From: Jason Wang <jasowang@redhat.com>
To: "Zhu, Lingshan" <lingshan.zhu@intel.com>, mst@redhat.com
Cc: virtualization@lists.linux-foundation.org,
netdev@vger.kernel.org, kvm@vger.kernel.org,
linux-kernel@vger.kernel.org
Subject: Re: [PATCH 2/2] vDPA/ifcvf: implement doorbell mapping for ifcvf
Date: Wed, 28 Apr 2021 21:08:35 +0800 [thread overview]
Message-ID: <bca66845-b16d-ee5e-807b-a3570e290813@redhat.com> (raw)
In-Reply-To: <f4cb4619-5634-e42d-0629-5c40f6b0dcd1@intel.com>
在 2021/4/28 下午6:20, Zhu, Lingshan 写道:
>
>
> On 4/28/2021 6:03 PM, Jason Wang wrote:
>>
>> 在 2021/4/28 下午5:56, Zhu, Lingshan 写道:
>>>
>>>
>>> On 4/28/2021 5:21 PM, Jason Wang wrote:
>>>>
>>>> 在 2021/4/28 下午4:59, Zhu, Lingshan 写道:
>>>>>
>>>>>
>>>>> On 4/28/2021 4:42 PM, Jason Wang wrote:
>>>>>>
>>>>>> 在 2021/4/28 下午4:21, Zhu Lingshan 写道:
>>>>>>> This commit implements doorbell mapping feature for ifcvf.
>>>>>>> This feature maps the notify page to userspace, to eliminate
>>>>>>> vmexit when kick a vq.
>>>>>>>
>>>>>>> Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
>>>>>>> ---
>>>>>>> drivers/vdpa/ifcvf/ifcvf_main.c | 18 ++++++++++++++++++
>>>>>>> 1 file changed, 18 insertions(+)
>>>>>>>
>>>>>>> diff --git a/drivers/vdpa/ifcvf/ifcvf_main.c
>>>>>>> b/drivers/vdpa/ifcvf/ifcvf_main.c
>>>>>>> index e48e6b74fe2e..afcb71bc0f51 100644
>>>>>>> --- a/drivers/vdpa/ifcvf/ifcvf_main.c
>>>>>>> +++ b/drivers/vdpa/ifcvf/ifcvf_main.c
>>>>>>> @@ -413,6 +413,23 @@ static int ifcvf_vdpa_get_vq_irq(struct
>>>>>>> vdpa_device *vdpa_dev,
>>>>>>> return vf->vring[qid].irq;
>>>>>>> }
>>>>>>> +static struct vdpa_notification_area
>>>>>>> ifcvf_get_vq_notification(struct vdpa_device *vdpa_dev,
>>>>>>> + u16 idx)
>>>>>>> +{
>>>>>>> + struct ifcvf_hw *vf = vdpa_to_vf(vdpa_dev);
>>>>>>> + struct vdpa_notification_area area;
>>>>>>> +
>>>>>>> + if (vf->notify_pa % PAGE_SIZE) {
>>>>>>> + area.addr = 0;
>>>>>>> + area.size = 0;
>>>>>>
>>>>>>
>>>>>> We don't need this since:
>>>>>>
>>>>>> 1) there's a check in the vhost vDPA
>>>>> I think you mean this code block in vdpa.c
>>>>> notify = ops->get_vq_notification(vdpa, index);
>>>>> if (notify.addr & (PAGE_SIZE - 1))
>>>>> return -EINVAL;
>>>>>
>>>>> This should work, however, I think the parent driver should ensure
>>>>> it passes a PAGE_SIZE aligned address to userspace, to be robust,
>>>>> to be reliable.
>>>>
>>>>
>>>> The point is parent is unaware of whether or not there's a userspace.
>>> when calling this, I think it targets a usersapce program, why
>>> kernel space need it, so IMHO no harm if we check this to keep the
>>> parent driver robust.
>>
>>
>> Again, vDPA device is unaware of what driver that is bound. It could
>> be virtio-vpda, vhost-vdpa or other in the future. It's only the vDPA
>> bus driver know how it is actually used.
>>
>>
>>>>
>>>>
>>>>>> 2) device is unaware of the bound driver, non page aligned
>>>>>> doorbell doesn't necessarily meant it can be used
>>>>> Yes, non page aligned doorbell can not be used, so there is a check.
>>>>
>>>>
>>>> Typo, what I meant is "it can't be used". That is to say, we should
>>>> let the vDPA bus driver to decide whether or not it can be used.
>>> If it is not page aligned, there would be extra complexities for
>>> vhost/qemu, I see it as a hardware defect,
>>
>>
>> It is allowed by the virtio spec, isn't it?
> The spec does not require the doorbell to be page size aligned,
> however it still a hardware defect if non page size aligned notify
> base present, I will leave a warning message here instead of the 0 value.
>
Another note is that, using PAGE_SIZE is wrong here since it varies
among archs (at most 64K on some one).
Thanks
> Thanks
> Zhu Lingshan
>>
>> Thanks
>>
>>
>>> why adapt to this kind of defects?
>>>
>>> Thanks
>>> Zhu Lingshan
>>>>
>>>> Thanks
>>>>
>>>>
>>>>>
>>>>> Thanks
>>>>> Zhu Lingshan
>>>>>>
>>>>>> Let's leave those polices to the driver.
>>>>>>
>>>>>> Thanks
>>>>>>
>>>>>>
>>>>>>> + } else {
>>>>>>> + area.addr = vf->notify_pa;
>>>>>>> + area.size = PAGE_SIZE;
>>>>>>> + }
>>>>>>> +
>>>>>>> + return area;
>>>>>>> +}
>>>>>>> +
>>>>>>> /*
>>>>>>> * IFCVF currently does't have on-chip IOMMU, so not
>>>>>>> * implemented set_map()/dma_map()/dma_unmap()
>>>>>>> @@ -440,6 +457,7 @@ static const struct vdpa_config_ops
>>>>>>> ifc_vdpa_ops ={
>>>>>>> .get_config = ifcvf_vdpa_get_config,
>>>>>>> .set_config = ifcvf_vdpa_set_config,
>>>>>>> .set_config_cb = ifcvf_vdpa_set_config_cb,
>>>>>>> + .get_vq_notification = ifcvf_get_vq_notification,
>>>>>>> };
>>>>>>> static int ifcvf_probe(struct pci_dev *pdev, const struct
>>>>>>> pci_device_id *id)
>>>>>>
>>>>>
>>>>
>>>
>>
>
next prev parent reply other threads:[~2021-04-28 13:11 UTC|newest]
Thread overview: 16+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-04-28 8:21 [PATCH 0/2] vDPA/ifcvf: implement doorbell mapping feature Zhu Lingshan
2021-04-28 8:21 ` [PATCH 1/2] vDPA/ifcvf: record virtio notify base Zhu Lingshan
2021-04-28 8:39 ` Jason Wang
2021-04-28 10:00 ` Zhu, Lingshan
2021-04-28 10:09 ` Jason Wang
2021-04-28 10:18 ` Zhu, Lingshan
2021-04-28 11:48 ` kernel test robot
2021-04-28 8:21 ` [PATCH 2/2] vDPA/ifcvf: implement doorbell mapping for ifcvf Zhu Lingshan
2021-04-28 8:42 ` Jason Wang
2021-04-28 8:59 ` Zhu, Lingshan
2021-04-28 9:21 ` Jason Wang
2021-04-28 9:56 ` Zhu, Lingshan
2021-04-28 10:03 ` Jason Wang
2021-04-28 10:20 ` Zhu, Lingshan
2021-04-28 13:08 ` Jason Wang [this message]
2021-04-29 11:26 ` Zhu, Lingshan
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=bca66845-b16d-ee5e-807b-a3570e290813@redhat.com \
--to=jasowang@redhat.com \
--cc=kvm@vger.kernel.org \
--cc=lingshan.zhu@intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mst@redhat.com \
--cc=netdev@vger.kernel.org \
--cc=virtualization@lists.linux-foundation.org \
/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).