Date:	Fri, 27 Mar 2009 18:57:31 -0400
From:	Rik van Riel <riel@...hat.com>
To:	Martin Schwidefsky <schwidefsky@...ibm.com>
CC:	linux-mm@...ck.org, linux-kernel@...r.kernel.org,
	virtualization@...ts.osdl.org, frankeh@...son.ibm.com,
	akpm@...l.org, nickpiggin@...oo.com.au, hugh@...itas.com
Subject: Re: [patch 1/6] Guest page hinting: core + volatile page cache.

Martin Schwidefsky wrote:

> The major obstacles that need to get addressed:
> * Concurrent page state changes:
>   To guard against concurrent page state updates some kind of lock
>   is needed. If page_make_volatile() has already done the 11 checks it
>   will issue the state change primitive. If in the meantime one of
>   the conditions has changed the user that requires that page in
>   stable state will have to wait in the page_make_stable() function
>   until the make volatile operation has finished. It is up to the
>   architecture to define how this is done with the three primitives
>   page_test_set_state_change, page_clear_state_change and
>   page_state_change.
>   There are some alternatives how this can be done, e.g. a global
>   lock, or lock per segment in the kernel page table, or the per page
>   bit PG_arch_1 if it is still free.

Can this be taken care of by memory barriers and
careful ordering of operations?

If we consider the states unused -> volatile -> stable
as progressively higher, "upgrades" can be done before
any kernel operation that requires the page to be in
that state (but after setting up the things that allow
it to be found), while downgrades can be done after the
kernel is done with needing the page at a higher level.

Since the downgrade checks for users that need the page
in a higher state, no lock should be required.

In fact, it may be possible to manage the page state
bitmap with compare-and-swap, without needing a call
to the hypervisor.

> Signed-off-by: Martin Schwidefsky <schwidefsky@...ibm.com>

Some comments and questions in line.

> @@ -601,6 +604,21 @@ copy_one_pte(struct mm_struct *dst_mm, s
>  
>  out_set_pte:
>  	set_pte_at(dst_mm, addr, dst_pte, pte);
> +	return;
> +
> +out_discard_pte:
> +	/*
> +	 * If the page referred by the pte has the PG_discarded bit set,
> +	 * copy_one_pte is racing with page_discard. The pte may not be
> +	 * copied or we can end up with a pte pointing to a page not
> +	 * in the page cache anymore. Do what try_to_unmap_one would do
> +	 * if the copy_one_pte had taken place before page_discard.
> +	 */
> +	if (page->index != linear_page_index(vma, addr))
> +		/* If nonlinear, store the file page offset in the pte. */
> +		set_pte_at(dst_mm, addr, dst_pte, pgoff_to_pte(page->index));
> +	else
> +		pte_clear(dst_mm, addr, dst_pte);
>  }

It would be good to document that PG_discarded can only happen for
file pages and NOT for eg. clean swap cache pages.

> @@ -1390,6 +1391,7 @@ int test_clear_page_writeback(struct pag
>  			radix_tree_tag_clear(&mapping->page_tree,
>  						page_index(page),
>  						PAGECACHE_TAG_WRITEBACK);
> +			page_make_volatile(page, 1);
>  			if (bdi_cap_account_writeback(bdi)) {
>  				__dec_bdi_stat(bdi, BDI_WRITEBACK);
>  				__bdi_writeout_inc(bdi);

Does this mark the page volatile before the IO writing the
dirty data back to disk has even started?  Is that OK?

-- 
All rights reversed.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists