Re: [PATCH v17 6/6] virtio-balloon: VIRTIO_BALLOON_F_FREE_PAGE_VQ

From: Wei Wang
Date: Mon Nov 20 2017 - 06:41:12 EST


On 11/14/2017 01:32 AM, Michael S. Tsirkin wrote:
You should Cc Nitesh who is working on a related feature.

OK, I'll do. We have two more issues which haven't been discussed yet, please have a check below.


On Mon, Nov 13, 2017 at 06:34:48PM +0800, Wei Wang wrote:
Ping for comments, thanks.

On 11/03/2017 04:13 PM, Wei Wang wrote:
+static void virtballoon_cmd_report_free_page_start(struct virtio_balloon *vb)
+{
+ unsigned long flags;
+
+ vb->report_free_page_stop = false;
this flag is used a lot outside any locks. Why is this safe?
Please add some comments explaining access to this flag.

I will revert the logic as suggested: vb->report_free_page. Also plan to simplify its usage as below.

The flag is set or cleared in the config handler according to the new_cmd_id given
by the host:

new_cmd_id=0: WRITE_ONCE(vb->report_free_page, false); // stop reporting
new_cmd_id != old_cmd_id: WRITE_ONCE(vb->report_free_page, true); // start reporting


The flag is read by virtio_balloon_send_free_pages() - the callback to report free pages:

if (!READ_ONCE(vb->report_free_page))
return false;

I don't find where it could be unsafe then (the flag is written by the config handler only).




+}
+
static inline s64 towards_target(struct virtio_balloon *vb)
{
s64 target;
@@ -597,42 +673,147 @@ static void update_balloon_size_func(struct work_struct *work)
queue_work(system_freezable_wq, work);
}
-static int init_vqs(struct virtio_balloon *vb)
+static bool virtio_balloon_send_free_pages(void *opaque, unsigned long pfn,
+ unsigned long nr_pages)
{
- struct virtqueue *vqs[3];
- vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
- static const char * const names[] = { "inflate", "deflate", "stats" };
- int err, nvqs;
+ struct virtio_balloon *vb = (struct virtio_balloon *)opaque;
+ void *addr = (void *)pfn_to_kaddr(pfn);
How do we know all free pages have a kaddr?

For x86_64, it works well since the kernel has all the physical memory mapped already. But for 32-bit kernel, yes, the high memory usually isn't mapped and thus no kaddr. Essentially, this pfn_to_kaddr convert isn't necessary, we do it here because the current API that virtio has is based on "struct scatterlist", which takes a kaddr, and this kaddr is then convert back to physical address in virtqueue_add() when assigning to desc->addr.

I think a better solution would be to add a new API, which directly assigns the caller's guest physical address to desc->addr, similar to the previous implementation "add_one_chunk()" (https://lists.gnu.org/archive/html/qemu-devel/2017-06/msg02452.html). But we can change that to a general virtio API:
virtqueue_add_one_desc(struct virtqueue *_vq, u64 base_addr, u32 size, bool in_desc, void *data);

What do you think?

Best,
Wei