Re: [Qemu-devel] [PATCH v2 4/8] exec: Factor out cpu_watchpoint_address_matches

[Richard Henderson]

On 8/29/19 10:20 AM, Philippe Mathieu-Daudé wrote:
>> +    /* Avoid trapping reads of pages with a write breakpoint. */
>> +    match = (prot & PAGE_READ ? BP_MEM_READ : 0)
>> +          | (prot & PAGE_WRITE ? BP_MEM_WRITE : 0);
> 
> Isn't it cheaper to do here:
> 
>        if (!match) {
>            return iotlb;
>        }
> 
> or
> 
>        if (match) {

Note that PROT_NONE pages never reach here; they always trap in tlb_fill.

The only way we can get match == 0 here is for the case of an execute-only
page.  Which is possible, but extremely unlikely.  Almost all targets merge the
text and rodata sections, which means that virtually all executable pages are
also readable.

(Although I must say that in this age of ROP-gadgets, leaving the rodata
section executable is probably a mistake, and tools should be updated to *not*
merge them.  That's still not necessarily execute-only for the text section,
but I don't see anything in principal that would prevent it.)


r~