[PATCH v2 0/2] Add support for BPF

[PATCH v2 0/2] Add support for BPF

[Richard Henderson]

[-- Attachment #1: Type: text/plain, Size: 2273 bytes --]

Changes since v2:
 (1) We've got an official number for EM_BPF,
 (2) A disassembly test case,
 (3) Added a configure test for <linux/bpf.h>,
 (4) Byte swapping of bpf_insns during disassembly,
 (5) ChangeLog entries.

I've tested 2-4 with a ppc64 Fedora 20 host, which doesn't natively
have <linux/bpf.h>.  I then forced configure to find the header I
copied over to the machine and re-tested to see that the little-endian
disassembly test case works properly with the big-endian host.


r~


Richard Henderson (2):
  Sync elf.h from glibc
  Add support for BPF

 ChangeLog                          |   4 +
 backends/ChangeLog                 |   9 ++
 backends/Makefile.am               |  10 +-
 backends/bpf_init.c                |  60 ++++++++
 backends/bpf_regs.c                |  64 +++++++++
 backends/bpf_reloc.def             |  31 ++++
 backends/common-reloc.c            |   7 +-
 configure.ac                       |   4 +
 libasm/ChangeLog                   |   4 +
 libasm/disasm_cb.c                 |   4 +-
 libcpu/ChangeLog                   |   7 +
 libcpu/Makefile.am                 |   6 +
 libcpu/bpf_disasm.c                | 288 +++++++++++++++++++++++++++++++++++++
 libcpu/i386_disasm.c               |   3 +-
 libebl/ChangeLog                   |   5 +
 libebl/ebl-hooks.h                 |   2 +-
 libebl/eblopenbackend.c            |   1 +
 libelf/ChangeLog                   |   4 +
 libelf/elf.h                       | 281 +++++++++++++++++++++++++-----------
 src/ChangeLog                      |   4 +
 src/elflint.c                      |   2 +-
 tests/ChangeLog                    |   9 ++
 tests/Makefile.am                  |   7 +-
 tests/run-disasm-bpf.sh            |  63 ++++++++
 tests/testfile-bpf-dis1.expect.bz2 | Bin 0 -> 1497 bytes
 tests/testfile-bpf-dis1.o.bz2      | Bin 0 -> 737 bytes
 26 files changed, 786 insertions(+), 93 deletions(-)
 create mode 100644 backends/bpf_init.c
 create mode 100644 backends/bpf_regs.c
 create mode 100644 backends/bpf_reloc.def
 create mode 100644 libcpu/bpf_disasm.c
 create mode 100755 tests/run-disasm-bpf.sh
 create mode 100644 tests/testfile-bpf-dis1.expect.bz2
 create mode 100644 tests/testfile-bpf-dis1.o.bz2

-- 
2.5.5

Re: [PATCH v2 0/2] Add support for BPF

[Mark Wielaard]

[-- Attachment #1: Type: text/plain, Size: 224 bytes --]

On Wed, 2016-06-22 at 08:57 -0700, Richard Henderson wrote:
> On 06/22/2016 07:54 AM, Mark Wielaard wrote:
> > Lets push this with that small tweak as soon as the glibc elf.h lands.
> 
> Ok.

Pushed!

Thanks,

Mark

Re: [PATCH v2 0/2] Add support for BPF

[Richard Henderson]

[-- Attachment #1: Type: text/plain, Size: 437 bytes --]

On 06/22/2016 07:54 AM, Mark Wielaard wrote:
> Only the disasm hook (which is already guarded) depends on cpu_bpf.
> Since that is only build when we have bpf.h this also needs a guard:
> 
> if HAVE_LINUX_BPF_H
> cpu_bpf = ../libcpu/libcpu_bpf.a
> else
> cpu_bpf = 
> endif

Good catch.  Not sure why I didn't see a problem with that...

> Lets push this with that small tweak as soon as the glibc elf.h lands.

Ok.


r~

Re: [PATCH v2 0/2] Add support for BPF

[Mark Wielaard]

[-- Attachment #1: Type: text/plain, Size: 1094 bytes --]

On Tue, 2016-06-21 at 15:10 -0700, Richard Henderson wrote:
> Changes since v2:
>  (1) We've got an official number for EM_BPF,
>  (2) A disassembly test case,
>  (3) Added a configure test for <linux/bpf.h>,
>  (4) Byte swapping of bpf_insns during disassembly,
>  (5) ChangeLog entries.

This all looks very good. Only one small build infrastructure thingy:

diff --git a/backends/Makefile.am b/backends/Makefile.am
> @@ -118,6 +118,11 @@ m68k_SRCS = m68k_init.c m68k_symbol.c m68k_regs.c \
>  libebl_m68k_pic_a_SOURCES = $(m68k_SRCS)
>  am_libebl_m68k_pic_a_OBJECTS = $(m68k_SRCS:.c=.os)
>  
> +bpf_SRCS = bpf_init.c bpf_regs.c
> +cpu_bpf = ../libcpu/libcpu_bpf.a
> +libebl_bpf_pic_a_SOURCES = $(bpf_SRCS)
> +am_libebl_bpf_pic_a_OBJECTS = $(bpf_SRCS:.c=.os)

Only the disasm hook (which is already guarded) depends on cpu_bpf.
Since that is only build when we have bpf.h this also needs a guard:

if HAVE_LINUX_BPF_H
cpu_bpf = ../libcpu/libcpu_bpf.a
else
cpu_bpf = 
endif

Lets push this with that small tweak as soon as the glibc elf.h lands.

Thanks,

Mark