Re: [Qemu-devel] [PATCH v4 24/26] tcg: Allocate a guard page after code_gen_buffer

[Richard Henderson <rth@twiddle.net>]

On 10/01/2015 02:33 AM, Aurelien Jarno wrote:
> On 2015-09-30 15:09, Richard Henderson wrote:
>> This will catch any overflow of the buffer.
>>
>> Add a native win32 alternative for alloc_code_gen_buffer;
>> remove the malloc alternative.
>>
>> Signed-off-by: Richard Henderson <rth@twiddle.net>
>> ---
>>   translate-all.c | 210 ++++++++++++++++++++++++++++++++------------------------
>>   1 file changed, 119 insertions(+), 91 deletions(-)
>
> I havent reviewed the patch in details, but I wonder if that could really
> happen? Given the size of the code generation buffer (a few MB at
> least), I don't think it's a problem if we don't use it to the last kB,
> and thus we could keep some safe margin if needed.

I don't believe that it can happen at the moment, given the size of the current 
"safe margin".  But the subject came up during review, and I thought we should 
indeed do something about it.

> Also what happens if an overflow really happens? In softmmu mode a
> segmentation fault will happen. In user-mode I guess the fault will
> be forwarded to the guest process, so this will likely wrongly
> be interpreted as a bug in the guest code.

You're right.  I had thought that we tested page_check_range on the given 
address in user-mode before passing it on to the guest.

In any case, that's how to fix the problem you indicate...


r~