Re: [Qemu-devel] [RFC 5/5] virtio-balloon: Safely handle BALLOON_PAGE_SIZE < host page size

[David Hildenbrand <david@redhat.com>]

On 12/10/2018 05:24, David Gibson wrote:
> The virtio-balloon always works in units of 4kiB (BALLOON_PAGE_SIZE), but
> on the host side, we can only actually discard memory in units of the host
> page size.
> 
> At present we handle this very badly: we silently ignore balloon requests
> that aren't host page aligned, and for requests that are host page aligned
> we discard the entire host page.  The latter potentially corrupts guest
> memory if its page size is smaller than the host's.
> 
> We could just disable the balloon if the host page size is not 4kiB, but
> that would break a the special case where host and guest have the same page
> size, but that's larger than 4kiB.  Thius case currently works by accident:
> when the guest puts its page into the balloon, it will submit balloon
> requests for each 4kiB subpage.  Most will be ignored, but the one which
> happens to be host page aligned will discard the whole lot.
> 
> This occurs in practice routinely for POWER KVM systems, since both host
> and guest typically use 64kiB pages.
> 
> To make this safe, without breaking that useful case, we need to
> accumulate 4kiB balloon requests until we have a whole contiguous host page
> at which point we can discard it.
> 
> We could in principle do that across all guest memory, but it would require
> a large bitmap to track.  This patch represents a compromise: instead we
> track ballooned subpages for a single contiguous host page at a time.  This
> means that if the guest discards all 4kiB chunks of a host page in
> succession, we will discard it.  In particular that means the balloon will
> continue to work for the (host page size) == (guest page size) > 4kiB case.
> 
> If the guest scatters 4kiB requests across different host pages, we don't
> discard anything, and issue a warning.  Not ideal, but at least we don't
> corrupt guest memory as the previous version could.
> 
> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
> ---
>  hw/virtio/virtio-balloon.c         | 67 +++++++++++++++++++++++++-----
>  include/hw/virtio/virtio-balloon.h |  3 ++
>  2 files changed, 60 insertions(+), 10 deletions(-)
> 
> diff --git a/hw/virtio/virtio-balloon.c b/hw/virtio/virtio-balloon.c
> index 4435905c87..39573ef2e3 100644
> --- a/hw/virtio/virtio-balloon.c
> +++ b/hw/virtio/virtio-balloon.c
> @@ -33,33 +33,80 @@
>  
>  #define BALLOON_PAGE_SIZE  (1 << VIRTIO_BALLOON_PFN_SHIFT)
>  
> +typedef struct PartiallyBalloonedPage {
> +    RAMBlock *rb;
> +    ram_addr_t base;
> +    unsigned long bitmap[];

BTW, it might be easier to only remember the last inflated page and
incrementing it when you see the successor.

initialize last_page to -1ull on realize/reset


if (QEMU_IS_ALIGNED(addr, PAGE_SIZE)) {
	/* start of a new potential page block */
	last_page == addr;
} else if (addr == last_page + BALLOON_PAGE_SIZE) {
	/* next successor */
	last_page == addr;
	if (QEMU_IS_ALIGNED(last_page + BALLOON_PAGE_SIZE, PAGE_SIZE)) {
		ramblock_discard()....
	}
} else {
	last_page = -1ull;
}


> +} PartiallyBalloonedPage;
> +
>  static void balloon_inflate_page(VirtIOBalloon *balloon,
>                                   MemoryRegion *mr, hwaddr offset)
>  {
>      void *addr = memory_region_get_ram_ptr(mr) + offset;
>      RAMBlock *rb;
>      size_t rb_page_size;
> -    ram_addr_t ram_offset;
> +    int subpages;
> +    ram_addr_t ram_offset, host_page_base;
>  
>      /* XXX is there a better way to get to the RAMBlock than via a
>       * host address? */
>      rb = qemu_ram_block_from_host(addr, false, &ram_offset);
>      rb_page_size = qemu_ram_pagesize(rb);
> +    host_page_base = ram_offset & ~(rb_page_size - 1);
> +
> +    if (rb_page_size == BALLOON_PAGE_SIZE) {
> +        /* Easy case */
>  
> -    /* Silently ignore hugepage RAM blocks */
> -    if (rb_page_size != getpagesize()) {
> +        ram_block_discard_range(rb, ram_offset, rb_page_size);
> +        /* We ignore errors from ram_block_discard_range(), because it
> +         * has already reported them, and failing to discard a balloon
> +         * page is not fatal */
>          return;
>      }
>  
> -    /* Silently ignore unaligned requests */
> -    if (ram_offset & (rb_page_size - 1)) {
> -        return;
> +    /* Hard case
> +     *
> +     * We've put a piece of a larger host page into the balloon - we
> +     * need to keep track until we have a whole host page to
> +     * discard
> +     */
> +    subpages = rb_page_size / BALLOON_PAGE_SIZE;
> +
> +    if (balloon->pbp
> +        && (rb != balloon->pbp->rb
> +            || host_page_base != balloon->pbp->base)) {
> +        /* We've partially ballooned part of a host page, but now
> +         * we're trying to balloon part of a different one.  Too hard,
> +         * give up on the old partial page */
> +        warn_report("Unable to insert a partial page into virtio-balloon");
> +        free(balloon->pbp);
> +        balloon->pbp = NULL;
>      }
>  
> -    ram_block_discard_range(rb, ram_offset, rb_page_size);
> -    /* We ignore errors from ram_block_discard_range(), because it has
> -     * already reported them, and failing to discard a balloon page is
> -     * not fatal */
> +    if (!balloon->pbp) {
> +        /* Starting on a new host page */
> +        size_t bitlen = BITS_TO_LONGS(subpages) * sizeof(unsigned long);
> +        balloon->pbp = g_malloc0(sizeof(PartiallyBalloonedPage) + bitlen);
> +        balloon->pbp->rb = rb;
> +        balloon->pbp->base = host_page_base;
> +    }
> +
> +    bitmap_set(balloon->pbp->bitmap,
> +               (ram_offset - balloon->pbp->base) / BALLOON_PAGE_SIZE,
> +               subpages);
> +
> +    if (bitmap_full(balloon->pbp->bitmap, subpages)) {
> +        /* We've accumulated a full host page, we can actually discard
> +         * it now */
> +
> +        ram_block_discard_range(rb, balloon->pbp->base, rb_page_size);
> +        /* We ignore errors from ram_block_discard_range(), because it
> +         * has already reported them, and failing to discard a balloon
> +         * page is not fatal */
> +
> +        free(balloon->pbp);
> +        balloon->pbp = NULL;
> +    }
>  }
>  
>  static const char *balloon_stat_names[] = {
> diff --git a/include/hw/virtio/virtio-balloon.h b/include/hw/virtio/virtio-balloon.h
> index e0df3528c8..99dcd6d105 100644
> --- a/include/hw/virtio/virtio-balloon.h
> +++ b/include/hw/virtio/virtio-balloon.h
> @@ -30,6 +30,8 @@ typedef struct virtio_balloon_stat_modern {
>         uint64_t val;
>  } VirtIOBalloonStatModern;
>  
> +typedef struct PartiallyBalloonedPage PartiallyBalloonedPage;
> +
>  typedef struct VirtIOBalloon {
>      VirtIODevice parent_obj;
>      VirtQueue *ivq, *dvq, *svq;
> @@ -42,6 +44,7 @@ typedef struct VirtIOBalloon {
>      int64_t stats_last_update;
>      int64_t stats_poll_interval;
>      uint32_t host_features;
> +    PartiallyBalloonedPage *pbp;
>  } VirtIOBalloon;
>  
>  #endif
> 


-- 

Thanks,

David / dhildenb