From: Hao Chen <chenh@yusur.tech>
To: Maxime Coquelin <maxime.coquelin@redhat.com>, dev@dpdk.org
Cc: zy@yusur.tech, huangml@yusur.tech,
Chenbo Xia <chenbo.xia@intel.com>,
xiao.w.wang@intel.com, vsrivast@xilinx.com
Subject: Re: [PATCH] vhost: fix vdpa driver multi-queue initialization failure
Date: Wed, 8 Feb 2023 17:56:12 +0800 [thread overview]
Message-ID: <9ed018e6-17bf-2e8e-3349-6a52bd89cb2f@yusur.tech> (raw)
In-Reply-To: <0f2f9d8c-486b-06cb-dd79-00de982f0f82@redhat.com>
On 2023/2/3 18:38, Maxime Coquelin wrote:
>
>
> On 1/28/23 04:06, Hao Chen wrote:
>> When 'virtio_is_Ready' returns 1, 'vdpa_dev->ops->dev_conf' will be
>> called, and the vdpa driver called 'rte_vhost_get_vhost_vring' to
>> get the vring addr info from 'vhost_devices->virtqueue[]'.
>> virtio_is_ready's nr_vring is VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY(2),
>> multi-queue's nr_vring is greater than 2.
>> Only 'vhost_devices->virtqueue[0]' and 'vhost_devices->virtqueue[1]'
>> are obtained in the for loop. Other queues of multiple queues cannot
>> obtain the corresponding 'vhost_devices->virtqueue[i]', which will
>> cause 'vdpa_dev->ops->dev_conf' to obtain QEMU vring addr information
>> from 'vhost_devices->virtqueue[i]' failed.
>> Here, nr_ving is modified to dev->nr_vring, so that multiple queues
>> can obtain the corresponding 'vhost_devices->virtqueue[i]'.
>>
>> Signed-off-by: Hao Chen <chenh@yusur.tech>
>> ---
>> lib/vhost/vhost_user.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/lib/vhost/vhost_user.c b/lib/vhost/vhost_user.c
>> index 9902ae9944..9249eafc06 100644
>> --- a/lib/vhost/vhost_user.c
>> +++ b/lib/vhost/vhost_user.c
>> @@ -1473,7 +1473,7 @@ virtio_is_ready(struct virtio_net *dev)
>> if (dev->nr_vring < nr_vring)
>> return 0;
>> - for (i = 0; i < nr_vring; i++) {
>> + for (i = 0; i < dev->nr_vring; i++) {
>> vq = dev->virtqueue[i];
>> if (!vq_is_ready(dev, vq))
>
> (reiterate my reply since the patch was posted twice, and I replied to
> the one discarded in PW)
>
> Nack.
>
> We can start processing the vrings as soon as the first queue pair is
> initialized. I think you may want to rely on
> vdpa_dev->ops->set_vring_state to initialize other vrings.
>
Does this mean that the queue should be initialized and enabled in
vdpa_dev->ops->set_vring_state, not in vdpa_dev->ops->dev_conf?
> What is the vDPA driver/device affected by this issue?
After tracking the code, I found that the dpdk-vdpa example's log shows
the calling order of multiple queues as follow:
"
VHOST_USER_SET_VRING_ADDR
VHOST_USER_SET_VRING_KICK
vdpa_dev->ops->set_vring_state (vq_idx=0)
VHOST_USER_SET_VRING_CALL
vdpa_dev->ops->set_vring_state (vq_idx=0)
...
VHOST_USER_SET_VRING_ADDR
VHOST_USER_SET_VRING_KICK
vdpa_dev->ops->set_vring_state (vq_idx=1)
VHOST_USER_SET_VRING_CALL
vdpa_dev->ops->set_vring_state (vq_idx=1)
"
then virtio_is_ready return 1, vdpa_dev->ops->dev_conf is called.
After that, other queues will call vdpa_dev->ops->set_vring_state (vq_idx=i)
The dev->virtqueue[]->desc/available/used is assigned in
VHOST_USER_SET_VRING_KICK's translate_ring_addresses() , so vdpa driver
should not initialize all queues in vdpa_dev->ops->dev_conf , otherwise
other queue's dev->virtqueue[i]->desc/available/used are NULL , vring's
hpa to gpa will fail by rte_vhost_get_vhost_vring().
dev->virtqueue[i]->desc/available/used should be accessed after
queue(i)'s VHOST_USER_SET_VRING_KICK is called.
I see that all of ifc/sfc vdpa driver's queues acessed all queues's
dev->virtqueue[i]->desc/available/used in vdpa_dev->ops->dev_conf, maybe
they should move the code?
>
> Regards,
> Maxime
>
prev parent reply other threads:[~2023-02-08 9:57 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-01-28 3:06 Hao Chen
2023-01-28 3:06 ` Hao Chen
2023-02-01 12:23 ` Maxime Coquelin
2023-02-03 10:38 ` Maxime Coquelin
2023-02-08 9:56 ` Hao Chen [this message]
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=9ed018e6-17bf-2e8e-3349-6a52bd89cb2f@yusur.tech \
--to=chenh@yusur.tech \
--cc=chenbo.xia@intel.com \
--cc=dev@dpdk.org \
--cc=huangml@yusur.tech \
--cc=maxime.coquelin@redhat.com \
--cc=vsrivast@xilinx.com \
--cc=xiao.w.wang@intel.com \
--cc=zy@yusur.tech \
/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).