(2014/08/16 10:38), Wang Nan wrote:
> On 2014/8/15 23:23, Masami Hiramatsu wrote:
>> (2014/08/12 13:56), Wang Nan wrote:
>>> +/* Caller must ensure addr & 3 == 0 */
>>> +static int can_optimize(unsigned long paddr)
>>> +{
>>> + return 1;
>>> +}
>>
>> As we have talked on another thread, we'd better filter-out all stack-pushing
>> instructions here, since (as you said) that will corrupt pt_regs on the
>> stack.
>>
>> Thank you,
>>
>
> So we need to identify the replaced instruction. I think some improvement on
> arm instruction decoder is required, else we have to implement another
> (although simpler)
> decoder for memory accessing instructions.
Since arm32 already have instruction emulator, I guess it's not so hard, we can
start with using emulator code to find which one will change sp.
> In this situation we are talking about, we need the decoder identify the
> addressing
> information for str/stm instroction.
No, sp register must be always the top of stack, or the code is just broken
(breaks
stack frame). So we need to identify the stm/str instructions which destination
is
sp register.
> However, decoder can bring up more information such as
> instruction type, source/destnation registers, memory access pattern ...
> With such information, we can further optimize our trampoline code.
> For example: doesn't protect destnation registers, and for some (most of, I
> think) instruction,
> we can direct execute them like x86_64.
Yeah, direct execution may reduce the overhead much :). But anyway, since we
need pt_regs,
we have to store all registers same as pt_regs.
Thank you,
--
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Research Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu...@hitachi.com
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
