On Thu, Jan 7, 2016 at 11:50 PM, Stephen Hemminger <stephen at networkplumber.org> wrote: > On Thu, 7 Jan 2016 22:03:06 +0530 > Santosh Shukla <sshukla at mvista.com> wrote: > >> +#ifdef RTE_EAL_VFIO >> + .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_DETACHABLE, >> +#else >> .drv_flags = RTE_PCI_DRV_DETACHABLE, >> +#endif > > Since VFIO is determined at runtime not compile time, the flags should > be updated at runtime not compile time. > > In general, Yes, Its a wrong approach i..e. Wrapping __need_mapping flag only for vfio case. I am thinking to add vfio parser routine something similar to virtio_xxx_xx_uio_xx() / virtio_xx_xx_ioport() currently exist. This will remove RTE_EAL_VFIO ifdef clutter for this patch and [08/12] patch and also virtio pmd driver can then initialize device for vfio mode..
_but_ I still need _MAPPING flag enabled for in virtio driver as because for vfio case - I want vfio_xx_mmap() routine to create vfio container/group_id and then create vfio_dev_fd for each virtio-net-pci interface. Let me know my approach aligned to your suggestion.
