vfio: change default per vfio_cfg map all dpdk segments for DMA

  From: Yunjian Wang <wangyunjian@huawei.com>

As all vfio_cfgs need to map all dpdk segments for DMA, but now only
default_vfio_cfg map it. This patch changes rte_vfio_setup_device()
as a default map, so that IOMMU progrming does notneed to map dpdk
segments separately.

Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
---
 lib/librte_eal/linux/eal_vfio.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
  

On 15-May-20 12:01 PM, wangyunjian wrote:
> From: Yunjian Wang <wangyunjian@huawei.com>
> 
> As all vfio_cfgs need to map all dpdk segments for DMA, but now only
> default_vfio_cfg map it. This patch changes rte_vfio_setup_device()
> as a default map, so that IOMMU progrming does notneed to map dpdk
> segments separately.
> 
> Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
> ---

I'm pretty sure this was intentional. The other containers were supposed 
to call DMA maps manually. That said, i would probably ask the original 
submitter for the multi-container feature as i have only a vague idea of 
usages for this API.

+Zhihong, as that's the closest i could find to a maintainer for this 
functionality :)
  

Hi, Yunjian

> -----Original Message-----
> From: dev <dev-bounces@dpdk.org> On Behalf Of wangyunjian
> Sent: Friday, May 15, 2020 7:02 PM
> To: dev@dpdk.org
> Cc: Burakov, Anatoly <anatoly.burakov@intel.com>; jerry.lilijun@huawei.com;
> xudingke@huawei.com; Yunjian Wang <wangyunjian@huawei.com>
> Subject: [dpdk-dev] [PATCH] vfio: change default per vfio_cfg map all dpdk
> segments for DMA
> 
> From: Yunjian Wang <wangyunjian@huawei.com>
> 
> As all vfio_cfgs need to map all dpdk segments for DMA, but now only
> default_vfio_cfg map it. This patch changes rte_vfio_setup_device()
> as a default map, so that IOMMU progrming does notneed to map dpdk
> segments separately.
> 
> Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
> ---
>  lib/librte_eal/linux/eal_vfio.c | 5 +----
>  1 file changed, 1 insertion(+), 4 deletions(-)
> 
> diff --git a/lib/librte_eal/linux/eal_vfio.c b/lib/librte_eal/linux/eal_vfio.c
> index d26e1649a..46160c328 100644
> --- a/lib/librte_eal/linux/eal_vfio.c
> +++ b/lib/librte_eal/linux/eal_vfio.c
> @@ -807,10 +807,7 @@ rte_vfio_setup_device(const char *sysfs_base, const char
> *dev_addr,
>  			 * after registering callback, to prevent races
>  			 */
>  			rte_mcfg_mem_read_lock();
> -			if (vfio_cfg == default_vfio_cfg)
> -				ret = t->dma_map_func(vfio_container_fd);
> -			else
> -				ret = 0;
> +			ret = t->dma_map_func(vfio_container_fd);

Anatoly is correct, this is intentional for ifc vdpa. For vdpa/ifc, we need another vfio container
for vdpa device and the dma map info comes from QEMU/virtio-user, so this container will not use dpdk
memsegs. Your patch will break current ifc vdpa driver.

Thanks,
Chenbo

>  			if (ret) {
>  				RTE_LOG(ERR, EAL,
>  					"  %s DMA remapping failed, error %i (%s)\n",
> --
> 2.23.0
>