mov [rax+rsp], rsi failed to assembly

[j123123]

Testcase rasm2 from radare2

$ rasm2 -b 64 -a x86.ks "mov [rax+rsp], rsi"
rasm2: /home/user/.config/radare2/r2pm/git/keystone/llvm/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp:558: void {anonymous}::X86MCCodeEmitter::EmitMemModRMByte(const llvm::MCInst&, unsigned int, unsigned int, uint64_t, unsigned int&, llvm::raw_ostream&, llvm::SmallVectorImpl<llvm::MCFixup>&, const llvm::MCSubtargetInfo&) const: Assertion `IndexReg.getReg() != X86::ESP && IndexReg.getReg() != X86::RSP && "Cannot use ESP as index reg!"' failed.
Aborted (core dumped)

[j123123]

but it works with [rsp+rax]

[aquynh]

would you mind sending in a testcase for this issue, like those under https://github.com/keystone-engine/keystone/tree/master/suite/regress?

[j123123]

I'm not very familiar with python. I can provide this example (terminal session):

$ cat sample.py 
#!/usr/bin/env python

from __future__ import print_function
from keystone import *

def test_ks(arch, mode, code, syntax=0):
    ks = Ks(arch, mode)
    if syntax != 0:
        ks.syntax = syntax

    encoding, count = ks.asm(code)

    print("%s = [ " % code, end='')
    for i in encoding:
        print("%02x " % i, end='')
    print("]")

if __name__ == '__main__':
    # X86
    test_ks(KS_ARCH_X86, KS_MODE_64, b"mov [rax+rsp], rsi")
    test_ks(KS_ARCH_X86, KS_MODE_64, b"mov [rsp+rax], rsi")
$ python sample.py 
mov [rax+rsp], rsi = [ 48 89 34 20 ]
mov [rsp+rax], rsi = [ 48 89 34 04 ]
$ rasm2 -a x86 -s intel -b 64 -d '48 89 34 20'
mov qword [rax], rsi
$ rasm2 -a x86 -s intel -b 64 -d '48 89 34 04'
mov qword [rsp + rax], rsi

Here I get invalid output when using python binding. rasm2 here uses capstone to disasm output from keystone