[PATCH 1/7] habanalabs: update firmware header files

[PATCH 1/7] habanalabs: update firmware header files

[Oded Gabbay]

From: Ofir Bitton <obitton@habana.ai>

Update recent changes made in firmware header files, which contain
a minor COMMS protocol change and new error status definitions.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
 drivers/misc/habanalabs/common/firmware_if.c  |  4 +--
 .../habanalabs/include/common/hl_boot_if.h    | 35 ++++++++++++++++---
 .../habanalabs/include/gaudi/gaudi_reg_map.h  |  2 --
 3 files changed, 33 insertions(+), 8 deletions(-)

diff --git a/drivers/misc/habanalabs/common/firmware_if.c b/drivers/misc/habanalabs/common/firmware_if.c
index 2e4d04ec6b53..bac25a60650f 100644
--- a/drivers/misc/habanalabs/common/firmware_if.c
+++ b/drivers/misc/habanalabs/common/firmware_if.c
@@ -1953,8 +1953,8 @@ 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");
diff --git a/drivers/misc/habanalabs/include/common/hl_boot_if.h b/drivers/misc/habanalabs/include/common/hl_boot_if.h
index fa8a5ad2d438..d762bb2f1204 100644
--- a/drivers/misc/habanalabs/include/common/hl_boot_if.h
+++ b/drivers/misc/habanalabs/include/common/hl_boot_if.h
@@ -78,6 +78,26 @@
  * CPU_BOOT_ERR0_DEVICE_UNUSABLE_FAIL	Device is unusable and customer support
  *					should be contacted.
  *
+ * CPU_BOOT_ERR0_ARC0_HALT_ACK_NOT_RCVD	HALT ACK from ARC0 is not received
+ *					within specified retries after issuing
+ *					HALT request. ARC0 appears to be in bad
+ *					reset.
+ *
+ * CPU_BOOT_ERR0_ARC1_HALT_ACK_NOT_RCVD	HALT ACK from ARC1 is not received
+ *					within specified retries after issuing
+ *					HALT request. ARC1 appears to be in bad
+ *					reset.
+ *
+ * CPU_BOOT_ERR0_ARC0_RUN_ACK_NOT_RCVD	RUN ACK from ARC0 is not received
+ *					within specified timeout after issuing
+ *					RUN request. ARC0 appears to be in bad
+ *					reset.
+ *
+ * CPU_BOOT_ERR0_ARC1_RUN_ACK_NOT_RCVD	RUN ACK from ARC1 is not received
+ *					within specified timeout after issuing
+ *					RUN request. ARC1 appears to be in bad
+ *					reset.
+ *
  * CPU_BOOT_ERR0_ENABLED		Error registers enabled.
  *					This is a main indication that the
  *					running FW populates the error
@@ -98,6 +118,10 @@
 #define CPU_BOOT_ERR0_SEC_IMG_VER_FAIL		(1 << 11)
 #define CPU_BOOT_ERR0_PLL_FAIL			(1 << 12)
 #define CPU_BOOT_ERR0_DEVICE_UNUSABLE_FAIL	(1 << 13)
+#define CPU_BOOT_ERR0_ARC0_HALT_ACK_NOT_RCVD	(1 << 14)
+#define CPU_BOOT_ERR0_ARC1_HALT_ACK_NOT_RCVD	(1 << 15)
+#define CPU_BOOT_ERR0_ARC0_RUN_ACK_NOT_RCVD	(1 << 16)
+#define CPU_BOOT_ERR0_ARC1_RUN_ACK_NOT_RCVD	(1 << 17)
 #define CPU_BOOT_ERR0_ENABLED			(1 << 31)
 #define CPU_BOOT_ERR1_ENABLED			(1 << 31)
 
@@ -313,10 +337,7 @@ struct cpu_dyn_regs {
 	__le32 hw_state;
 	__le32 kmd_msg_to_cpu;
 	__le32 cpu_cmd_status_to_host;
-	union {
-		__le32 gic_host_irq_ctrl;
-		__le32 gic_host_pi_upd_irq;
-	};
+	__le32 gic_host_pi_upd_irq;
 	__le32 gic_tpc_qm_irq_ctrl;
 	__le32 gic_mme_qm_irq_ctrl;
 	__le32 gic_dma_qm_irq_ctrl;
@@ -462,6 +483,11 @@ struct lkd_fw_comms_msg {
  *				Do not wait for BMC response.
  *
  * COMMS_LOW_PLL_OPP		Initialize PLLs for low OPP.
+ *
+ * COMMS_PREP_DESC_ELBI		Same as COMMS_PREP_DESC only that the memory
+ *				space is allocated in a ELBI access only
+ *				address range.
+ *
  */
 enum comms_cmd {
 	COMMS_NOOP = 0,
@@ -474,6 +500,7 @@ enum comms_cmd {
 	COMMS_GOTO_WFE = 7,
 	COMMS_SKIP_BMC = 8,
 	COMMS_LOW_PLL_OPP = 9,
+	COMMS_PREP_DESC_ELBI = 10,
 	COMMS_INVLD_LAST
 };
 
diff --git a/drivers/misc/habanalabs/include/gaudi/gaudi_reg_map.h b/drivers/misc/habanalabs/include/gaudi/gaudi_reg_map.h
index d95d4162ae2c..b9bd5a7f71eb 100644
--- a/drivers/misc/habanalabs/include/gaudi/gaudi_reg_map.h
+++ b/drivers/misc/habanalabs/include/gaudi/gaudi_reg_map.h
@@ -12,8 +12,6 @@
  * PSOC scratch-pad registers
  */
 #define mmHW_STATE			mmPSOC_GLOBAL_CONF_SCRATCHPAD_0
-/* TODO: remove mmGIC_HOST_IRQ_CTRL_POLL_REG */
-#define mmGIC_HOST_IRQ_CTRL_POLL_REG	mmPSOC_GLOBAL_CONF_SCRATCHPAD_1
 #define mmGIC_HOST_PI_UPD_IRQ_POLL_REG	mmPSOC_GLOBAL_CONF_SCRATCHPAD_1
 #define mmGIC_TPC_QM_IRQ_CTRL_POLL_REG	mmPSOC_GLOBAL_CONF_SCRATCHPAD_2
 #define mmGIC_MME_QM_IRQ_CTRL_POLL_REG	mmPSOC_GLOBAL_CONF_SCRATCHPAD_3
-- 
2.25.1

[PATCH 2/7] habanalabs: rename enum vm_type_t to vm_type

[Oded Gabbay]

We don't use typedefs so the enum name shouldn't end with _t

Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
 drivers/misc/habanalabs/common/debugfs.c    |  4 ++--
 drivers/misc/habanalabs/common/habanalabs.h |  8 ++++----
 drivers/misc/habanalabs/common/memory.c     | 14 ++++++--------
 3 files changed, 12 insertions(+), 14 deletions(-)

diff --git a/drivers/misc/habanalabs/common/debugfs.c b/drivers/misc/habanalabs/common/debugfs.c
index 703d79fb6f3f..fd44c3b66d3b 100644
--- a/drivers/misc/habanalabs/common/debugfs.c
+++ b/drivers/misc/habanalabs/common/debugfs.c
@@ -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;
@@ -492,7 +492,7 @@ static int device_va_to_pa(struct hl_device *hdev, u64 virt_addr, u32 size,
 	struct hl_ctx *ctx = hdev->compute_ctx;
 	struct hl_vm_hash_node *hnode;
 	struct hl_userptr *userptr;
-	enum vm_type_t *vm_type;
+	enum vm_type *vm_type;
 	bool valid = false;
 	u64 end_address;
 	u32 range_size;
diff --git a/drivers/misc/habanalabs/common/habanalabs.h b/drivers/misc/habanalabs/common/habanalabs.h
index 0edc72e678c9..31fda9e41fe7 100644
--- a/drivers/misc/habanalabs/common/habanalabs.h
+++ b/drivers/misc/habanalabs/common/habanalabs.h
@@ -317,11 +317,11 @@ struct hw_queue_properties {
 };
 
 /**
- * enum vm_type_t - virtual memory mapping request information.
+ * enum vm_type - virtual memory mapping request information.
  * @VM_TYPE_USERPTR: mapping of user memory to device virtual address.
  * @VM_TYPE_PHYS_PACK: mapping of DRAM memory to device virtual address.
  */
-enum vm_type_t {
+enum vm_type {
 	VM_TYPE_USERPTR = 0x1,
 	VM_TYPE_PHYS_PACK = 0x2
 };
@@ -1421,7 +1421,7 @@ struct hl_ctx_mgr {
  * @dma_mapped: true if the SG was mapped to DMA addresses, false otherwise.
  */
 struct hl_userptr {
-	enum vm_type_t		vm_type; /* must be first */
+	enum vm_type		vm_type; /* must be first */
 	struct list_head	job_node;
 	struct page		**pages;
 	unsigned int		npages;
@@ -1635,7 +1635,7 @@ struct hl_vm_hw_block_list_node {
  * @created_from_userptr: is product of host virtual address.
  */
 struct hl_vm_phys_pg_pack {
-	enum vm_type_t		vm_type; /* must be first */
+	enum vm_type		vm_type; /* must be first */
 	u64			*pages;
 	u64			npages;
 	u64			total_size;
diff --git a/drivers/misc/habanalabs/common/memory.c b/drivers/misc/habanalabs/common/memory.c
index d54fdd04be5b..57e980d1a84b 100644
--- a/drivers/misc/habanalabs/common/memory.c
+++ b/drivers/misc/habanalabs/common/memory.c
@@ -607,8 +607,7 @@ static u64 get_va_block(struct hl_device *hdev,
 			do_div(tmp_hint_addr, va_range->page_size))) {
 
 		if (force_hint) {
-			/* Hint must be repected, so here we just fail.
-			 */
+			/* Hint must be respected, so here we just fail */
 			dev_err(hdev->dev,
 				"Hint address 0x%llx is not page aligned - cannot be respected\n",
 				hint_addr);
@@ -1067,7 +1066,7 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 	struct hl_userptr *userptr = NULL;
 	struct hl_vm_hash_node *hnode;
 	struct hl_va_range *va_range;
-	enum vm_type_t *vm_type;
+	enum vm_type *vm_type;
 	u64 ret_vaddr, hint_addr;
 	u32 handle = 0, va_block_align;
 	int rc;
@@ -1098,7 +1097,7 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 			goto init_page_pack_err;
 		}
 
-		vm_type = (enum vm_type_t *) userptr;
+		vm_type = (enum vm_type *) userptr;
 		hint_addr = args->map_host.hint_addr;
 		handle = phys_pg_pack->handle;
 
@@ -1140,7 +1139,7 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 
 		spin_unlock(&vm->idr_lock);
 
-		vm_type = (enum vm_type_t *) phys_pg_pack;
+		vm_type = (enum vm_type *) phys_pg_pack;
 
 		hint_addr = args->map_device.hint_addr;
 
@@ -1171,8 +1170,7 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 
 	if (hint_addr && phys_pg_pack->offset) {
 		if (args->flags & HL_MEM_FORCE_HINT) {
-			/* If hint must be repected, since we can't - just fail.
-			 */
+			/* Fail if hint must be respected but it can't be */
 			dev_err(hdev->dev,
 				"Hint address 0x%llx cannot be respected because source memory is not aligned 0x%x\n",
 				hint_addr, phys_pg_pack->offset);
@@ -1273,7 +1271,7 @@ static int unmap_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 	struct hl_userptr *userptr = NULL;
 	struct hl_va_range *va_range;
 	u64 vaddr = args->unmap.device_virt_addr;
-	enum vm_type_t *vm_type;
+	enum vm_type *vm_type;
 	bool is_userptr;
 	int rc = 0;
 
-- 
2.25.1

[PATCH 3/7] habanalabs: re-init completion object upon retry

[Oded Gabbay]

In case user interrupt arrived but the completion value is less than
the target value, we want to retry the wait.

However, before the retry we must reinitialize the completion object,
under spin-lock, so the wait function won't exit immediately because
the completion object is already completed (from the previous
interrupt).

Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
 drivers/misc/habanalabs/common/command_submission.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/drivers/misc/habanalabs/common/command_submission.c b/drivers/misc/habanalabs/common/command_submission.c
index 80c60fb41bbc..12f20446e99a 100644
--- a/drivers/misc/habanalabs/common/command_submission.c
+++ b/drivers/misc/habanalabs/common/command_submission.c
@@ -2046,7 +2046,8 @@ static int _hl_interrupt_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx,
 		goto unlock_and_free_fence;
 	}
 
-	if (copy_from_user(&completion_value, u64_to_user_ptr(user_address), 4)) {
+	if (copy_from_user(&completion_value, u64_to_user_ptr(user_address),
+									4)) {
 		dev_err(hdev->dev,
 			"Failed to copy completion value from user\n");
 		rc = -EFAULT;
@@ -2077,18 +2078,28 @@ static int _hl_interrupt_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx,
 	 * If comparison fails, keep waiting until timeout expires
 	 */
 	if (completion_rc > 0) {
+		spin_lock(&interrupt->wait_list_lock);
+
 		if (copy_from_user(&completion_value,
 				u64_to_user_ptr(user_address), 4)) {
+
+			spin_unlock(&interrupt->wait_list_lock);
+
 			dev_err(hdev->dev,
 				"Failed to copy completion value from user\n");
 			rc = -EFAULT;
+
 			goto remove_pending_user_interrupt;
 		}
 
 		if (completion_value >= target_value) {
+			spin_unlock(&interrupt->wait_list_lock);
 			*status = CS_WAIT_STATUS_COMPLETED;
 		} else {
+			reinit_completion(&pend->fence.completion);
 			timeout = completion_rc;
+
+			spin_unlock(&interrupt->wait_list_lock);
 			goto wait_again;
 		}
 	} else {
-- 
2.25.1

[PATCH 4/7] habanalabs: release pending user interrupts on device fini

[Oded Gabbay]

In device fini there was missing a call to release all pending user
interrupts. That can cause a process to be stuck inside the driver's
IOCTL of wait for interrupts, in case the device is removed or
simulator is killed at the same time.

In addition, also call to remove inactive codec job was missing.

Moreover, to prevent such errors in the future (where code is added
to reset path but not to device fini), we moved some common parts
to two dedicated functions:
cleanup_resources
take_release_locks

Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
 drivers/misc/habanalabs/common/device.c | 100 +++++++++++-------------
 1 file changed, 46 insertions(+), 54 deletions(-)

diff --git a/drivers/misc/habanalabs/common/device.c b/drivers/misc/habanalabs/common/device.c
index ff4cbde289c0..43d0dcbd20e3 100644
--- a/drivers/misc/habanalabs/common/device.c
+++ b/drivers/misc/habanalabs/common/device.c
@@ -866,6 +866,48 @@ static void device_disable_open_processes(struct hl_device *hdev)
 	mutex_unlock(&hdev->fpriv_list_lock);
 }
 
+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 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)
+{
+	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);
+}
+
 /*
  * hl_device_reset - reset the device
  *
@@ -970,15 +1012,7 @@ 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");
 	}
@@ -999,31 +1033,7 @@ 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);
 
 kill_processes:
 	if (hard_reset) {
@@ -1567,31 +1577,13 @@ 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);
 
 	/* 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
-- 
2.25.1

[PATCH 5/7] habanalabs: handle case of interruptable wait

[Oded Gabbay]

Same as we handle it in the regular wait for CS, we need to handle the
case where the waiting for user interrupt was interrupted. In that case,
we need to return correct error code to the user.

Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
 drivers/misc/habanalabs/common/command_submission.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/misc/habanalabs/common/command_submission.c b/drivers/misc/habanalabs/common/command_submission.c
index 12f20446e99a..997a37e78241 100644
--- a/drivers/misc/habanalabs/common/command_submission.c
+++ b/drivers/misc/habanalabs/common/command_submission.c
@@ -2102,6 +2102,12 @@ static int _hl_interrupt_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx,
 			spin_unlock(&interrupt->wait_list_lock);
 			goto wait_again;
 		}
+	} else if (completion_rc == -ERESTARTSYS) {
+		dev_err_ratelimited(hdev->dev,
+			"user process got signal while waiting for interrupt ID %d\n",
+			interrupt->interrupt_id);
+		*status = HL_WAIT_CS_STATUS_INTERRUPTED;
+		rc = -EINTR;
 	} else {
 		*status = CS_WAIT_STATUS_BUSY;
 	}
@@ -2159,8 +2165,9 @@ static int hl_interrupt_wait_ioctl(struct hl_fpriv *hpriv, void *data)
 	memset(args, 0, sizeof(*args));
 
 	if (rc) {
-		dev_err_ratelimited(hdev->dev,
-			"interrupt_wait_ioctl failed (%d)\n", rc);
+		if (rc != -EINTR)
+			dev_err_ratelimited(hdev->dev,
+				"interrupt_wait_ioctl failed (%d)\n", rc);
 
 		return rc;
 	}
-- 
2.25.1

[PATCH 6/7] habanalabs: user mappings can be 64-bit

[Oded Gabbay]

Increase the size variable in the userptr structure to 64-bit. That
variable describes the size of the memory allocation of the user that
is now being mapped into the device. The mapping can be larger than
4GB, so we need to support it.

Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
 drivers/misc/habanalabs/common/debugfs.c    | 4 ++--
 drivers/misc/habanalabs/common/habanalabs.h | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/misc/habanalabs/common/debugfs.c b/drivers/misc/habanalabs/common/debugfs.c
index fd44c3b66d3b..77f7c2aa571d 100644
--- a/drivers/misc/habanalabs/common/debugfs.c
+++ b/drivers/misc/habanalabs/common/debugfs.c
@@ -213,7 +213,7 @@ static int userptr_show(struct seq_file *s, void *data)
 			seq_puts(s, "----------------------------------------------------------\n");
 		}
 		seq_printf(s,
-			"    0x%-14llx      %-10u    %-30s\n",
+			"    0x%-14llx      %-10llu    %-30s\n",
 			userptr->addr, userptr->size, dma_dir[userptr->dir]);
 	}
 
@@ -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;
diff --git a/drivers/misc/habanalabs/common/habanalabs.h b/drivers/misc/habanalabs/common/habanalabs.h
index 31fda9e41fe7..5ae95d2abaa6 100644
--- a/drivers/misc/habanalabs/common/habanalabs.h
+++ b/drivers/misc/habanalabs/common/habanalabs.h
@@ -1429,7 +1429,7 @@ struct hl_userptr {
 	enum dma_data_direction dir;
 	struct list_head	debugfs_list;
 	u64			addr;
-	u32			size;
+	u64			size;
 	u8			dma_mapped;
 };
 
-- 
2.25.1

[PATCH 7/7] habanalabs: allow disabling huge page use

[Oded Gabbay]

Sometimes we may need to disable optimization of using huge pages
in our memory management code. Add such a flag to the function that
creates the list of physical pages that would be programmed into the
device MMU.

Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
 drivers/misc/habanalabs/common/memory.c | 36 ++++++++++++++++---------
 1 file changed, 23 insertions(+), 13 deletions(-)

diff --git a/drivers/misc/habanalabs/common/memory.c b/drivers/misc/habanalabs/common/memory.c
index 57e980d1a84b..a05d98db4857 100644
--- a/drivers/misc/habanalabs/common/memory.c
+++ b/drivers/misc/habanalabs/common/memory.c
@@ -819,6 +819,10 @@ static u32 get_sg_info(struct scatterlist *sg, dma_addr_t *dma_addr)
  * @ctx: pointer to the context structure.
  * @userptr: userptr to initialize from.
  * @pphys_pg_pack: result pointer.
+ * @force_regular_page: tell the function to ignore huge page optimization,
+ *                      even if possible. Needed for cases where the device VA
+ *                      is allocated before we know the composition of the
+ *                      physical pages
  *
  * This function does the following:
  * - Pin the physical pages related to the given virtual block.
@@ -827,17 +831,18 @@ static u32 get_sg_info(struct scatterlist *sg, dma_addr_t *dma_addr)
  */
 static int init_phys_pg_pack_from_userptr(struct hl_ctx *ctx,
 				struct hl_userptr *userptr,
-				struct hl_vm_phys_pg_pack **pphys_pg_pack)
+				struct hl_vm_phys_pg_pack **pphys_pg_pack,
+				bool force_regular_page)
 {
+	u32 npages, page_size = PAGE_SIZE,
+		huge_page_size = ctx->hdev->asic_prop.pmmu_huge.page_size;
+	u32 pgs_in_huge_page = huge_page_size >> __ffs(page_size);
 	struct hl_vm_phys_pg_pack *phys_pg_pack;
+	bool first = true, is_huge_page_opt;
+	u64 page_mask, total_npages;
 	struct scatterlist *sg;
 	dma_addr_t dma_addr;
-	u64 page_mask, total_npages;
-	u32 npages, page_size = PAGE_SIZE,
-		huge_page_size = ctx->hdev->asic_prop.pmmu_huge.page_size;
-	bool first = true, is_huge_page_opt = true;
 	int rc, i, j;
-	u32 pgs_in_huge_page = huge_page_size >> __ffs(page_size);
 
 	phys_pg_pack = kzalloc(sizeof(*phys_pg_pack), GFP_KERNEL);
 	if (!phys_pg_pack)
@@ -848,6 +853,8 @@ static int init_phys_pg_pack_from_userptr(struct hl_ctx *ctx,
 	phys_pg_pack->asid = ctx->asid;
 	atomic_set(&phys_pg_pack->mapping_cnt, 1);
 
+	is_huge_page_opt = (force_regular_page ? false : true);
+
 	/* Only if all dma_addrs are aligned to 2MB and their
 	 * sizes is at least 2MB, we can use huge page mapping.
 	 * We limit the 2MB optimization to this condition,
@@ -1089,7 +1096,7 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 		}
 
 		rc = init_phys_pg_pack_from_userptr(ctx, userptr,
-				&phys_pg_pack);
+				&phys_pg_pack, false);
 		if (rc) {
 			dev_err(hdev->dev,
 				"unable to init page pack for vaddr 0x%llx\n",
@@ -1264,17 +1271,19 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 static int unmap_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 				bool ctx_free)
 {
-	struct hl_device *hdev = ctx->hdev;
-	struct asic_fixed_properties *prop = &hdev->asic_prop;
 	struct hl_vm_phys_pg_pack *phys_pg_pack = NULL;
+	u64 vaddr = args->unmap.device_virt_addr;
 	struct hl_vm_hash_node *hnode = NULL;
+	struct asic_fixed_properties *prop;
+	struct hl_device *hdev = ctx->hdev;
 	struct hl_userptr *userptr = NULL;
 	struct hl_va_range *va_range;
-	u64 vaddr = args->unmap.device_virt_addr;
 	enum vm_type *vm_type;
 	bool is_userptr;
 	int rc = 0;
 
+	prop = &hdev->asic_prop;
+
 	/* protect from double entrance */
 	mutex_lock(&ctx->mem_hash_lock);
 	hash_for_each_possible(ctx->mem_hash, hnode, node, (unsigned long)vaddr)
@@ -1297,8 +1306,9 @@ static int unmap_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 	if (*vm_type == VM_TYPE_USERPTR) {
 		is_userptr = true;
 		userptr = hnode->ptr;
-		rc = init_phys_pg_pack_from_userptr(ctx, userptr,
-							&phys_pg_pack);
+
+		rc = init_phys_pg_pack_from_userptr(ctx, userptr, &phys_pg_pack,
+							false);
 		if (rc) {
 			dev_err(hdev->dev,
 				"unable to init page pack for vaddr 0x%llx\n",
@@ -1382,7 +1392,7 @@ static int unmap_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 	kfree(hnode);
 
 	if (is_userptr) {
-		rc = free_phys_pg_pack(hdev, phys_pg_pack);
+		free_phys_pg_pack(hdev, phys_pg_pack);
 		dma_unmap_host_va(hdev, userptr);
 	}
 
-- 
2.25.1