Subversion Repositories Kolibri OS

Rev

Rev 5078 | Go to most recent revision | Only display areas with differences | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed

Rev 5078 Rev 5271
1
/*
1
/*
2
 * Copyright 2010 Advanced Micro Devices, Inc.
2
 * Copyright 2010 Advanced Micro Devices, Inc.
3
 *
3
 *
4
 * Permission is hereby granted, free of charge, to any person obtaining a
4
 * Permission is hereby granted, free of charge, to any person obtaining a
5
 * copy of this software and associated documentation files (the "Software"),
5
 * copy of this software and associated documentation files (the "Software"),
6
 * to deal in the Software without restriction, including without limitation
6
 * to deal in the Software without restriction, including without limitation
7
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
7
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8
 * and/or sell copies of the Software, and to permit persons to whom the
8
 * and/or sell copies of the Software, and to permit persons to whom the
9
 * Software is furnished to do so, subject to the following conditions:
9
 * Software is furnished to do so, subject to the following conditions:
10
 *
10
 *
11
 * The above copyright notice and this permission notice shall be included in
11
 * The above copyright notice and this permission notice shall be included in
12
 * all copies or substantial portions of the Software.
12
 * all copies or substantial portions of the Software.
13
 *
13
 *
14
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
16
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
17
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
18
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
19
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20
 * OTHER DEALINGS IN THE SOFTWARE.
20
 * OTHER DEALINGS IN THE SOFTWARE.
21
 *
21
 *
22
 * Authors: Alex Deucher
22
 * Authors: Alex Deucher
23
 */
23
 */
24
#include 
24
#include 
25
#include "radeon.h"
25
#include "radeon.h"
26
#include "radeon_asic.h"
26
#include "radeon_asic.h"
27
#include "evergreend.h"
27
#include "evergreend.h"
28
 
28
 
29
u32 evergreen_gpu_check_soft_reset(struct radeon_device *rdev);
29
u32 evergreen_gpu_check_soft_reset(struct radeon_device *rdev);
30
 
30
 
31
/**
31
/**
32
 * evergreen_dma_fence_ring_emit - emit a fence on the DMA ring
32
 * evergreen_dma_fence_ring_emit - emit a fence on the DMA ring
33
 *
33
 *
34
 * @rdev: radeon_device pointer
34
 * @rdev: radeon_device pointer
35
 * @fence: radeon fence object
35
 * @fence: radeon fence object
36
 *
36
 *
37
 * Add a DMA fence packet to the ring to write
37
 * Add a DMA fence packet to the ring to write
38
 * the fence seq number and DMA trap packet to generate
38
 * the fence seq number and DMA trap packet to generate
39
 * an interrupt if needed (evergreen-SI).
39
 * an interrupt if needed (evergreen-SI).
40
 */
40
 */
41
void evergreen_dma_fence_ring_emit(struct radeon_device *rdev,
41
void evergreen_dma_fence_ring_emit(struct radeon_device *rdev,
42
				   struct radeon_fence *fence)
42
				   struct radeon_fence *fence)
43
{
43
{
44
	struct radeon_ring *ring = &rdev->ring[fence->ring];
44
	struct radeon_ring *ring = &rdev->ring[fence->ring];
45
	u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
45
	u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
46
	/* write the fence */
46
	/* write the fence */
47
	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0));
47
	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0));
48
	radeon_ring_write(ring, addr & 0xfffffffc);
48
	radeon_ring_write(ring, addr & 0xfffffffc);
49
	radeon_ring_write(ring, (upper_32_bits(addr) & 0xff));
49
	radeon_ring_write(ring, (upper_32_bits(addr) & 0xff));
50
	radeon_ring_write(ring, fence->seq);
50
	radeon_ring_write(ring, fence->seq);
51
	/* generate an interrupt */
51
	/* generate an interrupt */
52
	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_TRAP, 0, 0));
52
	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_TRAP, 0, 0));
53
	/* flush HDP */
53
	/* flush HDP */
54
	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0));
54
	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0));
55
	radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
55
	radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
56
	radeon_ring_write(ring, 1);
56
	radeon_ring_write(ring, 1);
57
}
57
}
58
 
58
 
59
/**
59
/**
60
 * evergreen_dma_ring_ib_execute - schedule an IB on the DMA engine
60
 * evergreen_dma_ring_ib_execute - schedule an IB on the DMA engine
61
 *
61
 *
62
 * @rdev: radeon_device pointer
62
 * @rdev: radeon_device pointer
63
 * @ib: IB object to schedule
63
 * @ib: IB object to schedule
64
 *
64
 *
65
 * Schedule an IB in the DMA ring (evergreen).
65
 * Schedule an IB in the DMA ring (evergreen).
66
 */
66
 */
67
void evergreen_dma_ring_ib_execute(struct radeon_device *rdev,
67
void evergreen_dma_ring_ib_execute(struct radeon_device *rdev,
68
				   struct radeon_ib *ib)
68
				   struct radeon_ib *ib)
69
{
69
{
70
	struct radeon_ring *ring = &rdev->ring[ib->ring];
70
	struct radeon_ring *ring = &rdev->ring[ib->ring];
71
 
71
 
72
	if (rdev->wb.enabled) {
72
	if (rdev->wb.enabled) {
73
		u32 next_rptr = ring->wptr + 4;
73
		u32 next_rptr = ring->wptr + 4;
74
		while ((next_rptr & 7) != 5)
74
		while ((next_rptr & 7) != 5)
75
			next_rptr++;
75
			next_rptr++;
76
		next_rptr += 3;
76
		next_rptr += 3;
77
		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 1));
77
		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 1));
78
		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
78
		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
79
		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff);
79
		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff);
80
		radeon_ring_write(ring, next_rptr);
80
		radeon_ring_write(ring, next_rptr);
81
	}
81
	}
82
 
82
 
83
	/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
83
	/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
84
	 * Pad as necessary with NOPs.
84
	 * Pad as necessary with NOPs.
85
	 */
85
	 */
86
	while ((ring->wptr & 7) != 5)
86
	while ((ring->wptr & 7) != 5)
87
		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0));
87
		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0));
88
	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_INDIRECT_BUFFER, 0, 0));
88
	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_INDIRECT_BUFFER, 0, 0));
89
	radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
89
	radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
90
	radeon_ring_write(ring, (ib->length_dw << 12) | (upper_32_bits(ib->gpu_addr) & 0xFF));
90
	radeon_ring_write(ring, (ib->length_dw << 12) | (upper_32_bits(ib->gpu_addr) & 0xFF));
91
 
91
 
92
}
92
}
93
 
93
 
94
/**
94
/**
95
 * evergreen_copy_dma - copy pages using the DMA engine
95
 * evergreen_copy_dma - copy pages using the DMA engine
96
 *
96
 *
97
 * @rdev: radeon_device pointer
97
 * @rdev: radeon_device pointer
98
 * @src_offset: src GPU address
98
 * @src_offset: src GPU address
99
 * @dst_offset: dst GPU address
99
 * @dst_offset: dst GPU address
100
 * @num_gpu_pages: number of GPU pages to xfer
100
 * @num_gpu_pages: number of GPU pages to xfer
101
 * @fence: radeon fence object
101
 * @fence: radeon fence object
102
 *
102
 *
103
 * Copy GPU paging using the DMA engine (evergreen-cayman).
103
 * Copy GPU paging using the DMA engine (evergreen-cayman).
104
 * Used by the radeon ttm implementation to move pages if
104
 * Used by the radeon ttm implementation to move pages if
105
 * registered as the asic copy callback.
105
 * registered as the asic copy callback.
106
 */
106
 */
107
int evergreen_copy_dma(struct radeon_device *rdev,
107
struct radeon_fence *evergreen_copy_dma(struct radeon_device *rdev,
108
		       uint64_t src_offset, uint64_t dst_offset,
108
					uint64_t src_offset,
-
 
109
					uint64_t dst_offset,
109
		       unsigned num_gpu_pages,
110
		       unsigned num_gpu_pages,
110
		       struct radeon_fence **fence)
111
					struct reservation_object *resv)
111
{
112
{
112
	struct radeon_semaphore *sem = NULL;
113
	struct radeon_fence *fence;
-
 
114
	struct radeon_sync sync;
113
	int ring_index = rdev->asic->copy.dma_ring_index;
115
	int ring_index = rdev->asic->copy.dma_ring_index;
114
	struct radeon_ring *ring = &rdev->ring[ring_index];
116
	struct radeon_ring *ring = &rdev->ring[ring_index];
115
	u32 size_in_dw, cur_size_in_dw;
117
	u32 size_in_dw, cur_size_in_dw;
116
	int i, num_loops;
118
	int i, num_loops;
117
	int r = 0;
119
	int r = 0;
118
 
120
 
119
	r = radeon_semaphore_create(rdev, &sem);
-
 
120
	if (r) {
-
 
121
		DRM_ERROR("radeon: moving bo (%d).\n", r);
-
 
122
		return r;
-
 
123
	}
121
	radeon_sync_create(&sync);
124
 
122
 
125
	size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
123
	size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
126
	num_loops = DIV_ROUND_UP(size_in_dw, 0xfffff);
124
	num_loops = DIV_ROUND_UP(size_in_dw, 0xfffff);
127
	r = radeon_ring_lock(rdev, ring, num_loops * 5 + 11);
125
	r = radeon_ring_lock(rdev, ring, num_loops * 5 + 11);
128
	if (r) {
126
	if (r) {
129
		DRM_ERROR("radeon: moving bo (%d).\n", r);
127
		DRM_ERROR("radeon: moving bo (%d).\n", r);
130
		radeon_semaphore_free(rdev, &sem, NULL);
128
		radeon_sync_free(rdev, &sync, NULL);
131
		return r;
129
		return ERR_PTR(r);
132
	}
130
	}
133
 
131
 
134
	radeon_semaphore_sync_to(sem, *fence);
132
	radeon_sync_resv(rdev, &sync, resv, false);
135
	radeon_semaphore_sync_rings(rdev, sem, ring->idx);
133
	radeon_sync_rings(rdev, &sync, ring->idx);
136
 
134
 
137
	for (i = 0; i < num_loops; i++) {
135
	for (i = 0; i < num_loops; i++) {
138
		cur_size_in_dw = size_in_dw;
136
		cur_size_in_dw = size_in_dw;
139
		if (cur_size_in_dw > 0xFFFFF)
137
		if (cur_size_in_dw > 0xFFFFF)
140
			cur_size_in_dw = 0xFFFFF;
138
			cur_size_in_dw = 0xFFFFF;
141
		size_in_dw -= cur_size_in_dw;
139
		size_in_dw -= cur_size_in_dw;
142
		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, cur_size_in_dw));
140
		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, cur_size_in_dw));
143
		radeon_ring_write(ring, dst_offset & 0xfffffffc);
141
		radeon_ring_write(ring, dst_offset & 0xfffffffc);
144
		radeon_ring_write(ring, src_offset & 0xfffffffc);
142
		radeon_ring_write(ring, src_offset & 0xfffffffc);
145
		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
143
		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
146
		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
144
		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
147
		src_offset += cur_size_in_dw * 4;
145
		src_offset += cur_size_in_dw * 4;
148
		dst_offset += cur_size_in_dw * 4;
146
		dst_offset += cur_size_in_dw * 4;
149
	}
147
	}
150
 
148
 
151
	r = radeon_fence_emit(rdev, fence, ring->idx);
149
	r = radeon_fence_emit(rdev, &fence, ring->idx);
152
	if (r) {
150
	if (r) {
153
		radeon_ring_unlock_undo(rdev, ring);
151
		radeon_ring_unlock_undo(rdev, ring);
154
		radeon_semaphore_free(rdev, &sem, NULL);
152
		radeon_sync_free(rdev, &sync, NULL);
155
		return r;
153
		return ERR_PTR(r);
156
	}
154
	}
157
 
155
 
158
	radeon_ring_unlock_commit(rdev, ring, false);
156
	radeon_ring_unlock_commit(rdev, ring, false);
159
	radeon_semaphore_free(rdev, &sem, *fence);
157
	radeon_sync_free(rdev, &sync, fence);
160
 
158
 
161
	return r;
159
	return fence;
162
}
160
}
163
 
161
 
164
/**
162
/**
165
 * evergreen_dma_is_lockup - Check if the DMA engine is locked up
163
 * evergreen_dma_is_lockup - Check if the DMA engine is locked up
166
 *
164
 *
167
 * @rdev: radeon_device pointer
165
 * @rdev: radeon_device pointer
168
 * @ring: radeon_ring structure holding ring information
166
 * @ring: radeon_ring structure holding ring information
169
 *
167
 *
170
 * Check if the async DMA engine is locked up.
168
 * Check if the async DMA engine is locked up.
171
 * Returns true if the engine appears to be locked up, false if not.
169
 * Returns true if the engine appears to be locked up, false if not.
172
 */
170
 */
173
bool evergreen_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
171
bool evergreen_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
174
{
172
{
175
	u32 reset_mask = evergreen_gpu_check_soft_reset(rdev);
173
	u32 reset_mask = evergreen_gpu_check_soft_reset(rdev);
176
 
174
 
177
	if (!(reset_mask & RADEON_RESET_DMA)) {
175
	if (!(reset_mask & RADEON_RESET_DMA)) {
178
		radeon_ring_lockup_update(rdev, ring);
176
		radeon_ring_lockup_update(rdev, ring);
179
		return false;
177
		return false;
180
	}
178
	}
181
	return radeon_ring_test_lockup(rdev, ring);
179
	return radeon_ring_test_lockup(rdev, ring);
182
}
180
}