On Tue, 2011-08-16 at 21:09 -0700, Roland Dreier wrote:
> On Tue, Aug 16, 2011 at 2:35 PM, Boaz Harrosh <bharrosh@xxxxxxxxxxx> wrote:
> > On 08/16/2011 02:09 PM, Nicholas A. Bellinger wrote:
> > <snip>
> >>> +static enum dma_data_direction tcm_qla2xxx_mapping_dir(struct se_cmd *se_cmd)
> >>> +{
> >>> + if (se_cmd->t_tasks_bidi)
> >>> + return DMA_BIDIRECTIONAL;
> >>> +
> >>
> >> Mmmm, I don't think this is correct. It's my understanding that
> >> DMA_BIDIRECTIONAL is different from SCSI bidi operation, and that the
> >> individual bidi payloads should be retaining their individual
> >> DMA_FROM_DEVICE and DMA_TO_DEVICE usage.. (Boaz CC'ed)
> >
> > Yes! DMA_BIDIRECTIONAL means that the same direct-access buffer could
> > be written to and read from at the same time. For example a shared
> > host-DSP buffer could be that case. It tells the (true) IOMMU to
> > let both type of access on this single buffer.
>
> OK, fair enough. However I don't see where the cmd->data_direction gets
> changed from TO_DEVICE to FROM_DEVICE for the two phases. How does
> that happen?
>
I believe that DMA_TO_DEVICE (from the initiator's perpsective) is set
for bidi operations in struct scsi_cmnd->sc_data_direction, and
include/scsi/scsi_cmnd.h:scsi_in() is used to get the extra bidi READ
buffer. I'm not sure how this is going to work with qla2xxx/tcm_qla2xxx
data_direction yet, but I think it would make sense to do something
similar for qla_tgt_cmd->data_direction.
> Do bidi commands work at all with the qla2xxx target?
>
loopback is the only fabric that has been tested (and works) with bidi
commands thus far, and only with XDWRITEREAD via BSG using tomo's sg
utils tree.
--nab
--
To unsubscribe from this list: send the line "unsubscribe target-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
