1 /*
2  * Copyright 2019 Advanced Micro Devices, Inc.
3  * All Rights Reserved.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the
7  * "Software"), to deal in the Software without restriction, including
8  * without limitation the rights to use, copy, modify, merge, publish,
9  * distribute, sub license, and/or sell copies of the Software, and to
10  * permit persons to whom the Software is furnished to do so, subject to
11  * the following conditions:
12  *
13  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
16  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
17  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
18  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19  * USE OR OTHER DEALINGS IN THE SOFTWARE.
20  *
21  * The above copyright notice and this permission notice (including the
22  * next paragraph) shall be included in all copies or substantial portions
23  * of the Software.
24  *
25  */
26 
27 #include "amdgpu.h"
28 #include "amdgpu_jpeg.h"
29 #include "amdgpu_pm.h"
30 #include "soc15d.h"
31 #include "soc15_common.h"
32 
33 #define JPEG_IDLE_TIMEOUT	msecs_to_jiffies(1000)
34 
35 static void amdgpu_jpeg_idle_work_handler(struct work_struct *work);
36 
37 int amdgpu_jpeg_sw_init(struct amdgpu_device *adev)
38 {
39 	INIT_DELAYED_WORK(&adev->jpeg.idle_work, amdgpu_jpeg_idle_work_handler);
40 	mutex_init(&adev->jpeg.jpeg_pg_lock);
41 	atomic_set(&adev->jpeg.total_submission_cnt, 0);
42 
43 	return 0;
44 }
45 
46 int amdgpu_jpeg_sw_fini(struct amdgpu_device *adev)
47 {
48 	int i;
49 
50 	for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
51 		if (adev->jpeg.harvest_config & (1 << i))
52 			continue;
53 
54 		amdgpu_ring_fini(&adev->jpeg.inst[i].ring_dec);
55 	}
56 
57 	mutex_destroy(&adev->jpeg.jpeg_pg_lock);
58 
59 	return 0;
60 }
61 
62 int amdgpu_jpeg_suspend(struct amdgpu_device *adev)
63 {
64 	cancel_delayed_work_sync(&adev->jpeg.idle_work);
65 
66 	return 0;
67 }
68 
69 int amdgpu_jpeg_resume(struct amdgpu_device *adev)
70 {
71 	return 0;
72 }
73 
74 static void amdgpu_jpeg_idle_work_handler(struct work_struct *work)
75 {
76 	struct amdgpu_device *adev =
77 		container_of(work, struct amdgpu_device, jpeg.idle_work.work);
78 	unsigned int fences = 0;
79 	unsigned int i;
80 
81 	for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
82 		if (adev->jpeg.harvest_config & (1 << i))
83 			continue;
84 
85 		fences += amdgpu_fence_count_emitted(&adev->jpeg.inst[i].ring_dec);
86 	}
87 
88 	if (!fences && !atomic_read(&adev->jpeg.total_submission_cnt))
89 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_JPEG,
90 						       AMD_PG_STATE_GATE);
91 	else
92 		schedule_delayed_work(&adev->jpeg.idle_work, JPEG_IDLE_TIMEOUT);
93 }
94 
95 void amdgpu_jpeg_ring_begin_use(struct amdgpu_ring *ring)
96 {
97 	struct amdgpu_device *adev = ring->adev;
98 
99 	atomic_inc(&adev->jpeg.total_submission_cnt);
100 	cancel_delayed_work_sync(&adev->jpeg.idle_work);
101 
102 	mutex_lock(&adev->jpeg.jpeg_pg_lock);
103 	amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_JPEG,
104 						       AMD_PG_STATE_UNGATE);
105 	mutex_unlock(&adev->jpeg.jpeg_pg_lock);
106 }
107 
108 void amdgpu_jpeg_ring_end_use(struct amdgpu_ring *ring)
109 {
110 	atomic_dec(&ring->adev->jpeg.total_submission_cnt);
111 	schedule_delayed_work(&ring->adev->jpeg.idle_work, JPEG_IDLE_TIMEOUT);
112 }
113 
114 int amdgpu_jpeg_dec_ring_test_ring(struct amdgpu_ring *ring)
115 {
116 	struct amdgpu_device *adev = ring->adev;
117 	uint32_t tmp = 0;
118 	unsigned i;
119 	int r;
120 
121 	WREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch, 0xCAFEDEAD);
122 	r = amdgpu_ring_alloc(ring, 3);
123 	if (r)
124 		return r;
125 
126 	amdgpu_ring_write(ring, PACKET0(adev->jpeg.internal.jpeg_pitch, 0));
127 	amdgpu_ring_write(ring, 0xDEADBEEF);
128 	amdgpu_ring_commit(ring);
129 
130 	for (i = 0; i < adev->usec_timeout; i++) {
131 		tmp = RREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch);
132 		if (tmp == 0xDEADBEEF)
133 			break;
134 		udelay(1);
135 	}
136 
137 	if (i >= adev->usec_timeout)
138 		r = -ETIMEDOUT;
139 
140 	return r;
141 }
142 
143 static int amdgpu_jpeg_dec_set_reg(struct amdgpu_ring *ring, uint32_t handle,
144 		struct dma_fence **fence)
145 {
146 	struct amdgpu_device *adev = ring->adev;
147 	struct amdgpu_job *job;
148 	struct amdgpu_ib *ib;
149 	struct dma_fence *f = NULL;
150 	const unsigned ib_size_dw = 16;
151 	int i, r;
152 
153 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
154 					AMDGPU_IB_POOL_DIRECT, &job);
155 	if (r)
156 		return r;
157 
158 	ib = &job->ibs[0];
159 
160 	ib->ptr[0] = PACKETJ(adev->jpeg.internal.jpeg_pitch, 0, 0, PACKETJ_TYPE0);
161 	ib->ptr[1] = 0xDEADBEEF;
162 	for (i = 2; i < 16; i += 2) {
163 		ib->ptr[i] = PACKETJ(0, 0, 0, PACKETJ_TYPE6);
164 		ib->ptr[i+1] = 0;
165 	}
166 	ib->length_dw = 16;
167 
168 	r = amdgpu_job_submit_direct(job, ring, &f);
169 	if (r)
170 		goto err;
171 
172 	if (fence)
173 		*fence = dma_fence_get(f);
174 	dma_fence_put(f);
175 
176 	return 0;
177 
178 err:
179 	amdgpu_job_free(job);
180 	return r;
181 }
182 
183 int amdgpu_jpeg_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
184 {
185 	struct amdgpu_device *adev = ring->adev;
186 	uint32_t tmp = 0;
187 	unsigned i;
188 	struct dma_fence *fence = NULL;
189 	long r = 0;
190 
191 	r = amdgpu_jpeg_dec_set_reg(ring, 1, &fence);
192 	if (r)
193 		goto error;
194 
195 	r = dma_fence_wait_timeout(fence, false, timeout);
196 	if (r == 0) {
197 		r = -ETIMEDOUT;
198 		goto error;
199 	} else if (r < 0) {
200 		goto error;
201 	} else {
202 		r = 0;
203 	}
204 
205 	for (i = 0; i < adev->usec_timeout; i++) {
206 		tmp = RREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch);
207 		if (tmp == 0xDEADBEEF)
208 			break;
209 		udelay(1);
210 	}
211 
212 	if (i >= adev->usec_timeout)
213 		r = -ETIMEDOUT;
214 
215 	dma_fence_put(fence);
216 error:
217 	return r;
218 }
219