| 1 | |
| 2 | * Copyright 2014 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 |
|
| 24 | #include <linux/pci.h>
|
| 25 |
|
| 26 | #include "amdgpu.h"
|
| 27 | #include "amdgpu_ih.h"
|
| 28 | #include "vid.h"
|
| 29 |
|
| 30 | #include "oss/oss_3_0_d.h"
|
| 31 | #include "oss/oss_3_0_sh_mask.h"
|
| 32 |
|
| 33 | #include "bif/bif_5_1_d.h"
|
| 34 | #include "bif/bif_5_1_sh_mask.h"
|
| 35 |
|
| 36 | |
| 37 | * Interrupts
|
| 38 | * Starting with r6xx, interrupts are handled via a ring buffer.
|
| 39 | * Ring buffers are areas of GPU accessible memory that the GPU
|
| 40 | * writes interrupt vectors into and the host reads vectors out of.
|
| 41 | * There is a rptr (read pointer) that determines where the
|
| 42 | * host is currently reading, and a wptr (write pointer)
|
| 43 | * which determines where the GPU has written. When the
|
| 44 | * pointers are equal, the ring is idle. When the GPU
|
| 45 | * writes vectors to the ring buffer, it increments the
|
| 46 | * wptr. When there is an interrupt, the host then starts
|
| 47 | * fetching commands and processing them until the pointers are
|
| 48 | * equal again at which point it updates the rptr.
|
| 49 |
|
| 50 |
|
| 51 | static void tonga_ih_set_interrupt_funcs(struct amdgpu_device *adev);
|
| 52 |
|
| 53 | |
| 54 | * tonga_ih_enable_interrupts - Enable the interrupt ring buffer
|
| 55 | *
|
| 56 | * @adev: amdgpu_device pointer
|
| 57 | *
|
| 58 | * Enable the interrupt ring buffer (VI).
|
| 59 |
|
| 60 | static void tonga_ih_enable_interrupts(struct amdgpu_device *adev)
|
| 61 | {
|
| 62 | u32 ih_rb_cntl = RREG32(mmIH_RB_CNTL);
|
| 63 |
|
| 64 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 1);
|
| 65 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, ENABLE_INTR, 1);
|
| 66 | WREG32(mmIH_RB_CNTL, ih_rb_cntl);
|
| 67 | adev->irq.ih.enabled = true;
|
| 68 | }
|
| 69 |
|
| 70 | |
| 71 | * tonga_ih_disable_interrupts - Disable the interrupt ring buffer
|
| 72 | *
|
| 73 | * @adev: amdgpu_device pointer
|
| 74 | *
|
| 75 | * Disable the interrupt ring buffer (VI).
|
| 76 |
|
| 77 | static void tonga_ih_disable_interrupts(struct amdgpu_device *adev)
|
| 78 | {
|
| 79 | u32 ih_rb_cntl = RREG32(mmIH_RB_CNTL);
|
| 80 |
|
| 81 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 0);
|
| 82 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, ENABLE_INTR, 0);
|
| 83 | WREG32(mmIH_RB_CNTL, ih_rb_cntl);
|
| 84 |
|
| 85 | WREG32(mmIH_RB_RPTR, 0);
|
| 86 | WREG32(mmIH_RB_WPTR, 0);
|
| 87 | adev->irq.ih.enabled = false;
|
| 88 | adev->irq.ih.rptr = 0;
|
| 89 | }
|
| 90 |
|
| 91 | |
| 92 | * tonga_ih_irq_init - init and enable the interrupt ring
|
| 93 | *
|
| 94 | * @adev: amdgpu_device pointer
|
| 95 | *
|
| 96 | * Allocate a ring buffer for the interrupt controller,
|
| 97 | * enable the RLC, disable interrupts, enable the IH
|
| 98 | * ring buffer and enable it (VI).
|
| 99 | * Called at device load and reume.
|
| 100 | * Returns 0 for success, errors for failure.
|
| 101 |
|
| 102 | static int tonga_ih_irq_init(struct amdgpu_device *adev)
|
| 103 | {
|
| 104 | u32 interrupt_cntl, ih_rb_cntl, ih_doorbell_rtpr;
|
| 105 | struct amdgpu_ih_ring *ih = &adev->irq.ih;
|
| 106 | int rb_bufsz;
|
| 107 |
|
| 108 |
|
| 109 | tonga_ih_disable_interrupts(adev);
|
| 110 |
|
| 111 |
|
| 112 | WREG32(mmINTERRUPT_CNTL2, adev->dummy_page_addr >> 8);
|
| 113 | interrupt_cntl = RREG32(mmINTERRUPT_CNTL);
|
| 114 | |
| 115 | * INTERRUPT_CNTL__IH_DUMMY_RD_OVERRIDE_MASK=1 - dummy read controlled by IH_DUMMY_RD_EN
|
| 116 |
|
| 117 | interrupt_cntl = REG_SET_FIELD(interrupt_cntl, INTERRUPT_CNTL, IH_DUMMY_RD_OVERRIDE, 0);
|
| 118 |
|
| 119 | interrupt_cntl = REG_SET_FIELD(interrupt_cntl, INTERRUPT_CNTL, IH_REQ_NONSNOOP_EN, 0);
|
| 120 | WREG32(mmINTERRUPT_CNTL, interrupt_cntl);
|
| 121 |
|
| 122 |
|
| 123 | WREG32(mmIH_RB_BASE, ih->gpu_addr >> 8);
|
| 124 |
|
| 125 | rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4);
|
| 126 | ih_rb_cntl = REG_SET_FIELD(0, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
|
| 127 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_SIZE, rb_bufsz);
|
| 128 |
|
| 129 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1);
|
| 130 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_VMID, 0);
|
| 131 |
|
| 132 | if (adev->irq.msi_enabled)
|
| 133 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RPTR_REARM, 1);
|
| 134 |
|
| 135 | WREG32(mmIH_RB_CNTL, ih_rb_cntl);
|
| 136 |
|
| 137 |
|
| 138 | WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr));
|
| 139 | WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF);
|
| 140 |
|
| 141 |
|
| 142 | WREG32(mmIH_RB_RPTR, 0);
|
| 143 | WREG32(mmIH_RB_WPTR, 0);
|
| 144 |
|
| 145 | ih_doorbell_rtpr = RREG32(mmIH_DOORBELL_RPTR);
|
| 146 | if (adev->irq.ih.use_doorbell) {
|
| 147 | ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR,
|
| 148 | OFFSET, adev->irq.ih.doorbell_index);
|
| 149 | ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR,
|
| 150 | ENABLE, 1);
|
| 151 | } else {
|
| 152 | ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR,
|
| 153 | ENABLE, 0);
|
| 154 | }
|
| 155 | WREG32(mmIH_DOORBELL_RPTR, ih_doorbell_rtpr);
|
| 156 |
|
| 157 | pci_set_master(adev->pdev);
|
| 158 |
|
| 159 |
|
| 160 | tonga_ih_enable_interrupts(adev);
|
| 161 |
|
| 162 | if (adev->irq.ih_soft.ring_size)
|
| 163 | adev->irq.ih_soft.enabled = true;
|
| 164 |
|
| 165 | return 0;
|
| 166 | }
|
| 167 |
|
| 168 | |
| 169 | * tonga_ih_irq_disable - disable interrupts
|
| 170 | *
|
| 171 | * @adev: amdgpu_device pointer
|
| 172 | *
|
| 173 | * Disable interrupts on the hw (VI).
|
| 174 |
|
| 175 | static void tonga_ih_irq_disable(struct amdgpu_device *adev)
|
| 176 | {
|
| 177 | tonga_ih_disable_interrupts(adev);
|
| 178 |
|
| 179 |
|
| 180 | mdelay(1);
|
| 181 | }
|
| 182 |
|
| 183 | |
| 184 | * tonga_ih_get_wptr - get the IH ring buffer wptr
|
| 185 | *
|
| 186 | * @adev: amdgpu_device pointer
|
| 187 | * @ih: IH ring buffer to fetch wptr
|
| 188 | *
|
| 189 | * Get the IH ring buffer wptr from either the register
|
| 190 | * or the writeback memory buffer (VI). Also check for
|
| 191 | * ring buffer overflow and deal with it.
|
| 192 | * Used by cz_irq_process(VI).
|
| 193 | * Returns the value of the wptr.
|
| 194 |
|
| 195 | static u32 tonga_ih_get_wptr(struct amdgpu_device *adev,
|
| 196 | struct amdgpu_ih_ring *ih)
|
| 197 | {
|
| 198 | u32 wptr, tmp;
|
| 199 |
|
| 200 | wptr = le32_to_cpu(*ih->wptr_cpu);
|
| 201 |
|
| 202 | if (ih == &adev->irq.ih_soft)
|
| 203 | goto out;
|
| 204 |
|
| 205 | if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW))
|
| 206 | goto out;
|
| 207 |
|
| 208 |
|
| 209 | wptr = RREG32(mmIH_RB_WPTR);
|
| 210 |
|
| 211 | if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW))
|
| 212 | goto out;
|
| 213 |
|
| 214 | wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0);
|
| 215 |
|
| 216 | |
| 217 | * from the last not overwritten vector (wptr + 16). Hopefully
|
| 218 | * this should allow us to catchup.
|
| 219 |
|
| 220 |
|
| 221 | dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n",
|
| 222 | wptr, ih->rptr, (wptr + 16) & ih->ptr_mask);
|
| 223 | ih->rptr = (wptr + 16) & ih->ptr_mask;
|
| 224 | tmp = RREG32(mmIH_RB_CNTL);
|
| 225 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
|
| 226 | WREG32(mmIH_RB_CNTL, tmp);
|
| 227 |
|
| 228 | |
| 229 | * can be detected.
|
| 230 |
|
| 231 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 0);
|
| 232 | WREG32(mmIH_RB_CNTL, tmp);
|
| 233 |
|
| 234 | out:
|
| 235 | return (wptr & ih->ptr_mask);
|
| 236 | }
|
| 237 |
|
| 238 | |
| 239 | * tonga_ih_decode_iv - decode an interrupt vector
|
| 240 | *
|
| 241 | * @adev: amdgpu_device pointer
|
| 242 | * @ih: IH ring buffer to decode
|
| 243 | * @entry: IV entry to place decoded information into
|
| 244 | *
|
| 245 | * Decodes the interrupt vector at the current rptr
|
| 246 | * position and also advance the position.
|
| 247 |
|
| 248 | static void tonga_ih_decode_iv(struct amdgpu_device *adev,
|
| 249 | struct amdgpu_ih_ring *ih,
|
| 250 | struct amdgpu_iv_entry *entry)
|
| 251 | {
|
| 252 |
|
| 253 | u32 ring_index = ih->rptr >> 2;
|
| 254 | uint32_t dw[4];
|
| 255 |
|
| 256 | dw[0] = le32_to_cpu(ih->ring[ring_index + 0]);
|
| 257 | dw[1] = le32_to_cpu(ih->ring[ring_index + 1]);
|
| 258 | dw[2] = le32_to_cpu(ih->ring[ring_index + 2]);
|
| 259 | dw[3] = le32_to_cpu(ih->ring[ring_index + 3]);
|
| 260 |
|
| 261 | entry->client_id = AMDGPU_IRQ_CLIENTID_LEGACY;
|
| 262 | entry->src_id = dw[0] & 0xff;
|
| 263 | entry->src_data[0] = dw[1] & 0xfffffff;
|
| 264 | entry->ring_id = dw[2] & 0xff;
|
| 265 | entry->vmid = (dw[2] >> 8) & 0xff;
|
| 266 | entry->pasid = (dw[2] >> 16) & 0xffff;
|
| 267 |
|
| 268 |
|
| 269 | ih->rptr += 16;
|
| 270 | }
|
| 271 |
|
| 272 | |
| 273 | * tonga_ih_set_rptr - set the IH ring buffer rptr
|
| 274 | *
|
| 275 | * @adev: amdgpu_device pointer
|
| 276 | * @ih: IH ring buffer to set rptr
|
| 277 | *
|
| 278 | * Set the IH ring buffer rptr.
|
| 279 |
|
| 280 | static void tonga_ih_set_rptr(struct amdgpu_device *adev,
|
| 281 | struct amdgpu_ih_ring *ih)
|
| 282 | {
|
| 283 | if (ih->use_doorbell) {
|
| 284 |
|
| 285 | *ih->rptr_cpu = ih->rptr;
|
| 286 | WDOORBELL32(ih->doorbell_index, ih->rptr);
|
| 287 | } else {
|
| 288 | WREG32(mmIH_RB_RPTR, ih->rptr);
|
| 289 | }
|
| 290 | }
|
| 291 |
|
| 292 | static int tonga_ih_early_init(struct amdgpu_ip_block *ip_block)
|
| 293 | {
|
| 294 | struct amdgpu_device *adev = ip_block->adev;
|
| 295 | int ret;
|
| 296 |
|
| 297 | ret = amdgpu_irq_add_domain(adev);
|
| 298 | if (ret)
|
| 299 | return ret;
|
| 300 |
|
| 301 | tonga_ih_set_interrupt_funcs(adev);
|
| 302 |
|
| 303 | return 0;
|
| 304 | }
|
| 305 |
|
| 306 | static int tonga_ih_sw_init(struct amdgpu_ip_block *ip_block)
|
| 307 | {
|
| 308 | int r;
|
| 309 | struct amdgpu_device *adev = ip_block->adev;
|
| 310 |
|
| 311 | r = amdgpu_ih_ring_init(adev, &adev->irq.ih, 64 * 1024, true);
|
| 312 | if (r)
|
| 313 | return r;
|
| 314 |
|
| 315 | r = amdgpu_ih_ring_init(adev, &adev->irq.ih_soft, IH_SW_RING_SIZE, true);
|
| 316 | if (r)
|
| 317 | return r;
|
| 318 |
|
| 319 | adev->irq.ih.use_doorbell = true;
|
| 320 | adev->irq.ih.doorbell_index = adev->doorbell_index.ih;
|
| 321 |
|
| 322 | r = amdgpu_irq_init(adev);
|
| 323 |
|
| 324 | return r;
|
| 325 | }
|
| 326 |
|
| 327 | static int tonga_ih_sw_fini(struct amdgpu_ip_block *ip_block)
|
| 328 | {
|
| 329 | struct amdgpu_device *adev = ip_block->adev;
|
| 330 |
|
| 331 | amdgpu_irq_fini_sw(adev);
|
| 332 | amdgpu_irq_remove_domain(adev);
|
| 333 |
|
| 334 | return 0;
|
| 335 | }
|
| 336 |
|
| 337 | static int tonga_ih_hw_init(struct amdgpu_ip_block *ip_block)
|
| 338 | {
|
| 339 | int r;
|
| 340 | struct amdgpu_device *adev = ip_block->adev;
|
| 341 |
|
| 342 | r = tonga_ih_irq_init(adev);
|
| 343 | if (r)
|
| 344 | return r;
|
| 345 |
|
| 346 | return 0;
|
| 347 | }
|
| 348 |
|
| 349 | static int tonga_ih_hw_fini(struct amdgpu_ip_block *ip_block)
|
| 350 | {
|
| 351 | tonga_ih_irq_disable(ip_block->adev);
|
| 352 |
|
| 353 | return 0;
|
| 354 | }
|
| 355 |
|
| 356 | static int tonga_ih_suspend(struct amdgpu_ip_block *ip_block)
|
| 357 | {
|
| 358 | return tonga_ih_hw_fini(ip_block);
|
| 359 | }
|
| 360 |
|
| 361 | static int tonga_ih_resume(struct amdgpu_ip_block *ip_block)
|
| 362 | {
|
| 363 | return tonga_ih_hw_init(ip_block);
|
| 364 | }
|
| 365 |
|
| 366 | static bool tonga_ih_is_idle(struct amdgpu_ip_block *ip_block)
|
| 367 | {
|
| 368 | struct amdgpu_device *adev = ip_block->adev;
|
| 369 | u32 tmp = RREG32(mmSRBM_STATUS);
|
| 370 |
|
| 371 | if (REG_GET_FIELD(tmp, SRBM_STATUS, IH_BUSY))
|
| 372 | return false;
|
| 373 |
|
| 374 | return true;
|
| 375 | }
|
| 376 |
|
| 377 | static int tonga_ih_wait_for_idle(struct amdgpu_ip_block *ip_block)
|
| 378 | {
|
| 379 | unsigned i;
|
| 380 | u32 tmp;
|
| 381 | struct amdgpu_device *adev = ip_block->adev;
|
| 382 |
|
| 383 | for (i = 0; i < adev->usec_timeout; i++) {
|
| 384 |
|
| 385 | tmp = RREG32(mmSRBM_STATUS);
|
| 386 | if (!REG_GET_FIELD(tmp, SRBM_STATUS, IH_BUSY))
|
| 387 | return 0;
|
| 388 | udelay(1);
|
| 389 | }
|
| 390 | return -ETIMEDOUT;
|
| 391 | }
|
| 392 |
|
| 393 | static int tonga_ih_soft_reset(struct amdgpu_ip_block *ip_block)
|
| 394 | {
|
| 395 | struct amdgpu_device *adev = ip_block->adev;
|
| 396 | u32 srbm_soft_reset;
|
| 397 |
|
| 398 | if (!adev->irq.srbm_soft_reset)
|
| 399 | return 0;
|
| 400 | srbm_soft_reset = adev->irq.srbm_soft_reset;
|
| 401 |
|
| 402 | if (srbm_soft_reset) {
|
| 403 | u32 tmp;
|
| 404 |
|
| 405 | tmp = RREG32(mmSRBM_SOFT_RESET);
|
| 406 | tmp |= srbm_soft_reset;
|
| 407 | dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
|
| 408 | WREG32(mmSRBM_SOFT_RESET, tmp);
|
| 409 | tmp = RREG32(mmSRBM_SOFT_RESET);
|
| 410 |
|
| 411 | udelay(50);
|
| 412 |
|
| 413 | tmp &= ~srbm_soft_reset;
|
| 414 | WREG32(mmSRBM_SOFT_RESET, tmp);
|
| 415 | tmp = RREG32(mmSRBM_SOFT_RESET);
|
| 416 |
|
| 417 |
|
| 418 | udelay(50);
|
| 419 | }
|
| 420 |
|
| 421 | return 0;
|
| 422 | }
|
| 423 |
|
| 424 | static int tonga_ih_set_clockgating_state(struct amdgpu_ip_block *ip_block,
|
| 425 | enum amd_clockgating_state state)
|
| 426 | {
|
| 427 | return 0;
|
| 428 | }
|
| 429 |
|
| 430 | static int tonga_ih_set_powergating_state(struct amdgpu_ip_block *ip_block,
|
| 431 | enum amd_powergating_state state)
|
| 432 | {
|
| 433 | return 0;
|
| 434 | }
|
| 435 |
|
| 436 | static const struct amd_ip_funcs tonga_ih_ip_funcs = {
|
| 437 | .name = "tonga_ih",
|
| 438 | .early_init = tonga_ih_early_init,
|
| 439 | .sw_init = tonga_ih_sw_init,
|
| 440 | .sw_fini = tonga_ih_sw_fini,
|
| 441 | .hw_init = tonga_ih_hw_init,
|
| 442 | .hw_fini = tonga_ih_hw_fini,
|
| 443 | .suspend = tonga_ih_suspend,
|
| 444 | .resume = tonga_ih_resume,
|
| 445 | .is_idle = tonga_ih_is_idle,
|
| 446 | .wait_for_idle = tonga_ih_wait_for_idle,
|
| 447 | .soft_reset = tonga_ih_soft_reset,
|
| 448 | .set_clockgating_state = tonga_ih_set_clockgating_state,
|
| 449 | .set_powergating_state = tonga_ih_set_powergating_state,
|
| 450 | };
|
| 451 |
|
| 452 | static const struct amdgpu_ih_funcs tonga_ih_funcs = {
|
| 453 | .get_wptr = tonga_ih_get_wptr,
|
| 454 | .decode_iv = tonga_ih_decode_iv,
|
| 455 | .set_rptr = tonga_ih_set_rptr
|
| 456 | };
|
| 457 |
|
| 458 | static void tonga_ih_set_interrupt_funcs(struct amdgpu_device *adev)
|
| 459 | {
|
| 460 | adev->irq.ih_funcs = &tonga_ih_funcs;
|
| 461 | }
|
| 462 |
|
| 463 | const struct amdgpu_ip_block_version tonga_ih_ip_block = {
|
| 464 | .type = AMD_IP_BLOCK_TYPE_IH,
|
| 465 | .major = 3,
|
| 466 | .minor = 0,
|
| 467 | .rev = 0,
|
| 468 | .funcs = &tonga_ih_ip_funcs,
|
| 469 | };
|