Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / infiniband / hw / cxgb4 / t4.h
diff --git a/kernel/drivers/infiniband/hw/cxgb4/t4.h b/kernel/drivers/infiniband/hw/cxgb4/t4.h
new file mode 100644 (file)
index 0000000..7f2a6c2
--- /dev/null
@@ -0,0 +1,685 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __T4_H__
+#define __T4_H__
+
+#include "t4_hw.h"
+#include "t4_regs.h"
+#include "t4_msg.h"
+#include "t4fw_ri_api.h"
+
+#define T4_MAX_NUM_PD 65536
+#define T4_MAX_MR_SIZE (~0ULL)
+#define T4_PAGESIZE_MASK 0xffff000  /* 4KB-128MB */
+#define T4_STAG_UNSET 0xffffffff
+#define T4_FW_MAJ 0
+#define PCIE_MA_SYNC_A 0x30b4
+
+struct t4_status_page {
+       __be32 rsvd1;   /* flit 0 - hw owns */
+       __be16 rsvd2;
+       __be16 qid;
+       __be16 cidx;
+       __be16 pidx;
+       u8 qp_err;      /* flit 1 - sw owns */
+       u8 db_off;
+       u8 pad;
+       u16 host_wq_pidx;
+       u16 host_cidx;
+       u16 host_pidx;
+};
+
+#define T4_EQ_ENTRY_SIZE 64
+
+#define T4_SQ_NUM_SLOTS 5
+#define T4_SQ_NUM_BYTES (T4_EQ_ENTRY_SIZE * T4_SQ_NUM_SLOTS)
+#define T4_MAX_SEND_SGE ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_send_wr) - \
+                       sizeof(struct fw_ri_isgl)) / sizeof(struct fw_ri_sge))
+#define T4_MAX_SEND_INLINE ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_send_wr) - \
+                       sizeof(struct fw_ri_immd)))
+#define T4_MAX_WRITE_INLINE ((T4_SQ_NUM_BYTES - \
+                       sizeof(struct fw_ri_rdma_write_wr) - \
+                       sizeof(struct fw_ri_immd)))
+#define T4_MAX_WRITE_SGE ((T4_SQ_NUM_BYTES - \
+                       sizeof(struct fw_ri_rdma_write_wr) - \
+                       sizeof(struct fw_ri_isgl)) / sizeof(struct fw_ri_sge))
+#define T4_MAX_FR_IMMD ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_fr_nsmr_wr) - \
+                       sizeof(struct fw_ri_immd)) & ~31UL)
+#define T4_MAX_FR_IMMD_DEPTH (T4_MAX_FR_IMMD / sizeof(u64))
+#define T4_MAX_FR_DSGL 1024
+#define T4_MAX_FR_DSGL_DEPTH (T4_MAX_FR_DSGL / sizeof(u64))
+
+static inline int t4_max_fr_depth(int use_dsgl)
+{
+       return use_dsgl ? T4_MAX_FR_DSGL_DEPTH : T4_MAX_FR_IMMD_DEPTH;
+}
+
+#define T4_RQ_NUM_SLOTS 2
+#define T4_RQ_NUM_BYTES (T4_EQ_ENTRY_SIZE * T4_RQ_NUM_SLOTS)
+#define T4_MAX_RECV_SGE 4
+
+union t4_wr {
+       struct fw_ri_res_wr res;
+       struct fw_ri_wr ri;
+       struct fw_ri_rdma_write_wr write;
+       struct fw_ri_send_wr send;
+       struct fw_ri_rdma_read_wr read;
+       struct fw_ri_bind_mw_wr bind;
+       struct fw_ri_fr_nsmr_wr fr;
+       struct fw_ri_inv_lstag_wr inv;
+       struct t4_status_page status;
+       __be64 flits[T4_EQ_ENTRY_SIZE / sizeof(__be64) * T4_SQ_NUM_SLOTS];
+};
+
+union t4_recv_wr {
+       struct fw_ri_recv_wr recv;
+       struct t4_status_page status;
+       __be64 flits[T4_EQ_ENTRY_SIZE / sizeof(__be64) * T4_RQ_NUM_SLOTS];
+};
+
+static inline void init_wr_hdr(union t4_wr *wqe, u16 wrid,
+                              enum fw_wr_opcodes opcode, u8 flags, u8 len16)
+{
+       wqe->send.opcode = (u8)opcode;
+       wqe->send.flags = flags;
+       wqe->send.wrid = wrid;
+       wqe->send.r1[0] = 0;
+       wqe->send.r1[1] = 0;
+       wqe->send.r1[2] = 0;
+       wqe->send.len16 = len16;
+}
+
+/* CQE/AE status codes */
+#define T4_ERR_SUCCESS                     0x0
+#define T4_ERR_STAG                        0x1 /* STAG invalid: either the */
+                                               /* STAG is offlimt, being 0, */
+                                               /* or STAG_key mismatch */
+#define T4_ERR_PDID                        0x2 /* PDID mismatch */
+#define T4_ERR_QPID                        0x3 /* QPID mismatch */
+#define T4_ERR_ACCESS                      0x4 /* Invalid access right */
+#define T4_ERR_WRAP                        0x5 /* Wrap error */
+#define T4_ERR_BOUND                       0x6 /* base and bounds voilation */
+#define T4_ERR_INVALIDATE_SHARED_MR        0x7 /* attempt to invalidate a  */
+                                               /* shared memory region */
+#define T4_ERR_INVALIDATE_MR_WITH_MW_BOUND 0x8 /* attempt to invalidate a  */
+                                               /* shared memory region */
+#define T4_ERR_ECC                         0x9 /* ECC error detected */
+#define T4_ERR_ECC_PSTAG                   0xA /* ECC error detected when  */
+                                               /* reading PSTAG for a MW  */
+                                               /* Invalidate */
+#define T4_ERR_PBL_ADDR_BOUND              0xB /* pbl addr out of bounds:  */
+                                               /* software error */
+#define T4_ERR_SWFLUSH                    0xC  /* SW FLUSHED */
+#define T4_ERR_CRC                         0x10 /* CRC error */
+#define T4_ERR_MARKER                      0x11 /* Marker error */
+#define T4_ERR_PDU_LEN_ERR                 0x12 /* invalid PDU length */
+#define T4_ERR_OUT_OF_RQE                  0x13 /* out of RQE */
+#define T4_ERR_DDP_VERSION                 0x14 /* wrong DDP version */
+#define T4_ERR_RDMA_VERSION                0x15 /* wrong RDMA version */
+#define T4_ERR_OPCODE                      0x16 /* invalid rdma opcode */
+#define T4_ERR_DDP_QUEUE_NUM               0x17 /* invalid ddp queue number */
+#define T4_ERR_MSN                         0x18 /* MSN error */
+#define T4_ERR_TBIT                        0x19 /* tag bit not set correctly */
+#define T4_ERR_MO                          0x1A /* MO not 0 for TERMINATE  */
+                                               /* or READ_REQ */
+#define T4_ERR_MSN_GAP                     0x1B
+#define T4_ERR_MSN_RANGE                   0x1C
+#define T4_ERR_IRD_OVERFLOW                0x1D
+#define T4_ERR_RQE_ADDR_BOUND              0x1E /* RQE addr out of bounds:  */
+                                               /* software error */
+#define T4_ERR_INTERNAL_ERR                0x1F /* internal error (opcode  */
+                                               /* mismatch) */
+/*
+ * CQE defs
+ */
+struct t4_cqe {
+       __be32 header;
+       __be32 len;
+       union {
+               struct {
+                       __be32 stag;
+                       __be32 msn;
+               } rcqe;
+               struct {
+                       u32 nada1;
+                       u16 nada2;
+                       u16 cidx;
+               } scqe;
+               struct {
+                       __be32 wrid_hi;
+                       __be32 wrid_low;
+               } gen;
+       } u;
+       __be64 reserved;
+       __be64 bits_type_ts;
+};
+
+/* macros for flit 0 of the cqe */
+
+#define CQE_QPID_S        12
+#define CQE_QPID_M        0xFFFFF
+#define CQE_QPID_G(x)     ((((x) >> CQE_QPID_S)) & CQE_QPID_M)
+#define CQE_QPID_V(x)    ((x)<<CQE_QPID_S)
+
+#define CQE_SWCQE_S       11
+#define CQE_SWCQE_M       0x1
+#define CQE_SWCQE_G(x)    ((((x) >> CQE_SWCQE_S)) & CQE_SWCQE_M)
+#define CQE_SWCQE_V(x)   ((x)<<CQE_SWCQE_S)
+
+#define CQE_STATUS_S      5
+#define CQE_STATUS_M      0x1F
+#define CQE_STATUS_G(x)   ((((x) >> CQE_STATUS_S)) & CQE_STATUS_M)
+#define CQE_STATUS_V(x)   ((x)<<CQE_STATUS_S)
+
+#define CQE_TYPE_S        4
+#define CQE_TYPE_M        0x1
+#define CQE_TYPE_G(x)     ((((x) >> CQE_TYPE_S)) & CQE_TYPE_M)
+#define CQE_TYPE_V(x)     ((x)<<CQE_TYPE_S)
+
+#define CQE_OPCODE_S      0
+#define CQE_OPCODE_M      0xF
+#define CQE_OPCODE_G(x)   ((((x) >> CQE_OPCODE_S)) & CQE_OPCODE_M)
+#define CQE_OPCODE_V(x)   ((x)<<CQE_OPCODE_S)
+
+#define SW_CQE(x)         (CQE_SWCQE_G(be32_to_cpu((x)->header)))
+#define CQE_QPID(x)       (CQE_QPID_G(be32_to_cpu((x)->header)))
+#define CQE_TYPE(x)       (CQE_TYPE_G(be32_to_cpu((x)->header)))
+#define SQ_TYPE(x)       (CQE_TYPE((x)))
+#define RQ_TYPE(x)       (!CQE_TYPE((x)))
+#define CQE_STATUS(x)     (CQE_STATUS_G(be32_to_cpu((x)->header)))
+#define CQE_OPCODE(x)     (CQE_OPCODE_G(be32_to_cpu((x)->header)))
+
+#define CQE_SEND_OPCODE(x)( \
+       (CQE_OPCODE_G(be32_to_cpu((x)->header)) == FW_RI_SEND) || \
+       (CQE_OPCODE_G(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_SE) || \
+       (CQE_OPCODE_G(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_INV) || \
+       (CQE_OPCODE_G(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_SE_INV))
+
+#define CQE_LEN(x)        (be32_to_cpu((x)->len))
+
+/* used for RQ completion processing */
+#define CQE_WRID_STAG(x)  (be32_to_cpu((x)->u.rcqe.stag))
+#define CQE_WRID_MSN(x)   (be32_to_cpu((x)->u.rcqe.msn))
+
+/* used for SQ completion processing */
+#define CQE_WRID_SQ_IDX(x)     ((x)->u.scqe.cidx)
+
+/* generic accessor macros */
+#define CQE_WRID_HI(x)         (be32_to_cpu((x)->u.gen.wrid_hi))
+#define CQE_WRID_LOW(x)                (be32_to_cpu((x)->u.gen.wrid_low))
+
+/* macros for flit 3 of the cqe */
+#define CQE_GENBIT_S   63
+#define CQE_GENBIT_M   0x1
+#define CQE_GENBIT_G(x)        (((x) >> CQE_GENBIT_S) & CQE_GENBIT_M)
+#define CQE_GENBIT_V(x) ((x)<<CQE_GENBIT_S)
+
+#define CQE_OVFBIT_S   62
+#define CQE_OVFBIT_M   0x1
+#define CQE_OVFBIT_G(x)        ((((x) >> CQE_OVFBIT_S)) & CQE_OVFBIT_M)
+
+#define CQE_IQTYPE_S   60
+#define CQE_IQTYPE_M   0x3
+#define CQE_IQTYPE_G(x)        ((((x) >> CQE_IQTYPE_S)) & CQE_IQTYPE_M)
+
+#define CQE_TS_M       0x0fffffffffffffffULL
+#define CQE_TS_G(x)    ((x) & CQE_TS_M)
+
+#define CQE_OVFBIT(x)  ((unsigned)CQE_OVFBIT_G(be64_to_cpu((x)->bits_type_ts)))
+#define CQE_GENBIT(x)  ((unsigned)CQE_GENBIT_G(be64_to_cpu((x)->bits_type_ts)))
+#define CQE_TS(x)      (CQE_TS_G(be64_to_cpu((x)->bits_type_ts)))
+
+struct t4_swsqe {
+       u64                     wr_id;
+       struct t4_cqe           cqe;
+       int                     read_len;
+       int                     opcode;
+       int                     complete;
+       int                     signaled;
+       u16                     idx;
+       int                     flushed;
+       struct timespec         host_ts;
+       u64                     sge_ts;
+};
+
+static inline pgprot_t t4_pgprot_wc(pgprot_t prot)
+{
+#if defined(__i386__) || defined(__x86_64__) || defined(CONFIG_PPC64)
+       return pgprot_writecombine(prot);
+#else
+       return pgprot_noncached(prot);
+#endif
+}
+
+enum {
+       T4_SQ_ONCHIP = (1<<0),
+};
+
+struct t4_sq {
+       union t4_wr *queue;
+       dma_addr_t dma_addr;
+       DEFINE_DMA_UNMAP_ADDR(mapping);
+       unsigned long phys_addr;
+       struct t4_swsqe *sw_sq;
+       struct t4_swsqe *oldest_read;
+       u64 __iomem *udb;
+       size_t memsize;
+       u32 qid;
+       u16 in_use;
+       u16 size;
+       u16 cidx;
+       u16 pidx;
+       u16 wq_pidx;
+       u16 wq_pidx_inc;
+       u16 flags;
+       short flush_cidx;
+};
+
+struct t4_swrqe {
+       u64 wr_id;
+       struct timespec host_ts;
+       u64 sge_ts;
+};
+
+struct t4_rq {
+       union  t4_recv_wr *queue;
+       dma_addr_t dma_addr;
+       DEFINE_DMA_UNMAP_ADDR(mapping);
+       struct t4_swrqe *sw_rq;
+       u64 __iomem *udb;
+       size_t memsize;
+       u32 qid;
+       u32 msn;
+       u32 rqt_hwaddr;
+       u16 rqt_size;
+       u16 in_use;
+       u16 size;
+       u16 cidx;
+       u16 pidx;
+       u16 wq_pidx;
+       u16 wq_pidx_inc;
+};
+
+struct t4_wq {
+       struct t4_sq sq;
+       struct t4_rq rq;
+       void __iomem *db;
+       void __iomem *gts;
+       struct c4iw_rdev *rdev;
+       int flushed;
+};
+
+static inline int t4_rqes_posted(struct t4_wq *wq)
+{
+       return wq->rq.in_use;
+}
+
+static inline int t4_rq_empty(struct t4_wq *wq)
+{
+       return wq->rq.in_use == 0;
+}
+
+static inline int t4_rq_full(struct t4_wq *wq)
+{
+       return wq->rq.in_use == (wq->rq.size - 1);
+}
+
+static inline u32 t4_rq_avail(struct t4_wq *wq)
+{
+       return wq->rq.size - 1 - wq->rq.in_use;
+}
+
+static inline void t4_rq_produce(struct t4_wq *wq, u8 len16)
+{
+       wq->rq.in_use++;
+       if (++wq->rq.pidx == wq->rq.size)
+               wq->rq.pidx = 0;
+       wq->rq.wq_pidx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE);
+       if (wq->rq.wq_pidx >= wq->rq.size * T4_RQ_NUM_SLOTS)
+               wq->rq.wq_pidx %= wq->rq.size * T4_RQ_NUM_SLOTS;
+}
+
+static inline void t4_rq_consume(struct t4_wq *wq)
+{
+       wq->rq.in_use--;
+       wq->rq.msn++;
+       if (++wq->rq.cidx == wq->rq.size)
+               wq->rq.cidx = 0;
+}
+
+static inline u16 t4_rq_host_wq_pidx(struct t4_wq *wq)
+{
+       return wq->rq.queue[wq->rq.size].status.host_wq_pidx;
+}
+
+static inline u16 t4_rq_wq_size(struct t4_wq *wq)
+{
+               return wq->rq.size * T4_RQ_NUM_SLOTS;
+}
+
+static inline int t4_sq_onchip(struct t4_sq *sq)
+{
+       return sq->flags & T4_SQ_ONCHIP;
+}
+
+static inline int t4_sq_empty(struct t4_wq *wq)
+{
+       return wq->sq.in_use == 0;
+}
+
+static inline int t4_sq_full(struct t4_wq *wq)
+{
+       return wq->sq.in_use == (wq->sq.size - 1);
+}
+
+static inline u32 t4_sq_avail(struct t4_wq *wq)
+{
+       return wq->sq.size - 1 - wq->sq.in_use;
+}
+
+static inline void t4_sq_produce(struct t4_wq *wq, u8 len16)
+{
+       wq->sq.in_use++;
+       if (++wq->sq.pidx == wq->sq.size)
+               wq->sq.pidx = 0;
+       wq->sq.wq_pidx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE);
+       if (wq->sq.wq_pidx >= wq->sq.size * T4_SQ_NUM_SLOTS)
+               wq->sq.wq_pidx %= wq->sq.size * T4_SQ_NUM_SLOTS;
+}
+
+static inline void t4_sq_consume(struct t4_wq *wq)
+{
+       BUG_ON(wq->sq.in_use < 1);
+       if (wq->sq.cidx == wq->sq.flush_cidx)
+               wq->sq.flush_cidx = -1;
+       wq->sq.in_use--;
+       if (++wq->sq.cidx == wq->sq.size)
+               wq->sq.cidx = 0;
+}
+
+static inline u16 t4_sq_host_wq_pidx(struct t4_wq *wq)
+{
+       return wq->sq.queue[wq->sq.size].status.host_wq_pidx;
+}
+
+static inline u16 t4_sq_wq_size(struct t4_wq *wq)
+{
+               return wq->sq.size * T4_SQ_NUM_SLOTS;
+}
+
+/* This function copies 64 byte coalesced work request to memory
+ * mapped BAR2 space. For coalesced WRs, the SGE fetches data
+ * from the FIFO instead of from Host.
+ */
+static inline void pio_copy(u64 __iomem *dst, u64 *src)
+{
+       int count = 8;
+
+       while (count) {
+               writeq(*src, dst);
+               src++;
+               dst++;
+               count--;
+       }
+}
+
+static inline void t4_ring_sq_db(struct t4_wq *wq, u16 inc, u8 t5,
+                                union t4_wr *wqe)
+{
+
+       /* Flush host queue memory writes. */
+       wmb();
+       if (t5) {
+               if (inc == 1 && wqe) {
+                       PDBG("%s: WC wq->sq.pidx = %d\n",
+                            __func__, wq->sq.pidx);
+                       pio_copy(wq->sq.udb + 7, (void *)wqe);
+               } else {
+                       PDBG("%s: DB wq->sq.pidx = %d\n",
+                            __func__, wq->sq.pidx);
+                       writel(PIDX_T5_V(inc), wq->sq.udb);
+               }
+
+               /* Flush user doorbell area writes. */
+               wmb();
+               return;
+       }
+       writel(QID_V(wq->sq.qid) | PIDX_V(inc), wq->db);
+}
+
+static inline void t4_ring_rq_db(struct t4_wq *wq, u16 inc, u8 t5,
+                                union t4_recv_wr *wqe)
+{
+
+       /* Flush host queue memory writes. */
+       wmb();
+       if (t5) {
+               if (inc == 1 && wqe) {
+                       PDBG("%s: WC wq->rq.pidx = %d\n",
+                            __func__, wq->rq.pidx);
+                       pio_copy(wq->rq.udb + 7, (void *)wqe);
+               } else {
+                       PDBG("%s: DB wq->rq.pidx = %d\n",
+                            __func__, wq->rq.pidx);
+                       writel(PIDX_T5_V(inc), wq->rq.udb);
+               }
+
+               /* Flush user doorbell area writes. */
+               wmb();
+               return;
+       }
+       writel(QID_V(wq->rq.qid) | PIDX_V(inc), wq->db);
+}
+
+static inline int t4_wq_in_error(struct t4_wq *wq)
+{
+       return wq->rq.queue[wq->rq.size].status.qp_err;
+}
+
+static inline void t4_set_wq_in_error(struct t4_wq *wq)
+{
+       wq->rq.queue[wq->rq.size].status.qp_err = 1;
+}
+
+static inline void t4_disable_wq_db(struct t4_wq *wq)
+{
+       wq->rq.queue[wq->rq.size].status.db_off = 1;
+}
+
+static inline void t4_enable_wq_db(struct t4_wq *wq)
+{
+       wq->rq.queue[wq->rq.size].status.db_off = 0;
+}
+
+static inline int t4_wq_db_enabled(struct t4_wq *wq)
+{
+       return !wq->rq.queue[wq->rq.size].status.db_off;
+}
+
+enum t4_cq_flags {
+       CQ_ARMED        = 1,
+};
+
+struct t4_cq {
+       struct t4_cqe *queue;
+       dma_addr_t dma_addr;
+       DEFINE_DMA_UNMAP_ADDR(mapping);
+       struct t4_cqe *sw_queue;
+       void __iomem *gts;
+       struct c4iw_rdev *rdev;
+       u64 ugts;
+       size_t memsize;
+       __be64 bits_type_ts;
+       u32 cqid;
+       u32 qid_mask;
+       int vector;
+       u16 size; /* including status page */
+       u16 cidx;
+       u16 sw_pidx;
+       u16 sw_cidx;
+       u16 sw_in_use;
+       u16 cidx_inc;
+       u8 gen;
+       u8 error;
+       unsigned long flags;
+};
+
+static inline int t4_clear_cq_armed(struct t4_cq *cq)
+{
+       return test_and_clear_bit(CQ_ARMED, &cq->flags);
+}
+
+static inline int t4_arm_cq(struct t4_cq *cq, int se)
+{
+       u32 val;
+
+       set_bit(CQ_ARMED, &cq->flags);
+       while (cq->cidx_inc > CIDXINC_M) {
+               val = SEINTARM_V(0) | CIDXINC_V(CIDXINC_M) | TIMERREG_V(7) |
+                     INGRESSQID_V(cq->cqid & cq->qid_mask);
+               writel(val, cq->gts);
+               cq->cidx_inc -= CIDXINC_M;
+       }
+       val = SEINTARM_V(se) | CIDXINC_V(cq->cidx_inc) | TIMERREG_V(6) |
+             INGRESSQID_V(cq->cqid & cq->qid_mask);
+       writel(val, cq->gts);
+       cq->cidx_inc = 0;
+       return 0;
+}
+
+static inline void t4_swcq_produce(struct t4_cq *cq)
+{
+       cq->sw_in_use++;
+       if (cq->sw_in_use == cq->size) {
+               PDBG("%s cxgb4 sw cq overflow cqid %u\n", __func__, cq->cqid);
+               cq->error = 1;
+               BUG_ON(1);
+       }
+       if (++cq->sw_pidx == cq->size)
+               cq->sw_pidx = 0;
+}
+
+static inline void t4_swcq_consume(struct t4_cq *cq)
+{
+       BUG_ON(cq->sw_in_use < 1);
+       cq->sw_in_use--;
+       if (++cq->sw_cidx == cq->size)
+               cq->sw_cidx = 0;
+}
+
+static inline void t4_hwcq_consume(struct t4_cq *cq)
+{
+       cq->bits_type_ts = cq->queue[cq->cidx].bits_type_ts;
+       if (++cq->cidx_inc == (cq->size >> 4) || cq->cidx_inc == CIDXINC_M) {
+               u32 val;
+
+               val = SEINTARM_V(0) | CIDXINC_V(cq->cidx_inc) | TIMERREG_V(7) |
+                     INGRESSQID_V(cq->cqid & cq->qid_mask);
+               writel(val, cq->gts);
+               cq->cidx_inc = 0;
+       }
+       if (++cq->cidx == cq->size) {
+               cq->cidx = 0;
+               cq->gen ^= 1;
+       }
+}
+
+static inline int t4_valid_cqe(struct t4_cq *cq, struct t4_cqe *cqe)
+{
+       return (CQE_GENBIT(cqe) == cq->gen);
+}
+
+static inline int t4_next_hw_cqe(struct t4_cq *cq, struct t4_cqe **cqe)
+{
+       int ret;
+       u16 prev_cidx;
+
+       if (cq->cidx == 0)
+               prev_cidx = cq->size - 1;
+       else
+               prev_cidx = cq->cidx - 1;
+
+       if (cq->queue[prev_cidx].bits_type_ts != cq->bits_type_ts) {
+               ret = -EOVERFLOW;
+               cq->error = 1;
+               printk(KERN_ERR MOD "cq overflow cqid %u\n", cq->cqid);
+               BUG_ON(1);
+       } else if (t4_valid_cqe(cq, &cq->queue[cq->cidx])) {
+
+               /* Ensure CQE is flushed to memory */
+               rmb();
+               *cqe = &cq->queue[cq->cidx];
+               ret = 0;
+       } else
+               ret = -ENODATA;
+       return ret;
+}
+
+static inline struct t4_cqe *t4_next_sw_cqe(struct t4_cq *cq)
+{
+       if (cq->sw_in_use == cq->size) {
+               PDBG("%s cxgb4 sw cq overflow cqid %u\n", __func__, cq->cqid);
+               cq->error = 1;
+               BUG_ON(1);
+               return NULL;
+       }
+       if (cq->sw_in_use)
+               return &cq->sw_queue[cq->sw_cidx];
+       return NULL;
+}
+
+static inline int t4_next_cqe(struct t4_cq *cq, struct t4_cqe **cqe)
+{
+       int ret = 0;
+
+       if (cq->error)
+               ret = -ENODATA;
+       else if (cq->sw_in_use)
+               *cqe = &cq->sw_queue[cq->sw_cidx];
+       else
+               ret = t4_next_hw_cqe(cq, cqe);
+       return ret;
+}
+
+static inline int t4_cq_in_error(struct t4_cq *cq)
+{
+       return ((struct t4_status_page *)&cq->queue[cq->size])->qp_err;
+}
+
+static inline void t4_set_cq_in_error(struct t4_cq *cq)
+{
+       ((struct t4_status_page *)&cq->queue[cq->size])->qp_err = 1;
+}
+#endif
+
+struct t4_dev_status_page {
+       u8 db_off;
+};