Re: [PATCH] virtio_balloon: high order allocation

[Liu Shixin <liushixin2@huawei.com>]

Hi, Soichiro:

I'm interested with your patch. I'm looking at it and found some possible issues.


On 2022/12/23 17:35, Soichiro Ueda wrote:
> +			for (i = 0; i < num_pfns; i++) {
> +				set_page_pfns(vb, vb->pfns + vb->num_pfns + i,
> +					      nth_page(page, i));
> +			}
Since the interval of vb->num_pfns is VIRTIO_BALLOON_PAGES_PER_PAGE,
it seems that the pfns[] should be vb->pfns + vb->num_pfns + i * VIRTIO_BALLOON_PAGES_PER_PAGE.
> +			vb->num_pages +=
> +				num_pfns * VIRTIO_BALLOON_PAGES_PER_PAGE;
> +			if (!virtio_has_feature(
> +				    vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
> +				adjust_managed_page_count(page, -num_pfns);
num_pfns is of the unsigned int type so need be to converted to long manually.

root@syzkaller:~# free -m
              total        used        free      shared  buff/cache   available
Mem:           1975          78        1822           2          74        1869
Swap:             0           0           0
root@syzkaller:~# QEMU 7.2.0 monitor - type 'help' for more information
(qemu) info balloon
balloon: actual=2048
(qemu) balloon 2046
(qemu)

root@syzkaller:~# free -m
              total        used        free      shared  buff/cache   available
Mem:       33556405    33554510        1819           2          74        1867
Swap:             0           0           0
> +
> +		num_pfns = (1 << order);
> +		for (i = 0; i < num_pfns; i++)
> +			set_page_pfns(vb, vb->pfns + vb->num_pfns + i, nth_page(page, i));
The same problem mentioned above.


thanks,

Liu Shixin

.