Re: PTE access rules & abstraction

[Jeremy Fitzhardinge <jeremy@goop.org>]

Benjamin Herrenschmidt wrote:
> On Thu, 2008-09-25 at 11:15 -0700, Jeremy Fitzhardinge wrote:
>   
>> The ptep_modify_prot_start/commit pair specifies a single pte update in
>> such a way to allow more implementation flexibility - ie, there's no
>> naked requirement for an atomic fetch-and-clear operation.  I chose the
>> transaction-like terminology to emphasize that the start/commit
>> functions must be strictly paired; there's no way to fail or abort the
>> "transaction".  A whole group of those start/commit pairs can be batched
>> together without affecting their semantics.
>>     
>
> I still can't see the point of having now 3 functions instead of just
> one such as ptep_modify_protection(). I don't see what it buys you other
> than adding gratuituous new interfaces.
>   

Yeah, that would work too; that's pretty much how Xen implements it
anyway.  The main advantage of the start/commit pair is that the
resulting code was completely unchanged from the old code.  The mprotect
sequence using ptep_modify_protection would end up reading the pte twice
before writing it.

    J