[PATCH bpf] libbpf: Handle GCC built-in types for Arm NEON

Andrii Nakryiko andrii.nakryiko at gmail.com
Tue Aug 11 23:30:06 EDT 2020


On Mon, Aug 10, 2020 at 5:41 AM Jean-Philippe Brucker
<jean-philippe at linaro.org> wrote:
>
> When building Arm NEON (SIMD) code, GCC emits built-in types __PolyXX_t,
> which are not recognized by Clang. This causes build failures when
> including vmlinux.h generated from a kernel built with CONFIG_RAID6_PQ=y
> and CONFIG_KERNEL_MODE_NEON. Emit typedefs for these built-in types,
> based on the Clang definitions. poly64_t is unsigned long because it's
> only defined for 64-bit Arm.
>
> Including linux/kernel.h to use ARRAY_SIZE() incidentally redefined
> max(), causing a build bug due to different types, hence the seemingly
> unrelated change.
>
> Reported-by: Jakov Petrina <jakov.petrina at sartura.hr>
> Signed-off-by: Jean-Philippe Brucker <jean-philippe at linaro.org>
> ---

Thanks for sending small binaries (in little-endian, double thanks!)
for me to look at generated DWARF and BTF.

I have a bunch of naming nits below and a grudge against "long", but
the approach looks reasonable to me overall. It's unfortunate we have
to deal with GCC quirks like this.

>  tools/lib/bpf/btf_dump.c | 35 ++++++++++++++++++++++++++++++++++-
>  1 file changed, 34 insertions(+), 1 deletion(-)
>
> diff --git a/tools/lib/bpf/btf_dump.c b/tools/lib/bpf/btf_dump.c
> index cf711168d34a..3162d7b1880c 100644
> --- a/tools/lib/bpf/btf_dump.c
> +++ b/tools/lib/bpf/btf_dump.c
> @@ -13,6 +13,7 @@
>  #include <errno.h>
>  #include <linux/err.h>
>  #include <linux/btf.h>
> +#include <linux/kernel.h>
>  #include "btf.h"
>  #include "hashmap.h"
>  #include "libbpf.h"
> @@ -549,6 +550,9 @@ static int btf_dump_order_type(struct btf_dump *d, __u32 id, bool through_ptr)
>         }
>  }
>
> +static void btf_dump_emit_int_def(struct btf_dump *d, __u32 id,
> +                                 const struct btf_type *t);
> +
>  static void btf_dump_emit_struct_fwd(struct btf_dump *d, __u32 id,
>                                      const struct btf_type *t);
>  static void btf_dump_emit_struct_def(struct btf_dump *d, __u32 id,
> @@ -671,6 +675,9 @@ static void btf_dump_emit_type(struct btf_dump *d, __u32 id, __u32 cont_id)
>
>         switch (kind) {
>         case BTF_KIND_INT:
> +               /* Emit type alias definitions if necessary */
> +               btf_dump_emit_int_def(d, id, t);

let's call it btf_dump_emit_missing_aliases() or something like that,
so it's clear that it's some sort of compatibility/legacy compiler
handling. "emit_int_def" is way too generic and normal-looking.

> +
>                 tstate->emit_state = EMITTED;
>                 break;
>         case BTF_KIND_ENUM:
> @@ -870,7 +877,7 @@ static void btf_dump_emit_struct_def(struct btf_dump *d,
>                         btf_dump_printf(d, ": %d", m_sz);
>                         off = m_off + m_sz;
>                 } else {
> -                       m_sz = max(0, btf__resolve_size(d->btf, m->type));
> +                       m_sz = max(0LL, btf__resolve_size(d->btf, m->type));
>                         off = m_off + m_sz * 8;
>                 }
>                 btf_dump_printf(d, ";");
> @@ -890,6 +897,32 @@ static void btf_dump_emit_struct_def(struct btf_dump *d,
>                 btf_dump_printf(d, " __attribute__((packed))");
>  }
>
> +static const char *builtin_types[][2] = {

again, something like "missing_base_types" would be a bit more prominent

> +       /*
> +        * GCC emits typedefs to its internal __PolyXX_t types when compiling
> +        * Arm SIMD intrinsics. Alias them to the same standard types as Clang.
> +        */
> +       { "__Poly8_t",          "unsigned char" },
> +       { "__Poly16_t",         "unsigned short" },
> +       { "__Poly64_t",         "unsigned long" },

In the diff ([0]) that Daniel referenced, seems like they are adding
poly64_t to ARM32. What prevents GCC from doing that (or maybe they've
already done that). So instead of making unreliable assumptions, let's
define it as "unsigned long long" instead?

  [0] https://reviews.llvm.org/D79711

> +       { "__Poly128_t",        "unsigned __int128" },
> +};
> +
> +static void btf_dump_emit_int_def(struct btf_dump *d, __u32 id,
> +                                 const struct btf_type *t)
> +{
> +       const char *name = btf_dump_type_name(d, id);
> +       int i;
> +
> +       for (i = 0; i < ARRAY_SIZE(builtin_types); i++) {
> +               if (strcmp(name, builtin_types[i][0]) == 0) {
> +                       btf_dump_printf(d, "typedef %s %s;\n\n",
> +                                       builtin_types[i][1], name);
> +                       break;
> +               }
> +       }
> +}
> +
>  static void btf_dump_emit_enum_fwd(struct btf_dump *d, __u32 id,
>                                    const struct btf_type *t)
>  {
> --
> 2.27.0
>



More information about the linux-arm-kernel mailing list