> -----Original Message----- > From: Vijay Kumar Srivastava <[email protected]> > Sent: Tuesday, November 2, 2021 12:38 PM > To: Xia, Chenbo <[email protected]>; [email protected] > Cc: [email protected]; [email protected]; Praveen Kumar > Jain <[email protected]> > Subject: RE: [PATCH v3 02/10] vdpa/sfc: add support for device initialization > > Hi Chenbo, > > >-----Original Message----- > >From: Xia, Chenbo <[email protected]> > >Sent: Monday, November 1, 2021 5:19 PM > >To: Vijay Kumar Srivastava <[email protected]>; [email protected] > >Cc: [email protected]; [email protected]; Vijay > >Kumar Srivastava <[email protected]> > >Subject: RE: [PATCH v3 02/10] vdpa/sfc: add support for device initialization > > > >Hi Vijay, > > > >> -----Original Message----- > >> From: Vijay Srivastava <[email protected]> > >> Sent: Friday, October 29, 2021 10:47 PM > >> To: [email protected] > >> Cc: [email protected]; Xia, Chenbo <[email protected]>; > >> [email protected]; Vijay Kumar Srivastava > >> <[email protected]> > >> Subject: [PATCH v3 02/10] vdpa/sfc: add support for device > >> initialization > >> > >> From: Vijay Kumar Srivastava <[email protected]> > >> > >> Add HW initialization and vDPA device registration support. > >> > >> Signed-off-by: Vijay Kumar Srivastava <[email protected]> > >> Acked-by: Andrew Rybchenko <[email protected]> > >> --- > [SNIP] > >> + > >> + do { > >> + ret = rte_vfio_container_dma_map(sva->vfio_container_fd, > >> + (uint64_t)mz->addr, > >mcdi_iova, > >> + mcdi_buff_size); > >> + if (ret == 0) > >> + break; > >> + > >> + mcdi_iova = mcdi_iova >> 1; > >> + if (mcdi_iova < mcdi_buff_size) { > >> + sfc_vdpa_err(sva, > >> + "DMA mapping failed for MCDI : %s", > >> + rte_strerror(rte_errno)); > >> + rte_memzone_free(mz); > >> + return ret; > >> + } > >> + > >> + } while (ret < 0); > > > >So when QEMU iova and mcdi_iova conflicts, you just let vdpa dev failed to > >configure, right? > > > >Why not use re-mapping mcdi dma region as the solution? Any side-effect? > >Or you just assume conflict can hardly happen? > > MCDI configuration is being done at the very early point of initialization. > Conflict would be detected later when rte_vhost_get_mem_table() would be > invoked in .dev_conf callback and > then MCDI re-mapping can be done in case of conflict,
Agree. It should be done in dev_conf callback. > for this a patch is in > progress which would be submitted separately. OK for me, as the initial version, you can just let dev_conf fail if conflict happens. /Chenbo > > Regards, > Vijay > > [SNIP]
