bus/dpaa: support event dequeue and consumption
[dpdk.git] / drivers / bus / dpaa / base / qbman / qman.c
index d8fb25a..609bc76 100644 (file)
@@ -8,6 +8,8 @@
 #include "qman.h"
 #include <rte_branch_prediction.h>
 #include <rte_dpaa_bus.h>
+#include <rte_eventdev.h>
+#include <rte_byteorder.h>
 
 /* Compilation constants */
 #define DQRR_MAXFILL   15
@@ -1051,6 +1053,138 @@ u16 qman_affine_channel(int cpu)
        return affine_channels[cpu];
 }
 
+unsigned int qman_portal_poll_rx(unsigned int poll_limit,
+                                void **bufs,
+                                struct qman_portal *p)
+{
+       const struct qm_dqrr_entry *dq;
+       struct qman_fq *fq;
+       enum qman_cb_dqrr_result res;
+       unsigned int limit = 0;
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+       struct qm_dqrr_entry *shadow;
+#endif
+       unsigned int rx_number = 0;
+
+       do {
+               qm_dqrr_pvb_update(&p->p);
+               dq = qm_dqrr_current(&p->p);
+               if (unlikely(!dq))
+                       break;
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+       /* If running on an LE system the fields of the
+        * dequeue entry must be swapper.  Because the
+        * QMan HW will ignore writes the DQRR entry is
+        * copied and the index stored within the copy
+        */
+               shadow = &p->shadow_dqrr[DQRR_PTR2IDX(dq)];
+               *shadow = *dq;
+               dq = shadow;
+               shadow->fqid = be32_to_cpu(shadow->fqid);
+               shadow->contextB = be32_to_cpu(shadow->contextB);
+               shadow->seqnum = be16_to_cpu(shadow->seqnum);
+               hw_fd_to_cpu(&shadow->fd);
+#endif
+
+               /* SDQCR: context_b points to the FQ */
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+               fq = get_fq_table_entry(dq->contextB);
+#else
+               fq = (void *)(uintptr_t)dq->contextB;
+#endif
+               /* Now let the callback do its stuff */
+               res = fq->cb.dqrr_dpdk_cb(NULL, p, fq, dq, &bufs[rx_number]);
+               rx_number++;
+               /* Interpret 'dq' from a driver perspective. */
+               /*
+                * Parking isn't possible unless HELDACTIVE was set. NB,
+                * FORCEELIGIBLE implies HELDACTIVE, so we only need to
+                * check for HELDACTIVE to cover both.
+                */
+               DPAA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) ||
+                           (res != qman_cb_dqrr_park));
+               qm_dqrr_cdc_consume_1ptr(&p->p, dq, res == qman_cb_dqrr_park);
+               /* Move forward */
+               qm_dqrr_next(&p->p);
+               /*
+                * Entry processed and consumed, increment our counter.  The
+                * callback can request that we exit after consuming the
+                * entry, and we also exit if we reach our processing limit,
+                * so loop back only if neither of these conditions is met.
+                */
+       } while (likely(++limit < poll_limit));
+
+       return limit;
+}
+
+u32 qman_portal_dequeue(struct rte_event ev[], unsigned int poll_limit,
+                       void **bufs)
+{
+       const struct qm_dqrr_entry *dq;
+       struct qman_fq *fq;
+       enum qman_cb_dqrr_result res;
+       unsigned int limit = 0;
+       struct qman_portal *p = get_affine_portal();
+#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
+       struct qm_dqrr_entry *shadow;
+#endif
+       unsigned int rx_number = 0;
+
+       do {
+               qm_dqrr_pvb_update(&p->p);
+               dq = qm_dqrr_current(&p->p);
+               if (!dq)
+                       break;
+#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
+               /*
+                * If running on an LE system the fields of the
+                * dequeue entry must be swapper.  Because the
+                * QMan HW will ignore writes the DQRR entry is
+                * copied and the index stored within the copy
+                */
+               shadow = &p->shadow_dqrr[DQRR_PTR2IDX(dq)];
+               *shadow = *dq;
+               dq = shadow;
+               shadow->fqid = be32_to_cpu(shadow->fqid);
+               shadow->contextB = be32_to_cpu(shadow->contextB);
+               shadow->seqnum = be16_to_cpu(shadow->seqnum);
+               hw_fd_to_cpu(&shadow->fd);
+#endif
+
+              /* SDQCR: context_b points to the FQ */
+#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP
+               fq = get_fq_table_entry(dq->contextB);
+#else
+               fq = (void *)(uintptr_t)dq->contextB;
+#endif
+               /* Now let the callback do its stuff */
+               res = fq->cb.dqrr_dpdk_cb(&ev[rx_number], p, fq,
+                                        dq, &bufs[rx_number]);
+               rx_number++;
+               /* Interpret 'dq' from a driver perspective. */
+               /*
+                * Parking isn't possible unless HELDACTIVE was set. NB,
+                * FORCEELIGIBLE implies HELDACTIVE, so we only need to
+                * check for HELDACTIVE to cover both.
+                */
+               DPAA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) ||
+                           (res != qman_cb_dqrr_park));
+               if (res != qman_cb_dqrr_defer)
+                       qm_dqrr_cdc_consume_1ptr(&p->p, dq,
+                                                res == qman_cb_dqrr_park);
+               /* Move forward */
+               qm_dqrr_next(&p->p);
+               /*
+                * Entry processed and consumed, increment our counter.  The
+                * callback can request that we exit after consuming the
+                * entry, and we also exit if we reach our processing limit,
+                * so loop back only if neither of these conditions is met.
+                */
+       } while (++limit < poll_limit);
+
+       return limit;
+}
+
 struct qm_dqrr_entry *qman_dequeue(struct qman_fq *fq)
 {
        struct qman_portal *p = get_affine_portal();
@@ -1169,13 +1303,20 @@ u32 qman_static_dequeue_get(struct qman_portal *qp)
        return p->sdqcr;
 }
 
-void qman_dca(struct qm_dqrr_entry *dq, int park_request)
+void qman_dca(const struct qm_dqrr_entry *dq, int park_request)
 {
        struct qman_portal *p = get_affine_portal();
 
        qm_dqrr_cdc_consume_1ptr(&p->p, dq, park_request);
 }
 
+void qman_dca_index(u8 index, int park_request)
+{
+       struct qman_portal *p = get_affine_portal();
+
+       qm_dqrr_cdc_consume_1(&p->p, index, park_request);
+}
+
 /* Frame queue API */
 static const char *mcr_result_str(u8 result)
 {
@@ -1722,6 +1863,28 @@ int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np)
        return 0;
 }
 
+int qman_query_fq_frm_cnt(struct qman_fq *fq, u32 *frm_cnt)
+{
+       struct qm_mc_command *mcc;
+       struct qm_mc_result *mcr;
+       struct qman_portal *p = get_affine_portal();
+
+       mcc = qm_mc_start(&p->p);
+       mcc->queryfq.fqid = cpu_to_be32(fq->fqid);
+       qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP);
+       while (!(mcr = qm_mc_result(&p->p)))
+               cpu_relax();
+       DPAA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP);
+
+       if (mcr->result == QM_MCR_RESULT_OK)
+               *frm_cnt = be24_to_cpu(mcr->queryfq_np.frm_cnt);
+       else if (mcr->result == QM_MCR_RESULT_ERR_FQID)
+               return -ERANGE;
+       else if (mcr->result != QM_MCR_RESULT_OK)
+               return -EIO;
+       return 0;
+}
+
 int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq)
 {
        struct qm_mc_command *mcc;
@@ -2002,8 +2165,8 @@ int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags)
 }
 
 int qman_enqueue_multi(struct qman_fq *fq,
-                      const struct qm_fd *fd,
-                      int frames_to_send)
+                      const struct qm_fd *fd, u32 *flags,
+               int frames_to_send)
 {
        struct qman_portal *p = get_affine_portal();
        struct qm_portal *portal = &p->p;
@@ -2011,7 +2174,7 @@ int qman_enqueue_multi(struct qman_fq *fq,
        register struct qm_eqcr *eqcr = &portal->eqcr;
        struct qm_eqcr_entry *eq = eqcr->cursor, *prev_eq;
 
-       u8 i, diff, old_ci, sent = 0;
+       u8 i = 0, diff, old_ci, sent = 0;
 
        /* Update the available entries if no entry is free */
        if (!eqcr->available) {
@@ -2035,6 +2198,76 @@ int qman_enqueue_multi(struct qman_fq *fq,
                eq->fd.addr = cpu_to_be40(fd->addr);
                eq->fd.status = cpu_to_be32(fd->status);
                eq->fd.opaque = cpu_to_be32(fd->opaque);
+               if (flags[i] & QMAN_ENQUEUE_FLAG_DCA) {
+                       eq->dca = QM_EQCR_DCA_ENABLE |
+                               ((flags[i] >> 8) & QM_EQCR_DCA_IDXMASK);
+               }
+               i++;
+               eq = (void *)((unsigned long)(eq + 1) &
+                       (~(unsigned long)(QM_EQCR_SIZE << 6)));
+               eqcr->available--;
+               sent++;
+               fd++;
+       }
+       lwsync();
+
+       /* In order for flushes to complete faster, all lines are recorded in
+        * 32 bit word.
+        */
+       eq = eqcr->cursor;
+       for (i = 0; i < sent; i++) {
+               eq->__dont_write_directly__verb =
+                       QM_EQCR_VERB_CMD_ENQUEUE | eqcr->vbit;
+               prev_eq = eq;
+               eq = (void *)((unsigned long)(eq + 1) &
+                       (~(unsigned long)(QM_EQCR_SIZE << 6)));
+               if (unlikely((prev_eq + 1) != eq))
+                       eqcr->vbit ^= QM_EQCR_VERB_VBIT;
+       }
+
+       /* We need  to flush all the lines but without load/store operations
+        * between them
+        */
+       eq = eqcr->cursor;
+       for (i = 0; i < sent; i++) {
+               dcbf(eq);
+               eq = (void *)((unsigned long)(eq + 1) &
+                       (~(unsigned long)(QM_EQCR_SIZE << 6)));
+       }
+       /* Update cursor for the next call */
+       eqcr->cursor = eq;
+       return sent;
+}
+
+int
+qman_enqueue_multi_fq(struct qman_fq *fq[], const struct qm_fd *fd,
+                     int frames_to_send)
+{
+       struct qman_portal *p = get_affine_portal();
+       struct qm_portal *portal = &p->p;
+
+       register struct qm_eqcr *eqcr = &portal->eqcr;
+       struct qm_eqcr_entry *eq = eqcr->cursor, *prev_eq;
+
+       u8 i, diff, old_ci, sent = 0;
+
+       /* Update the available entries if no entry is free */
+       if (!eqcr->available) {
+               old_ci = eqcr->ci;
+               eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1);
+               diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci);
+               eqcr->available += diff;
+               if (!diff)
+                       return 0;
+       }
+
+       /* try to send as many frames as possible */
+       while (eqcr->available && frames_to_send--) {
+               eq->fqid = fq[sent]->fqid_le;
+               eq->fd.opaque_addr = fd->opaque_addr;
+               eq->fd.addr = cpu_to_be40(fd->addr);
+               eq->fd.status = cpu_to_be32(fd->status);
+               eq->fd.opaque = cpu_to_be32(fd->opaque);
 
                eq = (void *)((unsigned long)(eq + 1) &
                        (~(unsigned long)(QM_EQCR_SIZE << 6)));