This tag contains the following changes for kernel 5.9-rc1:

- Remove rate limiters from GAUDI configuration (no longer needed).
 - Set maximum amount of in-flight CS per ASIC type and increase
   the maximum amount for GAUDI.
 - Refactor signal/wait command submissions code
 - Calculate trace frequency from PLLs to show accurate profiling data
 - Rephrase error messages to make them more clear to the common user
 - Add statistics of dropped CS (counter per possible reason for drop)
 - Get ECC information from firmware
 - Remove support for partial SoC reset in Gaudi
 - Halt device CPU only when reset is certain to happen. Sometimes we abort
   the reset procedure and in that case we can't leave device CPU in halt
   mode.
 - set each CQ to its own work queue to prevent a race between completions
   on different CQs.
 - Use queue pi/ci in order to determine queue occupancy. This is done to
   make the code reusable between current and future ASICs.
 - Add more validations for user inputs.
 - Refactor PCIe controller configuration to make the code reusable between
   current and future ASICs.
 - Update firmware interface headers to latest version
 - Move all common code to a dedicated common sub-folder
 -----BEGIN PGP SIGNATURE-----
 
 iQFKBAABCgA0FiEE7TEboABC71LctBLFZR1NuKta54AFAl8bHekWHG9kZWQuZ2Fi
 YmF5QGdtYWlsLmNvbQAKCRBlHU24q1rngCwsCACGVS+NVbDO2E42TOpTxRst26Nu
 7V8s8HcICbdhfHbkJWnexsbPkjzRUvi1pXfklmafGtGtoTunwpb1AXx+oWeMqmnT
 8IoxVlp0A5doP9uh8WOdKPypDNWem8fL+89y+89C8/ImAHMeeC4X+v2hiIIoRBDu
 aoO3zgxPjLmNfcUGfUqcaLcar8w/EhVY/JIPjSc21PXouyUer7Jx9oe2MS1OkQ0h
 FIYBRW4oycEqZSYCs9SLog/ltBt84Kzk/TsWweTcW3V3EpXyhFjq5yG9nBmmNB9z
 U9wmJV4tpGOg/ehvy3MWMbCtv4xK/xS3R3pn25IJw79X2RWUKizBorgYlXu/
 =ZOVe
 -----END PGP SIGNATURE-----

Merge tag 'misc-habanalabs-next-2020-07-24' of git://people.freedesktop.org/~gabbayo/linux into char-misc-next

Oded writes:

This tag contains the following changes for kernel 5.9-rc1:

- Remove rate limiters from GAUDI configuration (no longer needed).
- Set maximum amount of in-flight CS per ASIC type and increase
  the maximum amount for GAUDI.
- Refactor signal/wait command submissions code
- Calculate trace frequency from PLLs to show accurate profiling data
- Rephrase error messages to make them more clear to the common user
- Add statistics of dropped CS (counter per possible reason for drop)
- Get ECC information from firmware
- Remove support for partial SoC reset in Gaudi
- Halt device CPU only when reset is certain to happen. Sometimes we abort
  the reset procedure and in that case we can't leave device CPU in halt
  mode.
- set each CQ to its own work queue to prevent a race between completions
  on different CQs.
- Use queue pi/ci in order to determine queue occupancy. This is done to
  make the code reusable between current and future ASICs.
- Add more validations for user inputs.
- Refactor PCIe controller configuration to make the code reusable between
  current and future ASICs.
- Update firmware interface headers to latest version
- Move all common code to a dedicated common sub-folder

* tag 'misc-habanalabs-next-2020-07-24' of git://people.freedesktop.org/~gabbayo/linux: (28 commits)
  habanalabs: Fix memory leak in error flow of context initialization
  habanalabs: use no flags on MMU cache invalidation
  habanalabs: enable device before hw_init()
  habanalabs: create internal CB pool
  habanalabs: update hl_boot_if.h from firmware
  habanalabs: create common folder
  habanalabs: check for DMA errors when clearing memory
  habanalabs: verify queue can contain all cs jobs
  habanalabs: Assign each CQ with its own work queue
  habanalabs: halt device CPU only upon certain reset
  habanalabs: remove unused hash
  habanalabs: use queue pi/ci in order to determine queue occupancy
  habanalabs: configure maximum queues per asic
  habanalabs: remove soft-reset support from GAUDI
  habanalabs: PCIe iATU refactoring
  habanalabs: Extract ECC information from FW
  habanalabs: Add dropped cs statistics info struct
  habanalabs: extract cpu boot status lookup
  habanalabs: rephrase error messages
  habanalabs: Increase queues depth
  ...

drivers/misc/habanalabs/Makefile

@@ -3,16 +3,15 @@
 # Makefile for HabanaLabs AI accelerators driver
 #
 
-obj-m	:= habanalabs.o
+obj-$(CONFIG_HABANA_AI) := habanalabs.o
 
-habanalabs-y := habanalabs_drv.o device.o context.o asid.o habanalabs_ioctl.o \
+include $(src)/common/Makefile
-		command_buffer.o hw_queue.o irq.o sysfs.o hwmon.o memory.o \
+habanalabs-y += $(HL_COMMON_FILES)
-		command_submission.o mmu.o firmware_if.o pci.o
-
-habanalabs-$(CONFIG_DEBUG_FS) += debugfs.o
 
 include $(src)/goya/Makefile
 habanalabs-y += $(HL_GOYA_FILES)
 
 include $(src)/gaudi/Makefile
 habanalabs-y += $(HL_GAUDI_FILES)
+
+habanalabs-$(CONFIG_DEBUG_FS) += common/debugfs.o

drivers/misc/habanalabs/common/Makefile

@@ -0,0 +1,9 @@
+# SPDX-License-Identifier: GPL-2.0-only
+subdir-ccflags-y += -I$(src)/common
+
+HL_COMMON_FILES := common/habanalabs_drv.o common/device.o common/context.o \
+		common/asid.o common/habanalabs_ioctl.o \
+		common/command_buffer.o common/hw_queue.o common/irq.o \
+		common/sysfs.o common/hwmon.o common/memory.o \
+		common/command_submission.o common/mmu.o common/firmware_if.o \
+		common/pci.o

drivers/misc/habanalabs/common/command_buffer.c

@@ -10,12 +10,18 @@
 
 #include <linux/mm.h>
 #include <linux/slab.h>
+#include <linux/genalloc.h>
 
 static void cb_fini(struct hl_device *hdev, struct hl_cb *cb)
 {
+	if (cb->is_internal)
+		gen_pool_free(hdev->internal_cb_pool,
+				cb->kernel_address, cb->size);
+	else
 		hdev->asic_funcs->asic_dma_free_coherent(hdev, cb->size,
 				(void *) (uintptr_t) cb->kernel_address,
 				cb->bus_address);
+
 	kfree(cb);
 }
 
@@ -44,9 +50,10 @@ static void cb_release(struct kref *ref)
 }
 
 static struct hl_cb *hl_cb_alloc(struct hl_device *hdev, u32 cb_size,
-					int ctx_id)
+					int ctx_id, bool internal_cb)
 {
 	struct hl_cb *cb;
+	u32 cb_offset;
 	void *p;
 
 	/*
@@ -65,13 +72,25 @@ static struct hl_cb *hl_cb_alloc(struct hl_device *hdev, u32 cb_size,
 	if (!cb)
 		return NULL;
 
-	if (ctx_id == HL_KERNEL_ASID_ID)
+	if (internal_cb) {
+		p = (void *) gen_pool_alloc(hdev->internal_cb_pool, cb_size);
+		if (!p) {
+			kfree(cb);
+			return NULL;
+		}
+
+		cb_offset = p - hdev->internal_cb_pool_virt_addr;
+		cb->is_internal = true;
+		cb->bus_address =  hdev->internal_cb_va_base + cb_offset;
+	} else if (ctx_id == HL_KERNEL_ASID_ID) {
 		p = hdev->asic_funcs->asic_dma_alloc_coherent(hdev, cb_size,
 						&cb->bus_address, GFP_ATOMIC);
-	else
+	} else {
 		p = hdev->asic_funcs->asic_dma_alloc_coherent(hdev, cb_size,
 						&cb->bus_address,
 						GFP_USER | __GFP_ZERO);
+	}
+
 	if (!p) {
 		dev_err(hdev->dev,
 			"failed to allocate %d of dma memory for CB\n",
@@ -87,7 +106,7 @@ static struct hl_cb *hl_cb_alloc(struct hl_device *hdev, u32 cb_size,
 }
 
 int hl_cb_create(struct hl_device *hdev, struct hl_cb_mgr *mgr,
-			u32 cb_size, u64 *handle, int ctx_id)
+			u32 cb_size, u64 *handle, int ctx_id, bool internal_cb)
 {
 	struct hl_cb *cb;
 	bool alloc_new_cb = true;
@@ -112,6 +131,7 @@ int hl_cb_create(struct hl_device *hdev, struct hl_cb_mgr *mgr,
 		goto out_err;
 	}
 
+	if (!internal_cb) {
 		/* Minimum allocation must be PAGE SIZE */
 		if (cb_size < PAGE_SIZE)
 			cb_size = PAGE_SIZE;
@@ -121,8 +141,8 @@ int hl_cb_create(struct hl_device *hdev, struct hl_cb_mgr *mgr,
 
 			spin_lock(&hdev->cb_pool_lock);
 			if (!list_empty(&hdev->cb_pool)) {
-			cb = list_first_entry(&hdev->cb_pool, typeof(*cb),
+				cb = list_first_entry(&hdev->cb_pool,
-					pool_list);
+						typeof(*cb), pool_list);
 				list_del(&cb->pool_list);
 				spin_unlock(&hdev->cb_pool_lock);
 				alloc_new_cb = false;
@@ -131,9 +151,10 @@ int hl_cb_create(struct hl_device *hdev, struct hl_cb_mgr *mgr,
 				dev_dbg(hdev->dev, "CB pool is empty\n");
 			}
 		}
+	}
 
 	if (alloc_new_cb) {
-		cb = hl_cb_alloc(hdev, cb_size, ctx_id);
+		cb = hl_cb_alloc(hdev, cb_size, ctx_id, internal_cb);
 		if (!cb) {
 			rc = -ENOMEM;
 			goto out_err;
@@ -230,7 +251,7 @@ int hl_cb_ioctl(struct hl_fpriv *hpriv, void *data)
 		} else {
 			rc = hl_cb_create(hdev, &hpriv->cb_mgr,
 					args->in.cb_size, &handle,
-						hpriv->ctx->asid);
+					hpriv->ctx->asid, false);
 		}
 
 		memset(args, 0, sizeof(*args));
@@ -398,14 +419,15 @@ void hl_cb_mgr_fini(struct hl_device *hdev, struct hl_cb_mgr *mgr)
 	idr_destroy(&mgr->cb_handles);
 }
 
-struct hl_cb *hl_cb_kernel_create(struct hl_device *hdev, u32 cb_size)
+struct hl_cb *hl_cb_kernel_create(struct hl_device *hdev, u32 cb_size,
+					bool internal_cb)
 {
 	u64 cb_handle;
 	struct hl_cb *cb;
 	int rc;
 
 	rc = hl_cb_create(hdev, &hdev->kernel_cb_mgr, cb_size, &cb_handle,
-			HL_KERNEL_ASID_ID);
+			HL_KERNEL_ASID_ID, internal_cb);
 	if (rc) {
 		dev_err(hdev->dev,
 			"Failed to allocate CB for the kernel driver %d\n", rc);
@@ -437,7 +459,7 @@ int hl_cb_pool_init(struct hl_device *hdev)
 
 	for (i = 0 ; i < hdev->asic_prop.cb_pool_cb_cnt ; i++) {
 		cb = hl_cb_alloc(hdev, hdev->asic_prop.cb_pool_cb_size,
-				HL_KERNEL_ASID_ID);
+				HL_KERNEL_ASID_ID, false);
 		if (cb) {
 			cb->is_pool = true;
 			list_add(&cb->pool_list, &hdev->cb_pool);

drivers/misc/habanalabs/common/command_submission.c

@@ -246,6 +246,18 @@ static void free_job(struct hl_device *hdev, struct hl_cs_job *job)
 	kfree(job);
 }
 
+static void cs_counters_aggregate(struct hl_device *hdev, struct hl_ctx *ctx)
+{
+	hdev->aggregated_cs_counters.device_in_reset_drop_cnt +=
+			ctx->cs_counters.device_in_reset_drop_cnt;
+	hdev->aggregated_cs_counters.out_of_mem_drop_cnt +=
+			ctx->cs_counters.out_of_mem_drop_cnt;
+	hdev->aggregated_cs_counters.parsing_drop_cnt +=
+			ctx->cs_counters.parsing_drop_cnt;
+	hdev->aggregated_cs_counters.queue_full_drop_cnt +=
+			ctx->cs_counters.queue_full_drop_cnt;
+}
+
 static void cs_do_release(struct kref *ref)
 {
 	struct hl_cs *cs = container_of(ref, struct hl_cs,
@@ -349,6 +361,9 @@ static void cs_do_release(struct kref *ref)
 	dma_fence_signal(cs->fence);
 	dma_fence_put(cs->fence);
 
+	cs_counters_aggregate(hdev, cs->ctx);
+
+	kfree(cs->jobs_in_queue_cnt);
 	kfree(cs);
 }
 
@@ -373,9 +388,9 @@ static void cs_timedout(struct work_struct *work)
 	hdev = cs->ctx->hdev;
 	ctx_asid = cs->ctx->asid;
 
-	/* TODO: add information about last signaled seq and last emitted seq */
+	dev_err(hdev->dev,
-	dev_err(hdev->dev, "User %d command submission %llu got stuck!\n",
+		"Command submission %llu has not finished in time!\n",
-		ctx_asid, cs->sequence);
+		cs->sequence);
 
 	cs_put(cs);
 
@@ -418,21 +433,29 @@ static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx,
 	spin_lock(&ctx->cs_lock);
 
 	cs_cmpl->cs_seq = ctx->cs_sequence;
-	other = ctx->cs_pending[cs_cmpl->cs_seq & (HL_MAX_PENDING_CS - 1)];
+	other = ctx->cs_pending[cs_cmpl->cs_seq &
+				(hdev->asic_prop.max_pending_cs - 1)];
 	if ((other) && (!dma_fence_is_signaled(other))) {
-		spin_unlock(&ctx->cs_lock);
 		dev_dbg(hdev->dev,
 			"Rejecting CS because of too many in-flights CS\n");
 		rc = -EAGAIN;
 		goto free_fence;
 	}
 
+	cs->jobs_in_queue_cnt = kcalloc(hdev->asic_prop.max_queues,
+			sizeof(*cs->jobs_in_queue_cnt), GFP_ATOMIC);
+	if (!cs->jobs_in_queue_cnt) {
+		rc = -ENOMEM;
+		goto free_fence;
+	}
+
 	dma_fence_init(&cs_cmpl->base_fence, &hl_fence_ops, &cs_cmpl->lock,
 			ctx->asid, ctx->cs_sequence);
 
 	cs->sequence = cs_cmpl->cs_seq;
 
-	ctx->cs_pending[cs_cmpl->cs_seq & (HL_MAX_PENDING_CS - 1)] =
+	ctx->cs_pending[cs_cmpl->cs_seq &
+			(hdev->asic_prop.max_pending_cs - 1)] =
 							&cs_cmpl->base_fence;
 	ctx->cs_sequence++;
 
@@ -447,6 +470,7 @@ static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx,
 	return 0;
 
 free_fence:
+	spin_unlock(&ctx->cs_lock);
 	kfree(cs_cmpl);
 free_cs:
 	kfree(cs);
@@ -463,10 +487,12 @@ static void cs_rollback(struct hl_device *hdev, struct hl_cs *cs)
 
 void hl_cs_rollback_all(struct hl_device *hdev)
 {
+	int i;
 	struct hl_cs *cs, *tmp;
 
 	/* flush all completions */
-	flush_workqueue(hdev->cq_wq);
+	for (i = 0 ; i < hdev->asic_prop.completion_queues_count ; i++)
+		flush_workqueue(hdev->cq_wq[i]);
 
 	/* Make sure we don't have leftovers in the H/W queues mirror list */
 	list_for_each_entry_safe(cs, tmp, &hdev->hw_queues_mirror_list,
@@ -499,10 +525,18 @@ static int validate_queue_index(struct hl_device *hdev,
 	struct asic_fixed_properties *asic = &hdev->asic_prop;
 	struct hw_queue_properties *hw_queue_prop;
 
+	/* This must be checked here to prevent out-of-bounds access to
+	 * hw_queues_props array
+	 */
+	if (chunk->queue_index >= asic->max_queues) {
+		dev_err(hdev->dev, "Queue index %d is invalid\n",
+			chunk->queue_index);
+		return -EINVAL;
+	}
+
 	hw_queue_prop = &asic->hw_queues_props[chunk->queue_index];
 
-	if ((chunk->queue_index >= HL_MAX_QUEUES) ||
+	if (hw_queue_prop->type == QUEUE_TYPE_NA) {
-			(hw_queue_prop->type == QUEUE_TYPE_NA)) {
 		dev_err(hdev->dev, "Queue index %d is invalid\n",
 			chunk->queue_index);
 		return -EINVAL;
@@ -630,12 +664,15 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks,
 
 		rc = validate_queue_index(hdev, chunk, &queue_type,
 						&is_kernel_allocated_cb);
-		if (rc)
+		if (rc) {
+			hpriv->ctx->cs_counters.parsing_drop_cnt++;
 			goto free_cs_object;
+		}
 
 		if (is_kernel_allocated_cb) {
 			cb = get_cb_from_cs_chunk(hdev, &hpriv->cb_mgr, chunk);
 			if (!cb) {
+				hpriv->ctx->cs_counters.parsing_drop_cnt++;
 				rc = -EINVAL;
 				goto free_cs_object;
 			}
@@ -649,6 +686,7 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks,
 		job = hl_cs_allocate_job(hdev, queue_type,
 						is_kernel_allocated_cb);
 		if (!job) {
+			hpriv->ctx->cs_counters.out_of_mem_drop_cnt++;
 			dev_err(hdev->dev, "Failed to allocate a new job\n");
 			rc = -ENOMEM;
 			if (is_kernel_allocated_cb)
@@ -681,6 +719,7 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks,
 
 		rc = cs_parser(hpriv, job);
 		if (rc) {
+			hpriv->ctx->cs_counters.parsing_drop_cnt++;
 			dev_err(hdev->dev,
 				"Failed to parse JOB %d.%llu.%d, err %d, rejecting the CS\n",
 				cs->ctx->asid, cs->sequence, job->id, rc);
@@ -689,6 +728,7 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks,
 	}
 
 	if (int_queues_only) {
+		hpriv->ctx->cs_counters.parsing_drop_cnt++;
 		dev_err(hdev->dev,
 			"Reject CS %d.%llu because only internal queues jobs are present\n",
 			cs->ctx->asid, cs->sequence);
@@ -738,6 +778,7 @@ static int cs_ioctl_signal_wait(struct hl_fpriv *hpriv, enum hl_cs_type cs_type,
 	struct hl_cs_job *job;
 	struct hl_cs *cs;
 	struct hl_cb *cb;
+	enum hl_queue_type q_type;
 	u64 *signal_seq_arr = NULL, signal_seq;
 	u32 size_to_copy, q_idx, signal_seq_arr_len, cb_size;
 	int rc;
@@ -770,9 +811,10 @@ static int cs_ioctl_signal_wait(struct hl_fpriv *hpriv, enum hl_cs_type cs_type,
 	chunk = &cs_chunk_array[0];
 	q_idx = chunk->queue_index;
 	hw_queue_prop = &hdev->asic_prop.hw_queues_props[q_idx];
+	q_type = hw_queue_prop->type;
 
-	if ((q_idx >= HL_MAX_QUEUES) ||
+	if ((q_idx >= hdev->asic_prop.max_queues) ||
-			(hw_queue_prop->type != QUEUE_TYPE_EXT)) {
+			(!hw_queue_prop->supports_sync_stream)) {
 		dev_err(hdev->dev, "Queue index %d is invalid\n", q_idx);
 		rc = -EINVAL;
 		goto free_cs_chunk_array;
@@ -869,25 +911,28 @@ static int cs_ioctl_signal_wait(struct hl_fpriv *hpriv, enum hl_cs_type cs_type,
 
 	*cs_seq = cs->sequence;
 
-	job = hl_cs_allocate_job(hdev, QUEUE_TYPE_EXT, true);
+	job = hl_cs_allocate_job(hdev, q_type, true);
 	if (!job) {
+		ctx->cs_counters.out_of_mem_drop_cnt++;
 		dev_err(hdev->dev, "Failed to allocate a new job\n");
 		rc = -ENOMEM;
 		goto put_cs;
 	}
 
-	cb = hl_cb_kernel_create(hdev, PAGE_SIZE);
-	if (!cb) {
-		kfree(job);
-		rc = -EFAULT;
-		goto put_cs;
-	}
-
 	if (cs->type == CS_TYPE_WAIT)
 		cb_size = hdev->asic_funcs->get_wait_cb_size(hdev);
 	else
 		cb_size = hdev->asic_funcs->get_signal_cb_size(hdev);
 
+	cb = hl_cb_kernel_create(hdev, cb_size,
+				q_type == QUEUE_TYPE_HW && hdev->mmu_enable);
+	if (!cb) {
+		ctx->cs_counters.out_of_mem_drop_cnt++;
+		kfree(job);
+		rc = -EFAULT;
+		goto put_cs;
+	}
+
 	job->id = 0;
 	job->cs = cs;
 	job->user_cb = cb;
@@ -1126,7 +1171,7 @@ static long _hl_cs_wait_ioctl(struct hl_device *hdev,
 		rc = PTR_ERR(fence);
 		if (rc == -EINVAL)
 			dev_notice_ratelimited(hdev->dev,
-				"Can't wait on seq %llu because current CS is at seq %llu\n",
+				"Can't wait on CS %llu because current CS is at seq %llu\n",
 				seq, ctx->cs_sequence);
 	} else if (fence) {
 		rc = dma_fence_wait_timeout(fence, true, timeout);
@@ -1159,15 +1204,21 @@ int hl_cs_wait_ioctl(struct hl_fpriv *hpriv, void *data)
 	memset(args, 0, sizeof(*args));
 
 	if (rc < 0) {
-		dev_err_ratelimited(hdev->dev,
-				"Error %ld on waiting for CS handle %llu\n",
-				rc, seq);
 		if (rc == -ERESTARTSYS) {
+			dev_err_ratelimited(hdev->dev,
+				"user process got signal while waiting for CS handle %llu\n",
+				seq);
 			args->out.status = HL_WAIT_CS_STATUS_INTERRUPTED;
 			rc = -EINTR;
 		} else if (rc == -ETIMEDOUT) {
+			dev_err_ratelimited(hdev->dev,
+				"CS %llu has timed-out while user process is waiting for it\n",
+				seq);
 			args->out.status = HL_WAIT_CS_STATUS_TIMEDOUT;
 		} else if (rc == -EIO) {
+			dev_err_ratelimited(hdev->dev,
+				"CS %llu has been aborted while user process is waiting for it\n",
+				seq);
 			args->out.status = HL_WAIT_CS_STATUS_ABORTED;
 		}
 		return rc;

drivers/misc/habanalabs/common/context.c

@@ -22,9 +22,11 @@ static void hl_ctx_fini(struct hl_ctx *ctx)
 	 * to this function unless the ref count is 0
 	 */
 
-	for (i = 0 ; i < HL_MAX_PENDING_CS ; i++)
+	for (i = 0 ; i < hdev->asic_prop.max_pending_cs ; i++)
 		dma_fence_put(ctx->cs_pending[i]);
 
+	kfree(ctx->cs_pending);
+
 	if (ctx->asid != HL_KERNEL_ASID_ID) {
 		/* The engines are stopped as there is no executing CS, but the
 		 * Coresight might be still working by accessing addresses
@@ -110,8 +112,7 @@ void hl_ctx_free(struct hl_device *hdev, struct hl_ctx *ctx)
 		return;
 
 	dev_warn(hdev->dev,
-		"Context %d closed or terminated but its CS are executing\n",
+		"user process released device but its command submissions are still executing\n");
-		ctx->asid);
 }
 
 int hl_ctx_init(struct hl_device *hdev, struct hl_ctx *ctx, bool is_kernel_ctx)
@@ -126,34 +127,49 @@ int hl_ctx_init(struct hl_device *hdev, struct hl_ctx *ctx, bool is_kernel_ctx)
 	spin_lock_init(&ctx->cs_lock);
 	atomic_set(&ctx->thread_ctx_switch_token, 1);
 	ctx->thread_ctx_switch_wait_token = 0;
+	ctx->cs_pending = kcalloc(hdev->asic_prop.max_pending_cs,
+				sizeof(struct dma_fence *),
+				GFP_KERNEL);
+	if (!ctx->cs_pending)
+		return -ENOMEM;
 
 	if (is_kernel_ctx) {
 		ctx->asid = HL_KERNEL_ASID_ID; /* Kernel driver gets ASID 0 */
 		rc = hl_mmu_ctx_init(ctx);
 		if (rc) {
 			dev_err(hdev->dev, "Failed to init mmu ctx module\n");
-			goto mem_ctx_err;
+			goto err_free_cs_pending;
 		}
 	} else {
 		ctx->asid = hl_asid_alloc(hdev);
 		if (!ctx->asid) {
 			dev_err(hdev->dev, "No free ASID, failed to create context\n");
-			return -ENOMEM;
+			rc = -ENOMEM;
+			goto err_free_cs_pending;
 		}
 
 		rc = hl_vm_ctx_init(ctx);
 		if (rc) {
 			dev_err(hdev->dev, "Failed to init mem ctx module\n");
 			rc = -ENOMEM;
-			goto mem_ctx_err;
+			goto err_asid_free;
+		}
+
+		rc = hdev->asic_funcs->ctx_init(ctx);
+		if (rc) {
+			dev_err(hdev->dev, "ctx_init failed\n");
+			goto err_vm_ctx_fini;
 		}
 	}
 
 	return 0;
 
-mem_ctx_err:
+err_vm_ctx_fini:
-	if (ctx->asid != HL_KERNEL_ASID_ID)
+	hl_vm_ctx_fini(ctx);
+err_asid_free:
 	hl_asid_free(hdev, ctx->asid);
+err_free_cs_pending:
+	kfree(ctx->cs_pending);
 
 	return rc;
 }
@@ -170,6 +186,7 @@ int hl_ctx_put(struct hl_ctx *ctx)
 
 struct dma_fence *hl_ctx_get_fence(struct hl_ctx *ctx, u64 seq)
 {
+	struct asic_fixed_properties *asic_prop = &ctx->hdev->asic_prop;
 	struct dma_fence *fence;
 
 	spin_lock(&ctx->cs_lock);
@@ -179,13 +196,13 @@ struct dma_fence *hl_ctx_get_fence(struct hl_ctx *ctx, u64 seq)
 		return ERR_PTR(-EINVAL);
 	}
 
-	if (seq + HL_MAX_PENDING_CS < ctx->cs_sequence) {
+	if (seq + asic_prop->max_pending_cs < ctx->cs_sequence) {
 		spin_unlock(&ctx->cs_lock);
 		return NULL;
 	}
 
 	fence = dma_fence_get(
-			ctx->cs_pending[seq & (HL_MAX_PENDING_CS - 1)]);
+			ctx->cs_pending[seq & (asic_prop->max_pending_cs - 1)]);
 	spin_unlock(&ctx->cs_lock);
 
 	return fence;

drivers/misc/habanalabs/common/device.c

@@ -249,7 +249,8 @@ static void device_cdev_sysfs_del(struct hl_device *hdev)
  */
 static int device_early_init(struct hl_device *hdev)
 {
-	int rc;
+	int i, rc;
+	char workq_name[32];
 
 	switch (hdev->asic_type) {
 	case ASIC_GOYA:
@@ -274,11 +275,24 @@ static int device_early_init(struct hl_device *hdev)
 	if (rc)
 		goto early_fini;
 
-	hdev->cq_wq = alloc_workqueue("hl-free-jobs", WQ_UNBOUND, 0);
+	if (hdev->asic_prop.completion_queues_count) {
+		hdev->cq_wq = kcalloc(hdev->asic_prop.completion_queues_count,
+				sizeof(*hdev->cq_wq),
+				GFP_ATOMIC);
+		if (!hdev->cq_wq) {
+			rc = -ENOMEM;
+			goto asid_fini;
+		}
+	}
+
+	for (i = 0 ; i < hdev->asic_prop.completion_queues_count ; i++) {
+		snprintf(workq_name, 32, "hl-free-jobs-%u", i);
+		hdev->cq_wq[i] = create_singlethread_workqueue(workq_name);
 		if (hdev->cq_wq == NULL) {
 			dev_err(hdev->dev, "Failed to allocate CQ workqueue\n");
 			rc = -ENOMEM;
-		goto asid_fini;
+			goto free_cq_wq;
+		}
 	}
 
 	hdev->eq_wq = alloc_workqueue("hl-events", WQ_UNBOUND, 0);
@@ -321,7 +335,10 @@ free_chip_info:
 free_eq_wq:
 	destroy_workqueue(hdev->eq_wq);
 free_cq_wq:
-	destroy_workqueue(hdev->cq_wq);
+	for (i = 0 ; i < hdev->asic_prop.completion_queues_count ; i++)
+		if (hdev->cq_wq[i])
+			destroy_workqueue(hdev->cq_wq[i]);
+	kfree(hdev->cq_wq);
 asid_fini:
 	hl_asid_fini(hdev);
 early_fini:
@@ -339,6 +356,8 @@ early_fini:
  */
 static void device_early_fini(struct hl_device *hdev)
 {
+	int i;
+
 	mutex_destroy(&hdev->mmu_cache_lock);
 	mutex_destroy(&hdev->debug_lock);
 	mutex_destroy(&hdev->send_cpu_message_lock);
@@ -351,7 +370,10 @@ static void device_early_fini(struct hl_device *hdev)
 	kfree(hdev->hl_chip_info);
 
 	destroy_workqueue(hdev->eq_wq);
-	destroy_workqueue(hdev->cq_wq);
+
+	for (i = 0 ; i < hdev->asic_prop.completion_queues_count ; i++)
+		destroy_workqueue(hdev->cq_wq[i]);
+	kfree(hdev->cq_wq);
 
 	hl_asid_fini(hdev);
 
@@ -838,6 +860,22 @@ int hl_device_reset(struct hl_device *hdev, bool hard_reset,
 		if (rc)
 			return 0;
 
+		if (hard_reset) {
+			/* Disable PCI access from device F/W so he won't send
+			 * us additional interrupts. We disable MSI/MSI-X at
+			 * the halt_engines function and we can't have the F/W
+			 * sending us interrupts after that. We need to disable
+			 * the access here because if the device is marked
+			 * disable, the message won't be send. Also, in case
+			 * of heartbeat, the device CPU is marked as disable
+			 * so this message won't be sent
+			 */
+			if (hl_fw_send_pci_access_msg(hdev,
+					ARMCP_PACKET_DISABLE_PCI_ACCESS))
+				dev_warn(hdev->dev,
+					"Failed to disable PCI access by F/W\n");
+		}
+
 		/* This also blocks future CS/VM/JOB completion operations */
 		hdev->disabled = true;
 
@@ -995,6 +1033,12 @@ again:
 		}
 	}
 
+	/* Device is now enabled as part of the initialization requires
+	 * communication with the device firmware to get information that
+	 * is required for the initialization itself
+	 */
+	hdev->disabled = false;
+
 	rc = hdev->asic_funcs->hw_init(hdev);
 	if (rc) {
 		dev_err(hdev->dev,
@@ -1002,8 +1046,6 @@ again:
 		goto out_err;
 	}
 
-	hdev->disabled = false;
-
 	/* Check that the communication with the device is working */
 	rc = hdev->asic_funcs->test_queues(hdev);
 	if (rc) {
@@ -1144,15 +1186,18 @@ int hl_device_init(struct hl_device *hdev, struct class *hclass)
 	 * because there the addresses of the completion queues are being
 	 * passed as arguments to request_irq
 	 */
+	if (cq_cnt) {
 		hdev->completion_queue = kcalloc(cq_cnt,
 				sizeof(*hdev->completion_queue),
 				GFP_KERNEL);
 
 		if (!hdev->completion_queue) {
-		dev_err(hdev->dev, "failed to allocate completion queues\n");
+			dev_err(hdev->dev,
+				"failed to allocate completion queues\n");
 			rc = -ENOMEM;
 			goto hw_queues_destroy;
 		}
+	}
 
 	for (i = 0, cq_ready_cnt = 0 ; i < cq_cnt ; i++, cq_ready_cnt++) {
 		rc = hl_cq_init(hdev, &hdev->completion_queue[i],
@@ -1162,6 +1207,7 @@ int hl_device_init(struct hl_device *hdev, struct class *hclass)
 				"failed to initialize completion queue\n");
 			goto cq_fini;
 		}
+		hdev->completion_queue[i].cq_idx = i;
 	}
 
 	/*
@@ -1219,6 +1265,12 @@ int hl_device_init(struct hl_device *hdev, struct class *hclass)
 	 */
 	add_cdev_sysfs_on_err = true;
 
+	/* Device is now enabled as part of the initialization requires
+	 * communication with the device firmware to get information that
+	 * is required for the initialization itself
+	 */
+	hdev->disabled = false;
+
 	rc = hdev->asic_funcs->hw_init(hdev);
 	if (rc) {
 		dev_err(hdev->dev, "failed to initialize the H/W\n");
@@ -1226,8 +1278,6 @@ int hl_device_init(struct hl_device *hdev, struct class *hclass)
 		goto out_disabled;
 	}
 
-	hdev->disabled = false;
-
 	/* Check that the communication with the device is working */
 	rc = hdev->asic_funcs->test_queues(hdev);
 	if (rc) {

drivers/misc/habanalabs/common/firmware_if.c

@@ -6,7 +6,7 @@
  */
 
 #include "habanalabs.h"
-#include "include/hl_boot_if.h"
+#include "include/common/hl_boot_if.h"
 
 #include <linux/firmware.h>
 #include <linux/genalloc.h>
@@ -289,7 +289,7 @@ int hl_fw_armcp_info_get(struct hl_device *hdev)
 					HL_ARMCP_INFO_TIMEOUT_USEC, &result);
 	if (rc) {
 		dev_err(hdev->dev,
-			"Failed to send ArmCP info pkt, error %d\n", rc);
+			"Failed to handle ArmCP info pkt, error %d\n", rc);
 		goto out;
 	}
 
@@ -340,7 +340,7 @@ int hl_fw_get_eeprom_data(struct hl_device *hdev, void *data, size_t max_size)
 
 	if (rc) {
 		dev_err(hdev->dev,
-			"Failed to send ArmCP EEPROM packet, error %d\n", rc);
+			"Failed to handle ArmCP EEPROM packet, error %d\n", rc);
 		goto out;
 	}
 
@@ -393,6 +393,53 @@ static void fw_read_errors(struct hl_device *hdev, u32 boot_err0_reg)
 			"Device boot error - NIC F/W initialization failed\n");
 }
 
+static void hl_detect_cpu_boot_status(struct hl_device *hdev, u32 status)
+{
+	switch (status) {
+	case CPU_BOOT_STATUS_NA:
+		dev_err(hdev->dev,
+			"Device boot error - BTL did NOT run\n");
+		break;
+	case CPU_BOOT_STATUS_IN_WFE:
+		dev_err(hdev->dev,
+			"Device boot error - Stuck inside WFE loop\n");
+		break;
+	case CPU_BOOT_STATUS_IN_BTL:
+		dev_err(hdev->dev,
+			"Device boot error - Stuck in BTL\n");
+		break;
+	case CPU_BOOT_STATUS_IN_PREBOOT:
+		dev_err(hdev->dev,
+			"Device boot error - Stuck in Preboot\n");
+		break;
+	case CPU_BOOT_STATUS_IN_SPL:
+		dev_err(hdev->dev,
+			"Device boot error - Stuck in SPL\n");
+		break;
+	case CPU_BOOT_STATUS_IN_UBOOT:
+		dev_err(hdev->dev,
+			"Device boot error - Stuck in u-boot\n");
+		break;
+	case CPU_BOOT_STATUS_DRAM_INIT_FAIL:
+		dev_err(hdev->dev,
+			"Device boot error - DRAM initialization failed\n");
+		break;
+	case CPU_BOOT_STATUS_UBOOT_NOT_READY:
+		dev_err(hdev->dev,
+			"Device boot error - u-boot stopped by user\n");
+		break;
+	case CPU_BOOT_STATUS_TS_INIT_FAIL:
+		dev_err(hdev->dev,
+			"Device boot error - Thermal Sensor initialization failed\n");
+		break;
+	default:
+		dev_err(hdev->dev,
+			"Device boot error - Invalid status code %d\n",
+			status);
+		break;
+	}
+}
+
 int hl_fw_init_cpu(struct hl_device *hdev, u32 cpu_boot_status_reg,
 			u32 msg_to_cpu_reg, u32 cpu_msg_status_reg,
 			u32 boot_err0_reg, bool skip_bmc,
@@ -466,50 +513,7 @@ int hl_fw_init_cpu(struct hl_device *hdev, u32 cpu_boot_status_reg,
 	 * versions but we keep them here for backward compatibility
 	 */
 	if (rc) {
-		switch (status) {
+		hl_detect_cpu_boot_status(hdev, status);
-		case CPU_BOOT_STATUS_NA:
-			dev_err(hdev->dev,
-				"Device boot error - BTL did NOT run\n");
-			break;
-		case CPU_BOOT_STATUS_IN_WFE:
-			dev_err(hdev->dev,
-				"Device boot error - Stuck inside WFE loop\n");
-			break;
-		case CPU_BOOT_STATUS_IN_BTL:
-			dev_err(hdev->dev,
-				"Device boot error - Stuck in BTL\n");
-			break;
-		case CPU_BOOT_STATUS_IN_PREBOOT:
-			dev_err(hdev->dev,
-				"Device boot error - Stuck in Preboot\n");
-			break;
-		case CPU_BOOT_STATUS_IN_SPL:
-			dev_err(hdev->dev,
-				"Device boot error - Stuck in SPL\n");
-			break;
-		case CPU_BOOT_STATUS_IN_UBOOT:
-			dev_err(hdev->dev,
-				"Device boot error - Stuck in u-boot\n");
-			break;
-		case CPU_BOOT_STATUS_DRAM_INIT_FAIL:
-			dev_err(hdev->dev,
-				"Device boot error - DRAM initialization failed\n");
-			break;
-		case CPU_BOOT_STATUS_UBOOT_NOT_READY:
-			dev_err(hdev->dev,
-				"Device boot error - u-boot stopped by user\n");
-			break;
-		case CPU_BOOT_STATUS_TS_INIT_FAIL:
-			dev_err(hdev->dev,
-				"Device boot error - Thermal Sensor initialization failed\n");
-			break;
-		default:
-			dev_err(hdev->dev,
-				"Device boot error - Invalid status code %d\n",
-				status);
-			break;
-		}
-
 		rc = -EIO;
 		goto out;
 	}
@@ -569,7 +573,8 @@ int hl_fw_init_cpu(struct hl_device *hdev, u32 cpu_boot_status_reg,
 				"Device reports FIT image is corrupted\n");
 		else
 			dev_err(hdev->dev,
-				"Device failed to load, %d\n", status);
+				"Failed to load firmware to device, %d\n",
+				status);
 
 		rc = -EIO;
 		goto out;

drivers/misc/habanalabs/common/habanalabs.h

@@ -8,8 +8,9 @@
 #ifndef HABANALABSP_H_
 #define HABANALABSP_H_
 
-#include "include/armcp_if.h"
+#include "include/common/armcp_if.h"
-#include "include/qman_if.h"
+#include "include/common/qman_if.h"
+#include <uapi/misc/habanalabs.h>
 
 #include <linux/cdev.h>
 #include <linux/iopoll.h>
@@ -40,11 +41,6 @@
 
 #define HL_SIM_MAX_TIMEOUT_US		10000000 /* 10s */
 
-#define HL_MAX_QUEUES			128
-
-/* MUST BE POWER OF 2 and larger than 1 */
-#define HL_MAX_PENDING_CS		64
-
 #define HL_IDLE_BUSY_TS_ARR_SIZE	4096
 
 /* Memory */
@@ -53,6 +49,10 @@
 /* MMU */
 #define MMU_HASH_TABLE_BITS		7 /* 1 << 7 buckets */
 
+/*
+ * HL_RSVD_SOBS 'sync stream' reserved sync objects per QMAN stream
+ * HL_RSVD_MONS 'sync stream' reserved monitors per QMAN stream
+ */
 #define HL_RSVD_SOBS			4
 #define HL_RSVD_MONS			2
 
@@ -61,6 +61,11 @@
 
 #define HL_MAX_SOB_VAL			(1 << 15)
 
+#define IS_POWER_OF_2(n)		(n != 0 && ((n & (n - 1)) == 0))
+#define IS_MAX_PENDING_CS_VALID(n)	(IS_POWER_OF_2(n) && (n > 1))
+
+#define HL_PCI_NUM_BARS			6
+
 /**
  * struct pgt_info - MMU hop page info.
  * @node: hash linked-list node for the pgts shadow hash of pgts.
@@ -85,6 +90,16 @@ struct pgt_info {
 struct hl_device;
 struct hl_fpriv;
 
+/**
+ * enum hl_pci_match_mode - pci match mode per region
+ * @PCI_ADDRESS_MATCH_MODE: address match mode
+ * @PCI_BAR_MATCH_MODE: bar match mode
+ */
+enum hl_pci_match_mode {
+	PCI_ADDRESS_MATCH_MODE,
+	PCI_BAR_MATCH_MODE
+};
+
 /**
  * enum hl_fw_component - F/W components to read version through registers.
  * @FW_COMP_UBOOT: u-boot.
@@ -120,6 +135,32 @@ enum hl_cs_type {
 	CS_TYPE_WAIT
 };
 
+/*
+ * struct hl_inbound_pci_region - inbound region descriptor
+ * @mode: pci match mode for this region
+ * @addr: region target address
+ * @size: region size in bytes
+ * @offset_in_bar: offset within bar (address match mode)
+ * @bar: bar id
+ */
+struct hl_inbound_pci_region {
+	enum hl_pci_match_mode	mode;
+	u64			addr;
+	u64			size;
+	u64			offset_in_bar;
+	u8			bar;
+};
+
+/*
+ * struct hl_outbound_pci_region - outbound region descriptor
+ * @addr: region target address
+ * @size: region size in bytes
+ */
+struct hl_outbound_pci_region {
+	u64	addr;
+	u64	size;
+};
+
 /*
  * struct hl_hw_sob - H/W SOB info.
  * @hdev: habanalabs device structure.
@@ -141,11 +182,13 @@ struct hl_hw_sob {
  *               false otherwise.
  * @requires_kernel_cb: true if a CB handle must be provided for jobs on this
  *                      queue, false otherwise (a CB address must be provided).
+ * @supports_sync_stream: True if queue supports sync stream
  */
 struct hw_queue_properties {
 	enum hl_queue_type	type;
 	u8			driver_only;
 	u8			requires_kernel_cb;
+	u8			supports_sync_stream;
 };
 
 /**
@@ -241,14 +284,19 @@ struct hl_mmu_properties {
  * @psoc_pci_pll_nf: PCI PLL NF value.
  * @psoc_pci_pll_od: PCI PLL OD value.
  * @psoc_pci_pll_div_factor: PCI PLL DIV FACTOR 1 value.
+ * @psoc_timestamp_frequency: frequency of the psoc timestamp clock.
  * @high_pll: high PLL frequency used by the device.
  * @cb_pool_cb_cnt: number of CBs in the CB pool.
  * @cb_pool_cb_size: size of each CB in the CB pool.
+ * @max_pending_cs: maximum of concurrent pending command submissions
+ * @max_queues: maximum amount of queues in the system
+ * @sync_stream_first_sob: first sync object available for sync stream use
+ * @sync_stream_first_mon: first monitor available for sync stream use
  * @tpc_enabled_mask: which TPCs are enabled.
  * @completion_queues_count: number of completion queues.
  */
 struct asic_fixed_properties {
-	struct hw_queue_properties	hw_queues_props[HL_MAX_QUEUES];
+	struct hw_queue_properties	*hw_queues_props;
 	struct armcp_info		armcp_info;
 	char				uboot_ver[VERSION_MAX_LEN];
 	char				preboot_ver[VERSION_MAX_LEN];
@@ -282,9 +330,14 @@ struct asic_fixed_properties {
 	u32				psoc_pci_pll_nf;
 	u32				psoc_pci_pll_od;
 	u32				psoc_pci_pll_div_factor;
+	u32				psoc_timestamp_frequency;
 	u32				high_pll;
 	u32				cb_pool_cb_cnt;
 	u32				cb_pool_cb_size;
+	u32				max_pending_cs;
+	u32				max_queues;
+	u16				sync_stream_first_sob;
+	u16				sync_stream_first_mon;
 	u8				tpc_enabled_mask;
 	u8				completion_queues_count;
 };
@@ -339,6 +392,7 @@ struct hl_cb_mgr {
  * @ctx_id: holds the ID of the owner's context.
  * @mmap: true if the CB is currently mmaped to user.
  * @is_pool: true if CB was acquired from the pool, false otherwise.
+ * @is_internal: internaly allocated
  */
 struct hl_cb {
 	struct kref		refcount;
@@ -355,6 +409,7 @@ struct hl_cb {
 	u32			ctx_id;
 	u8			mmap;
 	u8			is_pool;
+	u8			is_internal;
 };
 
 
@@ -364,38 +419,19 @@ struct hl_cb {
 
 struct hl_cs_job;
 
-/*
+/* Queue length of external and HW queues */
- * Currently, there are two limitations on the maximum length of a queue:
+#define HL_QUEUE_LENGTH			4096
- *
- * 1. The memory footprint of the queue. The current allocated space for the
- *    queue is PAGE_SIZE. Because each entry in the queue is HL_BD_SIZE,
- *    the maximum length of the queue can be PAGE_SIZE / HL_BD_SIZE,
- *    which currently is 4096/16 = 256 entries.
- *
- *    To increase that, we need either to decrease the size of the
- *    BD (difficult), or allocate more than a single page (easier).
- *
- * 2. Because the size of the JOB handle field in the BD CTL / completion queue
- *    is 10-bit, we can have up to 1024 open jobs per hardware queue.
- *    Therefore, each queue can hold up to 1024 entries.
- *
- * HL_QUEUE_LENGTH is in units of struct hl_bd.
- * HL_QUEUE_LENGTH * sizeof(struct hl_bd) should be <= HL_PAGE_SIZE
- */
-
-#define HL_PAGE_SIZE			4096 /* minimum page size */
-/* Must be power of 2 (HL_PAGE_SIZE / HL_BD_SIZE) */
-#define HL_QUEUE_LENGTH			256
 #define HL_QUEUE_SIZE_IN_BYTES		(HL_QUEUE_LENGTH * HL_BD_SIZE)
 
-/*
+#if (HL_MAX_JOBS_PER_CS > HL_QUEUE_LENGTH)
- * HL_CQ_LENGTH is in units of struct hl_cq_entry.
+#error "HL_QUEUE_LENGTH must be greater than HL_MAX_JOBS_PER_CS"
- * HL_CQ_LENGTH should be <= HL_PAGE_SIZE
+#endif
- */
+
+/* HL_CQ_LENGTH is in units of struct hl_cq_entry */
 #define HL_CQ_LENGTH			HL_QUEUE_LENGTH
 #define HL_CQ_SIZE_IN_BYTES		(HL_CQ_LENGTH * HL_CQ_ENTRY_SIZE)
 
-/* Must be power of 2 (HL_PAGE_SIZE / HL_EQ_ENTRY_SIZE) */
+/* Must be power of 2 */
 #define HL_EQ_LENGTH			64
 #define HL_EQ_SIZE_IN_BYTES		(HL_EQ_LENGTH * HL_EQ_ENTRY_SIZE)
 
@@ -422,6 +458,7 @@ struct hl_cs_job;
  *         exist).
  * @curr_sob_offset: the id offset to the currently used SOB from the
  *                   HL_RSVD_SOBS that are being used by this queue.
+ * @supports_sync_stream: True if queue supports sync stream
  */
 struct hl_hw_queue {
 	struct hl_hw_sob	hw_sob[HL_RSVD_SOBS];
@@ -430,7 +467,7 @@ struct hl_hw_queue {
 	u64			kernel_address;
 	dma_addr_t		bus_address;
 	u32			pi;
-	u32			ci;
+	atomic_t		ci;
 	u32			hw_queue_id;
 	u32			cq_id;
 	u32			msi_vec;
@@ -440,6 +477,7 @@ struct hl_hw_queue {
 	u16			base_mon_id;
 	u8			valid;
 	u8			curr_sob_offset;
+	u8			supports_sync_stream;
 };
 
 /**
@@ -447,6 +485,7 @@ struct hl_hw_queue {
  * @hdev: pointer to the device structure
  * @kernel_address: holds the queue's kernel virtual address
  * @bus_address: holds the queue's DMA address
+ * @cq_idx: completion queue index in array
  * @hw_queue_id: the id of the matching H/W queue
  * @ci: ci inside the queue
  * @pi: pi inside the queue
@@ -456,6 +495,7 @@ struct hl_cq {
 	struct hl_device	*hdev;
 	u64			kernel_address;
 	dma_addr_t		bus_address;
+	u32			cq_idx;
 	u32			hw_queue_id;
 	u32			ci;
 	u32			pi;
@@ -519,6 +559,15 @@ enum hl_pll_frequency {
 	PLL_LAST
 };
 
+#define PLL_REF_CLK 50
+
+enum div_select_defs {
+	DIV_SEL_REF_CLK = 0,
+	DIV_SEL_PLL_CLK = 1,
+	DIV_SEL_DIVIDED_REF = 2,
+	DIV_SEL_DIVIDED_PLL = 3,
+};
+
 /**
  * struct hl_asic_funcs - ASIC specific functions that are can be called from
  *                        common code.
@@ -596,14 +645,13 @@ enum hl_pll_frequency {
  * @rreg: Read a register. Needed for simulator support.
  * @wreg: Write a register. Needed for simulator support.
  * @halt_coresight: stop the ETF and ETR traces.
+ * @ctx_init: context dependent initialization.
  * @get_clk_rate: Retrieve the ASIC current and maximum clock rate in MHz
  * @get_queue_id_for_cq: Get the H/W queue id related to the given CQ index.
  * @read_device_fw_version: read the device's firmware versions that are
  *                          contained in registers
  * @load_firmware_to_device: load the firmware to the device's memory
  * @load_boot_fit_to_device: load boot fit to device's memory
- * @ext_queue_init: Initialize the given external queue.
- * @ext_queue_reset: Reset the given external queue.
  * @get_signal_cb_size: Get signal CB size.
  * @get_wait_cb_size: Get wait CB size.
  * @gen_signal_cb: Generate a signal CB.
@@ -700,14 +748,13 @@ struct hl_asic_funcs {
 	u32 (*rreg)(struct hl_device *hdev, u32 reg);
 	void (*wreg)(struct hl_device *hdev, u32 reg, u32 val);
 	void (*halt_coresight)(struct hl_device *hdev);
+	int (*ctx_init)(struct hl_ctx *ctx);
 	int (*get_clk_rate)(struct hl_device *hdev, u32 *cur_clk, u32 *max_clk);
 	u32 (*get_queue_id_for_cq)(struct hl_device *hdev, u32 cq_idx);
 	void (*read_device_fw_version)(struct hl_device *hdev,
 					enum hl_fw_component fwc);
 	int (*load_firmware_to_device)(struct hl_device *hdev);
 	int (*load_boot_fit_to_device)(struct hl_device *hdev);
-	void (*ext_queue_init)(struct hl_device *hdev, u32 hw_queue_id);
-	void (*ext_queue_reset)(struct hl_device *hdev, u32 hw_queue_id);
 	u32 (*get_signal_cb_size)(struct hl_device *hdev);
 	u32 (*get_wait_cb_size)(struct hl_device *hdev);
 	void (*gen_signal_cb)(struct hl_device *hdev, void *data, u16 sob_id);
@@ -743,7 +790,6 @@ struct hl_va_range {
  * struct hl_ctx - user/kernel context.
  * @mem_hash: holds mapping from virtual address to virtual memory area
  *		descriptor (hl_vm_phys_pg_list or hl_userptr).
- * @mmu_phys_hash: holds a mapping from physical address to pgt_info structure.
  * @mmu_shadow_hash: holds a mapping from shadow address to pgt_info structure.
  * @hpriv: pointer to the private (Kernel Driver) data of the process (fd).
  * @hdev: pointer to the device structure.
@@ -777,18 +823,18 @@ struct hl_va_range {
  */
 struct hl_ctx {
 	DECLARE_HASHTABLE(mem_hash, MEM_HASH_TABLE_BITS);
-	DECLARE_HASHTABLE(mmu_phys_hash, MMU_HASH_TABLE_BITS);
 	DECLARE_HASHTABLE(mmu_shadow_hash, MMU_HASH_TABLE_BITS);
 	struct hl_fpriv		*hpriv;
 	struct hl_device	*hdev;
 	struct kref		refcount;
-	struct dma_fence	*cs_pending[HL_MAX_PENDING_CS];
+	struct dma_fence	**cs_pending;
 	struct hl_va_range	*host_va_range;
 	struct hl_va_range	*host_huge_va_range;
 	struct hl_va_range	*dram_va_range;
 	struct mutex		mem_hash_lock;
 	struct mutex		mmu_lock;
 	struct list_head	debugfs_list;
+	struct hl_cs_counters	cs_counters;
 	u64			cs_sequence;
 	u64			*dram_default_hops;
 	spinlock_t		cs_lock;
@@ -863,7 +909,7 @@ struct hl_userptr {
  * @aborted: true if CS was aborted due to some device error.
  */
 struct hl_cs {
-	u16			jobs_in_queue_cnt[HL_MAX_QUEUES];
+	u16			*jobs_in_queue_cnt;
 	struct hl_ctx		*ctx;
 	struct list_head	job_list;
 	spinlock_t		job_lock;
@@ -1347,7 +1393,9 @@ struct hl_device_idle_busy_ts {
 /**
  * struct hl_device - habanalabs device structure.
  * @pdev: pointer to PCI device, can be NULL in case of simulator device.
- * @pcie_bar: array of available PCIe bars.
+ * @pcie_bar_phys: array of available PCIe bars physical addresses.
+ *		   (required only for PCI address match mode)
+ * @pcie_bar: array of available PCIe bars virtual addresses.
  * @rmmio: configuration area address on SRAM.
  * @cdev: related char device.
  * @cdev_ctrl: char device for control operations only (INFO IOCTL)
@@ -1358,7 +1406,8 @@ struct hl_device_idle_busy_ts {
  * @asic_name: ASIC specific nmae.
  * @asic_type: ASIC specific type.
  * @completion_queue: array of hl_cq.
- * @cq_wq: work queue of completion queues for executing work in process context
+ * @cq_wq: work queues of completion queues for executing work in process
+ *         context.
  * @eq_wq: work queue of event queue for executing work in process context.
  * @kernel_ctx: Kernel driver context structure.
  * @kernel_queues: array of hl_hw_queue.
@@ -1387,12 +1436,17 @@ struct hl_device_idle_busy_ts {
  * @hl_debugfs: device's debugfs manager.
  * @cb_pool: list of preallocated CBs.
  * @cb_pool_lock: protects the CB pool.
+ * @internal_cb_pool_virt_addr: internal command buffer pool virtual address.
+ * @internal_cb_pool_dma_addr: internal command buffer pool dma address.
+ * @internal_cb_pool: internal command buffer memory pool.
+ * @internal_cb_va_base: internal cb pool mmu virtual address base
  * @fpriv_list: list of file private data structures. Each structure is created
  *              when a user opens the device
  * @fpriv_list_lock: protects the fpriv_list
  * @compute_ctx: current compute context executing.
  * @idle_busy_ts_arr: array to hold time stamps of transitions from idle to busy
  *                    and vice-versa
+ * @aggregated_cs_counters: aggregated cs counters among all contexts
  * @dram_used_mem: current DRAM memory consumption.
  * @timeout_jiffies: device CS timeout value.
  * @max_power: the max power of the device, as configured by the sysadmin. This
@@ -1435,12 +1489,14 @@ struct hl_device_idle_busy_ts {
  * @cdev_sysfs_created: were char devices and sysfs nodes created.
  * @stop_on_err: true if engines should stop on error.
  * @supports_sync_stream: is sync stream supported.
+ * @sync_stream_queue_idx: helper index for sync stream queues initialization.
  * @supports_coresight: is CoreSight supported.
  * @supports_soft_reset: is soft reset supported.
  */
 struct hl_device {
 	struct pci_dev			*pdev;
-	void __iomem			*pcie_bar[6];
+	u64				pcie_bar_phys[HL_PCI_NUM_BARS];
+	void __iomem			*pcie_bar[HL_PCI_NUM_BARS];
 	void __iomem			*rmmio;
 	struct cdev			cdev;
 	struct cdev			cdev_ctrl;
@@ -1451,7 +1507,7 @@ struct hl_device {
 	char				asic_name[16];
 	enum hl_asic_type		asic_type;
 	struct hl_cq			*completion_queue;
-	struct workqueue_struct		*cq_wq;
+	struct workqueue_struct		**cq_wq;
 	struct workqueue_struct		*eq_wq;
 	struct hl_ctx			*kernel_ctx;
 	struct hl_hw_queue		*kernel_queues;
@@ -1483,6 +1539,11 @@ struct hl_device {
 	struct list_head		cb_pool;
 	spinlock_t			cb_pool_lock;
 
+	void				*internal_cb_pool_virt_addr;
+	dma_addr_t			internal_cb_pool_dma_addr;
+	struct gen_pool			*internal_cb_pool;
+	u64				internal_cb_va_base;
+
 	struct list_head		fpriv_list;
 	struct mutex			fpriv_list_lock;
 
@@ -1490,6 +1551,8 @@ struct hl_device {
 
 	struct hl_device_idle_busy_ts	*idle_busy_ts_arr;
 
+	struct hl_cs_counters		aggregated_cs_counters;
+
 	atomic64_t			dram_used_mem;
 	u64				timeout_jiffies;
 	u64				max_power;
@@ -1522,6 +1585,7 @@ struct hl_device {
 	u8				cdev_sysfs_created;
 	u8				stop_on_err;
 	u8				supports_sync_stream;
+	u8				sync_stream_queue_idx;
 	u8				supports_coresight;
 	u8				supports_soft_reset;
 
@@ -1690,7 +1754,7 @@ int hl_hwmon_init(struct hl_device *hdev);
 void hl_hwmon_fini(struct hl_device *hdev);
 
 int hl_cb_create(struct hl_device *hdev, struct hl_cb_mgr *mgr, u32 cb_size,
-		u64 *handle, int ctx_id);
+		u64 *handle, int ctx_id, bool internal_cb);
 int hl_cb_destroy(struct hl_device *hdev, struct hl_cb_mgr *mgr, u64 cb_handle);
 int hl_cb_mmap(struct hl_fpriv *hpriv, struct vm_area_struct *vma);
 struct hl_cb *hl_cb_get(struct hl_device *hdev,	struct hl_cb_mgr *mgr,
@@ -1698,7 +1762,8 @@ struct hl_cb *hl_cb_get(struct hl_device *hdev,	struct hl_cb_mgr *mgr,
 void hl_cb_put(struct hl_cb *cb);
 void hl_cb_mgr_init(struct hl_cb_mgr *mgr);
 void hl_cb_mgr_fini(struct hl_device *hdev, struct hl_cb_mgr *mgr);
-struct hl_cb *hl_cb_kernel_create(struct hl_device *hdev, u32 cb_size);
+struct hl_cb *hl_cb_kernel_create(struct hl_device *hdev, u32 cb_size,
+					bool internal_cb);
 int hl_cb_pool_init(struct hl_device *hdev);
 int hl_cb_pool_fini(struct hl_device *hdev);
 
@@ -1762,9 +1827,10 @@ int hl_pci_bars_map(struct hl_device *hdev, const char * const name[3],
 int hl_pci_iatu_write(struct hl_device *hdev, u32 addr, u32 data);
 int hl_pci_set_dram_bar_base(struct hl_device *hdev, u8 inbound_region, u8 bar,
 				u64 addr);
-int hl_pci_init_iatu(struct hl_device *hdev, u64 sram_base_address,
+int hl_pci_set_inbound_region(struct hl_device *hdev, u8 region,
-			u64 dram_base_address, u64 host_phys_base_address,
+		struct hl_inbound_pci_region *pci_region);
-			u64 host_phys_size);
+int hl_pci_set_outbound_region(struct hl_device *hdev,
+		struct hl_outbound_pci_region *pci_region);
 int hl_pci_init(struct hl_device *hdev);
 void hl_pci_fini(struct hl_device *hdev);
 

drivers/misc/habanalabs/common/habanalabs_drv.c

@@ -238,7 +238,6 @@ static void set_driver_behavior_per_device(struct hl_device *hdev)
 	hdev->axi_drain = 0;
 	hdev->sram_scrambler_enable = 1;
 	hdev->dram_scrambler_enable = 1;
-	hdev->rl_enable = 1;
 	hdev->bmc_enable = 1;
 	hdev->hard_reset_on_fw_events = 1;
 }

drivers/misc/habanalabs/common/habanalabs_ioctl.c

@@ -276,6 +276,27 @@ static int time_sync_info(struct hl_device *hdev, struct hl_info_args *args)
 		min((size_t) max_size, sizeof(time_sync))) ? -EFAULT : 0;
 }
 
+static int cs_counters_info(struct hl_fpriv *hpriv, struct hl_info_args *args)
+{
+	struct hl_device *hdev = hpriv->hdev;
+	struct hl_info_cs_counters cs_counters = {0};
+	u32 max_size = args->return_size;
+	void __user *out = (void __user *) (uintptr_t) args->return_pointer;
+
+	if ((!max_size) || (!out))
+		return -EINVAL;
+
+	memcpy(&cs_counters.cs_counters, &hdev->aggregated_cs_counters,
+			sizeof(struct hl_cs_counters));
+
+	if (hpriv->ctx)
+		memcpy(&cs_counters.ctx_cs_counters, &hpriv->ctx->cs_counters,
+				sizeof(struct hl_cs_counters));
+
+	return copy_to_user(out, &cs_counters,
+		min((size_t) max_size, sizeof(cs_counters))) ? -EFAULT : 0;
+}
+
 static int _hl_info_ioctl(struct hl_fpriv *hpriv, void *data,
 				struct device *dev)
 {
@@ -336,6 +357,9 @@ static int _hl_info_ioctl(struct hl_fpriv *hpriv, void *data,
 	case HL_INFO_TIME_SYNC:
 		return time_sync_info(hdev, args);
 
+	case HL_INFO_CS_COUNTERS:
+		return cs_counters_info(hpriv, args);
+
 	default:
 		dev_err(dev, "Invalid request %d\n", args->op);
 		rc = -ENOTTY;

drivers/misc/habanalabs/common/hw_queue.c

@@ -23,10 +23,14 @@ inline u32 hl_hw_queue_add_ptr(u32 ptr, u16 val)
 	ptr &= ((HL_QUEUE_LENGTH << 1) - 1);
 	return ptr;
 }
+static inline int queue_ci_get(atomic_t *ci, u32 queue_len)
+{
+	return atomic_read(ci) & ((queue_len << 1) - 1);
+}
 
 static inline int queue_free_slots(struct hl_hw_queue *q, u32 queue_len)
 {
-	int delta = (q->pi - q->ci);
+	int delta = (q->pi - queue_ci_get(&q->ci, queue_len));
 
 	if (delta >= 0)
 		return (queue_len - delta);
@@ -40,23 +44,16 @@ void hl_int_hw_queue_update_ci(struct hl_cs *cs)
 	struct hl_hw_queue *q;
 	int i;
 
-	hdev->asic_funcs->hw_queues_lock(hdev);
-
 	if (hdev->disabled)
-		goto out;
+		return;
 
 	q = &hdev->kernel_queues[0];
-	for (i = 0 ; i < HL_MAX_QUEUES ; i++, q++) {
+	for (i = 0 ; i < hdev->asic_prop.max_queues ; i++, q++) {
-		if (q->queue_type == QUEUE_TYPE_INT) {
+		if (q->queue_type == QUEUE_TYPE_INT)
-			q->ci += cs->jobs_in_queue_cnt[i];
+			atomic_add(cs->jobs_in_queue_cnt[i], &q->ci);
-			q->ci &= ((q->int_queue_len << 1) - 1);
 	}
 }
 
-out:
-	hdev->asic_funcs->hw_queues_unlock(hdev);
-}
-
 /*
  * ext_and_hw_queue_submit_bd() - Submit a buffer descriptor to an external or a
  *                                H/W queue.
@@ -161,6 +158,13 @@ static int int_queue_sanity_checks(struct hl_device *hdev,
 {
 	int free_slots_cnt;
 
+	if (num_of_entries > q->int_queue_len) {
+		dev_err(hdev->dev,
+			"Cannot populate queue %u with %u jobs\n",
+			q->hw_queue_id, num_of_entries);
+		return -ENOMEM;
+	}
+
 	/* Check we have enough space in the queue */
 	free_slots_cnt = queue_free_slots(q, q->int_queue_len);
 
@@ -174,38 +178,26 @@ static int int_queue_sanity_checks(struct hl_device *hdev,
 }
 
 /*
- * hw_queue_sanity_checks() - Perform some sanity checks on a H/W queue.
+ * hw_queue_sanity_checks() - Make sure we have enough space in the h/w queue
  * @hdev: Pointer to hl_device structure.
  * @q: Pointer to hl_hw_queue structure.
  * @num_of_entries: How many entries to check for space.
  *
- * Perform the following:
+ * Notice: We do not reserve queue entries so this function mustn't be called
- * - Make sure we have enough space in the completion queue.
+ *         more than once per CS for the same queue
- *   This check also ensures that there is enough space in the h/w queue, as
- *   both queues are of the same size.
- * - Reserve space in the completion queue (needs to be reversed if there
- *   is a failure down the road before the actual submission of work).
  *
- * Both operations are done using the "free_slots_cnt" field of the completion
- * queue. The CI counters of the queue and the completion queue are not
- * needed/used for the H/W queue type.
  */
 static int hw_queue_sanity_checks(struct hl_device *hdev, struct hl_hw_queue *q,
 					int num_of_entries)
 {
-	atomic_t *free_slots =
+	int free_slots_cnt;
-			&hdev->completion_queue[q->cq_id].free_slots_cnt;
 
-	/*
+	/* Check we have enough space in the queue */
-	 * Check we have enough space in the completion queue.
+	free_slots_cnt = queue_free_slots(q, HL_QUEUE_LENGTH);
-	 * Add -1 to counter (decrement) unless counter was already 0.
+
-	 * In that case, CQ is full so we can't submit a new CB.
+	if (free_slots_cnt < num_of_entries) {
-	 * atomic_add_unless will return 0 if counter was already 0.
+		dev_dbg(hdev->dev, "Queue %d doesn't have room for %d CBs\n",
-	 */
+			q->hw_queue_id, num_of_entries);
-	if (atomic_add_negative(num_of_entries * -1, free_slots)) {
-		dev_dbg(hdev->dev, "No space for %d entries on CQ %d\n",
-			num_of_entries, q->hw_queue_id);
-		atomic_add(num_of_entries, free_slots);
 		return -EAGAIN;
 	}
 
@@ -366,7 +358,6 @@ static void hw_queue_schedule_job(struct hl_cs_job *job)
 {
 	struct hl_device *hdev = job->cs->ctx->hdev;
 	struct hl_hw_queue *q = &hdev->kernel_queues[job->hw_queue_id];
-	struct hl_cq *cq;
 	u64 ptr;
 	u32 offset, ctl, len;
 
@@ -376,7 +367,7 @@ static void hw_queue_schedule_job(struct hl_cs_job *job)
 	 * write address offset in the SM block (QMAN LBW message).
 	 * The write address offset is calculated as "COMP_OFFSET << 2".
 	 */
-	offset = job->cs->sequence & (HL_MAX_PENDING_CS - 1);
+	offset = job->cs->sequence & (hdev->asic_prop.max_pending_cs - 1);
 	ctl = ((offset << BD_CTL_COMP_OFFSET_SHIFT) & BD_CTL_COMP_OFFSET_MASK) |
 		((q->pi << BD_CTL_COMP_DATA_SHIFT) & BD_CTL_COMP_DATA_MASK);
 
@@ -395,17 +386,6 @@ static void hw_queue_schedule_job(struct hl_cs_job *job)
 	else
 		ptr = (u64) (uintptr_t) job->user_cb;
 
-	/*
-	 * No need to protect pi_offset because scheduling to the
-	 * H/W queues is done under the scheduler mutex
-	 *
-	 * No need to check if CQ is full because it was already
-	 * checked in hw_queue_sanity_checks
-	 */
-	cq = &hdev->completion_queue[q->cq_id];
-
-	cq->pi = hl_cq_inc_ptr(cq->pi);
-
 	ext_and_hw_queue_submit_bd(hdev, q, ctl, len, ptr);
 }
 
@@ -509,19 +489,23 @@ int hl_hw_queue_schedule_cs(struct hl_cs *cs)
 	struct hl_device *hdev = ctx->hdev;
 	struct hl_cs_job *job, *tmp;
 	struct hl_hw_queue *q;
+	u32 max_queues;
 	int rc = 0, i, cq_cnt;
 
 	hdev->asic_funcs->hw_queues_lock(hdev);
 
 	if (hl_device_disabled_or_in_reset(hdev)) {
+		ctx->cs_counters.device_in_reset_drop_cnt++;
 		dev_err(hdev->dev,
 			"device is disabled or in reset, CS rejected!\n");
 		rc = -EPERM;
 		goto out;
 	}
 
+	max_queues = hdev->asic_prop.max_queues;
+
 	q = &hdev->kernel_queues[0];
-	for (i = 0, cq_cnt = 0 ; i < HL_MAX_QUEUES ; i++, q++) {
+	for (i = 0, cq_cnt = 0 ; i < max_queues ; i++, q++) {
 		if (cs->jobs_in_queue_cnt[i]) {
 			switch (q->queue_type) {
 			case QUEUE_TYPE_EXT:
@@ -543,11 +527,12 @@ int hl_hw_queue_schedule_cs(struct hl_cs *cs)
 				break;
 			}
 
-			if (rc)
+			if (rc) {
+				ctx->cs_counters.queue_full_drop_cnt++;
 				goto unroll_cq_resv;
+			}
 
-			if (q->queue_type == QUEUE_TYPE_EXT ||
+			if (q->queue_type == QUEUE_TYPE_EXT)
-					q->queue_type == QUEUE_TYPE_HW)
 				cq_cnt++;
 		}
 	}
@@ -598,10 +583,9 @@ int hl_hw_queue_schedule_cs(struct hl_cs *cs)
 
 unroll_cq_resv:
 	q = &hdev->kernel_queues[0];
-	for (i = 0 ; (i < HL_MAX_QUEUES) && (cq_cnt > 0) ; i++, q++) {
+	for (i = 0 ; (i < max_queues) && (cq_cnt > 0) ; i++, q++) {
-		if ((q->queue_type == QUEUE_TYPE_EXT ||
+		if ((q->queue_type == QUEUE_TYPE_EXT) &&
-				q->queue_type == QUEUE_TYPE_HW) &&
+						(cs->jobs_in_queue_cnt[i])) {
-				cs->jobs_in_queue_cnt[i]) {
 			atomic_t *free_slots =
 				&hdev->completion_queue[i].free_slots_cnt;
 			atomic_add(cs->jobs_in_queue_cnt[i], free_slots);
@@ -625,7 +609,7 @@ void hl_hw_queue_inc_ci_kernel(struct hl_device *hdev, u32 hw_queue_id)
 {
 	struct hl_hw_queue *q = &hdev->kernel_queues[hw_queue_id];
 
-	q->ci = hl_queue_inc_ptr(q->ci);
+	atomic_inc(&q->ci);
 }
 
 static int ext_and_cpu_queue_init(struct hl_device *hdev, struct hl_hw_queue *q,
@@ -660,12 +644,9 @@ static int ext_and_cpu_queue_init(struct hl_device *hdev, struct hl_hw_queue *q,
 	}
 
 	/* Make sure read/write pointers are initialized to start of queue */
-	q->ci = 0;
+	atomic_set(&q->ci, 0);
 	q->pi = 0;
 
-	if (!is_cpu_queue)
-		hdev->asic_funcs->ext_queue_init(hdev, q->hw_queue_id);
-
 	return 0;
 
 free_queue:
@@ -697,7 +678,7 @@ static int int_queue_init(struct hl_device *hdev, struct hl_hw_queue *q)
 
 	q->kernel_address = (u64) (uintptr_t) p;
 	q->pi = 0;
-	q->ci = 0;
+	atomic_set(&q->ci, 0);
 
 	return 0;
 }
@@ -726,12 +707,48 @@ static int hw_queue_init(struct hl_device *hdev, struct hl_hw_queue *q)
 	q->kernel_address = (u64) (uintptr_t) p;
 
 	/* Make sure read/write pointers are initialized to start of queue */
-	q->ci = 0;
+	atomic_set(&q->ci, 0);
 	q->pi = 0;
 
 	return 0;
 }
 
+static void sync_stream_queue_init(struct hl_device *hdev, u32 q_idx)
+{
+	struct hl_hw_queue *hw_queue = &hdev->kernel_queues[q_idx];
+	struct asic_fixed_properties *prop = &hdev->asic_prop;
+	struct hl_hw_sob *hw_sob;
+	int sob, queue_idx = hdev->sync_stream_queue_idx++;
+
+	hw_queue->base_sob_id =
+		prop->sync_stream_first_sob + queue_idx * HL_RSVD_SOBS;
+	hw_queue->base_mon_id =
+		prop->sync_stream_first_mon + queue_idx * HL_RSVD_MONS;
+	hw_queue->next_sob_val = 1;
+	hw_queue->curr_sob_offset = 0;
+
+	for (sob = 0 ; sob < HL_RSVD_SOBS ; sob++) {
+		hw_sob = &hw_queue->hw_sob[sob];
+		hw_sob->hdev = hdev;
+		hw_sob->sob_id = hw_queue->base_sob_id + sob;
+		hw_sob->q_idx = q_idx;
+		kref_init(&hw_sob->kref);
+	}
+}
+
+static void sync_stream_queue_reset(struct hl_device *hdev, u32 q_idx)
+{
+	struct hl_hw_queue *hw_queue = &hdev->kernel_queues[q_idx];
+
+	/*
+	 * In case we got here due to a stuck CS, the refcnt might be bigger
+	 * than 1 and therefore we reset it.
+	 */
+	kref_init(&hw_queue->hw_sob[hw_queue->curr_sob_offset].kref);
+	hw_queue->curr_sob_offset = 0;
+	hw_queue->next_sob_val = 1;
+}
+
 /*
  * queue_init - main initialization function for H/W queue object
  *
@@ -747,8 +764,6 @@ static int queue_init(struct hl_device *hdev, struct hl_hw_queue *q,
 {
 	int rc;
 
-	BUILD_BUG_ON(HL_QUEUE_SIZE_IN_BYTES > HL_PAGE_SIZE);
-
 	q->hw_queue_id = hw_queue_id;
 
 	switch (q->queue_type) {
@@ -774,6 +789,9 @@ static int queue_init(struct hl_device *hdev, struct hl_hw_queue *q,
 		break;
 	}
 
+	if (q->supports_sync_stream)
+		sync_stream_queue_init(hdev, q->hw_queue_id);
+
 	if (rc)
 		return rc;
 
@@ -835,7 +853,7 @@ int hl_hw_queues_create(struct hl_device *hdev)
 	struct hl_hw_queue *q;
 	int i, rc, q_ready_cnt;
 
-	hdev->kernel_queues = kcalloc(HL_MAX_QUEUES,
+	hdev->kernel_queues = kcalloc(asic->max_queues,
 				sizeof(*hdev->kernel_queues), GFP_KERNEL);
 
 	if (!hdev->kernel_queues) {
@@ -845,9 +863,11 @@ int hl_hw_queues_create(struct hl_device *hdev)
 
 	/* Initialize the H/W queues */
 	for (i = 0, q_ready_cnt = 0, q = hdev->kernel_queues;
-			i < HL_MAX_QUEUES ; i++, q_ready_cnt++, q++) {
+			i < asic->max_queues ; i++, q_ready_cnt++, q++) {
 
 		q->queue_type = asic->hw_queues_props[i].type;
+		q->supports_sync_stream =
+				asic->hw_queues_props[i].supports_sync_stream;
 		rc = queue_init(hdev, q, i);
 		if (rc) {
 			dev_err(hdev->dev,
@@ -870,9 +890,10 @@ release_queues:
 void hl_hw_queues_destroy(struct hl_device *hdev)
 {
 	struct hl_hw_queue *q;
+	u32 max_queues = hdev->asic_prop.max_queues;
 	int i;
 
-	for (i = 0, q = hdev->kernel_queues ; i < HL_MAX_QUEUES ; i++, q++)
+	for (i = 0, q = hdev->kernel_queues ; i < max_queues ; i++, q++)
 		queue_fini(hdev, q);
 
 	kfree(hdev->kernel_queues);
@@ -881,15 +902,17 @@ void hl_hw_queues_destroy(struct hl_device *hdev)
 void hl_hw_queue_reset(struct hl_device *hdev, bool hard_reset)
 {
 	struct hl_hw_queue *q;
+	u32 max_queues = hdev->asic_prop.max_queues;
 	int i;
 
-	for (i = 0, q = hdev->kernel_queues ; i < HL_MAX_QUEUES ; i++, q++) {
+	for (i = 0, q = hdev->kernel_queues ; i < max_queues ; i++, q++) {
 		if ((!q->valid) ||
 			((!hard_reset) && (q->queue_type == QUEUE_TYPE_CPU)))
 			continue;
-		q->pi = q->ci = 0;
+		q->pi = 0;
+		atomic_set(&q->ci, 0);
 
-		if (q->queue_type == QUEUE_TYPE_EXT)
+		if (q->supports_sync_stream)
-			hdev->asic_funcs->ext_queue_reset(hdev, q->hw_queue_id);
+			sync_stream_queue_reset(hdev, q->hw_queue_id);
 	}
 }

drivers/misc/habanalabs/common/irq.c

@@ -119,15 +119,10 @@ irqreturn_t hl_irq_handler_cq(int irq, void *arg)
 
 		if ((shadow_index_valid) && (!hdev->disabled)) {
 			job = queue->shadow_queue[hl_pi_2_offset(shadow_index)];
-			queue_work(hdev->cq_wq, &job->finish_work);
+			queue_work(hdev->cq_wq[cq->cq_idx], &job->finish_work);
 		}
 
-		/* Update ci of the context's queue. There is no
+		atomic_inc(&queue->ci);
-		 * need to protect it with spinlock because this update is
-		 * done only inside IRQ and there is a different IRQ per
-		 * queue
-		 */
-		queue->ci = hl_queue_inc_ptr(queue->ci);
 
 		/* Clear CQ entry ready bit */
 		cq_entry->data = cpu_to_le32(le32_to_cpu(cq_entry->data) &
@@ -220,8 +215,6 @@ int hl_cq_init(struct hl_device *hdev, struct hl_cq *q, u32 hw_queue_id)
 {
 	void *p;
 
-	BUILD_BUG_ON(HL_CQ_SIZE_IN_BYTES > HL_PAGE_SIZE);
-
 	p = hdev->asic_funcs->asic_dma_alloc_coherent(hdev, HL_CQ_SIZE_IN_BYTES,
 				&q->bus_address, GFP_KERNEL | __GFP_ZERO);
 	if (!p)
@@ -282,8 +275,6 @@ int hl_eq_init(struct hl_device *hdev, struct hl_eq *q)
 {
 	void *p;
 
-	BUILD_BUG_ON(HL_EQ_SIZE_IN_BYTES > HL_PAGE_SIZE);
-
 	p = hdev->asic_funcs->cpu_accessible_dma_pool_alloc(hdev,
 							HL_EQ_SIZE_IN_BYTES,
 							&q->bus_address);

drivers/misc/habanalabs/common/memory.c

@@ -1730,8 +1730,7 @@ void hl_vm_ctx_fini(struct hl_ctx *ctx)
 	 */
 	if (!hdev->hard_reset_pending && !hash_empty(ctx->mem_hash))
 		dev_notice(hdev->dev,
-				"ctx %d is freed while it has va in use\n",
+			"user released device without removing its memory mappings\n");
-				ctx->asid);
 
 	hash_for_each_safe(ctx->mem_hash, i, tmp_node, hnode, node) {
 		dev_dbg(hdev->dev,

drivers/misc/habanalabs/common/mmu.c

@@ -502,7 +502,6 @@ int hl_mmu_ctx_init(struct hl_ctx *ctx)
 		return 0;
 
 	mutex_init(&ctx->mmu_lock);
-	hash_init(ctx->mmu_phys_hash);
 	hash_init(ctx->mmu_shadow_hash);
 
 	return dram_default_mapping_init(ctx);

drivers/misc/habanalabs/common/pci.c

@@ -9,9 +9,15 @@
 #include "include/hw_ip/pci/pci_general.h"
 
 #include <linux/pci.h>
+#include <linux/bitfield.h>
 
 #define HL_PLDM_PCI_ELBI_TIMEOUT_MSEC	(HL_PCI_ELBI_TIMEOUT_MSEC * 10)
 
+#define IATU_REGION_CTRL_REGION_EN_MASK		BIT(31)
+#define IATU_REGION_CTRL_MATCH_MODE_MASK	BIT(30)
+#define IATU_REGION_CTRL_NUM_MATCH_EN_MASK	BIT(19)
+#define IATU_REGION_CTRL_BAR_NUM_MASK		GENMASK(10, 8)
+
 /**
  * hl_pci_bars_map() - Map PCI BARs.
  * @hdev: Pointer to hl_device structure.
@@ -187,110 +193,94 @@ static void hl_pci_reset_link_through_bridge(struct hl_device *hdev)
 }
 
 /**
- * hl_pci_set_dram_bar_base() - Set DDR BAR to map specific device address.
+ * hl_pci_set_inbound_region() - Configure inbound region
  * @hdev: Pointer to hl_device structure.
- * @inbound_region: Inbound region number.
+ * @region: Inbound region number.
- * @bar: PCI BAR number.
+ * @pci_region: Inbound region parameters.
- * @addr: Address in DRAM. Must be aligned to DRAM bar size.
  *
- * Configure the iATU so that the DRAM bar will start at the specified address.
+ * Configure the iATU inbound region.
  *
  * Return: 0 on success, negative value for failure.
  */
-int hl_pci_set_dram_bar_base(struct hl_device *hdev, u8 inbound_region, u8 bar,
+int hl_pci_set_inbound_region(struct hl_device *hdev, u8 region,
-				u64 addr)
+		struct hl_inbound_pci_region *pci_region)
 {
 	struct asic_fixed_properties *prop = &hdev->asic_prop;
-	u32 offset;
+	u64 bar_phys_base, region_base, region_end_address;
-	int rc;
+	u32 offset, ctrl_reg_val;
+	int rc = 0;
 
-	switch (inbound_region) {
+	/* region offset */
-	case 0:
+	offset = (0x200 * region) + 0x100;
-		offset = 0x100;
-		break;
-	case 1:
-		offset = 0x300;
-		break;
-	case 2:
-		offset = 0x500;
-		break;
-	default:
-		dev_err(hdev->dev, "Invalid inbound region %d\n",
-			inbound_region);
-		return -EINVAL;
-	}
 
-	if (bar != 0 && bar != 2 && bar != 4) {
+	if (pci_region->mode == PCI_ADDRESS_MATCH_MODE) {
-		dev_err(hdev->dev, "Invalid PCI BAR %d\n", bar);
+		bar_phys_base = hdev->pcie_bar_phys[pci_region->bar];
-		return -EINVAL;
+		region_base = bar_phys_base + pci_region->offset_in_bar;
+		region_end_address = region_base + pci_region->size - 1;
+
+		rc |= hl_pci_iatu_write(hdev, offset + 0x8,
+				lower_32_bits(region_base));
+		rc |= hl_pci_iatu_write(hdev, offset + 0xC,
+				upper_32_bits(region_base));
+		rc |= hl_pci_iatu_write(hdev, offset + 0x10,
+				lower_32_bits(region_end_address));
 	}
 
 	/* Point to the specified address */
-	rc = hl_pci_iatu_write(hdev, offset + 0x14, lower_32_bits(addr));
+	rc = hl_pci_iatu_write(hdev, offset + 0x14,
-	rc |= hl_pci_iatu_write(hdev, offset + 0x18, upper_32_bits(addr));
+			lower_32_bits(pci_region->addr));
+	rc |= hl_pci_iatu_write(hdev, offset + 0x18,
+			upper_32_bits(pci_region->addr));
 	rc |= hl_pci_iatu_write(hdev, offset + 0x0, 0);
-	/* Enable + BAR match + match enable + BAR number */
+
-	rc |= hl_pci_iatu_write(hdev, offset + 0x4, 0xC0080000 | (bar << 8));
+	/* Enable + bar/address match + match enable + bar number */
+	ctrl_reg_val = FIELD_PREP(IATU_REGION_CTRL_REGION_EN_MASK, 1);
+	ctrl_reg_val |= FIELD_PREP(IATU_REGION_CTRL_MATCH_MODE_MASK,
+			pci_region->mode);
+	ctrl_reg_val |= FIELD_PREP(IATU_REGION_CTRL_NUM_MATCH_EN_MASK, 1);
+
+	if (pci_region->mode == PCI_BAR_MATCH_MODE)
+		ctrl_reg_val |= FIELD_PREP(IATU_REGION_CTRL_BAR_NUM_MASK,
+				pci_region->bar);
+
+	rc |= hl_pci_iatu_write(hdev, offset + 0x4, ctrl_reg_val);
 
 	/* Return the DBI window to the default location */
 	rc |= hl_pci_elbi_write(hdev, prop->pcie_aux_dbi_reg_addr, 0);
 	rc |= hl_pci_elbi_write(hdev, prop->pcie_aux_dbi_reg_addr + 4, 0);
 
 	if (rc)
-		dev_err(hdev->dev, "failed to map DRAM bar to 0x%08llx\n",
+		dev_err(hdev->dev, "failed to map bar %u to 0x%08llx\n",
-			addr);
+				pci_region->bar, pci_region->addr);
 
 	return rc;
 }
 
 /**
- * hl_pci_init_iatu() - Initialize the iATU unit inside the PCI controller.
+ * hl_pci_set_outbound_region() - Configure outbound region 0
  * @hdev: Pointer to hl_device structure.
- * @sram_base_address: SRAM base address.
+ * @pci_region: Outbound region parameters.
- * @dram_base_address: DRAM base address.
- * @host_phys_base_address: Base physical address of host memory for device
- *                          transactions.
- * @host_phys_size: Size of host memory for device transactions.
  *
- * This is needed in case the firmware doesn't initialize the iATU.
+ * Configure the iATU outbound region 0.
  *
  * Return: 0 on success, negative value for failure.
  */
-int hl_pci_init_iatu(struct hl_device *hdev, u64 sram_base_address,
+int hl_pci_set_outbound_region(struct hl_device *hdev,
-			u64 dram_base_address, u64 host_phys_base_address,
+		struct hl_outbound_pci_region *pci_region)
-			u64 host_phys_size)
 {
 	struct asic_fixed_properties *prop = &hdev->asic_prop;
-	u64 host_phys_end_addr;
+	u64 outbound_region_end_address;
 	int rc = 0;
 
-	/* Inbound Region 0 - Bar 0 - Point to SRAM base address */
+	/* Outbound Region 0 */
-	rc  = hl_pci_iatu_write(hdev, 0x114, lower_32_bits(sram_base_address));
+	outbound_region_end_address =
-	rc |= hl_pci_iatu_write(hdev, 0x118, upper_32_bits(sram_base_address));
+			pci_region->addr + pci_region->size - 1;
-	rc |= hl_pci_iatu_write(hdev, 0x100, 0);
-	/* Enable + Bar match + match enable */
-	rc |= hl_pci_iatu_write(hdev, 0x104, 0xC0080000);
-
-	/* Return the DBI window to the default location */
-	rc |= hl_pci_elbi_write(hdev, prop->pcie_aux_dbi_reg_addr, 0);
-	rc |= hl_pci_elbi_write(hdev, prop->pcie_aux_dbi_reg_addr + 4, 0);
-
-	hdev->asic_funcs->set_dma_mask_from_fw(hdev);
-
-	/* Point to DRAM */
-	if (!hdev->asic_funcs->set_dram_bar_base)
-		return -EINVAL;
-	if (hdev->asic_funcs->set_dram_bar_base(hdev, dram_base_address) ==
-								U64_MAX)
-		return -EIO;
-
-	/* Outbound Region 0 - Point to Host */
-	host_phys_end_addr = host_phys_base_address + host_phys_size - 1;
 	rc |= hl_pci_iatu_write(hdev, 0x008,
-				lower_32_bits(host_phys_base_address));
+				lower_32_bits(pci_region->addr));
 	rc |= hl_pci_iatu_write(hdev, 0x00C,
-				upper_32_bits(host_phys_base_address));
+				upper_32_bits(pci_region->addr));
-	rc |= hl_pci_iatu_write(hdev, 0x010, lower_32_bits(host_phys_end_addr));
+	rc |= hl_pci_iatu_write(hdev, 0x010,
+				lower_32_bits(outbound_region_end_address));
 	rc |= hl_pci_iatu_write(hdev, 0x014, 0);
 
 	if ((hdev->power9_64bit_dma_enable) && (hdev->dma_mask == 64))
@@ -298,7 +288,8 @@ int hl_pci_init_iatu(struct hl_device *hdev, u64 sram_base_address,
 	else
 		rc |= hl_pci_iatu_write(hdev, 0x018, 0);
 
-	rc |= hl_pci_iatu_write(hdev, 0x020, upper_32_bits(host_phys_end_addr));
+	rc |= hl_pci_iatu_write(hdev, 0x020,
+				upper_32_bits(outbound_region_end_address));
 	/* Increase region size */
 	rc |= hl_pci_iatu_write(hdev, 0x000, 0x00002000);
 	/* Enable */
@@ -308,10 +299,7 @@ int hl_pci_init_iatu(struct hl_device *hdev, u64 sram_base_address,
 	rc |= hl_pci_elbi_write(hdev, prop->pcie_aux_dbi_reg_addr, 0);
 	rc |= hl_pci_elbi_write(hdev, prop->pcie_aux_dbi_reg_addr + 4, 0);
 
-	if (rc)
+	return rc;
-		return -EIO;
-
-	return 0;
 }
 
 /**

drivers/misc/habanalabs/common/sysfs.c

@@ -334,6 +334,9 @@ static ssize_t eeprom_read_handler(struct file *filp, struct kobject *kobj,
 	char *data;
 	int rc;
 
+	if (hl_device_disabled_or_in_reset(hdev))
+		return -ENODEV;
+
 	if (!max_size)
 		return -EINVAL;
 

drivers/misc/habanalabs/gaudi/Makefile

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: GPL-2.0-only
-subdir-ccflags-y += -I$(src)
+subdir-ccflags-y += -I$(src)/common
 
 HL_GAUDI_FILES := gaudi/gaudi.o gaudi/gaudi_hwmgr.o gaudi/gaudi_security.o \
 	gaudi/gaudi_coresight.o

drivers/misc/habanalabs/gaudi/gaudiP.h

@@ -10,7 +10,7 @@
 
 #include <uapi/misc/habanalabs.h>
 #include "habanalabs.h"
-#include "include/hl_boot_if.h"
+#include "include/common/hl_boot_if.h"
 #include "include/gaudi/gaudi_packets.h"
 #include "include/gaudi/gaudi.h"
 #include "include/gaudi/gaudi_async_events.h"
@@ -57,6 +57,12 @@
 
 #define GAUDI_DEFAULT_CARD_NAME		"HL2000"
 
+#define GAUDI_MAX_PENDING_CS		1024
+
+#if !IS_MAX_PENDING_CS_VALID(GAUDI_MAX_PENDING_CS)
+#error "GAUDI_MAX_PENDING_CS must be power of 2 and greater than 1"
+#endif
+
 #define PCI_DMA_NUMBER_OF_CHNLS		3
 #define HBM_DMA_NUMBER_OF_CHNLS		5
 #define DMA_NUMBER_OF_CHNLS		(PCI_DMA_NUMBER_OF_CHNLS + \
@@ -117,14 +123,14 @@
 
 /* Internal QMANs PQ sizes */
 
-#define MME_QMAN_LENGTH			64
+#define MME_QMAN_LENGTH			1024
 #define MME_QMAN_SIZE_IN_BYTES		(MME_QMAN_LENGTH * QMAN_PQ_ENTRY_SIZE)
 
-#define HBM_DMA_QMAN_LENGTH		64
+#define HBM_DMA_QMAN_LENGTH		1024
 #define HBM_DMA_QMAN_SIZE_IN_BYTES	\
 				(HBM_DMA_QMAN_LENGTH * QMAN_PQ_ENTRY_SIZE)
 
-#define TPC_QMAN_LENGTH			64
+#define TPC_QMAN_LENGTH			1024
 #define TPC_QMAN_SIZE_IN_BYTES		(TPC_QMAN_LENGTH * QMAN_PQ_ENTRY_SIZE)
 
 #define SRAM_USER_BASE_OFFSET  GAUDI_DRIVER_SRAM_RESERVED_SIZE_FROM_START
@@ -228,7 +234,6 @@ struct gaudi_internal_qman_info {
  *                      engine.
  * @multi_msi_mode: whether we are working in multi MSI single MSI mode.
  *                  Multi MSI is possible only with IOMMU enabled.
- * @ext_queue_idx: helper index for external queues initialization.
  * @mmu_cache_inv_pi: PI for MMU cache invalidation flow. The H/W expects an
  *                    8-bit value so use u8.
  */
@@ -249,7 +254,6 @@ struct gaudi_device {
 	u32				events_stat_aggregate[GAUDI_EVENT_SIZE];
 	u32				hw_cap_initialized;
 	u8				multi_msi_mode;
-	u8				ext_queue_idx;
 	u8				mmu_cache_inv_pi;
 };
 

drivers/misc/habanalabs/gaudi/gaudi_coresight.c

@@ -392,6 +392,7 @@ static int gaudi_config_stm(struct hl_device *hdev,
 {
 	struct hl_debug_params_stm *input;
 	u64 base_reg;
+	u32 frequency;
 	int rc;
 
 	if (params->reg_idx >= ARRAY_SIZE(debug_stm_regs)) {
@@ -420,7 +421,10 @@ static int gaudi_config_stm(struct hl_device *hdev,
 		WREG32(base_reg + 0xE00, lower_32_bits(input->sp_mask));
 		WREG32(base_reg + 0xEF4, input->id);
 		WREG32(base_reg + 0xDF4, 0x80);
-		WREG32(base_reg + 0xE8C, input->frequency);
+		frequency = hdev->asic_prop.psoc_timestamp_frequency;
+		if (frequency == 0)
+			frequency = input->frequency;
+		WREG32(base_reg + 0xE8C, frequency);
 		WREG32(base_reg + 0xE90, 0x7FF);
 
 		/* SW-2176 - SW WA for HW bug */

drivers/misc/habanalabs/goya/goya.c

@@ -337,11 +337,19 @@ static int goya_mmu_set_dram_default_page(struct hl_device *hdev);
 static int goya_mmu_add_mappings_for_device_cpu(struct hl_device *hdev);
 static void goya_mmu_prepare(struct hl_device *hdev, u32 asid);
 
-void goya_get_fixed_properties(struct hl_device *hdev)
+int goya_get_fixed_properties(struct hl_device *hdev)
 {
 	struct asic_fixed_properties *prop = &hdev->asic_prop;
 	int i;
 
+	prop->max_queues = GOYA_QUEUE_ID_SIZE;
+	prop->hw_queues_props = kcalloc(prop->max_queues,
+			sizeof(struct hw_queue_properties),
+			GFP_KERNEL);
+
+	if (!prop->hw_queues_props)
+		return -ENOMEM;
+
 	for (i = 0 ; i < NUMBER_OF_EXT_HW_QUEUES ; i++) {
 		prop->hw_queues_props[i].type = QUEUE_TYPE_EXT;
 		prop->hw_queues_props[i].driver_only = 0;
@@ -361,9 +369,6 @@ void goya_get_fixed_properties(struct hl_device *hdev)
 		prop->hw_queues_props[i].requires_kernel_cb = 0;
 	}
 
-	for (; i < HL_MAX_QUEUES; i++)
-		prop->hw_queues_props[i].type = QUEUE_TYPE_NA;
-
 	prop->completion_queues_count = NUMBER_OF_CMPLT_QUEUES;
 
 	prop->dram_base_address = DRAM_PHYS_BASE;
@@ -426,6 +431,10 @@ void goya_get_fixed_properties(struct hl_device *hdev)
 
 	strncpy(prop->armcp_info.card_name, GOYA_DEFAULT_CARD_NAME,
 		CARD_NAME_MAX_LEN);
+
+	prop->max_pending_cs = GOYA_MAX_PENDING_CS;
+
+	return 0;
 }
 
 /*
@@ -456,6 +465,7 @@ static int goya_pci_bars_map(struct hl_device *hdev)
 static u64 goya_set_ddr_bar_base(struct hl_device *hdev, u64 addr)
 {
 	struct goya_device *goya = hdev->asic_specific;
+	struct hl_inbound_pci_region pci_region;
 	u64 old_addr = addr;
 	int rc;
 
@@ -463,7 +473,10 @@ static u64 goya_set_ddr_bar_base(struct hl_device *hdev, u64 addr)
 		return old_addr;
 
 	/* Inbound Region 1 - Bar 4 - Point to DDR */
-	rc = hl_pci_set_dram_bar_base(hdev, 1, 4, addr);
+	pci_region.mode = PCI_BAR_MATCH_MODE;
+	pci_region.bar = DDR_BAR_ID;
+	pci_region.addr = addr;
+	rc = hl_pci_set_inbound_region(hdev, 1, &pci_region);
 	if (rc)
 		return U64_MAX;
 
@@ -485,8 +498,35 @@ static u64 goya_set_ddr_bar_base(struct hl_device *hdev, u64 addr)
  */
 static int goya_init_iatu(struct hl_device *hdev)
 {
-	return hl_pci_init_iatu(hdev, SRAM_BASE_ADDR, DRAM_PHYS_BASE,
+	struct hl_inbound_pci_region inbound_region;
-				HOST_PHYS_BASE, HOST_PHYS_SIZE);
+	struct hl_outbound_pci_region outbound_region;
+	int rc;
+
+	/* Inbound Region 0 - Bar 0 - Point to SRAM and CFG */
+	inbound_region.mode = PCI_BAR_MATCH_MODE;
+	inbound_region.bar = SRAM_CFG_BAR_ID;
+	inbound_region.addr = SRAM_BASE_ADDR;
+	rc = hl_pci_set_inbound_region(hdev, 0, &inbound_region);
+	if (rc)
+		goto done;
+
+	/* Inbound Region 1 - Bar 4 - Point to DDR */
+	inbound_region.mode = PCI_BAR_MATCH_MODE;
+	inbound_region.bar = DDR_BAR_ID;
+	inbound_region.addr = DRAM_PHYS_BASE;
+	rc = hl_pci_set_inbound_region(hdev, 1, &inbound_region);
+	if (rc)
+		goto done;
+
+	hdev->asic_funcs->set_dma_mask_from_fw(hdev);
+
+	/* Outbound Region 0 - Point to Host  */
+	outbound_region.addr = HOST_PHYS_BASE;
+	outbound_region.size = HOST_PHYS_SIZE;
+	rc = hl_pci_set_outbound_region(hdev, &outbound_region);
+
+done:
+	return rc;
 }
 
 /*
@@ -507,7 +547,11 @@ static int goya_early_init(struct hl_device *hdev)
 	u32 val;
 	int rc;
 
-	goya_get_fixed_properties(hdev);
+	rc = goya_get_fixed_properties(hdev);
+	if (rc) {
+		dev_err(hdev->dev, "Failed to get fixed properties\n");
+		return rc;
+	}
 
 	/* Check BAR sizes */
 	if (pci_resource_len(pdev, SRAM_CFG_BAR_ID) != CFG_BAR_SIZE) {
@@ -517,7 +561,8 @@ static int goya_early_init(struct hl_device *hdev)
 			(unsigned long long) pci_resource_len(pdev,
 							SRAM_CFG_BAR_ID),
 			CFG_BAR_SIZE);
-		return -ENODEV;
+		rc = -ENODEV;
+		goto free_queue_props;
 	}
 
 	if (pci_resource_len(pdev, MSIX_BAR_ID) != MSIX_BAR_SIZE) {
@@ -527,14 +572,15 @@ static int goya_early_init(struct hl_device *hdev)
 			(unsigned long long) pci_resource_len(pdev,
 								MSIX_BAR_ID),
 			MSIX_BAR_SIZE);
-		return -ENODEV;
+		rc = -ENODEV;
+		goto free_queue_props;
 	}
 
 	prop->dram_pci_bar_size = pci_resource_len(pdev, DDR_BAR_ID);
 
 	rc = hl_pci_init(hdev);
 	if (rc)
-		return rc;
+		goto free_queue_props;
 
 	if (!hdev->pldm) {
 		val = RREG32(mmPSOC_GLOBAL_CONF_BOOT_STRAP_PINS);
@@ -544,6 +590,10 @@ static int goya_early_init(struct hl_device *hdev)
 	}
 
 	return 0;
+
+free_queue_props:
+	kfree(hdev->asic_prop.hw_queues_props);
+	return rc;
 }
 
 /*
@@ -556,6 +606,7 @@ static int goya_early_init(struct hl_device *hdev)
  */
 static int goya_early_fini(struct hl_device *hdev)
 {
+	kfree(hdev->asic_prop.hw_queues_props);
 	hl_pci_fini(hdev);
 
 	return 0;
@@ -592,11 +643,36 @@ static void goya_qman0_set_security(struct hl_device *hdev, bool secure)
 static void goya_fetch_psoc_frequency(struct hl_device *hdev)
 {
 	struct asic_fixed_properties *prop = &hdev->asic_prop;
+	u32 trace_freq = 0;
+	u32 pll_clk = 0;
+	u32 div_fctr = RREG32(mmPSOC_PCI_PLL_DIV_FACTOR_1);
+	u32 div_sel = RREG32(mmPSOC_PCI_PLL_DIV_SEL_1);
+	u32 nr = RREG32(mmPSOC_PCI_PLL_NR);
+	u32 nf = RREG32(mmPSOC_PCI_PLL_NF);
+	u32 od = RREG32(mmPSOC_PCI_PLL_OD);
 
-	prop->psoc_pci_pll_nr = RREG32(mmPSOC_PCI_PLL_NR);
+	if (div_sel == DIV_SEL_REF_CLK || div_sel == DIV_SEL_DIVIDED_REF) {
-	prop->psoc_pci_pll_nf = RREG32(mmPSOC_PCI_PLL_NF);
+		if (div_sel == DIV_SEL_REF_CLK)
-	prop->psoc_pci_pll_od = RREG32(mmPSOC_PCI_PLL_OD);
+			trace_freq = PLL_REF_CLK;
-	prop->psoc_pci_pll_div_factor = RREG32(mmPSOC_PCI_PLL_DIV_FACTOR_1);
+		else
+			trace_freq = PLL_REF_CLK / (div_fctr + 1);
+	} else if (div_sel == DIV_SEL_PLL_CLK ||
+					div_sel == DIV_SEL_DIVIDED_PLL) {
+		pll_clk = PLL_REF_CLK * (nf + 1) / ((nr + 1) * (od + 1));
+		if (div_sel == DIV_SEL_PLL_CLK)
+			trace_freq = pll_clk;
+		else
+			trace_freq = pll_clk / (div_fctr + 1);
+	} else {
+		dev_warn(hdev->dev,
+			"Received invalid div select value: %d", div_sel);
+	}
+
+	prop->psoc_timestamp_frequency = trace_freq;
+	prop->psoc_pci_pll_nr = nr;
+	prop->psoc_pci_pll_nf = nf;
+	prop->psoc_pci_pll_od = od;
+	prop->psoc_pci_pll_div_factor = div_fctr;
 }
 
 int goya_late_init(struct hl_device *hdev)
@@ -2164,29 +2240,15 @@ static void goya_disable_timestamp(struct hl_device *hdev)
 
 static void goya_halt_engines(struct hl_device *hdev, bool hard_reset)
 {
-	u32 wait_timeout_ms, cpu_timeout_ms;
+	u32 wait_timeout_ms;
 
 	dev_info(hdev->dev,
 		"Halting compute engines and disabling interrupts\n");
 
-	if (hdev->pldm) {
+	if (hdev->pldm)
 		wait_timeout_ms = GOYA_PLDM_RESET_WAIT_MSEC;
-		cpu_timeout_ms = GOYA_PLDM_RESET_WAIT_MSEC;
+	else
-	} else {
 		wait_timeout_ms = GOYA_RESET_WAIT_MSEC;
-		cpu_timeout_ms = GOYA_CPU_RESET_WAIT_MSEC;
-	}
-
-	if (hard_reset) {
-		/*
-		 * I don't know what is the state of the CPU so make sure it is
-		 * stopped in any means necessary
-		 */
-		WREG32(mmPSOC_GLOBAL_CONF_UBOOT_MAGIC, KMD_MSG_GOTO_WFE);
-		WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR,
-			GOYA_ASYNC_EVENT_ID_HALT_MACHINE);
-		msleep(cpu_timeout_ms);
-	}
 
 	goya_stop_external_queues(hdev);
 	goya_stop_internal_queues(hdev);
@@ -2491,14 +2553,26 @@ disable_queues:
 static void goya_hw_fini(struct hl_device *hdev, bool hard_reset)
 {
 	struct goya_device *goya = hdev->asic_specific;
-	u32 reset_timeout_ms, status;
+	u32 reset_timeout_ms, cpu_timeout_ms, status;
 
-	if (hdev->pldm)
+	if (hdev->pldm) {
 		reset_timeout_ms = GOYA_PLDM_RESET_TIMEOUT_MSEC;
-	else
+		cpu_timeout_ms = GOYA_PLDM_RESET_WAIT_MSEC;
+	} else {
 		reset_timeout_ms = GOYA_RESET_TIMEOUT_MSEC;
+		cpu_timeout_ms = GOYA_CPU_RESET_WAIT_MSEC;
+	}
 
 	if (hard_reset) {
+		/* I don't know what is the state of the CPU so make sure it is
+		 * stopped in any means necessary
+		 */
+		WREG32(mmPSOC_GLOBAL_CONF_UBOOT_MAGIC, KMD_MSG_GOTO_WFE);
+		WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR,
+			GOYA_ASYNC_EVENT_ID_HALT_MACHINE);
+
+		msleep(cpu_timeout_ms);
+
 		goya_set_ddr_bar_base(hdev, DRAM_PHYS_BASE);
 		goya_disable_clk_rlx(hdev);
 		goya_set_pll_refclk(hdev);
@@ -3697,9 +3771,8 @@ static int goya_parse_cb_mmu(struct hl_device *hdev,
 	parser->patched_cb_size = parser->user_cb_size +
 			sizeof(struct packet_msg_prot) * 2;
 
-	rc = hl_cb_create(hdev, &hdev->kernel_cb_mgr,
+	rc = hl_cb_create(hdev, &hdev->kernel_cb_mgr, parser->patched_cb_size,
-				parser->patched_cb_size,
+			&patched_cb_handle, HL_KERNEL_ASID_ID, false);
-				&patched_cb_handle, HL_KERNEL_ASID_ID);
 
 	if (rc) {
 		dev_err(hdev->dev,
@@ -3771,9 +3844,8 @@ static int goya_parse_cb_no_mmu(struct hl_device *hdev,
 	if (rc)
 		goto free_userptr;
 
-	rc = hl_cb_create(hdev, &hdev->kernel_cb_mgr,
+	rc = hl_cb_create(hdev, &hdev->kernel_cb_mgr, parser->patched_cb_size,
-				parser->patched_cb_size,
+			&patched_cb_handle, HL_KERNEL_ASID_ID, false);
-				&patched_cb_handle, HL_KERNEL_ASID_ID);
 	if (rc) {
 		dev_err(hdev->dev,
 			"Failed to allocate patched CB for DMA CS %d\n", rc);
@@ -4619,7 +4691,7 @@ static int goya_memset_device_memory(struct hl_device *hdev, u64 addr, u64 size,
 	lin_dma_pkts_cnt = DIV_ROUND_UP_ULL(size, SZ_2G);
 	cb_size = lin_dma_pkts_cnt * sizeof(struct packet_lin_dma) +
 						sizeof(struct packet_msg_prot);
-	cb = hl_cb_kernel_create(hdev, cb_size);
+	cb = hl_cb_kernel_create(hdev, cb_size, false);
 	if (!cb)
 		return -ENOMEM;
 
@@ -5149,21 +5221,16 @@ static enum hl_device_hw_state goya_get_hw_state(struct hl_device *hdev)
 	return RREG32(mmHW_STATE);
 }
 
+int goya_ctx_init(struct hl_ctx *ctx)
+{
+	return 0;
+}
+
 u32 goya_get_queue_id_for_cq(struct hl_device *hdev, u32 cq_idx)
 {
 	return cq_idx;
 }
 
-static void goya_ext_queue_init(struct hl_device *hdev, u32 q_idx)
-{
-
-}
-
-static void goya_ext_queue_reset(struct hl_device *hdev, u32 q_idx)
-{
-
-}
-
 static u32 goya_get_signal_cb_size(struct hl_device *hdev)
 {
 	return 0;
@@ -5272,13 +5339,12 @@ static const struct hl_asic_funcs goya_funcs = {
 	.rreg = hl_rreg,
 	.wreg = hl_wreg,
 	.halt_coresight = goya_halt_coresight,
+	.ctx_init = goya_ctx_init,
 	.get_clk_rate = goya_get_clk_rate,
 	.get_queue_id_for_cq = goya_get_queue_id_for_cq,
 	.read_device_fw_version = goya_read_device_fw_version,
 	.load_firmware_to_device = goya_load_firmware_to_device,
 	.load_boot_fit_to_device = goya_load_boot_fit_to_device,
-	.ext_queue_init = goya_ext_queue_init,
-	.ext_queue_reset = goya_ext_queue_reset,
 	.get_signal_cb_size = goya_get_signal_cb_size,
 	.get_wait_cb_size = goya_get_wait_cb_size,
 	.gen_signal_cb = goya_gen_signal_cb,

drivers/misc/habanalabs/goya/goyaP.h

@@ -10,7 +10,7 @@
 
 #include <uapi/misc/habanalabs.h>
 #include "habanalabs.h"
-#include "include/hl_boot_if.h"
+#include "include/common/hl_boot_if.h"
 #include "include/goya/goya_packets.h"
 #include "include/goya/goya.h"
 #include "include/goya/goya_async_events.h"
@@ -31,10 +31,6 @@
  */
 #define NUMBER_OF_INTERRUPTS		(NUMBER_OF_CMPLT_QUEUES + 1)
 
-#if (NUMBER_OF_HW_QUEUES >= HL_MAX_QUEUES)
-#error "Number of H/W queues must be smaller than HL_MAX_QUEUES"
-#endif
-
 #if (NUMBER_OF_INTERRUPTS > GOYA_MSIX_ENTRIES)
 #error "Number of MSIX interrupts must be smaller or equal to GOYA_MSIX_ENTRIES"
 #endif
@@ -57,6 +53,12 @@
 
 #define GOYA_DEFAULT_CARD_NAME		"HL1000"
 
+#define GOYA_MAX_PENDING_CS		64
+
+#if !IS_MAX_PENDING_CS_VALID(GOYA_MAX_PENDING_CS)
+#error "GOYA_MAX_PENDING_CS must be power of 2 and greater than 1"
+#endif
+
 /* DRAM Memory Map */
 
 #define CPU_FW_IMAGE_SIZE		0x10000000	/* 256MB */
@@ -164,7 +166,7 @@ struct goya_device {
 	u8		device_cpu_mmu_mappings_done;
 };
 
-void goya_get_fixed_properties(struct hl_device *hdev);
+int goya_get_fixed_properties(struct hl_device *hdev);
 int goya_mmu_init(struct hl_device *hdev);
 void goya_init_dma_qmans(struct hl_device *hdev);
 void goya_init_mme_qmans(struct hl_device *hdev);

drivers/misc/habanalabs/goya/goya_coresight.c

@@ -232,6 +232,7 @@ static int goya_config_stm(struct hl_device *hdev,
 {
 	struct hl_debug_params_stm *input;
 	u64 base_reg;
+	u32 frequency;
 	int rc;
 
 	if (params->reg_idx >= ARRAY_SIZE(debug_stm_regs)) {
@@ -264,7 +265,10 @@ static int goya_config_stm(struct hl_device *hdev,
 		WREG32(base_reg + 0xE20, 0xFFFFFFFF);
 		WREG32(base_reg + 0xEF4, input->id);
 		WREG32(base_reg + 0xDF4, 0x80);
-		WREG32(base_reg + 0xE8C, input->frequency);
+		frequency = hdev->asic_prop.psoc_timestamp_frequency;
+		if (frequency == 0)
+			frequency = input->frequency;
+		WREG32(base_reg + 0xE8C, frequency);
 		WREG32(base_reg + 0xE90, 0x7FF);
 		WREG32(base_reg + 0xE80, 0x27 | (input->id << 16));
 	} else {

drivers/misc/habanalabs/include/common/armcp_if.h

@@ -19,10 +19,20 @@ struct hl_eq_header {
 	__le32 ctl;
 };
 
+struct hl_eq_ecc_data {
+	__le64 ecc_address;
+	__le64 ecc_syndrom;
+	__u8 memory_wrapper_idx;
+	__u8 pad[7];
+};
+
 struct hl_eq_entry {
 	struct hl_eq_header hdr;
+	union {
+		struct hl_eq_ecc_data ecc_data;
 		__le64 data[7];
 	};
+};
 
 #define HL_EQ_ENTRY_SIZE		sizeof(struct hl_eq_entry)
 
@@ -276,6 +286,8 @@ struct armcp_packet {
 		/* For get Armcp info/EEPROM data */
 		__le32 data_max_size;
 	};
+
+	__le32 reserved;
 };
 
 struct armcp_unmask_irq_arr_packet {

drivers/misc/habanalabs/include/common/hl_boot_if.h

@@ -44,6 +44,15 @@
  *					The NIC FW loading and initialization
  *					failed. This means NICs are not usable.
  *
+ * CPU_BOOT_ERR0_SECURITY_NOT_RDY	Chip security initialization has been
+ *					started, but is not ready yet - chip
+ *					cannot be accessed.
+ *
+ * CPU_BOOT_ERR0_SECURITY_FAIL		Security related tasks have failed.
+ *					The tasks are security init (root of
+ *					trust), boot authentication (chain of
+ *					trust), data packets authentication.
+ *
  * CPU_BOOT_ERR0_ENABLED		Error registers enabled.
  *					This is a main indication that the
  *					running FW populates the error
@@ -57,6 +66,8 @@
 #define CPU_BOOT_ERR0_BMC_WAIT_SKIPPED		(1 << 4)
 #define CPU_BOOT_ERR0_NIC_DATA_NOT_RDY		(1 << 5)
 #define CPU_BOOT_ERR0_NIC_FW_FAIL		(1 << 6)
+#define CPU_BOOT_ERR0_SECURITY_NOT_RDY		(1 << 7)
+#define CPU_BOOT_ERR0_SECURITY_FAIL		(1 << 8)
 #define CPU_BOOT_ERR0_ENABLED			(1 << 31)
 
 enum cpu_boot_status {
@@ -79,7 +90,10 @@ enum cpu_boot_status {
 	CPU_BOOT_STATUS_BMC_WAITING_SKIPPED, /* deprecated - will be removed */
 	/* Last boot loader progress status, ready to receive commands */
 	CPU_BOOT_STATUS_READY_TO_BOOT = 15,
+	/* Internal Boot finished, ready for boot-fit */
 	CPU_BOOT_STATUS_WAITING_FOR_BOOT_FIT = 16,
+	/* Internal Security has been initialized, device can be accessed */
+	CPU_BOOT_STATUS_SECURITY_READY = 17,
 };
 
 enum kmd_msg {

drivers/misc/habanalabs/include/gaudi/asic_reg/gaudi_regs.h

@@ -91,18 +91,16 @@
 
 #include "psoc_pci_pll_regs.h"
 #include "psoc_hbm_pll_regs.h"
+#include "psoc_cpu_pll_regs.h"
 
 #define GAUDI_ECC_MEM_SEL_OFFSET		0xF18
 #define GAUDI_ECC_ADDRESS_OFFSET		0xF1C
 #define GAUDI_ECC_SYNDROME_OFFSET		0xF20
+#define GAUDI_ECC_MEM_INFO_CLR_OFFSET		0xF28
+#define GAUDI_ECC_MEM_INFO_CLR_SERR_MASK	BIT(8)
+#define GAUDI_ECC_MEM_INFO_CLR_DERR_MASK	BIT(9)
 #define GAUDI_ECC_SERR0_OFFSET			0xF30
-#define GAUDI_ECC_SERR1_OFFSET		0xF34
-#define GAUDI_ECC_SERR2_OFFSET		0xF38
-#define GAUDI_ECC_SERR3_OFFSET		0xF3C
 #define GAUDI_ECC_DERR0_OFFSET			0xF40
-#define GAUDI_ECC_DERR1_OFFSET		0xF44
-#define GAUDI_ECC_DERR2_OFFSET		0xF48
-#define GAUDI_ECC_DERR3_OFFSET		0xF4C
 
 #define mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_SOB_OBJ_0                     0x492000
 #define mmSYNC_MNGR_W_S_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0               0x494000
@@ -294,6 +292,7 @@
 
 #define mmPCIE_DBI_DEVICE_ID_VENDOR_ID_REG                           0xC02000
 
+#define mmPCIE_AUX_FLR_CTRL                                          0xC07394
 #define mmPCIE_AUX_DBI                                               0xC07490
 
 #endif /* ASIC_REG_GAUDI_REGS_H_ */

drivers/misc/habanalabs/include/gaudi/asic_reg/psoc_cpu_pll_regs.h

@@ -0,0 +1,114 @@
+/* SPDX-License-Identifier: GPL-2.0
+ *
+ * Copyright 2016-2018 HabanaLabs, Ltd.
+ * All Rights Reserved.
+ *
+ */
+
+/************************************
+ ** This is an auto-generated file **
+ **       DO NOT EDIT BELOW        **
+ ************************************/
+
+#ifndef ASIC_REG_PSOC_CPU_PLL_REGS_H_
+#define ASIC_REG_PSOC_CPU_PLL_REGS_H_
+
+/*
+ *****************************************
+ *   PSOC_CPU_PLL (Prototype: PLL)
+ *****************************************
+ */
+
+#define mmPSOC_CPU_PLL_NR                                            0xC70100
+
+#define mmPSOC_CPU_PLL_NF                                            0xC70104
+
+#define mmPSOC_CPU_PLL_OD                                            0xC70108
+
+#define mmPSOC_CPU_PLL_NB                                            0xC7010C
+
+#define mmPSOC_CPU_PLL_CFG                                           0xC70110
+
+#define mmPSOC_CPU_PLL_LOSE_MASK                                     0xC70120
+
+#define mmPSOC_CPU_PLL_LOCK_INTR                                     0xC70128
+
+#define mmPSOC_CPU_PLL_LOCK_BYPASS                                   0xC7012C
+
+#define mmPSOC_CPU_PLL_DATA_CHNG                                     0xC70130
+
+#define mmPSOC_CPU_PLL_RST                                           0xC70134
+
+#define mmPSOC_CPU_PLL_SLIP_WD_CNTR                                  0xC70150
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_0                                  0xC70200
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_1                                  0xC70204
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_2                                  0xC70208
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_3                                  0xC7020C
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_CMD_0                              0xC70220
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_CMD_1                              0xC70224
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_CMD_2                              0xC70228
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_CMD_3                              0xC7022C
+
+#define mmPSOC_CPU_PLL_DIV_SEL_0                                     0xC70280
+
+#define mmPSOC_CPU_PLL_DIV_SEL_1                                     0xC70284
+
+#define mmPSOC_CPU_PLL_DIV_SEL_2                                     0xC70288
+
+#define mmPSOC_CPU_PLL_DIV_SEL_3                                     0xC7028C
+
+#define mmPSOC_CPU_PLL_DIV_EN_0                                      0xC702A0
+
+#define mmPSOC_CPU_PLL_DIV_EN_1                                      0xC702A4
+
+#define mmPSOC_CPU_PLL_DIV_EN_2                                      0xC702A8
+
+#define mmPSOC_CPU_PLL_DIV_EN_3                                      0xC702AC
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_BUSY_0                             0xC702C0
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_BUSY_1                             0xC702C4
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_BUSY_2                             0xC702C8
+
+#define mmPSOC_CPU_PLL_DIV_FACTOR_BUSY_3                             0xC702CC
+
+#define mmPSOC_CPU_PLL_CLK_GATER                                     0xC70300
+
+#define mmPSOC_CPU_PLL_CLK_RLX_0                                     0xC70310
+
+#define mmPSOC_CPU_PLL_CLK_RLX_1                                     0xC70314
+
+#define mmPSOC_CPU_PLL_CLK_RLX_2                                     0xC70318
+
+#define mmPSOC_CPU_PLL_CLK_RLX_3                                     0xC7031C
+
+#define mmPSOC_CPU_PLL_REF_CNTR_PERIOD                               0xC70400
+
+#define mmPSOC_CPU_PLL_REF_LOW_THRESHOLD                             0xC70410
+
+#define mmPSOC_CPU_PLL_REF_HIGH_THRESHOLD                            0xC70420
+
+#define mmPSOC_CPU_PLL_PLL_NOT_STABLE                                0xC70430
+
+#define mmPSOC_CPU_PLL_FREQ_CALC_EN                                  0xC70440
+
+#define mmPSOC_CPU_PLL_RLX_BITMAP_CFG                                0xC70500
+
+#define mmPSOC_CPU_PLL_RLX_BITMAP_0                                  0xC70510
+
+#define mmPSOC_CPU_PLL_RLX_BITMAP_1                                  0xC70514
+
+#define mmPSOC_CPU_PLL_RLX_BITMAP_2                                  0xC70518
+
+#define mmPSOC_CPU_PLL_RLX_BITMAP_3                                  0xC7051C
+
+#endif /* ASIC_REG_PSOC_CPU_PLL_REGS_H_ */

drivers/misc/habanalabs/include/gaudi/gaudi_masks.h

@@ -455,4 +455,7 @@ enum axi_id {
 					QM_ARB_ERR_MSG_EN_CHOISE_WDT_MASK |\
 					QM_ARB_ERR_MSG_EN_AXI_LBW_ERR_MASK)
 
+#define PCIE_AUX_FLR_CTRL_HW_CTRL_MASK                               0x1
+#define PCIE_AUX_FLR_CTRL_INT_MASK_MASK                              0x2
+
 #endif /* GAUDI_MASKS_H_ */

drivers/misc/habanalabs/include/gaudi/gaudi_packets.h

@@ -85,7 +85,7 @@ struct packet_msg_long {
 };
 
 #define GAUDI_PKT_SHORT_VAL_SOB_SYNC_VAL_SHIFT	0
-#define GAUDI_PKT_SHORT_VAL_SOB_SYNC_VAL_MASK	0x0000EFFF
+#define GAUDI_PKT_SHORT_VAL_SOB_SYNC_VAL_MASK	0x00007FFF
 
 #define GAUDI_PKT_SHORT_VAL_SOB_MOD_SHIFT	31
 #define GAUDI_PKT_SHORT_VAL_SOB_MOD_MASK	0x80000000
@@ -141,7 +141,7 @@ struct packet_msg_prot {
 #define GAUDI_PKT_FENCE_CFG_TARGET_VAL_MASK	0x00FF0000
 
 #define GAUDI_PKT_FENCE_CFG_ID_SHIFT		30
-#define GAUDI_PKT_FENCE_CFG_ID_MASK		0xC000000
+#define GAUDI_PKT_FENCE_CFG_ID_MASK		0xC0000000
 
 #define GAUDI_PKT_FENCE_CTL_PRED_SHIFT		0
 #define GAUDI_PKT_FENCE_CTL_PRED_MASK		0x0000001F

include/uapi/misc/habanalabs.h

@@ -263,6 +263,7 @@ enum hl_device_status {
  *                         time the driver was loaded.
  * HL_INFO_TIME_SYNC     - Retrieve the device's time alongside the host's time
  *                         for synchronization.
+ * HL_INFO_CS_COUNTERS   - Retrieve command submission counters
  */
 #define HL_INFO_HW_IP_INFO		0
 #define HL_INFO_HW_EVENTS		1
@@ -274,6 +275,7 @@ enum hl_device_status {
 #define HL_INFO_CLK_RATE		8
 #define HL_INFO_RESET_COUNT		9
 #define HL_INFO_TIME_SYNC		10
+#define HL_INFO_CS_COUNTERS		11
 
 #define HL_INFO_VERSION_MAX_LEN	128
 #define HL_INFO_CARD_NAME_MAX_LEN	16
@@ -338,6 +340,25 @@ struct hl_info_time_sync {
 	__u64 host_time;
 };
 
+/**
+ * struct hl_info_cs_counters - command submission counters
+ * @out_of_mem_drop_cnt: dropped due to memory allocation issue
+ * @parsing_drop_cnt: dropped due to error in packet parsing
+ * @queue_full_drop_cnt: dropped due to queue full
+ * @device_in_reset_drop_cnt: dropped due to device in reset
+ */
+struct hl_cs_counters {
+	__u64 out_of_mem_drop_cnt;
+	__u64 parsing_drop_cnt;
+	__u64 queue_full_drop_cnt;
+	__u64 device_in_reset_drop_cnt;
+};
+
+struct hl_info_cs_counters {
+	struct hl_cs_counters cs_counters;
+	struct hl_cs_counters ctx_cs_counters;
+};
+
 struct hl_info_args {
 	/* Location of relevant struct in userspace */
 	__u64 return_pointer;
@@ -530,13 +551,13 @@ union hl_wait_cs_args {
 	struct hl_wait_cs_out out;
 };
 
-/* Opcode to alloc device memory */
+/* Opcode to allocate device memory */
 #define HL_MEM_OP_ALLOC			0
 /* Opcode to free previously allocated device memory */
 #define HL_MEM_OP_FREE			1
-/* Opcode to map host memory */
+/* Opcode to map host and device memory */
 #define HL_MEM_OP_MAP			2
-/* Opcode to unmap previously mapped host memory */
+/* Opcode to unmap previously mapped host and device memory */
 #define HL_MEM_OP_UNMAP			3
 
 /* Memory flags */