Re: [Qemu-devel] [PATCH for-4.2 v10 07/15] virtio-iommu: Implement attach/detach command

[Auger Eric]

Hi Peter,
On 8/16/19 6:27 AM, Peter Xu wrote:
> On Tue, Jul 30, 2019 at 07:21:29PM +0200, Eric Auger wrote:
>> This patch implements the endpoint attach/detach to/from
>> a domain.
>>
>> Signed-off-by: Eric Auger <address@hidden>
>>
>> ---
>> ---
>>  hw/virtio/virtio-iommu.c | 40 ++++++++++++++++++++++++++++++++++------
>>  1 file changed, 34 insertions(+), 6 deletions(-)
>>
>> diff --git a/hw/virtio/virtio-iommu.c b/hw/virtio/virtio-iommu.c
>> index 77dccecc0a..5ea0930cc2 100644
>> --- a/hw/virtio/virtio-iommu.c
>> +++ b/hw/virtio/virtio-iommu.c
>> @@ -80,8 +80,8 @@ static void 
>> virtio_iommu_detach_endpoint_from_domain(viommu_endpoint *ep)
>>      ep->domain = NULL;
>>  }
>>  
>> -viommu_endpoint *virtio_iommu_get_endpoint(VirtIOIOMMU *s, uint32_t ep_id);
>> -viommu_endpoint *virtio_iommu_get_endpoint(VirtIOIOMMU *s, uint32_t ep_id)
> 
> These lines were just introduced in previous patch, I wanted to ask
> why the definition was needed but I don't know whether it'll be used
> in follow up patches.  Looks like it wasn't really used.
> 
> I would prefer patches like these to be squashed together not only to
> avoid the maintainance of diffs like this between patches, but also as
> a reviewer it'll be easier too when with all the contexts together.
> But I won't ask for it because it can be a personal preference only...

Yes that's a tradeoff. I tried to split the series to ease the review;
helpers were introduced separately in previous patch but not yet used.
Here I introduce call sites and they become static.
> 
>> +static viommu_endpoint *virtio_iommu_get_endpoint(VirtIOIOMMU *s,
>> +                                                  uint32_t ep_id)
>>  {
>>      viommu_endpoint *ep;
>>  
>> @@ -110,8 +110,8 @@ static void virtio_iommu_put_endpoint(gpointer data)
>>      g_free(ep);
>>  }
>>  
>> -viommu_domain *virtio_iommu_get_domain(VirtIOIOMMU *s, uint32_t domain_id);
>> -viommu_domain *virtio_iommu_get_domain(VirtIOIOMMU *s, uint32_t domain_id)
>> +static viommu_domain *virtio_iommu_get_domain(VirtIOIOMMU *s,
>> +                                              uint32_t domain_id)
>>  {
>>      viommu_domain *domain;
>>  
>> @@ -187,10 +187,27 @@ static int virtio_iommu_attach(VirtIOIOMMU *s,
>>  {
>>      uint32_t domain_id = le32_to_cpu(req->domain);
>>      uint32_t ep_id = le32_to_cpu(req->endpoint);
>> +    viommu_domain *domain;
>> +    viommu_endpoint *ep;
>>  
>>      trace_virtio_iommu_attach(domain_id, ep_id);
>>  
>> -    return VIRTIO_IOMMU_S_UNSUPP;
>> +    ep = virtio_iommu_get_endpoint(s, ep_id);
>> +    if (ep->domain) {
>> +        /*
>> +         * the device is already attached to a domain,
>> +         * detach it first
>> +         */
>> +        virtio_iommu_detach_endpoint_from_domain(ep);
> 
> Hmm... so this can be called without virtio_iommu_put_endpoint().
> Then I think we'd better move:
> 
>         g_tree_unref(ep->domain->mappings);
> 
> From virtio_iommu_put_endpoint() to inside
> virtio_iommu_detach_endpoint_from_domain() otherwise domain refs might
> leak?

I agree with you. I Also removed g_tree_destroy from
virtio_iommu_put_domain as detaching all its end points should now do
the job.

Thanks

Eric
> 
>> +    }
>> +
>> +    domain = virtio_iommu_get_domain(s, domain_id);
>> +    QLIST_INSERT_HEAD(&domain->endpoint_list, ep, next);
>> +
>> +    ep->domain = domain;
>> +    g_tree_ref(domain->mappings);
>> +
>> +    return VIRTIO_IOMMU_S_OK;
>>  }
> 
> Regards,
>