Re: [PATCH RFC 1/1] tcg: Always pass the full write size to notdirty_write()

[Richard Henderson <richard.henderson@linaro.org>]

On 8/8/23 02:59, Ilya Leoshkevich wrote:
> On Mon, 2023-08-07 at 11:21 -0700, Richard Henderson wrote:
>> IIRC there are some uses of probe_access_* that set size == 0.
>> Should we adjust addr+size to cover the whole page for that case?
>> That seems to be the intent, anyway.
> 
> There is a comment that says we shouldn't do watchpoint/smc detection
> in this case:
> 
>      /* Per the interface, size == 0 merely faults the access. */
>      if (size == 0) {
>          return NULL;
>      }
> 
> Come to think of it, qemu-user works this way too: SMC is detected on
> the actual access, not the probe:
> 
>      helper_vstl()
>        cpu_stq_be_data_ra()
>          ...
>             stq_he_p()
>               <signal handler called>
>                 host_signal_handler()
>                   handle_sigsegv_accerr_write()
>                     page_unprotect()
>                       tb_invalidate_phys_page_unwind()
>                     cpu_loop_exit_noexc()
> 
> So all this is probably fine, I now think it's better to leave the code
> as is, especially given that I cannot reproduce the original problem
> anymore.

Ok then.


r~