[bpf-next,06/10] bpf: Add bpf_mem_cache_is_mergeable() helper

Commit Message

Hou Tao Nov. 18, 2024, 1:08 a.m. UTC

From: Hou Tao <[e-mail]>

Add bpf_mem_cache_is_mergeable() to check whether two bpf mem allocator
for fixed-size objects are mergeable or not. The merging could reduce
the memory overhead of bpf mem allocator.

Signed-off-by: Hou Tao <[e-mail]>
---
 include/linux/bpf_mem_alloc.h |  1 +
 kernel/bpf/memalloc.c         | 12 ++++++++++++
 2 files changed, 13 insertions(+)

Comments

Thomas Weißschuh Nov. 18, 2024, 1:29 p.m. UTC | #1

On Mon, Nov 18, 2024 at 09:08:04AM +0800, Hou Tao wrote:
> From: Hou Tao <[e-mail]>
> 
> Add bpf_mem_cache_is_mergeable() to check whether two bpf mem allocator
> for fixed-size objects are mergeable or not. The merging could reduce
> the memory overhead of bpf mem allocator.
> 
> Signed-off-by: Hou Tao <[e-mail]>
> ---
>  include/linux/bpf_mem_alloc.h |  1 +
>  kernel/bpf/memalloc.c         | 12 ++++++++++++
>  2 files changed, 13 insertions(+)
> 
> diff --git a/include/linux/bpf_mem_alloc.h b/include/linux/bpf_mem_alloc.h
> index e45162ef59bb..faa54b9c7a04 100644
> --- a/include/linux/bpf_mem_alloc.h
> +++ b/include/linux/bpf_mem_alloc.h
> @@ -47,5 +47,6 @@ void bpf_mem_cache_free(struct bpf_mem_alloc *ma, void *ptr);
>  void bpf_mem_cache_free_rcu(struct bpf_mem_alloc *ma, void *ptr);
>  void bpf_mem_cache_raw_free(void *ptr);
>  void *bpf_mem_cache_alloc_flags(struct bpf_mem_alloc *ma, gfp_t flags);
> +bool bpf_mem_cache_is_mergeable(size_t size, size_t new_size, bool percpu);
>  
>  #endif /* _BPF_MEM_ALLOC_H */
> diff --git a/kernel/bpf/memalloc.c b/kernel/bpf/memalloc.c
> index 889374722d0a..49dd08ad1d4f 100644
> --- a/kernel/bpf/memalloc.c
> +++ b/kernel/bpf/memalloc.c
> @@ -1014,3 +1014,15 @@ int bpf_mem_alloc_check_size(bool percpu, size_t size)
>  
>  	return 0;
>  }
> +
> +bool bpf_mem_cache_is_mergeable(size_t size, size_t new_size, bool percpu)
> +{
> +	/* Only for fixed-size object allocator */
> +	if (!size || !new_size)
> +		return false;
> +
> +	return (percpu && ALIGN(size, PCPU_MIN_ALLOC_SIZE) ==
> +			  ALIGN(new_size, PCPU_MIN_ALLOC_SIZE)) ||
> +	       (!percpu && kmalloc_size_roundup(size + LLIST_NODE_SZ) ==
> +			   kmalloc_size_roundup(new_size + LLIST_NODE_SZ));

This would be easier to read:

if (percpu)
	return ALIGN() == ALIGN();
else
	return kmalloc_size_roundup() == kmalloc_size_roundup();

> +}
> -- 
> 2.29.2
>

Hou Tao Nov. 19, 2024, 1:06 a.m. UTC | #2

Hi,

On 11/18/2024 9:29 PM, Thomas Weißschuh wrote:
> On Mon, Nov 18, 2024 at 09:08:04AM +0800, Hou Tao wrote:
>> From: Hou Tao <[e-mail]>
>>
>> Add bpf_mem_cache_is_mergeable() to check whether two bpf mem allocator
>> for fixed-size objects are mergeable or not. The merging could reduce
>> the memory overhead of bpf mem allocator.
>>
>> Signed-off-by: Hou Tao <[e-mail]>
>> ---
>>  include/linux/bpf_mem_alloc.h |  1 +
>>  kernel/bpf/memalloc.c         | 12 ++++++++++++
>>  2 files changed, 13 insertions(+)
>>
>> diff --git a/include/linux/bpf_mem_alloc.h b/include/linux/bpf_mem_alloc.h
>> index e45162ef59bb..faa54b9c7a04 100644
>> --- a/include/linux/bpf_mem_alloc.h
>> +++ b/include/linux/bpf_mem_alloc.h
>> @@ -47,5 +47,6 @@ void bpf_mem_cache_free(struct bpf_mem_alloc *ma, void *ptr);
>>  void bpf_mem_cache_free_rcu(struct bpf_mem_alloc *ma, void *ptr);
>>  void bpf_mem_cache_raw_free(void *ptr);
>>  void *bpf_mem_cache_alloc_flags(struct bpf_mem_alloc *ma, gfp_t flags);
>> +bool bpf_mem_cache_is_mergeable(size_t size, size_t new_size, bool percpu);
>>  
>>  #endif /* _BPF_MEM_ALLOC_H */
>> diff --git a/kernel/bpf/memalloc.c b/kernel/bpf/memalloc.c
>> index 889374722d0a..49dd08ad1d4f 100644
>> --- a/kernel/bpf/memalloc.c
>> +++ b/kernel/bpf/memalloc.c
>> @@ -1014,3 +1014,15 @@ int bpf_mem_alloc_check_size(bool percpu, size_t size)
>>  
>>  	return 0;
>>  }
>> +
>> +bool bpf_mem_cache_is_mergeable(size_t size, size_t new_size, bool percpu)
>> +{
>> +	/* Only for fixed-size object allocator */
>> +	if (!size || !new_size)
>> +		return false;
>> +
>> +	return (percpu && ALIGN(size, PCPU_MIN_ALLOC_SIZE) ==
>> +			  ALIGN(new_size, PCPU_MIN_ALLOC_SIZE)) ||
>> +	       (!percpu && kmalloc_size_roundup(size + LLIST_NODE_SZ) ==
>> +			   kmalloc_size_roundup(new_size + LLIST_NODE_SZ));
> This would be easier to read:
>
> if (percpu)
> 	return ALIGN() == ALIGN();
> else
> 	return kmalloc_size_roundup() == kmalloc_size_roundup();

Indeed. Thanks for the suggestion. Will do in v2.
>> +}
>> -- 
>> 2.29.2
>>

Patch

diff --git a/include/linux/bpf_mem_alloc.h b/include/linux/bpf_mem_alloc.h
index e45162ef59bb..faa54b9c7a04 100644
--- a/include/linux/bpf_mem_alloc.h
+++ b/include/linux/bpf_mem_alloc.h
@@ -47,5 +47,6 @@  void bpf_mem_cache_free(struct bpf_mem_alloc *ma, void *ptr);
 void bpf_mem_cache_free_rcu(struct bpf_mem_alloc *ma, void *ptr);
 void bpf_mem_cache_raw_free(void *ptr);
 void *bpf_mem_cache_alloc_flags(struct bpf_mem_alloc *ma, gfp_t flags);
+bool bpf_mem_cache_is_mergeable(size_t size, size_t new_size, bool percpu);
 
 #endif /* _BPF_MEM_ALLOC_H */
diff --git a/kernel/bpf/memalloc.c b/kernel/bpf/memalloc.c
index 889374722d0a..49dd08ad1d4f 100644
--- a/kernel/bpf/memalloc.c
+++ b/kernel/bpf/memalloc.c
@@ -1014,3 +1014,15 @@  int bpf_mem_alloc_check_size(bool percpu, size_t size)
 
 	return 0;
 }
+
+bool bpf_mem_cache_is_mergeable(size_t size, size_t new_size, bool percpu)
+{
+	/* Only for fixed-size object allocator */
+	if (!size || !new_size)
+		return false;
+
+	return (percpu && ALIGN(size, PCPU_MIN_ALLOC_SIZE) ==
+			  ALIGN(new_size, PCPU_MIN_ALLOC_SIZE)) ||
+	       (!percpu && kmalloc_size_roundup(size + LLIST_NODE_SZ) ==
+			   kmalloc_size_roundup(new_size + LLIST_NODE_SZ));
+}