habanalabs: add device reset support

This patch adds support for doing various on-the-fly reset of Goya.

The driver supports two types of resets:
1. soft-reset
2. hard-reset

Soft-reset is done when the device detects a timeout of a command
submission that was given to the device. The soft-reset process only resets
the engines that are relevant for the submission of compute jobs, i.e. the
DMA channels, the TPCs and the MME. The purpose is to bring the device as
fast as possible to a working state.

Hard-reset is done in several cases:
1. After soft-reset is done but the device is not responding
2. When fatal errors occur inside the device, e.g. ECC error
3. When the driver is removed

Hard-reset performs a reset of the entire chip except for the PCI
controller and the PLLs. It is a much longer process then soft-reset but it
helps to recover the device without the need to reboot the Host.

After hard-reset, the driver will restore the max power attribute and in
case of manual power management, the frequencies that were set.

This patch also adds two entries to the sysfs, which allows the root user
to initiate a soft or hard reset.

Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

1 files changed, 323 insertions, 2 deletions

diff --git a/drivers/misc/habanalabs/device.c b/drivers/misc/habanalabs/device.c
index f879de310915..2aa8a68cdf76 100644
--- a/drivers/misc/habanalabs/device.c
+++ b/drivers/misc/habanalabs/device.c
@@ -8,9 +8,17 @@
 #include "habanalabs.h"
 
 #include <linux/pci.h>
-#include <linux/delay.h>
+#include <linux/sched/signal.h>
 #include <linux/hwmon.h>
 
+bool hl_device_disabled_or_in_reset(struct hl_device *hdev)
+{
+	if ((hdev->disabled) || (atomic_read(&hdev->in_reset)))
+		return true;
+	else
+		return false;
+}
+
 static void hpriv_release(struct kref *ref)
 {
 	struct hl_fpriv *hpriv;
@@ -200,6 +208,7 @@ static int device_early_init(struct hl_device *hdev)
 
 	mutex_init(&hdev->fd_open_cnt_lock);
 	mutex_init(&hdev->send_cpu_message_lock);
+	atomic_set(&hdev->in_reset, 0);
 	atomic_set(&hdev->fd_open_cnt, 0);
 
 	return 0;
@@ -254,6 +263,27 @@ static void set_freq_to_low_job(struct work_struct *work)
 			usecs_to_jiffies(HL_PLL_LOW_JOB_FREQ_USEC));
 }
 
+static void hl_device_heartbeat(struct work_struct *work)
+{
+	struct hl_device *hdev = container_of(work, struct hl_device,
+						work_heartbeat.work);
+
+	if (hl_device_disabled_or_in_reset(hdev))
+		goto reschedule;
+
+	if (!hdev->asic_funcs->send_heartbeat(hdev))
+		goto reschedule;
+
+	dev_err(hdev->dev, "Device heartbeat failed!\n");
+	hl_device_reset(hdev, true, false);
+
+	return;
+
+reschedule:
+	schedule_delayed_work(&hdev->work_heartbeat,
+			usecs_to_jiffies(HL_HEARTBEAT_PER_USEC));
+}
+
 /*
  * device_late_init - do late stuff initialization for the habanalabs device
  *
@@ -289,6 +319,12 @@ static int device_late_init(struct hl_device *hdev)
 	schedule_delayed_work(&hdev->work_freq,
 			usecs_to_jiffies(HL_PLL_LOW_JOB_FREQ_USEC));
 
+	if (hdev->heartbeat) {
+		INIT_DELAYED_WORK(&hdev->work_heartbeat, hl_device_heartbeat);
+		schedule_delayed_work(&hdev->work_heartbeat,
+				usecs_to_jiffies(HL_HEARTBEAT_PER_USEC));
+	}
+
 	hdev->late_init_done = true;
 
 	return 0;
@@ -306,6 +342,8 @@ static void device_late_fini(struct hl_device *hdev)
 		return;
 
 	cancel_delayed_work_sync(&hdev->work_freq);
+	if (hdev->heartbeat)
+		cancel_delayed_work_sync(&hdev->work_heartbeat);
 
 	if (hdev->asic_funcs->late_fini)
 		hdev->asic_funcs->late_fini(hdev);
@@ -413,6 +451,260 @@ int hl_device_resume(struct hl_device *hdev)
 	return 0;
 }
 
+static void hl_device_hard_reset_pending(struct work_struct *work)
+{
+	struct hl_device_reset_work *device_reset_work =
+		container_of(work, struct hl_device_reset_work, reset_work);
+	struct hl_device *hdev = device_reset_work->hdev;
+	u16 pending_cnt = HL_PENDING_RESET_PER_SEC;
+	struct task_struct *task = NULL;
+
+	/* Flush all processes that are inside hl_open */
+	mutex_lock(&hdev->fd_open_cnt_lock);
+
+	while ((atomic_read(&hdev->fd_open_cnt)) && (pending_cnt)) {
+
+		pending_cnt--;
+
+		dev_info(hdev->dev,
+			"Can't HARD reset, waiting for user to close FD\n");
+		ssleep(1);
+	}
+
+	if (atomic_read(&hdev->fd_open_cnt)) {
+		task = get_pid_task(hdev->user_ctx->hpriv->taskpid,
+					PIDTYPE_PID);
+		if (task) {
+			dev_info(hdev->dev, "Killing user processes\n");
+			send_sig(SIGKILL, task, 1);
+			msleep(100);
+
+			put_task_struct(task);
+		}
+	}
+
+	mutex_unlock(&hdev->fd_open_cnt_lock);
+
+	hl_device_reset(hdev, true, true);
+
+	kfree(device_reset_work);
+}
+
+/*
+ * hl_device_reset - reset the device
+ *
+ * @hdev: pointer to habanalabs device structure
+ * @hard_reset: should we do hard reset to all engines or just reset the
+ *              compute/dma engines
+ *
+ * Block future CS and wait for pending CS to be enqueued
+ * Call ASIC H/W fini
+ * Flush all completions
+ * Re-initialize all internal data structures
+ * Call ASIC H/W init, late_init
+ * Test queues
+ * Enable device
+ *
+ * Returns 0 for success or an error on failure.
+ */
+int hl_device_reset(struct hl_device *hdev, bool hard_reset,
+			bool from_hard_reset_thread)
+{
+	int i, rc;
+
+	if (!hdev->init_done) {
+		dev_err(hdev->dev,
+			"Can't reset before initialization is done\n");
+		return 0;
+	}
+
+	/*
+	 * Prevent concurrency in this function - only one reset should be
+	 * done at any given time. Only need to perform this if we didn't
+	 * get from the dedicated hard reset thread
+	 */
+	if (!from_hard_reset_thread) {
+		/* Block future CS/VM/JOB completion operations */
+		rc = atomic_cmpxchg(&hdev->in_reset, 0, 1);
+		if (rc)
+			return 0;
+
+		/* 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);
+
+		dev_err(hdev->dev, "Going to RESET device!\n");
+	}
+
+again:
+	if ((hard_reset) && (!from_hard_reset_thread)) {
+		struct hl_device_reset_work *device_reset_work;
+
+		if (!hdev->pdev) {
+			dev_err(hdev->dev,
+				"Reset action is NOT supported in simulator\n");
+			rc = -EINVAL;
+			goto out_err;
+		}
+
+		hdev->hard_reset_pending = true;
+
+		device_reset_work = kzalloc(sizeof(*device_reset_work),
+						GFP_ATOMIC);
+		if (!device_reset_work) {
+			rc = -ENOMEM;
+			goto out_err;
+		}
+
+		/*
+		 * Because the reset function can't run from interrupt or
+		 * from heartbeat work, we need to call the reset function
+		 * from a dedicated work
+		 */
+		INIT_WORK(&device_reset_work->reset_work,
+				hl_device_hard_reset_pending);
+		device_reset_work->hdev = hdev;
+		schedule_work(&device_reset_work->reset_work);
+
+		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);
+
+	if (hard_reset) {
+		/* Release kernel context */
+		if (hl_ctx_put(hdev->kernel_ctx) != 1) {
+			dev_err(hdev->dev,
+				"kernel ctx is alive during hard reset\n");
+			rc = -EBUSY;
+			goto out_err;
+		}
+
+		hdev->kernel_ctx = NULL;
+	}
+
+	/* Reset the H/W. It will be in idle state after this returns */
+	hdev->asic_funcs->hw_fini(hdev, hard_reset);
+
+	if (hard_reset)
+		hl_eq_reset(hdev, &hdev->event_queue);
+
+	/* Re-initialize PI,CI to 0 in all queues (hw queue, cq) */
+	hl_hw_queue_reset(hdev, hard_reset);
+	for (i = 0 ; i < hdev->asic_prop.completion_queues_count ; i++)
+		hl_cq_reset(hdev, &hdev->completion_queue[i]);
+
+	/* Finished tear-down, starting to re-initialize */
+
+	if (hard_reset) {
+		/* Allocate the kernel context */
+		hdev->kernel_ctx = kzalloc(sizeof(*hdev->kernel_ctx),
+						GFP_KERNEL);
+		if (!hdev->kernel_ctx) {
+			rc = -ENOMEM;
+			goto out_err;
+		}
+
+		hdev->user_ctx = NULL;
+
+		rc = hl_ctx_init(hdev, hdev->kernel_ctx, true);
+		if (rc) {
+			dev_err(hdev->dev,
+				"failed to init kernel ctx in hard reset\n");
+			kfree(hdev->kernel_ctx);
+			hdev->kernel_ctx = NULL;
+			goto out_err;
+		}
+	}
+
+	rc = hdev->asic_funcs->hw_init(hdev);
+	if (rc) {
+		dev_err(hdev->dev,
+			"failed to initialize the H/W after reset\n");
+		goto out_err;
+	}
+
+	hdev->disabled = false;
+
+	/* Check that the communication with the device is working */
+	rc = hdev->asic_funcs->test_queues(hdev);
+	if (rc) {
+		dev_err(hdev->dev,
+			"Failed to detect if device is alive after reset\n");
+		goto out_err;
+	}
+
+	if (hard_reset) {
+		rc = device_late_init(hdev);
+		if (rc) {
+			dev_err(hdev->dev,
+				"Failed late init after hard reset\n");
+			goto out_err;
+		}
+
+		hl_set_max_power(hdev, hdev->max_power);
+
+		hdev->hard_reset_pending = false;
+	} else {
+		rc = hdev->asic_funcs->soft_reset_late_init(hdev);
+		if (rc) {
+			dev_err(hdev->dev,
+				"Failed late init after soft reset\n");
+			goto out_err;
+		}
+	}
+
+	atomic_set(&hdev->in_reset, 0);
+
+	if (hard_reset)
+		hdev->hard_reset_cnt++;
+	else
+		hdev->soft_reset_cnt++;
+
+	return 0;
+
+out_err:
+	hdev->disabled = true;
+
+	if (hard_reset) {
+		dev_err(hdev->dev,
+			"Failed to reset! Device is NOT usable\n");
+		hdev->hard_reset_cnt++;
+	} else {
+		dev_err(hdev->dev,
+			"Failed to do soft-reset, trying hard reset\n");
+		hdev->soft_reset_cnt++;
+		hard_reset = true;
+		goto again;
+	}
+
+	atomic_set(&hdev->in_reset, 0);
+
+	return rc;
+}
+
 /*
  * hl_device_init - main initialization function for habanalabs device
  *
@@ -520,6 +812,12 @@ int hl_device_init(struct hl_device *hdev, struct class *hclass)
 		goto free_cb_pool;
 	}
 
+	if (hdev->asic_funcs->get_hw_state(hdev) == HL_DEVICE_HW_STATE_DIRTY) {
+		dev_info(hdev->dev,
+			"H/W state is dirty, must reset before initializing\n");
+		hdev->asic_funcs->hw_fini(hdev, true);
+	}
+
 	rc = hdev->asic_funcs->hw_init(hdev);
 	if (rc) {
 		dev_err(hdev->dev, "failed to initialize the H/W\n");
@@ -565,6 +863,8 @@ int hl_device_init(struct hl_device *hdev, struct class *hclass)
 	dev_notice(hdev->dev,
 		"Successfully added device to habanalabs driver\n");
 
+	hdev->init_done = true;
+
 	return 0;
 
 free_cb_pool:
@@ -612,9 +912,30 @@ out_disabled:
  */
 void hl_device_fini(struct hl_device *hdev)
 {
-	int i;
+	int i, rc;
+	ktime_t timeout;
+
 	dev_info(hdev->dev, "Removing device\n");
 
+	/*
+	 * This function is competing with the reset function, so try to
+	 * take the reset atomic and if we are already in middle of reset,
+	 * wait until reset function is finished. Reset function is designed
+	 * to always finish (could take up to a few seconds in worst case).
+	 */
+
+	timeout = ktime_add_us(ktime_get(),
+				HL_PENDING_RESET_PER_SEC * 1000 * 1000 * 4);
+	rc = atomic_cmpxchg(&hdev->in_reset, 0, 1);
+	while (rc) {
+		usleep_range(50, 200);
+		rc = atomic_cmpxchg(&hdev->in_reset, 0, 1);
+		if (ktime_compare(ktime_get(), timeout) > 0) {
+			WARN(1, "Failed to remove device because reset function did not finish\n");
+			return;
+		}
+	};
+
 	/* Mark device as disabled */
 	hdev->disabled = true;