Re: [Qemu-devel] [RFC PATCH] vhost-user: quit infinite loop while used memslots is more than the backend limit

[Jason Wang]

On 2017年12月13日 15:04, Jay Zhou wrote:
> Steps to 100% reproduce:
> (1) start a VM with two VFIO 82599 PCI NICs successfully
> (2) hot plug a RAM, which is attached successfully
> (3) hot plug another RAM, which is attached successfully
> (4) hot plug a vhost-user NIC, which is attached successfully
> (5) hot plug another vhost-user NIC, which is hanged
>
> then the qemu process infinitely and quickly print the logs below:
>      [...]
>      vhost backend memory slots limit is less than current number of present 
> memory slots
>      failed to init vhost_net for queue 0
>      vhost backend memory slots limit is less than current number of present 
> memory slots
>      failed to init vhost_net for queue 0
>      vhost backend memory slots limit is less than current number of present 
> memory slots
>      failed to init vhost_net for queue 0
>      [...]
>
> and the cpu utilization of the systemd-journal process on host is 100%,
> the reason is clear:
>      "used_memslots > hdev->vhost_ops->vhost_backend_memslots_limit(hdev)"
> is true in vhost_dev_init, and the main thread in the infinite do and
> while loop in net_vhost_user_init.
>
> One way is to increase the vhost backend memory slots limit, some
> discussions are here:
>      https://lists.gnu.org/archive/html/qemu-devel/2017-11/msg04572.html
>
> I think if
>      "used_memslots > hdev->vhost_ops->vhost_backend_memslots_limit(hdev)"
> is true, it does not make sense to try to initialize the vhost-user again
> and again, because the main thread can not handle other qmps at all,
> which means it can not reduce the number of memslots, it must be failed
> at last.

You should include the reason why we hit infinite loop in the commit 
log, and what's more important, is this issue specific to memslots only? 
What if you hit errors other than this in vhost_dev_init() for 
vhost-user? Is this better to disable event source in this case?

Thanks

> Signed-off-by: Jay Zhou <address@hidden>
> ---
>   hw/virtio/vhost.c         | 3 +++
>   include/hw/virtio/vhost.h | 2 ++
>   net/vhost-user.c          | 5 +++++
>   3 files changed, 10 insertions(+)
>
> diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c
> index e4290ce..1cc4a18 100644
> --- a/hw/virtio/vhost.c
> +++ b/hw/virtio/vhost.c
> @@ -47,6 +47,8 @@ static unsigned int used_memslots;
>   static QLIST_HEAD(, vhost_dev) vhost_devices =
>       QLIST_HEAD_INITIALIZER(vhost_devices);
>   
> +bool used_memslots_exceeded;
> +
>   bool vhost_has_free_slot(void)
>   {
>       unsigned int slots_limit = ~0U;
> @@ -1254,6 +1256,7 @@ int vhost_dev_init(struct vhost_dev *hdev, void *opaque,
>       if (used_memslots > hdev->vhost_ops->vhost_backend_memslots_limit(hdev)) {
>           error_report("vhost backend memory slots limit is less"
>                   " than current number of present memory slots");
> +        used_memslots_exceeded = true;
>           r = -1;
>           goto fail;
>       }
> diff --git a/include/hw/virtio/vhost.h b/include/hw/virtio/vhost.h
> index 467dc77..bcc66d9 100644
> --- a/include/hw/virtio/vhost.h
> +++ b/include/hw/virtio/vhost.h
> @@ -6,6 +6,8 @@
>   #include "hw/virtio/virtio.h"
>   #include "exec/memory.h"
>   
> +extern bool used_memslots_exceeded;
> +
>   /* Generic structures common for any vhost based device. */
>   struct vhost_virtqueue {
>       int kick;
> diff --git a/net/vhost-user.c b/net/vhost-user.c
> index c23927c..8b7bfff 100644
> --- a/net/vhost-user.c
> +++ b/net/vhost-user.c
> @@ -17,6 +17,7 @@
>   #include "qemu/error-report.h"
>   #include "qmp-commands.h"
>   #include "trace.h"
> +#include "include/hw/virtio/vhost.h"
>   
>   typedef struct VhostUserState {
>       NetClientState nc;
> @@ -311,6 +312,10 @@ static int net_vhost_user_init(NetClientState *peer, const 
> char *device,
>           qemu_chr_fe_set_handlers(&s->chr, NULL, NULL,
>                                    net_vhost_user_event, NULL, nc0->name, NULL,
>                                    true);
> +        if (used_memslots_exceeded) {
> +            error_report("used memslots exceeded the backend limit, quit 
> loop");
> +            return -1;
> +        }
>       } while (!s->started);
>   
>       assert(s->vhost_net);