[BPF] Mark FI_ri as isPseudo to avoid assertion during disassembly

[eddyz87]

Summary

When a specific sequence of bytes is present in the file during
disassembly the disassembler fails with the following assertion:

...
     0:	18 20 00 00 00 00 00 00	lea
... Assertion `idx < size()' failed.
...
llvm::SmallVectorTemplateCommon<...>::operator[](...) ...
llvm::MCInst::getOperand(unsigned int) ...
llvm::BPFInstPrinter::printOperand(...) ...
llvm::BPFInstPrinter::printInstruction() ...
llvm::BPFInstPrinter::printInst(...) ...
...

The byte sequence causing the error is (little endian):

18 20 00 00 00 00 00 00 00 00 00 00 00 00 00 00

The issue could be reproduced using the program bellow:

test.ir:

@G = constant
       [16 x i8]
       [i8 u0x18, i8 u0x20, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00,
        i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00],
       section "foo", align 8

Compiled and disassembled as follows:

cat test.ir | llc -march=bpfel -filetype=obj -o - \
            | llvm-objdump --arch=bpfel --section=foo -d -

This byte sequence corresponds to FI_ri instruction declared in the
BPFInstrInfo.td as follows:

def FI_ri
    : TYPE_LD_ST<BPF_IMM.Value, BPF_DW.Value,
                 (outs GPR:$dst),
                 (ins MEMri:$addr),
                 "lea\t$dst, $addr",
                 [(set i64:$dst, FIri:$addr)]> {
  // This is a tentative instruction, and will be replaced
  // with MOV_rr and ADD_ri in PEI phase
  let Inst{51-48} = 0;
  let Inst{55-52} = 2;
  let Inst{47-32} = 0;
  let Inst{31-0} = 0;
  let BPFClass = BPF_LD;
}

Notes:

• First byte (opcode) is formed as follows:
  • BPF_IMM.Value is 0x00
  • BPF_DW.Value is 0x18
  • BPF_LD is 0x00
• Second byte (registers) is formed as follows:
  • let Inst{55-52} = 2;
  • let Inst{51-48} = 0;

The FI_ri instruction is always replaced by MOV_rr ADD_ri instructions
pair in the BPFRegisterInfo::eliminateFrameIndex method. Thus, this
instruction should be invisible to disassembler. This patch achieves
this by adding "isPseudo" flag for this instruction.

The bug was found by decompiling of one of the BPF tests from Linux
kernel (llvm-objdump -D tools/testing/selftests/bpf/bpf_iter_sockmap.o)

[eddyz87]

Commit message fixes for better formatting

[eddyz87]

This comment was removed by eddyz87.

[eddyz87]

The following tests are passing with this patch:

• LLVM unit and regression tests (ninja check-all)
• BPF selftests (linux/tools/testing/selftests/bpf/vmtest.sh, bpf-next kernel repository, commit 20b87e7c29df)

[yonghong-song]

The change looks good to me. Do you have write permission to merge into llvm-project main repo? If not, I can do the merge for you.

[eddyz87]

In D125185#3503586, @yonghong-song wrote:

The change looks good to me. Do you have write permission to merge into llvm-project main repo? If not, I can do the merge for you.

Thank you,
I don't have the write permission, please merge it for me.

[yonghong-song]

Could you provide proper "<Name> <Email Address>" so I can change author to you when I push the commit to the main branch?

[eddyz87]

In D125185#3504146, @yonghong-song wrote:

Could you provide proper "<Name> <Email Address>" so I can change author to you when I push the commit to the main branch?

Eduard Zingerman [e-mail]

[ast]

a test got lost?

[yonghong-song]

In D125185#3505000, @ast wrote:

a test got lost?

My fault. Forgot to git add the file before the push. Will push the test shortly with a separate commit.

llvm/lib/Target/BPF/BPFInstrInfo.td

: TYPE_LD_ST<BPF_IMM.Value, BPF_DW.Value,
                  [(set i64:$dst, FIri:$addr)]> {
   // This is a tentative instruction, and will be replaced
   // with MOV_rr and ADD_ri in PEI phase
   let Inst{51-48} = 0;
   let Inst{55-52} = 2;
   let Inst{47-32} = 0;
   let Inst{31-0} = 0;
   let BPFClass = BPF_LD;
+  bit isPseudo = true;
 }
 
 def LD_pseudo
     : TYPE_LD_ST<BPF_IMM.Value, BPF_DW.Value,
                  (outs GPR:$dst),
                  (ins i64imm:$pseudo, u64imm:$imm),
                  "ld_pseudo\t$dst, $pseudo, $imm",
                  [(set GPR:$dst, (int_bpf_pseudo imm:$pseudo, imm:$imm))]> {

llvm/test/CodeGen/BPF/objdump_fi_ri.ll

+; RUN: llc -march=bpfel -filetype=obj -o - %s \
+; RUN:     | llvm-objdump --arch=bpfel --section=foo -d - \
+; RUN:     | FileCheck %s
+
+; This test was added because "isPseudo" flag was missing in FI_ri
+; instruction definition and certain byte sequence caused an assertion
+; in llvm-objdump tool.
+
+; The value "G" is byte by byte little endian representation of the FI_ri instruction,
+; as declared in the BPFInstrInfo.td.
+; The first byte encodes an opcode: BPF_IMM(0x00) | BPF_DW(0x18) | BPF_LD(0x00)
+; The second byte encodes source and destination registers: 2 and 0 respectively.
+; The rest of the bytes are zeroes to comply with the specification.
+; An additional 8 bytes follow the instruction as an immediate 64 bit argument,
+; (because of the BPF_IMM flag).
+
+; This is a pseudo instruction, meaning that it's not possible to
+; write it in assembly directly. Thus it is coded as a byte array.
+
+; Note the "bpfel" flags in the RUN command.
+
+@G = constant [16 x i8]
+              [i8 u0x18, i8 u0x20, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00,
+               i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00, i8 u0x00],
+              section "foo", align 8
+
+; CHECK-LABEL: G
+; CHECK:       0: 18 20 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ld_pseudo r0, 2, 0