Skip to contentSkip to conversation

bpf: Reduce smap->elem_size

PATCH Superseded Martin KaFai Lau 3 participants 4 messages Dec 16, 2022 – Dec 21, 2022 2 messages not in the archive BPF

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Martin KaFai Lau

author reviewer
#0 lore raw

PATCH 1/1 bpf: Reduce smap->elem_size

From: Martin KaFai Lau <[email protected]>

'struct bpf_local_storage_elem' has a 56 bytes padding at the end which can be used for attr->value_size. The current smap->elem_size calculation is unnecessarily inflated by 56 bytes.

The patch is to fix it by calculating the smap->elem_size with offsetof().

---
 kernel/bpf/bpf_local_storage.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
kernel/bpf/bpf_local_storage.c +2 -2
index b39a46e8fb08..cb43e70613b1 100644
@@ -580,8 +580,8 @@ static struct bpf_local_storage_map *__bpf_local_storage_map_alloc(union bpf_att
 		raw_spin_lock_init(&smap->buckets[i].lock);
 	}
 
-	smap->elem_size =
-		sizeof(struct bpf_local_storage_elem) + attr->value_size;
+	smap->elem_size = offsetof(struct bpf_local_storage_elem, sdata) +
+		offsetof(struct bpf_local_storage_data, data[attr->value_size]);
 
 	return smap;
 }
Signature · 1 line
2.30.2
Acked-by on v1 1/1
#1 lore raw
Quoted 4 lines · Martin KaFai Lau

From: Martin KaFai Lau <[email protected]>

'struct bpf_local_storage_elem' has a 56 bytes padding at the end which can be used for attr->value_size. The current smap->elem_size

'can be' => 'will be'?

Quoted 6 lines

calculation is unnecessarily inflated by 56 bytes.

The patch is to fix it by calculating the smap->elem_size with offsetof().

Quoted 19 lines
---
  kernel/bpf/bpf_local_storage.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
kernel/bpf/bpf_local_storage.c +2 -2
index b39a46e8fb08..cb43e70613b1 100644
@@ -580,8 +580,8 @@ static struct bpf_local_storage_map *__bpf_local_storage_map_alloc(union bpf_att
  		raw_spin_lock_init(&smap->buckets[i].lock);
  	}
 
-	smap->elem_size =
-		sizeof(struct bpf_local_storage_elem) + attr->value_size;
+	smap->elem_size = offsetof(struct bpf_local_storage_elem, sdata) +
+		offsetof(struct bpf_local_storage_data, data[attr->value_size]);
 
  	return smap;
  }

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): CAEf4BzbJGpkhyio9+S1U=bnYycaknw0SNada6orzNV_+VfPwGw@mail.gmail.com

author reviewer
#4 lore raw

[ Cc: the bpf list back. I dropped it by mistake in my last reply. ]

Quoted 24 lines · Andrii Nakryiko
Quoted 12 lines · Martin KaFai Lau
Quoted 9 lines · Yonghong Song
Quoted 4 lines · Martin KaFai Lau

From: Martin KaFai Lau <[email protected]>

'struct bpf_local_storage_elem' has a 56 bytes padding at the end which can be used for attr->value_size. The current smap->elem_size

'can be' => 'will be'?

Quoted 7 lines

I used 'can be' to describe the current situation that the padding is not used for the map's value. I may have used the wrong tense?

I can rephrase it to something like,

'struct bpf_local_storage_elem' has a 56 bytes padding at the end which is currently not used for the attr->value_size.

I actually found the use of attr->value_size to mean "value content" more confusing than can vs will be.

Quoted 8 lines

As a suggestion something like the below?

'struct bpf_local_storage_elem' has an unused 56 byte padding at the end due to struct's cache-line alignment requirement. This padding space is overlapped by storage value contents, so if we use sizeof() to calculate the total size, we overinflate it by 56 bytes. Use offsetof() instead to calculate more exact memory use.

SGTM.

Quoted 7 lines

btw, I think you can calculate the same arguably a bit more straightforwardly as:

smap->elem_size = offsetofend(struct bpf_local_storage_elem, sdata) + attr->value_size;

Sure. will change.

Quoted 5 lines

right?

but TIL that offsetof(struct bpf_local_storage_data, data[attr->value_size]) does the right thing

Yeah, I think I have seen other places using it also. I found it more intuitive to read with array[size] to tell there is a flexible array at the end. I am not super attached to it and will change to the way above.

#5 lore raw
Quoted 50 lines · Martin KaFai Lau

[ Cc: the bpf list back. I dropped it by mistake in my last reply. ]

Quoted 24 lines · Andrii Nakryiko
Quoted 12 lines · Martin KaFai Lau
Quoted 9 lines · Yonghong Song
Quoted 4 lines · Martin KaFai Lau

From: Martin KaFai Lau <[email protected]>

'struct bpf_local_storage_elem' has a 56 bytes padding at the end which can be used for attr->value_size. The current smap->elem_size

'can be' => 'will be'?

Quoted 7 lines

I used 'can be' to describe the current situation that the padding is not used for the map's value. I may have used the wrong tense?

I can rephrase it to something like,

'struct bpf_local_storage_elem' has a 56 bytes padding at the end which is currently not used for the attr->value_size.

I actually found the use of attr->value_size to mean "value content" more confusing than can vs will be.

Quoted 8 lines

As a suggestion something like the below?

'struct bpf_local_storage_elem' has an unused 56 byte padding at the end due to struct's cache-line alignment requirement. This padding space is overlapped by storage value contents, so if we use sizeof() to calculate the total size, we overinflate it by 56 bytes. Use offsetof() instead to calculate more exact memory use.

SGTM.

Quoted 7 lines

btw, I think you can calculate the same arguably a bit more straightforwardly as:

smap->elem_size = offsetofend(struct bpf_local_storage_elem, sdata) + attr->value_size;

Sure. will change.

Quoted 1 line
Quoted 9 lines
Quoted 4 lines

right?

but TIL that offsetof(struct bpf_local_storage_data, data[attr->value_size]) does the right thing

Yeah, I think I have seen other places using it also. I found it more intuitive to read with array[size] to tell there is a flexible array at the end. I am not super attached to it and will change to the way above.

I don't care either, was just surprised. But it just occurred to me that your change can be written equivalently (but now I do think it's cleaner) as:

smap->elem_size = offsetof(struct bpf_local_storage_elem, sdata.data[attr->value_size]);

sdata is embedded struct, no pointer dereference, so single offsetof() should be enough to peer inside it