drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c

/*
 * Copyright 2009 Jerome Glisse.
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 */
/*
 * Authors:
 *    Jerome Glisse <glisse@freedesktop.org>
 *    Dave Airlie
 */
#include <linux/seq_file.h>
#include <linux/atomic.h>
#include <linux/wait.h>
#include <linux/kref.h>
#include <linux/slab.h>
#include <linux/firmware.h>
#include <linux/pm_runtime.h>

#include <drm/drm_drv.h>
#include "amdgpu.h"
#include "amdgpu_trace.h"
#include "amdgpu_reset.h"

/*
 * Cast helper
 */
static const struct dma_fence_ops amdgpu_fence_ops;
static inline struct amdgpu_fence *to_amdgpu_fence(struct dma_fence *f)
{
	struct amdgpu_fence *__f = container_of(f, struct amdgpu_fence, base);

	return __f;
}

/**
 * amdgpu_fence_write - write a fence value
 *
 * @ring: ring the fence is associated with
 * @seq: sequence number to write
 *
 * Writes a fence value to memory (all asics).
 */
static void amdgpu_fence_write(struct amdgpu_ring *ring, u32 seq)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;

	if (drv->cpu_addr)
		*drv->cpu_addr = cpu_to_le32(seq);
}

/**
 * amdgpu_fence_read - read a fence value
 *
 * @ring: ring the fence is associated with
 *
 * Reads a fence value from memory (all asics).
 * Returns the value of the fence read from memory.
 */
static u32 amdgpu_fence_read(struct amdgpu_ring *ring)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	u32 seq = 0;

	if (drv->cpu_addr)
		seq = le32_to_cpu(*drv->cpu_addr);
	else
		seq = atomic_read(&drv->last_seq);

	return seq;
}

/**
 * amdgpu_fence_emit - emit a fence on the requested ring
 *
 * @ring: ring the fence is associated with
 * @af: amdgpu fence input
 * @flags: flags to pass into the subordinate .emit_fence() call
 *
 * Emits a fence command on the requested ring (all asics).
 */
void amdgpu_fence_emit(struct amdgpu_ring *ring, struct amdgpu_fence *af,
		       unsigned int flags)
{
	struct amdgpu_device *adev = ring->adev;
	struct dma_fence *fence;
	struct dma_fence __rcu **ptr;
	uint32_t seq;

	fence = &af->base;
	af->ring = ring;

	seq = ++ring->fence_drv.sync_seq;
	dma_fence_init(fence, &amdgpu_fence_ops,
		       &ring->fence_drv.lock,
		       adev->fence_context + ring->idx, seq);

	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
			       seq, flags | AMDGPU_FENCE_FLAG_INT);

	pm_runtime_get_noresume(adev_to_drm(adev)->dev);
	ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask];
	if (unlikely(rcu_dereference_protected(*ptr, 1))) {
		struct dma_fence *old;

		rcu_read_lock();
		old = dma_fence_get_rcu_safe(ptr);
		rcu_read_unlock();

		if (old) {
			/*
			 * dma_fence_wait(old, false) is not interruptible.
			 * It will not return an error in this case.
			 * So we can safely ignore the return value.
			 */
			dma_fence_wait(old, false);
			dma_fence_put(old);
		}
	}

	to_amdgpu_fence(fence)->start_timestamp = ktime_get();

	/* This function can't be called concurrently anyway, otherwise
	 * emitting the fence would mess up the hardware ring buffer.
	 */
	rcu_assign_pointer(*ptr, dma_fence_get(fence));
}

/**
 * amdgpu_fence_emit_polling - emit a fence on the requeste ring
 *
 * @ring: ring the fence is associated with
 * @s: resulting sequence number
 * @timeout: the timeout for waiting in usecs
 *
 * Emits a fence command on the requested ring (all asics).
 * Used For polling fence.
 * Returns 0 on success, -ENOMEM on failure.
 */
int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s,
			      uint32_t timeout)
{
	uint32_t seq;
	signed long r;

	if (!s)
		return -EINVAL;

	seq = ++ring->fence_drv.sync_seq;
	r = amdgpu_fence_wait_polling(ring,
				      seq - ring->fence_drv.num_fences_mask,
				      timeout);
	if (r < 1)
		return -ETIMEDOUT;

	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
			       seq, 0);

	*s = seq;

	return 0;
}

/**
 * amdgpu_fence_schedule_fallback - schedule fallback check
 *
 * @ring: pointer to struct amdgpu_ring
 *
 * Start a timer as fallback to our interrupts.
 */
static void amdgpu_fence_schedule_fallback(struct amdgpu_ring *ring)
{
	mod_timer(&ring->fence_drv.fallback_timer,
		  jiffies + AMDGPU_FENCE_JIFFIES_TIMEOUT);
}

/**
 * amdgpu_fence_process - check for fence activity
 *
 * @ring: pointer to struct amdgpu_ring
 *
 * Checks the current fence value and calculates the last
 * signalled fence value. Wakes the fence queue if the
 * sequence number has increased.
 *
 * Returns true if fence was processed
 */
bool amdgpu_fence_process(struct amdgpu_ring *ring)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	struct amdgpu_device *adev = ring->adev;
	uint32_t seq, last_seq;

	do {
		last_seq = atomic_read(&ring->fence_drv.last_seq);
		seq = amdgpu_fence_read(ring);

	} while (atomic_cmpxchg(&drv->last_seq, last_seq, seq) != last_seq);

	if (timer_delete(&ring->fence_drv.fallback_timer) &&
	    seq != ring->fence_drv.sync_seq)
		amdgpu_fence_schedule_fallback(ring);

	if (unlikely(seq == last_seq))
		return false;

	last_seq &= drv->num_fences_mask;
	seq &= drv->num_fences_mask;

	do {
		struct dma_fence *fence, **ptr;

		++last_seq;
		last_seq &= drv->num_fences_mask;
		ptr = &drv->fences[last_seq];

		/* There is always exactly one thread signaling this fence slot */
		fence = rcu_dereference_protected(*ptr, 1);
		RCU_INIT_POINTER(*ptr, NULL);

		if (!fence)
			continue;

		dma_fence_signal(fence);
		dma_fence_put(fence);
		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
	} while (last_seq != seq);

	return true;
}

/**
 * amdgpu_fence_fallback - fallback for hardware interrupts
 *
 * @t: timer context used to obtain the pointer to ring structure
 *
 * Checks for fence activity.
 */
static void amdgpu_fence_fallback(struct timer_list *t)
{
	struct amdgpu_ring *ring = timer_container_of(ring, t,
						      fence_drv.fallback_timer);

	if (amdgpu_fence_process(ring))
		dev_warn(ring->adev->dev,
			 "Fence fallback timer expired on ring %s\n",
			 ring->name);
}

/**
 * amdgpu_fence_wait_empty - wait for all fences to signal
 *
 * @ring: ring index the fence is associated with
 *
 * Wait for all fences on the requested ring to signal (all asics).
 * Returns 0 if the fences have passed, error for all other cases.
 */
int amdgpu_fence_wait_empty(struct amdgpu_ring *ring)
{
	uint64_t seq = READ_ONCE(ring->fence_drv.sync_seq);
	struct dma_fence *fence, **ptr;
	int r;

	if (!seq)
		return 0;

	ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask];
	rcu_read_lock();
	fence = rcu_dereference(*ptr);
	if (!fence || !dma_fence_get_rcu(fence)) {
		rcu_read_unlock();
		return 0;
	}
	rcu_read_unlock();

	r = dma_fence_wait(fence, false);
	dma_fence_put(fence);
	return r;
}

/**
 * amdgpu_fence_wait_polling - busy wait for givn sequence number
 *
 * @ring: ring index the fence is associated with
 * @wait_seq: sequence number to wait
 * @timeout: the timeout for waiting in usecs
 *
 * Wait for all fences on the requested ring to signal (all asics).
 * Returns left time if no timeout, 0 or minus if timeout.
 */
signed long amdgpu_fence_wait_polling(struct amdgpu_ring *ring,
				      uint32_t wait_seq,
				      signed long timeout)
{

	while ((int32_t)(wait_seq - amdgpu_fence_read(ring)) > 0 && timeout > 0) {
		udelay(2);
		timeout -= 2;
	}
	return timeout > 0 ? timeout : 0;
}
/**
 * amdgpu_fence_count_emitted - get the count of emitted fences
 *
 * @ring: ring the fence is associated with
 *
 * Get the number of fences emitted on the requested ring (all asics).
 * Returns the number of emitted fences on the ring.  Used by the
 * dynpm code to ring track activity.
 */
unsigned int amdgpu_fence_count_emitted(struct amdgpu_ring *ring)
{
	uint64_t emitted;

	/* We are not protected by ring lock when reading the last sequence
	 * but it's ok to report slightly wrong fence count here.
	 */
	emitted = 0x100000000ull;
	emitted -= atomic_read(&ring->fence_drv.last_seq);
	emitted += READ_ONCE(ring->fence_drv.sync_seq);
	return lower_32_bits(emitted);
}

/**
 * amdgpu_fence_last_unsignaled_time_us - the time fence emitted until now
 * @ring: ring the fence is associated with
 *
 * Find the earliest fence unsignaled until now, calculate the time delta
 * between the time fence emitted and now.
 */
u64 amdgpu_fence_last_unsignaled_time_us(struct amdgpu_ring *ring)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	struct dma_fence *fence;
	uint32_t last_seq, sync_seq;

	last_seq = atomic_read(&ring->fence_drv.last_seq);
	sync_seq = READ_ONCE(ring->fence_drv.sync_seq);
	if (last_seq == sync_seq)
		return 0;

	++last_seq;
	last_seq &= drv->num_fences_mask;
	fence = drv->fences[last_seq];
	if (!fence)
		return 0;

	return ktime_us_delta(ktime_get(),
		to_amdgpu_fence(fence)->start_timestamp);
}

/**
 * amdgpu_fence_update_start_timestamp - update the timestamp of the fence
 * @ring: ring the fence is associated with
 * @seq: the fence seq number to update.
 * @timestamp: the start timestamp to update.
 *
 * The function called at the time the fence and related ib is about to
 * resubmit to gpu in MCBP scenario. Thus we do not consider race condition
 * with amdgpu_fence_process to modify the same fence.
 */
void amdgpu_fence_update_start_timestamp(struct amdgpu_ring *ring, uint32_t seq, ktime_t timestamp)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	struct dma_fence *fence;

	seq &= drv->num_fences_mask;
	fence = drv->fences[seq];
	if (!fence)
		return;

	to_amdgpu_fence(fence)->start_timestamp = timestamp;
}

/**
 * amdgpu_fence_driver_start_ring - make the fence driver
 * ready for use on the requested ring.
 *
 * @ring: ring to start the fence driver on
 * @irq_src: interrupt source to use for this ring
 * @irq_type: interrupt type to use for this ring
 *
 * Make the fence driver ready for processing (all asics).
 * Not all asics have all rings, so each asic will only
 * start the fence driver on the rings it has.
 * Returns 0 for success, errors for failure.
 */
int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
				   struct amdgpu_irq_src *irq_src,
				   unsigned int irq_type)
{
	struct amdgpu_device *adev = ring->adev;
	uint64_t index;

	if (ring->funcs->type != AMDGPU_RING_TYPE_UVD) {
		ring->fence_drv.cpu_addr = ring->fence_cpu_addr;
		ring->fence_drv.gpu_addr = ring->fence_gpu_addr;
	} else {
		/* put fence directly behind firmware */
		index = ALIGN(adev->uvd.fw->size, 8);
		ring->fence_drv.cpu_addr = adev->uvd.inst[ring->me].cpu_addr + index;
		ring->fence_drv.gpu_addr = adev->uvd.inst[ring->me].gpu_addr + index;
	}
	amdgpu_fence_write(ring, atomic_read(&ring->fence_drv.last_seq));

	ring->fence_drv.irq_src = irq_src;
	ring->fence_drv.irq_type = irq_type;
	ring->fence_drv.initialized = true;

	DRM_DEV_DEBUG(adev->dev, "fence driver on ring %s use gpu addr 0x%016llx\n",
		      ring->name, ring->fence_drv.gpu_addr);
	return 0;
}

/**
 * amdgpu_fence_driver_init_ring - init the fence driver
 * for the requested ring.
 *
 * @ring: ring to init the fence driver on
 *
 * Init the fence driver for the requested ring (all asics).
 * Helper function for amdgpu_fence_driver_sw_init().
 */
int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring)
{
	struct amdgpu_device *adev = ring->adev;

	if (!adev)
		return -EINVAL;

	if (!is_power_of_2(ring->num_hw_submission))
		return -EINVAL;

	ring->fence_drv.cpu_addr = NULL;
	ring->fence_drv.gpu_addr = 0;
	ring->fence_drv.sync_seq = 0;
	atomic_set(&ring->fence_drv.last_seq, 0);
	ring->fence_drv.initialized = false;

	timer_setup(&ring->fence_drv.fallback_timer, amdgpu_fence_fallback, 0);

	ring->fence_drv.num_fences_mask = ring->num_hw_submission * 2 - 1;
	spin_lock_init(&ring->fence_drv.lock);
	ring->fence_drv.fences = kcalloc(ring->num_hw_submission * 2, sizeof(void *),
					 GFP_KERNEL);

	if (!ring->fence_drv.fences)
		return -ENOMEM;

	return 0;
}

/**
 * amdgpu_fence_driver_sw_init - init the fence driver
 * for all possible rings.
 *
 * @adev: amdgpu device pointer
 *
 * Init the fence driver for all possible rings (all asics).
 * Not all asics have all rings, so each asic will only
 * start the fence driver on the rings it has using
 * amdgpu_fence_driver_start_ring().
 * Returns 0 for success.
 */
int amdgpu_fence_driver_sw_init(struct amdgpu_device *adev)
{
	return 0;
}

/**
 * amdgpu_fence_need_ring_interrupt_restore - helper function to check whether
 * fence driver interrupts need to be restored.
 *
 * @ring: ring that to be checked
 *
 * Interrupts for rings that belong to GFX IP don't need to be restored
 * when the target power state is s0ix.
 *
 * Return true if need to restore interrupts, false otherwise.
 */
static bool amdgpu_fence_need_ring_interrupt_restore(struct amdgpu_ring *ring)
{
	struct amdgpu_device *adev = ring->adev;
	bool is_gfx_power_domain = false;

	switch (ring->funcs->type) {
	case AMDGPU_RING_TYPE_SDMA:
	/* SDMA 5.x+ is part of GFX power domain so it's covered by GFXOFF */
		if (amdgpu_ip_version(adev, SDMA0_HWIP, 0) >=
		    IP_VERSION(5, 0, 0))
			is_gfx_power_domain = true;
		break;
	case AMDGPU_RING_TYPE_GFX:
	case AMDGPU_RING_TYPE_COMPUTE:
	case AMDGPU_RING_TYPE_KIQ:
	case AMDGPU_RING_TYPE_MES:
		is_gfx_power_domain = true;
		break;
	default:
		break;
	}

	return !(adev->in_s0ix && is_gfx_power_domain);
}

/**
 * amdgpu_fence_driver_hw_fini - tear down the fence driver
 * for all possible rings.
 *
 * @adev: amdgpu device pointer
 *
 * Tear down the fence driver for all possible rings (all asics).
 */
void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev)
{
	int i, r;

	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];

		if (!ring || !ring->fence_drv.initialized)
			continue;

		/* You can't wait for HW to signal if it's gone */
		if (!drm_dev_is_unplugged(adev_to_drm(adev)))
			r = amdgpu_fence_wait_empty(ring);
		else
			r = -ENODEV;
		/* no need to trigger GPU reset as we are unloading */
		if (r)
			amdgpu_fence_driver_force_completion(ring, NULL);

		if (!drm_dev_is_unplugged(adev_to_drm(adev)) &&
		    ring->fence_drv.irq_src &&
		    amdgpu_fence_need_ring_interrupt_restore(ring))
			amdgpu_irq_put(adev, ring->fence_drv.irq_src,
				       ring->fence_drv.irq_type);

		timer_delete_sync(&ring->fence_drv.fallback_timer);
	}
}

/* Will either stop and flush handlers for amdgpu interrupt or reanble it */
void amdgpu_fence_driver_isr_toggle(struct amdgpu_device *adev, bool stop)
{
	int i;

	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];

		if (!ring || !ring->fence_drv.initialized || !ring->fence_drv.irq_src)
			continue;

		if (stop)
			disable_irq(adev->irq.irq);
		else
			enable_irq(adev->irq.irq);
	}
}

void amdgpu_fence_driver_sw_fini(struct amdgpu_device *adev)
{
	unsigned int i, j;

	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];

		if (!ring || !ring->fence_drv.initialized)
			continue;

		/*
		 * Notice we check for sched.ops since there's some
		 * override on the meaning of sched.ready by amdgpu.
		 * The natural check would be sched.ready, which is
		 * set as drm_sched_init() finishes...
		 */
		if (ring->sched.ops)
			drm_sched_fini(&ring->sched);

		for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j)
			dma_fence_put(ring->fence_drv.fences[j]);
		kfree(ring->fence_drv.fences);
		ring->fence_drv.fences = NULL;
		ring->fence_drv.initialized = false;
	}
}

/**
 * amdgpu_fence_driver_hw_init - enable the fence driver
 * for all possible rings.
 *
 * @adev: amdgpu device pointer
 *
 * Enable the fence driver for all possible rings (all asics).
 * Not all asics have all rings, so each asic will only
 * start the fence driver on the rings it has using
 * amdgpu_fence_driver_start_ring().
 * Returns 0 for success.
 */
void amdgpu_fence_driver_hw_init(struct amdgpu_device *adev)
{
	int i;

	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];

		if (!ring || !ring->fence_drv.initialized)
			continue;

		/* enable the interrupt */
		if (ring->fence_drv.irq_src &&
		    amdgpu_fence_need_ring_interrupt_restore(ring))
			amdgpu_irq_get(adev, ring->fence_drv.irq_src,
				       ring->fence_drv.irq_type);
	}
}

/**
 * amdgpu_fence_driver_set_error - set error code on fences
 * @ring: the ring which contains the fences
 * @error: the error code to set
 *
 * Set an error code to all the fences pending on the ring.
 */
void amdgpu_fence_driver_set_error(struct amdgpu_ring *ring, int error)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	unsigned long flags;

	spin_lock_irqsave(&drv->lock, flags);
	for (unsigned int i = 0; i <= drv->num_fences_mask; ++i) {
		struct dma_fence *fence;

		fence = rcu_dereference_protected(drv->fences[i],
						  lockdep_is_held(&drv->lock));
		if (fence && !dma_fence_is_signaled_locked(fence))
			dma_fence_set_error(fence, error);
	}
	spin_unlock_irqrestore(&drv->lock, flags);
}

/**
 * amdgpu_fence_driver_force_completion - force signal latest fence of ring
 *
 * @ring: fence of the ring to signal
 * @timedout_fence: fence of the timedout job
 *
 */
void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring,
					  struct dma_fence *timedout_fence)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	unsigned long flags;

	spin_lock_irqsave(&drv->lock, flags);
	for (unsigned int i = 0; i <= drv->num_fences_mask; ++i) {
		struct dma_fence *fence;

		fence = rcu_dereference_protected(drv->fences[i],
						  lockdep_is_held(&drv->lock));
		if (fence && !dma_fence_is_signaled_locked(fence)) {
			if (fence == timedout_fence)
				dma_fence_set_error(fence, -ETIME);
			else
				dma_fence_set_error(fence, -ECANCELED);
		}
	}
	spin_unlock_irqrestore(&drv->lock, flags);

	amdgpu_fence_write(ring, ring->fence_drv.sync_seq);
	amdgpu_fence_process(ring);
}

/*
 * Kernel queue reset handling
 *
 * The driver can reset individual queues for most engines, but those queues
 * may contain work from multiple contexts.  Resetting the queue will reset
 * lose all of that state.  In order to minimize the collateral damage, the
 * driver will save the ring contents which are not associated with the guilty
 * context prior to resetting the queue.  After resetting the queue the queue
 * contents from the other contexts is re-emitted to the rings so that it can
 * be processed by the engine.  To handle this, we save the queue's write
 * pointer (wptr) in the fences associated with each context.  If we get a
 * queue timeout, we can then use the wptrs from the fences to determine
 * which data needs to be saved out of the queue's ring buffer.
 */

/**
 * amdgpu_ring_set_fence_errors_and_reemit - Set dma_fence errors and reemit
 *
 * @ring: the ring to operate on
 * @guilty_fence: fence of the ring to update
 *
 */
void amdgpu_ring_set_fence_errors_and_reemit(struct amdgpu_ring *ring,
					     struct amdgpu_fence *guilty_fence)
{
	struct dma_fence *unprocessed;
	struct dma_fence __rcu **ptr;
	struct amdgpu_fence *fence;
	unsigned long flags;
	u32 seq, last_seq;
	unsigned int i;
	bool is_guilty_fence;
	bool is_guilty_context;

	last_seq = amdgpu_fence_read(ring);
	seq = ring->fence_drv.sync_seq;

	/* Hardware completion can precede the interrupt which signals fences. */
	if (last_seq == seq) {
		amdgpu_fence_process(ring);
		return;
	}

	/* If there is nothing to reemit, return early and set an error on the fence
	 * if applicable. If all of the fences are siganlled, this will be a nop.
	 * if there are still fences and ring_backup_entries_to_copy is 0, then
	 * we are skipping it on purpose.
	 */
	if (!ring->ring_backup_entries_to_copy) {
		if (guilty_fence)
			amdgpu_fence_driver_force_completion(ring,
						     &guilty_fence->base);
		return;
	}
	ring->reemit = true;
	amdgpu_ring_alloc(ring, ring->ring_backup_entries_to_copy);
	spin_lock_irqsave(&ring->fence_drv.lock, flags);
	do {
		last_seq++;

		ptr = &ring->fence_drv.fences[last_seq & ring->fence_drv.num_fences_mask];
		rcu_read_lock();
		unprocessed = rcu_dereference(*ptr);

		if (unprocessed && !dma_fence_is_signaled_locked(unprocessed)) {
			fence = container_of(unprocessed, struct amdgpu_fence, base);
			is_guilty_fence = fence == guilty_fence;
			is_guilty_context = guilty_fence ?
				(fence->context == guilty_fence->context) : false;

			/* mark all fences from the guilty context with an error */
			if (is_guilty_fence)
				dma_fence_set_error(&fence->base, -ETIME);
			else if (is_guilty_context)
				dma_fence_set_error(&fence->base, -ECANCELED);

			/* reemit the packet stream and update wptrs */
			fence->ib_wptr = ring->wptr;
			for (i = 0; i < fence->ib_dw_size; i++) {
				/* Skip the IB(s) for the guilty context. */
				if (is_guilty_context &&
				    i >= fence->skip_ib_dw_start_offset &&
				    i < fence->skip_ib_dw_end_offset)
					amdgpu_ring_write(ring, ring->funcs->nop);
				else
					amdgpu_ring_write(ring,
							  ring->ring_backup[fence->backup_idx + i]);
			}
		}
		rcu_read_unlock();
	} while (last_seq != seq);
	spin_unlock_irqrestore(&ring->fence_drv.lock, flags);
	amdgpu_ring_commit(ring);
	ring->reemit = false;
}

static void amdgpu_ring_backup_unprocessed_command(struct amdgpu_ring *ring,
						   struct amdgpu_fence *af)
{
	unsigned int first_idx = af->ib_wptr & ring->buf_mask;
	unsigned int dw_size = af->ib_dw_size;
	unsigned int i;

	af->backup_idx = ring->ring_backup_entries_to_copy;
	/* Backup the contents of the ring buffer. */
	for (i = first_idx; dw_size > 0; ++i, i &= ring->buf_mask, --dw_size)
		ring->ring_backup[ring->ring_backup_entries_to_copy++] = ring->ring[i];
}

void amdgpu_ring_backup_unprocessed_commands(struct amdgpu_ring *ring,
					     struct amdgpu_fence *guilty_fence)
{
	struct dma_fence *unprocessed;
	struct dma_fence __rcu **ptr;
	struct amdgpu_fence *fence;
	u32 seq, last_seq;

	last_seq = amdgpu_fence_read(ring);
	seq = ring->fence_drv.sync_seq;
	ring->ring_backup_entries_to_copy = 0;

	/*
	 * Avoid replay loops for a repeated guilty fence. NULL identifies a
	 * collateral ring, whose pending commands still need backup.
	 */
	if (guilty_fence && ring->guilty_fence == guilty_fence) {
		ring->guilty_fence = NULL;
		return;
	}
	ring->guilty_fence = guilty_fence;

	while (last_seq != seq) {
		last_seq++;

		ptr = &ring->fence_drv.fences[last_seq & ring->fence_drv.num_fences_mask];
		rcu_read_lock();
		unprocessed = rcu_dereference(*ptr);

		if (unprocessed && !dma_fence_is_signaled(unprocessed)) {
			fence = container_of(unprocessed, struct amdgpu_fence, base);

			amdgpu_ring_backup_unprocessed_command(ring, fence);
		}
		rcu_read_unlock();
	}
}

struct amdgpu_fence *
amdgpu_ring_find_guilty_fence(struct amdgpu_ring *ring)
{
	struct dma_fence *unprocessed;
	struct dma_fence __rcu **ptr;
	struct amdgpu_fence *fence;
	u32 seq, last_seq;

	last_seq = amdgpu_fence_read(ring) & ring->fence_drv.num_fences_mask;
	seq = ring->fence_drv.sync_seq & ring->fence_drv.num_fences_mask;

	do {
		last_seq++;
		last_seq &= ring->fence_drv.num_fences_mask;

		ptr = &ring->fence_drv.fences[last_seq];
		rcu_read_lock();
		unprocessed = rcu_dereference(*ptr);

		if (unprocessed && !dma_fence_is_signaled(unprocessed)) {
			fence = container_of(unprocessed, struct amdgpu_fence, base);
			rcu_read_unlock();
			return fence;
		}
		rcu_read_unlock();
	} while (last_seq != seq);

	return NULL;
}

/*
 * Common fence implementation
 */

static const char *amdgpu_fence_get_driver_name(struct dma_fence *fence)
{
	return "amdgpu";
}

static const char *amdgpu_fence_get_timeline_name(struct dma_fence *f)
{
	return (const char *)to_amdgpu_fence(f)->ring->name;
}

/**
 * amdgpu_fence_enable_signaling - enable signalling on fence
 * @f: fence
 *
 * This function is called with fence_queue lock held, and adds a callback
 * to fence_queue that checks if this fence is signaled, and if so it
 * signals the fence and removes itself.
 */
static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
{
	if (!timer_pending(&to_amdgpu_fence(f)->ring->fence_drv.fallback_timer))
		amdgpu_fence_schedule_fallback(to_amdgpu_fence(f)->ring);

	return true;
}

/**
 * amdgpu_fence_free - free up the fence memory
 *
 * @rcu: RCU callback head
 *
 * Free up the fence memory after the RCU grace period.
 */
static void amdgpu_fence_free(struct rcu_head *rcu)
{
	struct dma_fence *f = container_of(rcu, struct dma_fence, rcu);

	/* free fence_slab if it's separated fence*/
	kfree(to_amdgpu_fence(f));
}

/**
 * amdgpu_fence_release - callback that fence can be freed
 *
 * @f: fence
 *
 * This function is called when the reference count becomes zero.
 * It just RCU schedules freeing up the fence.
 */
static void amdgpu_fence_release(struct dma_fence *f)
{
	call_rcu(&f->rcu, amdgpu_fence_free);
}

static const struct dma_fence_ops amdgpu_fence_ops = {
	.get_driver_name = amdgpu_fence_get_driver_name,
	.get_timeline_name = amdgpu_fence_get_timeline_name,
	.enable_signaling = amdgpu_fence_enable_signaling,
	.release = amdgpu_fence_release,
};

/*
 * Fence debugfs
 */
#if defined(CONFIG_DEBUG_FS)
static int amdgpu_debugfs_fence_info_show(struct seq_file *m, void *unused)
{
	struct amdgpu_device *adev = m->private;
	int i;

	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
		struct amdgpu_ring *ring = adev->rings[i];

		if (!ring || !ring->fence_drv.initialized)
			continue;

		amdgpu_fence_process(ring);

		seq_printf(m, "--- ring %d (%s) ---\n", i, ring->name);
		seq_printf(m, "Last signaled fence          0x%08x\n",
			   atomic_read(&ring->fence_drv.last_seq));
		seq_printf(m, "Last emitted                 0x%08x\n",
			   ring->fence_drv.sync_seq);

		if (ring->funcs->type == AMDGPU_RING_TYPE_GFX ||
		    ring->funcs->type == AMDGPU_RING_TYPE_SDMA) {
			seq_printf(m, "Last signaled trailing fence 0x%08x\n",
				   le32_to_cpu(*ring->trail_fence_cpu_addr));
			seq_printf(m, "Last emitted                 0x%08x\n",
				   ring->trail_seq);
		}

		if (ring->funcs->type != AMDGPU_RING_TYPE_GFX)
			continue;

		/* set in CP_VMID_PREEMPT and preemption occurred */
		seq_printf(m, "Last preempted               0x%08x\n",
			   le32_to_cpu(*(ring->fence_drv.cpu_addr + 2)));
		/* set in CP_VMID_RESET and reset occurred */
		seq_printf(m, "Last reset                   0x%08x\n",
			   le32_to_cpu(*(ring->fence_drv.cpu_addr + 4)));
		/* Both preemption and reset occurred */
		seq_printf(m, "Last both                    0x%08x\n",
			   le32_to_cpu(*(ring->fence_drv.cpu_addr + 6)));
	}
	return 0;
}

/*
 * amdgpu_debugfs_gpu_recover - manually trigger a gpu reset & recover
 *
 * Manually trigger a gpu reset at the next fence wait.
 */
static int gpu_recover_get(void *data, u64 *val)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)data;
	struct drm_device *dev = adev_to_drm(adev);
	int r;

	r = pm_runtime_get_sync(dev->dev);
	if (r < 0) {
		pm_runtime_put_autosuspend(dev->dev);
		return 0;
	}

	if (amdgpu_reset_domain_schedule(adev->reset_domain, &adev->reset_work))
		flush_work(&adev->reset_work);

	*val = atomic_read(&adev->reset_domain->reset_res);

	pm_runtime_put_autosuspend(dev->dev);

	return 0;
}

DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_fence_info);
DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_debugfs_gpu_recover_fops, gpu_recover_get, NULL,
			 "%lld\n");

static void amdgpu_debugfs_reset_work(struct work_struct *work)
{
	struct amdgpu_device *adev = container_of(work, struct amdgpu_device,
						  reset_work);

	struct amdgpu_reset_context reset_context;

	memset(&reset_context, 0, sizeof(reset_context));

	reset_context.method = AMD_RESET_METHOD_NONE;
	reset_context.reset_req_dev = adev;
	reset_context.src = AMDGPU_RESET_SRC_USER;
	set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
	set_bit(AMDGPU_SKIP_COREDUMP, &reset_context.flags);

	amdgpu_device_gpu_recover(adev, NULL, &reset_context);
}

#endif

void amdgpu_debugfs_fence_init(struct amdgpu_device *adev)
{
#if defined(CONFIG_DEBUG_FS)
	struct drm_minor *minor = adev_to_drm(adev)->primary;
	struct dentry *root = minor->debugfs_root;

	debugfs_create_file("amdgpu_fence_info", 0444, root, adev,
			    &amdgpu_debugfs_fence_info_fops);

	if (!amdgpu_sriov_vf(adev)) {

		INIT_WORK(&adev->reset_work, amdgpu_debugfs_reset_work);
		debugfs_create_file("amdgpu_gpu_recover", 0444, root, adev,
				    &amdgpu_debugfs_gpu_recover_fops);
	}
#endif
}