Message-ID: <20121122233444.GE5121@bbox>
Date:	Fri, 23 Nov 2012 08:34:44 +0900
From:	Minchan Kim <minchan@...nel.org>
To:	Jerome Marchand <jmarchan@...hat.com>
Cc:	Nitin Gupta <ngupta@...are.org>,
	Seth Jennings <sjenning@...ux.vnet.ibm.com>,
	linux-mm@...ck.org, linux-kernel@...r.kernel.org,
	Pekka Enberg <penberg@...helsinki.fi>,
	Konrad Rzeszutek Wilk <konrad@...nok.org>,
	Dan Magenheimer <dan.magenheimer@...cle.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Greg Kroah-Hartman <gregkh@...uxfoundation.org>
Subject: Re: Lockdep complain for zram

On Thu, Nov 22, 2012 at 12:13:24PM +0100, Jerome Marchand wrote:
> On 11/21/2012 09:37 AM, Minchan Kim wrote:
> > Hi alls,
> > 
> > Today, I saw below complain of lockdep.
> > As a matter of fact, I knew it long time ago but forgot that.
> > The reason lockdep complains is that now zram uses GFP_KERNEL
> > in reclaim path(ex, __zram_make_request) :(
> > I can fix it via replacing GFP_KERNEL with GFP_NOIO.
> > But more big problem is vzalloc in zram_init_device which calls GFP_KERNEL.
> > Of course, I can change it with __vmalloc which can receive gfp_t.
> > But still we have a problem. Althoug __vmalloc can handle gfp_t, it calls
> > allocation of GFP_KERNEL. That's why I sent the patch.
> > https://lkml.org/lkml/2012/4/23/77
> > Since then, I forgot it, saw the bug today and poped the question again.
> > 
> > Yes. Fundamental problem is utter crap API vmalloc.
> > If we can fix it, everyone would be happy. But life isn't simple like seeing
> > my thread of the patch.
> > 
> > So next option is to move zram_init_device into setting disksize time.
> > But it makes unnecessary metadata waste until zram is used really(That's why
> > Nitin move zram_init_device from disksize setting time to make_request) and
> > it makes user should set the disksize before using, which are behavior change.
> > 
> > I would like to clean up this issue before promoting because it might change
> > usage behavior.
> > 
> > Do you have any idea?
> 
> This is a false positive due to the memory allocation in
> zram_init_device() called from zram_make_request(). It appears to
> lockdep that the allocation might trigger a request on the device that
> would try to take init_lock again, but in fact it doesn't. The device
> is not initialized yet, even less swapped on.

That's not a only swap case.
Let's think following usecase.

1) Booting
2) echo $((DISKSIZE)) > /sys/block/zram0/disksize
3) dd if=/dev/zero of=/dev/zram0 bs=4K count=1
4) Written 4K page(page-A) is still page cache and isn't submitted
   to zram block device.
5) Memory pressure happen by some memory hogger.
6) VM start to reclaim and write page-A to zram0.
7) zram_init_device is called at last.
8) allocate GFP_KERNEL in zram_init_device
9) goto reclaim path again.
10) deadlock.

So I think it's not false positive.
Even if it is, I think lock split isn't a good idea to just avoid
lockdep warn. It makes code unnecessary complicated and it would be more
error-prone. Let's not add another lock without performance trouble report
by the lock.

As I discussed with Nitin in this thread, lazy initialization don't have
much point and disksize setting option isn't consistent for user behavior.
And I expect Nitin will send patch "diet of table" soonish.

So just moving the initialzation part from reclaim context to process's one
is simple and clear solution, I believe.

> 
> The following (quickly tested) patch should prevent lockdep complain.  
> 
> Jerome
> 
> ---
> >From ebb3514c4ee18276da7c5ca08025991b493ac204 Mon Sep 17 00:00:00 2001
> From: Jerome Marchand <jmarchan@...hat.com>
> Date: Thu, 22 Nov 2012 09:07:40 +0100
> Subject: [PATCH] staging: zram: Avoid lockdep warning
> 
> zram triggers a lockdep warning. The cause of it is the call to
> zram_init_device() from zram_make_request(). The memory allocation in
> zram_init_device() could start a memory reclaim which in turn could
> cause swapout and (as it appears to lockdep) a call to
> zram_make_request(). However this is a false positive: an
> unititialized device can't be used as swap.
> A solution is to split init_lock in two lock. One mutex that protects
> init, reset and size setting and a rw_semaphore that protects requests
> and reset. Thus init and request would be protected by different locks
> and lockdep will be happy.
> 
> Signed-off-by: Jerome Marchand <jmarchan@...hat.com>
> ---
>  drivers/staging/zram/zram_drv.c   |   41 +++++++++++++++++++-----------------
>  drivers/staging/zram/zram_drv.h   |   16 ++++++++++---
>  drivers/staging/zram/zram_sysfs.c |   20 +++++++++---------
>  3 files changed, 44 insertions(+), 33 deletions(-)
> 
> diff --git a/drivers/staging/zram/zram_drv.c b/drivers/staging/zram/zram_drv.c
> index fb4a7c9..b3bc3c4 100644
> --- a/drivers/staging/zram/zram_drv.c
> +++ b/drivers/staging/zram/zram_drv.c
> @@ -470,11 +470,11 @@ static void zram_make_request(struct request_queue *queue, struct bio *bio)
>  {
>  	struct zram *zram = queue->queuedata;
>  
> -	if (unlikely(!zram->init_done) && zram_init_device(zram))
> +	if (unlikely(!is_initialized(zram)) && zram_init_device(zram))
>  		goto error;
>  
> -	down_read(&zram->init_lock);
> -	if (unlikely(!zram->init_done))
> +	down_read(&zram->req_lock);
> +	if (unlikely(!is_initialized(zram)))
>  		goto error_unlock;
>  
>  	if (!valid_io_request(zram, bio)) {
> @@ -483,12 +483,12 @@ static void zram_make_request(struct request_queue *queue, struct bio *bio)
>  	}
>  
>  	__zram_make_request(zram, bio, bio_data_dir(bio));
> -	up_read(&zram->init_lock);
> +	up_read(&zram->req_lock);
>  
>  	return;
>  
>  error_unlock:
> -	up_read(&zram->init_lock);
> +	up_read(&zram->req_lock);
>  error:
>  	bio_io_error(bio);
>  }
> @@ -497,7 +497,7 @@ void __zram_reset_device(struct zram *zram)
>  {
>  	size_t index;
>  
> -	zram->init_done = 0;
> +	atomic_set(&zram->init_done, 0);
>  
>  	/* Free various per-device buffers */
>  	kfree(zram->compress_workmem);
> @@ -529,9 +529,12 @@ void __zram_reset_device(struct zram *zram)
>  
>  void zram_reset_device(struct zram *zram)
>  {
> -	down_write(&zram->init_lock);
> -	__zram_reset_device(zram);
> -	up_write(&zram->init_lock);
> +	mutex_lock(&zram->init_lock);
> +	down_write(&zram->req_lock);
> +	if (is_initialized(zram))
> +		__zram_reset_device(zram);
> +	up_write(&zram->req_lock);
> +	mutex_unlock(&zram->init_lock);
>  }
>  
>  int zram_init_device(struct zram *zram)
> @@ -539,10 +542,10 @@ int zram_init_device(struct zram *zram)
>  	int ret;
>  	size_t num_pages;
>  
> -	down_write(&zram->init_lock);
> +	mutex_lock(&zram->init_lock);
>  
> -	if (zram->init_done) {
> -		up_write(&zram->init_lock);
> +	if (is_initialized(zram)) {
> +		mutex_unlock(&zram->init_lock);
>  		return 0;
>  	}
>  
> @@ -583,8 +586,8 @@ int zram_init_device(struct zram *zram)
>  		goto fail;
>  	}
>  
> -	zram->init_done = 1;
> -	up_write(&zram->init_lock);
> +	atomic_set(&zram->init_done, 1);
> +	mutex_unlock(&zram->init_lock);
>  
>  	pr_debug("Initialization done!\n");
>  	return 0;
> @@ -594,7 +597,7 @@ fail_no_table:
>  	zram->disksize = 0;
>  fail:
>  	__zram_reset_device(zram);
> -	up_write(&zram->init_lock);
> +	mutex_unlock(&zram->init_lock);
>  	pr_err("Initialization failed: err=%d\n", ret);
>  	return ret;
>  }
> @@ -619,7 +622,8 @@ static int create_device(struct zram *zram, int device_id)
>  	int ret = 0;
>  
>  	init_rwsem(&zram->lock);
> -	init_rwsem(&zram->init_lock);
> +	mutex_init(&zram->init_lock);
> +	init_rwsem(&zram->req_lock);
>  	spin_lock_init(&zram->stat64_lock);
>  
>  	zram->queue = blk_alloc_queue(GFP_KERNEL);
> @@ -672,7 +676,7 @@ static int create_device(struct zram *zram, int device_id)
>  		goto out;
>  	}
>  
> -	zram->init_done = 0;
> +	atomic_set(&zram->init_done, 0);
>  
>  out:
>  	return ret;
> @@ -755,8 +759,7 @@ static void __exit zram_exit(void)
>  		zram = &zram_devices[i];
>  
>  		destroy_device(zram);
> -		if (zram->init_done)
> -			zram_reset_device(zram);
> +		zram_reset_device(zram);
>  	}
>  
>  	unregister_blkdev(zram_major, "zram");
> diff --git a/drivers/staging/zram/zram_drv.h b/drivers/staging/zram/zram_drv.h
> index df2eec4..f6bcead 100644
> --- a/drivers/staging/zram/zram_drv.h
> +++ b/drivers/staging/zram/zram_drv.h
> @@ -96,9 +96,12 @@ struct zram {
>  				   * against concurrent read and writes */
>  	struct request_queue *queue;
>  	struct gendisk *disk;
> -	int init_done;
> -	/* Prevent concurrent execution of device init, reset and R/W request */
> -	struct rw_semaphore init_lock;
> +	atomic_t init_done;
> +	/* Prevent concurrent execution of device init, reset and
> +	 * disksize_store */
> +	struct mutex init_lock;
> +	/* Prevent concurent execution device reset and R/W requests */
> +	struct rw_semaphore req_lock;
>  	/*
>  	 * This is the limit on amount of *uncompressed* worth of data
>  	 * we can store in a disk.
> @@ -108,6 +111,11 @@ struct zram {
>  	struct zram_stats stats;
>  };
>  
> +static inline int is_initialized(struct zram *zram)
> +{
> +	return atomic_read(&zram->init_done);
> +}
> +
>  extern struct zram *zram_devices;
>  unsigned int zram_get_num_devices(void);
>  #ifdef CONFIG_SYSFS
> @@ -115,6 +123,6 @@ extern struct attribute_group zram_disk_attr_group;
>  #endif
>  
>  extern int zram_init_device(struct zram *zram);
> -extern void __zram_reset_device(struct zram *zram);
> +extern void zram_reset_device(struct zram *zram);
>  
>  #endif
> diff --git a/drivers/staging/zram/zram_sysfs.c b/drivers/staging/zram/zram_sysfs.c
> index de1eacf..b300881 100644
> --- a/drivers/staging/zram/zram_sysfs.c
> +++ b/drivers/staging/zram/zram_sysfs.c
> @@ -62,16 +62,19 @@ static ssize_t disksize_store(struct device *dev,
>  	if (!disksize)
>  		return -EINVAL;
>  
> -	down_write(&zram->init_lock);
> -	if (zram->init_done) {
> -		up_write(&zram->init_lock);
> +	mutex_lock(&zram->init_lock);
> +	down_write(&zram->req_lock);
> +	if (is_initialized(zram)) {
> +		up_write(&zram->req_lock);
> +		mutex_unlock(&zram->init_lock);
>  		pr_info("Cannot change disksize for initialized device\n");
>  		return -EBUSY;
>  	}
>  
>  	zram->disksize = PAGE_ALIGN(disksize);
>  	set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
> -	up_write(&zram->init_lock);
> +	up_write(&zram->req_lock);
> +	mutex_unlock(&zram->init_lock);
>  
>  	return len;
>  }
> @@ -81,7 +84,7 @@ static ssize_t initstate_show(struct device *dev,
>  {
>  	struct zram *zram = dev_to_zram(dev);
>  
> -	return sprintf(buf, "%u\n", zram->init_done);
> +	return sprintf(buf, "%u\n", atomic_read(&zram->init_done));
>  }
>  
>  static ssize_t reset_store(struct device *dev,
> @@ -110,10 +113,7 @@ static ssize_t reset_store(struct device *dev,
>  	if (bdev)
>  		fsync_bdev(bdev);
>  
> -	down_write(&zram->init_lock);
> -	if (zram->init_done)
> -		__zram_reset_device(zram);
> -	up_write(&zram->init_lock);
> +	zram_reset_device(zram);
>  
>  	return len;
>  }
> @@ -186,7 +186,7 @@ static ssize_t mem_used_total_show(struct device *dev,
>  	u64 val = 0;
>  	struct zram *zram = dev_to_zram(dev);
>  
> -	if (zram->init_done)
> +	if (is_initialized(zram))
>  		val = zs_get_total_size_bytes(zram->mem_pool);
>  
>  	return sprintf(buf, "%llu\n", val);
> -- 
> 1.7.7.6
> 
> --
> 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