Re: [dpdk-dev] [PATCH 3/4] dma/cnxk: add dma channel operations

[Radha Mohan <mohun106@gmail.com>]

On Tue, Oct 26, 2021 at 1:49 AM Jerin Jacob <jerinjacobk@gmail.com> wrote:
>
> On Tue, Oct 26, 2021 at 9:43 AM Radha Mohan Chintakuntla
> <radhac@marvell.com> wrote:
> >
> > Add functions for the dmadev vchan setup and DMA operations.
> >
> > Signed-off-by: Radha Mohan Chintakuntla <radhac@marvell.com>
> > ---
> >  drivers/dma/cnxk/cnxk_dmadev.c | 322 +++++++++++++++++++++++++++++++++
> >  drivers/dma/cnxk/cnxk_dmadev.h |  53 ++++++
> >  drivers/dma/cnxk/version.map   |   3 +
> >  3 files changed, 378 insertions(+)
> >  create mode 100644 drivers/dma/cnxk/version.map
> >
> > diff --git a/drivers/dma/cnxk/cnxk_dmadev.c b/drivers/dma/cnxk/cnxk_dmadev.c
> > index 620766743d..8434579aa2 100644
> > --- a/drivers/dma/cnxk/cnxk_dmadev.c
> > +++ b/drivers/dma/cnxk/cnxk_dmadev.c
> > @@ -18,6 +18,322 @@
> >  #include <roc_api.h>
> >  #include <cnxk_dmadev.h>
> >
> > +static int
> > +cnxk_dmadev_info_get(const struct rte_dma_dev *dev,
> > +                    struct rte_dma_info *dev_info, uint32_t size)
> > +{
> > +       RTE_SET_USED(dev);
> > +       RTE_SET_USED(size);
> > +
> > +       dev_info->max_vchans = 1;
> > +       dev_info->nb_vchans = 1;
> > +       dev_info->dev_capa = RTE_DMA_CAPA_MEM_TO_MEM |
> > +               RTE_DMA_CAPA_MEM_TO_DEV | RTE_DMA_CAPA_DEV_TO_MEM |
> > +               RTE_DMA_CAPA_OPS_COPY;
> > +       dev_info->max_desc = DPI_MAX_DESC;
> > +       dev_info->min_desc = 1;
> > +       dev_info->max_sges = DPI_MAX_POINTER;
> > +
> > +       return 0;
> > +}
> > +
> > +static int
> > +cnxk_dmadev_configure(struct rte_dma_dev *dev,
> > +                     const struct rte_dma_conf *conf, uint32_t conf_sz)
> > +{
> > +       struct cnxk_dpi_vf_s *dpivf = NULL;
> > +       int rc = 0;
> > +
> > +       RTE_SET_USED(conf);
> > +       RTE_SET_USED(conf);
> > +       RTE_SET_USED(conf_sz);
> > +       RTE_SET_USED(conf_sz);
> > +       dpivf = dev->fp_obj->dev_private;
> > +       rc = roc_dpi_queue_configure(&dpivf->rdpi);
> > +       if (rc < 0)
> > +               plt_err("DMA queue configure failed err = %d", rc);
> > +
> > +       return rc;
> > +}
> > +
> > +static int
> > +cnxk_dmadev_vchan_setup(struct rte_dma_dev *dev, uint16_t vchan,
> > +                       const struct rte_dma_vchan_conf *conf,
> > +                       uint32_t conf_sz)
> > +{
> > +       struct cnxk_dpi_vf_s *dpivf = dev->fp_obj->dev_private;
> > +       struct cnxk_dpi_compl_s *comp_data;
> > +       int i;
> > +
> > +       RTE_SET_USED(vchan);
> > +       RTE_SET_USED(conf_sz);
> > +
> > +       switch (conf->direction) {
> > +       case RTE_DMA_DIR_DEV_TO_MEM:
> > +               dpivf->conf.direction = DPI_XTYPE_INBOUND;
> > +               dpivf->conf.src_port = conf->src_port.pcie.coreid;
> > +               dpivf->conf.dst_port = 0;
> > +               break;
> > +       case RTE_DMA_DIR_MEM_TO_DEV:
> > +               dpivf->conf.direction = DPI_XTYPE_OUTBOUND;
> > +               dpivf->conf.src_port = 0;
> > +               dpivf->conf.dst_port = conf->dst_port.pcie.coreid;
> > +               break;
> > +       case RTE_DMA_DIR_MEM_TO_MEM:
> > +               dpivf->conf.direction = DPI_XTYPE_INTERNAL_ONLY;
> > +               dpivf->conf.src_port = 0;
> > +               dpivf->conf.dst_port = 0;
> > +               break;
> > +       case RTE_DMA_DIR_DEV_TO_DEV:
> > +               dpivf->conf.direction = DPI_XTYPE_EXTERNAL_ONLY;
> > +               dpivf->conf.src_port = conf->src_port.pcie.coreid;
> > +               dpivf->conf.dst_port = conf->src_port.pcie.coreid;
> > +       };
> > +
> > +       for (i = 0; i < conf->nb_desc; i++) {
> > +               comp_data = rte_zmalloc(NULL, sizeof(*comp_data), 0);
> > +               dpivf->conf.c_desc.compl_ptr[i] = comp_data;
> > +       };
> > +       dpivf->conf.c_desc.max_cnt = DPI_MAX_DESC;
> > +       dpivf->conf.c_desc.head = 0;
> > +       dpivf->conf.c_desc.tail = 0;
> > +
> > +       return 0;
> > +}
> > +
> > +static int
> > +cnxk_dmadev_start(struct rte_dma_dev *dev)
> > +{
> > +       struct cnxk_dpi_vf_s *dpivf = dev->fp_obj->dev_private;
> > +
> > +       roc_dpi_queue_start(&dpivf->rdpi);
> > +
> > +       return 0;
> > +}
> > +
> > +static int
> > +cnxk_dmadev_stop(struct rte_dma_dev *dev)
> > +{
> > +       struct cnxk_dpi_vf_s *dpivf = dev->fp_obj->dev_private;
> > +
> > +       roc_dpi_queue_stop(&dpivf->rdpi);
> > +
> > +       return 0;
> > +}
> > +
> > +static int
> > +cnxk_dmadev_close(struct rte_dma_dev *dev)
> > +{
> > +       struct cnxk_dpi_vf_s *dpivf = dev->fp_obj->dev_private;
> > +
> > +       roc_dpi_queue_stop(&dpivf->rdpi);
> > +       roc_dpi_dev_fini(&dpivf->rdpi);
> > +
> > +       return 0;
> > +}
> > +
> > +static inline int
> > +__dpi_queue_write(struct roc_dpi *dpi, uint64_t *cmds, int cmd_count)
> > +{
> > +       uint64_t *ptr = dpi->chunk_base;
> > +
> > +       if ((cmd_count < DPI_MIN_CMD_SIZE) || (cmd_count > DPI_MAX_CMD_SIZE) ||
> > +           cmds == NULL)
> > +               return -EINVAL;
> > +
> > +       /*
> > +        * Normally there is plenty of room in the current buffer for the
> > +        * command
> > +        */
> > +       if (dpi->chunk_head + cmd_count < dpi->pool_size_m1) {
> > +               ptr += dpi->chunk_head;
> > +               dpi->chunk_head += cmd_count;
> > +               while (cmd_count--)
> > +                       *ptr++ = *cmds++;
> > +       } else {
> > +               int count;
> > +               uint64_t *new_buff = dpi->chunk_next;
> > +
> > +               dpi->chunk_next =
> > +                       (void *)roc_npa_aura_op_alloc(dpi->aura_handle, 0);
> > +               if (!dpi->chunk_next) {
> > +                       plt_err("Failed to alloc next buffer from NPA");
> > +                       return -ENOMEM;
> > +               }
> > +
> > +               /*
> > +                * Figure out how many cmd words will fit in this buffer.
> > +                * One location will be needed for the next buffer pointer.
> > +                */
> > +               count = dpi->pool_size_m1 - dpi->chunk_head;
> > +               ptr += dpi->chunk_head;
> > +               cmd_count -= count;
> > +               while (count--)
> > +                       *ptr++ = *cmds++;
> > +
> > +               /*
> > +                * chunk next ptr is 2 DWORDS
> > +                * second DWORD is reserved.
> > +                */
> > +               *ptr++ = (uint64_t)new_buff;
> > +               *ptr = 0;
> > +
> > +               /*
> > +                * The current buffer is full and has a link to the next
> > +                * buffers. Time to write the rest of the commands into the new
> > +                * buffer.
> > +                */
> > +               dpi->chunk_base = new_buff;
> > +               dpi->chunk_head = cmd_count;
> > +               ptr = new_buff;
> > +               while (cmd_count--)
> > +                       *ptr++ = *cmds++;
> > +
> > +               /* queue index may be greater than pool size */
> > +               if (dpi->chunk_head >= dpi->pool_size_m1) {
> > +                       new_buff = dpi->chunk_next;
> > +                       dpi->chunk_next =
> > +                               (void *)roc_npa_aura_op_alloc(dpi->aura_handle,
> > +                                                             0);
> > +                       if (!dpi->chunk_next) {
> > +                               plt_err("Failed to alloc next buffer from NPA");
> > +                               return -ENOMEM;
> > +                       }
> > +                       /* Write next buffer address */
> > +                       *ptr = (uint64_t)new_buff;
> > +                       dpi->chunk_base = new_buff;
> > +                       dpi->chunk_head = 0;
> > +               }
> > +       }
> > +
> > +       return 0;
> > +}
> > +
> > +static int
> > +cnxk_dmadev_copy(void *dev_private, uint16_t vchan, rte_iova_t src,
> > +                rte_iova_t dst, uint32_t length, uint64_t flags)
> > +{
> > +       uint64_t cmd[DPI_MAX_CMD_SIZE] = {0};
> > +       union dpi_instr_hdr_s *header = (union dpi_instr_hdr_s *)&cmd[0];
> > +       rte_iova_t fptr, lptr;
> > +       struct cnxk_dpi_vf_s *dpivf = dev_private;
> > +       struct cnxk_dpi_compl_s *comp_ptr;
> > +       int num_words = 0;
> > +       int rc;
> > +
> > +       RTE_SET_USED(vchan);
> > +
> > +       header->s.xtype = dpivf->conf.direction;
> > +       header->s.pt = DPI_HDR_PT_ZBW_CA;
> > +       comp_ptr = dpivf->conf.c_desc.compl_ptr[dpivf->conf.c_desc.tail];
> > +       comp_ptr->cdata = DPI_REQ_CDATA;
> > +       header->s.ptr = (uint64_t)comp_ptr;
> > +       STRM_INC(dpivf->conf.c_desc);
> > +
> > +       /* pvfe should be set for inbound and outbound only */
> > +       if (header->s.xtype <= 1)
> > +               header->s.pvfe = 1;
> > +       num_words += 4;
> > +
> > +       header->s.nfst = 1;
> > +       header->s.nlst = 1;
>
> Including filling zeros in cmd and the rest of the filling can be
> moved to slow path..
>
> Please change the logic to populate the static items based on
> configure/channel setup
> in slowpath and update only per transfer-specific items to have better
> performance.
>
These are instruction header values that we are filling. If you look
at it there is really one 64bit field that can be filled beforehand
a.k.a slowpath in vchan_setup().
Rest of the header can only be filled here like nlst, nfst (these are
number of pointers to be DMA'ed) and completion pointer. So just for
that I do not see a value in moving around the code.

<snip>