1 /* 2 * Copyright 2016 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 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 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 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, 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 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 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23 #include <drm/drmP.h> 24 #include "amdgpu.h" 25 #include "amdgpu_ih.h" 26 #include "soc15.h" 27 28 29 #include "vega10/soc15ip.h" 30 #include "vega10/OSSSYS/osssys_4_0_offset.h" 31 #include "vega10/OSSSYS/osssys_4_0_sh_mask.h" 32 33 #include "soc15_common.h" 34 #include "vega10_ih.h" 35 36 37 38 static void vega10_ih_set_interrupt_funcs(struct amdgpu_device *adev); 39 40 /** 41 * vega10_ih_enable_interrupts - Enable the interrupt ring buffer 42 * 43 * @adev: amdgpu_device pointer 44 * 45 * Enable the interrupt ring buffer (VEGA10). 46 */ 47 static void vega10_ih_enable_interrupts(struct amdgpu_device *adev) 48 { 49 u32 ih_rb_cntl = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL)); 50 51 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 1); 52 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, ENABLE_INTR, 1); 53 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL), ih_rb_cntl); 54 adev->irq.ih.enabled = true; 55 } 56 57 /** 58 * vega10_ih_disable_interrupts - Disable the interrupt ring buffer 59 * 60 * @adev: amdgpu_device pointer 61 * 62 * Disable the interrupt ring buffer (VEGA10). 63 */ 64 static void vega10_ih_disable_interrupts(struct amdgpu_device *adev) 65 { 66 u32 ih_rb_cntl = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL)); 67 68 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 0); 69 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, ENABLE_INTR, 0); 70 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL), ih_rb_cntl); 71 /* set rptr, wptr to 0 */ 72 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR), 0); 73 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR), 0); 74 adev->irq.ih.enabled = false; 75 adev->irq.ih.rptr = 0; 76 } 77 78 /** 79 * vega10_ih_irq_init - init and enable the interrupt ring 80 * 81 * @adev: amdgpu_device pointer 82 * 83 * Allocate a ring buffer for the interrupt controller, 84 * enable the RLC, disable interrupts, enable the IH 85 * ring buffer and enable it (VI). 86 * Called at device load and reume. 87 * Returns 0 for success, errors for failure. 88 */ 89 static int vega10_ih_irq_init(struct amdgpu_device *adev) 90 { 91 int ret = 0; 92 int rb_bufsz; 93 u32 ih_rb_cntl, ih_doorbell_rtpr; 94 u32 tmp; 95 u64 wptr_off; 96 97 /* disable irqs */ 98 vega10_ih_disable_interrupts(adev); 99 100 if (adev->flags & AMD_IS_APU) 101 nbio_v7_0_ih_control(adev); 102 else 103 nbio_v6_1_ih_control(adev); 104 105 ih_rb_cntl = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL)); 106 /* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/ 107 if (adev->irq.ih.use_bus_addr) { 108 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE), adev->irq.ih.rb_dma_addr >> 8); 109 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_HI), ((u64)adev->irq.ih.rb_dma_addr >> 40) & 0xff); 110 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE, 1); 111 } else { 112 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE), adev->irq.ih.gpu_addr >> 8); 113 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_HI), (adev->irq.ih.gpu_addr >> 40) & 0xff); 114 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE, 4); 115 } 116 rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4); 117 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); 118 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_ENABLE, 1); 119 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_SIZE, rb_bufsz); 120 /* Ring Buffer write pointer writeback. If enabled, IH_RB_WPTR register value is written to memory */ 121 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1); 122 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SNOOP, 1); 123 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_RO, 0); 124 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_VMID, 0); 125 126 if (adev->irq.msi_enabled) 127 ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RPTR_REARM, 1); 128 129 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL), ih_rb_cntl); 130 131 /* set the writeback address whether it's enabled or not */ 132 if (adev->irq.ih.use_bus_addr) 133 wptr_off = adev->irq.ih.rb_dma_addr + (adev->irq.ih.wptr_offs * 4); 134 else 135 wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4); 136 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_ADDR_LO), lower_32_bits(wptr_off)); 137 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_ADDR_HI), upper_32_bits(wptr_off) & 0xFF); 138 139 /* set rptr, wptr to 0 */ 140 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR), 0); 141 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR), 0); 142 143 ih_doorbell_rtpr = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_DOORBELL_RPTR)); 144 if (adev->irq.ih.use_doorbell) { 145 ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR, 146 OFFSET, adev->irq.ih.doorbell_index); 147 ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR, 148 ENABLE, 1); 149 } else { 150 ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR, 151 ENABLE, 0); 152 } 153 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_DOORBELL_RPTR), ih_doorbell_rtpr); 154 if (adev->flags & AMD_IS_APU) 155 nbio_v7_0_ih_doorbell_range(adev, adev->irq.ih.use_doorbell, adev->irq.ih.doorbell_index); 156 else 157 nbio_v6_1_ih_doorbell_range(adev, adev->irq.ih.use_doorbell, adev->irq.ih.doorbell_index); 158 159 tmp = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL)); 160 tmp = REG_SET_FIELD(tmp, IH_STORM_CLIENT_LIST_CNTL, 161 CLIENT18_IS_STORM_CLIENT, 1); 162 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL), tmp); 163 164 tmp = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_INT_FLOOD_CNTL)); 165 tmp = REG_SET_FIELD(tmp, IH_INT_FLOOD_CNTL, FLOOD_CNTL_ENABLE, 1); 166 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_INT_FLOOD_CNTL), tmp); 167 168 pci_set_master(adev->pdev); 169 170 /* enable interrupts */ 171 vega10_ih_enable_interrupts(adev); 172 173 return ret; 174 } 175 176 /** 177 * vega10_ih_irq_disable - disable interrupts 178 * 179 * @adev: amdgpu_device pointer 180 * 181 * Disable interrupts on the hw (VEGA10). 182 */ 183 static void vega10_ih_irq_disable(struct amdgpu_device *adev) 184 { 185 vega10_ih_disable_interrupts(adev); 186 187 /* Wait and acknowledge irq */ 188 mdelay(1); 189 } 190 191 /** 192 * vega10_ih_get_wptr - get the IH ring buffer wptr 193 * 194 * @adev: amdgpu_device pointer 195 * 196 * Get the IH ring buffer wptr from either the register 197 * or the writeback memory buffer (VEGA10). Also check for 198 * ring buffer overflow and deal with it. 199 * Returns the value of the wptr. 200 */ 201 static u32 vega10_ih_get_wptr(struct amdgpu_device *adev) 202 { 203 u32 wptr, tmp; 204 205 if (adev->irq.ih.use_bus_addr) 206 wptr = le32_to_cpu(adev->irq.ih.ring[adev->irq.ih.wptr_offs]); 207 else 208 wptr = le32_to_cpu(adev->wb.wb[adev->irq.ih.wptr_offs]); 209 210 if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) { 211 wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0); 212 213 /* When a ring buffer overflow happen start parsing interrupt 214 * from the last not overwritten vector (wptr + 32). Hopefully 215 * this should allow us to catchup. 216 */ 217 tmp = (wptr + 32) & adev->irq.ih.ptr_mask; 218 dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", 219 wptr, adev->irq.ih.rptr, tmp); 220 adev->irq.ih.rptr = tmp; 221 222 tmp = RREG32_NO_KIQ(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL)); 223 tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); 224 WREG32_NO_KIQ(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL), tmp); 225 } 226 return (wptr & adev->irq.ih.ptr_mask); 227 } 228 229 /** 230 * vega10_ih_prescreen_iv - prescreen an interrupt vector 231 * 232 * @adev: amdgpu_device pointer 233 * 234 * Returns true if the interrupt vector should be further processed. 235 */ 236 static bool vega10_ih_prescreen_iv(struct amdgpu_device *adev) 237 { 238 u32 ring_index = adev->irq.ih.rptr >> 2; 239 u32 dw0, dw3, dw4, dw5; 240 u16 pasid; 241 u64 addr, key; 242 struct amdgpu_vm *vm; 243 int r; 244 245 dw0 = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]); 246 dw3 = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]); 247 dw4 = le32_to_cpu(adev->irq.ih.ring[ring_index + 4]); 248 dw5 = le32_to_cpu(adev->irq.ih.ring[ring_index + 5]); 249 250 /* Filter retry page faults, let only the first one pass. If 251 * there are too many outstanding faults, ignore them until 252 * some faults get cleared. 253 */ 254 switch (dw0 & 0xff) { 255 case AMDGPU_IH_CLIENTID_VMC: 256 case AMDGPU_IH_CLIENTID_UTCL2: 257 break; 258 default: 259 /* Not a VM fault */ 260 return true; 261 } 262 263 /* Not a retry fault */ 264 if (!(dw5 & 0x80)) 265 return true; 266 267 pasid = dw3 & 0xffff; 268 /* No PASID, can't identify faulting process */ 269 if (!pasid) 270 return true; 271 272 addr = ((u64)(dw5 & 0xf) << 44) | ((u64)dw4 << 12); 273 key = AMDGPU_VM_FAULT(pasid, addr); 274 r = amdgpu_ih_add_fault(adev, key); 275 276 /* Hash table is full or the fault is already being processed, 277 * ignore further page faults 278 */ 279 if (r != 0) 280 goto ignore_iv; 281 282 /* Track retry faults in per-VM fault FIFO. */ 283 spin_lock(&adev->vm_manager.pasid_lock); 284 vm = idr_find(&adev->vm_manager.pasid_idr, pasid); 285 spin_unlock(&adev->vm_manager.pasid_lock); 286 if (WARN_ON_ONCE(!vm)) { 287 /* VM not found, process it normally */ 288 amdgpu_ih_clear_fault(adev, key); 289 return true; 290 } 291 /* No locking required with single writer and single reader */ 292 r = kfifo_put(&vm->faults, key); 293 if (!r) { 294 /* FIFO is full. Ignore it until there is space */ 295 amdgpu_ih_clear_fault(adev, key); 296 goto ignore_iv; 297 } 298 299 /* It's the first fault for this address, process it normally */ 300 return true; 301 302 ignore_iv: 303 adev->irq.ih.rptr += 32; 304 return false; 305 } 306 307 /** 308 * vega10_ih_decode_iv - decode an interrupt vector 309 * 310 * @adev: amdgpu_device pointer 311 * 312 * Decodes the interrupt vector at the current rptr 313 * position and also advance the position. 314 */ 315 static void vega10_ih_decode_iv(struct amdgpu_device *adev, 316 struct amdgpu_iv_entry *entry) 317 { 318 /* wptr/rptr are in bytes! */ 319 u32 ring_index = adev->irq.ih.rptr >> 2; 320 uint32_t dw[8]; 321 322 dw[0] = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]); 323 dw[1] = le32_to_cpu(adev->irq.ih.ring[ring_index + 1]); 324 dw[2] = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]); 325 dw[3] = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]); 326 dw[4] = le32_to_cpu(adev->irq.ih.ring[ring_index + 4]); 327 dw[5] = le32_to_cpu(adev->irq.ih.ring[ring_index + 5]); 328 dw[6] = le32_to_cpu(adev->irq.ih.ring[ring_index + 6]); 329 dw[7] = le32_to_cpu(adev->irq.ih.ring[ring_index + 7]); 330 331 entry->client_id = dw[0] & 0xff; 332 entry->src_id = (dw[0] >> 8) & 0xff; 333 entry->ring_id = (dw[0] >> 16) & 0xff; 334 entry->vm_id = (dw[0] >> 24) & 0xf; 335 entry->vm_id_src = (dw[0] >> 31); 336 entry->timestamp = dw[1] | ((u64)(dw[2] & 0xffff) << 32); 337 entry->timestamp_src = dw[2] >> 31; 338 entry->pas_id = dw[3] & 0xffff; 339 entry->pasid_src = dw[3] >> 31; 340 entry->src_data[0] = dw[4]; 341 entry->src_data[1] = dw[5]; 342 entry->src_data[2] = dw[6]; 343 entry->src_data[3] = dw[7]; 344 345 346 /* wptr/rptr are in bytes! */ 347 adev->irq.ih.rptr += 32; 348 } 349 350 /** 351 * vega10_ih_set_rptr - set the IH ring buffer rptr 352 * 353 * @adev: amdgpu_device pointer 354 * 355 * Set the IH ring buffer rptr. 356 */ 357 static void vega10_ih_set_rptr(struct amdgpu_device *adev) 358 { 359 if (adev->irq.ih.use_doorbell) { 360 /* XXX check if swapping is necessary on BE */ 361 if (adev->irq.ih.use_bus_addr) 362 adev->irq.ih.ring[adev->irq.ih.rptr_offs] = adev->irq.ih.rptr; 363 else 364 adev->wb.wb[adev->irq.ih.rptr_offs] = adev->irq.ih.rptr; 365 WDOORBELL32(adev->irq.ih.doorbell_index, adev->irq.ih.rptr); 366 } else { 367 WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR), adev->irq.ih.rptr); 368 } 369 } 370 371 static int vega10_ih_early_init(void *handle) 372 { 373 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 374 375 vega10_ih_set_interrupt_funcs(adev); 376 return 0; 377 } 378 379 static int vega10_ih_sw_init(void *handle) 380 { 381 int r; 382 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 383 384 r = amdgpu_ih_ring_init(adev, 256 * 1024, true); 385 if (r) 386 return r; 387 388 adev->irq.ih.use_doorbell = true; 389 adev->irq.ih.doorbell_index = AMDGPU_DOORBELL64_IH << 1; 390 391 adev->irq.ih.faults = kmalloc(sizeof(*adev->irq.ih.faults), GFP_KERNEL); 392 if (!adev->irq.ih.faults) 393 return -ENOMEM; 394 INIT_CHASH_TABLE(adev->irq.ih.faults->hash, 395 AMDGPU_PAGEFAULT_HASH_BITS, 8, 0); 396 spin_lock_init(&adev->irq.ih.faults->lock); 397 adev->irq.ih.faults->count = 0; 398 399 r = amdgpu_irq_init(adev); 400 401 return r; 402 } 403 404 static int vega10_ih_sw_fini(void *handle) 405 { 406 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 407 408 amdgpu_irq_fini(adev); 409 amdgpu_ih_ring_fini(adev); 410 411 kfree(adev->irq.ih.faults); 412 adev->irq.ih.faults = NULL; 413 414 return 0; 415 } 416 417 static int vega10_ih_hw_init(void *handle) 418 { 419 int r; 420 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 421 422 r = vega10_ih_irq_init(adev); 423 if (r) 424 return r; 425 426 return 0; 427 } 428 429 static int vega10_ih_hw_fini(void *handle) 430 { 431 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 432 433 vega10_ih_irq_disable(adev); 434 435 return 0; 436 } 437 438 static int vega10_ih_suspend(void *handle) 439 { 440 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 441 442 return vega10_ih_hw_fini(adev); 443 } 444 445 static int vega10_ih_resume(void *handle) 446 { 447 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 448 449 return vega10_ih_hw_init(adev); 450 } 451 452 static bool vega10_ih_is_idle(void *handle) 453 { 454 /* todo */ 455 return true; 456 } 457 458 static int vega10_ih_wait_for_idle(void *handle) 459 { 460 /* todo */ 461 return -ETIMEDOUT; 462 } 463 464 static int vega10_ih_soft_reset(void *handle) 465 { 466 /* todo */ 467 468 return 0; 469 } 470 471 static int vega10_ih_set_clockgating_state(void *handle, 472 enum amd_clockgating_state state) 473 { 474 return 0; 475 } 476 477 static int vega10_ih_set_powergating_state(void *handle, 478 enum amd_powergating_state state) 479 { 480 return 0; 481 } 482 483 const struct amd_ip_funcs vega10_ih_ip_funcs = { 484 .name = "vega10_ih", 485 .early_init = vega10_ih_early_init, 486 .late_init = NULL, 487 .sw_init = vega10_ih_sw_init, 488 .sw_fini = vega10_ih_sw_fini, 489 .hw_init = vega10_ih_hw_init, 490 .hw_fini = vega10_ih_hw_fini, 491 .suspend = vega10_ih_suspend, 492 .resume = vega10_ih_resume, 493 .is_idle = vega10_ih_is_idle, 494 .wait_for_idle = vega10_ih_wait_for_idle, 495 .soft_reset = vega10_ih_soft_reset, 496 .set_clockgating_state = vega10_ih_set_clockgating_state, 497 .set_powergating_state = vega10_ih_set_powergating_state, 498 }; 499 500 static const struct amdgpu_ih_funcs vega10_ih_funcs = { 501 .get_wptr = vega10_ih_get_wptr, 502 .prescreen_iv = vega10_ih_prescreen_iv, 503 .decode_iv = vega10_ih_decode_iv, 504 .set_rptr = vega10_ih_set_rptr 505 }; 506 507 static void vega10_ih_set_interrupt_funcs(struct amdgpu_device *adev) 508 { 509 if (adev->irq.ih_funcs == NULL) 510 adev->irq.ih_funcs = &vega10_ih_funcs; 511 } 512 513 const struct amdgpu_ip_block_version vega10_ih_ip_block = 514 { 515 .type = AMD_IP_BLOCK_TYPE_IH, 516 .major = 4, 517 .minor = 0, 518 .rev = 0, 519 .funcs = &vega10_ih_ip_funcs, 520 }; 521