1/* 2 * Copyright 2015 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 * Authors: Alex Deucher 23 */ 24 25#include "amdgpu.h" 26#include "amdgpu_trace.h" 27#include "si.h" 28#include "sid.h" 29 30const u32 sdma_offsets[SDMA_MAX_INSTANCE] = 31{ 32 DMA0_REGISTER_OFFSET, 33 DMA1_REGISTER_OFFSET 34}; 35 36static void si_dma_set_ring_funcs(struct amdgpu_device *adev); 37static void si_dma_set_buffer_funcs(struct amdgpu_device *adev); 38static void si_dma_set_vm_pte_funcs(struct amdgpu_device *adev); 39static void si_dma_set_irq_funcs(struct amdgpu_device *adev); 40 41static uint64_t si_dma_ring_get_rptr(struct amdgpu_ring *ring) 42{ 43 return *ring->rptr_cpu_addr; 44} 45 46static uint64_t si_dma_ring_get_wptr(struct amdgpu_ring *ring) 47{ 48 struct amdgpu_device *adev = ring->adev; 49 u32 me = (ring == &adev->sdma.instance[0].ring) ? 0 : 1; 50 51 return (RREG32(DMA_RB_WPTR + sdma_offsets[me]) & 0x3fffc) >> 2; 52} 53 54static void si_dma_ring_set_wptr(struct amdgpu_ring *ring) 55{ 56 struct amdgpu_device *adev = ring->adev; 57 u32 me = (ring == &adev->sdma.instance[0].ring) ? 0 : 1; 58 59 WREG32(DMA_RB_WPTR + sdma_offsets[me], (ring->wptr << 2) & 0x3fffc); 60} 61 62static void si_dma_ring_emit_ib(struct amdgpu_ring *ring, 63 struct amdgpu_job *job, 64 struct amdgpu_ib *ib, 65 uint32_t flags) 66{ 67 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 68 /* The indirect buffer packet must end on an 8 DW boundary in the DMA ring. 69 * Pad as necessary with NOPs. 70 */ 71 while ((lower_32_bits(ring->wptr) & 7) != 5) 72 amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0)); 73 amdgpu_ring_write(ring, DMA_IB_PACKET(DMA_PACKET_INDIRECT_BUFFER, vmid, 0)); 74 amdgpu_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0)); 75 amdgpu_ring_write(ring, (ib->length_dw << 12) | (upper_32_bits(ib->gpu_addr) & 0xFF)); 76 77} 78 79/** 80 * si_dma_ring_emit_fence - emit a fence on the DMA ring 81 * 82 * @ring: amdgpu ring pointer 83 * @addr: address 84 * @seq: sequence number 85 * @flags: fence related flags 86 * 87 * Add a DMA fence packet to the ring to write 88 * the fence seq number and DMA trap packet to generate 89 * an interrupt if needed (VI). 90 */ 91static void si_dma_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, 92 unsigned flags) 93{ 94 95 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT; 96 /* write the fence */ 97 amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0, 0, 0)); 98 amdgpu_ring_write(ring, addr & 0xfffffffc); 99 amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xff)); 100 amdgpu_ring_write(ring, seq); 101 /* optionally write high bits as well */ 102 if (write64bit) { 103 addr += 4; 104 amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0, 0, 0)); 105 amdgpu_ring_write(ring, addr & 0xfffffffc); 106 amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xff)); 107 amdgpu_ring_write(ring, upper_32_bits(seq)); 108 } 109 /* generate an interrupt */ 110 amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_TRAP, 0, 0, 0, 0)); 111} 112 113static void si_dma_stop(struct amdgpu_device *adev) 114{ 115 u32 rb_cntl; 116 unsigned i; 117 118 amdgpu_sdma_unset_buffer_funcs_helper(adev); 119 120 for (i = 0; i < adev->sdma.num_instances; i++) { 121 /* dma0 */ 122 rb_cntl = RREG32(DMA_RB_CNTL + sdma_offsets[i]); 123 rb_cntl &= ~DMA_RB_ENABLE; 124 WREG32(DMA_RB_CNTL + sdma_offsets[i], rb_cntl); 125 } 126} 127 128static int si_dma_start(struct amdgpu_device *adev) 129{ 130 struct amdgpu_ring *ring; 131 u32 rb_cntl, dma_cntl, ib_cntl, rb_bufsz; 132 int i, r; 133 uint64_t rptr_addr; 134 135 for (i = 0; i < adev->sdma.num_instances; i++) { 136 ring = &adev->sdma.instance[i].ring; 137 138 WREG32(DMA_SEM_INCOMPLETE_TIMER_CNTL + sdma_offsets[i], 0); 139 WREG32(DMA_SEM_WAIT_FAIL_TIMER_CNTL + sdma_offsets[i], 0); 140 141 /* Set ring buffer size in dwords */ 142 rb_bufsz = order_base_2(ring->ring_size / 4); 143 rb_cntl = rb_bufsz << 1; 144#ifdef __BIG_ENDIAN 145 rb_cntl |= DMA_RB_SWAP_ENABLE | DMA_RPTR_WRITEBACK_SWAP_ENABLE; 146#endif 147 WREG32(DMA_RB_CNTL + sdma_offsets[i], rb_cntl); 148 149 /* Initialize the ring buffer's read and write pointers */ 150 WREG32(DMA_RB_RPTR + sdma_offsets[i], 0); 151 WREG32(DMA_RB_WPTR + sdma_offsets[i], 0); 152 153 rptr_addr = ring->rptr_gpu_addr; 154 155 WREG32(DMA_RB_RPTR_ADDR_LO + sdma_offsets[i], lower_32_bits(rptr_addr)); 156 WREG32(DMA_RB_RPTR_ADDR_HI + sdma_offsets[i], upper_32_bits(rptr_addr) & 0xFF); 157 158 rb_cntl |= DMA_RPTR_WRITEBACK_ENABLE; 159 160 WREG32(DMA_RB_BASE + sdma_offsets[i], ring->gpu_addr >> 8); 161 162 /* enable DMA IBs */ 163 ib_cntl = DMA_IB_ENABLE | CMD_VMID_FORCE; 164#ifdef __BIG_ENDIAN 165 ib_cntl |= DMA_IB_SWAP_ENABLE; 166#endif 167 WREG32(DMA_IB_CNTL + sdma_offsets[i], ib_cntl); 168 169 dma_cntl = RREG32(DMA_CNTL + sdma_offsets[i]); 170 dma_cntl &= ~CTXEMPTY_INT_ENABLE; 171 WREG32(DMA_CNTL + sdma_offsets[i], dma_cntl); 172 173 ring->wptr = 0; 174 WREG32(DMA_RB_WPTR + sdma_offsets[i], ring->wptr << 2); 175 WREG32(DMA_RB_CNTL + sdma_offsets[i], rb_cntl | DMA_RB_ENABLE); 176 177 r = amdgpu_ring_test_helper(ring); 178 if (r) 179 return r; 180 181 if (adev->mman.buffer_funcs_ring == ring) 182 amdgpu_ttm_set_buffer_funcs_status(adev, true); 183 } 184 185 return 0; 186} 187 188/** 189 * si_dma_ring_test_ring - simple async dma engine test 190 * 191 * @ring: amdgpu_ring structure holding ring information 192 * 193 * Test the DMA engine by writing using it to write an 194 * value to memory. (VI). 195 * Returns 0 for success, error for failure. 196 */ 197static int si_dma_ring_test_ring(struct amdgpu_ring *ring) 198{ 199 struct amdgpu_device *adev = ring->adev; 200 unsigned i; 201 unsigned index; 202 int r; 203 u32 tmp; 204 u64 gpu_addr; 205 206 r = amdgpu_device_wb_get(adev, &index); 207 if (r) 208 return r; 209 210 gpu_addr = adev->wb.gpu_addr + (index * 4); 211 tmp = 0xCAFEDEAD; 212 adev->wb.wb[index] = cpu_to_le32(tmp); 213 214 r = amdgpu_ring_alloc(ring, 4); 215 if (r) 216 goto error_free_wb; 217 218 amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, 1)); 219 amdgpu_ring_write(ring, lower_32_bits(gpu_addr)); 220 amdgpu_ring_write(ring, upper_32_bits(gpu_addr) & 0xff); 221 amdgpu_ring_write(ring, 0xDEADBEEF); 222 amdgpu_ring_commit(ring); 223 224 for (i = 0; i < adev->usec_timeout; i++) { 225 tmp = le32_to_cpu(adev->wb.wb[index]); 226 if (tmp == 0xDEADBEEF) 227 break; 228 udelay(1); 229 } 230 231 if (i >= adev->usec_timeout) 232 r = -ETIMEDOUT; 233 234error_free_wb: 235 amdgpu_device_wb_free(adev, index); 236 return r; 237} 238 239/** 240 * si_dma_ring_test_ib - test an IB on the DMA engine 241 * 242 * @ring: amdgpu_ring structure holding ring information 243 * @timeout: timeout value in jiffies, or MAX_SCHEDULE_TIMEOUT 244 * 245 * Test a simple IB in the DMA ring (VI). 246 * Returns 0 on success, error on failure. 247 */ 248static int si_dma_ring_test_ib(struct amdgpu_ring *ring, long timeout) 249{ 250 struct amdgpu_device *adev = ring->adev; 251 struct amdgpu_ib ib; 252 struct dma_fence *f = NULL; 253 unsigned index; 254 u32 tmp = 0; 255 u64 gpu_addr; 256 long r; 257 258 r = amdgpu_device_wb_get(adev, &index); 259 if (r) 260 return r; 261 262 gpu_addr = adev->wb.gpu_addr + (index * 4); 263 tmp = 0xCAFEDEAD; 264 adev->wb.wb[index] = cpu_to_le32(tmp); 265 memset(&ib, 0, sizeof(ib)); 266 r = amdgpu_ib_get(adev, NULL, 256, 267 AMDGPU_IB_POOL_DIRECT, &ib); 268 if (r) 269 goto err0; 270 271 ib.ptr[0] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, 1); 272 ib.ptr[1] = lower_32_bits(gpu_addr); 273 ib.ptr[2] = upper_32_bits(gpu_addr) & 0xff; 274 ib.ptr[3] = 0xDEADBEEF; 275 ib.length_dw = 4; 276 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 277 if (r) 278 goto err1; 279 280 r = dma_fence_wait_timeout(f, false, timeout); 281 if (r == 0) { 282 r = -ETIMEDOUT; 283 goto err1; 284 } else if (r < 0) { 285 goto err1; 286 } 287 tmp = le32_to_cpu(adev->wb.wb[index]); 288 if (tmp == 0xDEADBEEF) 289 r = 0; 290 else 291 r = -EINVAL; 292 293err1: 294 amdgpu_ib_free(adev, &ib, NULL); 295 dma_fence_put(f); 296err0: 297 amdgpu_device_wb_free(adev, index); 298 return r; 299} 300 301/** 302 * si_dma_vm_copy_pte - update PTEs by copying them from the GART 303 * 304 * @ib: indirect buffer to fill with commands 305 * @pe: addr of the page entry 306 * @src: src addr to copy from 307 * @count: number of page entries to update 308 * 309 * Update PTEs by copying them from the GART using DMA (SI). 310 */ 311static void si_dma_vm_copy_pte(struct amdgpu_ib *ib, 312 uint64_t pe, uint64_t src, 313 unsigned count) 314{ 315 unsigned bytes = count * 8; 316 317 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_COPY, 318 1, 0, 0, bytes); 319 ib->ptr[ib->length_dw++] = lower_32_bits(pe); 320 ib->ptr[ib->length_dw++] = lower_32_bits(src); 321 ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff; 322 ib->ptr[ib->length_dw++] = upper_32_bits(src) & 0xff; 323} 324 325/** 326 * si_dma_vm_write_pte - update PTEs by writing them manually 327 * 328 * @ib: indirect buffer to fill with commands 329 * @pe: addr of the page entry 330 * @value: dst addr to write into pe 331 * @count: number of page entries to update 332 * @incr: increase next addr by incr bytes 333 * 334 * Update PTEs by writing them manually using DMA (SI). 335 */ 336static void si_dma_vm_write_pte(struct amdgpu_ib *ib, uint64_t pe, 337 uint64_t value, unsigned count, 338 uint32_t incr) 339{ 340 unsigned ndw = count * 2; 341 342 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, ndw); 343 ib->ptr[ib->length_dw++] = lower_32_bits(pe); 344 ib->ptr[ib->length_dw++] = upper_32_bits(pe); 345 for (; ndw > 0; ndw -= 2) { 346 ib->ptr[ib->length_dw++] = lower_32_bits(value); 347 ib->ptr[ib->length_dw++] = upper_32_bits(value); 348 value += incr; 349 } 350} 351 352/** 353 * si_dma_vm_set_pte_pde - update the page tables using sDMA 354 * 355 * @ib: indirect buffer to fill with commands 356 * @pe: addr of the page entry 357 * @addr: dst addr to write into pe 358 * @count: number of page entries to update 359 * @incr: increase next addr by incr bytes 360 * @flags: access flags 361 * 362 * Update the page tables using sDMA (CIK). 363 */ 364static void si_dma_vm_set_pte_pde(struct amdgpu_ib *ib, 365 uint64_t pe, 366 uint64_t addr, unsigned count, 367 uint32_t incr, uint64_t flags) 368{ 369 uint64_t value; 370 unsigned ndw; 371 372 while (count) { 373 ndw = count * 2; 374 if (ndw > 0xFFFFE) 375 ndw = 0xFFFFE; 376 377 if (flags & AMDGPU_PTE_VALID) 378 value = addr; 379 else 380 value = 0; 381 382 /* for physically contiguous pages (vram) */ 383 ib->ptr[ib->length_dw++] = DMA_PTE_PDE_PACKET(ndw); 384 ib->ptr[ib->length_dw++] = pe; /* dst addr */ 385 ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff; 386 ib->ptr[ib->length_dw++] = lower_32_bits(flags); /* mask */ 387 ib->ptr[ib->length_dw++] = upper_32_bits(flags); 388 ib->ptr[ib->length_dw++] = value; /* value */ 389 ib->ptr[ib->length_dw++] = upper_32_bits(value); 390 ib->ptr[ib->length_dw++] = incr; /* increment size */ 391 ib->ptr[ib->length_dw++] = 0; 392 pe += ndw * 4; 393 addr += (ndw / 2) * incr; 394 count -= ndw / 2; 395 } 396} 397 398/** 399 * si_dma_ring_pad_ib - pad the IB to the required number of dw 400 * 401 * @ring: amdgpu_ring pointer 402 * @ib: indirect buffer to fill with padding 403 * 404 */ 405static void si_dma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) 406{ 407 while (ib->length_dw & 0x7) 408 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0); 409} 410 411/** 412 * si_dma_ring_emit_pipeline_sync - sync the pipeline 413 * 414 * @ring: amdgpu_ring pointer 415 * 416 * Make sure all previous operations are completed (CIK). 417 */ 418static void si_dma_ring_emit_pipeline_sync(struct amdgpu_ring *ring) 419{ 420 uint32_t seq = ring->fence_drv.sync_seq; 421 uint64_t addr = ring->fence_drv.gpu_addr; 422 423 /* wait for idle */ 424 amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_POLL_REG_MEM, 0, 0, 0, 0) | 425 (1 << 27)); /* Poll memory */ 426 amdgpu_ring_write(ring, lower_32_bits(addr)); 427 amdgpu_ring_write(ring, (0xff << 16) | upper_32_bits(addr)); /* retry, addr_hi */ 428 amdgpu_ring_write(ring, 0xffffffff); /* mask */ 429 amdgpu_ring_write(ring, seq); /* value */ 430 amdgpu_ring_write(ring, (3 << 28) | 0x20); /* func(equal) | poll interval */ 431} 432 433/** 434 * si_dma_ring_emit_vm_flush - cik vm flush using sDMA 435 * 436 * @ring: amdgpu_ring pointer 437 * @vmid: vmid number to use 438 * @pd_addr: address 439 * 440 * Update the page table base and flush the VM TLB 441 * using sDMA (VI). 442 */ 443static void si_dma_ring_emit_vm_flush(struct amdgpu_ring *ring, 444 unsigned vmid, uint64_t pd_addr) 445{ 446 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 447 448 /* wait for invalidate to complete */ 449 amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_POLL_REG_MEM, 0, 0, 0, 0)); 450 amdgpu_ring_write(ring, VM_INVALIDATE_REQUEST); 451 amdgpu_ring_write(ring, 0xff << 16); /* retry */ 452 amdgpu_ring_write(ring, 1 << vmid); /* mask */ 453 amdgpu_ring_write(ring, 0); /* value */ 454 amdgpu_ring_write(ring, (0 << 28) | 0x20); /* func(always) | poll interval */ 455} 456 457static void si_dma_ring_emit_wreg(struct amdgpu_ring *ring, 458 uint32_t reg, uint32_t val) 459{ 460 amdgpu_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0)); 461 amdgpu_ring_write(ring, (0xf << 16) | reg); 462 amdgpu_ring_write(ring, val); 463} 464 465static int si_dma_early_init(void *handle) 466{ 467 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 468 469 adev->sdma.num_instances = 2; 470 471 si_dma_set_ring_funcs(adev); 472 si_dma_set_buffer_funcs(adev); 473 si_dma_set_vm_pte_funcs(adev); 474 si_dma_set_irq_funcs(adev); 475 476 return 0; 477} 478 479static int si_dma_sw_init(void *handle) 480{ 481 struct amdgpu_ring *ring; 482 int r, i; 483 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 484 485 /* DMA0 trap event */ 486 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 224, 487 &adev->sdma.trap_irq); 488 if (r) 489 return r; 490 491 /* DMA1 trap event */ 492 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 244, 493 &adev->sdma.trap_irq); 494 if (r) 495 return r; 496 497 for (i = 0; i < adev->sdma.num_instances; i++) { 498 ring = &adev->sdma.instance[i].ring; 499 ring->ring_obj = NULL; 500 ring->use_doorbell = false; 501 sprintf(ring->name, "sdma%d", i); 502 r = amdgpu_ring_init(adev, ring, 1024, 503 &adev->sdma.trap_irq, 504 (i == 0) ? AMDGPU_SDMA_IRQ_INSTANCE0 : 505 AMDGPU_SDMA_IRQ_INSTANCE1, 506 AMDGPU_RING_PRIO_DEFAULT, NULL); 507 if (r) 508 return r; 509 } 510 511 return r; 512} 513 514static int si_dma_sw_fini(void *handle) 515{ 516 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 517 int i; 518 519 for (i = 0; i < adev->sdma.num_instances; i++) 520 amdgpu_ring_fini(&adev->sdma.instance[i].ring); 521 522 return 0; 523} 524 525static int si_dma_hw_init(void *handle) 526{ 527 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 528 529 return si_dma_start(adev); 530} 531 532static int si_dma_hw_fini(void *handle) 533{ 534 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 535 536 si_dma_stop(adev); 537 538 return 0; 539} 540 541static int si_dma_suspend(void *handle) 542{ 543 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 544 545 return si_dma_hw_fini(adev); 546} 547 548static int si_dma_resume(void *handle) 549{ 550 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 551 552 return si_dma_hw_init(adev); 553} 554 555static bool si_dma_is_idle(void *handle) 556{ 557 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 558 u32 tmp = RREG32(SRBM_STATUS2); 559 560 if (tmp & (DMA_BUSY_MASK | DMA1_BUSY_MASK)) 561 return false; 562 563 return true; 564} 565 566static int si_dma_wait_for_idle(void *handle) 567{ 568 unsigned i; 569 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 570 571 for (i = 0; i < adev->usec_timeout; i++) { 572 if (si_dma_is_idle(handle)) 573 return 0; 574 udelay(1); 575 } 576 return -ETIMEDOUT; 577} 578 579static int si_dma_soft_reset(void *handle) 580{ 581 DRM_INFO("si_dma_soft_reset --- not implemented !!!!!!!\n"); 582 return 0; 583} 584 585static int si_dma_set_trap_irq_state(struct amdgpu_device *adev, 586 struct amdgpu_irq_src *src, 587 unsigned type, 588 enum amdgpu_interrupt_state state) 589{ 590 u32 sdma_cntl; 591 592 switch (type) { 593 case AMDGPU_SDMA_IRQ_INSTANCE0: 594 switch (state) { 595 case AMDGPU_IRQ_STATE_DISABLE: 596 sdma_cntl = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET); 597 sdma_cntl &= ~TRAP_ENABLE; 598 WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, sdma_cntl); 599 break; 600 case AMDGPU_IRQ_STATE_ENABLE: 601 sdma_cntl = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET); 602 sdma_cntl |= TRAP_ENABLE; 603 WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, sdma_cntl); 604 break; 605 default: 606 break; 607 } 608 break; 609 case AMDGPU_SDMA_IRQ_INSTANCE1: 610 switch (state) { 611 case AMDGPU_IRQ_STATE_DISABLE: 612 sdma_cntl = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET); 613 sdma_cntl &= ~TRAP_ENABLE; 614 WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, sdma_cntl); 615 break; 616 case AMDGPU_IRQ_STATE_ENABLE: 617 sdma_cntl = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET); 618 sdma_cntl |= TRAP_ENABLE; 619 WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, sdma_cntl); 620 break; 621 default: 622 break; 623 } 624 break; 625 default: 626 break; 627 } 628 return 0; 629} 630 631static int si_dma_process_trap_irq(struct amdgpu_device *adev, 632 struct amdgpu_irq_src *source, 633 struct amdgpu_iv_entry *entry) 634{ 635 if (entry->src_id == 224) 636 amdgpu_fence_process(&adev->sdma.instance[0].ring); 637 else 638 amdgpu_fence_process(&adev->sdma.instance[1].ring); 639 return 0; 640} 641 642static int si_dma_set_clockgating_state(void *handle, 643 enum amd_clockgating_state state) 644{ 645 u32 orig, data, offset; 646 int i; 647 bool enable; 648 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 649 650 enable = (state == AMD_CG_STATE_GATE); 651 652 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_MGCG)) { 653 for (i = 0; i < adev->sdma.num_instances; i++) { 654 if (i == 0) 655 offset = DMA0_REGISTER_OFFSET; 656 else 657 offset = DMA1_REGISTER_OFFSET; 658 orig = data = RREG32(DMA_POWER_CNTL + offset); 659 data &= ~MEM_POWER_OVERRIDE; 660 if (data != orig) 661 WREG32(DMA_POWER_CNTL + offset, data); 662 WREG32(DMA_CLK_CTRL + offset, 0x00000100); 663 } 664 } else { 665 for (i = 0; i < adev->sdma.num_instances; i++) { 666 if (i == 0) 667 offset = DMA0_REGISTER_OFFSET; 668 else 669 offset = DMA1_REGISTER_OFFSET; 670 orig = data = RREG32(DMA_POWER_CNTL + offset); 671 data |= MEM_POWER_OVERRIDE; 672 if (data != orig) 673 WREG32(DMA_POWER_CNTL + offset, data); 674 675 orig = data = RREG32(DMA_CLK_CTRL + offset); 676 data = 0xff000000; 677 if (data != orig) 678 WREG32(DMA_CLK_CTRL + offset, data); 679 } 680 } 681 682 return 0; 683} 684 685static int si_dma_set_powergating_state(void *handle, 686 enum amd_powergating_state state) 687{ 688 u32 tmp; 689 690 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 691 692 WREG32(DMA_PGFSM_WRITE, 0x00002000); 693 WREG32(DMA_PGFSM_CONFIG, 0x100010ff); 694 695 for (tmp = 0; tmp < 5; tmp++) 696 WREG32(DMA_PGFSM_WRITE, 0); 697 698 return 0; 699} 700 701static const struct amd_ip_funcs si_dma_ip_funcs = { 702 .name = "si_dma", 703 .early_init = si_dma_early_init, 704 .late_init = NULL, 705 .sw_init = si_dma_sw_init, 706 .sw_fini = si_dma_sw_fini, 707 .hw_init = si_dma_hw_init, 708 .hw_fini = si_dma_hw_fini, 709 .suspend = si_dma_suspend, 710 .resume = si_dma_resume, 711 .is_idle = si_dma_is_idle, 712 .wait_for_idle = si_dma_wait_for_idle, 713 .soft_reset = si_dma_soft_reset, 714 .set_clockgating_state = si_dma_set_clockgating_state, 715 .set_powergating_state = si_dma_set_powergating_state, 716}; 717 718static const struct amdgpu_ring_funcs si_dma_ring_funcs = { 719 .type = AMDGPU_RING_TYPE_SDMA, 720 .align_mask = 0xf, 721 .nop = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0), 722 .support_64bit_ptrs = false, 723 .get_rptr = si_dma_ring_get_rptr, 724 .get_wptr = si_dma_ring_get_wptr, 725 .set_wptr = si_dma_ring_set_wptr, 726 .emit_frame_size = 727 3 + 3 + /* hdp flush / invalidate */ 728 6 + /* si_dma_ring_emit_pipeline_sync */ 729 SI_FLUSH_GPU_TLB_NUM_WREG * 3 + 6 + /* si_dma_ring_emit_vm_flush */ 730 9 + 9 + 9, /* si_dma_ring_emit_fence x3 for user fence, vm fence */ 731 .emit_ib_size = 7 + 3, /* si_dma_ring_emit_ib */ 732 .emit_ib = si_dma_ring_emit_ib, 733 .emit_fence = si_dma_ring_emit_fence, 734 .emit_pipeline_sync = si_dma_ring_emit_pipeline_sync, 735 .emit_vm_flush = si_dma_ring_emit_vm_flush, 736 .test_ring = si_dma_ring_test_ring, 737 .test_ib = si_dma_ring_test_ib, 738 .insert_nop = amdgpu_ring_insert_nop, 739 .pad_ib = si_dma_ring_pad_ib, 740 .emit_wreg = si_dma_ring_emit_wreg, 741}; 742 743static void si_dma_set_ring_funcs(struct amdgpu_device *adev) 744{ 745 int i; 746 747 for (i = 0; i < adev->sdma.num_instances; i++) 748 adev->sdma.instance[i].ring.funcs = &si_dma_ring_funcs; 749} 750 751static const struct amdgpu_irq_src_funcs si_dma_trap_irq_funcs = { 752 .set = si_dma_set_trap_irq_state, 753 .process = si_dma_process_trap_irq, 754}; 755 756static void si_dma_set_irq_funcs(struct amdgpu_device *adev) 757{ 758 adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; 759 adev->sdma.trap_irq.funcs = &si_dma_trap_irq_funcs; 760} 761 762/** 763 * si_dma_emit_copy_buffer - copy buffer using the sDMA engine 764 * 765 * @ib: indirect buffer to copy to 766 * @src_offset: src GPU address 767 * @dst_offset: dst GPU address 768 * @byte_count: number of bytes to xfer 769 * @tmz: is this a secure operation 770 * 771 * Copy GPU buffers using the DMA engine (VI). 772 * Used by the amdgpu ttm implementation to move pages if 773 * registered as the asic copy callback. 774 */ 775static void si_dma_emit_copy_buffer(struct amdgpu_ib *ib, 776 uint64_t src_offset, 777 uint64_t dst_offset, 778 uint32_t byte_count, 779 bool tmz) 780{ 781 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_COPY, 782 1, 0, 0, byte_count); 783 ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset); 784 ib->ptr[ib->length_dw++] = lower_32_bits(src_offset); 785 ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset) & 0xff; 786 ib->ptr[ib->length_dw++] = upper_32_bits(src_offset) & 0xff; 787} 788 789/** 790 * si_dma_emit_fill_buffer - fill buffer using the sDMA engine 791 * 792 * @ib: indirect buffer to copy to 793 * @src_data: value to write to buffer 794 * @dst_offset: dst GPU address 795 * @byte_count: number of bytes to xfer 796 * 797 * Fill GPU buffers using the DMA engine (VI). 798 */ 799static void si_dma_emit_fill_buffer(struct amdgpu_ib *ib, 800 uint32_t src_data, 801 uint64_t dst_offset, 802 uint32_t byte_count) 803{ 804 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_CONSTANT_FILL, 805 0, 0, 0, byte_count / 4); 806 ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset); 807 ib->ptr[ib->length_dw++] = src_data; 808 ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset) << 16; 809} 810 811 812static const struct amdgpu_buffer_funcs si_dma_buffer_funcs = { 813 .copy_max_bytes = 0xffff8, 814 .copy_num_dw = 5, 815 .emit_copy_buffer = si_dma_emit_copy_buffer, 816 817 .fill_max_bytes = 0xffff8, 818 .fill_num_dw = 4, 819 .emit_fill_buffer = si_dma_emit_fill_buffer, 820}; 821 822static void si_dma_set_buffer_funcs(struct amdgpu_device *adev) 823{ 824 adev->mman.buffer_funcs = &si_dma_buffer_funcs; 825 adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring; 826} 827 828static const struct amdgpu_vm_pte_funcs si_dma_vm_pte_funcs = { 829 .copy_pte_num_dw = 5, 830 .copy_pte = si_dma_vm_copy_pte, 831 832 .write_pte = si_dma_vm_write_pte, 833 .set_pte_pde = si_dma_vm_set_pte_pde, 834}; 835 836static void si_dma_set_vm_pte_funcs(struct amdgpu_device *adev) 837{ 838 unsigned i; 839 840 adev->vm_manager.vm_pte_funcs = &si_dma_vm_pte_funcs; 841 for (i = 0; i < adev->sdma.num_instances; i++) { 842 adev->vm_manager.vm_pte_scheds[i] = 843 &adev->sdma.instance[i].ring.sched; 844 } 845 adev->vm_manager.vm_pte_num_scheds = adev->sdma.num_instances; 846} 847 848const struct amdgpu_ip_block_version si_dma_ip_block = 849{ 850 .type = AMD_IP_BLOCK_TYPE_SDMA, 851 .major = 1, 852 .minor = 0, 853 .rev = 0, 854 .funcs = &si_dma_ip_funcs, 855}; 856