On Wed, Jun 05, 2019 at 05:02:09PM -0700, Sagi Grimberg wrote: > if the lld does not explicitly sets this, scsi takes BLK_MAX_SEGMENT_SIZE > and sets it using dma_set_max_seg_size(). In our case, this will affect > all the rdma device consumers. > > Fix it by setting shost max_segment_size according to the device > capability. > > Reported-by: Jason Gunthorpe <jgg@xxxxxxxx> > Signed-off-by: Sagi Grimberg <sagi@xxxxxxxxxxx> > --- > This goes on top of hch patchset: > "properly communicate queue limits to the DMA layer" > > Normally this should go through the rdma tree, so we can > either get it through jens with hch patchset. Alternatively > this is a fix that should go to rc anyways? I don't think this make sense. > > drivers/infiniband/ulp/iser/iscsi_iser.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.c b/drivers/infiniband/ulp/iser/iscsi_iser.c > index 56848232eb81..2984a366dd7d 100644 > --- a/drivers/infiniband/ulp/iser/iscsi_iser.c > +++ b/drivers/infiniband/ulp/iser/iscsi_iser.c > @@ -653,6 +653,7 @@ iscsi_iser_session_create(struct iscsi_endpoint *ep, > SHOST_DIX_GUARD_CRC); > } > > + shost->max_segment_size = ib_dma_max_seg_size(ib_dev); > if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG)) > shost->virt_boundary_mask = ~MASK_4K; We only really need this settings in the IB_DEVICE_SG_GAPS_REG case, as the segement size is unlimited on the PRP-like scheme used by the other MR types anyway, and set as such by the block layer. I.e.g this should become: if (ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG) shost->max_segment_size = ib_dma_max_seg_size(ib_dev); else shost->virt_boundary_mask = ~MASK_4K;
