Adding upstream version 4.19.249.upstream/4.19.249upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

1 files changed, 3423 insertions, 0 deletions

diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c
new file mode 100644
index 000000000..33ff9eca2
--- /dev/null
+++ b/drivers/infiniband/hw/hfi1/sdma.c
@@ -0,0 +1,3423 @@
+/*
+ * Copyright(c) 2015 - 2018 Intel Corporation.
+ *
+ * This file is provided under a dual BSD/GPLv2 license.  When using or
+ * redistributing this file, you may do so under either license.
+ *
+ * GPL LICENSE SUMMARY
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of version 2 of the GNU General Public License as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * BSD LICENSE
+ *
+ * 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.
+ *  - Neither the name of Intel Corporation nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+#include <linux/spinlock.h>
+#include <linux/seqlock.h>
+#include <linux/netdevice.h>
+#include <linux/moduleparam.h>
+#include <linux/bitops.h>
+#include <linux/timer.h>
+#include <linux/vmalloc.h>
+#include <linux/highmem.h>
+
+#include "hfi.h"
+#include "common.h"
+#include "qp.h"
+#include "sdma.h"
+#include "iowait.h"
+#include "trace.h"
+
+/* must be a power of 2 >= 64 <= 32768 */
+#define SDMA_DESCQ_CNT 2048
+#define SDMA_DESC_INTR 64
+#define INVALID_TAIL 0xffff
+#define SDMA_PAD max_t(size_t, MAX_16B_PADDING, sizeof(u32))
+
+static uint sdma_descq_cnt = SDMA_DESCQ_CNT;
+module_param(sdma_descq_cnt, uint, S_IRUGO);
+MODULE_PARM_DESC(sdma_descq_cnt, "Number of SDMA descq entries");
+
+static uint sdma_idle_cnt = 250;
+module_param(sdma_idle_cnt, uint, S_IRUGO);
+MODULE_PARM_DESC(sdma_idle_cnt, "sdma interrupt idle delay (ns,default 250)");
+
+uint mod_num_sdma;
+module_param_named(num_sdma, mod_num_sdma, uint, S_IRUGO);
+MODULE_PARM_DESC(num_sdma, "Set max number SDMA engines to use");
+
+static uint sdma_desct_intr = SDMA_DESC_INTR;
+module_param_named(desct_intr, sdma_desct_intr, uint, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(desct_intr, "Number of SDMA descriptor before interrupt");
+
+#define SDMA_WAIT_BATCH_SIZE 20
+/* max wait time for a SDMA engine to indicate it has halted */
+#define SDMA_ERR_HALT_TIMEOUT 10 /* ms */
+/* all SDMA engine errors that cause a halt */
+
+#define SD(name) SEND_DMA_##name
+#define ALL_SDMA_ENG_HALT_ERRS \
+	(SD(ENG_ERR_STATUS_SDMA_WRONG_DW_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_GEN_MISMATCH_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_TOO_LONG_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_TAIL_OUT_OF_BOUNDS_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_FIRST_DESC_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_MEM_READ_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_HALT_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_LENGTH_MISMATCH_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_PACKET_DESC_OVERFLOW_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_HEADER_SELECT_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_HEADER_ADDRESS_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_HEADER_LENGTH_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_TIMEOUT_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_DESC_TABLE_UNC_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_ASSEMBLY_UNC_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_PACKET_TRACKING_UNC_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_HEADER_STORAGE_UNC_ERR_SMASK) \
+	| SD(ENG_ERR_STATUS_SDMA_HEADER_REQUEST_FIFO_UNC_ERR_SMASK))
+
+/* sdma_sendctrl operations */
+#define SDMA_SENDCTRL_OP_ENABLE    BIT(0)
+#define SDMA_SENDCTRL_OP_INTENABLE BIT(1)
+#define SDMA_SENDCTRL_OP_HALT      BIT(2)
+#define SDMA_SENDCTRL_OP_CLEANUP   BIT(3)
+
+/* handle long defines */
+#define SDMA_EGRESS_PACKET_OCCUPANCY_SMASK \
+SEND_EGRESS_SEND_DMA_STATUS_SDMA_EGRESS_PACKET_OCCUPANCY_SMASK
+#define SDMA_EGRESS_PACKET_OCCUPANCY_SHIFT \
+SEND_EGRESS_SEND_DMA_STATUS_SDMA_EGRESS_PACKET_OCCUPANCY_SHIFT
+
+static const char * const sdma_state_names[] = {
+	[sdma_state_s00_hw_down]                = "s00_HwDown",
+	[sdma_state_s10_hw_start_up_halt_wait]  = "s10_HwStartUpHaltWait",
+	[sdma_state_s15_hw_start_up_clean_wait] = "s15_HwStartUpCleanWait",
+	[sdma_state_s20_idle]                   = "s20_Idle",
+	[sdma_state_s30_sw_clean_up_wait]       = "s30_SwCleanUpWait",
+	[sdma_state_s40_hw_clean_up_wait]       = "s40_HwCleanUpWait",
+	[sdma_state_s50_hw_halt_wait]           = "s50_HwHaltWait",
+	[sdma_state_s60_idle_halt_wait]         = "s60_IdleHaltWait",
+	[sdma_state_s80_hw_freeze]		= "s80_HwFreeze",
+	[sdma_state_s82_freeze_sw_clean]	= "s82_FreezeSwClean",
+	[sdma_state_s99_running]                = "s99_Running",
+};
+
+#ifdef CONFIG_SDMA_VERBOSITY
+static const char * const sdma_event_names[] = {
+	[sdma_event_e00_go_hw_down]   = "e00_GoHwDown",
+	[sdma_event_e10_go_hw_start]  = "e10_GoHwStart",
+	[sdma_event_e15_hw_halt_done] = "e15_HwHaltDone",
+	[sdma_event_e25_hw_clean_up_done] = "e25_HwCleanUpDone",
+	[sdma_event_e30_go_running]   = "e30_GoRunning",
+	[sdma_event_e40_sw_cleaned]   = "e40_SwCleaned",
+	[sdma_event_e50_hw_cleaned]   = "e50_HwCleaned",
+	[sdma_event_e60_hw_halted]    = "e60_HwHalted",
+	[sdma_event_e70_go_idle]      = "e70_GoIdle",
+	[sdma_event_e80_hw_freeze]    = "e80_HwFreeze",
+	[sdma_event_e81_hw_frozen]    = "e81_HwFrozen",
+	[sdma_event_e82_hw_unfreeze]  = "e82_HwUnfreeze",
+	[sdma_event_e85_link_down]    = "e85_LinkDown",
+	[sdma_event_e90_sw_halted]    = "e90_SwHalted",
+};
+#endif
+
+static const struct sdma_set_state_action sdma_action_table[] = {
+	[sdma_state_s00_hw_down] = {
+		.go_s99_running_tofalse = 1,
+		.op_enable = 0,
+		.op_intenable = 0,
+		.op_halt = 0,
+		.op_cleanup = 0,
+	},
+	[sdma_state_s10_hw_start_up_halt_wait] = {
+		.op_enable = 0,
+		.op_intenable = 0,
+		.op_halt = 1,
+		.op_cleanup = 0,
+	},
+	[sdma_state_s15_hw_start_up_clean_wait] = {
+		.op_enable = 0,
+		.op_intenable = 1,
+		.op_halt = 0,
+		.op_cleanup = 1,
+	},
+	[sdma_state_s20_idle] = {
+		.op_enable = 0,
+		.op_intenable = 1,
+		.op_halt = 0,
+		.op_cleanup = 0,
+	},
+	[sdma_state_s30_sw_clean_up_wait] = {
+		.op_enable = 0,
+		.op_intenable = 0,
+		.op_halt = 0,
+		.op_cleanup = 0,
+	},
+	[sdma_state_s40_hw_clean_up_wait] = {
+		.op_enable = 0,
+		.op_intenable = 0,
+		.op_halt = 0,
+		.op_cleanup = 1,
+	},
+	[sdma_state_s50_hw_halt_wait] = {
+		.op_enable = 0,
+		.op_intenable = 0,
+		.op_halt = 0,
+		.op_cleanup = 0,
+	},
+	[sdma_state_s60_idle_halt_wait] = {
+		.go_s99_running_tofalse = 1,
+		.op_enable = 0,
+		.op_intenable = 0,
+		.op_halt = 1,
+		.op_cleanup = 0,
+	},
+	[sdma_state_s80_hw_freeze] = {
+		.op_enable = 0,
+		.op_intenable = 0,
+		.op_halt = 0,
+		.op_cleanup = 0,
+	},
+	[sdma_state_s82_freeze_sw_clean] = {
+		.op_enable = 0,
+		.op_intenable = 0,
+		.op_halt = 0,
+		.op_cleanup = 0,
+	},
+	[sdma_state_s99_running] = {
+		.op_enable = 1,
+		.op_intenable = 1,
+		.op_halt = 0,
+		.op_cleanup = 0,
+		.go_s99_running_totrue = 1,
+	},
+};
+
+#define SDMA_TAIL_UPDATE_THRESH 0x1F
+
+/* declare all statics here rather than keep sorting */
+static void sdma_complete(struct kref *);
+static void sdma_finalput(struct sdma_state *);
+static void sdma_get(struct sdma_state *);
+static void sdma_hw_clean_up_task(unsigned long);
+static void sdma_put(struct sdma_state *);
+static void sdma_set_state(struct sdma_engine *, enum sdma_states);
+static void sdma_start_hw_clean_up(struct sdma_engine *);
+static void sdma_sw_clean_up_task(unsigned long);
+static void sdma_sendctrl(struct sdma_engine *, unsigned);
+static void init_sdma_regs(struct sdma_engine *, u32, uint);
+static void sdma_process_event(
+	struct sdma_engine *sde,
+	enum sdma_events event);
+static void __sdma_process_event(
+	struct sdma_engine *sde,
+	enum sdma_events event);
+static void dump_sdma_state(struct sdma_engine *sde);
+static void sdma_make_progress(struct sdma_engine *sde, u64 status);
+static void sdma_desc_avail(struct sdma_engine *sde, uint avail);
+static void sdma_flush_descq(struct sdma_engine *sde);
+
+/**
+ * sdma_state_name() - return state string from enum
+ * @state: state
+ */
+static const char *sdma_state_name(enum sdma_states state)
+{
+	return sdma_state_names[state];
+}
+
+static void sdma_get(struct sdma_state *ss)
+{
+	kref_get(&ss->kref);
+}
+
+static void sdma_complete(struct kref *kref)
+{
+	struct sdma_state *ss =
+		container_of(kref, struct sdma_state, kref);
+
+	complete(&ss->comp);
+}
+
+static void sdma_put(struct sdma_state *ss)
+{
+	kref_put(&ss->kref, sdma_complete);
+}
+
+static void sdma_finalput(struct sdma_state *ss)
+{
+	sdma_put(ss);
+	wait_for_completion(&ss->comp);
+}
+
+static inline void write_sde_csr(
+	struct sdma_engine *sde,
+	u32 offset0,
+	u64 value)
+{
+	write_kctxt_csr(sde->dd, sde->this_idx, offset0, value);
+}
+
+static inline u64 read_sde_csr(
+	struct sdma_engine *sde,
+	u32 offset0)
+{
+	return read_kctxt_csr(sde->dd, sde->this_idx, offset0);
+}
+
+/*
+ * sdma_wait_for_packet_egress() - wait for the VL FIFO occupancy for
+ * sdma engine 'sde' to drop to 0.
+ */
+static void sdma_wait_for_packet_egress(struct sdma_engine *sde,
+					int pause)
+{
+	u64 off = 8 * sde->this_idx;
+	struct hfi1_devdata *dd = sde->dd;
+	int lcnt = 0;
+	u64 reg_prev;
+	u64 reg = 0;
+
+	while (1) {
+		reg_prev = reg;
+		reg = read_csr(dd, off + SEND_EGRESS_SEND_DMA_STATUS);
+
+		reg &= SDMA_EGRESS_PACKET_OCCUPANCY_SMASK;
+		reg >>= SDMA_EGRESS_PACKET_OCCUPANCY_SHIFT;
+		if (reg == 0)
+			break;
+		/* counter is reest if accupancy count changes */
+		if (reg != reg_prev)
+			lcnt = 0;
+		if (lcnt++ > 500) {
+			/* timed out - bounce the link */
+			dd_dev_err(dd, "%s: engine %u timeout waiting for packets to egress, remaining count %u, bouncing link\n",
+				   __func__, sde->this_idx, (u32)reg);
+			queue_work(dd->pport->link_wq,
+				   &dd->pport->link_bounce_work);
+			break;
+		}
+		udelay(1);
+	}
+}
+
+/*
+ * sdma_wait() - wait for packet egress to complete for all SDMA engines,
+ * and pause for credit return.
+ */
+void sdma_wait(struct hfi1_devdata *dd)
+{
+	int i;
+
+	for (i = 0; i < dd->num_sdma; i++) {
+		struct sdma_engine *sde = &dd->per_sdma[i];
+
+		sdma_wait_for_packet_egress(sde, 0);
+	}
+}
+
+static inline void sdma_set_desc_cnt(struct sdma_engine *sde, unsigned cnt)
+{
+	u64 reg;
+
+	if (!(sde->dd->flags & HFI1_HAS_SDMA_TIMEOUT))
+		return;
+	reg = cnt;
+	reg &= SD(DESC_CNT_CNT_MASK);
+	reg <<= SD(DESC_CNT_CNT_SHIFT);
+	write_sde_csr(sde, SD(DESC_CNT), reg);
+}
+
+static inline void complete_tx(struct sdma_engine *sde,
+			       struct sdma_txreq *tx,
+			       int res)
+{
+	/* protect against complete modifying */
+	struct iowait *wait = tx->wait;
+	callback_t complete = tx->complete;
+
+#ifdef CONFIG_HFI1_DEBUG_SDMA_ORDER
+	trace_hfi1_sdma_out_sn(sde, tx->sn);
+	if (WARN_ON_ONCE(sde->head_sn != tx->sn))
+		dd_dev_err(sde->dd, "expected %llu got %llu\n",
+			   sde->head_sn, tx->sn);
+	sde->head_sn++;
+#endif
+	__sdma_txclean(sde->dd, tx);
+	if (complete)
+		(*complete)(tx, res);
+	if (wait && iowait_sdma_dec(wait))
+		iowait_drain_wakeup(wait);
+}
+
+/*
+ * Complete all the sdma requests with a SDMA_TXREQ_S_ABORTED status
+ *
+ * Depending on timing there can be txreqs in two places:
+ * - in the descq ring
+ * - in the flush list
+ *
+ * To avoid ordering issues the descq ring needs to be flushed
+ * first followed by the flush list.
+ *
+ * This routine is called from two places
+ * - From a work queue item
+ * - Directly from the state machine just before setting the
+ *   state to running
+ *
+ * Must be called with head_lock held
+ *
+ */
+static void sdma_flush(struct sdma_engine *sde)
+{
+	struct sdma_txreq *txp, *txp_next;
+	LIST_HEAD(flushlist);
+	unsigned long flags;
+
+	/* flush from head to tail */
+	sdma_flush_descq(sde);
+	spin_lock_irqsave(&sde->flushlist_lock, flags);
+	/* copy flush list */
+	list_splice_init(&sde->flushlist, &flushlist);
+	spin_unlock_irqrestore(&sde->flushlist_lock, flags);
+	/* flush from flush list */
+	list_for_each_entry_safe(txp, txp_next, &flushlist, list)
+		complete_tx(sde, txp, SDMA_TXREQ_S_ABORTED);
+}
+
+/*
+ * Fields a work request for flushing the descq ring
+ * and the flush list
+ *
+ * If the engine has been brought to running during
+ * the scheduling delay, the flush is ignored, assuming
+ * that the process of bringing the engine to running
+ * would have done this flush prior to going to running.
+ *
+ */
+static void sdma_field_flush(struct work_struct *work)
+{
+	unsigned long flags;
+	struct sdma_engine *sde =
+		container_of(work, struct sdma_engine, flush_worker);
+
+	write_seqlock_irqsave(&sde->head_lock, flags);
+	if (!__sdma_running(sde))
+		sdma_flush(sde);
+	write_sequnlock_irqrestore(&sde->head_lock, flags);
+}
+
+static void sdma_err_halt_wait(struct work_struct *work)
+{
+	struct sdma_engine *sde = container_of(work, struct sdma_engine,
+						err_halt_worker);
+	u64 statuscsr;
+	unsigned long timeout;
+
+	timeout = jiffies + msecs_to_jiffies(SDMA_ERR_HALT_TIMEOUT);
+	while (1) {
+		statuscsr = read_sde_csr(sde, SD(STATUS));
+		statuscsr &= SD(STATUS_ENG_HALTED_SMASK);
+		if (statuscsr)
+			break;
+		if (time_after(jiffies, timeout)) {
+			dd_dev_err(sde->dd,
+				   "SDMA engine %d - timeout waiting for engine to halt\n",
+				   sde->this_idx);
+			/*
+			 * Continue anyway.  This could happen if there was
+			 * an uncorrectable error in the wrong spot.
+			 */
+			break;
+		}
+		usleep_range(80, 120);
+	}
+
+	sdma_process_event(sde, sdma_event_e15_hw_halt_done);
+}
+
+static void sdma_err_progress_check_schedule(struct sdma_engine *sde)
+{
+	if (!is_bx(sde->dd) && HFI1_CAP_IS_KSET(SDMA_AHG)) {
+		unsigned index;
+		struct hfi1_devdata *dd = sde->dd;
+
+		for (index = 0; index < dd->num_sdma; index++) {
+			struct sdma_engine *curr_sdma = &dd->per_sdma[index];
+
+			if (curr_sdma != sde)
+				curr_sdma->progress_check_head =
+							curr_sdma->descq_head;
+		}
+		dd_dev_err(sde->dd,
+			   "SDMA engine %d - check scheduled\n",
+				sde->this_idx);
+		mod_timer(&sde->err_progress_check_timer, jiffies + 10);
+	}
+}
+
+static void sdma_err_progress_check(struct timer_list *t)
+{
+	unsigned index;
+	struct sdma_engine *sde = from_timer(sde, t, err_progress_check_timer);
+
+	dd_dev_err(sde->dd, "SDE progress check event\n");
+	for (index = 0; index < sde->dd->num_sdma; index++) {
+		struct sdma_engine *curr_sde = &sde->dd->per_sdma[index];
+		unsigned long flags;
+
+		/* check progress on each engine except the current one */
+		if (curr_sde == sde)
+			continue;
+		/*
+		 * We must lock interrupts when acquiring sde->lock,
+		 * to avoid a deadlock if interrupt triggers and spins on
+		 * the same lock on same CPU
+		 */
+		spin_lock_irqsave(&curr_sde->tail_lock, flags);
+		write_seqlock(&curr_sde->head_lock);
+
+		/* skip non-running queues */
+		if (curr_sde->state.current_state != sdma_state_s99_running) {
+			write_sequnlock(&curr_sde->head_lock);
+			spin_unlock_irqrestore(&curr_sde->tail_lock, flags);
+			continue;
+		}
+
+		if ((curr_sde->descq_head != curr_sde->descq_tail) &&
+		    (curr_sde->descq_head ==
+				curr_sde->progress_check_head))
+			__sdma_process_event(curr_sde,
+					     sdma_event_e90_sw_halted);
+		write_sequnlock(&curr_sde->head_lock);
+		spin_unlock_irqrestore(&curr_sde->tail_lock, flags);
+	}
+	schedule_work(&sde->err_halt_worker);
+}
+
+static void sdma_hw_clean_up_task(unsigned long opaque)
+{
+	struct sdma_engine *sde = (struct sdma_engine *)opaque;
+	u64 statuscsr;
+
+	while (1) {
+#ifdef CONFIG_SDMA_VERBOSITY
+		dd_dev_err(sde->dd, "CONFIG SDMA(%u) %s:%d %s()\n",
+			   sde->this_idx, slashstrip(__FILE__), __LINE__,
+			__func__);
+#endif
+		statuscsr = read_sde_csr(sde, SD(STATUS));
+		statuscsr &= SD(STATUS_ENG_CLEANED_UP_SMASK);
+		if (statuscsr)
+			break;
+		udelay(10);
+	}
+
+	sdma_process_event(sde, sdma_event_e25_hw_clean_up_done);
+}
+
+static inline struct sdma_txreq *get_txhead(struct sdma_engine *sde)
+{
+	return sde->tx_ring[sde->tx_head & sde->sdma_mask];
+}
+
+/*
+ * flush ring for recovery
+ */
+static void sdma_flush_descq(struct sdma_engine *sde)
+{
+	u16 head, tail;
+	int progress = 0;
+	struct sdma_txreq *txp = get_txhead(sde);
+
+	/* The reason for some of the complexity of this code is that
+	 * not all descriptors have corresponding txps.  So, we have to
+	 * be able to skip over descs until we wander into the range of
+	 * the next txp on the list.
+	 */
+	head = sde->descq_head & sde->sdma_mask;
+	tail = sde->descq_tail & sde->sdma_mask;
+	while (head != tail) {
+		/* advance head, wrap if needed */
+		head = ++sde->descq_head & sde->sdma_mask;
+		/* if now past this txp's descs, do the callback */
+		if (txp && txp->next_descq_idx == head) {
+			/* remove from list */
+			sde->tx_ring[sde->tx_head++ & sde->sdma_mask] = NULL;
+			complete_tx(sde, txp, SDMA_TXREQ_S_ABORTED);
+			trace_hfi1_sdma_progress(sde, head, tail, txp);
+			txp = get_txhead(sde);
+		}
+		progress++;
+	}
+	if (progress)
+		sdma_desc_avail(sde, sdma_descq_freecnt(sde));
+}
+
+static void sdma_sw_clean_up_task(unsigned long opaque)
+{
+	struct sdma_engine *sde = (struct sdma_engine *)opaque;
+	unsigned long flags;
+
+	spin_lock_irqsave(&sde->tail_lock, flags);
+	write_seqlock(&sde->head_lock);
+
+	/*
+	 * At this point, the following should always be true:
+	 * - We are halted, so no more descriptors are getting retired.
+	 * - We are not running, so no one is submitting new work.
+	 * - Only we can send the e40_sw_cleaned, so we can't start
+	 *   running again until we say so.  So, the active list and
+	 *   descq are ours to play with.
+	 */
+
+	/*
+	 * In the error clean up sequence, software clean must be called
+	 * before the hardware clean so we can use the hardware head in
+	 * the progress routine.  A hardware clean or SPC unfreeze will
+	 * reset the hardware head.
+	 *
+	 * Process all retired requests. The progress routine will use the
+	 * latest physical hardware head - we are not running so speed does
+	 * not matter.
+	 */
+	sdma_make_progress(sde, 0);
+
+	sdma_flush(sde);
+
+	/*
+	 * Reset our notion of head and tail.
+	 * Note that the HW registers have been reset via an earlier
+	 * clean up.
+	 */
+	sde->descq_tail = 0;
+	sde->descq_head = 0;
+	sde->desc_avail = sdma_descq_freecnt(sde);
+	*sde->head_dma = 0;
+
+	__sdma_process_event(sde, sdma_event_e40_sw_cleaned);
+
+	write_sequnlock(&sde->head_lock);
+	spin_unlock_irqrestore(&sde->tail_lock, flags);
+}
+
+static void sdma_sw_tear_down(struct sdma_engine *sde)
+{
+	struct sdma_state *ss = &sde->state;
+
+	/* Releasing this reference means the state machine has stopped. */
+	sdma_put(ss);
+
+	/* stop waiting for all unfreeze events to complete */
+	atomic_set(&sde->dd->sdma_unfreeze_count, -1);
+	wake_up_interruptible(&sde->dd->sdma_unfreeze_wq);
+}
+
+static void sdma_start_hw_clean_up(struct sdma_engine *sde)
+{
+	tasklet_hi_schedule(&sde->sdma_hw_clean_up_task);
+}
+
+static void sdma_set_state(struct sdma_engine *sde,
+			   enum sdma_states next_state)
+{
+	struct sdma_state *ss = &sde->state;
+	const struct sdma_set_state_action *action = sdma_action_table;
+	unsigned op = 0;
+
+	trace_hfi1_sdma_state(
+		sde,
+		sdma_state_names[ss->current_state],
+		sdma_state_names[next_state]);
+
+	/* debugging bookkeeping */
+	ss->previous_state = ss->current_state;
+	ss->previous_op = ss->current_op;
+	ss->current_state = next_state;
+
+	if (ss->previous_state != sdma_state_s99_running &&
+	    next_state == sdma_state_s99_running)
+		sdma_flush(sde);
+
+	if (action[next_state].op_enable)
+		op |= SDMA_SENDCTRL_OP_ENABLE;
+
+	if (action[next_state].op_intenable)
+		op |= SDMA_SENDCTRL_OP_INTENABLE;
+
+	if (action[next_state].op_halt)
+		op |= SDMA_SENDCTRL_OP_HALT;
+
+	if (action[next_state].op_cleanup)
+		op |= SDMA_SENDCTRL_OP_CLEANUP;
+
+	if (action[next_state].go_s99_running_tofalse)
+		ss->go_s99_running = 0;
+
+	if (action[next_state].go_s99_running_totrue)
+		ss->go_s99_running = 1;
+
+	ss->current_op = op;
+	sdma_sendctrl(sde, ss->current_op);
+}
+
+/**
+ * sdma_get_descq_cnt() - called when device probed
+ *
+ * Return a validated descq count.
+ *
+ * This is currently only used in the verbs initialization to build the tx
+ * list.
+ *
+ * This will probably be deleted in favor of a more scalable approach to
+ * alloc tx's.
+ *
+ */
+u16 sdma_get_descq_cnt(void)
+{
+	u16 count = sdma_descq_cnt;
+
+	if (!count)
+		return SDMA_DESCQ_CNT;
+	/* count must be a power of 2 greater than 64 and less than
+	 * 32768.   Otherwise return default.
+	 */
+	if (!is_power_of_2(count))
+		return SDMA_DESCQ_CNT;
+	if (count < 64 || count > 32768)
+		return SDMA_DESCQ_CNT;
+	return count;
+}
+
+/**
+ * sdma_engine_get_vl() - return vl for a given sdma engine
+ * @sde: sdma engine
+ *
+ * This function returns the vl mapped to a given engine, or an error if
+ * the mapping can't be found. The mapping fields are protected by RCU.
+ */
+int sdma_engine_get_vl(struct sdma_engine *sde)
+{
+	struct hfi1_devdata *dd = sde->dd;
+	struct sdma_vl_map *m;
+	u8 vl;
+
+	if (sde->this_idx >= TXE_NUM_SDMA_ENGINES)
+		return -EINVAL;
+
+	rcu_read_lock();
+	m = rcu_dereference(dd->sdma_map);
+	if (unlikely(!m)) {
+		rcu_read_unlock();
+		return -EINVAL;
+	}
+	vl = m->engine_to_vl[sde->this_idx];
+	rcu_read_unlock();
+
+	return vl;
+}
+
+/**
+ * sdma_select_engine_vl() - select sdma engine
+ * @dd: devdata
+ * @selector: a spreading factor
+ * @vl: this vl
+ *
+ *
+ * This function returns an engine based on the selector and a vl.  The
+ * mapping fields are protected by RCU.
+ */
+struct sdma_engine *sdma_select_engine_vl(
+	struct hfi1_devdata *dd,
+	u32 selector,
+	u8 vl)
+{
+	struct sdma_vl_map *m;
+	struct sdma_map_elem *e;
+	struct sdma_engine *rval;
+
+	/* NOTE This should only happen if SC->VL changed after the initial
+	 *      checks on the QP/AH
+	 *      Default will return engine 0 below
+	 */
+	if (vl >= num_vls) {
+		rval = NULL;
+		goto done;
+	}
+
+	rcu_read_lock();
+	m = rcu_dereference(dd->sdma_map);
+	if (unlikely(!m)) {
+		rcu_read_unlock();
+		return &dd->per_sdma[0];
+	}
+	e = m->map[vl & m->mask];
+	rval = e->sde[selector & e->mask];
+	rcu_read_unlock();
+
+done:
+	rval =  !rval ? &dd->per_sdma[0] : rval;
+	trace_hfi1_sdma_engine_select(dd, selector, vl, rval->this_idx);
+	return rval;
+}
+
+/**
+ * sdma_select_engine_sc() - select sdma engine
+ * @dd: devdata
+ * @selector: a spreading factor
+ * @sc5: the 5 bit sc
+ *
+ *
+ * This function returns an engine based on the selector and an sc.
+ */
+struct sdma_engine *sdma_select_engine_sc(
+	struct hfi1_devdata *dd,
+	u32 selector,
+	u8 sc5)
+{
+	u8 vl = sc_to_vlt(dd, sc5);
+
+	return sdma_select_engine_vl(dd, selector, vl);
+}
+
+struct sdma_rht_map_elem {
+	u32 mask;
+	u8 ctr;
+	struct sdma_engine *sde[0];
+};
+
+struct sdma_rht_node {
+	unsigned long cpu_id;
+	struct sdma_rht_map_elem *map[HFI1_MAX_VLS_SUPPORTED];
+	struct rhash_head node;
+};
+
+#define NR_CPUS_HINT 192
+
+static const struct rhashtable_params sdma_rht_params = {
+	.nelem_hint = NR_CPUS_HINT,
+	.head_offset = offsetof(struct sdma_rht_node, node),
+	.key_offset = offsetof(struct sdma_rht_node, cpu_id),
+	.key_len = FIELD_SIZEOF(struct sdma_rht_node, cpu_id),
+	.max_size = NR_CPUS,
+	.min_size = 8,
+	.automatic_shrinking = true,
+};
+
+/*
+ * sdma_select_user_engine() - select sdma engine based on user setup
+ * @dd: devdata
+ * @selector: a spreading factor
+ * @vl: this vl
+ *
+ * This function returns an sdma engine for a user sdma request.
+ * User defined sdma engine affinity setting is honored when applicable,
+ * otherwise system default sdma engine mapping is used. To ensure correct
+ * ordering, the mapping from <selector, vl> to sde must remain unchanged.
+ */
+struct sdma_engine *sdma_select_user_engine(struct hfi1_devdata *dd,
+					    u32 selector, u8 vl)
+{
+	struct sdma_rht_node *rht_node;
+	struct sdma_engine *sde = NULL;
+	const struct cpumask *current_mask = &current->cpus_allowed;
+	unsigned long cpu_id;
+
+	/*
+	 * To ensure that always the same sdma engine(s) will be
+	 * selected make sure the process is pinned to this CPU only.
+	 */
+	if (cpumask_weight(current_mask) != 1)
+		goto out;
+
+	cpu_id = smp_processor_id();
+	rcu_read_lock();
+	rht_node = rhashtable_lookup_fast(dd->sdma_rht, &cpu_id,
+					  sdma_rht_params);
+
+	if (rht_node && rht_node->map[vl]) {
+		struct sdma_rht_map_elem *map = rht_node->map[vl];
+
+		sde = map->sde[selector & map->mask];
+	}
+	rcu_read_unlock();
+
+	if (sde)
+		return sde;
+
+out:
+	return sdma_select_engine_vl(dd, selector, vl);
+}
+
+static void sdma_populate_sde_map(struct sdma_rht_map_elem *map)
+{
+	int i;
+
+	for (i = 0; i < roundup_pow_of_two(map->ctr ? : 1) - map->ctr; i++)
+		map->sde[map->ctr + i] = map->sde[i];
+}
+
+static void sdma_cleanup_sde_map(struct sdma_rht_map_elem *map,
+				 struct sdma_engine *sde)
+{
+	unsigned int i, pow;
+
+	/* only need to check the first ctr entries for a match */
+	for (i = 0; i < map->ctr; i++) {
+		if (map->sde[i] == sde) {
+			memmove(&map->sde[i], &map->sde[i + 1],
+				(map->ctr - i - 1) * sizeof(map->sde[0]));
+			map->ctr--;
+			pow = roundup_pow_of_two(map->ctr ? : 1);
+			map->mask = pow - 1;
+			sdma_populate_sde_map(map);
+			break;
+		}
+	}
+}
+
+/*
+ * Prevents concurrent reads and writes of the sdma engine cpu_mask
+ */
+static DEFINE_MUTEX(process_to_sde_mutex);
+
+ssize_t sdma_set_cpu_to_sde_map(struct sdma_engine *sde, const char *buf,
+				size_t count)
+{
+	struct hfi1_devdata *dd = sde->dd;
+	cpumask_var_t mask, new_mask;
+	unsigned long cpu;
+	int ret, vl, sz;
+	struct sdma_rht_node *rht_node;
+
+	vl = sdma_engine_get_vl(sde);
+	if (unlikely(vl < 0 || vl >= ARRAY_SIZE(rht_node->map)))
+		return -EINVAL;
+
+	ret = zalloc_cpumask_var(&mask, GFP_KERNEL);
+	if (!ret)
+		return -ENOMEM;
+
+	ret = zalloc_cpumask_var(&new_mask, GFP_KERNEL);
+	if (!ret) {
+		free_cpumask_var(mask);
+		return -ENOMEM;
+	}
+	ret = cpulist_parse(buf, mask);
+	if (ret)
+		goto out_free;
+
+	if (!cpumask_subset(mask, cpu_online_mask)) {
+		dd_dev_warn(sde->dd, "Invalid CPU mask\n");
+		ret = -EINVAL;
+		goto out_free;
+	}
+
+	sz = sizeof(struct sdma_rht_map_elem) +
+			(TXE_NUM_SDMA_ENGINES * sizeof(struct sdma_engine *));
+
+	mutex_lock(&process_to_sde_mutex);
+
+	for_each_cpu(cpu, mask) {
+		/* Check if we have this already mapped */
+		if (cpumask_test_cpu(cpu, &sde->cpu_mask)) {
+			cpumask_set_cpu(cpu, new_mask);
+			continue;
+		}
+
+		rht_node = rhashtable_lookup_fast(dd->sdma_rht, &cpu,
+						  sdma_rht_params);
+		if (!rht_node) {
+			rht_node = kzalloc(sizeof(*rht_node), GFP_KERNEL);
+			if (!rht_node) {
+				ret = -ENOMEM;
+				goto out;
+			}
+
+			rht_node->map[vl] = kzalloc(sz, GFP_KERNEL);
+			if (!rht_node->map[vl]) {
+				kfree(rht_node);
+				ret = -ENOMEM;
+				goto out;
+			}
+			rht_node->cpu_id = cpu;
+			rht_node->map[vl]->mask = 0;
+			rht_node->map[vl]->ctr = 1;
+			rht_node->map[vl]->sde[0] = sde;
+
+			ret = rhashtable_insert_fast(dd->sdma_rht,
+						     &rht_node->node,
+						     sdma_rht_params);
+			if (ret) {
+				kfree(rht_node->map[vl]);
+				kfree(rht_node);
+				dd_dev_err(sde->dd, "Failed to set process to sde affinity for cpu %lu\n",
+					   cpu);
+				goto out;
+			}
+
+		} else {
+			int ctr, pow;
+
+			/* Add new user mappings */
+			if (!rht_node->map[vl])
+				rht_node->map[vl] = kzalloc(sz, GFP_KERNEL);
+
+			if (!rht_node->map[vl]) {
+				ret = -ENOMEM;
+				goto out;
+			}
+
+			rht_node->map[vl]->ctr++;
+			ctr = rht_node->map[vl]->ctr;
+			rht_node->map[vl]->sde[ctr - 1] = sde;
+			pow = roundup_pow_of_two(ctr);
+			rht_node->map[vl]->mask = pow - 1;
+
+			/* Populate the sde map table */
+			sdma_populate_sde_map(rht_node->map[vl]);
+		}
+		cpumask_set_cpu(cpu, new_mask);
+	}
+
+	/* Clean up old mappings */
+	for_each_cpu(cpu, cpu_online_mask) {
+		struct sdma_rht_node *rht_node;
+
+		/* Don't cleanup sdes that are set in the new mask */
+		if (cpumask_test_cpu(cpu, mask))
+			continue;
+
+		rht_node = rhashtable_lookup_fast(dd->sdma_rht, &cpu,
+						  sdma_rht_params);
+		if (rht_node) {
+			bool empty = true;
+			int i;
+
+			/* Remove mappings for old sde */
+			for (i = 0; i < HFI1_MAX_VLS_SUPPORTED; i++)
+				if (rht_node->map[i])
+					sdma_cleanup_sde_map(rht_node->map[i],
+							     sde);
+
+			/* Free empty hash table entries */
+			for (i = 0; i < HFI1_MAX_VLS_SUPPORTED; i++) {
+				if (!rht_node->map[i])
+					continue;
+
+				if (rht_node->map[i]->ctr) {
+					empty = false;
+					break;
+				}
+			}
+
+			if (empty) {
+				ret = rhashtable_remove_fast(dd->sdma_rht,
+							     &rht_node->node,
+							     sdma_rht_params);
+				WARN_ON(ret);
+
+				for (i = 0; i < HFI1_MAX_VLS_SUPPORTED; i++)
+					kfree(rht_node->map[i]);
+
+				kfree(rht_node);
+			}
+		}
+	}
+
+	cpumask_copy(&sde->cpu_mask, new_mask);
+out:
+	mutex_unlock(&process_to_sde_mutex);
+out_free:
+	free_cpumask_var(mask);
+	free_cpumask_var(new_mask);
+	return ret ? : strnlen(buf, PAGE_SIZE);
+}
+
+ssize_t sdma_get_cpu_to_sde_map(struct sdma_engine *sde, char *buf)
+{
+	mutex_lock(&process_to_sde_mutex);
+	if (cpumask_empty(&sde->cpu_mask))
+		snprintf(buf, PAGE_SIZE, "%s\n", "empty");
+	else
+		cpumap_print_to_pagebuf(true, buf, &sde->cpu_mask);
+	mutex_unlock(&process_to_sde_mutex);
+	return strnlen(buf, PAGE_SIZE);
+}
+
+static void sdma_rht_free(void *ptr, void *arg)
+{
+	struct sdma_rht_node *rht_node = ptr;
+	int i;
+
+	for (i = 0; i < HFI1_MAX_VLS_SUPPORTED; i++)
+		kfree(rht_node->map[i]);
+
+	kfree(rht_node);
+}
+
+/**
+ * sdma_seqfile_dump_cpu_list() - debugfs dump the cpu to sdma mappings
+ * @s: seq file
+ * @dd: hfi1_devdata
+ * @cpuid: cpu id
+ *
+ * This routine dumps the process to sde mappings per cpu
+ */
+void sdma_seqfile_dump_cpu_list(struct seq_file *s,
+				struct hfi1_devdata *dd,
+				unsigned long cpuid)
+{
+	struct sdma_rht_node *rht_node;
+	int i, j;
+
+	rht_node = rhashtable_lookup_fast(dd->sdma_rht, &cpuid,
+					  sdma_rht_params);
+	if (!rht_node)
+		return;
+
+	seq_printf(s, "cpu%3lu: ", cpuid);
+	for (i = 0; i < HFI1_MAX_VLS_SUPPORTED; i++) {
+		if (!rht_node->map[i] || !rht_node->map[i]->ctr)
+			continue;
+
+		seq_printf(s, " vl%d: [", i);
+
+		for (j = 0; j < rht_node->map[i]->ctr; j++) {
+			if (!rht_node->map[i]->sde[j])
+				continue;
+
+			if (j > 0)
+				seq_puts(s, ",");
+
+			seq_printf(s, " sdma%2d",
+				   rht_node->map[i]->sde[j]->this_idx);
+		}
+		seq_puts(s, " ]");
+	}
+
+	seq_puts(s, "\n");
+}
+
+/*
+ * Free the indicated map struct
+ */
+static void sdma_map_free(struct sdma_vl_map *m)
+{
+	int i;
+
+	for (i = 0; m && i < m->actual_vls; i++)
+		kfree(m->map[i]);
+	kfree(m);
+}
+
+/*
+ * Handle RCU callback
+ */
+static void sdma_map_rcu_callback(struct rcu_head *list)
+{
+	struct sdma_vl_map *m = container_of(list, struct sdma_vl_map, list);
+
+	sdma_map_free(m);
+}
+
+/**
+ * sdma_map_init - called when # vls change
+ * @dd: hfi1_devdata
+ * @port: port number
+ * @num_vls: number of vls
+ * @vl_engines: per vl engine mapping (optional)
+ *
+ * This routine changes the mapping based on the number of vls.
+ *
+ * vl_engines is used to specify a non-uniform vl/engine loading. NULL
+ * implies auto computing the loading and giving each VLs a uniform
+ * distribution of engines per VL.
+ *
+ * The auto algorithm computes the sde_per_vl and the number of extra
+ * engines.  Any extra engines are added from the last VL on down.
+ *
+ * rcu locking is used here to control access to the mapping fields.
+ *
+ * If either the num_vls or num_sdma are non-power of 2, the array sizes
+ * in the struct sdma_vl_map and the struct sdma_map_elem are rounded
+ * up to the next highest power of 2 and the first entry is reused
+ * in a round robin fashion.
+ *
+ * If an error occurs the map change is not done and the mapping is
+ * not changed.
+ *
+ */
+int sdma_map_init(struct hfi1_devdata *dd, u8 port, u8 num_vls, u8 *vl_engines)
+{
+	int i, j;
+	int extra, sde_per_vl;
+	int engine = 0;
+	u8 lvl_engines[OPA_MAX_VLS];
+	struct sdma_vl_map *oldmap, *newmap;
+
+	if (!(dd->flags & HFI1_HAS_SEND_DMA))
+		return 0;
+
+	if (!vl_engines) {
+		/* truncate divide */
+		sde_per_vl = dd->num_sdma / num_vls;
+		/* extras */
+		extra = dd->num_sdma % num_vls;
+		vl_engines = lvl_engines;
+		/* add extras from last vl down */
+		for (i = num_vls - 1; i >= 0; i--, extra--)
+			vl_engines[i] = sde_per_vl + (extra > 0 ? 1 : 0);
+	}
+	/* build new map */
+	newmap = kzalloc(
+		sizeof(struct sdma_vl_map) +
+			roundup_pow_of_two(num_vls) *
+			sizeof(struct sdma_map_elem *),
+		GFP_KERNEL);
+	if (!newmap)
+		goto bail;
+	newmap->actual_vls = num_vls;
+	newmap->vls = roundup_pow_of_two(num_vls);
+	newmap->mask = (1 << ilog2(newmap->vls)) - 1;
+	/* initialize back-map */
+	for (i = 0; i < TXE_NUM_SDMA_ENGINES; i++)
+		newmap->engine_to_vl[i] = -1;
+	for (i = 0; i < newmap->vls; i++) {
+		/* save for wrap around */
+		int first_engine = engine;
+
+		if (i < newmap->actual_vls) {
+			int sz = roundup_pow_of_two(vl_engines[i]);
+
+			/* only allocate once */
+			newmap->map[i] = kzalloc(
+				sizeof(struct sdma_map_elem) +
+					sz * sizeof(struct sdma_engine *),
+				GFP_KERNEL);
+			if (!newmap->map[i])
+				goto bail;
+			newmap->map[i]->mask = (1 << ilog2(sz)) - 1;
+			/* assign engines */
+			for (j = 0; j < sz; j++) {
+				newmap->map[i]->sde[j] =
+					&dd->per_sdma[engine];
+				if (++engine >= first_engine + vl_engines[i])
+					/* wrap back to first engine */
+					engine = first_engine;
+			}
+			/* assign back-map */
+			for (j = 0; j < vl_engines[i]; j++)
+				newmap->engine_to_vl[first_engine + j] = i;
+		} else {
+			/* just re-use entry without allocating */
+			newmap->map[i] = newmap->map[i % num_vls];
+		}
+		engine = first_engine + vl_engines[i];
+	}
+	/* newmap in hand, save old map */
+	spin_lock_irq(&dd->sde_map_lock);
+	oldmap = rcu_dereference_protected(dd->sdma_map,
+					   lockdep_is_held(&dd->sde_map_lock));
+
+	/* publish newmap */
+	rcu_assign_pointer(dd->sdma_map, newmap);
+
+	spin_unlock_irq(&dd->sde_map_lock);
+	/* success, free any old map after grace period */
+	if (oldmap)
+		call_rcu(&oldmap->list, sdma_map_rcu_callback);
+	return 0;
+bail:
+	/* free any partial allocation */
+	sdma_map_free(newmap);
+	return -ENOMEM;
+}
+
+/**
+ * sdma_clean()  Clean up allocated memory
+ * @dd:          struct hfi1_devdata
+ * @num_engines: num sdma engines
+ *
+ * This routine can be called regardless of the success of
+ * sdma_init()
+ */
+void sdma_clean(struct hfi1_devdata *dd, size_t num_engines)
+{
+	size_t i;
+	struct sdma_engine *sde;
+
+	if (dd->sdma_pad_dma) {
+		dma_free_coherent(&dd->pcidev->dev, SDMA_PAD,
+				  (void *)dd->sdma_pad_dma,
+				  dd->sdma_pad_phys);
+		dd->sdma_pad_dma = NULL;
+		dd->sdma_pad_phys = 0;
+	}
+	if (dd->sdma_heads_dma) {
+		dma_free_coherent(&dd->pcidev->dev, dd->sdma_heads_size,
+				  (void *)dd->sdma_heads_dma,
+				  dd->sdma_heads_phys);
+		dd->sdma_heads_dma = NULL;
+		dd->sdma_heads_phys = 0;
+	}
+	for (i = 0; dd->per_sdma && i < num_engines; ++i) {
+		sde = &dd->per_sdma[i];
+
+		sde->head_dma = NULL;
+		sde->head_phys = 0;
+
+		if (sde->descq) {
+			dma_free_coherent(
+				&dd->pcidev->dev,
+				sde->descq_cnt * sizeof(u64[2]),
+				sde->descq,
+				sde->descq_phys
+			);
+			sde->descq = NULL;
+			sde->descq_phys = 0;
+		}
+		kvfree(sde->tx_ring);
+		sde->tx_ring = NULL;
+	}
+	if (rcu_access_pointer(dd->sdma_map)) {
+		spin_lock_irq(&dd->sde_map_lock);
+		sdma_map_free(rcu_access_pointer(dd->sdma_map));
+		RCU_INIT_POINTER(dd->sdma_map, NULL);
+		spin_unlock_irq(&dd->sde_map_lock);
+		synchronize_rcu();
+	}
+	kfree(dd->per_sdma);
+	dd->per_sdma = NULL;
+
+	if (dd->sdma_rht) {
+		rhashtable_free_and_destroy(dd->sdma_rht, sdma_rht_free, NULL);
+		kfree(dd->sdma_rht);
+		dd->sdma_rht = NULL;
+	}
+}
+
+/**
+ * sdma_init() - called when device probed
+ * @dd: hfi1_devdata
+ * @port: port number (currently only zero)
+ *
+ * Initializes each sde and its csrs.
+ * Interrupts are not required to be enabled.
+ *
+ * Returns:
+ * 0 - success, -errno on failure
+ */
+int sdma_init(struct hfi1_devdata *dd, u8 port)
+{
+	unsigned this_idx;
+	struct sdma_engine *sde;
+	struct rhashtable *tmp_sdma_rht;
+	u16 descq_cnt;
+	void *curr_head;
+	struct hfi1_pportdata *ppd = dd->pport + port;
+	u32 per_sdma_credits;
+	uint idle_cnt = sdma_idle_cnt;
+	size_t num_engines = chip_sdma_engines(dd);
+	int ret = -ENOMEM;
+
+	if (!HFI1_CAP_IS_KSET(SDMA)) {
+		HFI1_CAP_CLEAR(SDMA_AHG);
+		return 0;
+	}
+	if (mod_num_sdma &&
+	    /* can't exceed chip support */
+	    mod_num_sdma <= chip_sdma_engines(dd) &&
+	    /* count must be >= vls */
+	    mod_num_sdma >= num_vls)
+		num_engines = mod_num_sdma;
+
+	dd_dev_info(dd, "SDMA mod_num_sdma: %u\n", mod_num_sdma);
+	dd_dev_info(dd, "SDMA chip_sdma_engines: %u\n", chip_sdma_engines(dd));
+	dd_dev_info(dd, "SDMA chip_sdma_mem_size: %u\n",
+		    chip_sdma_mem_size(dd));
+
+	per_sdma_credits =
+		chip_sdma_mem_size(dd) / (num_engines * SDMA_BLOCK_SIZE);
+
+	/* set up freeze waitqueue */
+	init_waitqueue_head(&dd->sdma_unfreeze_wq);
+	atomic_set(&dd->sdma_unfreeze_count, 0);
+
+	descq_cnt = sdma_get_descq_cnt();
+	dd_dev_info(dd, "SDMA engines %zu descq_cnt %u\n",
+		    num_engines, descq_cnt);
+
+	/* alloc memory for array of send engines */
+	dd->per_sdma = kcalloc_node(num_engines, sizeof(*dd->per_sdma),
+				    GFP_KERNEL, dd->node);
+	if (!dd->per_sdma)
+		return ret;
+
+	idle_cnt = ns_to_cclock(dd, idle_cnt);
+	if (idle_cnt)
+		dd->default_desc1 =
+			SDMA_DESC1_HEAD_TO_HOST_FLAG;
+	else
+		dd->default_desc1 =
+			SDMA_DESC1_INT_REQ_FLAG;
+
+	if (!sdma_desct_intr)
+		sdma_desct_intr = SDMA_DESC_INTR;
+
+	/* Allocate memory for SendDMA descriptor FIFOs */
+	for (this_idx = 0; this_idx < num_engines; ++this_idx) {
+		sde = &dd->per_sdma[this_idx];
+		sde->dd = dd;
+		sde->ppd = ppd;
+		sde->this_idx = this_idx;
+		sde->descq_cnt = descq_cnt;
+		sde->desc_avail = sdma_descq_freecnt(sde);
+		sde->sdma_shift = ilog2(descq_cnt);
+		sde->sdma_mask = (1 << sde->sdma_shift) - 1;
+
+		/* Create a mask specifically for each interrupt source */
+		sde->int_mask = (u64)1 << (0 * TXE_NUM_SDMA_ENGINES +
+					   this_idx);
+		sde->progress_mask = (u64)1 << (1 * TXE_NUM_SDMA_ENGINES +
+						this_idx);
+		sde->idle_mask = (u64)1 << (2 * TXE_NUM_SDMA_ENGINES +
+					    this_idx);
+		/* Create a combined mask to cover all 3 interrupt sources */
+		sde->imask = sde->int_mask | sde->progress_mask |
+			     sde->idle_mask;
+
+		spin_lock_init(&sde->tail_lock);
+		seqlock_init(&sde->head_lock);
+		spin_lock_init(&sde->senddmactrl_lock);
+		spin_lock_init(&sde->flushlist_lock);
+		/* insure there is always a zero bit */
+		sde->ahg_bits = 0xfffffffe00000000ULL;
+
+		sdma_set_state(sde, sdma_state_s00_hw_down);
+
+		/* set up reference counting */
+		kref_init(&sde->state.kref);
+		init_completion(&sde->state.comp);
+
+		INIT_LIST_HEAD(&sde->flushlist);
+		INIT_LIST_HEAD(&sde->dmawait);
+
+		sde->tail_csr =
+			get_kctxt_csr_addr(dd, this_idx, SD(TAIL));
+
+		tasklet_init(&sde->sdma_hw_clean_up_task, sdma_hw_clean_up_task,
+			     (unsigned long)sde);
+
+		tasklet_init(&sde->sdma_sw_clean_up_task, sdma_sw_clean_up_task,
+			     (unsigned long)sde);
+		INIT_WORK(&sde->err_halt_worker, sdma_err_halt_wait);
+		INIT_WORK(&sde->flush_worker, sdma_field_flush);
+
+		sde->progress_check_head = 0;
+
+		timer_setup(&sde->err_progress_check_timer,
+			    sdma_err_progress_check, 0);
+
+		sde->descq = dma_zalloc_coherent(
+			&dd->pcidev->dev,
+			descq_cnt * sizeof(u64[2]),
+			&sde->descq_phys,
+			GFP_KERNEL
+		);
+		if (!sde->descq)
+			goto bail;
+		sde->tx_ring =
+			kvzalloc_node(array_size(descq_cnt,
+						 sizeof(struct sdma_txreq *)),
+				      GFP_KERNEL, dd->node);
+		if (!sde->tx_ring)
+			goto bail;
+	}
+
+	dd->sdma_heads_size = L1_CACHE_BYTES * num_engines;
+	/* Allocate memory for DMA of head registers to memory */
+	dd->sdma_heads_dma = dma_zalloc_coherent(
+		&dd->pcidev->dev,
+		dd->sdma_heads_size,
+		&dd->sdma_heads_phys,
+		GFP_KERNEL
+	);
+	if (!dd->sdma_heads_dma) {
+		dd_dev_err(dd, "failed to allocate SendDMA head memory\n");
+		goto bail;
+	}
+
+	/* Allocate memory for pad */
+	dd->sdma_pad_dma = dma_zalloc_coherent(
+		&dd->pcidev->dev,
+		SDMA_PAD,
+		&dd->sdma_pad_phys,
+		GFP_KERNEL
+	);
+	if (!dd->sdma_pad_dma) {
+		dd_dev_err(dd, "failed to allocate SendDMA pad memory\n");
+		goto bail;
+	}
+
+	/* assign each engine to different cacheline and init registers */
+	curr_head = (void *)dd->sdma_heads_dma;
+	for (this_idx = 0; this_idx < num_engines; ++this_idx) {
+		unsigned long phys_offset;
+
+		sde = &dd->per_sdma[this_idx];
+
+		sde->head_dma = curr_head;
+		curr_head += L1_CACHE_BYTES;
+		phys_offset = (unsigned long)sde->head_dma -
+			      (unsigned long)dd->sdma_heads_dma;
+		sde->head_phys = dd->sdma_heads_phys + phys_offset;
+		init_sdma_regs(sde, per_sdma_credits, idle_cnt);
+	}
+	dd->flags |= HFI1_HAS_SEND_DMA;
+	dd->flags |= idle_cnt ? HFI1_HAS_SDMA_TIMEOUT : 0;
+	dd->num_sdma = num_engines;
+	ret = sdma_map_init(dd, port, ppd->vls_operational, NULL);
+	if (ret < 0)
+		goto bail;
+
+	tmp_sdma_rht = kzalloc(sizeof(*tmp_sdma_rht), GFP_KERNEL);
+	if (!tmp_sdma_rht) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	ret = rhashtable_init(tmp_sdma_rht, &sdma_rht_params);
+	if (ret < 0) {
+		kfree(tmp_sdma_rht);
+		goto bail;
+	}
+
+	dd->sdma_rht = tmp_sdma_rht;
+
+	dd_dev_info(dd, "SDMA num_sdma: %u\n", dd->num_sdma);
+	return 0;
+
+bail:
+	sdma_clean(dd, num_engines);
+	return ret;
+}
+
+/**
+ * sdma_all_running() - called when the link goes up
+ * @dd: hfi1_devdata
+ *
+ * This routine moves all engines to the running state.
+ */
+void sdma_all_running(struct hfi1_devdata *dd)
+{
+	struct sdma_engine *sde;
+	unsigned int i;
+
+	/* move all engines to running */
+	for (i = 0; i < dd->num_sdma; ++i) {
+		sde = &dd->per_sdma[i];
+		sdma_process_event(sde, sdma_event_e30_go_running);
+	}
+}
+
+/**
+ * sdma_all_idle() - called when the link goes down
+ * @dd: hfi1_devdata
+ *
+ * This routine moves all engines to the idle state.
+ */
+void sdma_all_idle(struct hfi1_devdata *dd)
+{
+	struct sdma_engine *sde;
+	unsigned int i;
+
+	/* idle all engines */
+	for (i = 0; i < dd->num_sdma; ++i) {
+		sde = &dd->per_sdma[i];
+		sdma_process_event(sde, sdma_event_e70_go_idle);
+	}
+}
+
+/**
+ * sdma_start() - called to kick off state processing for all engines
+ * @dd: hfi1_devdata
+ *
+ * This routine is for kicking off the state processing for all required
+ * sdma engines.  Interrupts need to be working at this point.
+ *
+ */
+void sdma_start(struct hfi1_devdata *dd)
+{
+	unsigned i;
+	struct sdma_engine *sde;
+
+	/* kick off the engines state processing */
+	for (i = 0; i < dd->num_sdma; ++i) {
+		sde = &dd->per_sdma[i];
+		sdma_process_event(sde, sdma_event_e10_go_hw_start);
+	}
+}
+
+/**
+ * sdma_exit() - used when module is removed
+ * @dd: hfi1_devdata
+ */
+void sdma_exit(struct hfi1_devdata *dd)
+{
+	unsigned this_idx;
+	struct sdma_engine *sde;
+
+	for (this_idx = 0; dd->per_sdma && this_idx < dd->num_sdma;
+			++this_idx) {
+		sde = &dd->per_sdma[this_idx];
+		if (!list_empty(&sde->dmawait))
+			dd_dev_err(dd, "sde %u: dmawait list not empty!\n",
+				   sde->this_idx);
+		sdma_process_event(sde, sdma_event_e00_go_hw_down);
+
+		del_timer_sync(&sde->err_progress_check_timer);
+
+		/*
+		 * This waits for the state machine to exit so it is not
+		 * necessary to kill the sdma_sw_clean_up_task to make sure
+		 * it is not running.
+		 */
+		sdma_finalput(&sde->state);
+	}
+}
+
+/*
+ * unmap the indicated descriptor
+ */
+static inline void sdma_unmap_desc(
+	struct hfi1_devdata *dd,
+	struct sdma_desc *descp)
+{
+	switch (sdma_mapping_type(descp)) {
+	case SDMA_MAP_SINGLE:
+		dma_unmap_single(
+			&dd->pcidev->dev,
+			sdma_mapping_addr(descp),
+			sdma_mapping_len(descp),
+			DMA_TO_DEVICE);
+		break;
+	case SDMA_MAP_PAGE:
+		dma_unmap_page(
+			&dd->pcidev->dev,
+			sdma_mapping_addr(descp),
+			sdma_mapping_len(descp),
+			DMA_TO_DEVICE);
+		break;
+	}
+}
+
+/*
+ * return the mode as indicated by the first
+ * descriptor in the tx.
+ */
+static inline u8 ahg_mode(struct sdma_txreq *tx)
+{
+	return (tx->descp[0].qw[1] & SDMA_DESC1_HEADER_MODE_SMASK)
+		>> SDMA_DESC1_HEADER_MODE_SHIFT;
+}
+
+/**
+ * __sdma_txclean() - clean tx of mappings, descp *kmalloc's
+ * @dd: hfi1_devdata for unmapping
+ * @tx: tx request to clean
+ *
+ * This is used in the progress routine to clean the tx or
+ * by the ULP to toss an in-process tx build.
+ *
+ * The code can be called multiple times without issue.
+ *
+ */
+void __sdma_txclean(
+	struct hfi1_devdata *dd,
+	struct sdma_txreq *tx)
+{
+	u16 i;
+
+	if (tx->num_desc) {
+		u8 skip = 0, mode = ahg_mode(tx);
+
+		/* unmap first */
+		sdma_unmap_desc(dd, &tx->descp[0]);
+		/* determine number of AHG descriptors to skip */
+		if (mode > SDMA_AHG_APPLY_UPDATE1)
+			skip = mode >> 1;
+		for (i = 1 + skip; i < tx->num_desc; i++)
+			sdma_unmap_desc(dd, &tx->descp[i]);
+		tx->num_desc = 0;
+	}
+	kfree(tx->coalesce_buf);
+	tx->coalesce_buf = NULL;
+	/* kmalloc'ed descp */
+	if (unlikely(tx->desc_limit > ARRAY_SIZE(tx->descs))) {
+		tx->desc_limit = ARRAY_SIZE(tx->descs);
+		kfree(tx->descp);
+	}
+}
+
+static inline u16 sdma_gethead(struct sdma_engine *sde)
+{
+	struct hfi1_devdata *dd = sde->dd;
+	int use_dmahead;
+	u16 hwhead;
+
+#ifdef CONFIG_SDMA_VERBOSITY
+	dd_dev_err(sde->dd, "CONFIG SDMA(%u) %s:%d %s()\n",
+		   sde->this_idx, slashstrip(__FILE__), __LINE__, __func__);
+#endif
+
+retry:
+	use_dmahead = HFI1_CAP_IS_KSET(USE_SDMA_HEAD) && __sdma_running(sde) &&
+					(dd->flags & HFI1_HAS_SDMA_TIMEOUT);
+	hwhead = use_dmahead ?
+		(u16)le64_to_cpu(*sde->head_dma) :
+		(u16)read_sde_csr(sde, SD(HEAD));
+
+	if (unlikely(HFI1_CAP_IS_KSET(SDMA_HEAD_CHECK))) {
+		u16 cnt;
+		u16 swtail;
+		u16 swhead;
+		int sane;
+
+		swhead = sde->descq_head & sde->sdma_mask;
+		/* this code is really bad for cache line trading */
+		swtail = READ_ONCE(sde->descq_tail) & sde->sdma_mask;
+		cnt = sde->descq_cnt;
+
+		if (swhead < swtail)
+			/* not wrapped */
+			sane = (hwhead >= swhead) & (hwhead <= swtail);
+		else if (swhead > swtail)
+			/* wrapped around */
+			sane = ((hwhead >= swhead) && (hwhead < cnt)) ||
+				(hwhead <= swtail);
+		else
+			/* empty */
+			sane = (hwhead == swhead);
+
+		if (unlikely(!sane)) {
+			dd_dev_err(dd, "SDMA(%u) bad head (%s) hwhd=%hu swhd=%hu swtl=%hu cnt=%hu\n",
+				   sde->this_idx,
+				   use_dmahead ? "dma" : "kreg",
+				   hwhead, swhead, swtail, cnt);
+			if (use_dmahead) {
+				/* try one more time, using csr */
+				use_dmahead = 0;
+				goto retry;
+			}
+			/* proceed as if no progress */
+			hwhead = swhead;
+		}
+	}
+	return hwhead;
+}
+
+/*
+ * This is called when there are send DMA descriptors that might be
+ * available.
+ *
+ * This is called with head_lock held.
+ */
+static void sdma_desc_avail(struct sdma_engine *sde, uint avail)
+{
+	struct iowait *wait, *nw;
+	struct iowait *waits[SDMA_WAIT_BATCH_SIZE];
+	uint i, n = 0, seq, max_idx = 0;
+	struct sdma_txreq *stx;
+	struct hfi1_ibdev *dev = &sde->dd->verbs_dev;
+	u8 max_starved_cnt = 0;
+
+#ifdef CONFIG_SDMA_VERBOSITY
+	dd_dev_err(sde->dd, "CONFIG SDMA(%u) %s:%d %s()\n", sde->this_idx,
+		   slashstrip(__FILE__), __LINE__, __func__);
+	dd_dev_err(sde->dd, "avail: %u\n", avail);
+#endif
+
+	do {
+		seq = read_seqbegin(&dev->iowait_lock);
+		if (!list_empty(&sde->dmawait)) {
+			/* at least one item */
+			write_seqlock(&dev->iowait_lock);
+			/* Harvest waiters wanting DMA descriptors */
+			list_for_each_entry_safe(
+					wait,
+					nw,
+					&sde->dmawait,
+					list) {
+				u16 num_desc = 0;
+
+				if (!wait->wakeup)
+					continue;
+				if (n == ARRAY_SIZE(waits))
+					break;
+				if (!list_empty(&wait->tx_head)) {
+					stx = list_first_entry(
+						&wait->tx_head,
+						struct sdma_txreq,
+						list);
+					num_desc = stx->num_desc;
+				}
+				if (num_desc > avail)
+					break;
+				avail -= num_desc;
+				/* Find the most starved wait memeber */
+				iowait_starve_find_max(wait, &max_starved_cnt,
+						       n, &max_idx);
+				list_del_init(&wait->list);
+				waits[n++] = wait;
+			}
+			write_sequnlock(&dev->iowait_lock);
+			break;
+		}
+	} while (read_seqretry(&dev->iowait_lock, seq));
+
+	/* Schedule the most starved one first */
+	if (n)
+		waits[max_idx]->wakeup(waits[max_idx], SDMA_AVAIL_REASON);
+
+	for (i = 0; i < n; i++)
+		if (i != max_idx)
+			waits[i]->wakeup(waits[i], SDMA_AVAIL_REASON);
+}
+
+/* head_lock must be held */
+static void sdma_make_progress(struct sdma_engine *sde, u64 status)
+{
+	struct sdma_txreq *txp = NULL;
+	int progress = 0;
+	u16 hwhead, swhead;
+	int idle_check_done = 0;
+
+	hwhead = sdma_gethead(sde);
+
+	/* The reason for some of the complexity of this code is that
+	 * not all descriptors have corresponding txps.  So, we have to
+	 * be able to skip over descs until we wander into the range of
+	 * the next txp on the list.
+	 */
+
+retry:
+	txp = get_txhead(sde);
+	swhead = sde->descq_head & sde->sdma_mask;
+	trace_hfi1_sdma_progress(sde, hwhead, swhead, txp);
+	while (swhead != hwhead) {
+		/* advance head, wrap if needed */
+		swhead = ++sde->descq_head & sde->sdma_mask;
+
+		/* if now past this txp's descs, do the callback */
+		if (txp && txp->next_descq_idx == swhead) {
+			/* remove from list */
+			sde->tx_ring[sde->tx_head++ & sde->sdma_mask] = NULL;
+			complete_tx(sde, txp, SDMA_TXREQ_S_OK);
+			/* see if there is another txp */
+			txp = get_txhead(sde);
+		}
+		trace_hfi1_sdma_progress(sde, hwhead, swhead, txp);
+		progress++;
+	}
+
+	/*
+	 * The SDMA idle interrupt is not guaranteed to be ordered with respect
+	 * to updates to the the dma_head location in host memory. The head
+	 * value read might not be fully up to date. If there are pending
+	 * descriptors and the SDMA idle interrupt fired then read from the
+	 * CSR SDMA head instead to get the latest value from the hardware.
+	 * The hardware SDMA head should be read at most once in this invocation
+	 * of sdma_make_progress(..) which is ensured by idle_check_done flag
+	 */
+	if ((status & sde->idle_mask) && !idle_check_done) {
+		u16 swtail;
+
+		swtail = READ_ONCE(sde->descq_tail) & sde->sdma_mask;
+		if (swtail != hwhead) {
+			hwhead = (u16)read_sde_csr(sde, SD(HEAD));
+			idle_check_done = 1;
+			goto retry;
+		}
+	}
+
+	sde->last_status = status;
+	if (progress)
+		sdma_desc_avail(sde, sdma_descq_freecnt(sde));
+}
+
+/*
+ * sdma_engine_interrupt() - interrupt handler for engine
+ * @sde: sdma engine
+ * @status: sdma interrupt reason
+ *
+ * Status is a mask of the 3 possible interrupts for this engine.  It will
+ * contain bits _only_ for this SDMA engine.  It will contain at least one
+ * bit, it may contain more.
+ */
+void sdma_engine_interrupt(struct sdma_engine *sde, u64 status)
+{
+	trace_hfi1_sdma_engine_interrupt(sde, status);
+	write_seqlock(&sde->head_lock);
+	sdma_set_desc_cnt(sde, sdma_desct_intr);
+	if (status & sde->idle_mask)
+		sde->idle_int_cnt++;
+	else if (status & sde->progress_mask)
+		sde->progress_int_cnt++;
+	else if (status & sde->int_mask)
+		sde->sdma_int_cnt++;
+	sdma_make_progress(sde, status);
+	write_sequnlock(&sde->head_lock);
+}
+
+/**
+ * sdma_engine_error() - error handler for engine
+ * @sde: sdma engine
+ * @status: sdma interrupt reason
+ */
+void sdma_engine_error(struct sdma_engine *sde, u64 status)
+{
+	unsigned long flags;
+
+#ifdef CONFIG_SDMA_VERBOSITY
+	dd_dev_err(sde->dd, "CONFIG SDMA(%u) error status 0x%llx state %s\n",
+		   sde->this_idx,
+		   (unsigned long long)status,
+		   sdma_state_names[sde->state.current_state]);
+#endif
+	spin_lock_irqsave(&sde->tail_lock, flags);
+	write_seqlock(&sde->head_lock);
+	if (status & ALL_SDMA_ENG_HALT_ERRS)
+		__sdma_process_event(sde, sdma_event_e60_hw_halted);
+	if (status & ~SD(ENG_ERR_STATUS_SDMA_HALT_ERR_SMASK)) {
+		dd_dev_err(sde->dd,
+			   "SDMA (%u) engine error: 0x%llx state %s\n",
+			   sde->this_idx,
+			   (unsigned long long)status,
+			   sdma_state_names[sde->state.current_state]);
+		dump_sdma_state(sde);
+	}
+	write_sequnlock(&sde->head_lock);
+	spin_unlock_irqrestore(&sde->tail_lock, flags);
+}
+
+static void sdma_sendctrl(struct sdma_engine *sde, unsigned op)
+{
+	u64 set_senddmactrl = 0;
+	u64 clr_senddmactrl = 0;
+	unsigned long flags;
+
+#ifdef CONFIG_SDMA_VERBOSITY
+	dd_dev_err(sde->dd, "CONFIG SDMA(%u) senddmactrl E=%d I=%d H=%d C=%d\n",
+		   sde->this_idx,
+		   (op & SDMA_SENDCTRL_OP_ENABLE) ? 1 : 0,
+		   (op & SDMA_SENDCTRL_OP_INTENABLE) ? 1 : 0,
+		   (op & SDMA_SENDCTRL_OP_HALT) ? 1 : 0,
+		   (op & SDMA_SENDCTRL_OP_CLEANUP) ? 1 : 0);
+#endif
+
+	if (op & SDMA_SENDCTRL_OP_ENABLE)
+		set_senddmactrl |= SD(CTRL_SDMA_ENABLE_SMASK);
+	else
+		clr_senddmactrl |= SD(CTRL_SDMA_ENABLE_SMASK);
+
+	if (op & SDMA_SENDCTRL_OP_INTENABLE)
+		set_senddmactrl |= SD(CTRL_SDMA_INT_ENABLE_SMASK);
+	else
+		clr_senddmactrl |= SD(CTRL_SDMA_INT_ENABLE_SMASK);
+
+	if (op & SDMA_SENDCTRL_OP_HALT)
+		set_senddmactrl |= SD(CTRL_SDMA_HALT_SMASK);
+	else
+		clr_senddmactrl |= SD(CTRL_SDMA_HALT_SMASK);
+
+	spin_lock_irqsave(&sde->senddmactrl_lock, flags);
+
+	sde->p_senddmactrl |= set_senddmactrl;
+	sde->p_senddmactrl &= ~clr_senddmactrl;
+
+	if (op & SDMA_SENDCTRL_OP_CLEANUP)
+		write_sde_csr(sde, SD(CTRL),
+			      sde->p_senddmactrl |
+			      SD(CTRL_SDMA_CLEANUP_SMASK));
+	else
+		write_sde_csr(sde, SD(CTRL), sde->p_senddmactrl);
+
+	spin_unlock_irqrestore(&sde->senddmactrl_lock, flags);
+
+#ifdef CONFIG_SDMA_VERBOSITY
+	sdma_dumpstate(sde);
+#endif
+}
+
+static void sdma_setlengen(struct sdma_engine *sde)
+{
+#ifdef CONFIG_SDMA_VERBOSITY
+	dd_dev_err(sde->dd, "CONFIG SDMA(%u) %s:%d %s()\n",
+		   sde->this_idx, slashstrip(__FILE__), __LINE__, __func__);
+#endif
+
+	/*
+	 * Set SendDmaLenGen and clear-then-set the MSB of the generation
+	 * count to enable generation checking and load the internal
+	 * generation counter.
+	 */
+	write_sde_csr(sde, SD(LEN_GEN),
+		      (sde->descq_cnt / 64) << SD(LEN_GEN_LENGTH_SHIFT));
+	write_sde_csr(sde, SD(LEN_GEN),
+		      ((sde->descq_cnt / 64) << SD(LEN_GEN_LENGTH_SHIFT)) |
+		      (4ULL << SD(LEN_GEN_GENERATION_SHIFT)));
+}
+
+static inline void sdma_update_tail(struct sdma_engine *sde, u16 tail)
+{
+	/* Commit writes to memory and advance the tail on the chip */
+	smp_wmb(); /* see get_txhead() */
+	writeq(tail, sde->tail_csr);
+}
+
+/*
+ * This is called when changing to state s10_hw_start_up_halt_wait as
+ * a result of send buffer errors or send DMA descriptor errors.
+ */
+static void sdma_hw_start_up(struct sdma_engine *sde)
+{
+	u64 reg;
+
+#ifdef CONFIG_SDMA_VERBOSITY
+	dd_dev_err(sde->dd, "CONFIG SDMA(%u) %s:%d %s()\n",
+		   sde->this_idx, slashstrip(__FILE__), __LINE__, __func__);
+#endif
+
+	sdma_setlengen(sde);
+	sdma_update_tail(sde, 0); /* Set SendDmaTail */
+	*sde->head_dma = 0;
+
+	reg = SD(ENG_ERR_CLEAR_SDMA_HEADER_REQUEST_FIFO_UNC_ERR_MASK) <<
+	      SD(ENG_ERR_CLEAR_SDMA_HEADER_REQUEST_FIFO_UNC_ERR_SHIFT);
+	write_sde_csr(sde, SD(ENG_ERR_CLEAR), reg);
+}
+
+/*
+ * set_sdma_integrity
+ *
+ * Set the SEND_DMA_CHECK_ENABLE register for send DMA engine 'sde'.
+ */
+static void set_sdma_integrity(struct sdma_engine *sde)
+{
+	struct hfi1_devdata *dd = sde->dd;
+
+	write_sde_csr(sde, SD(CHECK_ENABLE),
+		      hfi1_pkt_base_sdma_integrity(dd));
+}
+
+static void init_sdma_regs(
+	struct sdma_engine *sde,
+	u32 credits,
+	uint idle_cnt)
+{
+	u8 opval, opmask;
+#ifdef CONFIG_SDMA_VERBOSITY
+	struct hfi1_devdata *dd = sde->dd;
+
+	dd_dev_err(dd, "CONFIG SDMA(%u) %s:%d %s()\n",
+		   sde->this_idx, slashstrip(__FILE__), __LINE__, __func__);
+#endif
+
+	write_sde_csr(sde, SD(BASE_ADDR), sde->descq_phys);
+	sdma_setlengen(sde);
+	sdma_update_tail(sde, 0); /* Set SendDmaTail */
+	write_sde_csr(sde, SD(RELOAD_CNT), idle_cnt);
+	write_sde_csr(sde, SD(DESC_CNT), 0);
+	write_sde_csr(sde, SD(HEAD_ADDR), sde->head_phys);
+	write_sde_csr(sde, SD(MEMORY),
+		      ((u64)credits << SD(MEMORY_SDMA_MEMORY_CNT_SHIFT)) |
+		      ((u64)(credits * sde->this_idx) <<
+		       SD(MEMORY_SDMA_MEMORY_INDEX_SHIFT)));
+	write_sde_csr(sde, SD(ENG_ERR_MASK), ~0ull);
+	set_sdma_integrity(sde);
+	opmask = OPCODE_CHECK_MASK_DISABLED;
+	opval = OPCODE_CHECK_VAL_DISABLED;
+	write_sde_csr(sde, SD(CHECK_OPCODE),
+		      (opmask << SEND_CTXT_CHECK_OPCODE_MASK_SHIFT) |
+		      (opval << SEND_CTXT_CHECK_OPCODE_VALUE_SHIFT));
+}
+
+#ifdef CONFIG_SDMA_VERBOSITY
+
+#define sdma_dumpstate_helper0(reg) do { \
+		csr = read_csr(sde->dd, reg); \
+		dd_dev_err(sde->dd, "%36s     0x%016llx\n", #reg, csr); \
+	} while (0)
+
+#define sdma_dumpstate_helper(reg) do { \
+		csr = read_sde_csr(sde, reg); \
+		dd_dev_err(sde->dd, "%36s[%02u] 0x%016llx\n", \
+			#reg, sde->this_idx, csr); \
+	} while (0)
+
+#define sdma_dumpstate_helper2(reg) do { \
+		csr = read_csr(sde->dd, reg + (8 * i)); \
+		dd_dev_err(sde->dd, "%33s_%02u     0x%016llx\n", \
+				#reg, i, csr); \
+	} while (0)
+
+void sdma_dumpstate(struct sdma_engine *sde)
+{
+	u64 csr;
+	unsigned i;
+
+	sdma_dumpstate_helper(SD(CTRL));
+	sdma_dumpstate_helper(SD(STATUS));
+	sdma_dumpstate_helper0(SD(ERR_STATUS));
+	sdma_dumpstate_helper0(SD(ERR_MASK));
+	sdma_dumpstate_helper(SD(ENG_ERR_STATUS));
+	sdma_dumpstate_helper(SD(ENG_ERR_MASK));
+
+	for (i = 0; i < CCE_NUM_INT_CSRS; ++i) {
+		sdma_dumpstate_helper2(CCE_INT_STATUS);
+		sdma_dumpstate_helper2(CCE_INT_MASK);
+		sdma_dumpstate_helper2(CCE_INT_BLOCKED);
+	}
+
+	sdma_dumpstate_helper(SD(TAIL));
+	sdma_dumpstate_helper(SD(HEAD));
+	sdma_dumpstate_helper(SD(PRIORITY_THLD));
+	sdma_dumpstate_helper(SD(IDLE_CNT));
+	sdma_dumpstate_helper(SD(RELOAD_CNT));
+	sdma_dumpstate_helper(SD(DESC_CNT));
+	sdma_dumpstate_helper(SD(DESC_FETCHED_CNT));
+	sdma_dumpstate_helper(SD(MEMORY));
+	sdma_dumpstate_helper0(SD(ENGINES));
+	sdma_dumpstate_helper0(SD(MEM_SIZE));
+	/* sdma_dumpstate_helper(SEND_EGRESS_SEND_DMA_STATUS);  */
+	sdma_dumpstate_helper(SD(BASE_ADDR));
+	sdma_dumpstate_helper(SD(LEN_GEN));
+	sdma_dumpstate_helper(SD(HEAD_ADDR));
+	sdma_dumpstate_helper(SD(CHECK_ENABLE));
+	sdma_dumpstate_helper(SD(CHECK_VL));
+	sdma_dumpstate_helper(SD(CHECK_JOB_KEY));
+	sdma_dumpstate_helper(SD(CHECK_PARTITION_KEY));
+	sdma_dumpstate_helper(SD(CHECK_SLID));
+	sdma_dumpstate_helper(SD(CHECK_OPCODE));
+}
+#endif
+
+static void dump_sdma_state(struct sdma_engine *sde)
+{
+	struct hw_sdma_desc *descqp;
+	u64 desc[2];
+	u64 addr;
+	u8 gen;
+	u16 len;
+	u16 head, tail, cnt;
+
+	head = sde->descq_head & sde->sdma_mask;
+	tail = sde->descq_tail & sde->sdma_mask;
+	cnt = sdma_descq_freecnt(sde);
+
+	dd_dev_err(sde->dd,
+		   "SDMA (%u) descq_head: %u descq_tail: %u freecnt: %u FLE %d\n",
+		   sde->this_idx, head, tail, cnt,
+		   !list_empty(&sde->flushlist));
+
+	/* print info for each entry in the descriptor queue */
+	while (head != tail) {
+		char flags[6] = { 'x', 'x', 'x', 'x', 0 };
+
+		descqp = &sde->descq[head];
+		desc[0] = le64_to_cpu(descqp->qw[0]);
+		desc[1] = le64_to_cpu(descqp->qw[1]);
+		flags[0] = (desc[1] & SDMA_DESC1_INT_REQ_FLAG) ? 'I' : '-';
+		flags[1] = (desc[1] & SDMA_DESC1_HEAD_TO_HOST_FLAG) ?
+				'H' : '-';
+		flags[2] = (desc[0] & SDMA_DESC0_FIRST_DESC_FLAG) ? 'F' : '-';
+		flags[3] = (desc[0] & SDMA_DESC0_LAST_DESC_FLAG) ? 'L' : '-';
+		addr = (desc[0] >> SDMA_DESC0_PHY_ADDR_SHIFT)
+			& SDMA_DESC0_PHY_ADDR_MASK;
+		gen = (desc[1] >> SDMA_DESC1_GENERATION_SHIFT)
+			& SDMA_DESC1_GENERATION_MASK;
+		len = (desc[0] >> SDMA_DESC0_BYTE_COUNT_SHIFT)
+			& SDMA_DESC0_BYTE_COUNT_MASK;
+		dd_dev_err(sde->dd,
+			   "SDMA sdmadesc[%u]: flags:%s addr:0x%016llx gen:%u len:%u bytes\n",
+			   head, flags, addr, gen, len);
+		dd_dev_err(sde->dd,
+			   "\tdesc0:0x%016llx desc1 0x%016llx\n",
+			   desc[0], desc[1]);
+		if (desc[0] & SDMA_DESC0_FIRST_DESC_FLAG)
+			dd_dev_err(sde->dd,
+				   "\taidx: %u amode: %u alen: %u\n",
+				   (u8)((desc[1] &
+					 SDMA_DESC1_HEADER_INDEX_SMASK) >>
+					SDMA_DESC1_HEADER_INDEX_SHIFT),
+				   (u8)((desc[1] &
+					 SDMA_DESC1_HEADER_MODE_SMASK) >>
+					SDMA_DESC1_HEADER_MODE_SHIFT),
+				   (u8)((desc[1] &
+					 SDMA_DESC1_HEADER_DWS_SMASK) >>
+					SDMA_DESC1_HEADER_DWS_SHIFT));
+		head++;
+		head &= sde->sdma_mask;
+	}
+}
+
+#define SDE_FMT \
+	"SDE %u CPU %d STE %s C 0x%llx S 0x%016llx E 0x%llx T(HW) 0x%llx T(SW) 0x%x H(HW) 0x%llx H(SW) 0x%x H(D) 0x%llx DM 0x%llx GL 0x%llx R 0x%llx LIS 0x%llx AHGI 0x%llx TXT %u TXH %u DT %u DH %u FLNE %d DQF %u SLC 0x%llx\n"
+/**
+ * sdma_seqfile_dump_sde() - debugfs dump of sde
+ * @s: seq file
+ * @sde: send dma engine to dump
+ *
+ * This routine dumps the sde to the indicated seq file.
+ */
+void sdma_seqfile_dump_sde(struct seq_file *s, struct sdma_engine *sde)
+{
+	u16 head, tail;
+	struct hw_sdma_desc *descqp;
+	u64 desc[2];
+	u64 addr;
+	u8 gen;
+	u16 len;
+
+	head = sde->descq_head & sde->sdma_mask;
+	tail = READ_ONCE(sde->descq_tail) & sde->sdma_mask;
+	seq_printf(s, SDE_FMT, sde->this_idx,
+		   sde->cpu,
+		   sdma_state_name(sde->state.current_state),
+		   (unsigned long long)read_sde_csr(sde, SD(CTRL)),
+		   (unsigned long long)read_sde_csr(sde, SD(STATUS)),
+		   (unsigned long long)read_sde_csr(sde, SD(ENG_ERR_STATUS)),
+		   (unsigned long long)read_sde_csr(sde, SD(TAIL)), tail,
+		   (unsigned long long)read_sde_csr(sde, SD(HEAD)), head,
+		   (unsigned long long)le64_to_cpu(*sde->head_dma),
+		   (unsigned long long)read_sde_csr(sde, SD(MEMORY)),
+		   (unsigned long long)read_sde_csr(sde, SD(LEN_GEN)),
+		   (unsigned long long)read_sde_csr(sde, SD(RELOAD_CNT)),
+		   (unsigned long long)sde->last_status,
+		   (unsigned long long)sde->ahg_bits,
+		   sde->tx_tail,
+		   sde->tx_head,
+		   sde->descq_tail,
+		   sde->descq_head,
+		   !list_empty(&sde->flushlist),
+		   sde->descq_full_count,
+		   (unsigned long long)read_sde_csr(sde, SEND_DMA_CHECK_SLID));
+
+	/* print info for each entry in the descriptor queue */
+	while (head != tail) {
+		char flags[6] = { 'x', 'x', 'x', 'x', 0 };
+
+		descqp = &sde->descq[head];
+		desc[0] = le64_to_cpu(descqp->qw[0]);
+		desc[1] = le64_to_cpu(descqp->qw[1]);
+		flags[0] = (desc[1] & SDMA_DESC1_INT_REQ_FLAG) ? 'I' : '-';
+		flags[1] = (desc[1] & SDMA_DESC1_HEAD_TO_HOST_FLAG) ?
+				'H' : '-';
+		flags[2] = (desc[0] & SDMA_DESC0_FIRST_DESC_FLAG) ? 'F' : '-';
+		flags[3] = (desc[0] & SDMA_DESC0_LAST_DESC_FLAG) ? 'L' : '-';
+		addr = (desc[0] >> SDMA_DESC0_PHY_ADDR_SHIFT)
+			& SDMA_DESC0_PHY_ADDR_MASK;
+		gen = (desc[1] >> SDMA_DESC1_GENERATION_SHIFT)
+			& SDMA_DESC1_GENERATION_MASK;
+		len = (desc[0] >> SDMA_DESC0_BYTE_COUNT_SHIFT)
+			& SDMA_DESC0_BYTE_COUNT_MASK;
+		seq_printf(s,
+			   "\tdesc[%u]: flags:%s addr:0x%016llx gen:%u len:%u bytes\n",
+			   head, flags, addr, gen, len);
+		if (desc[0] & SDMA_DESC0_FIRST_DESC_FLAG)
+			seq_printf(s, "\t\tahgidx: %u ahgmode: %u\n",
+				   (u8)((desc[1] &
+					 SDMA_DESC1_HEADER_INDEX_SMASK) >>
+					SDMA_DESC1_HEADER_INDEX_SHIFT),
+				   (u8)((desc[1] &
+					 SDMA_DESC1_HEADER_MODE_SMASK) >>
+					SDMA_DESC1_HEADER_MODE_SHIFT));
+		head = (head + 1) & sde->sdma_mask;
+	}
+}
+
+/*
+ * add the generation number into
+ * the qw1 and return
+ */
+static inline u64 add_gen(struct sdma_engine *sde, u64 qw1)
+{
+	u8 generation = (sde->descq_tail >> sde->sdma_shift) & 3;
+
+	qw1 &= ~SDMA_DESC1_GENERATION_SMASK;
+	qw1 |= ((u64)generation & SDMA_DESC1_GENERATION_MASK)
+			<< SDMA_DESC1_GENERATION_SHIFT;
+	return qw1;
+}
+
+/*
+ * This routine submits the indicated tx
+ *
+ * Space has already been guaranteed and
+ * tail side of ring is locked.
+ *
+ * The hardware tail update is done
+ * in the caller and that is facilitated
+ * by returning the new tail.
+ *
+ * There is special case logic for ahg
+ * to not add the generation number for
+ * up to 2 descriptors that follow the
+ * first descriptor.
+ *
+ */
+static inline u16 submit_tx(struct sdma_engine *sde, struct sdma_txreq *tx)
+{
+	int i;
+	u16 tail;
+	struct sdma_desc *descp = tx->descp;
+	u8 skip = 0, mode = ahg_mode(tx);
+
+	tail = sde->descq_tail & sde->sdma_mask;
+	sde->descq[tail].qw[0] = cpu_to_le64(descp->qw[0]);
+	sde->descq[tail].qw[1] = cpu_to_le64(add_gen(sde, descp->qw[1]));
+	trace_hfi1_sdma_descriptor(sde, descp->qw[0], descp->qw[1],
+				   tail, &sde->descq[tail]);
+	tail = ++sde->descq_tail & sde->sdma_mask;
+	descp++;
+	if (mode > SDMA_AHG_APPLY_UPDATE1)
+		skip = mode >> 1;
+	for (i = 1; i < tx->num_desc; i++, descp++) {
+		u64 qw1;
+
+		sde->descq[tail].qw[0] = cpu_to_le64(descp->qw[0]);
+		if (skip) {
+			/* edits don't have generation */
+			qw1 = descp->qw[1];
+			skip--;
+		} else {
+			/* replace generation with real one for non-edits */
+			qw1 = add_gen(sde, descp->qw[1]);
+		}
+		sde->descq[tail].qw[1] = cpu_to_le64(qw1);
+		trace_hfi1_sdma_descriptor(sde, descp->qw[0], qw1,
+					   tail, &sde->descq[tail]);
+		tail = ++sde->descq_tail & sde->sdma_mask;
+	}
+	tx->next_descq_idx = tail;
+#ifdef CONFIG_HFI1_DEBUG_SDMA_ORDER
+	tx->sn = sde->tail_sn++;
+	trace_hfi1_sdma_in_sn(sde, tx->sn);
+	WARN_ON_ONCE(sde->tx_ring[sde->tx_tail & sde->sdma_mask]);
+#endif
+	sde->tx_ring[sde->tx_tail++ & sde->sdma_mask] = tx;
+	sde->desc_avail -= tx->num_desc;
+	return tail;
+}
+
+/*
+ * Check for progress
+ */
+static int sdma_check_progress(
+	struct sdma_engine *sde,
+	struct iowait *wait,
+	struct sdma_txreq *tx,
+	bool pkts_sent)
+{
+	int ret;
+
+	sde->desc_avail = sdma_descq_freecnt(sde);
+	if (tx->num_desc <= sde->desc_avail)
+		return -EAGAIN;
+	/* pulse the head_lock */
+	if (wait && wait->sleep) {
+		unsigned seq;
+
+		seq = raw_seqcount_begin(
+			(const seqcount_t *)&sde->head_lock.seqcount);
+		ret = wait->sleep(sde, wait, tx, seq, pkts_sent);
+		if (ret == -EAGAIN)
+			sde->desc_avail = sdma_descq_freecnt(sde);
+	} else {
+		ret = -EBUSY;
+	}
+	return ret;
+}
+
+/**
+ * sdma_send_txreq() - submit a tx req to ring
+ * @sde: sdma engine to use
+ * @wait: wait structure to use when full (may be NULL)
+ * @tx: sdma_txreq to submit
+ * @pkts_sent: has any packet been sent yet?
+ *
+ * The call submits the tx into the ring.  If a iowait structure is non-NULL
+ * the packet will be queued to the list in wait.
+ *
+ * Return:
+ * 0 - Success, -EINVAL - sdma_txreq incomplete, -EBUSY - no space in
+ * ring (wait == NULL)
+ * -EIOCBQUEUED - tx queued to iowait, -ECOMM bad sdma state
+ */
+int sdma_send_txreq(struct sdma_engine *sde,
+		    struct iowait *wait,
+		    struct sdma_txreq *tx,
+		    bool pkts_sent)
+{
+	int ret = 0;
+	u16 tail;
+	unsigned long flags;
+
+	/* user should have supplied entire packet */
+	if (unlikely(tx->tlen))
+		return -EINVAL;
+	tx->wait = wait;
+	spin_lock_irqsave(&sde->tail_lock, flags);
+retry:
+	if (unlikely(!__sdma_running(sde)))
+		goto unlock_noconn;
+	if (unlikely(tx->num_desc > sde->desc_avail))
+		goto nodesc;
+	tail = submit_tx(sde, tx);
+	if (wait)
+		iowait_sdma_inc(wait);
+	sdma_update_tail(sde, tail);
+unlock:
+	spin_unlock_irqrestore(&sde->tail_lock, flags);
+	return ret;
+unlock_noconn:
+	if (wait)
+		iowait_sdma_inc(wait);
+	tx->next_descq_idx = 0;
+#ifdef CONFIG_HFI1_DEBUG_SDMA_ORDER
+	tx->sn = sde->tail_sn++;
+	trace_hfi1_sdma_in_sn(sde, tx->sn);
+#endif
+	spin_lock(&sde->flushlist_lock);
+	list_add_tail(&tx->list, &sde->flushlist);
+	spin_unlock(&sde->flushlist_lock);
+	if (wait) {
+		wait->tx_count++;
+		wait->count += tx->num_desc;
+	}
+	queue_work_on(sde->cpu, system_highpri_wq, &sde->flush_worker);
+	ret = -ECOMM;
+	goto unlock;
+nodesc:
+	ret = sdma_check_progress(sde, wait, tx, pkts_sent);
+	if (ret == -EAGAIN) {
+		ret = 0;
+		goto retry;
+	}
+	sde->descq_full_count++;
+	goto unlock;
+}
+
+/**
+ * sdma_send_txlist() - submit a list of tx req to ring
+ * @sde: sdma engine to use
+ * @wait: wait structure to use when full (may be NULL)
+ * @tx_list: list of sdma_txreqs to submit
+ * @count: pointer to a u32 which, after return will contain the total number of
+ *         sdma_txreqs removed from the tx_list. This will include sdma_txreqs
+ *         whose SDMA descriptors are submitted to the ring and the sdma_txreqs
+ *         which are added to SDMA engine flush list if the SDMA engine state is
+ *         not running.
+ *
+ * The call submits the list into the ring.
+ *
+ * If the iowait structure is non-NULL and not equal to the iowait list
+ * the unprocessed part of the list  will be appended to the list in wait.
+ *
+ * In all cases, the tx_list will be updated so the head of the tx_list is
+ * the list of descriptors that have yet to be transmitted.
+ *
+ * The intent of this call is to provide a more efficient
+ * way of submitting multiple packets to SDMA while holding the tail
+ * side locking.
+ *
+ * Return:
+ * 0 - Success,
+ * -EINVAL - sdma_txreq incomplete, -EBUSY - no space in ring (wait == NULL)
+ * -EIOCBQUEUED - tx queued to iowait, -ECOMM bad sdma state
+ */
+int sdma_send_txlist(struct sdma_engine *sde, struct iowait *wait,
+		     struct list_head *tx_list, u32 *count_out)
+{
+	struct sdma_txreq *tx, *tx_next;
+	int ret = 0;
+	unsigned long flags;
+	u16 tail = INVALID_TAIL;
+	u32 submit_count = 0, flush_count = 0, total_count;
+
+	spin_lock_irqsave(&sde->tail_lock, flags);
+retry:
+	list_for_each_entry_safe(tx, tx_next, tx_list, list) {
+		tx->wait = wait;
+		if (unlikely(!__sdma_running(sde)))
+			goto unlock_noconn;
+		if (unlikely(tx->num_desc > sde->desc_avail))
+			goto nodesc;
+		if (unlikely(tx->tlen)) {
+			ret = -EINVAL;
+			goto update_tail;
+		}
+		list_del_init(&tx->list);
+		tail = submit_tx(sde, tx);
+		submit_count++;
+		if (tail != INVALID_TAIL &&
+		    (submit_count & SDMA_TAIL_UPDATE_THRESH) == 0) {
+			sdma_update_tail(sde, tail);
+			tail = INVALID_TAIL;
+		}
+	}
+update_tail:
+	total_count = submit_count + flush_count;
+	if (wait) {
+		iowait_sdma_add(wait, total_count);
+		iowait_starve_clear(submit_count > 0, wait);
+	}
+	if (tail != INVALID_TAIL)
+		sdma_update_tail(sde, tail);
+	spin_unlock_irqrestore(&sde->tail_lock, flags);
+	*count_out = total_count;
+	return ret;
+unlock_noconn:
+	spin_lock(&sde->flushlist_lock);
+	list_for_each_entry_safe(tx, tx_next, tx_list, list) {
+		tx->wait = wait;
+		list_del_init(&tx->list);
+		tx->next_descq_idx = 0;
+#ifdef CONFIG_HFI1_DEBUG_SDMA_ORDER
+		tx->sn = sde->tail_sn++;
+		trace_hfi1_sdma_in_sn(sde, tx->sn);
+#endif
+		list_add_tail(&tx->list, &sde->flushlist);
+		flush_count++;
+		if (wait) {
+			wait->tx_count++;
+			wait->count += tx->num_desc;
+		}
+	}
+	spin_unlock(&sde->flushlist_lock);
+	queue_work_on(sde->cpu, system_highpri_wq, &sde->flush_worker);
+	ret = -ECOMM;
+	goto update_tail;
+nodesc:
+	ret = sdma_check_progress(sde, wait, tx, submit_count > 0);
+	if (ret == -EAGAIN) {
+		ret = 0;
+		goto retry;
+	}
+	sde->descq_full_count++;
+	goto update_tail;
+}
+
+static void sdma_process_event(struct sdma_engine *sde, enum sdma_events event)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&sde->tail_lock, flags);
+	write_seqlock(&sde->head_lock);
+
+	__sdma_process_event(sde, event);
+
+	if (sde->state.current_state == sdma_state_s99_running)
+		sdma_desc_avail(sde, sdma_descq_freecnt(sde));
+
+	write_sequnlock(&sde->head_lock);
+	spin_unlock_irqrestore(&sde->tail_lock, flags);
+}
+
+static void __sdma_process_event(struct sdma_engine *sde,
+				 enum sdma_events event)
+{
+	struct sdma_state *ss = &sde->state;
+	int need_progress = 0;
+
+	/* CONFIG SDMA temporary */
+#ifdef CONFIG_SDMA_VERBOSITY
+	dd_dev_err(sde->dd, "CONFIG SDMA(%u) [%s] %s\n", sde->this_idx,
+		   sdma_state_names[ss->current_state],
+		   sdma_event_names[event]);
+#endif
+
+	switch (ss->current_state) {
+	case sdma_state_s00_hw_down:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			break;
+		case sdma_event_e30_go_running:
+			/*
+			 * If down, but running requested (usually result
+			 * of link up, then we need to start up.
+			 * This can happen when hw down is requested while
+			 * bringing the link up with traffic active on
+			 * 7220, e.g.
+			 */
+			ss->go_s99_running = 1;
+			/* fall through -- and start dma engine */
+		case sdma_event_e10_go_hw_start:
+			/* This reference means the state machine is started */
+			sdma_get(&sde->state);
+			sdma_set_state(sde,
+				       sdma_state_s10_hw_start_up_halt_wait);
+			break;
+		case sdma_event_e15_hw_halt_done:
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			break;
+		case sdma_event_e40_sw_cleaned:
+			sdma_sw_tear_down(sde);
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			break;
+		case sdma_event_e70_go_idle:
+			break;
+		case sdma_event_e80_hw_freeze:
+			break;
+		case sdma_event_e81_hw_frozen:
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			break;
+		case sdma_event_e85_link_down:
+			break;
+		case sdma_event_e90_sw_halted:
+			break;
+		}
+		break;
+
+	case sdma_state_s10_hw_start_up_halt_wait:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			sdma_set_state(sde, sdma_state_s00_hw_down);
+			sdma_sw_tear_down(sde);
+			break;
+		case sdma_event_e10_go_hw_start:
+			break;
+		case sdma_event_e15_hw_halt_done:
+			sdma_set_state(sde,
+				       sdma_state_s15_hw_start_up_clean_wait);
+			sdma_start_hw_clean_up(sde);
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			break;
+		case sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case sdma_event_e40_sw_cleaned:
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			schedule_work(&sde->err_halt_worker);
+			break;
+		case sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e80_hw_freeze:
+			break;
+		case sdma_event_e81_hw_frozen:
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			break;
+		case sdma_event_e85_link_down:
+			break;
+		case sdma_event_e90_sw_halted:
+			break;
+		}
+		break;
+
+	case sdma_state_s15_hw_start_up_clean_wait:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			sdma_set_state(sde, sdma_state_s00_hw_down);
+			sdma_sw_tear_down(sde);
+			break;
+		case sdma_event_e10_go_hw_start:
+			break;
+		case sdma_event_e15_hw_halt_done:
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			sdma_hw_start_up(sde);
+			sdma_set_state(sde, ss->go_s99_running ?
+				       sdma_state_s99_running :
+				       sdma_state_s20_idle);
+			break;
+		case sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case sdma_event_e40_sw_cleaned:
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			break;
+		case sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e80_hw_freeze:
+			break;
+		case sdma_event_e81_hw_frozen:
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			break;
+		case sdma_event_e85_link_down:
+			break;
+		case sdma_event_e90_sw_halted:
+			break;
+		}
+		break;
+
+	case sdma_state_s20_idle:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			sdma_set_state(sde, sdma_state_s00_hw_down);
+			sdma_sw_tear_down(sde);
+			break;
+		case sdma_event_e10_go_hw_start:
+			break;
+		case sdma_event_e15_hw_halt_done:
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			break;
+		case sdma_event_e30_go_running:
+			sdma_set_state(sde, sdma_state_s99_running);
+			ss->go_s99_running = 1;
+			break;
+		case sdma_event_e40_sw_cleaned:
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			sdma_set_state(sde, sdma_state_s50_hw_halt_wait);
+			schedule_work(&sde->err_halt_worker);
+			break;
+		case sdma_event_e70_go_idle:
+			break;
+		case sdma_event_e85_link_down:
+			/* fall through */
+		case sdma_event_e80_hw_freeze:
+			sdma_set_state(sde, sdma_state_s80_hw_freeze);
+			atomic_dec(&sde->dd->sdma_unfreeze_count);
+			wake_up_interruptible(&sde->dd->sdma_unfreeze_wq);
+			break;
+		case sdma_event_e81_hw_frozen:
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			break;
+		case sdma_event_e90_sw_halted:
+			break;
+		}
+		break;
+
+	case sdma_state_s30_sw_clean_up_wait:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			sdma_set_state(sde, sdma_state_s00_hw_down);
+			break;
+		case sdma_event_e10_go_hw_start:
+			break;
+		case sdma_event_e15_hw_halt_done:
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			break;
+		case sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case sdma_event_e40_sw_cleaned:
+			sdma_set_state(sde, sdma_state_s40_hw_clean_up_wait);
+			sdma_start_hw_clean_up(sde);
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			break;
+		case sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e80_hw_freeze:
+			break;
+		case sdma_event_e81_hw_frozen:
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			break;
+		case sdma_event_e85_link_down:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e90_sw_halted:
+			break;
+		}
+		break;
+
+	case sdma_state_s40_hw_clean_up_wait:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			sdma_set_state(sde, sdma_state_s00_hw_down);
+			tasklet_hi_schedule(&sde->sdma_sw_clean_up_task);
+			break;
+		case sdma_event_e10_go_hw_start:
+			break;
+		case sdma_event_e15_hw_halt_done:
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			sdma_hw_start_up(sde);
+			sdma_set_state(sde, ss->go_s99_running ?
+				       sdma_state_s99_running :
+				       sdma_state_s20_idle);
+			break;
+		case sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case sdma_event_e40_sw_cleaned:
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			break;
+		case sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e80_hw_freeze:
+			break;
+		case sdma_event_e81_hw_frozen:
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			break;
+		case sdma_event_e85_link_down:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e90_sw_halted:
+			break;
+		}
+		break;
+
+	case sdma_state_s50_hw_halt_wait:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			sdma_set_state(sde, sdma_state_s00_hw_down);
+			tasklet_hi_schedule(&sde->sdma_sw_clean_up_task);
+			break;
+		case sdma_event_e10_go_hw_start:
+			break;
+		case sdma_event_e15_hw_halt_done:
+			sdma_set_state(sde, sdma_state_s30_sw_clean_up_wait);
+			tasklet_hi_schedule(&sde->sdma_sw_clean_up_task);
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			break;
+		case sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case sdma_event_e40_sw_cleaned:
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			schedule_work(&sde->err_halt_worker);
+			break;
+		case sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e80_hw_freeze:
+			break;
+		case sdma_event_e81_hw_frozen:
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			break;
+		case sdma_event_e85_link_down:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e90_sw_halted:
+			break;
+		}
+		break;
+
+	case sdma_state_s60_idle_halt_wait:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			sdma_set_state(sde, sdma_state_s00_hw_down);
+			tasklet_hi_schedule(&sde->sdma_sw_clean_up_task);
+			break;
+		case sdma_event_e10_go_hw_start:
+			break;
+		case sdma_event_e15_hw_halt_done:
+			sdma_set_state(sde, sdma_state_s30_sw_clean_up_wait);
+			tasklet_hi_schedule(&sde->sdma_sw_clean_up_task);
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			break;
+		case sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case sdma_event_e40_sw_cleaned:
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			schedule_work(&sde->err_halt_worker);
+			break;
+		case sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e80_hw_freeze:
+			break;
+		case sdma_event_e81_hw_frozen:
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			break;
+		case sdma_event_e85_link_down:
+			break;
+		case sdma_event_e90_sw_halted:
+			break;
+		}
+		break;
+
+	case sdma_state_s80_hw_freeze:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			sdma_set_state(sde, sdma_state_s00_hw_down);
+			tasklet_hi_schedule(&sde->sdma_sw_clean_up_task);
+			break;
+		case sdma_event_e10_go_hw_start:
+			break;
+		case sdma_event_e15_hw_halt_done:
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			break;
+		case sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case sdma_event_e40_sw_cleaned:
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			break;
+		case sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e80_hw_freeze:
+			break;
+		case sdma_event_e81_hw_frozen:
+			sdma_set_state(sde, sdma_state_s82_freeze_sw_clean);
+			tasklet_hi_schedule(&sde->sdma_sw_clean_up_task);
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			break;
+		case sdma_event_e85_link_down:
+			break;
+		case sdma_event_e90_sw_halted:
+			break;
+		}
+		break;
+
+	case sdma_state_s82_freeze_sw_clean:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			sdma_set_state(sde, sdma_state_s00_hw_down);
+			tasklet_hi_schedule(&sde->sdma_sw_clean_up_task);
+			break;
+		case sdma_event_e10_go_hw_start:
+			break;
+		case sdma_event_e15_hw_halt_done:
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			break;
+		case sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case sdma_event_e40_sw_cleaned:
+			/* notify caller this engine is done cleaning */
+			atomic_dec(&sde->dd->sdma_unfreeze_count);
+			wake_up_interruptible(&sde->dd->sdma_unfreeze_wq);
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			break;
+		case sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case sdma_event_e80_hw_freeze:
+			break;
+		case sdma_event_e81_hw_frozen:
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			sdma_hw_start_up(sde);
+			sdma_set_state(sde, ss->go_s99_running ?
+				       sdma_state_s99_running :
+				       sdma_state_s20_idle);
+			break;
+		case sdma_event_e85_link_down:
+			break;
+		case sdma_event_e90_sw_halted:
+			break;
+		}
+		break;
+
+	case sdma_state_s99_running:
+		switch (event) {
+		case sdma_event_e00_go_hw_down:
+			sdma_set_state(sde, sdma_state_s00_hw_down);
+			tasklet_hi_schedule(&sde->sdma_sw_clean_up_task);
+			break;
+		case sdma_event_e10_go_hw_start:
+			break;
+		case sdma_event_e15_hw_halt_done:
+			break;
+		case sdma_event_e25_hw_clean_up_done:
+			break;
+		case sdma_event_e30_go_running:
+			break;
+		case sdma_event_e40_sw_cleaned:
+			break;
+		case sdma_event_e50_hw_cleaned:
+			break;
+		case sdma_event_e60_hw_halted:
+			need_progress = 1;
+			sdma_err_progress_check_schedule(sde);
+			/* fall through */
+		case sdma_event_e90_sw_halted:
+			/*
+			* SW initiated halt does not perform engines
+			* progress check
+			*/
+			sdma_set_state(sde, sdma_state_s50_hw_halt_wait);
+			schedule_work(&sde->err_halt_worker);
+			break;
+		case sdma_event_e70_go_idle:
+			sdma_set_state(sde, sdma_state_s60_idle_halt_wait);
+			break;
+		case sdma_event_e85_link_down:
+			ss->go_s99_running = 0;
+			/* fall through */
+		case sdma_event_e80_hw_freeze:
+			sdma_set_state(sde, sdma_state_s80_hw_freeze);
+			atomic_dec(&sde->dd->sdma_unfreeze_count);
+			wake_up_interruptible(&sde->dd->sdma_unfreeze_wq);
+			break;
+		case sdma_event_e81_hw_frozen:
+			break;
+		case sdma_event_e82_hw_unfreeze:
+			break;
+		}
+		break;
+	}
+
+	ss->last_event = event;
+	if (need_progress)
+		sdma_make_progress(sde, 0);
+}
+
+/*
+ * _extend_sdma_tx_descs() - helper to extend txreq
+ *
+ * This is called once the initial nominal allocation
+ * of descriptors in the sdma_txreq is exhausted.
+ *
+ * The code will bump the allocation up to the max
+ * of MAX_DESC (64) descriptors. There doesn't seem
+ * much point in an interim step. The last descriptor
+ * is reserved for coalesce buffer in order to support
+ * cases where input packet has >MAX_DESC iovecs.
+ *
+ */
+static int _extend_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx)
+{
+	int i;
+	struct sdma_desc *descp;
+
+	/* Handle last descriptor */
+	if (unlikely((tx->num_desc == (MAX_DESC - 1)))) {
+		/* if tlen is 0, it is for padding, release last descriptor */
+		if (!tx->tlen) {
+			tx->desc_limit = MAX_DESC;
+		} else if (!tx->coalesce_buf) {
+			/* allocate coalesce buffer with space for padding */
+			tx->coalesce_buf = kmalloc(tx->tlen + sizeof(u32),
+						   GFP_ATOMIC);
+			if (!tx->coalesce_buf)
+				goto enomem;
+			tx->coalesce_idx = 0;
+		}
+		return 0;
+	}
+
+	if (unlikely(tx->num_desc == MAX_DESC))
+		goto enomem;
+
+	descp = kmalloc_array(MAX_DESC, sizeof(struct sdma_desc), GFP_ATOMIC);
+	if (!descp)
+		goto enomem;
+	tx->descp = descp;
+
+	/* reserve last descriptor for coalescing */
+	tx->desc_limit = MAX_DESC - 1;
+	/* copy ones already built */
+	for (i = 0; i < tx->num_desc; i++)
+		tx->descp[i] = tx->descs[i];
+	return 0;
+enomem:
+	__sdma_txclean(dd, tx);
+	return -ENOMEM;
+}
+
+/*
+ * ext_coal_sdma_tx_descs() - extend or coalesce sdma tx descriptors
+ *
+ * This is called once the initial nominal allocation of descriptors
+ * in the sdma_txreq is exhausted.
+ *
+ * This function calls _extend_sdma_tx_descs to extend or allocate
+ * coalesce buffer. If there is a allocated coalesce buffer, it will
+ * copy the input packet data into the coalesce buffer. It also adds
+ * coalesce buffer descriptor once when whole packet is received.
+ *
+ * Return:
+ * <0 - error
+ * 0 - coalescing, don't populate descriptor
+ * 1 - continue with populating descriptor
+ */
+int ext_coal_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx,
+			   int type, void *kvaddr, struct page *page,
+			   unsigned long offset, u16 len)
+{
+	int pad_len, rval;
+	dma_addr_t addr;
+
+	rval = _extend_sdma_tx_descs(dd, tx);
+	if (rval) {
+		__sdma_txclean(dd, tx);
+		return rval;
+	}
+
+	/* If coalesce buffer is allocated, copy data into it */
+	if (tx->coalesce_buf) {
+		if (type == SDMA_MAP_NONE) {
+			__sdma_txclean(dd, tx);
+			return -EINVAL;
+		}
+
+		if (type == SDMA_MAP_PAGE) {
+			kvaddr = kmap(page);
+			kvaddr += offset;
+		} else if (WARN_ON(!kvaddr)) {
+			__sdma_txclean(dd, tx);
+			return -EINVAL;
+		}
+
+		memcpy(tx->coalesce_buf + tx->coalesce_idx, kvaddr, len);
+		tx->coalesce_idx += len;
+		if (type == SDMA_MAP_PAGE)
+			kunmap(page);
+
+		/* If there is more data, return */
+		if (tx->tlen - tx->coalesce_idx)
+			return 0;
+
+		/* Whole packet is received; add any padding */
+		pad_len = tx->packet_len & (sizeof(u32) - 1);
+		if (pad_len) {
+			pad_len = sizeof(u32) - pad_len;
+			memset(tx->coalesce_buf + tx->coalesce_idx, 0, pad_len);
+			/* padding is taken care of for coalescing case */
+			tx->packet_len += pad_len;
+			tx->tlen += pad_len;
+		}
+
+		/* dma map the coalesce buffer */
+		addr = dma_map_single(&dd->pcidev->dev,
+				      tx->coalesce_buf,
+				      tx->tlen,
+				      DMA_TO_DEVICE);
+
+		if (unlikely(dma_mapping_error(&dd->pcidev->dev, addr))) {
+			__sdma_txclean(dd, tx);
+			return -ENOSPC;
+		}
+
+		/* Add descriptor for coalesce buffer */
+		tx->desc_limit = MAX_DESC;
+		return _sdma_txadd_daddr(dd, SDMA_MAP_SINGLE, tx,
+					 addr, tx->tlen);
+	}
+
+	return 1;
+}
+
+/* Update sdes when the lmc changes */
+void sdma_update_lmc(struct hfi1_devdata *dd, u64 mask, u32 lid)
+{
+	struct sdma_engine *sde;
+	int i;
+	u64 sreg;
+
+	sreg = ((mask & SD(CHECK_SLID_MASK_MASK)) <<
+		SD(CHECK_SLID_MASK_SHIFT)) |
+		(((lid & mask) & SD(CHECK_SLID_VALUE_MASK)) <<
+		SD(CHECK_SLID_VALUE_SHIFT));
+
+	for (i = 0; i < dd->num_sdma; i++) {
+		hfi1_cdbg(LINKVERB, "SendDmaEngine[%d].SLID_CHECK = 0x%x",
+			  i, (u32)sreg);
+		sde = &dd->per_sdma[i];
+		write_sde_csr(sde, SD(CHECK_SLID), sreg);
+	}
+}
+
+/* tx not dword sized - pad */
+int _pad_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx)
+{
+	int rval = 0;
+
+	tx->num_desc++;
+	if ((unlikely(tx->num_desc == tx->desc_limit))) {
+		rval = _extend_sdma_tx_descs(dd, tx);
+		if (rval) {
+			__sdma_txclean(dd, tx);
+			return rval;
+		}
+	}
+	/* finish the one just added */
+	make_tx_sdma_desc(
+		tx,
+		SDMA_MAP_NONE,
+		dd->sdma_pad_phys,
+		sizeof(u32) - (tx->packet_len & (sizeof(u32) - 1)));
+	_sdma_close_tx(dd, tx);
+	return rval;
+}
+
+/*
+ * Add ahg to the sdma_txreq
+ *
+ * The logic will consume up to 3
+ * descriptors at the beginning of
+ * sdma_txreq.
+ */
+void _sdma_txreq_ahgadd(
+	struct sdma_txreq *tx,
+	u8 num_ahg,
+	u8 ahg_entry,
+	u32 *ahg,
+	u8 ahg_hlen)
+{
+	u32 i, shift = 0, desc = 0;
+	u8 mode;
+
+	WARN_ON_ONCE(num_ahg > 9 || (ahg_hlen & 3) || ahg_hlen == 4);
+	/* compute mode */
+	if (num_ahg == 1)
+		mode = SDMA_AHG_APPLY_UPDATE1;
+	else if (num_ahg <= 5)
+		mode = SDMA_AHG_APPLY_UPDATE2;
+	else
+		mode = SDMA_AHG_APPLY_UPDATE3;
+	tx->num_desc++;
+	/* initialize to consumed descriptors to zero */
+	switch (mode) {
+	case SDMA_AHG_APPLY_UPDATE3:
+		tx->num_desc++;
+		tx->descs[2].qw[0] = 0;
+		tx->descs[2].qw[1] = 0;
+		/* FALLTHROUGH */
+	case SDMA_AHG_APPLY_UPDATE2:
+		tx->num_desc++;
+		tx->descs[1].qw[0] = 0;
+		tx->descs[1].qw[1] = 0;
+		break;
+	}
+	ahg_hlen >>= 2;
+	tx->descs[0].qw[1] |=
+		(((u64)ahg_entry & SDMA_DESC1_HEADER_INDEX_MASK)
+			<< SDMA_DESC1_HEADER_INDEX_SHIFT) |
+		(((u64)ahg_hlen & SDMA_DESC1_HEADER_DWS_MASK)
+			<< SDMA_DESC1_HEADER_DWS_SHIFT) |
+		(((u64)mode & SDMA_DESC1_HEADER_MODE_MASK)
+			<< SDMA_DESC1_HEADER_MODE_SHIFT) |
+		(((u64)ahg[0] & SDMA_DESC1_HEADER_UPDATE1_MASK)
+			<< SDMA_DESC1_HEADER_UPDATE1_SHIFT);
+	for (i = 0; i < (num_ahg - 1); i++) {
+		if (!shift && !(i & 2))
+			desc++;
+		tx->descs[desc].qw[!!(i & 2)] |=
+			(((u64)ahg[i + 1])
+				<< shift);
+		shift = (shift + 32) & 63;
+	}
+}
+
+/**
+ * sdma_ahg_alloc - allocate an AHG entry
+ * @sde: engine to allocate from
+ *
+ * Return:
+ * 0-31 when successful, -EOPNOTSUPP if AHG is not enabled,
+ * -ENOSPC if an entry is not available
+ */
+int sdma_ahg_alloc(struct sdma_engine *sde)
+{
+	int nr;
+	int oldbit;
+
+	if (!sde) {
+		trace_hfi1_ahg_allocate(sde, -EINVAL);
+		return -EINVAL;
+	}
+	while (1) {
+		nr = ffz(READ_ONCE(sde->ahg_bits));
+		if (nr > 31) {
+			trace_hfi1_ahg_allocate(sde, -ENOSPC);
+			return -ENOSPC;
+		}
+		oldbit = test_and_set_bit(nr, &sde->ahg_bits);
+		if (!oldbit)
+			break;
+		cpu_relax();
+	}
+	trace_hfi1_ahg_allocate(sde, nr);
+	return nr;
+}
+
+/**
+ * sdma_ahg_free - free an AHG entry
+ * @sde: engine to return AHG entry
+ * @ahg_index: index to free
+ *
+ * This routine frees the indicate AHG entry.
+ */
+void sdma_ahg_free(struct sdma_engine *sde, int ahg_index)
+{
+	if (!sde)
+		return;
+	trace_hfi1_ahg_deallocate(sde, ahg_index);
+	if (ahg_index < 0 || ahg_index > 31)
+		return;
+	clear_bit(ahg_index, &sde->ahg_bits);
+}
+
+/*
+ * SPC freeze handling for SDMA engines.  Called when the driver knows
+ * the SPC is going into a freeze but before the freeze is fully
+ * settled.  Generally an error interrupt.
+ *
+ * This event will pull the engine out of running so no more entries can be
+ * added to the engine's queue.
+ */
+void sdma_freeze_notify(struct hfi1_devdata *dd, int link_down)
+{
+	int i;
+	enum sdma_events event = link_down ? sdma_event_e85_link_down :
+					     sdma_event_e80_hw_freeze;
+
+	/* set up the wait but do not wait here */
+	atomic_set(&dd->sdma_unfreeze_count, dd->num_sdma);
+
+	/* tell all engines to stop running and wait */
+	for (i = 0; i < dd->num_sdma; i++)
+		sdma_process_event(&dd->per_sdma[i], event);
+
+	/* sdma_freeze() will wait for all engines to have stopped */
+}
+
+/*
+ * SPC freeze handling for SDMA engines.  Called when the driver knows
+ * the SPC is fully frozen.
+ */
+void sdma_freeze(struct hfi1_devdata *dd)
+{
+	int i;
+	int ret;
+
+	/*
+	 * Make sure all engines have moved out of the running state before
+	 * continuing.
+	 */
+	ret = wait_event_interruptible(dd->sdma_unfreeze_wq,
+				       atomic_read(&dd->sdma_unfreeze_count) <=
+				       0);
+	/* interrupted or count is negative, then unloading - just exit */
+	if (ret || atomic_read(&dd->sdma_unfreeze_count) < 0)
+		return;
+
+	/* set up the count for the next wait */
+	atomic_set(&dd->sdma_unfreeze_count, dd->num_sdma);
+
+	/* tell all engines that the SPC is frozen, they can start cleaning */
+	for (i = 0; i < dd->num_sdma; i++)
+		sdma_process_event(&dd->per_sdma[i], sdma_event_e81_hw_frozen);
+
+	/*
+	 * Wait for everyone to finish software clean before exiting.  The
+	 * software clean will read engine CSRs, so must be completed before
+	 * the next step, which will clear the engine CSRs.
+	 */
+	(void)wait_event_interruptible(dd->sdma_unfreeze_wq,
+				atomic_read(&dd->sdma_unfreeze_count) <= 0);
+	/* no need to check results - done no matter what */
+}
+
+/*
+ * SPC freeze handling for the SDMA engines.  Called after the SPC is unfrozen.
+ *
+ * The SPC freeze acts like a SDMA halt and a hardware clean combined.  All
+ * that is left is a software clean.  We could do it after the SPC is fully
+ * frozen, but then we'd have to add another state to wait for the unfreeze.
+ * Instead, just defer the software clean until the unfreeze step.
+ */
+void sdma_unfreeze(struct hfi1_devdata *dd)
+{
+	int i;
+
+	/* tell all engines start freeze clean up */
+	for (i = 0; i < dd->num_sdma; i++)
+		sdma_process_event(&dd->per_sdma[i],
+				   sdma_event_e82_hw_unfreeze);
+}
+
+/**
+ * _sdma_engine_progress_schedule() - schedule progress on engine
+ * @sde: sdma_engine to schedule progress
+ *
+ */
+void _sdma_engine_progress_schedule(
+	struct sdma_engine *sde)
+{
+	trace_hfi1_sdma_engine_progress(sde, sde->progress_mask);
+	/* assume we have selected a good cpu */
+	write_csr(sde->dd,
+		  CCE_INT_FORCE + (8 * (IS_SDMA_START / 64)),
+		  sde->progress_mask);
+}