netdev - Re: [PATCH V2 RESEND 2/2] vDPA/ifcvf: implement doorbell mapping for ifcvf

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <91c64c0c-7b78-4c41-a6d7-6d9f084c7cc5@redhat.com>
Date:   Tue, 1 Jun 2021 11:31:56 +0800
From:   Jason Wang <jasowang@...hat.com>
To:     "Zhu, Lingshan" <lingshan.zhu@...el.com>, mst@...hat.com
Cc:     virtualization@...ts.linux-foundation.org, netdev@...r.kernel.org,
        kvm@...r.kernel.org
Subject: Re: [PATCH V2 RESEND 2/2] vDPA/ifcvf: implement doorbell mapping for
 ifcvf


在 2021/6/1 上午11:27, Zhu, Lingshan 写道:
>
>
> On 5/31/2021 3:56 PM, Jason Wang wrote:
>>
>> 在 2021/5/31 下午3:33, 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@...el.com>
>>> ---
>>>   drivers/vdpa/ifcvf/ifcvf_main.c | 17 +++++++++++++++++
>>>   1 file changed, 17 insertions(+)
>>>
>>> diff --git a/drivers/vdpa/ifcvf/ifcvf_main.c 
>>> b/drivers/vdpa/ifcvf/ifcvf_main.c
>>> index ab0ab5cf0f6e..effb0e549135 100644
>>> --- a/drivers/vdpa/ifcvf/ifcvf_main.c
>>> +++ b/drivers/vdpa/ifcvf/ifcvf_main.c
>>> @@ -413,6 +413,22 @@ 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_adapter *adapter = vdpa_to_adapter(vdpa_dev);
>>> +    struct ifcvf_hw *vf = vdpa_to_vf(vdpa_dev);
>>> +    struct pci_dev *pdev = adapter->pdev;
>>> +    struct vdpa_notification_area area;
>>> +
>>> +    area.addr = vf->vring[idx].notify_pa;
>>> +    area.size = PAGE_SIZE;
>>> +    if (area.addr % PAGE_SIZE)
>>> +        IFCVF_DBG(pdev, "vq %u doorbell address is not PAGE_SIZE 
>>> aligned\n", idx);
>>
>>
>> Let's leave the decision to upper layer by: (see 
>> vp_vdpa_get_vq_notification)
>>
>> area.addr = notify_pa;
>> area.size = notify_offset_multiplier;
>>
>> Thanks
>
> Hi Jason,
>
> notify_offset_multiplier can be zero, means vqs share the same 
> doorbell address, distinguished by qid.
> and in vdpa.c:
>
>         if (vma->vm_end - vma->vm_start != notify.size)
>                 return -ENOTSUPP;
>
> so a zero size would cause this feature failure.
> mmap should work on at least a page, so if we really want "area.size = 
> notify_offset_multiplier;"
> I think we should add some code in vdpa.c, like:
>
> if(!notify.size)
>     notify.size = PAGE_SIZE;
>
> sounds good?


It's the responsibility of the driver to report a correct one. So I 
think it's better to tweak it as:

area.size = notify_offset_multiplier ?: PAGE_SIZE;

Thanks


>
> Thanks
> Zhu Lingshan
>>
>>
>>> +
>>> +    return area;
>>> +}
>>> +
>>>   /*
>>>    * IFCVF currently does't have on-chip IOMMU, so not
>>>    * implemented set_map()/dma_map()/dma_unmap()
>>> @@ -440,6 +456,7 @@ 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)
>>
>