Re: [PATCH v2 1/2] RISC-V: Fix buffer overflow on print_insn_riscv

[Jan Beulich <jbeulich@suse.com>]

On 04.10.2022 12:16, Jan Beulich via Binutils wrote:
> On 04.10.2022 12:13, Tsukasa OI wrote:
>> On 2022/10/04 18:58, Jan Beulich wrote:
>>> On 04.10.2022 11:45, Tsukasa OI wrote:
>>>> --- a/include/opcode/riscv.h
>>>> +++ b/include/opcode/riscv.h
>>>> @@ -55,6 +55,8 @@ static const char * const riscv_pred_succ[16] =
>>>>    "i", "iw", "ir", "irw", "io", "iow", "ior", "iorw"
>>>>  };
>>>>  
>>>> +#define RISCV_MAX_INSN_LEN 22  /* max 176-bit encoding.  */
>>>
>>> To be honest this still doesn't look sufficient to me: There's still
>>> no connection between this constant and riscv_insn_length(). Yet both
>>> want changing at the same time when it comes to insn length aspects.
>>> As said in reply to v1 - comments may be one way of dealing with this.
>>> We don't have BUILD_BUG_ON() or alike (and even if we had it wouldn't
>>> be usable in a portable way), so an actual build time check might not
>>> be feasible. A runtime check also doesn't look realistic, as
>>>
>>>     gas_assert (riscv_insn_length(~0) == RISCV_MAX_INSN_LEN);
>>>
>>> wouldn't be correct, and I'm unconvinced of using other than the most
>>> simple ~0 as an argument here.
>>
>> I have to agree that the constant with no direct connection with
>> riscv_insn_length is not good but I don't come up with better solution
>> than this (with given constraints).
>> In any case, keeping this stack buffer overflow is definitely a bad idea
>> and we have to do something to deal with it in a days.
> 
> Agreed. Hence could you add cross-referencing comments at both sides
> while introducing the #define, as a minimal measure?

Or wait - why don't you move the #define _into_ riscv_insn_length(),
placed right at the position that would need touching when adding
support for wider insns (or when deciding to reduce support again)?

Jan