[PATCH v3 4/4] arm64: mte: Optimize mte_assign_mem_tag_range()

[Vincenzo Frascino]

On 1/18/21 10:41 AM, Mark Rutland wrote:
> On Sun, Jan 17, 2021 at 12:27:08PM +0000, Vincenzo Frascino wrote:
>> Hi Mark,
>>
>> On 1/16/21 2:22 PM, Vincenzo Frascino wrote:
>>>> Is there any chance that this can be used for the last bytes of the
>>>> virtual address space? This might need to change to `_addr == _end` if
>>>> that is possible, otherwise it'll terminate early in that case.
>>>>
>>> Theoretically it is a possibility. I will change the condition and add a note
>>> for that.
>>>
>>
>> I was thinking to the end of the virtual address space scenario and I forgot
>> that if I use a condition like `_addr == _end` the tagging operation overflows
>> to the first granule of the next allocation. This disrupts tagging accesses for
>> that memory area hence I think that `_addr < _end` is the way to go.
> 
> I think it implies `_addr != _end` is necessary. Otherwise, if `addr` is
> PAGE_SIZE from the end of memory, and `size` is PAGE_SIZE, `_end` will
> be 0, so using `_addr < _end` will mean the loop will terminate after a
> single MTE tag granule rather than the whole page.
> 
> Generally, for some addr/increment/size combination (where all are
> suitably aligned), you need a pattern like:
> 
> | do {
> |       thing(addr);
> |       addr += increment;
> | } while (addr != end);
> 
> ... or:
> 
> | for (addr = start; addr != end; addr += increment) {
> |       thing(addr);
> | }
> 
> ... to correctly handle working at the very end of the VA space.
> 
> We do similar for page tables, e.g. when we use pmd_addr_end().
>

Good point! I agree it wraps around otherwise. I will change it accordingly.

Thanks!

> Thanks,
> Mark.
> 

-- 
Regards,
Vincenzo