Message-ID: <alpine.LSU.2.11.1402271054390.7000@eggly.anvils>
Date:	Thu, 27 Feb 2014 16:33:43 -0800 (PST)
From:	Hugh Dickins <hughd@...gle.com>
To:	Kelley Nielsen <kelleynnn@...il.com>
cc:	riel@...riel.com, akpm@...ux-foundation.org,
	gnomes@...rguk.ukuu.org.uk, josh@...htriplett.org,
	linux-kernel@...r.kernel.org, linux-mm@...ck.org,
	opw-kernel@...glegroups.com, jamieliu@...gle.com,
	sjenning@...ux.vnet.ibm.com
Subject: Re: [RFC] mm:prototype for the updated swapoff implementation

On Tue, 18 Feb 2014, Kelley Nielsen wrote:

> The function try_to_unuse() is of quadratic complexity, with a lot of
> wasted effort. It unuses swap entries one by one, potentially iterating
> over all the page tables for all the processes in the system for each
> one.

You've chosen a good target, and I like the look of what you've done.
But I'm afraid it will have to get uglier before it's ready, and I'm
unsure whether your approach will prove to be a clear win or not.

I am sure that it's an experiment worth trying:
there's plenty to dislike about what we have there at present.

> 
> This new proposed implementation of try_to_unuse simplifies its
> complexity to linear. It iterates over the system's mms once, unusing
> all the affected entries as it walks each set of page tables. It also
> makes similar changes to shmem_unuse.
> 
> Improvement
> 
> swapoff was called on a swap partition containing about 50M of data,
> and calls to the function unuse_pte_range were counted.
> 
> Present implementation....about 22.5M calls.
> Prototype.................about  7.0K   calls.

That's nice, but mostly it's the time spent that matters.

I should explain why we've left the try_to_unuse() implementation as is
for so many years: it's a matter of tradeoff between fast cpu and slow
seeking disk.

(Plus a reluctance to expend more memory on just making swapoff faster:
you're preserving that tradition, but it might be time to break from it
a little.  It's useful to see the responses from Rik and Alan: I've not
seen real user complaints demanding change for years.)

I'll be surprised if your approach does not improve swapoff from SSD
(and brd and zram and zswap) very significantly; but the case to worry
about is swapoff from hard disk.  You are changing swapoff to use the
cpu much more efficiently; but now that you no longer move linearly up
the swap_map, you are making the disk head seek around very much more.

How much that matters is unclear to me; and there's a danger that
we will end up stuck in trading worst-case scenarios without a clear
resolution.  But I hope that your approach will quickly convince me
once it's ready.

You're right to be using swapin_readahead()
to reduce those additional seeks.  From time to time it has been
proposed to replace the read_swap_cache_async() in try_to_unuse();
but, given the linearity of the current implementation, each time
I tried, I found any improvement it made at best risible compared
with time spent searching processes and shmem files.  With your
approach it becomes much more important.

Beware 3.14-rc2's 579f82901f6f "swap: add a simple detector for
inappropriate swapin readahead".  I believe what it does is appropriate
for your usage too, tuning the extent of readahead to its success rate
(which itself depends on how much memory is free or freeable).  But it
may alter your results, and you may want to experiment with it reverted,
or extended to suit you better (maybe try longer extents for swapoff).

> 
> Details
> 
> In shmem_unuse, iterate over the shmem_swaplist and, for each
> shmem_inode_info that contains a swap entry, pass it to shmem_unuse_inode,
> along with the swap type. In shmem_unuse_inode, iterate over its associated
> radix tree, and pass each exceptional entry to shmem_getpage_gfp by way of
> a new function called shmem_unuse_inode_index.
> 
> In try_to_unuse, instead of iterating over the entries in the type and
> unusing them one by one, perhaps walking all the page tables for all the
> processes for each one, iterate over the mmlist, making one pass. Pass
> each mm to unuse_mm to begin its page table walk, and during the walk,
> unuse all the ptes that have backing store in the swap type received by
> try_to_unuse.
> 
> Change unuse_mm and the intervening walk functions down to unuse_pte_range
> to take the type as a parameter, and to iterate over their entire range,
> calling the next function down on every iteration. In unuse_pte_range,
> make a swap entry from each pte in the range using the passed in type.
> If it has backing store in the type, call swapin_readahead to retrieve
> the page, and then pass this page to unuse_pte.
> 
> TODO

I usually like to try out the patch, put it under test, while I'm
looking at it; but this is quite a long TODO list, and some of the
things I noticed in the current RFC patch have deterred me from
trying it at all.

> 
> * Find and correct where swap entries are being left behind
> * Probably related: handle case of remaining reference in try_to_unuse
> * Remove find_next_to_unuse, and the call to it in try_to_unuse,
>   when the previous item has been resolved
> * Handle the failure of swapin_readahead in unuse_pte_range
> * make sure unuse_pte_range is handling multiple ptes in the best way
> * clean up after failure of unuse_pte in unuse_pte_range
> * Determine the proper place for the mmlist locks in try_to_unuse
> * Handle count of unused pages for frontswap
> * Determine what kind of housekeeping shmem_unuse needs
> * Tighten up the access control for all the various data structures
>   (for instance, the mutex on shmem_swaplist is held throughout the
>   entire process, which is probably not only unneccesary but problematic)

You're right to be concerned about shmem_swaplist_mutex and mmap_sem
hold times; and I'm ashamed of how swapoff races (with unlink of tmpfs
file or unmount of tmpfs) have led me to rely more than is healthy on
that shmem_swaplist_mutex, now it's always taken in shmem_writepage()
on the tmpfs reclaim path.  Not good; but something to return to at
a later time - the important thing is first to have a correct
implementation, with convincing timings that justify further effort.

> * Prevent radix entries with zero indices from being passed to
>   shmem_unuse_inode_index
> * Decide if shmem_unuse_inode* should be combined into one function
> * Find cases in which the errors returned from shmem_getpage_gfp can be
>   gracefully handled in shmem_unuse_inode_index, instead of just failing

Yes, I've not checked through them either, but I expect that any error
other than -ENOMEM (in this swapoff case, when you know you're only
dealing with existing user data, not adding to it) would just come
from a race, and be reason to continue on to the next entry.  With
prospect of occasional second (and third, and...) circuit through
all tmpfsfiles and processes later, if inuse_pages still not 0.

swapoff is a huge consumer of memory (is there any other system call
which can take so much in a single call?), so it is important to get out
quickly when there's no more memory left for it to take, whether it's
-ENOMEM or OOM-killed (you probably noticed the set_current_oom_origin()
which begs OOM-killer to kill it first); or even just that the user got
tired of waiting for it (signal_pending() checks both).

For testing, I usually swapon a fresh area of the same size immediately
before swapoff of an area, with heavy memory load running concurrently:
in that case the swapoff ought not to OOM, but the memory it's taking
be swapped out to the new area while swapoff of old progresses.  It
doesn't matter if your approach slows down such a test somewhat, it's
not a case to optimize for at all; but it does need to work correctly.

> * determine when old comments and housekeeping are no longer needed
>   (there are still some to serve as reminders of the housekeeping that
>    needs to be accounted for)
> ---
>  include/linux/shmem_fs.h |   2 +-
>  mm/shmem.c               | 146 +++++++-------------
>  mm/swapfile.c            | 352 ++++++++++++++++++++---------------------------
>  3 files changed, 206 insertions(+), 294 deletions(-)

That's good, we all love deletions, Linus especially;
but I'm afraid I'm going to spoil it for you below.

> 
> diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
> index 9d55438..af78151 100644
> --- a/include/linux/shmem_fs.h
> +++ b/include/linux/shmem_fs.h
> @@ -55,7 +55,7 @@ extern void shmem_unlock_mapping(struct address_space *mapping);
>  extern struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
>  					pgoff_t index, gfp_t gfp_mask);
>  extern void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end);
> -extern int shmem_unuse(swp_entry_t entry, struct page *page);
> +extern int shmem_unuse(unsigned int type);
>  
>  static inline struct page *shmem_read_mapping_page(
>  				struct address_space *mapping, pgoff_t index)
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 1f18c9d..802456e 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -650,127 +650,87 @@ static void shmem_evict_inode(struct inode *inode)
>  /*
>   * If swap found in inode, free it and move page from swapcache to filecache.
>   */
> -static int shmem_unuse_inode(struct shmem_inode_info *info,
> -			     swp_entry_t swap, struct page **pagep)
> +/* TODO Since there's hardly anything left of this function
> + * now that the things shmem_getpage_gfp does have been removed,
> + * just incorporate its actions into shmem_unuse_inode?
> + */
> +static int shmem_unuse_inode_index(struct shmem_inode_info *info,
> +			     pgoff_t index)
>  {
>  	struct address_space *mapping = info->vfs_inode.i_mapping;
> -	void *radswap;
> -	pgoff_t index;
> +	struct page *pagep;
>  	gfp_t gfp;
>  	int error = 0;
>  
> -	radswap = swp_to_radix_entry(swap);
> -	index = radix_tree_locate_item(&mapping->page_tree, radswap);
> -	if (index == -1)
> -		return 0;
> -
> -	/*
> -	 * Move _head_ to start search for next from here.
> -	 * But be careful: shmem_evict_inode checks list_empty without taking
> -	 * mutex, and there's an instant in list_move_tail when info->swaplist
> -	 * would appear empty, if it were the only one on shmem_swaplist.
> +	gfp = mapping_gfp_mask(mapping);
> +	error = shmem_getpage_gfp(&info->vfs_inode, index, &pagep, SGP_CACHE,
> +			gfp, NULL);
> +	/* TODO: go through all the possible error returns
> +	 * in shmem_getpage_gfp, and determine whether
> +	 * we need to fail, or whether we can gracefully recover.
> +	 * (for instance, if the page was swapped in from somewhere
> +	 * else in the kernel between the start of swapoff and now,
> +	 * and can be safely let go.)
> +	 * For now, send failure up the call chain for all errors.
>  	 */
> -	if (shmem_swaplist.next != &info->swaplist)
> -		list_move_tail(&shmem_swaplist, &info->swaplist);
> +	return error;
> +}
>  
> -	gfp = mapping_gfp_mask(mapping);
> -	if (shmem_should_replace_page(*pagep, gfp)) {
> -		mutex_unlock(&shmem_swaplist_mutex);
> -		error = shmem_replace_page(pagep, gfp, info, index);
> -		mutex_lock(&shmem_swaplist_mutex);
> -		/*
> -		 * We needed to drop mutex to make that restrictive page
> -		 * allocation, but the inode might have been freed while we
> -		 * dropped it: although a racing shmem_evict_inode() cannot
> -		 * complete without emptying the radix_tree, our page lock
> -		 * on this swapcache page is not enough to prevent that -
> -		 * free_swap_and_cache() of our swap entry will only
> -		 * trylock_page(), removing swap from radix_tree whatever.
> -		 *
> -		 * We must not proceed to shmem_add_to_page_cache() if the
> -		 * inode has been freed, but of course we cannot rely on
> -		 * inode or mapping or info to check that.  However, we can
> -		 * safely check if our swap entry is still in use (and here
> -		 * it can't have got reused for another page): if it's still
> -		 * in use, then the inode cannot have been freed yet, and we
> -		 * can safely proceed (if it's no longer in use, that tells
> -		 * nothing about the inode, but we don't need to unuse swap).
> -		 */
> -		if (!page_swapcount(*pagep))
> -			error = -ENOENT;
> -	}
> +/* TODO some pages with a null index are slipping through
> + * and being passed to shmem_unuse_inode_index
> + */
> +static int shmem_unuse_inode(struct shmem_inode_info *info, unsigned int type){
> +	struct address_space *mapping = info->vfs_inode.i_mapping;
> +	void **slot;
> +	struct radix_tree_iter iter;
> +	int error = 0;
>  
> -	/*
> -	 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
> -	 * but also to hold up shmem_evict_inode(): so inode cannot be freed
> -	 * beneath us (pagelock doesn't help until the page is in pagecache).
> -	 */
> -	if (!error)
> -		error = shmem_add_to_page_cache(*pagep, mapping, index,
> -						GFP_NOWAIT, radswap);
> -	if (error != -ENOMEM) {
> -		/*
> -		 * Truncation and eviction use free_swap_and_cache(), which
> -		 * only does trylock page: if we raced, best clean up here.
> -		 */
> -		delete_from_swap_cache(*pagep);
> -		set_page_dirty(*pagep);
> -		if (!error) {
> -			spin_lock(&info->lock);
> -			info->swapped--;
> -			spin_unlock(&info->lock);
> -			swap_free(swap);
> +	rcu_read_lock();
> +	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, 0){
> +		struct page *page;
> +		pgoff_t index;
> +		swp_entry_t entry;
> +		unsigned int stype;
> +
> +		index = iter.index;
> +		page = radix_tree_deref_slot(slot);
> +		if (unlikely(!page))
> +			continue;
> +		if (radix_tree_exceptional_entry(page)) {
> +			entry = radix_to_swp_entry(page);
> +			stype = swp_type(entry);
> +			if (stype == type){
> +				error = shmem_unuse_inode_index(info, index);

Oh dear.  It was a nice idea to use shmem_getpage_gfp() to do all
the work, including allocating page of memory and reading into it from
swap; but here you call it between rcu_read_lock() and rcu_read_unlock().

(I'm surprised that you didn't have debug config options turned on,
which would have warned you about that.  CONFIG_DEBUG_KERNEL, DEBUG_VM,
SCHED_DEBUG, DEBUG_PREEMPT, DEBUG_SPINLOCK, DEBUG_MUTEXES, ATOMIC_SLEEP,
PROVE_LOCKING, PROVE_RCU are particularly useful when reorganizing code
in this way.  Though some of those, notably PROVE_LOCKING, have quite
a high cost, that you'd prefer to turn off when doing measurements.
And beware that none of them can see the page bit-spin-lock.)

Sorry, no, that will have to be restructured.  I think what you will
want to end up with is allocating a batch of pages upfront (perhaps
SWAP_CLUSTER_MAX 32, but falling back to 1 to make forward progress)
while holding no locks or mutexes, then pass that batch down for use 
by shmem_getpage_gfp-replacement and swapin_readahead-replacement.

I doubt it would turn out well to modify shmem_getpage_gfp() to handle
that (though just today I see Kirill Shutemov beginning to think of
faulting multiple shmem pages at once: perhaps later on you could share
his work): I think you'd do better to make a copy of it, then edit it
down to the swap parts you need, then change it to draw from the batch
of pages passed in (which will need to use an extension or modified
copy of swapin_readahead() too).

Of course, you can't read from swap under the rcu_read_lock() either:
that radix_tree_for_each_slot() will probably have to be converted to
a pagevec kind of interface, where you get the radix-tree to fill you
a vector of exceptional entries (of the "type" you want, I suppose:
even though that would be a layering violation), before you get to
work on them after the rcu_read_unlock().

Hmm, perhaps I'm wrong to send you down that path of a preallocated
batch of pages: maybe you can forget that idea, and continue to use
shmem_getpage_gfp() and swapin_readahead() as is, one page at a
time from the entries in this vector.  If you can safely drop
shmem_swaplist_mutex also across that work, it should be fine;
and if you cannot, it may still be good enough for a prototype.

> +			}
>  		}
> -		error = 1;	/* not an error, but entry was found */
> +		if (error)
> +		break;
>  	}
> +	rcu_read_unlock();
>  	return error;
>  }
>  
> -/*
> - * Search through swapped inodes to find and replace swap by page.
> +/* unuse all the shared memory swap entries that
> + * have backing store in the designated swap type.
>   */
> -int shmem_unuse(swp_entry_t swap, struct page *page)
> +int shmem_unuse(unsigned int type)
>  {
>  	struct list_head *this, *next;
>  	struct shmem_inode_info *info;
> -	int found = 0;
>  	int error = 0;
>  
> -	/*
> -	 * There's a faint possibility that swap page was replaced before
> -	 * caller locked it: caller will come back later with the right page.
> -	 */
> -	if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
> -		goto out;
> -
> -	/*
> -	 * Charge page using GFP_KERNEL while we can wait, before taking
> -	 * the shmem_swaplist_mutex which might hold up shmem_writepage().
> -	 * Charged back to the user (not to caller) when swap account is used.
> -	 */
> -	error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
> -	if (error)
> -		goto out;
> -	/* No radix_tree_preload: swap entry keeps a place for page in tree */
> -
>  	mutex_lock(&shmem_swaplist_mutex);
>  	list_for_each_safe(this, next, &shmem_swaplist) {
>  		info = list_entry(this, struct shmem_inode_info, swaplist);
>  		if (info->swapped)
> -			found = shmem_unuse_inode(info, swap, &page);
> +			error = shmem_unuse_inode(info, type);
>  		else
>  			list_del_init(&info->swaplist);
>  		cond_resched();
> -		if (found)
> +		if (error)
>  			break;
>  	}
>  	mutex_unlock(&shmem_swaplist_mutex);
> -
> -	if (found < 0)
> -		error = found;
> -out:
> -	unlock_page(page);
> -	page_cache_release(page);
>  	return error;
>  }
>  
> @@ -2873,7 +2833,7 @@ int __init shmem_init(void)
>  	return 0;
>  }
>  
> -int shmem_unuse(swp_entry_t swap, struct page *page)
> +int shmem_unuse(unsigned int type)
>  {
>  	return 0;
>  }
> diff --git a/mm/swapfile.c b/mm/swapfile.c
> index 4a7f7e6..b69e319 100644
> --- a/mm/swapfile.c
> +++ b/mm/swapfile.c
> @@ -68,6 +68,9 @@ static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
>  /* Activity counter to indicate that a swapon or swapoff has occurred */
>  static atomic_t proc_poll_event = ATOMIC_INIT(0);
>  
> +/* count instances of unuse_pte for the changelog */
> +static long unusepte_calls = 0;
> +
>  static inline unsigned char swap_count(unsigned char ent)
>  {
>  	return ent & ~SWAP_HAS_CACHE;	/* may include SWAP_HAS_CONT flag */
> @@ -1167,13 +1170,18 @@ out_nolock:
>  
>  static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
>  				unsigned long addr, unsigned long end,
> -				swp_entry_t entry, struct page *page)
> +				unsigned int type)
>  {
> -	pte_t swp_pte = swp_entry_to_pte(entry);
> +	struct page * page;
> +	swp_entry_t entry;
> +	unsigned int found_type;
>  	pte_t *pte;
>  	int ret = 0;
>  
> +	unusepte_calls++;
> +
>  	/*
> +	 * TODO comment left from original:
>  	 * We don't actually need pte lock while scanning for swp_pte: since
>  	 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
>  	 * page table while we're scanning; though it could get zapped, and on
> @@ -1184,16 +1192,71 @@ static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
>  	 */
>  	pte = pte_offset_map(pmd, addr);
>  	do {
> +		if (is_swap_pte(*pte)){
> +			entry = pte_to_swp_entry(*pte);
> +			found_type = swp_type(entry);
> +		}
> +		else {
> +			continue;
> +		}
> +		if (found_type == type){
> +			entry = pte_to_swp_entry(*pte);
> +			page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
> +					vma, addr);

A similar problem when we come here.  You are calling swapin_readahead()
in between pte_offset_map() and pte_unmap().  It's less surprising that
went undetected, since it will pass unnoticed on x86_64 (or x86_32
without CONFIG_HIGHMEM and CONFIG_HIGHPTE).  If you're using x86_64
for your testing, you may prefer not to set up another partition or
VM with 32-bit: I think if you hack your pte_offset_map() and pte_unmap()
to include preempt_disable() and preempt_enable() respectively, that
should give you the same debug warnings (along with the config options
suggested before - I think CONFIG_ATOMIC_SLEEP is the relevant one).

> +			if (!page){
> +				/* TODO not sure yet what to do here, or
> +				 * how great the chance of the page
> +				 * not existing actually is.
> +				 * There is a comment in try_to_unuse
> +				 * about the page possibly being freed
> +				 * independently, etc
> +				 */
> +				printk("unuse_pte tried to swap in an invalid page\n");
> +				continue;
> +			}
>  		/*
> -		 * swapoff spends a _lot_ of time in this loop!
> -		 * Test inline before going to call unuse_pte.
> +		 * Wait for and lock page.  When do_swap_page races with
> +		 * try_to_unuse, do_swap_page can handle the fault much
> +		 * faster than try_to_unuse can locate the entry.  This
> +		 * apparently redundant "wait_on_page_locked" lets try_to_unuse
> +		 * defer to do_swap_page in such a case - in some tests,
> +		 * do_swap_page and try_to_unuse repeatedly compete.
>  		 */
> -		if (unlikely(maybe_same_pte(*pte, swp_pte))) {
> +			wait_on_page_locked(page);
> +			wait_on_page_writeback(page);
> +			lock_page(page);
> +			wait_on_page_writeback(page);
>  			pte_unmap(pte);

And of course that pte_unmap(pte) needs to be before the
wait_on_page_locked(page).

>  			ret = unuse_pte(vma, pmd, addr, entry, page);
> -			if (ret)
> -				goto out;
> -			pte = pte_offset_map(pmd, addr);
> +		/* TODO fix
> +		 * in the new way, we unuse
> +		 * all ptes in the range or fail before returning.
> +		 * For now, leave the return from unuse_pte as is,
> +		 * move on and unuse the next pte.
> +		 */
> +		if (ret < 1){
> +			/* TODO for now, we're just returning
> +			 * the error if unuse_pte fails.
> +			 * we need to clean up the allocated page,
> +			 * plus all the rest of the mess
> +			 */
> +			unlock_page(page);
> +			goto out;
> +		}
> +		/*
> +		 * TODO moved here from try_to_unuse--still relevant?:
> +		 * It is conceivable that a racing task removed this page from
> +		 * swap cache just before we acquired the page lock at the top,
> +		 * or while we dropped it in unuse_mm().  The page might even
> +		 * be back in swap cache on another swap area: that we must not
> +		 * delete, since it may not have been written out to swap yet.
> +		 */
> +		if (PageSwapCache(page) &&
> +		    likely(page_private(page) == entry.val))
> +			delete_from_swap_cache(page);
> +		SetPageDirty(page);
> +		unlock_page(page);
> +		page_cache_release(page);
>  		}
>  	} while (pte++, addr += PAGE_SIZE, addr != end);
>  	pte_unmap(pte - 1);
> @@ -1203,7 +1266,7 @@ out:
>  
>  static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
>  				unsigned long addr, unsigned long end,
> -				swp_entry_t entry, struct page *page)
> +				unsigned int type)
>  {
>  	pmd_t *pmd;
>  	unsigned long next;
> @@ -1214,8 +1277,8 @@ static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
>  		next = pmd_addr_end(addr, end);
>  		if (pmd_none_or_trans_huge_or_clear_bad(pmd))
>  			continue;
> -		ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
> -		if (ret)
> +		ret = unuse_pte_range(vma, pmd, addr, next, type);
> +		if (ret < 0)
>  			return ret;
>  	} while (pmd++, addr = next, addr != end);
>  	return 0;
> @@ -1223,7 +1286,7 @@ static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
>  
>  static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
>  				unsigned long addr, unsigned long end,
> -				swp_entry_t entry, struct page *page)
> +				unsigned int type)
>  {
>  	pud_t *pud;
>  	unsigned long next;
> @@ -1234,67 +1297,52 @@ static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
>  		next = pud_addr_end(addr, end);
>  		if (pud_none_or_clear_bad(pud))
>  			continue;
> -		ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
> -		if (ret)
> +		ret = unuse_pmd_range(vma, pud, addr, next, type);
> +		if (ret < 0)
>  			return ret;
>  	} while (pud++, addr = next, addr != end);
>  	return 0;
>  }
>  
> -static int unuse_vma(struct vm_area_struct *vma,
> -				swp_entry_t entry, struct page *page)
> +static int unuse_vma(struct vm_area_struct *vma, unsigned int type)
>  {
>  	pgd_t *pgd;
>  	unsigned long addr, end, next;
>  	int ret;
>  
> -	if (page_anon_vma(page)) {
> -		addr = page_address_in_vma(page, vma);
> -		if (addr == -EFAULT)
> -			return 0;
> -		else
> -			end = addr + PAGE_SIZE;
> -	} else {
> -		addr = vma->vm_start;
> -		end = vma->vm_end;
> -	}
> +	addr = vma->vm_start;
> +	end = vma->vm_end;
>  
>  	pgd = pgd_offset(vma->vm_mm, addr);
>  	do {
>  		next = pgd_addr_end(addr, end);
>  		if (pgd_none_or_clear_bad(pgd))
>  			continue;
> -		ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
> -		if (ret)
> +		ret = unuse_pud_range(vma, pgd, addr, next, type);
> +		if (ret < 0)
>  			return ret;
>  	} while (pgd++, addr = next, addr != end);
>  	return 0;
>  }
>  
> -static int unuse_mm(struct mm_struct *mm,
> -				swp_entry_t entry, struct page *page)
> +static int unuse_mm(struct mm_struct *mm, unsigned int type)
>  {
>  	struct vm_area_struct *vma;
>  	int ret = 0;
>  
> -	if (!down_read_trylock(&mm->mmap_sem)) {
> -		/*
> -		 * Activate page so shrink_inactive_list is unlikely to unmap
> -		 * its ptes while lock is dropped, so swapoff can make progress.
> -		 */
> -		activate_page(page);
> -		unlock_page(page);
> -		down_read(&mm->mmap_sem);
> -		lock_page(page);
> -	}
> +	down_read(&mm->mmap_sem);

Correct, you don't have page lock so you don't need down_read_trylock().

You may encounter some surprises through not holding page lock across
all this (of course you don't even have a page to lock here): it did
serialize against various races, which may surface elsewhere in your
implementation; but it's too early to think about those yet, we can
worry about them later when the correct framework is complete.

>  	for (vma = mm->mmap; vma; vma = vma->vm_next) {
> -		if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
> +		if (vma->anon_vma && (ret = unuse_vma(vma, type)))
>  			break;
>  	}
>  	up_read(&mm->mmap_sem);
>  	return (ret < 0)? ret: 0;
>  }
>  
> +/* TODO: this whole function is no longer necessary 
> + * useful for checking that the swap area is clean,
> + * so leaving until these changes are submitted
> + */
>  /*
>   * Scan swap_map (or frontswap_map if frontswap parameter is true)
>   * from current position to next entry still in use.
> @@ -1341,29 +1389,31 @@ static unsigned int find_next_to_unuse(struct swap_info_struct *si,
>  }
>  
>  /*
> - * We completely avoid races by reading each swap page in advance,
> - * and then search for the process using it.  All the necessary
> - * page table adjustments can then be made atomically.
> - *
>   * if the boolean frontswap is true, only unuse pages_to_unuse pages;
>   * pages_to_unuse==0 means all pages; ignored if frontswap is false
>   */
>  int try_to_unuse(unsigned int type, bool frontswap,
>  		 unsigned long pages_to_unuse)
>  {
> -	struct swap_info_struct *si = swap_info[type];
>  	struct mm_struct *start_mm;
> -	volatile unsigned char *swap_map; /* swap_map is accessed without
> -					   * locking. Mark it as volatile
> -					   * to prevent compiler doing
> -					   * something odd.
> -					   */
> -	unsigned char swcount;
> -	struct page *page;
> -	swp_entry_t entry;
> -	unsigned int i = 0;
> +	struct mm_struct *mm;
> +	struct list_head *p;
>  	int retval = 0;
>  
> +	/* TODO for checking if any entries are left
> +	 * after swapoff finishes
> +	 * for debug purposes, remove before submitting */
> +	struct swap_info_struct *si = swap_info[type];
> +	int i = 0;
> +
> +	/* TODO shmem_unuse needs its housekeeping
> +	 * exactly what needs to be done is not yet
> +	 * determined
> +	 */
> +	retval = shmem_unuse(type);
> +	if (retval)
> +		goto out;
> +
>  	/*
>  	 * When searching mms for an entry, a good strategy is to
>  	 * start at the first mm we freed the previous entry from
> @@ -1380,48 +1430,17 @@ int try_to_unuse(unsigned int type, bool frontswap,
>  	 */
>  	start_mm = &init_mm;
>  	atomic_inc(&init_mm.mm_users);
> +	p = &start_mm->mmlist;
>  
> -	/*
> -	 * Keep on scanning until all entries have gone.  Usually,
> -	 * one pass through swap_map is enough, but not necessarily:
> -	 * there are races when an instance of an entry might be missed.
> +	/* TODO: why do we protect the mmlist? (noob QUESTION)
> +	 * Where should the locks actually go?
>  	 */
> -	while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
> +	spin_lock(&mmlist_lock);
> +	while (!retval && (p = p->next) != &start_mm->mmlist) {
>  		if (signal_pending(current)) {
>  			retval = -EINTR;
>  			break;
>  		}
> -
> -		/*
> -		 * Get a page for the entry, using the existing swap
> -		 * cache page if there is one.  Otherwise, get a clean
> -		 * page and read the swap into it.
> -		 */
> -		swap_map = &si->swap_map[i];
> -		entry = swp_entry(type, i);
> -		page = read_swap_cache_async(entry,
> -					GFP_HIGHUSER_MOVABLE, NULL, 0);
> -		if (!page) {
> -			/*
> -			 * Either swap_duplicate() failed because entry
> -			 * has been freed independently, and will not be
> -			 * reused since sys_swapoff() already disabled
> -			 * allocation from here, or alloc_page() failed.
> -			 */
> -			swcount = *swap_map;
> -			/*
> -			 * We don't hold lock here, so the swap entry could be
> -			 * SWAP_MAP_BAD (when the cluster is discarding).
> -			 * Instead of fail out, We can just skip the swap
> -			 * entry because swapoff will wait for discarding
> -			 * finish anyway.
> -			 */
> -			if (!swcount || swcount == SWAP_MAP_BAD)
> -				continue;
> -			retval = -ENOMEM;
> -			break;
> -		}
> -
>  		/*
>  		 * Don't hold on to start_mm if it looks like exiting.
>  		 */
> @@ -1431,80 +1450,36 @@ int try_to_unuse(unsigned int type, bool frontswap,
>  			atomic_inc(&init_mm.mm_users);
>  		}
>  
> -		/*
> -		 * Wait for and lock page.  When do_swap_page races with
> -		 * try_to_unuse, do_swap_page can handle the fault much
> -		 * faster than try_to_unuse can locate the entry.  This
> -		 * apparently redundant "wait_on_page_locked" lets try_to_unuse
> -		 * defer to do_swap_page in such a case - in some tests,
> -		 * do_swap_page and try_to_unuse repeatedly compete.
> -		 */
> -		wait_on_page_locked(page);
> -		wait_on_page_writeback(page);
> -		lock_page(page);
> -		wait_on_page_writeback(page);
> +		mm = list_entry(p, struct mm_struct, mmlist);
> +		if (!atomic_inc_not_zero(&mm->mm_users))
> +			continue;
> +		spin_unlock(&mmlist_lock);
> +
> +		cond_resched();
> +
> +		retval = unuse_mm(mm, type);
> +		mmput(mm);
> +		if (retval)
> +			break;
>  
>  		/*
> -		 * Remove all references to entry.
> +		 * Make sure that we aren't completely killing
> +		 * interactive performance.
>  		 */
> -		swcount = *swap_map;
> -		if (swap_count(swcount) == SWAP_MAP_SHMEM) {
> -			retval = shmem_unuse(entry, page);
> -			/* page has already been unlocked and released */
> -			if (retval < 0)
> +		cond_resched();
> +		/* TODO we need another way to count these,
> +		 * because we will now be unusing all an mm's pages
> +		 * on each pass through the loop
> +		 * Ignoring frontswap for now

Quite right, ignore frontswap for now.

> +		 */
> +		if (frontswap && pages_to_unuse > 0) {
> +			if (!--pages_to_unuse)
>  				break;
> -			continue;
>  		}
> -		if (swap_count(swcount) && start_mm != &init_mm)
> -			retval = unuse_mm(start_mm, entry, page);
> -
> -		if (swap_count(*swap_map)) {
> -			int set_start_mm = (*swap_map >= swcount);
> -			struct list_head *p = &start_mm->mmlist;
> -			struct mm_struct *new_start_mm = start_mm;
> -			struct mm_struct *prev_mm = start_mm;
> -			struct mm_struct *mm;
> -
> -			atomic_inc(&new_start_mm->mm_users);
> -			atomic_inc(&prev_mm->mm_users);
> -			spin_lock(&mmlist_lock);
> -			while (swap_count(*swap_map) && !retval &&
> -					(p = p->next) != &start_mm->mmlist) {
> -				mm = list_entry(p, struct mm_struct, mmlist);
> -				if (!atomic_inc_not_zero(&mm->mm_users))
> -					continue;
> -				spin_unlock(&mmlist_lock);
> -				mmput(prev_mm);
> -				prev_mm = mm;
>  
> -				cond_resched();
> -
> -				swcount = *swap_map;
> -				if (!swap_count(swcount)) /* any usage ? */
> -					;
> -				else if (mm == &init_mm)
> -					set_start_mm = 1;
> -				else
> -					retval = unuse_mm(mm, entry, page);
> -
> -				if (set_start_mm && *swap_map < swcount) {
> -					mmput(new_start_mm);
> -					atomic_inc(&mm->mm_users);
> -					new_start_mm = mm;
> -					set_start_mm = 0;
> -				}
> -				spin_lock(&mmlist_lock);
> -			}
> -			spin_unlock(&mmlist_lock);
> -			mmput(prev_mm);
> -			mmput(start_mm);
> -			start_mm = new_start_mm;
> -		}
> -		if (retval) {
> -			unlock_page(page);
> -			page_cache_release(page);
> -			break;
> -		}
> +		spin_lock(&mmlist_lock);
> +	}
> +	spin_unlock(&mmlist_lock);
>  
>  		/*
>  		 * If a reference remains (rare), we would like to leave
> @@ -1524,50 +1499,27 @@ int try_to_unuse(unsigned int type, bool frontswap,
>  		 * this splitting happens to be just what is needed to
>  		 * handle where KSM pages have been swapped out: re-reading
>  		 * is unnecessarily slow, but we can fix that later on.
> +		 * TODO move this to unuse_pte_range?
>  		 */
> -		if (swap_count(*swap_map) &&
> -		     PageDirty(page) && PageSwapCache(page)) {
> -			struct writeback_control wbc = {
> -				.sync_mode = WB_SYNC_NONE,
> -			};
> -
> -			swap_writepage(page, &wbc);
> -			lock_page(page);
> -			wait_on_page_writeback(page);
> -		}
> -
> -		/*
> -		 * It is conceivable that a racing task removed this page from
> -		 * swap cache just before we acquired the page lock at the top,
> -		 * or while we dropped it in unuse_mm().  The page might even
> -		 * be back in swap cache on another swap area: that we must not
> -		 * delete, since it may not have been written out to swap yet.
> -		 */
> -		if (PageSwapCache(page) &&
> -		    likely(page_private(page) == entry.val))
> -			delete_from_swap_cache(page);
> -
> -		/*
> -		 * So we could skip searching mms once swap count went
> -		 * to 1, we did not mark any present ptes as dirty: must
> -		 * mark page dirty so shrink_page_list will preserve it.
> -		 */
> -		SetPageDirty(page);
> -		unlock_page(page);
> -		page_cache_release(page);
> -
> -		/*
> -		 * Make sure that we aren't completely killing
> -		 * interactive performance.
> -		 */
> -		cond_resched();
> -		if (frontswap && pages_to_unuse > 0) {
> -			if (!--pages_to_unuse)
> -				break;
> -		}
> -	}
> +/*		if (swap_count(*swap_map) &&
> +*		     PageDirty(page) && PageSwapCache(page)) {
> +*			struct writeback_control wbc = {
> +*				.sync_mode = WB_SYNC_NONE,
> +*			};
> +*
> +*			swap_writepage(page, &wbc);
> +*			lock_page(page);
> +*			wait_on_page_writeback(page);
> +*		}
> +*/
>  
> +	/* TODO check if there are any swap entries we failed to clean up. */
> +	if ((i = find_next_to_unuse(si, i, frontswap)) != 0)
> +		printk("swap entries remain, type not clean\n");
> +	printk("Leaving try_to_unuse\n");
> +	printk("Calls made to unuse_pte: %lu\n", unusepte_calls);
>  	mmput(start_mm);
> +out:
>  	return retval;
>  }
>  
> -- 
> 1.8.3.2

Good luck with the next version!

Hugh
--
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