From: Jason Wang <jasowang@redhat.com>
To: Eugenio Perez Martin <eperezma@redhat.com>,
"Michael S. Tsirkin" <mst@redhat.com>
Cc: qemu-devel@nongnu.org, si-wei.liu@oracle.com,
Liuxiangdong <liuxiangdong5@huawei.com>,
Zhu Lingshan <lingshan.zhu@intel.com>,
"Gonglei (Arei)" <arei.gonglei@huawei.com>,
alvaro.karsz@solid-run.com, Shannon Nelson <snelson@pensando.io>,
Laurent Vivier <lvivier@redhat.com>,
Harpreet Singh Anand <hanand@xilinx.com>,
Gautam Dawar <gdawar@xilinx.com>,
Stefano Garzarella <sgarzare@redhat.com>,
Cornelia Huck <cohuck@redhat.com>, Cindy Lu <lulu@redhat.com>,
Eli Cohen <eli@mellanox.com>, Paolo Bonzini <pbonzini@redhat.com>,
Stefan Hajnoczi <stefanha@redhat.com>,
Parav Pandit <parav@mellanox.com>
Subject: Re: [RFC v2 05/13] vdpa net: add migration blocker if cannot migrate cvq
Date: Tue, 17 Jan 2023 13:42:05 +0800 [thread overview]
Message-ID: <2616f0cd-f9e8-d183-ea78-db1be4825d9c@redhat.com> (raw)
In-Reply-To: <CAJaqyWd=Lt+41Zrn70v7rN_0ociiRcvRSXuS4ZpevCUFrajEJQ@mail.gmail.com>
在 2023/1/16 17:33, Eugenio Perez Martin 写道:
> On Mon, Jan 16, 2023 at 6:24 AM Michael S. Tsirkin <mst@redhat.com> wrote:
>> On Mon, Jan 16, 2023 at 11:34:20AM +0800, Jason Wang wrote:
>>> 在 2023/1/13 15:46, Eugenio Perez Martin 写道:
>>>> On Fri, Jan 13, 2023 at 5:25 AM Jason Wang <jasowang@redhat.com> wrote:
>>>>> 在 2023/1/13 01:24, Eugenio Pérez 写道:
>>>>>> A vdpa net device must initialize with SVQ in order to be migratable,
>>>>>> and initialization code verifies conditions. If the device is not
>>>>>> initialized with the x-svq parameter, it will not expose _F_LOG so vhost
>>>>>> sybsystem will block VM migration from its initialization.
>>>>>>
>>>>>> Next patches change this. Net data VQs will be shadowed only at
>>>>>> migration time and vdpa net devices need to expose _F_LOG as long as it
>>>>>> can go to SVQ.
>>>>>>
>>>>>> Since we don't know that at initialization time but at start, add an
>>>>>> independent blocker at CVQ.
>>>>>>
>>>>>> Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
>>>>>> ---
>>>>>> net/vhost-vdpa.c | 35 +++++++++++++++++++++++++++++------
>>>>>> 1 file changed, 29 insertions(+), 6 deletions(-)
>>>>>>
>>>>>> diff --git a/net/vhost-vdpa.c b/net/vhost-vdpa.c
>>>>>> index 631424d9c4..2ca93e850a 100644
>>>>>> --- a/net/vhost-vdpa.c
>>>>>> +++ b/net/vhost-vdpa.c
>>>>>> @@ -26,12 +26,14 @@
>>>>>> #include <err.h>
>>>>>> #include "standard-headers/linux/virtio_net.h"
>>>>>> #include "monitor/monitor.h"
>>>>>> +#include "migration/blocker.h"
>>>>>> #include "hw/virtio/vhost.h"
>>>>>>
>>>>>> /* Todo:need to add the multiqueue support here */
>>>>>> typedef struct VhostVDPAState {
>>>>>> NetClientState nc;
>>>>>> struct vhost_vdpa vhost_vdpa;
>>>>>> + Error *migration_blocker;
>>>>> Any reason we can't use the mivration_blocker in vhost_dev structure?
>>>>>
>>>>> I believe we don't need to wait until start to know we can't migrate.
>>>>>
>>>> Device migratability also depends on features that the guest acks.
>>>
>>> This sounds a little bit tricky, more below:
>>>
>>>
>>>> For example, if the device does not support ASID it can be migrated as
>>>> long as _F_CVQ is not acked.
>>>
>>> The management may notice a non-consistent behavior in this case. I wonder
>>> if we can simply check the host features.
>>>
> That's right, and I can see how that can be an issue.
>
> However, the check for the ASID is based on queue indexes at the
> moment. If we want to register the blocker at the initialization
> moment the only option I see is to do two features ack & reset cycle:
> one with MQ and another one without MQ.
That's should be fine, or any issue you saw?
Thanks
>
> Would it be more correct to assume the device will assign the right
> ASID only probing one configuration? I don't think so but I'm ok to
> leave the code that way if we agree it is more viable.
>
>>> Thanks
>>
>> Yes the issue is that ack can happen after migration started.
>> I don't think this kind of blocker appearing during migration
>> is currently expected/supported well. Is it?
>>
> In that case the guest cannot DRIVER_OK the device, because the call
> to migrate_add_blocker fails and the error propagates from
> vhost_net_start up to the virtio device.
>
> But I can also see how this is inconvenient and to add a migration
> blocker at initialization can simplify things here. As long as we
> agree on the right way to probe I can send a new version that way for
> sure.
>
> Thanks!
>
>>>> Thanks!
>>>>
>>>>> Thanks
>>>>>
>>>>>
>>>>>> VHostNetState *vhost_net;
>>>>>>
>>>>>> /* Control commands shadow buffers */
>>>>>> @@ -433,9 +435,15 @@ static int vhost_vdpa_net_cvq_start(NetClientState *nc)
>>>>>> g_strerror(errno), errno);
>>>>>> return -1;
>>>>>> }
>>>>>> - if (!(backend_features & BIT_ULL(VHOST_BACKEND_F_IOTLB_ASID)) ||
>>>>>> - !vhost_vdpa_net_valid_svq_features(v->dev->features, NULL)) {
>>>>>> - return 0;
>>>>>> + if (!(backend_features & BIT_ULL(VHOST_BACKEND_F_IOTLB_ASID))) {
>>>>>> + error_setg(&s->migration_blocker,
>>>>>> + "vdpa device %s does not support ASID",
>>>>>> + nc->name);
>>>>>> + goto out;
>>>>>> + }
>>>>>> + if (!vhost_vdpa_net_valid_svq_features(v->dev->features,
>>>>>> + &s->migration_blocker)) {
>>>>>> + goto out;
>>>>>> }
>>>>>>
>>>>>> /*
>>>>>> @@ -455,7 +463,10 @@ static int vhost_vdpa_net_cvq_start(NetClientState *nc)
>>>>>> }
>>>>>>
>>>>>> if (group == cvq_group) {
>>>>>> - return 0;
>>>>>> + error_setg(&s->migration_blocker,
>>>>>> + "vdpa %s vq %d group %"PRId64" is the same as cvq group "
>>>>>> + "%"PRId64, nc->name, i, group, cvq_group);
>>>>>> + goto out;
>>>>>> }
>>>>>> }
>>>>>>
>>>>>> @@ -468,8 +479,15 @@ static int vhost_vdpa_net_cvq_start(NetClientState *nc)
>>>>>> s->vhost_vdpa.address_space_id = VHOST_VDPA_NET_CVQ_ASID;
>>>>>>
>>>>>> out:
>>>>>> - if (!s->vhost_vdpa.shadow_vqs_enabled) {
>>>>>> - return 0;
>>>>>> + if (s->migration_blocker) {
>>>>>> + Error *errp = NULL;
>>>>>> + r = migrate_add_blocker(s->migration_blocker, &errp);
>>>>>> + if (unlikely(r != 0)) {
>>>>>> + g_clear_pointer(&s->migration_blocker, error_free);
>>>>>> + error_report_err(errp);
>>>>>> + }
>>>>>> +
>>>>>> + return r;
>>>>>> }
>>>>>>
>>>>>> s0 = vhost_vdpa_net_first_nc_vdpa(s);
>>>>>> @@ -513,6 +531,11 @@ static void vhost_vdpa_net_cvq_stop(NetClientState *nc)
>>>>>> vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->status);
>>>>>> }
>>>>>>
>>>>>> + if (s->migration_blocker) {
>>>>>> + migrate_del_blocker(s->migration_blocker);
>>>>>> + g_clear_pointer(&s->migration_blocker, error_free);
>>>>>> + }
>>>>>> +
>>>>>> vhost_vdpa_net_client_stop(nc);
>>>>>> }
>>>>>>
next prev parent reply other threads:[~2023-01-17 5:42 UTC|newest]
Thread overview: 76+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-01-12 17:24 [RFC v2 00/13] Dinamycally switch to vhost shadow virtqueues at vdpa net migration Eugenio Pérez
2023-01-12 17:24 ` [RFC v2 01/13] vdpa: fix VHOST_BACKEND_F_IOTLB_ASID flag check Eugenio Pérez
2023-01-13 3:12 ` Jason Wang
2023-01-13 6:42 ` Eugenio Perez Martin
2023-01-16 3:01 ` Jason Wang
2023-01-12 17:24 ` [RFC v2 02/13] vdpa net: move iova tree creation from init to start Eugenio Pérez
2023-01-13 3:53 ` Jason Wang
2023-01-13 7:28 ` Eugenio Perez Martin
2023-01-16 3:05 ` Jason Wang
2023-01-16 9:14 ` Eugenio Perez Martin
2023-01-17 4:30 ` Jason Wang
2023-01-12 17:24 ` [RFC v2 03/13] vdpa: copy cvq shadow_data from data vqs, not from x-svq Eugenio Pérez
2023-01-12 17:24 ` [RFC v2 04/13] vdpa: rewind at get_base, not set_base Eugenio Pérez
2023-01-13 4:09 ` Jason Wang
2023-01-13 7:40 ` Eugenio Perez Martin
2023-01-16 3:32 ` Jason Wang
2023-01-16 9:53 ` Eugenio Perez Martin
2023-01-17 4:38 ` Jason Wang
2023-01-17 6:57 ` Eugenio Perez Martin
2023-01-12 17:24 ` [RFC v2 05/13] vdpa net: add migration blocker if cannot migrate cvq Eugenio Pérez
2023-01-13 4:24 ` Jason Wang
2023-01-13 7:46 ` Eugenio Perez Martin
2023-01-16 3:34 ` Jason Wang
2023-01-16 5:23 ` Michael S. Tsirkin
2023-01-16 9:33 ` Eugenio Perez Martin
2023-01-17 5:42 ` Jason Wang [this message]
2023-01-12 17:24 ` [RFC v2 06/13] vhost: delay set_vring_ready after DRIVER_OK Eugenio Pérez
2023-01-13 4:36 ` Jason Wang
2023-01-13 8:19 ` Eugenio Perez Martin
2023-01-13 9:51 ` Stefano Garzarella
2023-01-13 10:03 ` Eugenio Perez Martin
2023-01-13 10:37 ` Stefano Garzarella
2023-01-17 15:15 ` Maxime Coquelin
2023-01-16 6:36 ` Jason Wang
2023-01-16 16:16 ` Eugenio Perez Martin
2023-01-17 5:36 ` Jason Wang
2023-01-12 17:24 ` [RFC v2 07/13] vdpa: " Eugenio Pérez
2023-01-12 17:24 ` [RFC v2 08/13] vdpa: Negotiate _F_SUSPEND feature Eugenio Pérez
2023-01-13 4:39 ` Jason Wang
2023-01-13 8:45 ` Eugenio Perez Martin
2023-01-16 6:48 ` Jason Wang
2023-01-16 16:17 ` Eugenio Perez Martin
2023-01-12 17:24 ` [RFC v2 09/13] vdpa: add feature_log parameter to vhost_vdpa Eugenio Pérez
2023-01-12 17:24 ` [RFC v2 10/13] vdpa net: allow VHOST_F_LOG_ALL Eugenio Pérez
2023-01-13 4:42 ` Jason Wang
2023-01-12 17:24 ` [RFC v2 11/13] vdpa: add vdpa net migration state notifier Eugenio Pérez
2023-01-13 4:54 ` Jason Wang
2023-01-13 9:00 ` Eugenio Perez Martin
2023-01-16 6:51 ` Jason Wang
2023-01-16 15:21 ` Eugenio Perez Martin
2023-01-17 9:58 ` Dr. David Alan Gilbert
2023-01-17 10:23 ` Eugenio Perez Martin
2023-01-17 12:54 ` Dr. David Alan Gilbert
2023-02-02 1:52 ` Si-Wei Liu
2023-02-02 15:28 ` Eugenio Perez Martin
2023-02-04 2:03 ` Si-Wei Liu
2023-02-13 9:47 ` Eugenio Perez Martin
2023-02-13 22:36 ` Si-Wei Liu
2023-02-14 18:51 ` Eugenio Perez Martin
2023-02-12 14:31 ` Eli Cohen
2023-01-12 17:24 ` [RFC v2 12/13] vdpa: preemptive kick at enable Eugenio Pérez
2023-01-13 2:31 ` Jason Wang
2023-01-13 3:25 ` Zhu, Lingshan
2023-01-13 3:39 ` Jason Wang
2023-01-13 9:06 ` Eugenio Perez Martin
2023-01-16 7:02 ` Jason Wang
2023-02-02 16:55 ` Eugenio Perez Martin
2023-02-02 0:56 ` Si-Wei Liu
2023-02-02 16:53 ` Eugenio Perez Martin
2023-02-04 11:04 ` Si-Wei Liu
2023-02-05 10:00 ` Michael S. Tsirkin
2023-02-06 5:08 ` Si-Wei Liu
2023-01-12 17:24 ` [RFC v2 13/13] vdpa: Conditionally expose _F_LOG in vhost_net devices Eugenio Pérez
2023-02-02 1:00 ` [RFC v2 00/13] Dinamycally switch to vhost shadow virtqueues at vdpa net migration Si-Wei Liu
2023-02-02 11:27 ` Eugenio Perez Martin
2023-02-03 5:08 ` Si-Wei Liu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=2616f0cd-f9e8-d183-ea78-db1be4825d9c@redhat.com \
--to=jasowang@redhat.com \
--cc=alvaro.karsz@solid-run.com \
--cc=arei.gonglei@huawei.com \
--cc=cohuck@redhat.com \
--cc=eli@mellanox.com \
--cc=eperezma@redhat.com \
--cc=gdawar@xilinx.com \
--cc=hanand@xilinx.com \
--cc=lingshan.zhu@intel.com \
--cc=liuxiangdong5@huawei.com \
--cc=lulu@redhat.com \
--cc=lvivier@redhat.com \
--cc=mst@redhat.com \
--cc=parav@mellanox.com \
--cc=pbonzini@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=sgarzare@redhat.com \
--cc=si-wei.liu@oracle.com \
--cc=snelson@pensando.io \
--cc=stefanha@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).