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, j; 49 50 for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { 51 if (adev->jpeg.harvest_config & (1 << i)) 52 continue; 53 54 for (j = 0; j < adev->jpeg.num_jpeg_rings; ++j) 55 amdgpu_ring_fini(&adev->jpeg.inst[i].ring_dec[j]); 56 } 57 58 mutex_destroy(&adev->jpeg.jpeg_pg_lock); 59 60 return 0; 61 } 62 63 int amdgpu_jpeg_suspend(struct amdgpu_device *adev) 64 { 65 cancel_delayed_work_sync(&adev->jpeg.idle_work); 66 67 return 0; 68 } 69 70 int amdgpu_jpeg_resume(struct amdgpu_device *adev) 71 { 72 return 0; 73 } 74 75 static void amdgpu_jpeg_idle_work_handler(struct work_struct *work) 76 { 77 struct amdgpu_device *adev = 78 container_of(work, struct amdgpu_device, jpeg.idle_work.work); 79 unsigned int fences = 0; 80 unsigned int i, j; 81 82 for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { 83 if (adev->jpeg.harvest_config & (1 << i)) 84 continue; 85 86 for (j = 0; j < adev->jpeg.num_jpeg_rings; ++j) 87 fences += amdgpu_fence_count_emitted(&adev->jpeg.inst[i].ring_dec[j]); 88 } 89 90 if (!fences && !atomic_read(&adev->jpeg.total_submission_cnt)) 91 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_JPEG, 92 AMD_PG_STATE_GATE); 93 else 94 schedule_delayed_work(&adev->jpeg.idle_work, JPEG_IDLE_TIMEOUT); 95 } 96 97 void amdgpu_jpeg_ring_begin_use(struct amdgpu_ring *ring) 98 { 99 struct amdgpu_device *adev = ring->adev; 100 101 atomic_inc(&adev->jpeg.total_submission_cnt); 102 cancel_delayed_work_sync(&adev->jpeg.idle_work); 103 104 mutex_lock(&adev->jpeg.jpeg_pg_lock); 105 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_JPEG, 106 AMD_PG_STATE_UNGATE); 107 mutex_unlock(&adev->jpeg.jpeg_pg_lock); 108 } 109 110 void amdgpu_jpeg_ring_end_use(struct amdgpu_ring *ring) 111 { 112 atomic_dec(&ring->adev->jpeg.total_submission_cnt); 113 schedule_delayed_work(&ring->adev->jpeg.idle_work, JPEG_IDLE_TIMEOUT); 114 } 115 116 int amdgpu_jpeg_dec_ring_test_ring(struct amdgpu_ring *ring) 117 { 118 struct amdgpu_device *adev = ring->adev; 119 uint32_t tmp = 0; 120 unsigned i; 121 int r; 122 123 /* JPEG in SRIOV does not support direct register read/write */ 124 if (amdgpu_sriov_vf(adev)) 125 return 0; 126 127 r = amdgpu_ring_alloc(ring, 3); 128 if (r) 129 return r; 130 131 WREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch[ring->pipe], 0xCAFEDEAD); 132 /* Add a read register to make sure the write register is executed. */ 133 RREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch[ring->pipe]); 134 135 amdgpu_ring_write(ring, PACKET0(adev->jpeg.internal.jpeg_pitch[ring->pipe], 0)); 136 amdgpu_ring_write(ring, 0xABADCAFE); 137 amdgpu_ring_commit(ring); 138 139 for (i = 0; i < adev->usec_timeout; i++) { 140 tmp = RREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch[ring->pipe]); 141 if (tmp == 0xABADCAFE) 142 break; 143 udelay(1); 144 } 145 146 if (i >= adev->usec_timeout) 147 r = -ETIMEDOUT; 148 149 return r; 150 } 151 152 static int amdgpu_jpeg_dec_set_reg(struct amdgpu_ring *ring, uint32_t handle, 153 struct dma_fence **fence) 154 { 155 struct amdgpu_device *adev = ring->adev; 156 struct amdgpu_job *job; 157 struct amdgpu_ib *ib; 158 struct dma_fence *f = NULL; 159 const unsigned ib_size_dw = 16; 160 int i, r; 161 162 r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, ib_size_dw * 4, 163 AMDGPU_IB_POOL_DIRECT, &job); 164 if (r) 165 return r; 166 167 ib = &job->ibs[0]; 168 169 ib->ptr[0] = PACKETJ(adev->jpeg.internal.jpeg_pitch[ring->pipe], 0, 0, PACKETJ_TYPE0); 170 ib->ptr[1] = 0xDEADBEEF; 171 for (i = 2; i < 16; i += 2) { 172 ib->ptr[i] = PACKETJ(0, 0, 0, PACKETJ_TYPE6); 173 ib->ptr[i+1] = 0; 174 } 175 ib->length_dw = 16; 176 177 r = amdgpu_job_submit_direct(job, ring, &f); 178 if (r) 179 goto err; 180 181 if (fence) 182 *fence = dma_fence_get(f); 183 dma_fence_put(f); 184 185 return 0; 186 187 err: 188 amdgpu_job_free(job); 189 return r; 190 } 191 192 int amdgpu_jpeg_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout) 193 { 194 struct amdgpu_device *adev = ring->adev; 195 uint32_t tmp = 0; 196 unsigned i; 197 struct dma_fence *fence = NULL; 198 long r = 0; 199 200 r = amdgpu_jpeg_dec_set_reg(ring, 1, &fence); 201 if (r) 202 goto error; 203 204 r = dma_fence_wait_timeout(fence, false, timeout); 205 if (r == 0) { 206 r = -ETIMEDOUT; 207 goto error; 208 } else if (r < 0) { 209 goto error; 210 } else { 211 r = 0; 212 } 213 if (!amdgpu_sriov_vf(adev)) { 214 for (i = 0; i < adev->usec_timeout; i++) { 215 tmp = RREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch[ring->pipe]); 216 if (tmp == 0xDEADBEEF) 217 break; 218 udelay(1); 219 } 220 221 if (i >= adev->usec_timeout) 222 r = -ETIMEDOUT; 223 } 224 225 dma_fence_put(fence); 226 error: 227 return r; 228 } 229 230 int amdgpu_jpeg_process_poison_irq(struct amdgpu_device *adev, 231 struct amdgpu_irq_src *source, 232 struct amdgpu_iv_entry *entry) 233 { 234 struct ras_common_if *ras_if = adev->jpeg.ras_if; 235 struct ras_dispatch_if ih_data = { 236 .entry = entry, 237 }; 238 239 if (!ras_if) 240 return 0; 241 242 ih_data.head = *ras_if; 243 amdgpu_ras_interrupt_dispatch(adev, &ih_data); 244 245 return 0; 246 } 247 248 int amdgpu_jpeg_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block) 249 { 250 int r, i; 251 252 r = amdgpu_ras_block_late_init(adev, ras_block); 253 if (r) 254 return r; 255 256 if (amdgpu_ras_is_supported(adev, ras_block->block)) { 257 for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { 258 if (adev->jpeg.harvest_config & (1 << i)) 259 continue; 260 261 r = amdgpu_irq_get(adev, &adev->jpeg.inst[i].ras_poison_irq, 0); 262 if (r) 263 goto late_fini; 264 } 265 } 266 return 0; 267 268 late_fini: 269 amdgpu_ras_block_late_fini(adev, ras_block); 270 return r; 271 } 272 273 int amdgpu_jpeg_ras_sw_init(struct amdgpu_device *adev) 274 { 275 int err; 276 struct amdgpu_jpeg_ras *ras; 277 278 if (!adev->jpeg.ras) 279 return 0; 280 281 ras = adev->jpeg.ras; 282 err = amdgpu_ras_register_ras_block(adev, &ras->ras_block); 283 if (err) { 284 dev_err(adev->dev, "Failed to register jpeg ras block!\n"); 285 return err; 286 } 287 288 strcpy(ras->ras_block.ras_comm.name, "jpeg"); 289 ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__JPEG; 290 ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__POISON; 291 adev->jpeg.ras_if = &ras->ras_block.ras_comm; 292 293 if (!ras->ras_block.ras_late_init) 294 ras->ras_block.ras_late_init = amdgpu_jpeg_ras_late_init; 295 296 return 0; 297 } 298