scsi: elx: efct: Hardware queue creation and deletion
authorJames Smart <jsmart2021@gmail.com>
Tue, 1 Jun 2021 23:55:00 +0000 (16:55 -0700)
committerMartin K. Petersen <martin.petersen@oracle.com>
Wed, 16 Jun 2021 03:39:31 +0000 (23:39 -0400)
Add routines for queue creation, deletion, and configuration.

Link: https://lore.kernel.org/r/20210601235512.20104-20-jsmart2021@gmail.com
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Co-developed-by: Ram Vegesna <ram.vegesna@broadcom.com>
Signed-off-by: Ram Vegesna <ram.vegesna@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/elx/efct/efct_hw.h
drivers/scsi/elx/efct/efct_hw_queues.c [new file with mode: 0644]

index 6c8ed6a..7440461 100644 (file)
@@ -593,6 +593,10 @@ efct_hw_setup(struct efct_hw *hw, void *os, struct pci_dev *pdev);
 int efct_hw_init(struct efct_hw *hw);
 int
 efct_hw_parse_filter(struct efct_hw *hw, void *value);
+int
+efct_hw_init_queues(struct efct_hw *hw);
+int
+efct_hw_map_wq_cpu(struct efct_hw *hw);
 uint64_t
 efct_get_wwnn(struct efct_hw *hw);
 uint64_t
diff --git a/drivers/scsi/elx/efct/efct_hw_queues.c b/drivers/scsi/elx/efct/efct_hw_queues.c
new file mode 100644 (file)
index 0000000..3a1d1a5
--- /dev/null
@@ -0,0 +1,677 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2021 Broadcom. All Rights Reserved. The term
+ * “Broadcom” refers to Broadcom Inc. and/or its subsidiaries.
+ */
+
+#include "efct_driver.h"
+#include "efct_hw.h"
+#include "efct_unsol.h"
+
+int
+efct_hw_init_queues(struct efct_hw *hw)
+{
+       struct hw_eq *eq = NULL;
+       struct hw_cq *cq = NULL;
+       struct hw_wq *wq = NULL;
+       struct hw_mq *mq = NULL;
+
+       struct hw_eq *eqs[EFCT_HW_MAX_NUM_EQ];
+       struct hw_cq *cqs[EFCT_HW_MAX_NUM_EQ];
+       struct hw_rq *rqs[EFCT_HW_MAX_NUM_EQ];
+       u32 i = 0, j;
+
+       hw->eq_count = 0;
+       hw->cq_count = 0;
+       hw->mq_count = 0;
+       hw->wq_count = 0;
+       hw->rq_count = 0;
+       hw->hw_rq_count = 0;
+       INIT_LIST_HEAD(&hw->eq_list);
+
+       for (i = 0; i < hw->config.n_eq; i++) {
+               /* Create EQ */
+               eq = efct_hw_new_eq(hw, EFCT_HW_EQ_DEPTH);
+               if (!eq) {
+                       efct_hw_queue_teardown(hw);
+                       return -ENOMEM;
+               }
+
+               eqs[i] = eq;
+
+               /* Create one MQ */
+               if (!i) {
+                       cq = efct_hw_new_cq(eq,
+                                           hw->num_qentries[SLI4_QTYPE_CQ]);
+                       if (!cq) {
+                               efct_hw_queue_teardown(hw);
+                               return -ENOMEM;
+                       }
+
+                       mq = efct_hw_new_mq(cq, EFCT_HW_MQ_DEPTH);
+                       if (!mq) {
+                               efct_hw_queue_teardown(hw);
+                               return -ENOMEM;
+                       }
+               }
+
+               /* Create WQ */
+               cq = efct_hw_new_cq(eq, hw->num_qentries[SLI4_QTYPE_CQ]);
+               if (!cq) {
+                       efct_hw_queue_teardown(hw);
+                       return -ENOMEM;
+               }
+
+               wq = efct_hw_new_wq(cq, hw->num_qentries[SLI4_QTYPE_WQ]);
+               if (!wq) {
+                       efct_hw_queue_teardown(hw);
+                       return -ENOMEM;
+               }
+       }
+
+       /* Create CQ set */
+       if (efct_hw_new_cq_set(eqs, cqs, i, hw->num_qentries[SLI4_QTYPE_CQ])) {
+               efct_hw_queue_teardown(hw);
+               return -EIO;
+       }
+
+       /* Create RQ set */
+       if (efct_hw_new_rq_set(cqs, rqs, i, EFCT_HW_RQ_ENTRIES_DEF)) {
+               efct_hw_queue_teardown(hw);
+               return -EIO;
+       }
+
+       for (j = 0; j < i ; j++) {
+               rqs[j]->filter_mask = 0;
+               rqs[j]->is_mrq = true;
+               rqs[j]->base_mrq_id = rqs[0]->hdr->id;
+       }
+
+       hw->hw_mrq_count = i;
+
+       return 0;
+}
+
+int
+efct_hw_map_wq_cpu(struct efct_hw *hw)
+{
+       struct efct *efct = hw->os;
+       u32 cpu = 0, i;
+
+       /* Init cpu_map array */
+       hw->wq_cpu_array = kcalloc(num_possible_cpus(), sizeof(void *),
+                                  GFP_KERNEL);
+       if (!hw->wq_cpu_array)
+               return -ENOMEM;
+
+       for (i = 0; i < hw->config.n_eq; i++) {
+               const struct cpumask *maskp;
+
+               /* Get a CPU mask for all CPUs affinitized to this vector */
+               maskp = pci_irq_get_affinity(efct->pci, i);
+               if (!maskp) {
+                       efc_log_debug(efct, "maskp null for vector:%d\n", i);
+                       continue;
+               }
+
+               /* Loop through all CPUs associated with vector idx */
+               for_each_cpu_and(cpu, maskp, cpu_present_mask) {
+                       efc_log_debug(efct, "CPU:%d irq vector:%d\n", cpu, i);
+                       hw->wq_cpu_array[cpu] = hw->hw_wq[i];
+               }
+       }
+
+       return 0;
+}
+
+struct hw_eq *
+efct_hw_new_eq(struct efct_hw *hw, u32 entry_count)
+{
+       struct hw_eq *eq = kzalloc(sizeof(*eq), GFP_KERNEL);
+
+       if (!eq)
+               return NULL;
+
+       eq->type = SLI4_QTYPE_EQ;
+       eq->hw = hw;
+       eq->entry_count = entry_count;
+       eq->instance = hw->eq_count++;
+       eq->queue = &hw->eq[eq->instance];
+       INIT_LIST_HEAD(&eq->cq_list);
+
+       if (sli_queue_alloc(&hw->sli, SLI4_QTYPE_EQ, eq->queue, entry_count,
+                           NULL)) {
+               efc_log_err(hw->os, "EQ[%d] alloc failure\n", eq->instance);
+               kfree(eq);
+               return NULL;
+       }
+
+       sli_eq_modify_delay(&hw->sli, eq->queue, 1, 0, 8);
+       hw->hw_eq[eq->instance] = eq;
+       INIT_LIST_HEAD(&eq->list_entry);
+       list_add_tail(&eq->list_entry, &hw->eq_list);
+       efc_log_debug(hw->os, "create eq[%2d] id %3d len %4d\n", eq->instance,
+                     eq->queue->id, eq->entry_count);
+       return eq;
+}
+
+struct hw_cq *
+efct_hw_new_cq(struct hw_eq *eq, u32 entry_count)
+{
+       struct efct_hw *hw = eq->hw;
+       struct hw_cq *cq = kzalloc(sizeof(*cq), GFP_KERNEL);
+
+       if (!cq)
+               return NULL;
+
+       cq->eq = eq;
+       cq->type = SLI4_QTYPE_CQ;
+       cq->instance = eq->hw->cq_count++;
+       cq->entry_count = entry_count;
+       cq->queue = &hw->cq[cq->instance];
+
+       INIT_LIST_HEAD(&cq->q_list);
+
+       if (sli_queue_alloc(&hw->sli, SLI4_QTYPE_CQ, cq->queue,
+                           cq->entry_count, eq->queue)) {
+               efc_log_err(hw->os, "CQ[%d] allocation failure len=%d\n",
+                           eq->instance, eq->entry_count);
+               kfree(cq);
+               return NULL;
+       }
+
+       hw->hw_cq[cq->instance] = cq;
+       INIT_LIST_HEAD(&cq->list_entry);
+       list_add_tail(&cq->list_entry, &eq->cq_list);
+       efc_log_debug(hw->os, "create cq[%2d] id %3d len %4d\n", cq->instance,
+                     cq->queue->id, cq->entry_count);
+       return cq;
+}
+
+u32
+efct_hw_new_cq_set(struct hw_eq *eqs[], struct hw_cq *cqs[],
+                  u32 num_cqs, u32 entry_count)
+{
+       u32 i;
+       struct efct_hw *hw = eqs[0]->hw;
+       struct sli4 *sli4 = &hw->sli;
+       struct hw_cq *cq = NULL;
+       struct sli4_queue *qs[SLI4_MAX_CQ_SET_COUNT];
+       struct sli4_queue *assefct[SLI4_MAX_CQ_SET_COUNT];
+
+       /* Initialise CQS pointers to NULL */
+       for (i = 0; i < num_cqs; i++)
+               cqs[i] = NULL;
+
+       for (i = 0; i < num_cqs; i++) {
+               cq = kzalloc(sizeof(*cq), GFP_KERNEL);
+               if (!cq)
+                       goto error;
+
+               cqs[i]          = cq;
+               cq->eq          = eqs[i];
+               cq->type        = SLI4_QTYPE_CQ;
+               cq->instance    = hw->cq_count++;
+               cq->entry_count = entry_count;
+               cq->queue       = &hw->cq[cq->instance];
+               qs[i]           = cq->queue;
+               assefct[i]       = eqs[i]->queue;
+               INIT_LIST_HEAD(&cq->q_list);
+       }
+
+       if (sli_cq_alloc_set(sli4, qs, num_cqs, entry_count, assefct)) {
+               efc_log_err(hw->os, "Failed to create CQ Set.\n");
+               goto error;
+       }
+
+       for (i = 0; i < num_cqs; i++) {
+               hw->hw_cq[cqs[i]->instance] = cqs[i];
+               INIT_LIST_HEAD(&cqs[i]->list_entry);
+               list_add_tail(&cqs[i]->list_entry, &cqs[i]->eq->cq_list);
+       }
+
+       return 0;
+
+error:
+       for (i = 0; i < num_cqs; i++) {
+               kfree(cqs[i]);
+               cqs[i] = NULL;
+       }
+       return -EIO;
+}
+
+struct hw_mq *
+efct_hw_new_mq(struct hw_cq *cq, u32 entry_count)
+{
+       struct efct_hw *hw = cq->eq->hw;
+       struct hw_mq *mq = kzalloc(sizeof(*mq), GFP_KERNEL);
+
+       if (!mq)
+               return NULL;
+
+       mq->cq = cq;
+       mq->type = SLI4_QTYPE_MQ;
+       mq->instance = cq->eq->hw->mq_count++;
+       mq->entry_count = entry_count;
+       mq->entry_size = EFCT_HW_MQ_DEPTH;
+       mq->queue = &hw->mq[mq->instance];
+
+       if (sli_queue_alloc(&hw->sli, SLI4_QTYPE_MQ, mq->queue, mq->entry_size,
+                           cq->queue)) {
+               efc_log_err(hw->os, "MQ allocation failure\n");
+               kfree(mq);
+               return NULL;
+       }
+
+       hw->hw_mq[mq->instance] = mq;
+       INIT_LIST_HEAD(&mq->list_entry);
+       list_add_tail(&mq->list_entry, &cq->q_list);
+       efc_log_debug(hw->os, "create mq[%2d] id %3d len %4d\n", mq->instance,
+                     mq->queue->id, mq->entry_count);
+       return mq;
+}
+
+struct hw_wq *
+efct_hw_new_wq(struct hw_cq *cq, u32 entry_count)
+{
+       struct efct_hw *hw = cq->eq->hw;
+       struct hw_wq *wq = kzalloc(sizeof(*wq), GFP_KERNEL);
+
+       if (!wq)
+               return NULL;
+
+       wq->hw = cq->eq->hw;
+       wq->cq = cq;
+       wq->type = SLI4_QTYPE_WQ;
+       wq->instance = cq->eq->hw->wq_count++;
+       wq->entry_count = entry_count;
+       wq->queue = &hw->wq[wq->instance];
+       wq->wqec_set_count = EFCT_HW_WQEC_SET_COUNT;
+       wq->wqec_count = wq->wqec_set_count;
+       wq->free_count = wq->entry_count - 1;
+       INIT_LIST_HEAD(&wq->pending_list);
+
+       if (sli_queue_alloc(&hw->sli, SLI4_QTYPE_WQ, wq->queue,
+                           wq->entry_count, cq->queue)) {
+               efc_log_err(hw->os, "WQ allocation failure\n");
+               kfree(wq);
+               return NULL;
+       }
+
+       hw->hw_wq[wq->instance] = wq;
+       INIT_LIST_HEAD(&wq->list_entry);
+       list_add_tail(&wq->list_entry, &cq->q_list);
+       efc_log_debug(hw->os, "create wq[%2d] id %3d len %4d cls %d\n",
+                     wq->instance, wq->queue->id, wq->entry_count, wq->class);
+       return wq;
+}
+
+u32
+efct_hw_new_rq_set(struct hw_cq *cqs[], struct hw_rq *rqs[],
+                  u32 num_rq_pairs, u32 entry_count)
+{
+       struct efct_hw *hw = cqs[0]->eq->hw;
+       struct hw_rq *rq = NULL;
+       struct sli4_queue *qs[SLI4_MAX_RQ_SET_COUNT * 2] = { NULL };
+       u32 i, q_count, size;
+
+       /* Initialise RQS pointers */
+       for (i = 0; i < num_rq_pairs; i++)
+               rqs[i] = NULL;
+
+       /*
+        * Allocate an RQ object SET, where each element in set
+        * encapsulates 2 SLI queues (for rq pair)
+        */
+       for (i = 0, q_count = 0; i < num_rq_pairs; i++, q_count += 2) {
+               rq = kzalloc(sizeof(*rq), GFP_KERNEL);
+               if (!rq)
+                       goto error;
+
+               rqs[i] = rq;
+               rq->instance = hw->hw_rq_count++;
+               rq->cq = cqs[i];
+               rq->type = SLI4_QTYPE_RQ;
+               rq->entry_count = entry_count;
+
+               /* Header RQ */
+               rq->hdr = &hw->rq[hw->rq_count];
+               rq->hdr_entry_size = EFCT_HW_RQ_HEADER_SIZE;
+               hw->hw_rq_lookup[hw->rq_count] = rq->instance;
+               hw->rq_count++;
+               qs[q_count] = rq->hdr;
+
+               /* Data RQ */
+               rq->data = &hw->rq[hw->rq_count];
+               rq->data_entry_size = hw->config.rq_default_buffer_size;
+               hw->hw_rq_lookup[hw->rq_count] = rq->instance;
+               hw->rq_count++;
+               qs[q_count + 1] = rq->data;
+
+               rq->rq_tracker = NULL;
+       }
+
+       if (sli_fc_rq_set_alloc(&hw->sli, num_rq_pairs, qs,
+                               cqs[0]->queue->id,
+                           rqs[0]->entry_count,
+                           rqs[0]->hdr_entry_size,
+                           rqs[0]->data_entry_size)) {
+               efc_log_err(hw->os, "RQ Set alloc failure for base CQ=%d\n",
+                           cqs[0]->queue->id);
+               goto error;
+       }
+
+       for (i = 0; i < num_rq_pairs; i++) {
+               hw->hw_rq[rqs[i]->instance] = rqs[i];
+               INIT_LIST_HEAD(&rqs[i]->list_entry);
+               list_add_tail(&rqs[i]->list_entry, &cqs[i]->q_list);
+               size = sizeof(struct efc_hw_sequence *) * rqs[i]->entry_count;
+               rqs[i]->rq_tracker = kzalloc(size, GFP_KERNEL);
+               if (!rqs[i]->rq_tracker)
+                       goto error;
+       }
+
+       return 0;
+
+error:
+       for (i = 0; i < num_rq_pairs; i++) {
+               if (rqs[i]) {
+                       kfree(rqs[i]->rq_tracker);
+                       kfree(rqs[i]);
+               }
+       }
+
+       return -EIO;
+}
+
+void
+efct_hw_del_eq(struct hw_eq *eq)
+{
+       struct hw_cq *cq;
+       struct hw_cq *cq_next;
+
+       if (!eq)
+               return;
+
+       list_for_each_entry_safe(cq, cq_next, &eq->cq_list, list_entry)
+               efct_hw_del_cq(cq);
+       list_del(&eq->list_entry);
+       eq->hw->hw_eq[eq->instance] = NULL;
+       kfree(eq);
+}
+
+void
+efct_hw_del_cq(struct hw_cq *cq)
+{
+       struct hw_q *q;
+       struct hw_q *q_next;
+
+       if (!cq)
+               return;
+
+       list_for_each_entry_safe(q, q_next, &cq->q_list, list_entry) {
+               switch (q->type) {
+               case SLI4_QTYPE_MQ:
+                       efct_hw_del_mq((struct hw_mq *)q);
+                       break;
+               case SLI4_QTYPE_WQ:
+                       efct_hw_del_wq((struct hw_wq *)q);
+                       break;
+               case SLI4_QTYPE_RQ:
+                       efct_hw_del_rq((struct hw_rq *)q);
+                       break;
+               default:
+                       break;
+               }
+       }
+       list_del(&cq->list_entry);
+       cq->eq->hw->hw_cq[cq->instance] = NULL;
+       kfree(cq);
+}
+
+void
+efct_hw_del_mq(struct hw_mq *mq)
+{
+       if (!mq)
+               return;
+
+       list_del(&mq->list_entry);
+       mq->cq->eq->hw->hw_mq[mq->instance] = NULL;
+       kfree(mq);
+}
+
+void
+efct_hw_del_wq(struct hw_wq *wq)
+{
+       if (!wq)
+               return;
+
+       list_del(&wq->list_entry);
+       wq->cq->eq->hw->hw_wq[wq->instance] = NULL;
+       kfree(wq);
+}
+
+void
+efct_hw_del_rq(struct hw_rq *rq)
+{
+       struct efct_hw *hw = NULL;
+
+       if (!rq)
+               return;
+       /* Free RQ tracker */
+       kfree(rq->rq_tracker);
+       rq->rq_tracker = NULL;
+       list_del(&rq->list_entry);
+       hw = rq->cq->eq->hw;
+       hw->hw_rq[rq->instance] = NULL;
+       kfree(rq);
+}
+
+void
+efct_hw_queue_teardown(struct efct_hw *hw)
+{
+       struct hw_eq *eq;
+       struct hw_eq *eq_next;
+
+       if (!hw->eq_list.next)
+               return;
+
+       list_for_each_entry_safe(eq, eq_next, &hw->eq_list, list_entry)
+               efct_hw_del_eq(eq);
+}
+
+static inline int
+efct_hw_rqpair_find(struct efct_hw *hw, u16 rq_id)
+{
+       return efct_hw_queue_hash_find(hw->rq_hash, rq_id);
+}
+
+static struct efc_hw_sequence *
+efct_hw_rqpair_get(struct efct_hw *hw, u16 rqindex, u16 bufindex)
+{
+       struct sli4_queue *rq_hdr = &hw->rq[rqindex];
+       struct efc_hw_sequence *seq = NULL;
+       struct hw_rq *rq = hw->hw_rq[hw->hw_rq_lookup[rqindex]];
+       unsigned long flags = 0;
+
+       if (bufindex >= rq_hdr->length) {
+               efc_log_err(hw->os,
+                           "RQidx %d bufidx %d exceed ring len %d for id %d\n",
+                           rqindex, bufindex, rq_hdr->length, rq_hdr->id);
+               return NULL;
+       }
+
+       /* rq_hdr lock also covers rqindex+1 queue */
+       spin_lock_irqsave(&rq_hdr->lock, flags);
+
+       seq = rq->rq_tracker[bufindex];
+       rq->rq_tracker[bufindex] = NULL;
+
+       if (!seq) {
+               efc_log_err(hw->os,
+                           "RQbuf NULL, rqidx %d, bufidx %d, cur q idx = %d\n",
+                           rqindex, bufindex, rq_hdr->index);
+       }
+
+       spin_unlock_irqrestore(&rq_hdr->lock, flags);
+       return seq;
+}
+
+int
+efct_hw_rqpair_process_rq(struct efct_hw *hw, struct hw_cq *cq,
+                         u8 *cqe)
+{
+       u16 rq_id;
+       u32 index;
+       int rqindex;
+       int rq_status;
+       u32 h_len;
+       u32 p_len;
+       struct efc_hw_sequence *seq;
+       struct hw_rq *rq;
+
+       rq_status = sli_fc_rqe_rqid_and_index(&hw->sli, cqe,
+                                             &rq_id, &index);
+       if (rq_status != 0) {
+               switch (rq_status) {
+               case SLI4_FC_ASYNC_RQ_BUF_LEN_EXCEEDED:
+               case SLI4_FC_ASYNC_RQ_DMA_FAILURE:
+                       /* just get RQ buffer then return to chip */
+                       rqindex = efct_hw_rqpair_find(hw, rq_id);
+                       if (rqindex < 0) {
+                               efc_log_debug(hw->os,
+                                             "status=%#x: lookup fail id=%#x\n",
+                                            rq_status, rq_id);
+                               break;
+                       }
+
+                       /* get RQ buffer */
+                       seq = efct_hw_rqpair_get(hw, rqindex, index);
+
+                       /* return to chip */
+                       if (efct_hw_rqpair_sequence_free(hw, seq)) {
+                               efc_log_debug(hw->os,
+                                             "status=%#x,fail rtrn buf to RQ\n",
+                                            rq_status);
+                               break;
+                       }
+                       break;
+               case SLI4_FC_ASYNC_RQ_INSUFF_BUF_NEEDED:
+               case SLI4_FC_ASYNC_RQ_INSUFF_BUF_FRM_DISC:
+                       /*
+                        * since RQ buffers were not consumed, cannot return
+                        * them to chip
+                        */
+                       efc_log_debug(hw->os, "Warning: RCQE status=%#x,\n",
+                                     rq_status);
+                       fallthrough;
+               default:
+                       break;
+               }
+               return -EIO;
+       }
+
+       rqindex = efct_hw_rqpair_find(hw, rq_id);
+       if (rqindex < 0) {
+               efc_log_debug(hw->os, "Error: rq_id lookup failed for id=%#x\n",
+                             rq_id);
+               return -EIO;
+       }
+
+       rq = hw->hw_rq[hw->hw_rq_lookup[rqindex]];
+       rq->use_count++;
+
+       seq = efct_hw_rqpair_get(hw, rqindex, index);
+       if (WARN_ON(!seq))
+               return -EIO;
+
+       seq->hw = hw;
+
+       sli_fc_rqe_length(&hw->sli, cqe, &h_len, &p_len);
+       seq->header->dma.len = h_len;
+       seq->payload->dma.len = p_len;
+       seq->fcfi = sli_fc_rqe_fcfi(&hw->sli, cqe);
+       seq->hw_priv = cq->eq;
+
+       efct_unsolicited_cb(hw->os, seq);
+
+       return 0;
+}
+
+static int
+efct_hw_rqpair_put(struct efct_hw *hw, struct efc_hw_sequence *seq)
+{
+       struct sli4_queue *rq_hdr = &hw->rq[seq->header->rqindex];
+       struct sli4_queue *rq_payload = &hw->rq[seq->payload->rqindex];
+       u32 hw_rq_index = hw->hw_rq_lookup[seq->header->rqindex];
+       struct hw_rq *rq = hw->hw_rq[hw_rq_index];
+       u32 phys_hdr[2];
+       u32 phys_payload[2];
+       int qindex_hdr;
+       int qindex_payload;
+       unsigned long flags = 0;
+
+       /* Update the RQ verification lookup tables */
+       phys_hdr[0] = upper_32_bits(seq->header->dma.phys);
+       phys_hdr[1] = lower_32_bits(seq->header->dma.phys);
+       phys_payload[0] = upper_32_bits(seq->payload->dma.phys);
+       phys_payload[1] = lower_32_bits(seq->payload->dma.phys);
+
+       /* rq_hdr lock also covers payload / header->rqindex+1 queue */
+       spin_lock_irqsave(&rq_hdr->lock, flags);
+
+       /*
+        * Note: The header must be posted last for buffer pair mode because
+        *       posting on the header queue posts the payload queue as well.
+        *       We do not ring the payload queue independently in RQ pair mode.
+        */
+       qindex_payload = sli_rq_write(&hw->sli, rq_payload,
+                                     (void *)phys_payload);
+       qindex_hdr = sli_rq_write(&hw->sli, rq_hdr, (void *)phys_hdr);
+       if (qindex_hdr < 0 ||
+           qindex_payload < 0) {
+               efc_log_err(hw->os, "RQ_ID=%#x write failed\n", rq_hdr->id);
+               spin_unlock_irqrestore(&rq_hdr->lock, flags);
+               return -EIO;
+       }
+
+       /* ensure the indexes are the same */
+       WARN_ON(qindex_hdr != qindex_payload);
+
+       /* Update the lookup table */
+       if (!rq->rq_tracker[qindex_hdr]) {
+               rq->rq_tracker[qindex_hdr] = seq;
+       } else {
+               efc_log_debug(hw->os,
+                             "expected rq_tracker[%d][%d] buffer to be NULL\n",
+                             hw_rq_index, qindex_hdr);
+       }
+
+       spin_unlock_irqrestore(&rq_hdr->lock, flags);
+       return 0;
+}
+
+int
+efct_hw_rqpair_sequence_free(struct efct_hw *hw, struct efc_hw_sequence *seq)
+{
+       int rc = 0;
+
+       /*
+        * Post the data buffer first. Because in RQ pair mode, ringing the
+        * doorbell of the header ring will post the data buffer as well.
+        */
+       if (efct_hw_rqpair_put(hw, seq)) {
+               efc_log_err(hw->os, "error writing buffers\n");
+               return -EIO;
+       }
+
+       return rc;
+}
+
+int
+efct_efc_hw_sequence_free(struct efc *efc, struct efc_hw_sequence *seq)
+{
+       struct efct *efct = efc->base;
+
+       return efct_hw_rqpair_sequence_free(&efct->hw, seq);
+}