On Fri, Aug 18, 2023 at 02:37:41PM -0400, David Marchevsky wrote:
> On 8/14/23 1:28 PM, Yonghong Song wrote:
> > BPF_KPTR_PERCPU_REF represents a percpu field type like below
> >
> > struct val_t {
> > ... fields ...
> > };
> > struct t {
> > ...
> > struct val_t __percpu *percpu_data_ptr;
> > ...
> > };
> >
> > where
> > #define __percpu __attribute__((btf_type_tag("percpu")))
>
> nit: Maybe this should be __percpu_kptr (and similar for the actual tag)?
+1.
I think it might conflict with kernel:
include/linux/compiler_types.h:# define __percpu BTF_TYPE_TAG(percpu)
It's the same tag name, but the kernel semantics are different from our kptr
semantics inside bpf prog.
I think we have to use a different tag like:
#define __percpu_kptr __attribute__((btf_type_tag("percpu_kptr")))
> > index 60e80e90c37d..e6348fd0a785 100644
> > --- a/include/linux/bpf.h
> > +++ b/include/linux/bpf.h
> > @@ -180,14 +180,15 @@ enum btf_field_type {
> > BPF_TIMER = (1 << 1),
> > BPF_KPTR_UNREF = (1 << 2),
> > BPF_KPTR_REF = (1 << 3),
> > - BPF_KPTR = BPF_KPTR_UNREF | BPF_KPTR_REF,
> > - BPF_LIST_HEAD = (1 << 4),
> > - BPF_LIST_NODE = (1 << 5),
> > - BPF_RB_ROOT = (1 << 6),
> > - BPF_RB_NODE = (1 << 7),
> > + BPF_KPTR_PERCPU_REF = (1 << 4),
I think _REF is redundant here. _UNREF is obsolete. We might remove it and
rename BPF_KPTR_REF to just BPF_KPTR.
BPF_KPTR_PERCPU should be clear enough.
