Message-ID: <20121226034600.GB2453@blaptop>
Date:	Wed, 26 Dec 2012 12:46:00 +0900
From:	Minchan Kim <minchan@...nel.org>
To:	Kamezawa Hiroyuki <kamezawa.hiroyu@...fujitsu.com>
Cc:	Andrew Morton <akpm@...ux-foundation.org>,
	linux-kernel@...r.kernel.org, linux-mm@...ck.org,
	Michael Kerrisk <mtk.manpages@...il.com>,
	Arun Sharma <asharma@...com>, sanjay@...gle.com,
	Paul Turner <pjt@...gle.com>,
	David Rientjes <rientjes@...gle.com>,
	John Stultz <john.stultz@...aro.org>,
	Christoph Lameter <cl@...ux.com>,
	Android Kernel Team <kernel-team@...roid.com>,
	Robert Love <rlove@...gle.com>, Mel Gorman <mel@....ul.ie>,
	Hugh Dickins <hughd@...gle.com>,
	Dave Hansen <dave@...ux.vnet.ibm.com>,
	Rik van Riel <riel@...hat.com>,
	Dave Chinner <david@...morbit.com>, Neil Brown <neilb@...e.de>,
	Mike Hommey <mh@...ndium.org>, Taras Glek <tglek@...illa.com>,
	KOSAKI Motohiro <kosaki.motohiro@...il.com>
Subject: Re: [RFC v4 0/3] Support volatile for anonymous range

Hi Kame,

What are you doing these holiday season? :)
I can't believe you sit down in front of computer.

On Wed, Dec 26, 2012 at 11:37:02AM +0900, Kamezawa Hiroyuki wrote:
> (2012/12/18 15:47), Minchan Kim wrote:
> > This is still RFC because we need more input from user-space
> > people and discussion about interface/reclaim policy of volatile
> > pages and I want to expand this concept to tmpfs volatile range
> > if it is possbile without big performance drop of anonymous volatile
> > rnage (Let's define our term. anon volatile VS tmpfs volatile? John?)
> > 
> > NOTE: I didn't consider THP/KSM so for test, you should disable them.
> > 
> > I hope more inputs from user-space allocator people and test patch
> > with their allocator because it might need design change of arena
> > management for getting real vaule.
> > 
> > Changelog from v4
> > 
> >   * Add new system call mvolatile/mnovolatile
> >   * Add sigbus when user try to access volatile range
> >   * Rebased on v3.7
> >   * Applied bug fix from John Stultz, Thanks!
> > 
> > Changelog from v3
> > 
> >   * Removing madvise(addr, length, MADV_NOVOLATILE).
> >   * add vmstat about the number of discarded volatile pages
> >   * discard volatile pages without promotion in reclaim path
> > 
> > This is based on v3.7
> > 
> > - What's the mvolatile(addr, length)?
> > 
> >    It's a hint that user deliver to kernel so kernel can *discard*
> >    pages in a range anytime.
> > 
> 
> This can work against both of PRIVATE and SHARED mapping  ?

Yes.

> 
> What happens at fork() ? VOLATILE ranges are copied ?

Just child vma would have a VM_VOLATILE flag.
If a page is shared like above, the page could be discarded only when
all vmas pointing to the page are VM_VOLATILE.

> 
> 
> > - What happens if user access page(ie, virtual address) discarded
> >    by kernel?
> > 
> >    The user can encounter SIGBUS.
> > 
> > - What should user do for avoding SIGBUS?
> >    He should call mnovolatie(addr, length) before accessing the range
> >    which was called by mvolatile.
> > 
> Will mnovolatile() return whether the range is discarded or not ?

Absolutely.

> 
> What the user should do in signal handler ?

It depends on usecase.
Please read John's mail. http://lwn.net/Articles/518130/
Quote from the link
"
But one interesting new tweak on this design, suggested by the Taras
Glek and others at Mozilla, is as follows:

Instead of leaving volatile data access as being undefined , when
accessing volatile data, either the data expected will be returned
if it has not been purged, or the application will get a SIGBUS when
it accesses volatile data that has been purged.

Everything else remains the same (error on marking non-volatile
if data was purged, etc). This model allows applications to avoid
having to unmark volatile data when it wants to access it, then
immediately re-mark it as volatile when its done. It is in effect
"lazy" with its marking, allowing the kernel to hit it with a signal
when it gets unlucky and touches purged data. From the signal handler,
the application can note the address it faulted on, unmark the range,
and regenerate the needed data before returning to execution.

Since this approach avoids the more explicit unmark/access/mark
pattern, it avoids the extra overhead required to ensure data is
non-volatile before being accessed.

However, If applications don't want to deal with handling the
sigbus, they can use the more straightforward (but more costly)
unmark/access/mark pattern in the same way as my earlier proposals.

This allows folks to balance the cost vs complexity in their
application appropriately.

So that's a general overview of how the idea I'm proposing could
be used.
"

> Can the all expected opereations be done in signal-safe manner ?
> (IOW, can user do enough job easily without taking any locks in userland ?)

It depends on design of user application but some user space guys want
it so it could be done enoughly, I think. Expecially, Android have used it
by ashmem where was another interface for same goal but it works only tmpfs pages
but mine is normal anonymous page but the goal is to support both.

> 
> > - What happens if user access page(ie, virtual address) doesn't
> >    discarded by kernel?
> > 
> >    The user can see old data without page fault.
> > 
> 
> What happens when ther user calls mvolatile() against mlock()'d range or
> calling mlock() against mvolatile()'d range ?

-EINVAL

> 
> Hm, by the way, the user need to attach pages to the process by causing page-fault
> (as you do by memset()) before calling mvolatile() ?

For effectiveness, Yes.

> 
> I think your approach is interesting, anyway.

Thanks for your interest, Kame.

あけましておめでとう.

> 
> Thanks,
> -Kame
> 
> 
> > - What's different with madvise(DONTNEED)?
> > 
> >    System call semantic
> > 
> >    DONTNEED makes sure user always can see zero-fill pages after
> >    he calls madvise while mvolatile can see old data or encounter
> >    SIGBUS.
> > 
> >    Internal implementation
> > 
> >    The madvise(DONTNEED) should zap all mapped pages in range so
> >    overhead is increased linearly with the number of mapped pages.
> >    Even, if user access zapped pages as write mode, page fault +
> >    page allocation + memset should be happened.
> > 
> >    The mvolatile just marks the flag in a range(ie, VMA) instead of
> >    zapping all of pte in the vma so it doesn't touch ptes any more.
> > 
> > - What's the benefit compared to DONTNEED?
> > 
> >    1. The system call overhead is smaller because mvolatile just marks
> >       the flag to VMA instead of zapping all the page in a range so
> >       overhead should be very small.
> > 
> >    2. It has a chance to eliminate overheads (ex, zapping pte + page fault
> >       + page allocation + memset(PAGE_SIZE)) if memory pressure isn't
> >       severe.
> > 
> >    3. It has a potential to zap all ptes and free the pages if memory
> >       pressure is severe so reclaim overhead could be disappear - TODO
> > 
> > - Isn't there any drawback?
> > 
> >    Madvise(DONTNEED) doesn't need exclusive mmap_sem so concurrent page
> >    fault of other threads could be allowed. But m[no]volatile needs
> >    exclusive mmap_sem so other thread would be blocked if they try to
> >    access not-yet-mapped pages. That's why I design m[no]volatile
> >    overhead should be small as far as possible.
> > 
> >    It could suffer from max rss usage increasement because madvise(DONTNEED)
> >    deallocates pages instantly when the system call is issued while mvoatile
> >    delays it until memory pressure happens so if memory pressure is severe by
> >    max rss incresement, system would suffer. First of all, allocator needs
> >    some balance logic for that or kernel might handle it by zapping pages
> >    although user calls mvolatile if memory pressure is severe.
> >    The problem is how we know memory pressure is severe.
> >    One of solution is to see kswapd is active or not. Another solution is
> >    Anton's mempressure so allocator can handle it.
> > 
> > - What's for targetting?
> > 
> >    Firstly, user-space allocator like ptmalloc, tcmalloc or heap management
> >    of virtual machine like Dalvik. Also, it comes in handy for embedded
> >    which doesn't have swap device so they can't reclaim anonymous pages.
> >    By discarding instead of swapout, it could be used in the non-swap system.
> >    For it, we have to age anon lru list although we don't have swap because
> >    I don't want to discard volatile pages by top priority when memory pressure
> >    happens as volatile in this patch means "We don't need to swap out because
> >    user can handle the situation which data are disappear suddenly", NOT
> >    "They are useless so hurry up to reclaim them". So I want to apply same
> >    aging rule of nomal pages to them.
> > 
> >    Anonymous page background aging of non-swap system would be a trade-off
> >    for getting good feature. Even, we had done it two years ago until merge
> >    [1] and I believe gain of this patch will beat loss of anon lru aging's
> >    overead once all of allocator start to use madvise.
> >    (This patch doesn't include background aging in case of non-swap system
> >    but it's trivial if we decide)
> > 
> >    As another choice, we can zap the range like madvise(DONTNEED) when mvolatile
> >    is called if we don't have swap space.
> > 
> > - Stupid performance test
> >    I attach test program/script which are utter crap and I don't expect
> >    current smart allocator never have done it so we need more practical data
> >    with real allocator.
> > 
> >    KVM - 8 core, 2G
> > 
> > VOLATILE test
> > 13.16user 7.58system 0:06.04elapsed 343%CPU (0avgtext+0avgdata 2624096maxresident)k
> > 0inputs+0outputs (0major+164050minor)pagefaults 0swaps
> > 
> > DONTNEED test
> > 23.30user 228.92system 0:33.10elapsed 762%CPU (0avgtext+0avgdata 213088maxresident)k
> > 0inputs+0outputs (0major+16384210minor)pagefaults 0swaps
> > 
> >    x86-64 - 12 core, 2G
> > 
> > VOLATILE test
> > 33.38user 0.44system 0:02.87elapsed 1178%CPU (0avgtext+0avgdata 3935008maxresident)k
> > 0inputs+0outputs (0major+245989minor)pagefaults 0swaps
> > 
> > DONTNEED test
> > 28.02user 41.25system 0:05.80elapsed 1192%CPU (0avgtext+0avgdata 387776maxresident)k
> > 
> > [1] 74e3f3c3, vmscan: prevent background aging of anon page in no swap system
> > 
> > Any comments are welcome!
> > 
> > Cc: Michael Kerrisk <mtk.manpages@...il.com>
> > Cc: Arun Sharma <asharma@...com>
> > Cc: sanjay@...gle.com
> > Cc: Paul Turner <pjt@...gle.com>
> > CC: David Rientjes <rientjes@...gle.com>
> > Cc: John Stultz <john.stultz@...aro.org>
> > Cc: Andrew Morton <akpm@...ux-foundation.org>
> > Cc: Christoph Lameter <cl@...ux.com>
> > Cc: Android Kernel Team <kernel-team@...roid.com>
> > Cc: Robert Love <rlove@...gle.com>
> > Cc: Mel Gorman <mel@....ul.ie>
> > Cc: Hugh Dickins <hughd@...gle.com>
> > Cc: Dave Hansen <dave@...ux.vnet.ibm.com>
> > Cc: Rik van Riel <riel@...hat.com>
> > Cc: Dave Chinner <david@...morbit.com>
> > Cc: Neil Brown <neilb@...e.de>
> > Cc: Mike Hommey <mh@...ndium.org>
> > Cc: Taras Glek <tglek@...illa.com>
> > Cc: KOSAKI Motohiro <kosaki.motohiro@...il.com>
> > Cc: Christoph Lameter <cl@...ux.com>
> > Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>
> > 
> > Minchan Kim (3):
> >    Introduce new system call mvolatile
> >    Discard volatile page
> >    add PGVOLATILE vmstat count
> > 
> >   arch/x86/syscalls/syscall_64.tbl |    3 +-
> >   include/linux/mm.h               |    1 +
> >   include/linux/mm_types.h         |    2 +
> >   include/linux/rmap.h             |    3 +
> >   include/linux/syscalls.h         |    2 +
> >   include/linux/vm_event_item.h    |    2 +-
> >   mm/Makefile                      |    4 +-
> >   mm/huge_memory.c                 |    9 +-
> >   mm/ksm.c                         |    3 +-
> >   mm/memory.c                      |    2 +
> >   mm/migrate.c                     |    6 +-
> >   mm/mlock.c                       |    5 +-
> >   mm/mmap.c                        |    2 +-
> >   mm/mvolatile.c                   |  396 ++++++++++++++++++++++++++++++++++++++
> >   mm/rmap.c                        |   97 +++++++++-
> >   mm/vmscan.c                      |    4 +
> >   mm/vmstat.c                      |    1 +
> >   17 files changed, 527 insertions(+), 15 deletions(-)
> >   create mode 100644 mm/mvolatile.c
> > 
> > ================== 8< =============================
> > 
> > #define _GNU_SOURCE
> > #include <stdio.h>
> > #include <pthread.h>
> > #include <sched.h>
> > #include <sys/mman.h>
> > #include <sys/types.h>
> > #include <stdlib.h>
> > #include <string.h>
> > #include <unistd.h>
> > #include <sys/syscall.h>
> > 
> > #define SYS_mvolatile 313
> > #define SYS_mnovolatile 314
> > 
> > #define ALLOC_SIZE (8 << 20)
> > #define MAP_SIZE  (ALLOC_SIZE * 10)
> > #define PAGE_SIZE (1 << 12)
> > #define RETRY 100
> > 
> > pthread_barrier_t barrier;
> > int mode;
> > #define VOLATILE_MODE 1
> > 
> > static int mvolatile(void *addr, size_t length)
> > {
> > 	return syscall(SYS_mvolatile, addr, length);
> > }
> > 
> > static int mnovolatile(void *addr, size_t length)
> > {
> > 	return syscall(SYS_mnovolatile, addr, length);
> > }
> > 
> > void *thread_entry(void *data)
> > {
> > 	unsigned long i;
> > 	cpu_set_t set;
> > 	int cpu = *(int*)data;
> > 	void *mmap_area;
> > 	int retry = RETRY;
> > 
> > 	CPU_ZERO(&set);
> > 	CPU_SET(cpu, &set);
> > 	sched_setaffinity(0, sizeof(set), &set);
> > 
> > 	mmap(NULL, PAGE_SIZE, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS, 0, 0);
> > 	mmap_area = mmap(NULL, MAP_SIZE, PROT_READ|PROT_WRITE,
> > 					MAP_PRIVATE|MAP_ANONYMOUS, 0, 0);
> > 	if (mmap_area == MAP_FAILED) {
> > 		fprintf(stderr, "Fail to mmap [%d]\n", *(int*)data);
> > 		exit(1);
> > 	}
> > 
> > 	pthread_barrier_wait(&barrier);
> > 
> > 	while(retry--) {
> > 		if (mode == VOLATILE_MODE) {
> > 			mvolatile(mmap_area, MAP_SIZE);
> > 			for (i = 0; i < MAP_SIZE; i+= ALLOC_SIZE) {
> > 				mnovolatile(mmap_area + i, ALLOC_SIZE);
> > 				memset(mmap_area + i, i, ALLOC_SIZE);
> > 				mvolatile(mmap_area + i, ALLOC_SIZE);
> > 			}
> > 		} else {
> > 			for (i = 0; i < MAP_SIZE; i += ALLOC_SIZE) {
> > 				memset(mmap_area + i, i, ALLOC_SIZE);
> > 				madvise(mmap_area + i, ALLOC_SIZE, MADV_DONTNEED);
> > 			}
> > 		}
> > 	}
> > 	return NULL;
> > }
> > 
> > int main(int argc, char *argv[])
> > {
> > 	int i, nr_thread;
> > 	int *data;
> > 
> > 	if (argc < 3)
> > 		return 1;
> > 
> > 	nr_thread = atoi(argv[1]);
> > 	mode = atoi(argv[2]);
> > 
> > 	pthread_t *thread = malloc(sizeof(pthread_t) * nr_thread);
> > 	data = malloc(sizeof(int) * nr_thread);
> > 	pthread_barrier_init(&barrier, NULL, nr_thread);
> > 
> > 	for (i = 0; i < nr_thread; i++) {
> > 		data[i] = i;
> > 		if (pthread_create(&thread[i], NULL, thread_entry, &data[i])) {
> > 			perror("Fail to create thread\n");
> > 			exit(1);
> > 		}
> > 	}
> > 
> > 	for (i = 0; i < nr_thread; i++) {
> > 		if (pthread_join(thread[i], NULL))
> > 			perror("Fail to join thread\n");
> > 		printf("[%d] thread done\n", i);
> > 	}
> > 
> > 	return 0;
> > }
> > 
> 
> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@...ck.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@...ck.org"> email@...ck.org </a>

-- 
Kind regards,
Minchan Kim
--
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