On 06.03.19 04:06, David Gibson wrote:
> Prior to f6deb6d9 "virtio-balloon: Remove unnecessary MADV_WILLNEED on
> deflate", the balloon device issued an madvise() MADV_WILLNEED on
> pages removed from the balloon. That would hint to the host kernel
> that the pages were likely to be needed by the guest in the near
> future.
>
> It's unclear if this is actually valuable or not, and so f6deb6d9
> removed this, essentially ignoring balloon deflate requests. However,
> concerns have been raised that this might cause a performance
> regression by causing extra latency for the guest in certain
> configurations.
>
> So, until we can get actual benchmark data to see if that's the case,
> this restores the old behaviour, issuing a MADV_WILLNEED when a page is
> removed from the balloon.
>
> Signed-off-by: David Gibson <da...@gibson.dropbear.id.au>
> ---
> hw/virtio/virtio-balloon.c | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
> diff --git a/hw/virtio/virtio-balloon.c b/hw/virtio/virtio-balloon.c
> index 7412bf8c85..ac36988605 100644
> --- a/hw/virtio/virtio-balloon.c
> +++ b/hw/virtio/virtio-balloon.c
> @@ -118,6 +118,8 @@ static void balloon_deflate_page(VirtIOBalloon *balloon,
> RAMBlock *rb;
> size_t rb_page_size;
> ram_addr_t ram_offset, host_page_base;
> + void *host_addr;
> + int ret;
>
> /* XXX is there a better way to get to the RAMBlock than via a
> * host address? */
> @@ -146,6 +148,17 @@ static void balloon_deflate_page(VirtIOBalloon *balloon,
> balloon->pbp = NULL;
> }
> }
> +
> + host_addr = (void *)((uintptr_t)addr & ~(rb_page_size - 1));
> +
> + /* When a page is deflated, we hint the whole host page it lives
> + * on, since we can't do anything smaller */
> + ret = qemu_madvise(host_addr, rb_page_size, QEMU_MADV_WILLNEED);
> + if (ret != 0) {
> + warn_report("Couldn't MADV_WILLNEED on balloon deflate: %s",
> + strerror(errno));
> + /* Otherwise ignore, failing to page hint shouldn't be fatal */
> + }
As far as I understand the man page, only MADV_DONTNEED will complain
about locked, huge or VM_PFNMAP pages. So I guess MADV_WILLNEED will
work for these cases (in the worst case be a NOP).
--
Thanks,
David / dhildenb
