Merge tag 'char-misc-5.15-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc

Pull habanalabs updates from Greg KH:
 "Here is another round of misc driver patches for 5.15-rc1.

  In here is only updates for the Habanalabs driver. This request is
  late because the previously-objected-to dma-buf patches are all
  removed and some fixes that you and others found are now included in
  here as well.

  All of these have been in linux-next for well over a week with no
  reports of problems, and they are all self-contained to only this one
  driver. Full details are in the shortlog"

* tag 'char-misc-5.15-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc: (61 commits)
  habanalabs/gaudi: hwmon default card name
  habanalabs: add support for f/w reset
  habanalabs/gaudi: block ICACHE_BASE_ADDERESS_HIGH in TPC
  habanalabs: cannot sleep while holding spinlock
  habanalabs: never copy_from_user inside spinlock
  habanalabs: remove unnecessary device status check
  habanalabs: disable IRQ in user interrupts spinlock
  habanalabs: add "in device creation" status
  habanalabs/gaudi: invalidate PMMU mem cache on init
  habanalabs/gaudi: size should be printed in decimal
  habanalabs/gaudi: define DC POWER for secured PMC
  habanalabs/gaudi: unmask out of bounds SLM access interrupt
  habanalabs: add userptr_lookup node in debugfs
  habanalabs/gaudi: fetch TPC/MME ECC errors from F/W
  habanalabs: modify multi-CS to wait on stream masters
  habanalabs/gaudi: add monitored SOBs to state dump
  habanalabs/gaudi: restore user registers when context opens
  habanalabs/gaudi: increase boot fit timeout
  habanalabs: update to latest firmware headers
  habanalabs/gaudi: minimize number of register reads
  ...

Documentation/ABI/testing/debugfs-driver-habanalabs

@@ -215,6 +215,17 @@ Description:    Sets the skip reset on timeout option for the device. Value of
                 "0" means device will be reset in case some CS has timed out,
                 otherwise it will not be reset.
 
+What:           /sys/kernel/debug/habanalabs/hl<n>/state_dump
+Date:           Oct 2021
+KernelVersion:  5.15
+Contact:        ynudelman@habana.ai
+Description:    Gets the state dump occurring on a CS timeout or failure.
+                State dump is used for debug and is created each time in case of
+                a problem in a CS execution, before reset.
+                Reading from the node returns the newest state dump available.
+                Writing an integer X discards X state dumps, so that the
+                next read would return X+1-st newest state dump.
+
 What:           /sys/kernel/debug/habanalabs/hl<n>/stop_on_err
 Date:           Mar 2020
 KernelVersion:  5.6
@@ -230,6 +241,14 @@ Description:    Displays a list with information about the currently user
                 pointers (user virtual addresses) that are pinned and mapped
                 to DMA addresses
 
+What:           /sys/kernel/debug/habanalabs/hl<n>/userptr_lookup
+Date:           Aug 2021
+KernelVersion:  5.15
+Contact:        ogabbay@kernel.org
+Description:    Allows to search for specific user pointers (user virtual
+                addresses) that are pinned and mapped to DMA addresses, and see
+                their resolution to the specific dma address.
+
 What:           /sys/kernel/debug/habanalabs/hl<n>/vm
 Date:           Jan 2019
 KernelVersion:  5.1

drivers/misc/habanalabs/common/Makefile

@@ -10,4 +10,5 @@ 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/firmware_if.o
+		common/command_submission.o common/firmware_if.o \
+		common/state_dump.o

drivers/misc/habanalabs/common/command_buffer.c

@@ -314,8 +314,6 @@ int hl_cb_create(struct hl_device *hdev, struct hl_cb_mgr *mgr,
 
 	spin_lock(&mgr->cb_lock);
 	rc = idr_alloc(&mgr->cb_handles, cb, 1, 0, GFP_ATOMIC);
-	if (rc < 0)
-		rc = idr_alloc(&mgr->cb_handles, cb, 1, 0, GFP_KERNEL);
 	spin_unlock(&mgr->cb_lock);
 
 	if (rc < 0) {
@@ -552,7 +550,7 @@ int hl_cb_mmap(struct hl_fpriv *hpriv, struct vm_area_struct *vma)
 
 	vma->vm_private_data = cb;
 
-	rc = hdev->asic_funcs->cb_mmap(hdev, vma, cb->kernel_address,
+	rc = hdev->asic_funcs->mmap(hdev, vma, cb->kernel_address,
 					cb->bus_address, cb->size);
 	if (rc) {
 		spin_lock(&cb->lock);

drivers/misc/habanalabs/common/context.c

@@ -9,16 +9,70 @@
 
 #include <linux/slab.h>
 
+void hl_encaps_handle_do_release(struct kref *ref)
+{
+	struct hl_cs_encaps_sig_handle *handle =
+		container_of(ref, struct hl_cs_encaps_sig_handle, refcount);
+	struct hl_ctx *ctx = handle->hdev->compute_ctx;
+	struct hl_encaps_signals_mgr *mgr = &ctx->sig_mgr;
+
+	spin_lock(&mgr->lock);
+	idr_remove(&mgr->handles, handle->id);
+	spin_unlock(&mgr->lock);
+
+	kfree(handle);
+}
+
+static void hl_encaps_handle_do_release_sob(struct kref *ref)
+{
+	struct hl_cs_encaps_sig_handle *handle =
+		container_of(ref, struct hl_cs_encaps_sig_handle, refcount);
+	struct hl_ctx *ctx = handle->hdev->compute_ctx;
+	struct hl_encaps_signals_mgr *mgr = &ctx->sig_mgr;
+
+	/* if we're here, then there was a signals reservation but cs with
+	 * encaps signals wasn't submitted, so need to put refcount
+	 * to hw_sob taken at the reservation.
+	 */
+	hw_sob_put(handle->hw_sob);
+
+	spin_lock(&mgr->lock);
+	idr_remove(&mgr->handles, handle->id);
+	spin_unlock(&mgr->lock);
+
+	kfree(handle);
+}
+
+static void hl_encaps_sig_mgr_init(struct hl_encaps_signals_mgr *mgr)
+{
+	spin_lock_init(&mgr->lock);
+	idr_init(&mgr->handles);
+}
+
+static void hl_encaps_sig_mgr_fini(struct hl_device *hdev,
+			struct hl_encaps_signals_mgr *mgr)
+{
+	struct hl_cs_encaps_sig_handle *handle;
+	struct idr *idp;
+	u32 id;
+
+	idp = &mgr->handles;
+
+	if (!idr_is_empty(idp)) {
+		dev_warn(hdev->dev, "device released while some encaps signals handles are still allocated\n");
+		idr_for_each_entry(idp, handle, id)
+			kref_put(&handle->refcount,
+					hl_encaps_handle_do_release_sob);
+	}
+
+	idr_destroy(&mgr->handles);
+}
+
 static void hl_ctx_fini(struct hl_ctx *ctx)
 {
 	struct hl_device *hdev = ctx->hdev;
 	int i;
 
-	/* Release all allocated pending cb's, those cb's were never
-	 * scheduled so it is safe to release them here
-	 */
-	hl_pending_cb_list_flush(ctx);
-
 	/* Release all allocated HW block mapped list entries and destroy
 	 * the mutex.
 	 */
@@ -53,6 +107,7 @@ static void hl_ctx_fini(struct hl_ctx *ctx)
 		hl_cb_va_pool_fini(ctx);
 		hl_vm_ctx_fini(ctx);
 		hl_asid_free(hdev, ctx->asid);
+		hl_encaps_sig_mgr_fini(hdev, &ctx->sig_mgr);
 
 		/* Scrub both SRAM and DRAM */
 		hdev->asic_funcs->scrub_device_mem(hdev, 0, 0);
@@ -130,9 +185,6 @@ void hl_ctx_free(struct hl_device *hdev, struct hl_ctx *ctx)
 {
 	if (kref_put(&ctx->refcount, hl_ctx_do_release) == 1)
 		return;
-
-	dev_warn(hdev->dev,
-		"user process released device but its command submissions are still executing\n");
 }
 
 int hl_ctx_init(struct hl_device *hdev, struct hl_ctx *ctx, bool is_kernel_ctx)
@@ -144,11 +196,8 @@ int hl_ctx_init(struct hl_device *hdev, struct hl_ctx *ctx, bool is_kernel_ctx)
 	kref_init(&ctx->refcount);
 
 	ctx->cs_sequence = 1;
-	INIT_LIST_HEAD(&ctx->pending_cb_list);
-	spin_lock_init(&ctx->pending_cb_lock);
 	spin_lock_init(&ctx->cs_lock);
 	atomic_set(&ctx->thread_ctx_switch_token, 1);
-	atomic_set(&ctx->thread_pending_cb_token, 1);
 	ctx->thread_ctx_switch_wait_token = 0;
 	ctx->cs_pending = kcalloc(hdev->asic_prop.max_pending_cs,
 				sizeof(struct hl_fence *),
@@ -200,6 +249,8 @@ int hl_ctx_init(struct hl_device *hdev, struct hl_ctx *ctx, bool is_kernel_ctx)
 			goto err_cb_va_pool_fini;
 		}
 
+		hl_encaps_sig_mgr_init(&ctx->sig_mgr);
+
 		dev_dbg(hdev->dev, "create user context %d\n", ctx->asid);
 	}
 
@@ -229,31 +280,86 @@ int hl_ctx_put(struct hl_ctx *ctx)
 	return kref_put(&ctx->refcount, hl_ctx_do_release);
 }
 
-struct hl_fence *hl_ctx_get_fence(struct hl_ctx *ctx, u64 seq)
+/*
+ * hl_ctx_get_fence_locked - get CS fence under CS lock
+ *
+ * @ctx: pointer to the context structure.
+ * @seq: CS sequences number
+ *
+ * @return valid fence pointer on success, NULL if fence is gone, otherwise
+ *         error pointer.
+ *
+ * NOTE: this function shall be called with cs_lock locked
+ */
+static struct hl_fence *hl_ctx_get_fence_locked(struct hl_ctx *ctx, u64 seq)
 {
 	struct asic_fixed_properties *asic_prop = &ctx->hdev->asic_prop;
 	struct hl_fence *fence;
 
-	spin_lock(&ctx->cs_lock);
-
-	if (seq >= ctx->cs_sequence) {
-		spin_unlock(&ctx->cs_lock);
+	if (seq >= ctx->cs_sequence)
 		return ERR_PTR(-EINVAL);
-	}
 
-	if (seq + asic_prop->max_pending_cs < ctx->cs_sequence) {
-		spin_unlock(&ctx->cs_lock);
+	if (seq + asic_prop->max_pending_cs < ctx->cs_sequence)
 		return NULL;
-	}
 
 	fence = ctx->cs_pending[seq & (asic_prop->max_pending_cs - 1)];
 	hl_fence_get(fence);
+	return fence;
+}
+
+struct hl_fence *hl_ctx_get_fence(struct hl_ctx *ctx, u64 seq)
+{
+	struct hl_fence *fence;
+
+	spin_lock(&ctx->cs_lock);
+
+	fence = hl_ctx_get_fence_locked(ctx, seq);
 
 	spin_unlock(&ctx->cs_lock);
 
 	return fence;
 }
 
+/*
+ * hl_ctx_get_fences - get multiple CS fences under the same CS lock
+ *
+ * @ctx: pointer to the context structure.
+ * @seq_arr: array of CS sequences to wait for
+ * @fence: fence array to store the CS fences
+ * @arr_len: length of seq_arr and fence_arr
+ *
+ * @return 0 on success, otherwise non 0 error code
+ */
+int hl_ctx_get_fences(struct hl_ctx *ctx, u64 *seq_arr,
+				struct hl_fence **fence, u32 arr_len)
+{
+	struct hl_fence **fence_arr_base = fence;
+	int i, rc = 0;
+
+	spin_lock(&ctx->cs_lock);
+
+	for (i = 0; i < arr_len; i++, fence++) {
+		u64 seq = seq_arr[i];
+
+		*fence = hl_ctx_get_fence_locked(ctx, seq);
+
+		if (IS_ERR(*fence)) {
+			dev_err(ctx->hdev->dev,
+				"Failed to get fence for CS with seq 0x%llx\n",
+					seq);
+			rc = PTR_ERR(*fence);
+			break;
+		}
+	}
+
+	spin_unlock(&ctx->cs_lock);
+
+	if (rc)
+		hl_fences_put(fence_arr_base, i);
+
+	return rc;
+}
+
 /*
  * hl_ctx_mgr_init - initialize the context manager
  *

drivers/misc/habanalabs/common/debugfs.c

@@ -209,12 +209,12 @@ static int userptr_show(struct seq_file *s, void *data)
 		if (first) {
 			first = false;
 			seq_puts(s, "\n");
-			seq_puts(s, " user virtual address     size             dma dir\n");
+			seq_puts(s, " pid      user virtual address     size             dma dir\n");
 			seq_puts(s, "----------------------------------------------------------\n");
 		}
-		seq_printf(s,
-			"    0x%-14llx      %-10u    %-30s\n",
-			userptr->addr, userptr->size, dma_dir[userptr->dir]);
+		seq_printf(s, " %-7d  0x%-14llx      %-10llu    %-30s\n",
+				userptr->pid, userptr->addr, userptr->size,
+				dma_dir[userptr->dir]);
 	}
 
 	spin_unlock(&dev_entry->userptr_spinlock);
@@ -235,7 +235,7 @@ static int vm_show(struct seq_file *s, void *data)
 	struct hl_vm_hash_node *hnode;
 	struct hl_userptr *userptr;
 	struct hl_vm_phys_pg_pack *phys_pg_pack = NULL;
-	enum vm_type_t *vm_type;
+	enum vm_type *vm_type;
 	bool once = true;
 	u64 j;
 	int i;
@@ -261,7 +261,7 @@ static int vm_show(struct seq_file *s, void *data)
 			if (*vm_type == VM_TYPE_USERPTR) {
 				userptr = hnode->ptr;
 				seq_printf(s,
-					"    0x%-14llx      %-10u\n",
+					"    0x%-14llx      %-10llu\n",
 					hnode->vaddr, userptr->size);
 			} else {
 				phys_pg_pack = hnode->ptr;
@@ -320,6 +320,77 @@ static int vm_show(struct seq_file *s, void *data)
 	return 0;
 }
 
+static int userptr_lookup_show(struct seq_file *s, void *data)
+{
+	struct hl_debugfs_entry *entry = s->private;
+	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
+	struct scatterlist *sg;
+	struct hl_userptr *userptr;
+	bool first = true;
+	u64 total_npages, npages, sg_start, sg_end;
+	dma_addr_t dma_addr;
+	int i;
+
+	spin_lock(&dev_entry->userptr_spinlock);
+
+	list_for_each_entry(userptr, &dev_entry->userptr_list, debugfs_list) {
+		if (dev_entry->userptr_lookup >= userptr->addr &&
+		dev_entry->userptr_lookup < userptr->addr + userptr->size) {
+			total_npages = 0;
+			for_each_sg(userptr->sgt->sgl, sg, userptr->sgt->nents,
+					i) {
+				npages = hl_get_sg_info(sg, &dma_addr);
+				sg_start = userptr->addr +
+					total_npages * PAGE_SIZE;
+				sg_end = userptr->addr +
+					(total_npages + npages) * PAGE_SIZE;
+
+				if (dev_entry->userptr_lookup >= sg_start &&
+				    dev_entry->userptr_lookup < sg_end) {
+					dma_addr += (dev_entry->userptr_lookup -
+							sg_start);
+					if (first) {
+						first = false;
+						seq_puts(s, "\n");
+						seq_puts(s, " user virtual address         dma address       pid        region start     region size\n");
+						seq_puts(s, "---------------------------------------------------------------------------------------\n");
+					}
+					seq_printf(s, " 0x%-18llx  0x%-16llx  %-8u  0x%-16llx %-12llu\n",
+						dev_entry->userptr_lookup,
+						(u64)dma_addr, userptr->pid,
+						userptr->addr, userptr->size);
+				}
+				total_npages += npages;
+			}
+		}
+	}
+
+	spin_unlock(&dev_entry->userptr_spinlock);
+
+	if (!first)
+		seq_puts(s, "\n");
+
+	return 0;
+}
+
+static ssize_t userptr_lookup_write(struct file *file, const char __user *buf,
+		size_t count, loff_t *f_pos)
+{
+	struct seq_file *s = file->private_data;
+	struct hl_debugfs_entry *entry = s->private;
+	struct hl_dbg_device_entry *dev_entry = entry->dev_entry;
+	ssize_t rc;
+	u64 value;
+
+	rc = kstrtoull_from_user(buf, count, 16, &value);
+	if (rc)
+		return rc;
+
+	dev_entry->userptr_lookup = value;
+
+	return count;
+}
+
 static int mmu_show(struct seq_file *s, void *data)
 {
 	struct hl_debugfs_entry *entry = s->private;
@@ -349,7 +420,7 @@ static int mmu_show(struct seq_file *s, void *data)
 		return 0;
 	}
 
-	phys_addr = hops_info.hop_info[hops_info.used_hops - 1].hop_pte_val;
+	hl_mmu_va_to_pa(ctx, virt_addr, &phys_addr);
 
 	if (hops_info.scrambled_vaddr &&
 		(dev_entry->mmu_addr != hops_info.scrambled_vaddr))
@@ -491,11 +562,10 @@ static int device_va_to_pa(struct hl_device *hdev, u64 virt_addr, u32 size,
 	struct hl_vm_phys_pg_pack *phys_pg_pack;
 	struct hl_ctx *ctx = hdev->compute_ctx;
 	struct hl_vm_hash_node *hnode;
+	u64 end_address, range_size;
 	struct hl_userptr *userptr;
-	enum vm_type_t *vm_type;
+	enum vm_type *vm_type;
 	bool valid = false;
-	u64 end_address;
-	u32 range_size;
 	int i, rc = 0;
 
 	if (!ctx) {
@@ -1043,6 +1113,60 @@ static ssize_t hl_security_violations_read(struct file *f, char __user *buf,
 	return 0;
 }
 
+static ssize_t hl_state_dump_read(struct file *f, char __user *buf,
+					size_t count, loff_t *ppos)
+{
+	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
+	ssize_t rc;
+
+	down_read(&entry->state_dump_sem);
+	if (!entry->state_dump[entry->state_dump_head])
+		rc = 0;
+	else
+		rc = simple_read_from_buffer(
+			buf, count, ppos,
+			entry->state_dump[entry->state_dump_head],
+			strlen(entry->state_dump[entry->state_dump_head]));
+	up_read(&entry->state_dump_sem);
+
+	return rc;
+}
+
+static ssize_t hl_state_dump_write(struct file *f, const char __user *buf,
+					size_t count, loff_t *ppos)
+{
+	struct hl_dbg_device_entry *entry = file_inode(f)->i_private;
+	struct hl_device *hdev = entry->hdev;
+	ssize_t rc;
+	u32 size;
+	int i;
+
+	rc = kstrtouint_from_user(buf, count, 10, &size);
+	if (rc)
+		return rc;
+
+	if (size <= 0 || size >= ARRAY_SIZE(entry->state_dump)) {
+		dev_err(hdev->dev, "Invalid number of dumps to skip\n");
+		return -EINVAL;
+	}
+
+	if (entry->state_dump[entry->state_dump_head]) {
+		down_write(&entry->state_dump_sem);
+		for (i = 0; i < size; ++i) {
+			vfree(entry->state_dump[entry->state_dump_head]);
+			entry->state_dump[entry->state_dump_head] = NULL;
+			if (entry->state_dump_head > 0)
+				entry->state_dump_head--;
+			else
+				entry->state_dump_head =
+					ARRAY_SIZE(entry->state_dump) - 1;
+		}
+		up_write(&entry->state_dump_sem);
+	}
+
+	return count;
+}
+
 static const struct file_operations hl_data32b_fops = {
 	.owner = THIS_MODULE,
 	.read = hl_data_read32,
@@ -1110,12 +1234,19 @@ static const struct file_operations hl_security_violations_fops = {
 	.read = hl_security_violations_read
 };
 
+static const struct file_operations hl_state_dump_fops = {
+	.owner = THIS_MODULE,
+	.read = hl_state_dump_read,
+	.write = hl_state_dump_write
+};
+
 static const struct hl_info_list hl_debugfs_list[] = {
 	{"command_buffers", command_buffers_show, NULL},
 	{"command_submission", command_submission_show, NULL},
 	{"command_submission_jobs", command_submission_jobs_show, NULL},
 	{"userptr", userptr_show, NULL},
 	{"vm", vm_show, NULL},
+	{"userptr_lookup", userptr_lookup_show, userptr_lookup_write},
 	{"mmu", mmu_show, mmu_asid_va_write},
 	{"engines", engines_show, NULL}
 };
@@ -1172,6 +1303,7 @@ void hl_debugfs_add_device(struct hl_device *hdev)
 	INIT_LIST_HEAD(&dev_entry->userptr_list);
 	INIT_LIST_HEAD(&dev_entry->ctx_mem_hash_list);
 	mutex_init(&dev_entry->file_mutex);
+	init_rwsem(&dev_entry->state_dump_sem);
 	spin_lock_init(&dev_entry->cb_spinlock);
 	spin_lock_init(&dev_entry->cs_spinlock);
 	spin_lock_init(&dev_entry->cs_job_spinlock);
@@ -1283,6 +1415,12 @@ void hl_debugfs_add_device(struct hl_device *hdev)
 				dev_entry->root,
 				&hdev->skip_reset_on_timeout);
 
+	debugfs_create_file("state_dump",
+				0600,
+				dev_entry->root,
+				dev_entry,
+				&hl_state_dump_fops);
+
 	for (i = 0, entry = dev_entry->entry_arr ; i < count ; i++, entry++) {
 		debugfs_create_file(hl_debugfs_list[i].name,
 					0444,
@@ -1297,6 +1435,7 @@ void hl_debugfs_add_device(struct hl_device *hdev)
 void hl_debugfs_remove_device(struct hl_device *hdev)
 {
 	struct hl_dbg_device_entry *entry = &hdev->hl_debugfs;
+	int i;
 
 	debugfs_remove_recursive(entry->root);
 
@@ -1304,6 +1443,9 @@ void hl_debugfs_remove_device(struct hl_device *hdev)
 
 	vfree(entry->blob_desc.data);
 
+	for (i = 0; i < ARRAY_SIZE(entry->state_dump); ++i)
+		vfree(entry->state_dump[i]);
+
 	kfree(entry->entry_arr);
 }
 
@@ -1416,6 +1558,28 @@ void hl_debugfs_remove_ctx_mem_hash(struct hl_device *hdev, struct hl_ctx *ctx)
 	spin_unlock(&dev_entry->ctx_mem_hash_spinlock);
 }
 
+/**
+ * hl_debugfs_set_state_dump - register state dump making it accessible via
+ *                             debugfs
+ * @hdev: pointer to the device structure
+ * @data: the actual dump data
+ * @length: the length of the data
+ */
+void hl_debugfs_set_state_dump(struct hl_device *hdev, char *data,
+					unsigned long length)
+{
+	struct hl_dbg_device_entry *dev_entry = &hdev->hl_debugfs;
+
+	down_write(&dev_entry->state_dump_sem);
+
+	dev_entry->state_dump_head = (dev_entry->state_dump_head + 1) %
+					ARRAY_SIZE(dev_entry->state_dump);
+	vfree(dev_entry->state_dump[dev_entry->state_dump_head]);
+	dev_entry->state_dump[dev_entry->state_dump_head] = data;
+
+	up_write(&dev_entry->state_dump_sem);
+}
+
 void __init hl_debugfs_init(void)
 {
 	hl_debug_root = debugfs_create_dir("habanalabs", NULL);

drivers/misc/habanalabs/common/device.c

@@ -7,11 +7,11 @@
 
 #define pr_fmt(fmt)			"habanalabs: " fmt
 
+#include <uapi/misc/habanalabs.h>
 #include "habanalabs.h"
 
 #include <linux/pci.h>
 #include <linux/hwmon.h>
-#include <uapi/misc/habanalabs.h>
 
 enum hl_device_status hl_device_status(struct hl_device *hdev)
 {
@@ -23,6 +23,8 @@ enum hl_device_status hl_device_status(struct hl_device *hdev)
 		status = HL_DEVICE_STATUS_NEEDS_RESET;
 	else if (hdev->disabled)
 		status = HL_DEVICE_STATUS_MALFUNCTION;
+	else if (!hdev->init_done)
+		status = HL_DEVICE_STATUS_IN_DEVICE_CREATION;
 	else
 		status = HL_DEVICE_STATUS_OPERATIONAL;
 
@@ -44,6 +46,7 @@ bool hl_device_operational(struct hl_device *hdev,
 	case HL_DEVICE_STATUS_NEEDS_RESET:
 		return false;
 	case HL_DEVICE_STATUS_OPERATIONAL:
+	case HL_DEVICE_STATUS_IN_DEVICE_CREATION:
 	default:
 		return true;
 	}
@@ -129,8 +132,8 @@ static int hl_device_release(struct inode *inode, struct file *filp)
 	hl_ctx_mgr_fini(hdev, &hpriv->ctx_mgr);
 
 	if (!hl_hpriv_put(hpriv))
-		dev_warn(hdev->dev,
-			"Device is still in use because there are live CS and/or memory mappings\n");
+		dev_notice(hdev->dev,
+			"User process closed FD but device still in use\n");
 
 	hdev->last_open_session_duration_jif =
 		jiffies - hdev->last_successful_open_jif;
@@ -308,9 +311,15 @@ static void device_hard_reset_pending(struct work_struct *work)
 		container_of(work, struct hl_device_reset_work,
 				reset_work.work);
 	struct hl_device *hdev = device_reset_work->hdev;
+	u32 flags;
 	int rc;
 
-	rc = hl_device_reset(hdev, HL_RESET_HARD | HL_RESET_FROM_RESET_THREAD);
+	flags = HL_RESET_HARD | HL_RESET_FROM_RESET_THREAD;
+
+	if (device_reset_work->fw_reset)
+		flags |= HL_RESET_FW;
+
+	rc = hl_device_reset(hdev, flags);
 	if ((rc == -EBUSY) && !hdev->device_fini_pending) {
 		dev_info(hdev->dev,
 			"Could not reset device. will try again in %u seconds",
@@ -682,6 +691,44 @@ int hl_device_set_debug_mode(struct hl_device *hdev, bool enable)
 	return rc;
 }
 
+static void take_release_locks(struct hl_device *hdev)
+{
+	/* Flush anyone that is inside the critical section of enqueue
+	 * jobs to the H/W
+	 */
+	hdev->asic_funcs->hw_queues_lock(hdev);
+	hdev->asic_funcs->hw_queues_unlock(hdev);
+
+	/* Flush processes that are sending message to CPU */
+	mutex_lock(&hdev->send_cpu_message_lock);
+	mutex_unlock(&hdev->send_cpu_message_lock);
+
+	/* Flush anyone that is inside device open */
+	mutex_lock(&hdev->fpriv_list_lock);
+	mutex_unlock(&hdev->fpriv_list_lock);
+}
+
+static void cleanup_resources(struct hl_device *hdev, bool hard_reset, bool fw_reset)
+{
+	if (hard_reset)
+		device_late_fini(hdev);
+
+	/*
+	 * Halt the engines and disable interrupts so we won't get any more
+	 * completions from H/W and we won't have any accesses from the
+	 * H/W to the host machine
+	 */
+	hdev->asic_funcs->halt_engines(hdev, hard_reset, fw_reset);
+
+	/* Go over all the queues, release all CS and their jobs */
+	hl_cs_rollback_all(hdev);
+
+	/* Release all pending user interrupts, each pending user interrupt
+	 * holds a reference to user context
+	 */
+	hl_release_pending_user_interrupts(hdev);
+}
+
 /*
  * hl_device_suspend - initiate device suspend
  *
@@ -707,16 +754,7 @@ int hl_device_suspend(struct hl_device *hdev)
 	/* This blocks all other stuff that is not blocked by in_reset */
 	hdev->disabled = true;
 
-	/*
-	 * Flush anyone that is inside the critical section of enqueue
-	 * jobs to the H/W
-	 */
-	hdev->asic_funcs->hw_queues_lock(hdev);
-	hdev->asic_funcs->hw_queues_unlock(hdev);
-
-	/* Flush processes that are sending message to CPU */
-	mutex_lock(&hdev->send_cpu_message_lock);
-	mutex_unlock(&hdev->send_cpu_message_lock);
+	take_release_locks(hdev);
 
 	rc = hdev->asic_funcs->suspend(hdev);
 	if (rc)
@@ -819,6 +857,11 @@ static int device_kill_open_processes(struct hl_device *hdev, u32 timeout)
 			usleep_range(1000, 10000);
 
 			put_task_struct(task);
+		} else {
+			dev_warn(hdev->dev,
+				"Can't get task struct for PID so giving up on killing process\n");
+			mutex_unlock(&hdev->fpriv_list_lock);
+			return -ETIME;
 		}
 	}
 
@@ -885,7 +928,7 @@ static void device_disable_open_processes(struct hl_device *hdev)
 int hl_device_reset(struct hl_device *hdev, u32 flags)
 {
 	u64 idle_mask[HL_BUSY_ENGINES_MASK_EXT_SIZE] = {0};
-	bool hard_reset, from_hard_reset_thread, hard_instead_soft = false;
+	bool hard_reset, from_hard_reset_thread, fw_reset, hard_instead_soft = false;
 	int i, rc;
 
 	if (!hdev->init_done) {
@@ -894,8 +937,9 @@ int hl_device_reset(struct hl_device *hdev, u32 flags)
 		return 0;
 	}
 
-	hard_reset = (flags & HL_RESET_HARD) != 0;
-	from_hard_reset_thread = (flags & HL_RESET_FROM_RESET_THREAD) != 0;
+	hard_reset = !!(flags & HL_RESET_HARD);
+	from_hard_reset_thread = !!(flags & HL_RESET_FROM_RESET_THREAD);
+	fw_reset = !!(flags & HL_RESET_FW);
 
 	if (!hard_reset && !hdev->supports_soft_reset) {
 		hard_instead_soft = true;
@@ -947,11 +991,13 @@ int hl_device_reset(struct hl_device *hdev, u32 flags)
 		else
 			hdev->curr_reset_cause = HL_RESET_CAUSE_UNKNOWN;
 
-		/*
-		 * if reset is due to heartbeat, device CPU is no responsive in
-		 * which case no point sending PCI disable message to it
+		/* If reset is due to heartbeat, device CPU is no responsive in
+		 * which case no point sending PCI disable message to it.
+		 *
+		 * If F/W is performing the reset, no need to send it a message to disable
+		 * PCI access
 		 */
-		if (hard_reset && !(flags & HL_RESET_HEARTBEAT)) {
+		if (hard_reset && !(flags & (HL_RESET_HEARTBEAT | HL_RESET_FW))) {
 			/* 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
@@ -970,15 +1016,7 @@ int hl_device_reset(struct hl_device *hdev, u32 flags)
 		/* This also blocks future CS/VM/JOB completion operations */
 		hdev->disabled = true;
 
-		/* Flush anyone that is inside the critical section of enqueue
-		 * jobs to the H/W
-		 */
-		hdev->asic_funcs->hw_queues_lock(hdev);
-		hdev->asic_funcs->hw_queues_unlock(hdev);
-
-		/* Flush anyone that is inside device open */
-		mutex_lock(&hdev->fpriv_list_lock);
-		mutex_unlock(&hdev->fpriv_list_lock);
+		take_release_locks(hdev);
 
 		dev_err(hdev->dev, "Going to RESET device!\n");
 	}
@@ -989,6 +1027,8 @@ int hl_device_reset(struct hl_device *hdev, u32 flags)
 
 		hdev->process_kill_trial_cnt = 0;
 
+		hdev->device_reset_work.fw_reset = fw_reset;
+
 		/*
 		 * Because the reset function can't run from heartbeat work,
 		 * we need to call the reset function from a dedicated work.
@@ -999,31 +1039,7 @@ int hl_device_reset(struct hl_device *hdev, u32 flags)
 		return 0;
 	}
 
-	if (hard_reset) {
-		device_late_fini(hdev);
-
-		/*
-		 * Now that the heartbeat thread is closed, flush processes
-		 * which are sending messages to CPU
-		 */
-		mutex_lock(&hdev->send_cpu_message_lock);
-		mutex_unlock(&hdev->send_cpu_message_lock);
-	}
-
-	/*
-	 * Halt the engines and disable interrupts so we won't get any more
-	 * completions from H/W and we won't have any accesses from the
-	 * H/W to the host machine
-	 */
-	hdev->asic_funcs->halt_engines(hdev, hard_reset);
-
-	/* Go over all the queues, release all CS and their jobs */
-	hl_cs_rollback_all(hdev);
-
-	/* Release all pending user interrupts, each pending user interrupt
-	 * holds a reference to user context
-	 */
-	hl_release_pending_user_interrupts(hdev);
+	cleanup_resources(hdev, hard_reset, fw_reset);
 
 kill_processes:
 	if (hard_reset) {
@@ -1057,12 +1073,15 @@ int hl_device_reset(struct hl_device *hdev, u32 flags)
 	}
 
 	/* Reset the H/W. It will be in idle state after this returns */
-	hdev->asic_funcs->hw_fini(hdev, hard_reset);
+	hdev->asic_funcs->hw_fini(hdev, hard_reset, fw_reset);
 
 	if (hard_reset) {
+		hdev->fw_loader.linux_loaded = false;
+
 		/* Release kernel context */
 		if (hdev->kernel_ctx && hl_ctx_put(hdev->kernel_ctx) == 1)
 			hdev->kernel_ctx = NULL;
+
 		hl_vm_fini(hdev);
 		hl_mmu_fini(hdev);
 		hl_eq_reset(hdev, &hdev->event_queue);
@@ -1292,6 +1311,10 @@ int hl_device_init(struct hl_device *hdev, struct class *hclass)
 	if (rc)
 		goto user_interrupts_fini;
 
+
+	/* initialize completion structure for multi CS wait */
+	hl_multi_cs_completion_init(hdev);
+
 	/*
 	 * Initialize the H/W queues. Must be done before hw_init, because
 	 * there the addresses of the kernel queue are being written to the
@@ -1361,6 +1384,8 @@ int hl_device_init(struct hl_device *hdev, struct class *hclass)
 
 	hdev->compute_ctx = NULL;
 
+	hdev->asic_funcs->state_dump_init(hdev);
+
 	hl_debugfs_add_device(hdev);
 
 	/* debugfs nodes are created in hl_ctx_init so it must be called after
@@ -1567,31 +1592,13 @@ void hl_device_fini(struct hl_device *hdev)
 	/* Mark device as disabled */
 	hdev->disabled = true;
 
-	/* Flush anyone that is inside the critical section of enqueue
-	 * jobs to the H/W
-	 */
-	hdev->asic_funcs->hw_queues_lock(hdev);
-	hdev->asic_funcs->hw_queues_unlock(hdev);
-
-	/* Flush anyone that is inside device open */
-	mutex_lock(&hdev->fpriv_list_lock);
-	mutex_unlock(&hdev->fpriv_list_lock);
+	take_release_locks(hdev);
 
 	hdev->hard_reset_pending = true;
 
 	hl_hwmon_fini(hdev);
 
-	device_late_fini(hdev);
-
-	/*
-	 * Halt the engines and disable interrupts so we won't get any more
-	 * completions from H/W and we won't have any accesses from the
-	 * H/W to the host machine
-	 */
-	hdev->asic_funcs->halt_engines(hdev, true);
-
-	/* Go over all the queues, release all CS and their jobs */
-	hl_cs_rollback_all(hdev);
+	cleanup_resources(hdev, true, false);
 
 	/* Kill processes here after CS rollback. This is because the process
 	 * can't really exit until all its CSs are done, which is what we
@@ -1610,7 +1617,9 @@ void hl_device_fini(struct hl_device *hdev)
 	hl_cb_pool_fini(hdev);
 
 	/* Reset the H/W. It will be in idle state after this returns */
-	hdev->asic_funcs->hw_fini(hdev, true);
+	hdev->asic_funcs->hw_fini(hdev, true, false);
+
+	hdev->fw_loader.linux_loaded = false;
 
 	/* Release kernel context */
 	if ((hdev->kernel_ctx) && (hl_ctx_put(hdev->kernel_ctx) != 1))

drivers/misc/habanalabs/common/firmware_if.c

 // SPDX-License-Identifier: GPL-2.0
 
 /*
- * Copyright 2016-2019 HabanaLabs, Ltd.
+ * Copyright 2016-2021 HabanaLabs, Ltd.
  * All Rights Reserved.
  */
 
@@ -240,11 +240,15 @@ int hl_fw_send_cpu_message(struct hl_device *hdev, u32 hw_queue_id, u32 *msg,
 	/* set fence to a non valid value */
 	pkt->fence = cpu_to_le32(UINT_MAX);
 
-	rc = hl_hw_queue_send_cb_no_cmpl(hdev, hw_queue_id, len, pkt_dma_addr);
-	if (rc) {
-		dev_err(hdev->dev, "Failed to send CB on CPU PQ (%d)\n", rc);
-		goto out;
-	}
+	/*
+	 * The CPU queue is a synchronous queue with an effective depth of
+	 * a single entry (although it is allocated with room for multiple
+	 * entries). We lock on it using 'send_cpu_message_lock' which
+	 * serializes accesses to the CPU queue.
+	 * Which means that we don't need to lock the access to the entire H/W
+	 * queues module when submitting a JOB to the CPU queue.
+	 */
+	hl_hw_queue_submit_bd(hdev, queue, 0, len, pkt_dma_addr);
 
 	if (prop->fw_app_cpu_boot_dev_sts0 & CPU_BOOT_DEV_STS0_PKT_PI_ACK_EN)
 		expected_ack_val = queue->pi;
@@ -663,17 +667,15 @@ int hl_fw_cpucp_info_get(struct hl_device *hdev,
 	hdev->event_queue.check_eqe_index = false;
 
 	/* Read FW application security bits again */
-	if (hdev->asic_prop.fw_cpu_boot_dev_sts0_valid) {
-		hdev->asic_prop.fw_app_cpu_boot_dev_sts0 =
-						RREG32(sts_boot_dev_sts0_reg);
-		if (hdev->asic_prop.fw_app_cpu_boot_dev_sts0 &
+	if (prop->fw_cpu_boot_dev_sts0_valid) {
+		prop->fw_app_cpu_boot_dev_sts0 = RREG32(sts_boot_dev_sts0_reg);
+		if (prop->fw_app_cpu_boot_dev_sts0 &
 				CPU_BOOT_DEV_STS0_EQ_INDEX_EN)
 			hdev->event_queue.check_eqe_index = true;
 	}
 
-	if (hdev->asic_prop.fw_cpu_boot_dev_sts1_valid)
-		hdev->asic_prop.fw_app_cpu_boot_dev_sts1 =
-						RREG32(sts_boot_dev_sts1_reg);
+	if (prop->fw_cpu_boot_dev_sts1_valid)
+		prop->fw_app_cpu_boot_dev_sts1 = RREG32(sts_boot_dev_sts1_reg);
 
 out:
 	hdev->asic_funcs->cpu_accessible_dma_pool_free(hdev,
@@ -1008,6 +1010,11 @@ void hl_fw_ask_halt_machine_without_linux(struct hl_device *hdev)
 	} else {
 		WREG32(static_loader->kmd_msg_to_cpu_reg, KMD_MSG_GOTO_WFE);
 		msleep(static_loader->cpu_reset_wait_msec);
+
+		/* Must clear this register in order to prevent preboot
+		 * from reading WFE after reboot
+		 */
+		WREG32(static_loader->kmd_msg_to_cpu_reg, KMD_MSG_NA);
 	}
 
 	hdev->device_cpu_is_halted = true;
@@ -1055,6 +1062,10 @@ static void detect_cpu_boot_status(struct hl_device *hdev, u32 status)
 		dev_err(hdev->dev,
 			"Device boot progress - Thermal Sensor initialization failed\n");
 		break;
+	case CPU_BOOT_STATUS_SECURITY_READY:
+		dev_err(hdev->dev,
+			"Device boot progress - Stuck in preboot after security initialization\n");
+		break;
 	default:
 		dev_err(hdev->dev,
 			"Device boot progress - Invalid status code %d\n",
@@ -1238,11 +1249,6 @@ static void hl_fw_preboot_update_state(struct hl_device *hdev)
 	 *               b. Check whether hard reset is done by boot cpu
 	 * 3. FW application - a. Fetch fw application security status
 	 *                     b. Check whether hard reset is done by fw app
-	 *
-	 * Preboot:
-	 * Check security status bit (CPU_BOOT_DEV_STS0_ENABLED). If set, then-
-	 * check security enabled bit (CPU_BOOT_DEV_STS0_SECURITY_EN)
-	 * If set, then mark GIC controller to be disabled.
 	 */
 	prop->hard_reset_done_by_fw =
 		!!(cpu_boot_dev_sts0 & CPU_BOOT_DEV_STS0_FW_HARD_RST_EN);
@@ -1953,8 +1959,8 @@ static void hl_fw_dynamic_update_linux_interrupt_if(struct hl_device *hdev)
 	if (!hdev->asic_prop.gic_interrupts_enable &&
 			!(hdev->asic_prop.fw_app_cpu_boot_dev_sts0 &
 				CPU_BOOT_DEV_STS0_MULTI_IRQ_POLL_EN)) {
-		dyn_regs->gic_host_halt_irq = dyn_regs->gic_host_irq_ctrl;
-		dyn_regs->gic_host_ints_irq = dyn_regs->gic_host_irq_ctrl;
+		dyn_regs->gic_host_halt_irq = dyn_regs->gic_host_pi_upd_irq;
+		dyn_regs->gic_host_ints_irq = dyn_regs->gic_host_pi_upd_irq;
 
 		dev_warn(hdev->dev,
 			"Using a single interrupt interface towards cpucp");
@@ -2122,8 +2128,7 @@ static void hl_fw_linux_update_state(struct hl_device *hdev,
 
 	/* Read FW application security bits */
 	if (prop->fw_cpu_boot_dev_sts0_valid) {
-		prop->fw_app_cpu_boot_dev_sts0 =
-				RREG32(cpu_boot_dev_sts0_reg);
+		prop->fw_app_cpu_boot_dev_sts0 = RREG32(cpu_boot_dev_sts0_reg);
 
 		if (prop->fw_app_cpu_boot_dev_sts0 &
 				CPU_BOOT_DEV_STS0_FW_HARD_RST_EN)
@@ -2143,8 +2148,7 @@ static void hl_fw_linux_update_state(struct hl_device *hdev,
 	}
 
 	if (prop->fw_cpu_boot_dev_sts1_valid) {
-		prop->fw_app_cpu_boot_dev_sts1 =
-				RREG32(cpu_boot_dev_sts1_reg);
+		prop->fw_app_cpu_boot_dev_sts1 = RREG32(cpu_boot_dev_sts1_reg);
 
 		dev_dbg(hdev->dev,
 			"Firmware application CPU status1 %#x\n",
@@ -2235,6 +2239,10 @@ static int hl_fw_dynamic_init_cpu(struct hl_device *hdev,
 	dev_info(hdev->dev,
 		"Loading firmware to device, may take some time...\n");
 
+	/*
+	 * In this stage, "cpu_dyn_regs" contains only LKD's hard coded values!
+	 * It will be updated from FW after hl_fw_dynamic_request_descriptor().
+	 */
 	dyn_regs = &fw_loader->dynamic_loader.comm_desc.cpu_dyn_regs;
 
 	rc = hl_fw_dynamic_send_protocol_cmd(hdev, fw_loader, COMMS_RST_STATE,

drivers/misc/habanalabs/common/habanalabs_drv.c

@@ -141,7 +141,7 @@ int hl_device_open(struct inode *inode, struct file *filp)
 	hl_cb_mgr_init(&hpriv->cb_mgr);
 	hl_ctx_mgr_init(&hpriv->ctx_mgr);
 
-	hpriv->taskpid = find_get_pid(current->pid);
+	hpriv->taskpid = get_task_pid(current, PIDTYPE_PID);
 
 	mutex_lock(&hdev->fpriv_list_lock);
 
@@ -194,7 +194,6 @@ int hl_device_open(struct inode *inode, struct file *filp)
 
 out_err:
 	mutex_unlock(&hdev->fpriv_list_lock);
-
 	hl_cb_mgr_fini(hpriv->hdev, &hpriv->cb_mgr);
 	hl_ctx_mgr_fini(hpriv->hdev, &hpriv->ctx_mgr);
 	filp->private_data = NULL;
@@ -318,12 +317,16 @@ int create_hdev(struct hl_device **dev, struct pci_dev *pdev,
 		hdev->asic_prop.fw_security_enabled = false;
 
 	/* Assign status description string */
-	strncpy(hdev->status[HL_DEVICE_STATUS_MALFUNCTION],
-					"disabled", HL_STR_MAX);
+	strncpy(hdev->status[HL_DEVICE_STATUS_OPERATIONAL],
+					"operational", HL_STR_MAX);
 	strncpy(hdev->status[HL_DEVICE_STATUS_IN_RESET],
 					"in reset", HL_STR_MAX);
+	strncpy(hdev->status[HL_DEVICE_STATUS_MALFUNCTION],
+					"disabled", HL_STR_MAX);
 	strncpy(hdev->status[HL_DEVICE_STATUS_NEEDS_RESET],
 					"needs reset", HL_STR_MAX);
+	strncpy(hdev->status[HL_DEVICE_STATUS_IN_DEVICE_CREATION],
+					"in device creation", HL_STR_MAX);
 
 	hdev->major = hl_major;
 	hdev->reset_on_lockup = reset_on_lockup;
@@ -532,7 +535,7 @@ hl_pci_err_detected(struct pci_dev *pdev, pci_channel_state_t state)
 		result = PCI_ERS_RESULT_NONE;
 	}
 
-	hdev->asic_funcs->halt_engines(hdev, true);
+	hdev->asic_funcs->halt_engines(hdev, true, false);
 
 	return result;
 }

drivers/misc/habanalabs/common/habanalabs_ioctl.c

@@ -94,6 +94,8 @@ static int hw_ip_info(struct hl_device *hdev, struct hl_info_args *args)
 
 	hw_ip.first_available_interrupt_id =
 			prop->first_available_user_msix_interrupt;
+	hw_ip.server_type = prop->server_type;
+
 	return copy_to_user(out, &hw_ip,
 		min((size_t) size, sizeof(hw_ip))) ? -EFAULT : 0;
 }

drivers/misc/habanalabs/common/mmu/mmu_v1.c

@@ -470,13 +470,13 @@ static void hl_mmu_v1_fini(struct hl_device *hdev)
 	if (!ZERO_OR_NULL_PTR(hdev->mmu_priv.hr.mmu_shadow_hop0)) {
 		kvfree(hdev->mmu_priv.dr.mmu_shadow_hop0);
 		gen_pool_destroy(hdev->mmu_priv.dr.mmu_pgt_pool);
-	}
 
-	/* Make sure that if we arrive here again without init was called we
-	 * won't cause kernel panic. This can happen for example if we fail
-	 * during hard reset code at certain points
+		/* Make sure that if we arrive here again without init was
+		 * called we won't cause kernel panic. This can happen for
+		 * example if we fail during hard reset code at certain points
 		 */
 		hdev->mmu_priv.dr.mmu_shadow_hop0 = NULL;
+	}
 }
 
 /**

drivers/misc/habanalabs/common/pci/pci.c

@@ -436,6 +436,8 @@ int hl_pci_init(struct hl_device *hdev)
 		goto unmap_pci_bars;
 	}
 
+	dma_set_max_seg_size(&pdev->dev, U32_MAX);
+
 	return 0;
 
 unmap_pci_bars: