[dpdk-dev] [PATCH 07/10] linuxapp/eal_vfio: honor iova mode before mapping

santosh santosh.shukla at caviumnetworks.com
Thu Jul 6 16:13:13 CEST 2017


On Thursday 06 July 2017 06:41 PM, Maxime Coquelin wrote:

>
> On 07/06/2017 03:08 PM, Maxime Coquelin wrote:
>>
>>
>> On 07/06/2017 01:19 PM, santosh wrote:
>>> On Thursday 06 July 2017 04:29 PM, Maxime Coquelin wrote:
>>>
>>>>
>>>> On 07/06/2017 11:49 AM, Jerin Jacob wrote:
>>>>> -----Original Message-----
>>>>>> Date: Thu, 6 Jul 2017 09:58:41 +0200
>>>>>> From: Maxime Coquelin <maxime.coquelin at redhat.com>
>>>>>> To: Jerin Jacob <jerin.jacob at caviumnetworks.com>
>>>>>> CC: Santosh Shukla <santosh.shukla at caviumnetworks.com>,
>>>>>>    thomas at monjalon.net, bruce.richardson at intel.com, dev at dpdk.org,
>>>>>>    hemant.agrawal at nxp.com, shreyansh.jain at nxp.com, gaetan.rivet at 6wind.com
>>>>>> Subject: Re: [dpdk-dev] [PATCH 07/10] linuxapp/eal_vfio: honor iova mode
>>>>>>    before mapping
>>>>>> User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
>>>>>>    Thunderbird/52.1.0
>>>>>>
>>>>>>
>>>>>>
>>>>>> On 07/05/2017 05:43 PM, Jerin Jacob wrote:
>>>>>>> -----Original Message-----
>>>>>>>> Date: Wed, 5 Jul 2017 11:14:01 +0200
>>>>>>>> From: Maxime Coquelin <maxime.coquelin at redhat.com>
>>>>>>>> To: Santosh Shukla <santosh.shukla at caviumnetworks.com>,
>>>>>>>>     thomas at monjalon.net, bruce.richardson at intel.com, dev at dpdk.org
>>>>>>>> CC: jerin.jacob at caviumnetworks.com, hemant.agrawal at nxp.com,
>>>>>>>>     shreyansh.jain at nxp.com, gaetan.rivet at 6wind.com
>>>>>>>> Subject: Re: [dpdk-dev] [PATCH 07/10] linuxapp/eal_vfio: honor iova mode
>>>>>>>>     before mapping
>>>>>>>> User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
>>>>>>>>     Thunderbird/52.1.0
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> On 06/08/2017 01:05 PM, Santosh Shukla wrote:
>>>>>>>>> Check iova mode and accordingly map iova to pa or va.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Santosh Shukla<santosh.shukla at caviumnetworks.com>
>>>>>>>>> Signed-off-by: Jerin Jacob<jerin.jacob at caviumnetworks.com>
>>>>>>>>> ---
>>>>>>>>>      lib/librte_eal/linuxapp/eal/eal_vfio.c | 10 ++++++++--
>>>>>>>>>      1 file changed, 8 insertions(+), 2 deletions(-)
>>>>>>>>>
>>>>>>>>> diff --git a/lib/librte_eal/linuxapp/eal/eal_vfio.c b/lib/librte_eal/linuxapp/eal/eal_vfio.c
>>>>>>>>> index 04914406f..348b7a7f4 100644
>>>>>>>>> --- a/lib/librte_eal/linuxapp/eal/eal_vfio.c
>>>>>>>>> +++ b/lib/librte_eal/linuxapp/eal/eal_vfio.c
>>>>>>>>> @@ -706,7 +706,10 @@ vfio_type1_dma_map(int vfio_container_fd)
>>>>>>>>>              dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map);
>>>>>>>>>              dma_map.vaddr = ms[i].addr_64;
>>>>>>>>>              dma_map.size = ms[i].len;
>>>>>>>>> -        dma_map.iova = ms[i].phys_addr;
>>>>>>>>> +        if (rte_eal_iova_mode() == RTE_IOVA_VA)
>>>>>>>>> +            dma_map.iova = dma_map.vaddr;
>>>>>>>>> +        else
>>>>>>>>> +            dma_map.iova = ms[i].phys_addr;
>>>>>>>>>              dma_map.flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE;
>>>>>>>>
>>>>>>>> IIUC, it is changing default behavior for VFIO devices.
>>>>>>>>
>>>>>>>> I see a possible problem, but I'm not sure the case is valid.
>>>>>>>>
>>>>>>>> Imagine you have two devices in the iommu group, and the two devices are
>>>>>>>> used in separate processes. Each process could try two different
>>>>>>>> physical addresses at the same virtual address, and so the second map
>>>>>>>> would fail.
>>>>>>>
>>>>>>> IMO, Doesn't look like a problem. Here is the data flow
>>>>>>>
>>>>>>> 1) The vfio DMA map function(vfio_type1_dma_map()) will be called only
>>>>>>> on primary process
>>>>>>> http://dpdk.org/browse/dpdk/tree/lib/librte_eal/linuxapp/eal/eal_vfio.c#n359
>>>>>>>
>>>>>>> 2) On secondary process, DPDK rte_eal_huge_page_attach() will make sure
>>>>>>> that, the Secondary process has the _same_ virtual address as primary or
>>>>>>> exit from on attach.
>>>>>>> http://dpdk.org/browse/dpdk/tree/lib/librte_eal/linuxapp/eal/eal_memory.c#n1452
>>>>>>>
>>>>>>> 3) Since secondary process adds the mapped the virtual address in step (2).
>>>>>>> in the page table in OS. On SMMU entry miss(When device
>>>>>>> request from I/O transaction), OS will load the mapping and update the SMMU
>>>>>>> "context" with page tables from MMU.
>>>>>>
>>>>>> Ok thanks for the detailed info, but what about the case where the same
>>>>>> iommu group is used by two primary processes?
>>>>>
>>>>> Does that case exist with DPDK? We always need to blacklist same BDF in
>>>>> the secondary process to make things work with existing DPDK setup. Which
>>>>> make sense as well. Only primary process configures the HW blocks.
>>>>
>>>> I meant the case when two BDF are in the same IOMMU group (if ACS is not
>>>> supported at some point in the hierarchy). And I meant two primary
>>>> processes running, like for example two containers running each a DPDK
>>>> application.
>>>>
>>>> Maybe this is not a valid use-case (it is not secure, as it would break
>>>> isolation between the two containers), but it seems that it is something
>>>> DPDK allows today, if I'm not mistaken.
>>>>
>>> I'm not sure how two primary process could run, as because latter primary process
>>> would try accessing /var/run/.rte_config and would fail at this [1] point.
>>>
>>> It's not valid use-case for dpdk (imo).
>>> [1] http://dpdk.org/browse/dpdk/tree/lib/librte_eal/linuxapp/eal/eal.c#n204
>>
>> Yes this is possible. I had never used it before, but Thomas told me it
>> is supported by setting--file-prefix option. I had a trial, and I
>> confirm it works:
>> session 1> ./install/bin/testpmd -l 0,2 --socket-mem=1024 -w 0000:05:00.0 --proc-type=primary --file-prefix=app1 -- --disable-hw-vlan -i --rxq=1 --txq=1         --nb-cores=1 --forward-mode=io
>> session 2> ./install/bin/testpmd -l 0,3 --socket-mem=1024 -w 0000:05:00.1 --proc-type=primary --file-prefix=app2 -- --disable-hw-vlan -i --rxq=1 --txq=1         --nb-cores=1 --forward-mode=io
>>
>> In the above example, two ports of the same card is used by two
>> processes. Note that in this case, ACS is supproted and both ports have
>> their own iommu group.
>
> # ls -al /var/run/.app*
> -rw-r-----. 1 root root 208420 Jul  6 09:08 /var/run/.app1_config
> -rw-r--r--. 1 root root  49728 Jul  6 09:08 /var/run/.app1_hugepage_info
> srwxr-xr-x. 1 root root      0 Jul  6 09:08 /var/run/.app1_mp_socket
> -rw-r-----. 1 root root 208420 Jul  6 09:08 /var/run/.app2_config
> -rw-r--r--. 1 root root  45584 Jul  6 09:08 /var/run/.app2_hugepage_info
> srwxr-xr-x. 1 root root      0 Jul  6 09:08 /var/run/.app2_mp_socket
>
Yes, You're right, you can start two primary process, I missed that point. 
Use-case which you mentioned is ok, because they are under two different iommu
group so proposed scheme will work. It may not work for the case when ACS not present,
so its bypass mode which falls under vfio-noiommu category.

Having said that: Per discussion on [1]. The proposed scheme where 
bus makes decision based on pci_id and/or pci_drv will be a full proof
solution, and that way other types of devices will not be impacted. Right?

[1] https://www.mail-archive.com/dev@dpdk.org/msg70283.html




More information about the dev mailing list