/*
* 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 <drm/drmP.h>
#include "amdgpu.h"
#include "amdgpu_trace.h"
/*
* Fences
* Fences mark an event in the GPUs pipeline and are used
* for GPU/CPU synchronization. When the fence is written,
* it is expected that all buffers associated with that fence
* are no longer in use by the associated ring on the GPU and
* that the the relevant GPU caches have been flushed.
*/
/**
* 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 = lower_32_bits(atomic64_read(&drv->last_seq));
return seq;
}
/**
* amdgpu_fence_schedule_check - schedule lockup check
*
* @ring: pointer to struct amdgpu_ring
*
* Queues a delayed work item to check for lockups.
*/
static void amdgpu_fence_schedule_check(struct amdgpu_ring *ring)
{
/*
* Do not reset the timer here with mod_delayed_work,
* this can livelock in an interaction with TTM delayed destroy.
*/
queue_delayed_work(system_power_efficient_wq,
&ring->fence_drv.lockup_work,
AMDGPU_FENCE_JIFFIES_TIMEOUT);
}
/**
* amdgpu_fence_emit - emit a fence on the requested ring
*
* @ring: ring the fence is associated with
* @owner: creator of the fence
* @fence: amdgpu fence object
*
* Emits a fence command on the requested ring (all asics).
* Returns 0 on success, -ENOMEM on failure.
*/
int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner,
struct amdgpu_fence **fence)
{
struct amdgpu_device *adev = ring->adev;
/* we are protected by the ring emission mutex */
*fence = kmalloc(sizeof(struct amdgpu_fence), GFP_KERNEL);
if ((*fence) == NULL) {
return -ENOMEM;
}
(*fence)->seq = ++ring->fence_drv.sync_seq[ring->idx];
(*fence)->ring = ring;
(*fence)->owner = owner;
fence_init(&(*fence)->base, &amdgpu_fence_ops,
&ring->fence_drv.fence_queue.lock,
adev->fence_context + ring->idx,
(*fence)->seq);
amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
(*fence)->seq,
AMDGPU_FENCE_FLAG_INT);
trace_amdgpu_fence_emit(ring->adev->ddev, ring->idx, (*fence)->seq);
return 0;
}
/**
* amdgpu_fence_check_signaled - callback from fence_queue
*
* this function is called with fence_queue lock held, which is also used
* for the fence locking itself, so unlocked variants are used for
* fence_signal, and remove_wait_queue.
*/
static int amdgpu_fence_check_signaled(wait_queue_t *wait, unsigned mode, int flags, void *key)
{
struct amdgpu_fence *fence;
struct amdgpu_device *adev;
u64 seq;
int ret;
fence = container_of(wait, struct amdgpu_fence, fence_wake);
adev = fence->ring->adev;
/*
* We cannot use amdgpu_fence_process here because we're already
* in the waitqueue, in a call from wake_up_all.
*/
seq = atomic64_read(&fence->ring->fence_drv.last_seq);
if (seq >= fence->seq) {
ret = fence_signal_locked(&fence->base);
if (!ret)
FENCE_TRACE(&fence->base, "signaled from irq context\n");
else
FENCE_TRACE(&fence->base, "was already signaled\n");
__remove_wait_queue(&fence->ring->fence_drv.fence_queue, &fence->fence_wake);
fence_put(&fence