Re: [Patch 3/3] prepopulate/cache cleared pages

[Nick Piggin <nickpiggin@yahoo.com.au>]

Ingo Molnar wrote:
> * Nick Piggin <nickpiggin@yahoo.com.au> wrote:
> 
> 
>>>couldnt the new pte be flipped in atomically via cmpxchg? That way 
>>>we could do the page clearing close to where we are doing it now,
>>>but without holding the mmap_sem.
>>
>>We have nothing to pin the pte page with if we're not holding the 
>>mmap_sem.
> 
> 
> why does it have to be pinned? The page is mostly private to this thread 
> until it manages to flip it into the pte. Since there's no pte presence, 
> there's no swapout possible [here i'm assuming anonymous malloc() 
> memory, which is the main focus of Arjan's patch]. Any parallel 
> unmapping of that page will be caught and the installation of the page 
> will be prevented by the 'bit-spin-lock' embedded in the pte.
> 

No, I was talking about page table pages, rather than the newly
allocated page.

But I didn't realise you wanted the bit lock to go the other way
as well (ie. a real bit spinlock). Seems like that would have to
add overhead somewhere.

> 
>>But even in that case, there is nothing in the mmu gather / tlb flush 
>>interface that guarantees an architecture cannot free the page table 
>>pages immediately (ie without waiting for the flush IPI). This would 
>>make sense on architectures that don't walk the page tables in 
>>hardware.
> 
> 
> but the page wont be found by any other CPU, so it wont be freed! It is 
> private to this CPU. The page has no pte presence. It will only be 
> present and lookupable as a result of the cmpxchg() flipping the page 
> into the pte.
> 

Yeah, as I said above, the newly allocated page is fine, it is the
page table pages I'm worried about.

-- 
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com