Re: [PATCH v3 1/1] objdump, as: Add callx support for BPF CPU v1

[Will Hawkins <hawkinsw@obs.cr>]

On Wed, Feb 14, 2024 at 11:30 AM Jose E. Marchesi
<jose.marchesi@oracle.com> wrote:
>
>
> Hi Will.
> Thanks for the patch.  Please see some comments below.
>
> > Add support for (dis)assembling the callx instruction back to CPU v1.
> >
> > gas/ChangeLog:
> >
> >       * testsuite/gas/bpf/indcall-1-pseudoc.d: Refactor tests ...
> >       * testsuite/gas/bpf/indcall-1-pseudoc.s: ... to visually match ...
> >       * testsuite/gas/bpf/indcall-1.d: ... equivalent test in ...
> >       * testsuite/gas/bpf/indcall-1.s: ... clang/llvm.
> >
> > include/ChangeLog:
> >
> >       * opcode/bpf.h (enum bpf_insn_id): BPF_INSN_CALLR to BPF_INSN_CALLX
> >       * (for consistency) and add it to the v1 ISA variant.
> >
> > opcodes/ChangeLog:
> >
> >       * bpf-opc.c: Use BPF_INSN_CALLX instead of BPF_INSN_CALLR.
> >
> > Signed-off-by: Will Hawkins <hawkinsw@obs.cr>
> > ---
> >  gas/testsuite/gas/bpf/indcall-1-pseudoc.d |  8 ++++----
> >  gas/testsuite/gas/bpf/indcall-1-pseudoc.s |  6 +++---
> >  gas/testsuite/gas/bpf/indcall-1.d         | 10 +++++-----
> >  gas/testsuite/gas/bpf/indcall-1.s         |  6 +++---
> >  include/opcode/bpf.h                      |  2 +-
> >  opcodes/bpf-opc.c                         |  4 ++--
> >  6 files changed, 18 insertions(+), 18 deletions(-)
> >
> > diff --git a/gas/testsuite/gas/bpf/indcall-1-pseudoc.d b/gas/testsuite/gas/bpf/indcall-1-pseudoc.d
> > index 7a95bad8e65..12f9d6a9d49 100644
> > --- a/gas/testsuite/gas/bpf/indcall-1-pseudoc.d
> > +++ b/gas/testsuite/gas/bpf/indcall-1-pseudoc.d
> > @@ -1,4 +1,4 @@
> > -#as: -EL -mdialect=pseudoc -misa-spec=xbpf
> > +#as: -EL -mdialect=pseudoc -misa-spec=v1
>
> I think this test can now omit -misa-spec entirely.
>
> >  #objdump: -M xbpf,pseudoc,dec -dr
> >  #source: indcall-1-pseudoc.s
> >  #name: BPF indirect call 1, pseudoc syntax
> > @@ -10,11 +10,11 @@ Disassembly of section \.text:
> >  0000000000000000 <main>:
> >     0:        b7 00 00 00 01 00 00 00         r0=1
> >     8:        b7 01 00 00 01 00 00 00         r1=1
> > -  10:        b7 02 00 00 02 00 00 00         r2=2
> > -  18:        18 06 00 00 38 00 00 00         r6=56 ll
> > +  10:        b7 03 00 00 03 00 00 00         r3=3
> > +  18:        18 02 00 00 38 00 00 00         r2=56 ll
> >    20:        00 00 00 00 00 00 00 00[    ]*
> >                       18: R_BPF_64_64 .text
> > -  28:        8d 06 00 00 00 00 00 00         callx r6
> > +  28:        8d 02 00 00 00 00 00 00         callx r2
> >    30:        95 00 00 00 00 00 00 00         exit
>
> I don't see any reason for changing the test bodies like this.  If it is
> about matching some llvm test, IMO consistency between binutils and llvm
> testsuites is not a general goal we can (or should) aim to.
>
> Same applies to the other tests changed in the thunk below.

As I said, I am happy to do whatever you like! I did it just so that I
could visually check that the two matched and then I left it. In v4 I
will revert this particular piece!

>
> >  0000000000000038 <bar>:
> > diff --git a/gas/testsuite/gas/bpf/indcall-1-pseudoc.s b/gas/testsuite/gas/bpf/indcall-1-pseudoc.s
> > index 2042697f15b..5639e288869 100644
> > --- a/gas/testsuite/gas/bpf/indcall-1-pseudoc.s
> > +++ b/gas/testsuite/gas/bpf/indcall-1-pseudoc.s
> > @@ -4,9 +4,9 @@
> >  main:
> >       r0 = 1
> >       r1 = 1
> > -     r2 = 2
> > -     r6 = bar ll
> > -     callx r6
> > +     r3 = 3
> > +     r2 = bar ll
> > +     callx r2
> >       exit
> >  bar:
> >       r0 = 0
> > diff --git a/gas/testsuite/gas/bpf/indcall-1.d b/gas/testsuite/gas/bpf/indcall-1.d
> > index 51103bba2a1..1a2c36999b1 100644
> > --- a/gas/testsuite/gas/bpf/indcall-1.d
> > +++ b/gas/testsuite/gas/bpf/indcall-1.d
> > @@ -1,5 +1,5 @@
> > -#as: -EL -misa-spec=xbpf
> > -#objdump: -dr -M xbpf,dec
> > +#as: -EL -misa-spec=v1
> > +#objdump: -dr -M v1,dec
>
> Likewise, I think this test can just omit the -M v1 and -ima-spec=v1
> arguments now.
>

Ack!

> >  #source: indcall-1.s
> >  #name: BPF indirect call 1, normal syntax
> >
> > @@ -10,11 +10,11 @@ Disassembly of section \.text:
> >  0000000000000000 <main>:
> >     0:        b7 00 00 00 01 00 00 00         mov %r0,1
> >     8:        b7 01 00 00 01 00 00 00         mov %r1,1
> > -  10:        b7 02 00 00 02 00 00 00         mov %r2,2
> > -  18:        18 06 00 00 38 00 00 00         lddw %r6,56
> > +  10:        b7 03 00 00 03 00 00 00         mov %r3,3
> > +  18:        18 02 00 00 38 00 00 00         lddw %r2,56
> >    20:        00 00 00 00 00 00 00 00[    ]*
> >                       18: R_BPF_64_64 .text
> > -  28:        8d 06 00 00 00 00 00 00         call %r6
> > +  28:        8d 02 00 00 00 00 00 00         call %r2
> >    30:        95 00 00 00 00 00 00 00         exit
> >
> >  0000000000000038 <bar>:
> > diff --git a/gas/testsuite/gas/bpf/indcall-1.s b/gas/testsuite/gas/bpf/indcall-1.s
> > index 5d49e41040a..7fbeeeb9a38 100644
> > --- a/gas/testsuite/gas/bpf/indcall-1.s
> > +++ b/gas/testsuite/gas/bpf/indcall-1.s
> > @@ -4,9 +4,9 @@
> >  main:
> >      mov %r0, 1
> >      mov %r1, 1
> > -    mov %r2, 2
> > -    lddw %r6, bar
> > -    call %r6
> > +    mov %r3, 3
> > +    lddw %r2, bar
> > +    call %r2
> >      exit
> >
> >  bar:
> > diff --git a/include/opcode/bpf.h b/include/opcode/bpf.h
> > index df1e3bd0918..97e25053175 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_CALLX, BPF_INSN_CALL, BPF_INSN_EXIT,
>
> I don't think it is a good idea to rename BPF_INSN_CALLR to
> BPF_INSN_CALLX.
>
> At the moment the assembler uses the `callx' mnemonic for it and GCC
> uses the same name because that is what clang does, but if/when the
> corresponding instruction gets incorporated into BPF, I indend to push
> for a better name, certainly not something as unmemorable and
> indescriptive as `callx'.

I will drop this change, as well! I know that Dave is working on this
naming now on the standardization mailing list. I look forward to
watching that argument! (I am definitely going to get butter with the
popcorn!).

I will have a v4 later today -- the day job calls for the next few hours! Sorry!
Will


>
> >    /* 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..23473fc0cd9 100644
> > --- a/opcodes/bpf-opc.c
> > +++ b/opcodes/bpf-opc.c
> > @@ -272,8 +272,8 @@ const struct bpf_opcode bpf_opcodes[] =
> >     BPF_V1, BPF_CODE, BPF_CLASS_JMP|BPF_CODE_JSET|BPF_SRC_X},
> >    {BPF_INSN_JNER, "jne%W%dr , %sr , %d16", "if%w%dr != %sr%wgoto%w%d16",
> >     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, "call%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",