Re: Note describing poor dcache utilization under high memory pressure

[Daniel Phillips <phillips@bonn-fries.net>]

On January 29, 2002 02:29 am, Oliver Xymoron wrote:
> On Mon, 28 Jan 2002, Daniel Phillips wrote:
> 
> >   copy_page_range
> >     Intead of copying the page tables, just increment their use counts
> >
> >   zap_page_range:
> >     If a page table is shared according to its use count, just decrement
> >     the use count and otherwise leave it alone.
> >
> >   handle_mm_fault:
> >     If a page table is shared according to its use count and the faulting
> >     instruction is a write, allocate a new page table and do the work that
> >     would have normally been done by copy_page_range at fork time.
> >     Decrement the use count of the (perhaps formerly) shared page table.
> 
> Somewhere in here, the pages have got to all be marked read-only or
> something.

Yes, that's an essential detail I omitted: when a page table's use count 
transitions from 1 to 2, mark all the CoW pages on the page table RO.

> If they're not, then either parent or child writing to
> non-faulting addresses will be writing to shared memory.

Yes, and after all, the whole point is to generalize CoW of pages to include 
instantiation of page tables.

> I think something more is needed, such as creating a minimal page table
> for the child process with read-only mappings to the current %EIP and %EBP
> pages in it. This gets us past the fork/exec hurdle. Without the exec, we
> copy over chunks when they're accessed as above in handle_mm_fault. But
> you can't actually _share_ the page tables without marking the pages
> themselves readonly.

Oh yes, it's what I intended, thanks.  Um, and I think you just told me what 
one of my bugs is.

-- 
Daniel