| Message ID | 20230227152032.12359-3-laoar.shao@gmail.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | BPF |
| Headers | show |
| Series | bpf: bpf memory usage | expand |
From: Hou Tao <houtao1@huawei.com> Hi, > trie_mem_usage() is introduced to calculate the lpm_trie memory usage. > Some small memory allocations are ignored. The inner node is also > ignored. > > The result as follows, > > - before > 10: lpm_trie flags 0x1 > key 8B value 8B max_entries 65536 memlock 1048576B > > - after > 10: lpm_trie flags 0x1 > key 8B value 8B max_entries 65536 memlock 2291536B > > Signed-off-by: Yafang Shao <laoar.shao@gmail.com> > --- > kernel/bpf/lpm_trie.c | 11 +++++++++++ > 1 file changed, 11 insertions(+) > > diff --git a/kernel/bpf/lpm_trie.c b/kernel/bpf/lpm_trie.c > index d833496..e0ca08e 100644 > --- a/kernel/bpf/lpm_trie.c > +++ b/kernel/bpf/lpm_trie.c > @@ -720,6 +720,16 @@ static int trie_check_btf(const struct bpf_map *map, > -EINVAL : 0; > } > > +static u64 trie_mem_usage(const struct bpf_map *map) > +{ > + struct lpm_trie *trie = container_of(map, struct lpm_trie, map); > + u64 elem_size; > + > + elem_size = sizeof(struct lpm_trie_node) + trie->data_size + > + trie->map.value_size; > + return elem_size * trie->n_entries; Need to use READ_ONCE(trie->n_entries) because all updates of n_entries are protected by trie->lock and here it is a lockless read. > +} > + > BTF_ID_LIST_SINGLE(trie_map_btf_ids, struct, lpm_trie) > const struct bpf_map_ops trie_map_ops = { > .map_meta_equal = bpf_map_meta_equal, > @@ -733,5 +743,6 @@ static int trie_check_btf(const struct bpf_map *map, > .map_update_batch = generic_map_update_batch, > .map_delete_batch = generic_map_delete_batch, > .map_check_btf = trie_check_btf, > + .map_mem_usage = trie_mem_usage, > .map_btf_id = &trie_map_btf_ids[0], > }; > -- > 1.8.3.1 >
On Wed, Mar 1, 2023 at 10:31 AM Hou Tao <houtao@huaweicloud.com> wrote: > > From: Hou Tao <houtao1@huawei.com> > > Hi, > > > trie_mem_usage() is introduced to calculate the lpm_trie memory usage. > > Some small memory allocations are ignored. The inner node is also > > ignored. > > > > The result as follows, > > > > - before > > 10: lpm_trie flags 0x1 > > key 8B value 8B max_entries 65536 memlock 1048576B > > > > - after > > 10: lpm_trie flags 0x1 > > key 8B value 8B max_entries 65536 memlock 2291536B > > > > Signed-off-by: Yafang Shao <laoar.shao@gmail.com> > > --- > > kernel/bpf/lpm_trie.c | 11 +++++++++++ > > 1 file changed, 11 insertions(+) > > > > diff --git a/kernel/bpf/lpm_trie.c b/kernel/bpf/lpm_trie.c > > index d833496..e0ca08e 100644 > > --- a/kernel/bpf/lpm_trie.c > > +++ b/kernel/bpf/lpm_trie.c > > @@ -720,6 +720,16 @@ static int trie_check_btf(const struct bpf_map *map, > > -EINVAL : 0; > > } > > > > +static u64 trie_mem_usage(const struct bpf_map *map) > > +{ > > + struct lpm_trie *trie = container_of(map, struct lpm_trie, map); > > + u64 elem_size; > > + > > + elem_size = sizeof(struct lpm_trie_node) + trie->data_size + > > + trie->map.value_size; > > + return elem_size * trie->n_entries; > Need to use READ_ONCE(trie->n_entries) because all updates of n_entries are protected by trie->lock and here it is a lockless read. Indeed. Will change it in the next version. Thanks for your review.
diff --git a/kernel/bpf/lpm_trie.c b/kernel/bpf/lpm_trie.c index d833496..e0ca08e 100644 --- a/kernel/bpf/lpm_trie.c +++ b/kernel/bpf/lpm_trie.c @@ -720,6 +720,16 @@ static int trie_check_btf(const struct bpf_map *map, -EINVAL : 0; } +static u64 trie_mem_usage(const struct bpf_map *map) +{ + struct lpm_trie *trie = container_of(map, struct lpm_trie, map); + u64 elem_size; + + elem_size = sizeof(struct lpm_trie_node) + trie->data_size + + trie->map.value_size; + return elem_size * trie->n_entries; +} + BTF_ID_LIST_SINGLE(trie_map_btf_ids, struct, lpm_trie) const struct bpf_map_ops trie_map_ops = { .map_meta_equal = bpf_map_meta_equal, @@ -733,5 +743,6 @@ static int trie_check_btf(const struct bpf_map *map, .map_update_batch = generic_map_update_batch, .map_delete_batch = generic_map_delete_batch, .map_check_btf = trie_check_btf, + .map_mem_usage = trie_mem_usage, .map_btf_id = &trie_map_btf_ids[0], };
trie_mem_usage() is introduced to calculate the lpm_trie memory usage. Some small memory allocations are ignored. The inner node is also ignored. The result as follows, - before 10: lpm_trie flags 0x1 key 8B value 8B max_entries 65536 memlock 1048576B - after 10: lpm_trie flags 0x1 key 8B value 8B max_entries 65536 memlock 2291536B Signed-off-by: Yafang Shao <laoar.shao@gmail.com> --- kernel/bpf/lpm_trie.c | 11 +++++++++++ 1 file changed, 11 insertions(+)