Andrii Nakryiko wrote:
> Refactor internals of struct btf to remove assumptions that BTF header, type
> data, and string data are layed out contiguously in a memory in a single
> memory allocation. Now we have three separate pointers pointing to the start
> of each respective are: header, types, strings. In the next patches, these
> pointers will be re-assigned to point to independently allocated memory areas,
> if BTF needs to be modified.
>
> Signed-off-by: Andrii Nakryiko <andr...@fb.com>
> ---
[...]
> --- a/tools/lib/bpf/btf.c
> +++ b/tools/lib/bpf/btf.c
> @@ -27,18 +27,37 @@
> static struct btf_type btf_void;
>
> struct btf {
> - union {
> - struct btf_header *hdr;
> - void *data;
> - };
> + void *raw_data;
> + __u32 raw_size;
> +
> + /*
> + * When BTF is loaded from ELF or raw memory it is stored
> + * in contiguous memory block, pointed to by raw_data pointer, and
> + * hdr, types_data, and strs_data point inside that memory region to
> + * respective parts of BTF representation:
I find the above comment a bit confusing. The picture though is great. How
about something like,
When BTF is loaded from an ELF or raw memory it is stored
in a continguous memory block. The hdr, type_data, and strs_data
point inside that memory region to their respective parts of BTF
representation
> + *
> + * +--------------------------------+
> + * | Header | Types | Strings |
> + * +--------------------------------+
> + * ^ ^ ^
> + * | | |
> + * hdr | |
> + * types_data-+ |
> + * strs_data------------+
> + */
> + struct btf_header *hdr;
> + void *types_data;
> + void *strs_data;
> +
> + /* type ID to `struct btf_type *` lookup index */
> __u32 *type_offs;
> __u32 type_offs_cap;
> - const char *strings;
> - void *nohdr_data;
> - void *types_data;
> __u32 nr_types;
> - __u32 data_size;
> +
> + /* BTF object FD, if loaded into kernel */
> int fd;
> +
> + /* Pointer size (in bytes) for a target architecture of this BTF */
> int ptr_sz;
> };
>
Thanks,
John
