Re: [PATCH bpf-next 1/2] btf: resolve enum fwds in btf_dedup

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, Mar 11, 2019 at 12:02 AM Yonghong Song <yhs@xxxxxx> wrote:
>
>
>
> On 3/10/19 5:44 PM, Andrii Nakryiko wrote:
> > GCC and clang support enum forward declarations as an extension. Such
> > forward-declared enums will be represented as normal BTF_KIND_ENUM types with
> > vlen=0. This patch adds ability to resolve such enums to their corresponding
>
> Thanks Andrii for reporting and trying to fix forward-declared enum
> issue. The current approach to represent forward enum with
> BTF_KIND_ENUM with vlen = 0 does work. "enum A {};" is illegal for C,
> so vlen=0 must be a forward declaration.
>
> But this is probably not the best since we have BTF_KIND_FWD to make
> forward declaration explicit.

Yeah, that's true. See below, though, why for enum it's a bit inconvenient.

In retrospect, I'm wondering if having a special "declaration vs
definition" bit in btf_type->info for struct/union would be better
approach, kind of like DWARF's DW_AT_declaration attribute. But it's a
bit late now.

>
> Currently BTF_KIND_FWD uses (btf_type.info >> 31) to distinguish
> struct/union. If btf_type.info >> 31 == 0, it is a struct, otherwise
> it is a union.
>
> We could extend this to cover forward enum as well with backward
> compability. We can extend by using both bit 30 and bit 31:
>    btf_type.info >> 31   (btf_type.info >> 30) & 0x1
>          0                  0                  <== struct
>          1                  0                  <== union
>          0                  1                  <== enum
>
> Maybe this may be reasonable?

The only thing we should consider is that enums can have different
sizes. And enum size is part of enum's forward declaration. So unlike
struct/union fwd, enum's fwd has extra info. I don't think it's
possible to specify in C (enum is always 4 bytes), but C++ definitely
allows to specify custom underlying integer type.

So I don't know, do you think we should just ignore size and go with
BTF_KIND_FWD?

>
> This will require pahole and llvm change to support, besides kernel
> support. I will try to prototype llvm support tomorrow.
>
>
> > fully defined enums. This helps to avoid duplicated BTF type graphs which only
> > differ by some types referencing forward-declared enum vs full enum.
> >
> > One such example in kernel is enum irqchip_irq_state, defined in
> > include/linux/interrupt.h and forward-declared in include/linux/irq.h. This
> > causes entire struct task_struct and all referenced types to be duplicated in
> > btf_dedup output. This patch eliminates such duplication cases.
> >
> > Signed-off-by: Andrii Nakryiko <andriin@xxxxxx>
> > ---
> >   tools/lib/bpf/btf.c | 51 +++++++++++++++++++++++++++++++++------------
> >   1 file changed, 38 insertions(+), 13 deletions(-)
> >
> > diff --git a/tools/lib/bpf/btf.c b/tools/lib/bpf/btf.c
> > index 1b8d8cdd3575..87e3020ac1bc 100644
> > --- a/tools/lib/bpf/btf.c
> > +++ b/tools/lib/bpf/btf.c
> > @@ -1602,16 +1602,12 @@ static bool btf_equal_int(struct btf_type *t1, struct btf_type *t2)
> >   /* Calculate type signature hash of ENUM. */
> >   static __u32 btf_hash_enum(struct btf_type *t)
> >   {
> > -     struct btf_enum *member = (struct btf_enum *)(t + 1);
> > -     __u32 vlen = BTF_INFO_VLEN(t->info);
> > -     __u32 h = btf_hash_common(t);
> > -     int i;
> > +     __u32 h;
> >
> > -     for (i = 0; i < vlen; i++) {
> > -             h = hash_combine(h, member->name_off);
> > -             h = hash_combine(h, member->val);
> > -             member++;
> > -     }
> > +     /* don't hash vlen and enum members to support enum fwd resolving */
> > +     h = hash_combine(0, t->name_off);
> > +     h = hash_combine(h, t->info & ~0xffff);
> > +     h = hash_combine(h, t->size);
> >       return h;
> >   }
> >
> > @@ -1637,6 +1633,22 @@ static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
> >       return true;
> >   }
> >
> > +static inline bool btf_is_enum_fwd(struct btf_type *t)
> > +{
> > +     return BTF_INFO_KIND(t->info) == BTF_KIND_ENUM &&
> > +            BTF_INFO_VLEN(t->info) == 0;
> > +}
> > +
> > +static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
> > +{
> > +     if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
> > +             return btf_equal_enum(t1, t2);
> > +     /* ignore vlen when comparing */
> > +     return t1->name_off == t2->name_off &&
> > +            (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
> > +            t1->size == t2->size;
> > +}
> > +
> >   /*
> >    * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
> >    * as referenced type IDs equivalence is established separately during type
> > @@ -1860,6 +1872,17 @@ static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
> >                               new_id = cand_node->type_id;
> >                               break;
> >                       }
> > +                     if (d->opts.dont_resolve_fwds)
> > +                             continue;
> > +                     if (btf_compat_enum(t, cand)) {
> > +                             if (btf_is_enum_fwd(t)) {
> > +                                     /* resolve fwd to full enum */
> > +                                     new_id = cand_node->type_id;
> > +                                     break;
> > +                             }
> > +                             /* resolve canonical enum fwd to full enum */
> > +                             d->map[cand_node->type_id] = type_id;
> > +                     }
> >               }
> >               break;
> >
> > @@ -2084,15 +2107,15 @@ static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
> >               return fwd_kind == real_kind;
> >       }
> >
> > -     if (cand_type->info != canon_type->info)
> > -             return 0;
> > -
> >       switch (cand_kind) {
> >       case BTF_KIND_INT:
> >               return btf_equal_int(cand_type, canon_type);
> >
> >       case BTF_KIND_ENUM:
> > -             return btf_equal_enum(cand_type, canon_type);
> > +             if (d->opts.dont_resolve_fwds)
> > +                     return btf_equal_enum(cand_type, canon_type);
> > +             else
> > +                     return btf_compat_enum(cand_type, canon_type);
> >
> >       case BTF_KIND_FWD:
> >               return btf_equal_common(cand_type, canon_type);
> > @@ -2103,6 +2126,8 @@ static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
> >       case BTF_KIND_PTR:
> >       case BTF_KIND_TYPEDEF:
> >       case BTF_KIND_FUNC:
> > +             if (cand_type->info != canon_type->info)
> > +                     return 0;
> >               return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
> >
> >       case BTF_KIND_ARRAY: {
> >



[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux