Re: [Qemu-devel] [PATCH] target-*: Advance pc after recognizing a breakpoint

[Sergey Fedorov <serge.fdrv@gmail.com>]

On 19.10.2015 01:46, Richard Henderson wrote:
> On 10/16/2015 04:08 AM, Sergey Fedorov wrote:
>> On 16.10.2015 04:14, Richard Henderson wrote:
>>> On 10/16/2015 03:36 AM, Peter Maydell wrote:
>>>> On 14 October 2015 at 22:02, Richard Henderson <rth@twiddle.net>
>>>> wrote:
>>>>> On 10/15/2015 06:34 AM, Peter Maydell wrote:
>>>>>>
>>>>>> This is still the same cryptic comment we have in the
>>>>>> targets which do do this. Can we have something
>>>>>> that is a bit more explanatory about what is going on and
>>>>>> why we need to do this, please?
>>>>>
>>>>>
>>>>> Suggestions?
>>>>
>>>> ...well, I don't entirely understand the problem it's
>>>> fixing, which is why I'm asking for a better comment :-)
>>>
>>> Heh.  Fair enough.  How about
>>>
>>>    /* The address covered by the breakpoint must be included in
>>>       [tb->pc, tb->pc + tb->size) in order to for it to be
>>>       properly cleared -- thus we increment the PC here so that
>>>       the logic setting tb->size below does the right thing.  */
>>>
>>> There are two edge cases that cause the problem with clearing that
>>> could be described, but I think that the comment becomes too bulky, as
>>> well as confuses the situation for someone cutting-and-pasting the
>>> logic to a new port.
>>
>> Maybe we could rather fix that condition in
>> tb_invalidate_phys_page_range()? It seems weird that it can't handle a
>> zero-sized TB.
>
> We also need to be able to handle a TB which crosses a page.  E.g. the
> breakpoint is at the page boundary, and we fall through into it from
> the top. This will be true on e.g. x86.  This is not simply true for
> breakpoint insertion/removal, but also page invalidation.
>
> The same fix, adding a byte to the size, handles this as well.

It's clear except that instructions crossing a page boundary can be
different in size. AFAIK, x86 instructions can be up to 15-byte long.
What if only the very last byte of instruction crosses a page boundary?

Best regards,
Sergey