Hey Sergey,
On Sun, Feb 01, 2015 at 11:50:36PM +0900, Sergey Senozhatsky wrote:
> Hello Minchan,
>
> the idea looks good and this is something I was trying to do, except
> that I used kref.
>
> some review nitpicks are below. I also posted modified version of your
> patch so that will save some time.
Thanks a lot!
>
>
> > static inline int init_done(struct zram *zram)
> > {
> > - return zram->meta != NULL;
> > + return zram->disksize != 0;
>
> we don't set ->disksize to 0 when create device. and I think
> it's better to use refcount here, but set it to 0 during device creation.
> (see the patch below)
There was a reason I didn't use refcount there.
I should have written down it.
We need something to prevent further I/O handling on other CPUs.
Otherwise, it's livelock. For example, new 'A' I/O rw path on CPU 1
can see non-zero refcount if another CPU is going on rw.
Then, another new 'B' I/O rw path on CPU 2 can see non-zero refcount
if A I/O is going on. Then, another new 'C' I/O rw path on CPU 3 can
see non-zero refcount if B I/O is going on. Finally, 'A' IO is done
on CPU 1 and next I/O 'D' on CPU 1 can see non-zero refcount because
'C' on CPU 3 is going on. Infinite loop.
>
> > +static inline bool zram_meta_get(struct zram_meta *meta)
> > +{
> > + if (!atomic_inc_not_zero(&meta->refcount))
> > + return false;
> > + return true;
> > +}
>
> I've changed it to likely case first: `if recount return true'
Thanks!
>
> > static void zram_reset_device(struct zram *zram, bool reset_capacity)
> > {
> > + struct zram_meta *meta;
> > + u64 disksize;
>
> not needed. (see the patch below).
>
> > +
> > down_write(&zram->init_lock);
> >
> > zram->limit_pages = 0;
> > @@ -728,14 +750,20 @@ static void zram_reset_device(struct zram *zram, bool reset_capacity)
> > return;
> > }
> >
> > + meta = zram->meta;
> > +
> > zcomp_destroy(zram->comp);
>
> we can't destoy zcomp before we see IO completion.
True.
>
> > zram->max_comp_streams = 1;
>
> we better keep original comp_streams number before we see IO completion.
> we don't know how many RW ops we have, so completion can happen earlier.
Yeb.
>
> > - zram_meta_free(zram->meta, zram->disksize);
> > - zram->meta = NULL;
> > + disksize = zram->disksize;
> > + zram_meta_put(meta);
> > + /* Read/write handler will not handle further I/O operation. */
> > + zram->disksize = 0;
>
> I keep it on its current position. (see below)
>
> > + wait_for_completion(&meta->complete);
> > + /* I/O operation under all of CPU are done so let's free */
> > + zram_meta_free(zram->meta, disksize);
> > /* Reset stats */
> > memset(&zram->stats, 0, sizeof(zram->stats));
> >
> > - zram->disksize = 0;
> > if (reset_capacity)
> > set_capacity(zram->disk, 0);
> >
> > @@ -908,23 +936,25 @@ static void zram_make_request(struct request_queue *queue, struct bio *bio)
> > {
> > struct zram *zram = queue->queuedata;
> >
> > - down_read(&zram->init_lock);
> > - if (unlikely(!init_done(zram)))
> > + if (unlikely(!zram_meta_get(zram->meta)))
> > goto error;
> >
> > + if (unlikely(!init_done(zram)))
> > + goto put_meta;
> > +
>
> here and later:
> we can't take zram_meta_get() first and then check for init_done(zram),
> because ->meta can be NULL, so it fill be ->NULL->refcount.
True.
Actually, it was totally RFC I forgot adding the tag in the night but I can't
escape from my shame with the escuse. Thanks!
>
> let's keep ->completion and ->refcount in zram and rename zram_meta_[get|put]
> to zram_[get|put].
Good idea but still want to name it as zram_meta_get/put because zram_get naming
might confuse struct zram's refcount rather than zram_meta. :)
>
>
>
>
> please review a bit modified version of your patch.
>
> /* the patch also reogranizes a bit order of struct zram members, to move
> member that we use more often together and to avoid paddings. nothing
> critical here. */
>
>
> next action items are:
> -- we actually can now switch from init_lock in some _show() fucntion to
> zram_get()/zram_put()
> -- address that theoretical and very unlikely in real live race condition
> of umount-reset vs. mount-rw.
>
>
> no concerns about performance of this version -- we probably will not get
> any faster than that.
>
>
> thanks a lot for your effort!
>
> ---
>
> drivers/block/zram/zram_drv.c | 82 ++++++++++++++++++++++++++++++-------------
> drivers/block/zram/zram_drv.h | 17 ++++-----
> 2 files changed, 66 insertions(+), 33 deletions(-)
>
> diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
> index aa5a4c5..6916790 100644
> --- a/drivers/block/zram/zram_drv.c
> +++ b/drivers/block/zram/zram_drv.c
> @@ -44,7 +44,7 @@ static const char *default_compressor = "lzo";
> static unsigned int num_devices = 1;
>
> #define ZRAM_ATTR_RO(name) \
> -static ssize_t name##_show(struct device *d, \
> +static ssize_t name##_show(struct device *d, \
> struct device_attribute *attr, char *b) \
> { \
> struct zram *zram = dev_to_zram(d); \
> @@ -55,7 +55,7 @@ static DEVICE_ATTR_RO(name);
>
> static inline int init_done(struct zram *zram)
> {
> - return zram->meta != NULL;
> + return atomic_read(&zram->refcount);
> }
>
> static inline struct zram *dev_to_zram(struct device *dev)
> @@ -358,6 +358,23 @@ out_error:
> return NULL;
> }
>
> +static inline bool zram_get(struct zram *zram)
> +{
> + if (atomic_inc_not_zero(&zram->refcount))
> + return true;
> + return false;
> +}
> +
> +/*
> + * We want to free zram_meta in process context to avoid
> + * deadlock between reclaim path and any other locks
> + */
> +static inline void zram_put(struct zram *zram)
> +{
> + if (atomic_dec_and_test(&zram->refcount))
> + complete(&zram->io_done);
> +}
> +
> static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
> {
> if (*offset + bvec->bv_len >= PAGE_SIZE)
> @@ -719,6 +736,9 @@ static void zram_bio_discard(struct zram *zram, u32 index,
>
> static void zram_reset_device(struct zram *zram, bool reset_capacity)
> {
> + struct zram_meta *meta;
> + struct zcomp *comp;
> +
> down_write(&zram->init_lock);
>
> zram->limit_pages = 0;
> @@ -728,14 +748,21 @@ static void zram_reset_device(struct zram *zram, bool reset_capacity)
> return;
> }
>
> - zcomp_destroy(zram->comp);
> - zram->max_comp_streams = 1;
> - zram_meta_free(zram->meta, zram->disksize);
> - zram->meta = NULL;
> + meta = zram->meta;
> + comp = zram->comp;
> + /* ->refcount will go down to 0 eventually */
> + zram_put(zram);
> +
> + wait_for_completion(&zram->io_done);
> + /* I/O operation under all of CPU are done so let's free */
> + zram_meta_free(meta, disksize);
> + zcomp_destroy(comp);
> +
> /* Reset stats */
> memset(&zram->stats, 0, sizeof(zram->stats));
> -
> zram->disksize = 0;
> + zram->max_comp_streams = 1;
> +
> if (reset_capacity)
> set_capacity(zram->disk, 0);
>
> @@ -783,6 +810,8 @@ static ssize_t disksize_store(struct device *dev,
> goto out_destroy_comp;
> }
>
> + init_completion(&zram->io_done);
> + atomic_set(&zram->refcount, 1);
> zram->meta = meta;
> zram->comp = comp;
> zram->disksize = disksize;
> @@ -795,7 +824,6 @@ static ssize_t disksize_store(struct device *dev,
> * so that revalidate_disk always sees up-to-date capacity.
> */
> revalidate_disk(zram->disk);
> -
> return len;
>
> out_destroy_comp:
> @@ -908,23 +936,24 @@ static void zram_make_request(struct request_queue *queue, struct bio *bio)
> {
> struct zram *zram = queue->queuedata;
>
> - down_read(&zram->init_lock);
> - if (unlikely(!init_done(zram)))
> + if (unlikely(!zram_get(zram)))
> goto error;
>
> + if (unlikely(!init_done(zram)))
> + goto put_zram;
> +
> if (!valid_io_request(zram, bio->bi_iter.bi_sector,
> bio->bi_iter.bi_size)) {
> atomic64_inc(&zram->stats.invalid_io);
> - goto error;
> + goto put_zram;
> }
>
> __zram_make_request(zram, bio);
> - up_read(&zram->init_lock);
> -
> + zram_put(zram);
> return;
> -
> +put_zram:
> + zram_put(zram);
> error:
> - up_read(&zram->init_lock);
> bio_io_error(bio);
> }
>
> @@ -946,21 +975,22 @@ static void zram_slot_free_notify(struct block_device *bdev,
> static int zram_rw_page(struct block_device *bdev, sector_t sector,
> struct page *page, int rw)
> {
> - int offset, err;
> + int offset, err = -EIO;
> u32 index;
> struct zram *zram;
> struct bio_vec bv;
>
> zram = bdev->bd_disk->private_data;
> + if (unlikely(!zram_get(zram)))
> + goto out;
> +
> + if (unlikely(!init_done(zram)))
> + goto put_zram;
> +
> if (!valid_io_request(zram, sector, PAGE_SIZE)) {
> atomic64_inc(&zram->stats.invalid_io);
> - return -EINVAL;
> - }
> -
> - down_read(&zram->init_lock);
> - if (unlikely(!init_done(zram))) {
> - err = -EIO;
> - goto out_unlock;
> + err = -EINVAL;
> + goto put_zram;
> }
>
> index = sector >> SECTORS_PER_PAGE_SHIFT;
> @@ -971,8 +1001,9 @@ static int zram_rw_page(struct block_device *bdev, sector_t sector,
> bv.bv_offset = 0;
>
> err = zram_bvec_rw(zram, &bv, index, offset, rw);
> -out_unlock:
> - up_read(&zram->init_lock);
> +put_zram:
> + zram_put(zram);
> +out:
> /*
> * If I/O fails, just return error(ie, non-zero) without
> * calling page_endio.
> @@ -1041,6 +1072,7 @@ static int create_device(struct zram *zram, int device_id)
> int ret = -ENOMEM;
>
> init_rwsem(&zram->init_lock);
> + atomic_set(&zram->refcount, 0);
>
> zram->queue = blk_alloc_queue(GFP_KERNEL);
> if (!zram->queue) {
> diff --git a/drivers/block/zram/zram_drv.h b/drivers/block/zram/zram_drv.h
> index b05a816..7138c82 100644
> --- a/drivers/block/zram/zram_drv.h
> +++ b/drivers/block/zram/zram_drv.h
> @@ -100,24 +100,25 @@ struct zram_meta {
>
> struct zram {
> struct zram_meta *meta;
> + struct zcomp *comp;
> struct request_queue *queue;
> struct gendisk *disk;
> - struct zcomp *comp;
> -
> /* Prevent concurrent execution of device init, reset and R/W request */
> struct rw_semaphore init_lock;
> /*
> - * This is the limit on amount of *uncompressed* worth of data
> - * we can store in a disk.
> + * the number of pages zram can consume for storing compressed data
> */
> - u64 disksize; /* bytes */
> + unsigned long limit_pages;
> + atomic_t refcount;
> int max_comp_streams;
> +
> struct zram_stats stats;
> + struct completion io_done; /* notify IO under all of cpu are done */
> /*
> - * the number of pages zram can consume for storing compressed data
> + * This is the limit on amount of *uncompressed* worth of data
> + * we can store in a disk.
> */
> - unsigned long limit_pages;
> -
> + u64 disksize; /* bytes */
> char compressor[10];
> };
> #endif
>
--
Kind regards,
Minchan Kim
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>