Re: [PATCH kernel v8 2/4] virtio-balloon: VIRTIO_BALLOON_F_CHUNK_TRANSFER

[Michael S. Tsirkin]

On Wed, Apr 05, 2017 at 03:31:36AM +0000, Wang, Wei W wrote:
> On Thursday, March 16, 2017 3:09 PM Wei Wang wrote:
> > The implementation of the current virtio-balloon is not very efficient, because
> > the ballooned pages are transferred to the host one by one. Here is the
> > breakdown of the time in percentage spent on each step of the balloon inflating
> > process (inflating 7GB of an 8GB idle guest).
> > 
> > 1) allocating pages (6.5%)
> > 2) sending PFNs to host (68.3%)
> > 3) address translation (6.1%)
> > 4) madvise (19%)
> > 
> > It takes about 4126ms for the inflating process to complete.
> > The above profiling shows that the bottlenecks are stage 2) and stage 4).
> > 
> > This patch optimizes step 2) by transferring pages to the host in chunks. A chunk
> > consists of guest physically continuous pages, and it is offered to the host via a
> > base PFN (i.e. the start PFN of those physically continuous pages) and the size
> > (i.e. the total number of the pages). A chunk is formated as below:
> > 
> > --------------------------------------------------------
> > |                 Base (52 bit)        | Rsvd (12 bit) |
> > --------------------------------------------------------
> > --------------------------------------------------------
> > |                 Size (52 bit)        | Rsvd (12 bit) |
> > --------------------------------------------------------
> > 
> > By doing so, step 4) can also be optimized by doing address translation and
> > madvise() in chunks rather than page by page.
> > 
> > This optimization requires the negotiation of a new feature bit,
> > VIRTIO_BALLOON_F_CHUNK_TRANSFER.
> > 
> > With this new feature, the above ballooning process takes ~590ms resulting in
> > an improvement of ~85%.
> > 
> > TODO: optimize stage 1) by allocating/freeing a chunk of pages instead of a
> > single page each time.
> > 
> > Signed-off-by: Liang Li <liang.z.li@xxxxxxxxx>
> > Signed-off-by: Wei Wang <wei.w.wang@xxxxxxxxx>
> > Suggested-by: Michael S. Tsirkin <mst@xxxxxxxxxx>
> > ---
> >  drivers/virtio/virtio_balloon.c     | 371 +++++++++++++++++++++++++++++++++-
> > --
> >  include/uapi/linux/virtio_balloon.h |   9 +
> >  2 files changed, 353 insertions(+), 27 deletions(-)
> > 
> > diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c index
> > f59cb4f..3f4a161 100644
> > --- a/drivers/virtio/virtio_balloon.c
> > +++ b/drivers/virtio/virtio_balloon.c
> > @@ -42,6 +42,10 @@
> >  #define OOM_VBALLOON_DEFAULT_PAGES 256
> >  #define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
> > 
> > +#define PAGE_BMAP_SIZE	(8 * PAGE_SIZE)
> > +#define PFNS_PER_PAGE_BMAP	(PAGE_BMAP_SIZE * BITS_PER_BYTE)
> > +#define PAGE_BMAP_COUNT_MAX	32
> > +
> >  static int oom_pages = OOM_VBALLOON_DEFAULT_PAGES;
> > module_param(oom_pages, int, S_IRUSR | S_IWUSR);
> > MODULE_PARM_DESC(oom_pages, "pages to free on OOM"); @@ -50,6 +54,14
> > @@ MODULE_PARM_DESC(oom_pages, "pages to free on OOM");  static struct
> > vfsmount *balloon_mnt;  #endif
> > 
> > +#define BALLOON_CHUNK_BASE_SHIFT 12
> > +#define BALLOON_CHUNK_SIZE_SHIFT 12
> > +struct balloon_page_chunk {
> > +	__le64 base;
> > +	__le64 size;
> > +};
> > +
> > +typedef __le64 resp_data_t;
> >  struct virtio_balloon {
> >  	struct virtio_device *vdev;
> >  	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq; @@ -67,6 +79,31
> > @@ struct virtio_balloon {
> > 
> >  	/* Number of balloon pages we've told the Host we're not using. */
> >  	unsigned int num_pages;
> > +	/* Pointer to the response header. */
> > +	struct virtio_balloon_resp_hdr *resp_hdr;
> > +	/* Pointer to the start address of response data. */
> > +	resp_data_t *resp_data;
> 
> I think the implementation has an issue here - both the balloon pages and the unused pages use the same buffer ("resp_data" above) to store chunks. It would cause a race in this case: live migration starts while ballooning is also in progress. I plan to use separate buffers for CHUNKS_OF_BALLOON_PAGES and CHUNKS_OF_UNUSED_PAGES. Please let me know if you have a different suggestion. Thanks.
> 
> Best,
> Wei

Is only one resp data ever in flight for each kind?
If not you want as many buffers as vq allows.

-- 
MST