[DTrace-devel] [PATCH] bpf, btf: store BTF dict id for kernel modules
Nick Alcock
nick.alcock at oracle.com
Fri Apr 26 11:37:48 UTC 2024
On 25 Apr 2024, Kris Van Hees said:
> BPF requires a file descriptor (provided by the kernel) to be associated
> with BTF type ids for loadble kernel modules. This file descriptor can
s/loadble/loadable/
> only be obtained from the kernel based on a BTF dict id for the module,
> and these ids are assigned dynamically at runtime when modules are
> loaded.
... were they *trying* to make this unnecessarily painful to use?
> Signed-off-by: Kris Van Hees <kris.van.hees at oracle.com>
Reviewed-by: Nick Alcock <nick.alcock at oracle.com>
with a few niggles
> diff --git a/libdtrace/dt_btf.h b/libdtrace/dt_btf.h
> index 0146386b..60256f12 100644
> --- a/libdtrace/dt_btf.h
> +++ b/libdtrace/dt_btf.h
> @@ -15,7 +15,8 @@
> extern "C" {
> #endif
>
> -typedef struct dt_btf dt_btf_t;
> +typedef struct dt_btf dt_btf_t;
> +typedef struct bpf_btf_info btf_info_t;
This sort of thing will start throwing compiler errors whenever any user
#includes the header that gives the concrete definition of any of those
types, because we're probably typedeffing them there too, and unlike
struct definitions you cannot repeat typedefs.
(Not a problem now, but just noting that this sort of pattern tends to
store up trouble for the future.)
> @@ -339,6 +340,8 @@ struct dtrace_hdl {
> char *dt_ctfa_path; /* path to vmlinux.ctfa */
> ctf_archive_t *dt_ctfa; /* ctf archive for the entire kernel tree */
> struct dt_btf *dt_shared_btf; /* BTF data for the kernel (shared) */
> + int dt_btf_cnt; /* BTF id count */
> + int *dt_btf_fds; /* BTF id-to-fd mapping */
> ctf_file_t *dt_shared_ctf; /* Handle to the shared CTF */
> dt_htab_t *dt_kernpaths; /* hash table of dt_kern_path_t's */
> dt_module_t *dt_exec; /* pointer to executable module */
These don't seem to be used anywhere in this patch: did you swap between
storing globally and storing per-module?
--
NULL && (void)
More information about the DTrace-devel
mailing list