Re: [dpdk-dev] [PATCH v2] vfio: allow secondary process to query IOMMU type

["Burakov, Anatoly" <anatoly.burakov@intel.com>]

On 21-Jan-19 10:21 AM, Thomas Monjalon wrote:
> 21/01/2019 11:13, Burakov, Anatoly:
>> On 19-Jan-19 3:23 AM, Wang, Xiao W wrote:
>>> Hi Anatoly,
>>>
>>> From: Burakov, Anatoly
>>>>
>>>> It is only possible to know IOMMU type of a given VFIO container
>>>> by attempting to initialize it. Since secondary process never
>>>> attempts to set up VFIO container itself (because they're shared
>>>> between primary and secondary), it never knows which IOMMU type
>>>> the container is using, and never sets up the appropriate config
>>>> structures. This results in inability to perform DMA mappings in
>>>> secondary process.
>>>>
>>>> Fix this by allowing secondary process to query IOMMU type of
>>>> primary's default container at device initialization.
>>>>
>>>> Note that this fix is assuming we're only interested in default
>>>> container.
>>>>
>>>> Bugzilla ID: 174
>>>>
>>>> Fixes: 6bcb7c95fe14 ("vfio: share default container in multi-process")
>>>> Cc: dariusz.stojaczyk@intel.com
>>>> Cc: stable@dpdk.org
>>>>
>>>> Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
>>>> ---
>>>>
>>>> Notes:
>>>>       v2:
>>>>       - Check if we found our IOMMU type within list of IOMMU types
>>>>       - Don't request new default container fd as this should have
>>>>         been done during rte_vfio_enable()
>>>>
>>>>    lib/librte_eal/linuxapp/eal/eal_vfio.c        | 88 +++++++++++++++++++
>>>>    lib/librte_eal/linuxapp/eal/eal_vfio.h        | 12 ++-
>>>>    .../linuxapp/eal/eal_vfio_mp_sync.c           | 16 ++++
>>>>    3 files changed, 115 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/lib/librte_eal/linuxapp/eal/eal_vfio.c
>>>> b/lib/librte_eal/linuxapp/eal/eal_vfio.c
>>>> index 72cc65151..c821e8382 100644
>>>> --- a/lib/librte_eal/linuxapp/eal/eal_vfio.c
>>>> +++ b/lib/librte_eal/linuxapp/eal/eal_vfio.c
>>>> @@ -549,6 +549,65 @@ vfio_mem_event_callback(enum rte_mem_event
>>>> type, const void *addr, size_t len,
>>>>    	}
>>>>    }
>>>>
>>>> +static int
>>>> +vfio_sync_default_container(void)
>>>> +{
>>>> +	struct rte_mp_msg mp_req, *mp_rep;
>>>> +	struct rte_mp_reply mp_reply;
>>>> +	struct timespec ts = {.tv_sec = 5, .tv_nsec = 0};
>>>> +	struct vfio_mp_param *p = (struct vfio_mp_param *)mp_req.param;
>>>> +	int iommu_type_id;
>>>> +	unsigned int i;
>>>> +
>>>> +	/* cannot be called from primary */
>>>> +	if (rte_eal_process_type() != RTE_PROC_SECONDARY)
>>>> +		return -1;
>>>> +
>>>> +	/* default container fd should have been opened in rte_vfio_enable()
>>>> */
>>>> +	if (!default_vfio_cfg->vfio_enabled ||
>>>> +			default_vfio_cfg->vfio_container_fd < 0) {
>>>> +		RTE_LOG(ERR, EAL, "VFIO support is not initialized\n");
>>>> +		return -1;
>>>> +	}
>>>> +
>>>> +	/* find default container's IOMMU type */
>>>> +	p->req = SOCKET_REQ_IOMMU_TYPE;
>>>
>>> Since this function is to sync IOMMU type for the default container, should we make the req type as
>>> SOCKET_REQ_DEFAULT_IOMMU_TYPE?
>>
>> Hi,
>>
>> Sure, that can be done. However, i don't think it warrants a respin
>> unless there's more important stuff to fix also. This patch is a
>> stop-gap, and this stuff will be rewritten for 19.05, so getting this
>> right is not that important :).
> 
> Anatoly,
> I don't understand what is the use case.
> It does not look critical enough to be merged late in 19.02.
> 

This isn't "a use case", it's a bug fix. This was supposed to work, but 
doesn't.

-- 
Thanks,
Anatoly