On Thu, Jan 17, 2019 at 01:54:55PM -0500, Devesh Sharma wrote:
> The new chip series has 64 bit doorbell for notification
> queues. Thus, both control and data path event queues
> need new routines to write 64 bit doorbell. Adding the
> same. There is new doorbell interface between the chip
> and driver. Changing the chip specific data structure
> definitions.
>
> Additional significant changes are listed below
> - bnxt_re_net_ring_free/alloc takes a new argument
> - bnxt_qplib_enable_nq and enable_rcfw uses new doorbell offset
> for new chip.
> - DB mapping for NQ and CREQ now maps 8 bytes.
> - DBR_DBR_* macros renames to DBC_DBC_*
> - store nq_db_offset in a 32bit data type.
>
> Signed-off-by: Selvin Xavier <selvin.xavier@xxxxxxxxxxxx>
> Signed-off-by: Devesh Sharma <devesh.sharma@xxxxxxxxxxxx>
> drivers/infiniband/hw/bnxt_re/Kconfig | 1 +
> drivers/infiniband/hw/bnxt_re/ib_verbs.c | 4 +-
> drivers/infiniband/hw/bnxt_re/main.c | 85 +++++++++++++++----------
> drivers/infiniband/hw/bnxt_re/qplib_fp.c | 99 ++++++++++++++++--------------
> drivers/infiniband/hw/bnxt_re/qplib_fp.h | 46 ++++++++++++--
> drivers/infiniband/hw/bnxt_re/qplib_rcfw.c | 31 +++++++---
> drivers/infiniband/hw/bnxt_re/qplib_rcfw.h | 46 ++++++++++++--
> drivers/infiniband/hw/bnxt_re/qplib_res.h | 13 ++++
> drivers/infiniband/hw/bnxt_re/roce_hsi.h | 96 ++++++++++++++++-------------
> 9 files changed, 279 insertions(+), 142 deletions(-)
>
> diff --git a/drivers/infiniband/hw/bnxt_re/Kconfig b/drivers/infiniband/hw/bnxt_re/Kconfig
> index 18f5ed0..be2fdad 100644
> +++ b/drivers/infiniband/hw/bnxt_re/Kconfig
> @@ -1,5 +1,6 @@
> config INFINIBAND_BNXT_RE
> tristate "Broadcom Netxtreme HCA support"
> + depends on 64BIT
> depends on ETHERNET && NETDEVICES && PCI && INET && DCB
> depends on MAY_USE_DEVLINK
> select NET_VENDOR_BROADCOM
> diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
> index 9bc637e..3e3e481 100644
> +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c
> @@ -3299,10 +3299,10 @@ int bnxt_re_req_notify_cq(struct ib_cq *ib_cq,
> spin_lock_irqsave(&cq->cq_lock, flags);
> /* Trigger on the very next completion */
> if (ib_cqn_flags & IB_CQ_NEXT_COMP)
> - type = DBR_DBR_TYPE_CQ_ARMALL;
> + type = DBC_DBC_TYPE_CQ_ARMALL;
> /* Trigger on the next solicited completion */
> else if (ib_cqn_flags & IB_CQ_SOLICITED)
> - type = DBR_DBR_TYPE_CQ_ARMSE;
> + type = DBC_DBC_TYPE_CQ_ARMSE;
>
> /* Poll to see if there are missed events */
> if ((ib_cqn_flags & IB_CQ_REPORT_MISSED_EVENTS) &&
> diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
> index 16af3f1..a837229 100644
> +++ b/drivers/infiniband/hw/bnxt_re/main.c
> @@ -369,7 +369,8 @@ static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
> fw_msg->timeout = timeout;
> }
>
> -static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, u16 fw_ring_id)
> +static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
> + u16 fw_ring_id, int type)
> {
> struct bnxt_en_dev *en_dev = rdev->en_dev;
> struct hwrm_ring_free_input req = {0};
> @@ -383,7 +384,7 @@ static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, u16 fw_ring_id)
> memset(&fw_msg, 0, sizeof(fw_msg));
>
> bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
> - req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
> + req.ring_type = type;
> req.ring_id = cpu_to_le16(fw_ring_id);
> bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
> sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
> @@ -420,7 +421,7 @@ static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr,
> /* Association of ring index with doorbell index and MSIX number */
> req.logical_id = cpu_to_le16(map_index);
> req.length = cpu_to_le32(ring_mask + 1);
> - req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
> + req.ring_type = type;
> req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
> bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
> sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
> @@ -884,6 +885,12 @@ static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
> return 0;
> }
>
> +static u32 bnxt_re_get_nqdb_offset(struct bnxt_re_dev *rdev, u16 indx)
> +{
> + return bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx) ?
> + 0x10000 : rdev->msix_entries[indx].db_offset;
> +}
> +
> static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
> {
> int i;
> @@ -897,18 +904,18 @@ static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
>
> static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
> {
> - int rc = 0, i;
> int num_vec_enabled = 0;
> + int rc = 0, i;
> + u32 db_offt;
>
> bnxt_qplib_init_res(&rdev->qplib_res);
>
> for (i = 1; i < rdev->num_msix ; i++) {
> + db_offt = bnxt_re_get_nqdb_offset(rdev, i);
> rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
> i - 1, rdev->msix_entries[i].vector,
> - rdev->msix_entries[i].db_offset,
> - &bnxt_re_cqn_handler,
> + db_offt, &bnxt_re_cqn_handler,
> &bnxt_re_srqn_handler);
> -
> if (rc) {
> dev_err(rdev_to_dev(rdev),
> "Failed to enable NQ with rc = 0x%x", rc);
> @@ -920,17 +927,18 @@ static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
> fail:
> for (i = num_vec_enabled; i >= 0; i--)
> bnxt_qplib_disable_nq(&rdev->nq[i]);
> -
> return rc;
> }
>
> static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
> {
> + u8 type;
> int i;
>
> for (i = 0; i < rdev->num_msix - 1; i++) {
> + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
> + bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
> rdev->nq[i].res = NULL;
> - bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id);
> bnxt_qplib_free_nq(&rdev->nq[i]);
> }
> }
> @@ -952,8 +960,11 @@ static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
>
> static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
> {
> - int rc = 0, i;
> int num_vec_created = 0;
> + dma_addr_t *pg_map;
> + int rc = 0, i;
> + int pages;
> + u8 type;
>
> /* Configure and allocate resources for qplib */
> rdev->qplib_res.rcfw = &rdev->rcfw;
> @@ -983,13 +994,13 @@ static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
> i, rc);
> goto free_nq;
> }
> - rc = bnxt_re_net_ring_alloc
> - (rdev, rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr,
> - rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count,
> - HWRM_RING_ALLOC_CMPL,
> - BNXT_QPLIB_NQE_MAX_CNT - 1,
> - rdev->msix_entries[i + 1].ring_idx,
> - &rdev->nq[i].ring_id);
> + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
> + pg_map = rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr;
> + pages = rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count;
> + rc = bnxt_re_net_ring_alloc(rdev, pg_map, pages, type,
> + BNXT_QPLIB_NQE_MAX_CNT - 1,
> + rdev->msix_entries[i + 1].ring_idx,
> + &rdev->nq[i].ring_id);
> if (rc) {
> dev_err(rdev_to_dev(rdev),
> "Failed to allocate NQ fw id with rc = 0x%x",
> @@ -1002,7 +1013,8 @@ static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
> return 0;
> free_nq:
> for (i = num_vec_created; i >= 0; i--) {
> - bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id);
> + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
> + bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
> bnxt_qplib_free_nq(&rdev->nq[i]);
> }
> bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
> @@ -1256,6 +1268,7 @@ static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
>
> static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev)
> {
> + u8 type;
> int rc;
>
> if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
> @@ -1279,7 +1292,8 @@ static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev)
> bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
> bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
> bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
> - bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id);
> + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
> + bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, type);
> bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
> }
> if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
> @@ -1310,9 +1324,12 @@ static void bnxt_re_worker(struct work_struct *work)
>
> static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
> {
> - int rc;
> -
> + dma_addr_t *pg_map;
> + u32 db_offt, ridx;
> + int pages, vid;
> bool locked;
> + u8 type;
> + int rc;
>
> /* Acquire rtnl lock through out this function */
> rtnl_lock();
> @@ -1356,21 +1373,22 @@ static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
> pr_err("Failed to allocate RCFW Channel: %#x\n", rc);
> goto fail;
> }
> - rc = bnxt_re_net_ring_alloc
> - (rdev, rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr,
> - rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count,
> - HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_CREQE_MAX_CNT - 1,
> - rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx,
> - &rdev->rcfw.creq_ring_id);
> + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
> + pg_map = rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr;
> + pages = rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count;
> + ridx = rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx;
> + rc = bnxt_re_net_ring_alloc(rdev, pg_map, pages, type,
> + BNXT_QPLIB_CREQE_MAX_CNT - 1,
> + ridx, &rdev->rcfw.creq_ring_id);
> if (rc) {
> pr_err("Failed to allocate CREQ: %#x\n", rc);
> goto free_rcfw;
> }
> - rc = bnxt_qplib_enable_rcfw_channel
> - (rdev->en_dev->pdev, &rdev->rcfw,
> - rdev->msix_entries[BNXT_RE_AEQ_IDX].vector,
> - rdev->msix_entries[BNXT_RE_AEQ_IDX].db_offset,
> - rdev->is_virtfn, &bnxt_re_aeq_handler);
> + db_offt = bnxt_re_get_nqdb_offset(rdev, BNXT_RE_AEQ_IDX);
> + vid = rdev->msix_entries[BNXT_RE_AEQ_IDX].vector;
> + rc = bnxt_qplib_enable_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
> + vid, db_offt, rdev->is_virtfn,
> + &bnxt_re_aeq_handler);
> if (rc) {
> pr_err("Failed to enable RCFW channel: %#x\n", rc);
> goto free_ring;
> @@ -1454,7 +1472,8 @@ static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
> disable_rcfw:
> bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
> free_ring:
> - bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id);
> + type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
> + bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, type);
> free_rcfw:
> bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
> fail:
> diff --git a/drivers/infiniband/hw/bnxt_re/qplib_fp.c b/drivers/infiniband/hw/bnxt_re/qplib_fp.c
> index b98b054..49bac02 100644
> +++ b/drivers/infiniband/hw/bnxt_re/qplib_fp.c
> @@ -244,6 +244,7 @@ static void bnxt_qplib_service_nq(unsigned long data)
> u16 type;
> int budget = nq->budget;
> uintptr_t q_handle;
> + bool gen_p5 = bnxt_qplib_is_chip_gen_p5(nq->res->cctx);
>
> /* Service the NQ until empty */
> raw_cons = hwq->cons;
> @@ -290,7 +291,7 @@ static void bnxt_qplib_service_nq(unsigned long data)
> q_handle |= (u64)le32_to_cpu(nqsrqe->srq_handle_high)
> << 32;
> bnxt_qplib_arm_srq((struct bnxt_qplib_srq *)q_handle,
> - DBR_DBR_TYPE_SRQ_ARMENA);
> + DBC_DBC_TYPE_SRQ_ARMENA);
> if (!nq->srqn_handler(nq,
> (struct bnxt_qplib_srq *)q_handle,
> nqsrqe->event))
> @@ -312,7 +313,9 @@ static void bnxt_qplib_service_nq(unsigned long data)
> }
> if (hwq->cons != raw_cons) {
> hwq->cons = raw_cons;
> - NQ_DB_REARM(nq->bar_reg_iomem, hwq->cons, hwq->max_elements);
> + bnxt_qplib_ring_nq_db_rearm(nq->bar_reg_iomem, hwq->cons,
> + hwq->max_elements, nq->ring_id,
> + gen_p5);
> }
> }
>
> @@ -336,9 +339,11 @@ static irqreturn_t bnxt_qplib_nq_irq(int irq, void *dev_instance)
>
> void bnxt_qplib_nq_stop_irq(struct bnxt_qplib_nq *nq, bool kill)
> {
> + bool gen_p5 = bnxt_qplib_is_chip_gen_p5(nq->res->cctx);
> tasklet_disable(&nq->worker);
> /* Mask h/w interrupt */
> - NQ_DB(nq->bar_reg_iomem, nq->hwq.cons, nq->hwq.max_elements);
> + bnxt_qplib_ring_nq_db(nq->bar_reg_iomem, nq->hwq.cons,
> + nq->hwq.max_elements, nq->ring_id, gen_p5);
> /* Sync with last running IRQ handler */
> synchronize_irq(nq->vector);
> if (kill)
> @@ -373,6 +378,7 @@ void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq)
> int bnxt_qplib_nq_start_irq(struct bnxt_qplib_nq *nq, int nq_indx,
> int msix_vector, bool need_init)
> {
> + bool gen_p5 = bnxt_qplib_is_chip_gen_p5(nq->res->cctx);
> int rc;
>
> if (nq->requested)
> @@ -399,7 +405,8 @@ int bnxt_qplib_nq_start_irq(struct bnxt_qplib_nq *nq, int nq_indx,
> nq->vector, nq_indx);
> }
> nq->requested = true;
> - NQ_DB_REARM(nq->bar_reg_iomem, nq->hwq.cons, nq->hwq.max_elements);
> + bnxt_qplib_ring_nq_db_rearm(nq->bar_reg_iomem, nq->hwq.cons,
> + nq->hwq.max_elements, nq->ring_id, gen_p5);
>
> return rc;
> }
> @@ -433,7 +440,8 @@ int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq,
> rc = -ENOMEM;
> goto fail;
> }
> - nq->bar_reg_iomem = ioremap_nocache(nq_base + nq->bar_reg_off, 4);
> + /* Unconditionally map 8 bytes to support 57500 series */
> + nq->bar_reg_iomem = ioremap_nocache(nq_base + nq->bar_reg_off, 8);
> if (!nq->bar_reg_iomem) {
> rc = -ENOMEM;
> goto fail;
> @@ -462,15 +470,17 @@ void bnxt_qplib_free_nq(struct bnxt_qplib_nq *nq)
>
> int bnxt_qplib_alloc_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq)
> {
> + u8 hwq_type;
> +
> nq->pdev = pdev;
> if (!nq->hwq.max_elements ||
> nq->hwq.max_elements > BNXT_QPLIB_NQE_MAX_CNT)
> nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT;
> -
> + hwq_type = bnxt_qplib_get_hwq_type(nq->res);
> if (bnxt_qplib_alloc_init_hwq(nq->pdev, &nq->hwq, NULL, 0,
> &nq->hwq.max_elements,
> BNXT_QPLIB_MAX_NQE_ENTRY_SIZE, 0,
> - PAGE_SIZE, HWQ_TYPE_L2_CMPL))
> + PAGE_SIZE, hwq_type))
> return -ENOMEM;
>
> nq->budget = 8;
> @@ -481,19 +491,19 @@ int bnxt_qplib_alloc_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq)
> static void bnxt_qplib_arm_srq(struct bnxt_qplib_srq *srq, u32 arm_type)
> {
> struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
> - struct dbr_dbr db_msg = { 0 };
> + struct dbc_dbc db_msg = { 0 };
> void __iomem *db;
> u32 sw_prod = 0;
>
> /* Ring DB */
> - sw_prod = (arm_type == DBR_DBR_TYPE_SRQ_ARM) ? srq->threshold :
> - HWQ_CMP(srq_hwq->prod, srq_hwq);
> - db_msg.index = cpu_to_le32((sw_prod << DBR_DBR_INDEX_SFT) &
> - DBR_DBR_INDEX_MASK);
> - db_msg.type_xid = cpu_to_le32(((srq->id << DBR_DBR_XID_SFT) &
> - DBR_DBR_XID_MASK) | arm_type);
> - db = (arm_type == DBR_DBR_TYPE_SRQ_ARMENA) ?
> - srq->dbr_base : srq->dpi->dbr;
> + sw_prod = (arm_type == DBC_DBC_TYPE_SRQ_ARM) ?
> + srq->threshold : HWQ_CMP(srq_hwq->prod, srq_hwq);
> + db_msg.index = cpu_to_le32((sw_prod << DBC_DBC_INDEX_SFT) &
> + DBC_DBC_INDEX_MASK);
> + db_msg.type_path_xid = cpu_to_le32(((srq->id << DBC_DBC_XID_SFT) &
> + DBC_DBC_XID_MASK) | arm_type);
> + db = (arm_type == DBC_DBC_TYPE_SRQ_ARMENA) ? srq->dbr_base :
> + srq->dpi->dbr;
> wmb(); /* barrier before db ring */
> __iowrite64_copy(db, &db_msg, sizeof(db_msg) / sizeof(u64));
If the driver now requires writeq() why not use it for all these
__iowrite64_copy's too?
Jason
