[PATCH 0/1] objdump: Add callx support for BPF CPU v1

[PATCH 0/1] objdump: Add callx support for BPF CPU v1

[Will Hawkins]

Hello everyone!

First time patch submitter but longtime GNU enthusiast. Thank you for
all the work that you all do to maintain these vital pieces of software
for the community.

I am one of the maintainers of ubpf, a userspace eBPF runtime. As part
of the ISA standardization process at the IETF, Dave Thaler discovered
that clang is now emiting the callx instruction when targeting CPU
versions all the way back to v1. I know from Jose that gcc emits that
instruction in a special mode. 

There is a discussion at the ebpf-for-windows GitHub about adding callx
support to prevail (and other related tools):

https://github.com/microsoft/ebpf-for-windows/issues/3237

This patch adds the ability for objdump to decode the callx instruction
for ELF-encoded binaries targeting CPU v1 and greater.

I tried very hard to follow best practices for creating this patch
submission but I am sure that I did something wrong. Please let me know
what I can do differently or better.

Again, thank you for your effort maintaining these programs for the
community!

Sincerely,
Will

Will Hawkins (1):
  objdump: Add callx support for BPF CPU v1

 include/opcode/bpf.h | 2 +-
 opcodes/bpf-opc.c    | 2 ++
 sim/bpf/bpf-sim.c    | 4 ++++
 3 files changed, 7 insertions(+), 1 deletion(-)

-- 
2.43.0

[PATCH 1/1] objdump: Add callx support for BPF CPU v1

[Will Hawkins]

Add support for disassembling the callx instruction back to CPU v1.

include/ChangeLog:

	* opcode/bpf.h (enum bpf_insn_id):

opcodes/ChangeLog:

	* bpf-opc.c:

ChangeLog:

	* sim/bpf/bpf-sim.c (execute):

Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
---
 include/opcode/bpf.h | 2 +-
 opcodes/bpf-opc.c    | 2 ++
 sim/bpf/bpf-sim.c    | 4 ++++
 3 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/include/opcode/bpf.h b/include/opcode/bpf.h
index df1e3bd0918..48cef368e4e 100644
--- a/include/opcode/bpf.h
+++ b/include/opcode/bpf.h
@@ -202,7 +202,7 @@ enum bpf_insn_id
   BPF_INSN_JAR, BPF_INSN_JEQR, BPF_INSN_JGTR, BPF_INSN_JSGTR,
   BPF_INSN_JGER, BPF_INSN_JSGER, BPF_INSN_JLTR, BPF_INSN_JSLTR,
   BPF_INSN_JSLER, BPF_INSN_JLER, BPF_INSN_JSETR, BPF_INSN_JNER,
-  BPF_INSN_CALLR, BPF_INSN_CALL, BPF_INSN_EXIT,
+  BPF_INSN_CALLR, BPF_INSN_CALLX, BPF_INSN_CALL, BPF_INSN_EXIT,
   /* Compare-and-jump instructions (reg OP imm.)  */
   BPF_INSN_JEQI, BPF_INSN_JGTI, BPF_INSN_JSGTI,
   BPF_INSN_JGEI, BPF_INSN_JSGEI, BPF_INSN_JLTI, BPF_INSN_JSLTI,
diff --git a/opcodes/bpf-opc.c b/opcodes/bpf-opc.c
index 19e096501a2..3f030a39c1b 100644
--- a/opcodes/bpf-opc.c
+++ b/opcodes/bpf-opc.c
@@ -274,6 +274,8 @@ const struct bpf_opcode bpf_opcodes[] =
    BPF_V1, BPF_CODE, BPF_CLASS_JMP|BPF_CODE_JNE|BPF_SRC_X},
   {BPF_INSN_CALLR, "call%W%dr", "callx%w%dr",
    BPF_XBPF, BPF_CODE, BPF_CLASS_JMP|BPF_CODE_CALL|BPF_SRC_X},
+  {BPF_INSN_CALLX, "callx%W%dr", "callx%w%dr",
+   BPF_V1, BPF_CODE, BPF_CLASS_JMP|BPF_CODE_CALL|BPF_SRC_X},
   {BPF_INSN_CALL, "call%W%d32", "call%w%d32",
    BPF_V1, BPF_CODE, BPF_CLASS_JMP|BPF_CODE_CALL|BPF_SRC_K},
   {BPF_INSN_EXIT, "exit", "exit",
diff --git a/sim/bpf/bpf-sim.c b/sim/bpf/bpf-sim.c
index c1f103823fb..a3976d0b4bf 100644
--- a/sim/bpf/bpf-sim.c
+++ b/sim/bpf/bpf-sim.c
@@ -1096,6 +1096,10 @@ execute (SIM_CPU *cpu, struct bpf_insn *insn)
       BPF_TRACE ("BPF_INSN_CALLR\n");
       bpf_call (cpu, DISP (bpf_regs[insn->dst]), insn->src);
       break;
+    case BPF_INSN_CALLX:
+      BPF_TRACE ("BPF_INSN_CALLX\n");
+      bpf_call (cpu, DISP (bpf_regs[insn->dst]), insn->src);
+      break;
     case BPF_INSN_CALL:
       BPF_TRACE ("BPF_INSN_CALL\n");
       bpf_call (cpu, insn->imm32, insn->src);
-- 
2.43.0

Re: [PATCH 1/1] objdump: Add callx support for BPF CPU v1

[Hans-Peter Nilsson]

On Fri, 9 Feb 2024, Will Hawkins wrote:

> Add support for disassembling the callx instruction back to CPU v1.
> 
> include/ChangeLog:
> 
> 	* opcode/bpf.h (enum bpf_insn_id):
> 
> opcodes/ChangeLog:
> 
> 	* bpf-opc.c:
> 
> ChangeLog:
> 
> 	* sim/bpf/bpf-sim.c (execute):

Not commenting on the rest of the patch but there's missing text 
after ":" above. Beware: gcc/contrib/mklog.py doesn't help with 
that part. :)

Happy hacking!

brgds, H-P

Re: [PATCH 1/1] objdump: Add callx support for BPF CPU v1

[Will Hawkins]

On Fri, Feb 9, 2024 at 11:47 PM Hans-Peter Nilsson <hp@bitrange.com> wrote:
>
> On Fri, 9 Feb 2024, Will Hawkins wrote:
>
> > Add support for disassembling the callx instruction back to CPU v1.
> >
> > include/ChangeLog:
> >
> >       * opcode/bpf.h (enum bpf_insn_id):
> >
> > opcodes/ChangeLog:
> >
> >       * bpf-opc.c:
> >
> > ChangeLog:
> >
> >       * sim/bpf/bpf-sim.c (execute):
>
> Not commenting on the rest of the patch but there's missing text
> after ":" above. Beware: gcc/contrib/mklog.py doesn't help with
> that part. :)
>

Thank you! I thought something looked suspicious but because it was
what contrib/mklog.py gave me, I assumed you all had special
requirements! I will fix it and resubmit.

Thank you again!
Will



> Happy hacking!
>
> brgds, H-P

Re: [PATCH 1/1] objdump: Add callx support for BPF CPU v1

[Will Hawkins]

On Sat, Feb 10, 2024 at 1:13 AM Will Hawkins <hawkinsw@obs.cr> wrote:
>
> On Fri, Feb 9, 2024 at 11:47 PM Hans-Peter Nilsson <hp@bitrange.com> wrote:
> >
> > On Fri, 9 Feb 2024, Will Hawkins wrote:
> >
> > > Add support for disassembling the callx instruction back to CPU v1.
> > >
> > > include/ChangeLog:
> > >
> > >       * opcode/bpf.h (enum bpf_insn_id):
> > >
> > > opcodes/ChangeLog:
> > >
> > >       * bpf-opc.c:
> > >
> > > ChangeLog:
> > >
> > >       * sim/bpf/bpf-sim.c (execute):
> >
> > Not commenting on the rest of the patch but there's missing text
> > after ":" above. Beware: gcc/contrib/mklog.py doesn't help with
> > that part. :)
> >
>
> Thank you! I thought something looked suspicious but because it was
> what contrib/mklog.py gave me, I assumed you all had special
> requirements! I will fix it and resubmit.
>
> Thank you again!
> Will
>
>

Please disregard this version of the patch. I will resubmit a v2 patch
soon. There is a discrepancy between the way that gcc encodes the
callx (aka callr) and the way that clang encodes the callx. I am
working with the author of the IETF standardization document for the
BPF ISA to understand how the instruction encoding will be specified.
Once that discussion has concluded, I will resubmit the patch.

Sorry for the extra emails.

Will


>
> > Happy hacking!
> >
> > brgds, H-P

Re: [PATCH 0/1] objdump: Add callx support for BPF CPU v1

[Nick Clifton]

Hi Will,

> First time patch submitter but long time GNU enthusiast. Thank you for
> all the work that you all do to maintain these vital pieces of software
> for the community.

Welcome on board and thank you contributing.


> I tried very hard to follow best practices for creating this patch
> submission but I am sure that I did something wrong. Please let me know
> what I can do differently or better.

No, no, what you did was excellent.  My only suggestions is that
with your v2 patch you consider adding a new assembler test (or
extending an already existing one) that checks that the instruction
can be assembled and then disassembled.  For example you might
like to add it to gas/testsuite/gas/bpf/call.s (and the associated
call.d file for the disassembly).

Cheers
   Nick