Re: [dpdk-dev] [PATCH v6 2/2] vhost: enable IOMMU for async vhost

["Hu, Jiayu" <jiayu.hu@intel.com>]

> -----Original Message-----
> From: Burakov, Anatoly <anatoly.burakov@intel.com>
> Sent: Wednesday, September 29, 2021 5:39 PM
> To: Hu, Jiayu <jiayu.hu@intel.com>; Ding, Xuan <xuan.ding@intel.com>;
> dev@dpdk.org; maxime.coquelin@redhat.com; Xia, Chenbo
> <chenbo.xia@intel.com>
> Cc: Jiang, Cheng1 <cheng1.jiang@intel.com>; Richardson, Bruce
> <bruce.richardson@intel.com>; Pai G, Sunil <sunil.pai.g@intel.com>; Wang,
> Yinan <yinan.wang@intel.com>; Yang, YvonneX <yvonnex.yang@intel.com>
> Subject: Re: [PATCH v6 2/2] vhost: enable IOMMU for async vhost
> 
> On 29-Sep-21 7:12 AM, Hu, Jiayu wrote:
> > Hi Xuan,
> >
> >> -----Original Message-----
> >> From: Ding, Xuan <xuan.ding@intel.com>
> >> Sent: Wednesday, September 29, 2021 10:41 AM
> >> To: dev@dpdk.org; Burakov, Anatoly <anatoly.burakov@intel.com>;
> >> maxime.coquelin@redhat.com; Xia, Chenbo <chenbo.xia@intel.com>
> >> Cc: Hu, Jiayu <jiayu.hu@intel.com>; Jiang, Cheng1
> >> <cheng1.jiang@intel.com>; Richardson, Bruce
> >> <bruce.richardson@intel.com>; Pai G, Sunil <sunil.pai.g@intel.com>;
> >> Wang, Yinan <yinan.wang@intel.com>; Yang, YvonneX
> >> <yvonnex.yang@intel.com>; Ding, Xuan <xuan.ding@intel.com>
> >> Subject: [PATCH v6 2/2] vhost: enable IOMMU for async vhost
> >>
> >> The use of IOMMU has many advantages, such as isolation and address
> >> translation. This patch extends the capbility of DMA engine to use
> >> IOMMU if the DMA engine is bound to vfio.
> >>
> >> When set memory table, the guest memory will be mapped into the
> >> default container of DPDK.
> >>
> >> Signed-off-by: Xuan Ding <xuan.ding@intel.com>
> >> ---
> >> +async_dma_map(struct rte_vhost_mem_region *region, bool
> >> +*dma_map_success, bool do_map) { uint64_t host_iova; int ret = 0;
> >> +
> >> +host_iova = rte_mem_virt2iova((void *)(uintptr_t)region-
> >>> host_user_addr);
> >> +if (do_map) {
> >> +/* Add mapped region into the default container of DPDK. */ ret =
> >> rte_vfio_container_dma_map(RTE_VFIO_DEFAULT_CONTAINER_FD,
> >> + region->host_user_addr,
> >> + host_iova,
> >> + region->size);
> >> +*dma_map_success = ret == 0;
> >> +
> >> +if (ret) {
> >> +/*
> >> + * DMA device may bind with kernel driver, in this
> >> case,
> >> + * we don't need to program IOMMU manually.
> >> However, if no
> >> + * device is bound with vfio/uio in DPDK, and vfio
> >> kernel
> >> + * module is loaded, the API will still be called and
> >> return
> >> + * with ENODEV/ENOSUP.
> >> + *
> >> + * DPDK vfio only returns ENODEV/ENOSUP in very
> >> similar
> >> + * situations(vfio either unsupported, or supported
> >> + * but no devices found). Either way, no mappings
> >> could be
> >> + * performed. We treat it as normal case in async
> >> path.
> >> + */
> >
> >
> > What do you mean by saying "vfio either unsupported"? Does it mean
> > platform doesn't support iommu?
> 
> Unsupported as in the VFIO driver is not loaded. We don't really care if the
> *system* supports VFIO as much as whether it's *accessible to us*.
> I'm sure you would agree that scenario "VFIO is not supported" is exactly
> equivalent to "VFIO driver is not loaded", because from our perspective, in
> both cases the VFIO driver is not loaded :)

OK, that's clear.

Thanks,
Jiayu