check_and_init_map_value(&htab->map, dst_val + off);
-				off += size;
+			if (elem_map_flags & BPF_F_CPU) {
+				cpu = elem_map_flags >> 32;
+				copy_map_value(&htab->map, dst_val, per_cpu_ptr(pptr, cpu));
+				check_and_init_map_value(&htab->map, dst_val);
+			} else {
+				for_each_possible_cpu(cpu) {
+					copy_map_value_long(&htab->map, dst_val + off,
+							    per_cpu_ptr(pptr, cpu));
+					check_and_init_map_value(&htab->map, dst_val + off);
+					off += size;
+				}
 			}
 		} else {
 			value = htab_elem_value(l, key_size);
@@ -2368,7 +2393,7 @@ static void *htab_lru_percpu_map_lookup_percpu_elem(struct bpf_map *map, void *k
 	return NULL;
 }
 
-int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value)
+int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value, u64 map_flags)
 {
 	struct htab_elem *l;
 	void __percpu *pptr;
@@ -2385,16 +2410,22 @@ int bpf_percpu_hash_copy(struct bpf_map *map, void *key, void *value)
 	l = __htab_map_lookup_elem(map, key);
 	if (!l)
 		goto out;
+	ret = 0;
 	/* We do not mark LRU map element here in order to not mess up
 	 * eviction heuristics when user space does a map walk.
 	 */
 	pptr = htab_elem_get_ptr(l, map->key_size);
+	if (map_flags & BPF_F_CPU) {
+		cpu = map_flags >> 32;
+		copy_map_value_long(map, value, per_cpu_ptr(pptr, cpu));
+		check_and_init_map_value(map, value);
+		goto out;
+	}
 	for_each_possible_cpu(cpu) {
 		copy_map_value_long(map, value + off, per_cpu_ptr(pptr, cpu));
 		check_and_init_map_value(map, value + off);
 		off += size;
 	}
-	ret = 0;
 out:
 	rcu_read_unlock();
 	return ret;
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index 91868fded80e2..f336ca1f48d9b 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -316,7 +316,7 @@ static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 	bpf_disable_instrumentation();
 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
-		err = bpf_percpu_hash_copy(map, key, value);
+		err = bpf_percpu_hash_copy(map, key, value, flags);
 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 		err = bpf_percpu_array_copy(map, key, value, flags);
 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
-- 
2.50.1

[PATCH bpf-next v8 5/7] bpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu_cgroup_storage maps

[Leon Hwang]

Introduce BPF_F_ALL_CPUS flag support for percpu_cgroup_storage maps to
allow updating values for all CPUs with a single value for update_elem
API.

Introduce BPF_F_CPU flag support for percpu_cgroup_storage maps to
allow:

* update value for specified CPU for update_elem API.
* lookup value for specified CPU for lookup_elem API.

The BPF_F_CPU flag is passed via map_flags along with embedded cpu info.

Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
---
 include/linux/bpf-cgroup.h |  4 ++--
 include/linux/bpf.h        |  1 +
 kernel/bpf/local_storage.c | 22 +++++++++++++++++++---
 kernel/bpf/syscall.c       |  2 +-
 4 files changed, 23 insertions(+), 6 deletions(-)

diff --git a/include/linux/bpf-cgroup.h b/include/linux/bpf-cgroup.h
index aedf573bdb426..013f4db9903fd 100644
--- a/include/linux/bpf-cgroup.h
+++ b/include/linux/bpf-cgroup.h
@@ -172,7 +172,7 @@ void bpf_cgroup_storage_link(struct bpf_cgroup_storage *storage,
 void bpf_cgroup_storage_unlink(struct bpf_cgroup_storage *storage);
 int bpf_cgroup_storage_assign(struct bpf_prog_aux *aux, struct bpf_map *map);
 
-int bpf_percpu_cgroup_storage_copy(struct bpf_map *map, void *key, void *value);
+int bpf_percpu_cgroup_storage_copy(struct bpf_map *map, void *key, void *value, u64 flags);
 int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key,
 				     void *value, u64 flags);
 
@@ -467,7 +467,7 @@ static inline struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(
 static inline void bpf_cgroup_storage_free(
 	struct bpf_cgroup_storage *storage) {}
 static inline int bpf_percpu_cgroup_storage_copy(struct bpf_map *map, void *key,
-						 void *value) {
+						 void *value, u64 flags) {
 	return 0;
 }
 static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,
diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 04041dd35d42f..c19237cd122ac 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -3774,6 +3774,7 @@ static inline bool bpf_map_supports_cpu_flags(enum bpf_map_type map_type)
 	case BPF_MAP_TYPE_PERCPU_ARRAY:
 	case BPF_MAP_TYPE_PERCPU_HASH:
 	case BPF_MAP_TYPE_LRU_PERCPU_HASH:
+	case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
 		return true;
 	default:
 		return false;
diff --git a/kernel/bpf/local_storage.c b/kernel/bpf/local_storage.c
index c93a756e035c0..6887a78b4823a 100644
--- a/kernel/bpf/local_storage.c
+++ b/kernel/bpf/local_storage.c
@@ -180,7 +180,7 @@ static long cgroup_storage_update_elem(struct bpf_map *map, void *key,
 }
 
 int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *key,
-				   void *value)
+				   void *value, u64 map_flags)
 {
 	struct bpf_cgroup_storage_map *map = map_to_storage(_map);
 	struct bpf_cgroup_storage *storage;
@@ -199,11 +199,17 @@ int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *key,
 	 * will not leak any kernel data
 	 */
 	size = round_up(_map->value_size, 8);
+	if (map_flags & BPF_F_CPU) {
+		cpu = map_flags >> 32;
+		bpf_long_memcpy(value, per_cpu_ptr(storage->percpu_buf, cpu), size);
+		goto unlock;
+	}
 	for_each_possible_cpu(cpu) {
 		bpf_long_memcpy(value + off,
 				per_cpu_ptr(storage->percpu_buf, cpu), size);
 		off += size;
 	}
+unlock:
 	rcu_read_unlock();
 	return 0;
 }
@@ -216,7 +222,7 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *key,
 	int cpu, off = 0;
 	u32 size;
 
-	if (map_flags != BPF_ANY && map_flags != BPF_EXIST)
+	if ((u32)map_flags & ~(BPF_ANY | BPF_EXIST | BPF_F_CPU | BPF_F_ALL_CPUS))
 		return -EINVAL;
 
 	rcu_read_lock();
@@ -233,11 +239,21 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *key,
 	 * so no kernel data leaks possible
 	 */
 	size = round_up(_map->value_size, 8);
+	if (map_flags & BPF_F_CPU) {
+		cpu = map_flags >> 32;
+		bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu), value, size);
+		goto unlock;
+	}
 	for_each_possible_cpu(cpu) {
 		bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu),
 				value + off, size);
-		off += size;
+		/* same user-provided value is used if BPF_F_ALL_CPUS is
+		 * specified, otherwise value is an array of per-CPU values.
+		 */
+		if (!(map_flags & BPF_F_ALL_CPUS))
+			off += size;
 	}
+unlock:
 	rcu_read_unlock();
 	return 0;
 }
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index f336ca1f48d9b..c227994c065ff 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -320,7 +320,7 @@ static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 		err = bpf_percpu_array_copy(map, key, value, flags);
 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
-		err = bpf_percpu_cgroup_storage_copy(map, key, value);
+		err = bpf_percpu_cgroup_storage_copy(map, key, value, flags);
 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 		err = bpf_stackmap_copy(map, key, value);
 	} else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
-- 
2.50.1

[PATCH bpf-next v8 6/7] libbpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu maps

[Leon Hwang]

Add libbpf support for the BPF_F_CPU flag for percpu maps by embedding the
cpu info into the high 32 bits of:

1. **flags**: bpf_map_lookup_elem_flags(), bpf_map__lookup_elem(),
   bpf_map_update_elem() and bpf_map__update_elem()
2. **opts->elem_flags**: bpf_map_lookup_batch() and
   bpf_map_update_batch()

And the flag can be BPF_F_ALL_CPUS, but cannot be
'BPF_F_CPU | BPF_F_ALL_CPUS'.

Behavior:

* If the flag is BPF_F_ALL_CPUS, the update is applied across all CPUs.
* If the flag is BPF_F_CPU, it updates value only to the specified CPU.
* If the flag is BPF_F_CPU, lookup value only from the specified CPU.
* lookup does not support BPF_F_ALL_CPUS.

Acked-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
---
 tools/lib/bpf/bpf.h    |  8 ++++++++
 tools/lib/bpf/libbpf.c | 26 ++++++++++++++++++++------
 tools/lib/bpf/libbpf.h | 21 ++++++++-------------
 3 files changed, 36 insertions(+), 19 deletions(-)

diff --git a/tools/lib/bpf/bpf.h b/tools/lib/bpf/bpf.h
index e983a3e40d612..ffd93feffd71d 100644
--- a/tools/lib/bpf/bpf.h
+++ b/tools/lib/bpf/bpf.h
@@ -289,6 +289,14 @@ LIBBPF_API int bpf_map_lookup_and_delete_batch(int fd, void *in_batch,
  *    Update spin_lock-ed map elements. This must be
  *    specified if the map value contains a spinlock.
  *
+ * **BPF_F_CPU**
+ *    As for percpu maps, update value on the specified CPU. And the cpu
+ *    info is embedded into the high 32 bits of **opts->elem_flags**.
+ *
+ * **BPF_F_ALL_CPUS**
+ *    As for percpu maps, update value across all CPUs. This flag cannot
+ *    be used with BPF_F_CPU at the same time.
+ *
  * @param fd BPF map file descriptor
  * @param keys pointer to an array of *count* keys
  * @param values pointer to an array of *count* values
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
index 5161c2b398753..61983c603fa17 100644
--- a/tools/lib/bpf/libbpf.c
+++ b/tools/lib/bpf/libbpf.c
@@ -10675,7 +10675,7 @@ bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
 }
 
 static int validate_map_op(const struct bpf_map *map, size_t key_sz,
-			   size_t value_sz, bool check_value_sz)
+			   size_t value_sz, bool check_value_sz, __u64 flags)
 {
 	if (!map_is_created(map)) /* map is not yet created */
 		return -ENOENT;
@@ -10702,6 +10702,20 @@ static int validate_map_op(const struct bpf_map *map, size_t key_sz,
 		int num_cpu = libbpf_num_possible_cpus();
 		size_t elem_sz = roundup(map->def.value_size, 8);
 
+		if (flags & (BPF_F_CPU | BPF_F_ALL_CPUS)) {
+			if ((flags & BPF_F_CPU) && (flags & BPF_F_ALL_CPUS)) {
+				pr_warn("map '%s': BPF_F_CPU and BPF_F_ALL_CPUS are mutually exclusive\n",
+					map->name);
+				return -EINVAL;
+			}
+			if (value_sz != elem_sz) {
+				pr_warn("map '%s': unexpected value size %zu provided for per-CPU map, expected %zu\n",
+					map->name, value_sz, elem_sz);
+				return -EINVAL;
+			}
+			break;
+		}
+
 		if (value_sz != num_cpu * elem_sz) {
 			pr_warn("map '%s': unexpected value size %zu provided for per-CPU map, expected %d * %zu = %zd\n",
 				map->name, value_sz, num_cpu, elem_sz, num_cpu * elem_sz);
@@ -10726,7 +10740,7 @@ int bpf_map__lookup_elem(const struct bpf_map *map,
 {
 	int err;
 
-	err = validate_map_op(map, key_sz, value_sz, true);
+	err = validate_map_op(map, key_sz, value_sz, true, flags);
 	if (err)
 		return libbpf_err(err);
 
@@ -10739,7 +10753,7 @@ int bpf_map__update_elem(const struct bpf_map *map,
 {
 	int err;
 
-	err = validate_map_op(map, key_sz, value_sz, true);
+	err = validate_map_op(map, key_sz, value_sz, true, flags);
 	if (err)
 		return libbpf_err(err);
 
@@ -10751,7 +10765,7 @@ int bpf_map__delete_elem(const struct bpf_map *map,
 {
 	int err;
 
-	err = validate_map_op(map, key_sz, 0, false /* check_value_sz */);
+	err = validate_map_op(map, key_sz, 0, false /* check_value_sz */, flags);
 	if (err)
 		return libbpf_err(err);
 
@@ -10764,7 +10778,7 @@ int bpf_map__lookup_and_delete_elem(const struct bpf_map *map,
 {
 	int err;
 
-	err = validate_map_op(map, key_sz, value_sz, true);
+	err = validate_map_op(map, key_sz, value_sz, true, flags);
 	if (err)
 		return libbpf_err(err);
 
@@ -10776,7 +10790,7 @@ int bpf_map__get_next_key(const struct bpf_map *map,
 {
 	int err;
 
-	err = validate_map_op(map, key_sz, 0, false /* check_value_sz */);
+	err = validate_map_op(map, key_sz, 0, false /* check_value_sz */, 0);
 	if (err)
 		return libbpf_err(err);
 
diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
index 5118d0a90e243..8c06a5cb7569f 100644
--- a/tools/lib/bpf/libbpf.h
+++ b/tools/lib/bpf/libbpf.h
@@ -1196,12 +1196,13 @@ LIBBPF_API struct bpf_map *bpf_map__inner_map(struct bpf_map *map);
  * @param key_sz size in bytes of key data, needs to match BPF map definition's **key_size**
  * @param value pointer to memory in which looked up value will be stored
  * @param value_sz size in byte of value data memory; it has to match BPF map
- * definition's **value_size**. For per-CPU BPF maps value size has to be
- * a product of BPF map value size and number of possible CPUs in the system
- * (could be fetched with **libbpf_num_possible_cpus()**). Note also that for
- * per-CPU values value size has to be aligned up to closest 8 bytes for
- * alignment reasons, so expected size is: `round_up(value_size, 8)
- * * libbpf_num_possible_cpus()`.
+ * definition's **value_size**. For per-CPU BPF maps, value size can be
+ * `round_up(value_size, 8)` if either **BPF_F_CPU** or **BPF_F_ALL_CPUS** is
+ * specified in **flags**, otherwise a product of BPF map value size and number
+ * of possible CPUs in the system (could be fetched with
+ * **libbpf_num_possible_cpus()**). Note also that for per-CPU values value
+ * size has to be aligned up to closest 8 bytes, so expected size is:
+ * `round_up(value_size, 8) * libbpf_num_possible_cpus()`.
  * @flags extra flags passed to kernel for this operation
  * @return 0, on success; negative error, otherwise
  *
@@ -1219,13 +1220,7 @@ LIBBPF_API int bpf_map__lookup_elem(const struct bpf_map *map,
  * @param key pointer to memory containing bytes of the key
  * @param key_sz size in bytes of key data, needs to match BPF map definition's **key_size**
  * @param value pointer to memory containing bytes of the value
- * @param value_sz size in byte of value data memory; it has to match BPF map
- * definition's **value_size**. For per-CPU BPF maps value size has to be
- * a product of BPF map value size and number of possible CPUs in the system
- * (could be fetched with **libbpf_num_possible_cpus()**). Note also that for
- * per-CPU values value size has to be aligned up to closest 8 bytes for
- * alignment reasons, so expected size is: `round_up(value_size, 8)
- * * libbpf_num_possible_cpus()`.
+ * @param value_sz refer to **bpf_map__lookup_elem**'s description.'
  * @flags extra flags passed to kernel for this operation
  * @return 0, on success; negative error, otherwise
  *
-- 
2.50.1

[PATCH bpf-next v8 7/7] selftests/bpf: Add cases to test BPF_F_CPU and BPF_F_ALL_CPUS flags

[Leon Hwang]

Add test coverage for the new BPF_F_CPU and BPF_F_ALL_CPUS flags support
in percpu maps. The following APIs are exercised:

* bpf_map_update_batch()
* bpf_map_lookup_batch()
* bpf_map_update_elem()
* bpf_map__update_elem()
* bpf_map_lookup_elem_flags()
* bpf_map__lookup_elem()

Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
---
 .../selftests/bpf/prog_tests/percpu_alloc.c   | 234 ++++++++++++++++++
 .../selftests/bpf/progs/percpu_alloc_array.c  |  32 +++
 2 files changed, 266 insertions(+)

diff --git a/tools/testing/selftests/bpf/prog_tests/percpu_alloc.c b/tools/testing/selftests/bpf/prog_tests/percpu_alloc.c
index 343da65864d6d..69f881a8cfc47 100644
--- a/tools/testing/selftests/bpf/prog_tests/percpu_alloc.c
+++ b/tools/testing/selftests/bpf/prog_tests/percpu_alloc.c
@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0
 #include <test_progs.h>
+#include "cgroup_helpers.h"
 #include "percpu_alloc_array.skel.h"
 #include "percpu_alloc_cgrp_local_storage.skel.h"
 #include "percpu_alloc_fail.skel.h"
@@ -115,6 +116,231 @@ static void test_failure(void) {
 	RUN_TESTS(percpu_alloc_fail);
 }
 
+static void test_percpu_map_op_cpu_flag(struct bpf_map *map, void *keys, size_t key_sz,
+					u32 max_entries, bool test_batch)
+{
+	int i, j, cpu, map_fd, value_size, nr_cpus, err;
+	u64 *values = NULL, batch = 0, flags;
+	const u64 value = 0xDEADC0DE;
+	size_t value_sz = sizeof(u64);
+	u32 count = max_entries;
+	LIBBPF_OPTS(bpf_map_batch_opts, batch_opts);
+
+	nr_cpus = libbpf_num_possible_cpus();
+	if (!ASSERT_GT(nr_cpus, 0, "libbpf_num_possible_cpus"))
+		return;
+
+	value_size = value_sz * nr_cpus;
+	values = calloc(max_entries, value_size);
+	if (!ASSERT_OK_PTR(values, "calloc values"))
+		goto out;
+	memset(values, 0, value_size * max_entries);
+
+	map_fd = bpf_map__fd(map);
+	flags = BPF_F_CPU | BPF_F_ALL_CPUS;
+	err = bpf_map_lookup_elem_flags(map_fd, keys, values, flags);
+	if (!ASSERT_ERR(err, "bpf_map_lookup_elem_flags err"))
+		goto out;
+
+	err = bpf_map_update_elem(map_fd, keys, values, flags);
+	if (!ASSERT_ERR(err, "bpf_map_update_elem err"))
+		goto out;
+
+	flags = (u64)nr_cpus << 32 | BPF_F_CPU;
+	err = bpf_map_update_elem(map_fd, keys, values, flags);
+	if (!ASSERT_EQ(err, -ERANGE, "bpf_map_update_elem -ERANGE"))
+		goto out;
+
+	err = bpf_map__update_elem(map, keys, key_sz, values, value_sz, flags);
+	if (!ASSERT_EQ(err, -ERANGE, "bpf_map__update_elem -ERANGE"))
+		goto out;
+
+	err = bpf_map_lookup_elem_flags(map_fd, keys, values, flags);
+	if (!ASSERT_EQ(err, -ERANGE, "bpf_map_lookup_elem_flags -ERANGE"))
+		goto out;
+
+	err = bpf_map__lookup_elem(map, keys, key_sz, values, value_sz, flags);
+	if (!ASSERT_EQ(err, -ERANGE, "bpf_map__lookup_elem -ERANGE"))
+		goto out;
+
+	for (cpu = 0; cpu < nr_cpus; cpu++) {
+		/* clear value on all cpus */
+		values[0] = 0;
+		flags = BPF_F_ALL_CPUS;
+		for (i = 0; i < max_entries; i++) {
+			err = bpf_map__update_elem(map, keys + i * key_sz, key_sz, values,
+						   value_sz, flags);
+			if (!ASSERT_OK(err, "bpf_map__update_elem all_cpus"))
+				goto out;
+		}
+
+		/* update value on specified cpu */
+		for (i = 0; i < max_entries; i++) {
+			values[0] = value;
+			flags = (u64)cpu << 32 | BPF_F_CPU;
+			err = bpf_map__update_elem(map, keys + i * key_sz, key_sz, values,
+						   value_sz, flags);
+			if (!ASSERT_OK(err, "bpf_map__update_elem specified cpu"))
+				goto out;
+
+			/* lookup then check value on CPUs */
+			for (j = 0; j < nr_cpus; j++) {
+				flags = (u64)j << 32 | BPF_F_CPU;
+				err = bpf_map__lookup_elem(map, keys + i * key_sz, key_sz, values,
+							   value_sz, flags);
+				if (!ASSERT_OK(err, "bpf_map__lookup_elem specified cpu"))
+					goto out;
+				if (!ASSERT_EQ(values[0], j != cpu ? 0 : value,
+					       "bpf_map__lookup_elem value on specified cpu"))
+					goto out;
+			}
+		}
+	}
+
+	if (!test_batch)
+		goto out;
+
+	batch_opts.elem_flags = (u64)nr_cpus << 32 | BPF_F_CPU;
+	err = bpf_map_update_batch(map_fd, keys, values, &max_entries, &batch_opts);
+	if (!ASSERT_EQ(err, -ERANGE, "bpf_map_update_batch -ERANGE"))
+		goto out;
+
+	for (cpu = 0; cpu < nr_cpus; cpu++) {
+		memset(values, 0, max_entries * value_size);
+
+		/* clear values across all CPUs */
+		batch_opts.elem_flags = BPF_F_ALL_CPUS;
+		err = bpf_map_update_batch(map_fd, keys, values, &max_entries, &batch_opts);
+		if (!ASSERT_OK(err, "bpf_map_update_batch all_cpus"))
+			goto out;
+
+		/* update values on specified CPU */
+		for (i = 0; i < max_entries; i++)
+			values[i] = value;
+
+		batch_opts.elem_flags = (u64)cpu << 32 | BPF_F_CPU;
+		err = bpf_map_update_batch(map_fd, keys, values, &max_entries, &batch_opts);
+		if (!ASSERT_OK(err, "bpf_map_update_batch specified cpu"))
+			goto out;
+
+		/* lookup values on specified CPU */
+		memset(values, 0, max_entries * value_sz);
+		err = bpf_map_lookup_batch(map_fd, NULL, &batch, keys, values, &count, &batch_opts);
+		if (!ASSERT_TRUE(!err || err == -ENOENT, "bpf_map_lookup_batch specified cpu"))
+			goto out;
+
+		for (i = 0; i < max_entries; i++)
+			if (!ASSERT_EQ(values[i], value,
+				       "bpf_map_lookup_batch value on specified cpu"))
+				goto out;
+
+		/* lookup values from all CPUs */
+		batch_opts.elem_flags = 0;
+		memset(values, 0, max_entries * value_size);
+		err = bpf_map_lookup_batch(map_fd, NULL, &batch, keys, values, &count, &batch_opts);
+		if (!ASSERT_TRUE(!err || err == -ENOENT, "bpf_map_lookup_batch all_cpus"))
+			goto out;
+
+		for (i = 0; i < max_entries; i++) {
+			for (j = 0; j < nr_cpus; j++) {
+				if (!ASSERT_EQ(values[i*nr_cpus + j], j != cpu ? 0 : value,
+					       "bpf_map_lookup_batch value all_cpus"))
+					goto out;
+			}
+		}
+	}
+
+out:
+	if (values)
+		free(values);
+}
+
+static void test_percpu_map_cpu_flag(enum bpf_map_type map_type)
+{
+	struct percpu_alloc_array *skel;
+	size_t key_sz = sizeof(int);
+	int *keys = NULL, i, err;
+	struct bpf_map *map;
+	u32 max_entries;
+
+	skel = percpu_alloc_array__open();
+	if (!ASSERT_OK_PTR(skel, "percpu_alloc_array__open"))
+		return;
+
+	map = skel->maps.percpu;
+	bpf_map__set_type(map, map_type);
+
+	err = percpu_alloc_array__load(skel);
+	if (!ASSERT_OK(err, "test_percpu_alloc__load"))
+		goto out;
+
+	max_entries = bpf_map__max_entries(map);
+	keys = calloc(max_entries, key_sz);
+	if (!ASSERT_OK_PTR(keys, "calloc keys"))
+		goto out;
+
+	for (i = 0; i < max_entries; i++)
+		keys[i] = i;
+
+	test_percpu_map_op_cpu_flag(map, keys, key_sz, max_entries, true);
+out:
+	if (keys)
+		free(keys);
+	percpu_alloc_array__destroy(skel);
+}
+
+static void test_percpu_array_cpu_flag(void)
+{
+	test_percpu_map_cpu_flag(BPF_MAP_TYPE_PERCPU_ARRAY);
+}
+
+static void test_percpu_hash_cpu_flag(void)
+{
+	test_percpu_map_cpu_flag(BPF_MAP_TYPE_PERCPU_HASH);
+}
+
+static void test_lru_percpu_hash_cpu_flag(void)
+{
+	test_percpu_map_cpu_flag(BPF_MAP_TYPE_LRU_PERCPU_HASH);
+}
+
+static void test_percpu_cgroup_storage_cpu_flag(void)
+{
+	struct bpf_cgroup_storage_key key;
+	struct percpu_alloc_array *skel;
+	int cgroup = -1, prog_fd, err;
+	struct bpf_map *map;
+
+	skel = percpu_alloc_array__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "percpu_alloc_array__open_and_load"))
+		return;
+
+	cgroup = create_and_get_cgroup("/cg_percpu");
+	if (!ASSERT_GE(cgroup, 0, "create_and_get_cgroup"))
+		goto out;
+
+	err = join_cgroup("/cg_percpu");
+	if (!ASSERT_OK(err, "join_cgroup"))
+		goto out;
+
+	prog_fd = bpf_program__fd(skel->progs.cgroup_egress);
+	err = bpf_prog_attach(prog_fd, cgroup, BPF_CGROUP_INET_EGRESS, 0);
+	if (!ASSERT_OK(err, "bpf_prog_attach"))
+		goto out;
+
+	map = skel->maps.percpu_cgroup_storage;
+	err = bpf_map_get_next_key(bpf_map__fd(map), NULL, &key);
+	if (!ASSERT_OK(err, "bpf_map_get_next_key"))
+		goto out;
+
+	test_percpu_map_op_cpu_flag(map, &key, sizeof(key), 1, false);
+out:
+	bpf_prog_detach2(-1, cgroup, BPF_CGROUP_INET_EGRESS);
+	close(cgroup);
+	cleanup_cgroup_environment();
+	percpu_alloc_array__destroy(skel);
+}
+
 void test_percpu_alloc(void)
 {
 	if (test__start_subtest("array"))
@@ -125,4 +351,12 @@ void test_percpu_alloc(void)
 		test_cgrp_local_storage();
 	if (test__start_subtest("failure_tests"))
 		test_failure();
+	if (test__start_subtest("cpu_flag_percpu_array"))
+		test_percpu_array_cpu_flag();
+	if (test__start_subtest("cpu_flag_percpu_hash"))
+		test_percpu_hash_cpu_flag();
+	if (test__start_subtest("cpu_flag_lru_percpu_hash"))
+		test_lru_percpu_hash_cpu_flag();
+	if (test__start_subtest("cpu_flag_percpu_cgroup_storage"))
+		test_percpu_cgroup_storage_cpu_flag();
 }
diff --git a/tools/testing/selftests/bpf/progs/percpu_alloc_array.c b/tools/testing/selftests/bpf/progs/percpu_alloc_array.c
index 37c2d2608ec0b..427301909c349 100644
--- a/tools/testing/selftests/bpf/progs/percpu_alloc_array.c
+++ b/tools/testing/selftests/bpf/progs/percpu_alloc_array.c
@@ -187,4 +187,36 @@ int BPF_PROG(test_array_map_10)
 	return 0;
 }
 
+struct {
+	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
+	__uint(max_entries, 2);
+	__type(key, int);
+	__type(value, u64);
+} percpu SEC(".maps");
+
+SEC("?fentry/bpf_fentry_test1")
+int BPF_PROG(test_percpu_array, int x)
+{
+	u64 value = 0xDEADC0DE;
+	int key = 0;
+
+	bpf_map_update_elem(&percpu, &key, &value, BPF_ANY);
+	return 0;
+}
+
+struct {
+	__uint(type, BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE);
+	__type(key, struct bpf_cgroup_storage_key);
+	__type(value, u64);
+} percpu_cgroup_storage SEC(".maps");
+
+SEC("cgroup_skb/egress")
+int cgroup_egress(struct __sk_buff *skb)
+{
+	u64 *val = bpf_get_local_storage(&percpu_cgroup_storage, 0);
+
+	__sync_fetch_and_add(val, 1);
+	return 1;
+}
+
 char _license[] SEC("license") = "GPL";
-- 
2.50.1

Re: [PATCH bpf-next v8 2/7] bpf: Introduce BPF_F_CPU and BPF_F_ALL_CPUS flags

[Andrii Nakryiko]

On Thu, Sep 25, 2025 at 8:38 AM Leon Hwang <leon.hwang@linux.dev> wrote:
>
> Introduce BPF_F_CPU and BPF_F_ALL_CPUS flags and check them for
> following APIs:
>
> * 'map_lookup_elem()'
> * 'map_update_elem()'
> * 'generic_map_lookup_batch()'
> * 'generic_map_update_batch()'
>
> And, get the correct value size for these APIs.
>
> Acked-by: Andrii Nakryiko <andrii@kernel.org>
> Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
> ---
>  include/linux/bpf.h            | 23 ++++++++++++++++++++++-
>  include/uapi/linux/bpf.h       |  2 ++
>  kernel/bpf/syscall.c           | 31 +++++++++++++++++--------------
>  tools/include/uapi/linux/bpf.h |  2 ++
>  4 files changed, 43 insertions(+), 15 deletions(-)
>

[...]

> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> index 84261a0211c51..d1bbc309b62ff 100644
> --- a/kernel/bpf/syscall.c
> +++ b/kernel/bpf/syscall.c
> @@ -133,12 +133,14 @@ bool bpf_map_write_active(const struct bpf_map *map)
>         return atomic64_read(&map->writecnt) != 0;
>  }
>
> -static u32 bpf_map_value_size(const struct bpf_map *map)
> -{
> -       if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
> -           map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
> -           map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
> -           map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
> +static u32 bpf_map_value_size(const struct bpf_map *map, u64 flags)
> +{
> +       if (flags & (BPF_F_CPU | BPF_F_ALL_CPUS))
> +               return round_up(map->value_size, 8);

with the idea that BPF_F_CPU (and BPF_F_ALL_CPUS for updates) turns
per-CPU lookup/update into a singular non-per-CPU lookup, shouldn't
this be just `return map->value_size;`?

pw-bot: cr

> +       else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
> +                map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
> +                map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
> +                map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
>                 return round_up(map->value_size, 8) * num_possible_cpus();
>         else if (IS_FD_MAP(map))
>                 return sizeof(u32);

[...]

Re: [PATCH bpf-next v8 5/7] bpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu_cgroup_storage maps

[Andrii Nakryiko]

On Thu, Sep 25, 2025 at 8:38 AM Leon Hwang <leon.hwang@linux.dev> wrote:
>
> Introduce BPF_F_ALL_CPUS flag support for percpu_cgroup_storage maps to
> allow updating values for all CPUs with a single value for update_elem
> API.
>
> Introduce BPF_F_CPU flag support for percpu_cgroup_storage maps to
> allow:
>
> * update value for specified CPU for update_elem API.
> * lookup value for specified CPU for lookup_elem API.
>
> The BPF_F_CPU flag is passed via map_flags along with embedded cpu info.
>
> Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
> ---
>  include/linux/bpf-cgroup.h |  4 ++--
>  include/linux/bpf.h        |  1 +
>  kernel/bpf/local_storage.c | 22 +++++++++++++++++++---
>  kernel/bpf/syscall.c       |  2 +-
>  4 files changed, 23 insertions(+), 6 deletions(-)
>

[...]

>  int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *key,
> -                                  void *value)
> +                                  void *value, u64 map_flags)
>  {
>         struct bpf_cgroup_storage_map *map = map_to_storage(_map);
>         struct bpf_cgroup_storage *storage;
> @@ -199,11 +199,17 @@ int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *key,
>          * will not leak any kernel data
>          */
>         size = round_up(_map->value_size, 8);

um... same issue with rounding up value_size when BPF_F_CPU is set, no?

> +       if (map_flags & BPF_F_CPU) {
> +               cpu = map_flags >> 32;
> +               bpf_long_memcpy(value, per_cpu_ptr(storage->percpu_buf, cpu), size);
> +               goto unlock;
> +       }
>         for_each_possible_cpu(cpu) {
>                 bpf_long_memcpy(value + off,
>                                 per_cpu_ptr(storage->percpu_buf, cpu), size);
>                 off += size;
>         }
> +unlock:
>         rcu_read_unlock();
>         return 0;
>  }
> @@ -216,7 +222,7 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *key,
>         int cpu, off = 0;
>         u32 size;
>
> -       if (map_flags != BPF_ANY && map_flags != BPF_EXIST)
> +       if ((u32)map_flags & ~(BPF_ANY | BPF_EXIST | BPF_F_CPU | BPF_F_ALL_CPUS))
>                 return -EINVAL;
>
>         rcu_read_lock();
> @@ -233,11 +239,21 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *key,
>          * so no kernel data leaks possible
>          */
>         size = round_up(_map->value_size, 8);
> +       if (map_flags & BPF_F_CPU) {
> +               cpu = map_flags >> 32;
> +               bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu), value, size);

ditto

> +               goto unlock;
> +       }
>         for_each_possible_cpu(cpu) {
>                 bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu),
>                                 value + off, size);
> -               off += size;
> +               /* same user-provided value is used if BPF_F_ALL_CPUS is
> +                * specified, otherwise value is an array of per-CPU values.
> +                */
> +               if (!(map_flags & BPF_F_ALL_CPUS))
> +                       off += size;

btw, given we'll need another revision to fix up all those round_up()
issues, what do you think about make this offset logic completely
stateless (and, in my opinion, more obvious):

for_each_possible_cpu(cpu) {
    p = (map_flags & BPF_F_ALL_CPUS) ? value : value + size * cpu;
    memcpy(per_cpu_ptr(storage->percpu_buf, cpu), p, size);
}

seems more straightforward to me

P.S. And please update this everywhere for consistency ;)


>         }
> +unlock:
>         rcu_read_unlock();
>         return 0;
>  }
> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> index f336ca1f48d9b..c227994c065ff 100644
> --- a/kernel/bpf/syscall.c
> +++ b/kernel/bpf/syscall.c
> @@ -320,7 +320,7 @@ static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
>         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
>                 err = bpf_percpu_array_copy(map, key, value, flags);
>         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
> -               err = bpf_percpu_cgroup_storage_copy(map, key, value);
> +               err = bpf_percpu_cgroup_storage_copy(map, key, value, flags);
>         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
>                 err = bpf_stackmap_copy(map, key, value);
>         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
> --
> 2.50.1
>

Re: [PATCH bpf-next v8 6/7] libbpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu maps

[Andrii Nakryiko]

On Thu, Sep 25, 2025 at 8:38 AM Leon Hwang <leon.hwang@linux.dev> wrote:
>
> Add libbpf support for the BPF_F_CPU flag for percpu maps by embedding the
> cpu info into the high 32 bits of:
>
> 1. **flags**: bpf_map_lookup_elem_flags(), bpf_map__lookup_elem(),
>    bpf_map_update_elem() and bpf_map__update_elem()
> 2. **opts->elem_flags**: bpf_map_lookup_batch() and
>    bpf_map_update_batch()
>
> And the flag can be BPF_F_ALL_CPUS, but cannot be
> 'BPF_F_CPU | BPF_F_ALL_CPUS'.
>
> Behavior:
>
> * If the flag is BPF_F_ALL_CPUS, the update is applied across all CPUs.
> * If the flag is BPF_F_CPU, it updates value only to the specified CPU.
> * If the flag is BPF_F_CPU, lookup value only from the specified CPU.
> * lookup does not support BPF_F_ALL_CPUS.
>
> Acked-by: Andrii Nakryiko <andrii@kernel.org>
> Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
> ---
>  tools/lib/bpf/bpf.h    |  8 ++++++++
>  tools/lib/bpf/libbpf.c | 26 ++++++++++++++++++++------
>  tools/lib/bpf/libbpf.h | 21 ++++++++-------------
>  3 files changed, 36 insertions(+), 19 deletions(-)
>

[...]

> diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
> index 5118d0a90e243..8c06a5cb7569f 100644
> --- a/tools/lib/bpf/libbpf.h
> +++ b/tools/lib/bpf/libbpf.h
> @@ -1196,12 +1196,13 @@ LIBBPF_API struct bpf_map *bpf_map__inner_map(struct bpf_map *map);
>   * @param key_sz size in bytes of key data, needs to match BPF map definition's **key_size**
>   * @param value pointer to memory in which looked up value will be stored
>   * @param value_sz size in byte of value data memory; it has to match BPF map
> - * definition's **value_size**. For per-CPU BPF maps value size has to be
> - * a product of BPF map value size and number of possible CPUs in the system
> - * (could be fetched with **libbpf_num_possible_cpus()**). Note also that for
> - * per-CPU values value size has to be aligned up to closest 8 bytes for
> - * alignment reasons, so expected size is: `round_up(value_size, 8)
> - * * libbpf_num_possible_cpus()`.
> + * definition's **value_size**. For per-CPU BPF maps, value size can be
> + * `round_up(value_size, 8)` if either **BPF_F_CPU** or **BPF_F_ALL_CPUS** is

please update, it's not round_up(), as we decided earlier


> + * specified in **flags**, otherwise a product of BPF map value size and number
> + * of possible CPUs in the system (could be fetched with
> + * **libbpf_num_possible_cpus()**). Note also that for per-CPU values value
> + * size has to be aligned up to closest 8 bytes, so expected size is:
> + * `round_up(value_size, 8) * libbpf_num_possible_cpus()`.
>   * @flags extra flags passed to kernel for this operation
>   * @return 0, on success; negative error, otherwise
>   *
> @@ -1219,13 +1220,7 @@ LIBBPF_API int bpf_map__lookup_elem(const struct bpf_map *map,
>   * @param key pointer to memory containing bytes of the key
>   * @param key_sz size in bytes of key data, needs to match BPF map definition's **key_size**
>   * @param value pointer to memory containing bytes of the value
> - * @param value_sz size in byte of value data memory; it has to match BPF map
> - * definition's **value_size**. For per-CPU BPF maps value size has to be
> - * a product of BPF map value size and number of possible CPUs in the system
> - * (could be fetched with **libbpf_num_possible_cpus()**). Note also that for
> - * per-CPU values value size has to be aligned up to closest 8 bytes for
> - * alignment reasons, so expected size is: `round_up(value_size, 8)
> - * * libbpf_num_possible_cpus()`.
> + * @param value_sz refer to **bpf_map__lookup_elem**'s description.'
>   * @flags extra flags passed to kernel for this operation
>   * @return 0, on success; negative error, otherwise
>   *
> --
> 2.50.1
>

Re: [PATCH bpf-next v8 5/7] bpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu_cgroup_storage maps

[Leon Hwang]

On 2025/9/28 10:42, Andrii Nakryiko wrote:
> On Thu, Sep 25, 2025 at 8:38 AM Leon Hwang <leon.hwang@linux.dev> wrote:
>>
>> Introduce BPF_F_ALL_CPUS flag support for percpu_cgroup_storage maps to
>> allow updating values for all CPUs with a single value for update_elem
>> API.
>>
>> Introduce BPF_F_CPU flag support for percpu_cgroup_storage maps to
>> allow:
>>
>> * update value for specified CPU for update_elem API.
>> * lookup value for specified CPU for lookup_elem API.
>>
>> The BPF_F_CPU flag is passed via map_flags along with embedded cpu info.
>>
>> Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
>> ---
>>  include/linux/bpf-cgroup.h |  4 ++--
>>  include/linux/bpf.h        |  1 +
>>  kernel/bpf/local_storage.c | 22 +++++++++++++++++++---
>>  kernel/bpf/syscall.c       |  2 +-
>>  4 files changed, 23 insertions(+), 6 deletions(-)
>>
>
> [...]
>
>>  int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *key,
>> -                                  void *value)
>> +                                  void *value, u64 map_flags)
>>  {
>>         struct bpf_cgroup_storage_map *map = map_to_storage(_map);
>>         struct bpf_cgroup_storage *storage;
>> @@ -199,11 +199,17 @@ int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *key,
>>          * will not leak any kernel data
>>          */
>>         size = round_up(_map->value_size, 8);
>
> um... same issue with rounding up value_size when BPF_F_CPU is set, no?
>
>> +       if (map_flags & BPF_F_CPU) {
>> +               cpu = map_flags >> 32;
>> +               bpf_long_memcpy(value, per_cpu_ptr(storage->percpu_buf, cpu), size);
>> +               goto unlock;
>> +       }
>>         for_each_possible_cpu(cpu) {
>>                 bpf_long_memcpy(value + off,
>>                                 per_cpu_ptr(storage->percpu_buf, cpu), size);
>>                 off += size;
>>         }
>> +unlock:
>>         rcu_read_unlock();
>>         return 0;
>>  }
>> @@ -216,7 +222,7 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *key,
>>         int cpu, off = 0;
>>         u32 size;
>>
>> -       if (map_flags != BPF_ANY && map_flags != BPF_EXIST)
>> +       if ((u32)map_flags & ~(BPF_ANY | BPF_EXIST | BPF_F_CPU | BPF_F_ALL_CPUS))
>>                 return -EINVAL;
>>
>>         rcu_read_lock();
>> @@ -233,11 +239,21 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *key,
>>          * so no kernel data leaks possible
>>          */
>>         size = round_up(_map->value_size, 8);
>> +       if (map_flags & BPF_F_CPU) {
>> +               cpu = map_flags >> 32;
>> +               bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu), value, size);
>
> ditto
>
>> +               goto unlock;
>> +       }
>>         for_each_possible_cpu(cpu) {
>>                 bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu),
>>                                 value + off, size);
>> -               off += size;
>> +               /* same user-provided value is used if BPF_F_ALL_CPUS is
>> +                * specified, otherwise value is an array of per-CPU values.
>> +                */
>> +               if (!(map_flags & BPF_F_ALL_CPUS))
>> +                       off += size;
>
> btw, given we'll need another revision to fix up all those round_up()
> issues, what do you think about make this offset logic completely
> stateless (and, in my opinion, more obvious):
>
> for_each_possible_cpu(cpu) {
>     p = (map_flags & BPF_F_ALL_CPUS) ? value : value + size * cpu;
>     memcpy(per_cpu_ptr(storage->percpu_buf, cpu), p, size);
> }
>
> seems more straightforward to me

lgtm.

But I think the correct memcpy() should look like this:

memcpy(per_cpu_ptr(storage->percpu_buf, cpu), p,
       (map_flags & BPF_F_ALL_CPUS) ? _map->value_size : size);

because 'size' is 8-byte aligned and can’t be used directly when
'map_flags & BPF_F_ALL_CPUS' is set.

So the more accurate version would be:

for_each_possible_cpu(cpu) {
    p = (map_flags & BPF_F_ALL_CPUS) ? value : value + size * cpu;
    s = (map_flags & BPF_F_ALL_CPUS) ? _map->value_size : size;
    memcpy(per_cpu_ptr(storage->percpu_buf, cpu), p, s);
}

Isn’t this the correct approach?

Thanks,
Leon

[...]

Re: [PATCH bpf-next v8 5/7] bpf: Add BPF_F_CPU and BPF_F_ALL_CPUS flags support for percpu_cgroup_storage maps

[Andrii Nakryiko]

On Sun, Sep 28, 2025 at 8:06 AM Leon Hwang <leon.hwang@linux.dev> wrote:
>
>
>
> On 2025/9/28 10:42, Andrii Nakryiko wrote:
> > On Thu, Sep 25, 2025 at 8:38 AM Leon Hwang <leon.hwang@linux.dev> wrote:
> >>
> >> Introduce BPF_F_ALL_CPUS flag support for percpu_cgroup_storage maps to
> >> allow updating values for all CPUs with a single value for update_elem
> >> API.
> >>
> >> Introduce BPF_F_CPU flag support for percpu_cgroup_storage maps to
> >> allow:
> >>
> >> * update value for specified CPU for update_elem API.
> >> * lookup value for specified CPU for lookup_elem API.
> >>
> >> The BPF_F_CPU flag is passed via map_flags along with embedded cpu info.
> >>
> >> Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
> >> ---
> >>  include/linux/bpf-cgroup.h |  4 ++--
> >>  include/linux/bpf.h        |  1 +
> >>  kernel/bpf/local_storage.c | 22 +++++++++++++++++++---
> >>  kernel/bpf/syscall.c       |  2 +-
> >>  4 files changed, 23 insertions(+), 6 deletions(-)
> >>
> >
> > [...]
> >
> >>  int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *key,
> >> -                                  void *value)
> >> +                                  void *value, u64 map_flags)
> >>  {
> >>         struct bpf_cgroup_storage_map *map = map_to_storage(_map);
> >>         struct bpf_cgroup_storage *storage;
> >> @@ -199,11 +199,17 @@ int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *key,
> >>          * will not leak any kernel data
> >>          */
> >>         size = round_up(_map->value_size, 8);
> >
> > um... same issue with rounding up value_size when BPF_F_CPU is set, no?
> >
> >> +       if (map_flags & BPF_F_CPU) {
> >> +               cpu = map_flags >> 32;
> >> +               bpf_long_memcpy(value, per_cpu_ptr(storage->percpu_buf, cpu), size);
> >> +               goto unlock;
> >> +       }
> >>         for_each_possible_cpu(cpu) {
> >>                 bpf_long_memcpy(value + off,
> >>                                 per_cpu_ptr(storage->percpu_buf, cpu), size);
> >>                 off += size;
> >>         }
> >> +unlock:
> >>         rcu_read_unlock();
> >>         return 0;
> >>  }
> >> @@ -216,7 +222,7 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *key,
> >>         int cpu, off = 0;
> >>         u32 size;
> >>
> >> -       if (map_flags != BPF_ANY && map_flags != BPF_EXIST)
> >> +       if ((u32)map_flags & ~(BPF_ANY | BPF_EXIST | BPF_F_CPU | BPF_F_ALL_CPUS))
> >>                 return -EINVAL;
> >>
> >>         rcu_read_lock();
> >> @@ -233,11 +239,21 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *key,
> >>          * so no kernel data leaks possible
> >>          */
> >>         size = round_up(_map->value_size, 8);
> >> +       if (map_flags & BPF_F_CPU) {
> >> +               cpu = map_flags >> 32;
> >> +               bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu), value, size);
> >
> > ditto
> >
> >> +               goto unlock;
> >> +       }
> >>         for_each_possible_cpu(cpu) {
> >>                 bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu),
> >>                                 value + off, size);
> >> -               off += size;
> >> +               /* same user-provided value is used if BPF_F_ALL_CPUS is
> >> +                * specified, otherwise value is an array of per-CPU values.
> >> +                */
> >> +               if (!(map_flags & BPF_F_ALL_CPUS))
> >> +                       off += size;
> >
> > btw, given we'll need another revision to fix up all those round_up()
> > issues, what do you think about make this offset logic completely
> > stateless (and, in my opinion, more obvious):
> >
> > for_each_possible_cpu(cpu) {
> >     p = (map_flags & BPF_F_ALL_CPUS) ? value : value + size * cpu;
> >     memcpy(per_cpu_ptr(storage->percpu_buf, cpu), p, size);
> > }
> >
> > seems more straightforward to me
>
> lgtm.
>
> But I think the correct memcpy() should look like this:
>
> memcpy(per_cpu_ptr(storage->percpu_buf, cpu), p,
>        (map_flags & BPF_F_ALL_CPUS) ? _map->value_size : size);
>
> because 'size' is 8-byte aligned and can’t be used directly when
> 'map_flags & BPF_F_ALL_CPUS' is set.
>
> So the more accurate version would be:
>
> for_each_possible_cpu(cpu) {
>     p = (map_flags & BPF_F_ALL_CPUS) ? value : value + size * cpu;
>     s = (map_flags & BPF_F_ALL_CPUS) ? _map->value_size : size;
>     memcpy(per_cpu_ptr(storage->percpu_buf, cpu), p, s);
> }
>
> Isn’t this the correct approach?
>

Yes, but I think it would be better to update:

size = round_up(_map->value_size, 8);

to take into account BPF_F_ALL_CPUS instead and do it once outside of
the loop. But same idea, of course, yes.

> Thanks,
> Leon
>
> [...]