diff options
author | Alan Modra <amodra@gmail.com> | 2024-07-26 08:49:48 +0930 |
---|---|---|
committer | Alan Modra <amodra@gmail.com> | 2024-07-26 10:34:54 +0930 |
commit | e26ff6c44e7546b9f761e00bb7cf76a3fa8bfc5d (patch) | |
tree | 0b517db8c2a89950a86ae2b0700b02bc2afd3190 | |
parent | 5a7f5aa29b33e9cd01a0062bcf8d2aa4d76852f9 (diff) | |
download | gdb-e26ff6c44e7546b9f761e00bb7cf76a3fa8bfc5d.zip gdb-e26ff6c44e7546b9f761e00bb7cf76a3fa8bfc5d.tar.gz gdb-e26ff6c44e7546b9f761e00bb7cf76a3fa8bfc5d.tar.bz2 |
ARM print_insn_mve assertion
This corrects objdump -d -m armv8.1-m.main output for a testcase found
by oss-fuzz, .inst 0xee2fee79, which hits an assertion.
Obviously the switch case constants should be binary, not hex.
Correcting that is enough to cure this assertion, but I don't see any
point in singling out the invalid case 0b10. In fact, it is just plain
wrong to print "undefined instruction: size equals zero undefined
instruction: size equals two".
I also don't see the need for defensive programming here as is done
elsewhere in checking that "value" is in range before indexing
mve_vec_sizename. There is exactly one MVE_VSHLL_T2 entry in
mve_opcodes. It is easy to verify that "value" is only two bits.
-rw-r--r-- | opcodes/arm-dis.c | 19 |
1 files changed, 2 insertions, 17 deletions
diff --git a/opcodes/arm-dis.c b/opcodes/arm-dis.c index 480e4c2..d1d7ca3 100644 --- a/opcodes/arm-dis.c +++ b/opcodes/arm-dis.c @@ -9887,23 +9887,8 @@ print_insn_mve (struct disassemble_info *info, long given) if (mve_shift_insn_p (insn->mve_op)) print_mve_shift_n (info, given, insn->mve_op); else if (insn->mve_op == MVE_VSHLL_T2) - { - switch (value) - { - case 0x00: - func (stream, dis_style_immediate, "8"); - break; - case 0x01: - func (stream, dis_style_immediate, "16"); - break; - case 0x10: - print_mve_undefined (info, UNDEF_SIZE_0); - break; - default: - assert (0); - break; - } - } + func (stream, dis_style_immediate, "%s", + mve_vec_sizename[value]); else { if (insn->mve_op == MVE_VSHLC && value == 0) |