| 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 | |
| 24 | #include <linux/pci.h> |
| 25 | |
| 26 | #include "amdgpu.h" |
| 27 | #include "amdgpu_ih.h" |
| 28 | #include "soc15.h" |
| 29 | |
| 30 | #include "oss/osssys_4_0_offset.h" |
| 31 | #include "oss/osssys_4_0_sh_mask.h" |
| 32 | |
| 33 | #include "soc15_common.h" |
| 34 | #include "vega10_ih.h" |
| 35 | |
| 36 | #define MAX_REARM_RETRY 10 |
| 37 | |
| 38 | static void vega10_ih_set_interrupt_funcs(struct amdgpu_device *adev); |
| 39 | |
| 40 | /** |
| 41 | * vega10_ih_init_register_offset - Initialize register offset for ih rings |
| 42 | * |
| 43 | * @adev: amdgpu_device pointer |
| 44 | * |
| 45 | * Initialize register offset ih rings (VEGA10). |
| 46 | */ |
| 47 | static void vega10_ih_init_register_offset(struct amdgpu_device *adev) |
| 48 | { |
| 49 | struct amdgpu_ih_regs *ih_regs; |
| 50 | |
| 51 | if (adev->irq.ih.ring_size) { |
| 52 | ih_regs = &adev->irq.ih.ih_regs; |
| 53 | ih_regs->ih_rb_base = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE); |
| 54 | ih_regs->ih_rb_base_hi = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_HI); |
| 55 | ih_regs->ih_rb_cntl = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL); |
| 56 | ih_regs->ih_rb_wptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR); |
| 57 | ih_regs->ih_rb_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR); |
| 58 | ih_regs->ih_doorbell_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_DOORBELL_RPTR); |
| 59 | ih_regs->ih_rb_wptr_addr_lo = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_ADDR_LO); |
| 60 | ih_regs->ih_rb_wptr_addr_hi = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_ADDR_HI); |
| 61 | ih_regs->psp_reg_id = PSP_REG_IH_RB_CNTL; |
| 62 | } |
| 63 | |
| 64 | if (adev->irq.ih1.ring_size) { |
| 65 | ih_regs = &adev->irq.ih1.ih_regs; |
| 66 | ih_regs->ih_rb_base = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_RING1); |
| 67 | ih_regs->ih_rb_base_hi = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_HI_RING1); |
| 68 | ih_regs->ih_rb_cntl = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL_RING1); |
| 69 | ih_regs->ih_rb_wptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_RING1); |
| 70 | ih_regs->ih_rb_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR_RING1); |
| 71 | ih_regs->ih_doorbell_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_DOORBELL_RPTR_RING1); |
| 72 | ih_regs->psp_reg_id = PSP_REG_IH_RB_CNTL_RING1; |
| 73 | } |
| 74 | |
| 75 | if (adev->irq.ih2.ring_size) { |
| 76 | ih_regs = &adev->irq.ih2.ih_regs; |
| 77 | ih_regs->ih_rb_base = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_RING2); |
| 78 | ih_regs->ih_rb_base_hi = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_HI_RING2); |
| 79 | ih_regs->ih_rb_cntl = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL_RING2); |
| 80 | ih_regs->ih_rb_wptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_RING2); |
| 81 | ih_regs->ih_rb_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR_RING2); |
| 82 | ih_regs->ih_doorbell_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_DOORBELL_RPTR_RING2); |
| 83 | ih_regs->psp_reg_id = PSP_REG_IH_RB_CNTL_RING2; |
| 84 | } |
| 85 | } |
| 86 | |
| 87 | /** |
| 88 | * vega10_ih_toggle_ring_interrupts - toggle the interrupt ring buffer |
| 89 | * |
| 90 | * @adev: amdgpu_device pointer |
| 91 | * @ih: amdgpu_ih_ring pointet |
| 92 | * @enable: true - enable the interrupts, false - disable the interrupts |
| 93 | * |
| 94 | * Toggle the interrupt ring buffer (VEGA10) |
| 95 | */ |
| 96 | static int vega10_ih_toggle_ring_interrupts(struct amdgpu_device *adev, |
| 97 | struct amdgpu_ih_ring *ih, |
| 98 | bool enable) |
| 99 | { |
| 100 | struct amdgpu_ih_regs *ih_regs; |
| 101 | uint32_t tmp; |
| 102 | |
| 103 | ih_regs = &ih->ih_regs; |
| 104 | |
| 105 | tmp = RREG32(ih_regs->ih_rb_cntl); |
| 106 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RB_ENABLE, (enable ? 1 : 0)); |
| 107 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RB_GPU_TS_ENABLE, 1); |
| 108 | /* enable_intr field is only valid in ring0 */ |
| 109 | if (ih == &adev->irq.ih) |
| 110 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, ENABLE_INTR, (enable ? 1 : 0)); |
| 111 | if (amdgpu_sriov_vf(adev)) { |
| 112 | if (psp_reg_program(psp: &adev->psp, reg: ih_regs->psp_reg_id, value: tmp)) { |
| 113 | dev_err(adev->dev, "PSP program IH_RB_CNTL failed!\n" ); |
| 114 | return -ETIMEDOUT; |
| 115 | } |
| 116 | } else { |
| 117 | WREG32(ih_regs->ih_rb_cntl, tmp); |
| 118 | } |
| 119 | |
| 120 | if (enable) { |
| 121 | ih->enabled = true; |
| 122 | } else { |
| 123 | /* set rptr, wptr to 0 */ |
| 124 | WREG32(ih_regs->ih_rb_rptr, 0); |
| 125 | WREG32(ih_regs->ih_rb_wptr, 0); |
| 126 | ih->enabled = false; |
| 127 | ih->rptr = 0; |
| 128 | } |
| 129 | |
| 130 | return 0; |
| 131 | } |
| 132 | |
| 133 | /** |
| 134 | * vega10_ih_toggle_interrupts - Toggle all the available interrupt ring buffers |
| 135 | * |
| 136 | * @adev: amdgpu_device pointer |
| 137 | * @enable: enable or disable interrupt ring buffers |
| 138 | * |
| 139 | * Toggle all the available interrupt ring buffers (VEGA10). |
| 140 | */ |
| 141 | static int vega10_ih_toggle_interrupts(struct amdgpu_device *adev, bool enable) |
| 142 | { |
| 143 | struct amdgpu_ih_ring *ih[] = {&adev->irq.ih, &adev->irq.ih1, &adev->irq.ih2}; |
| 144 | int i; |
| 145 | int r; |
| 146 | |
| 147 | for (i = 0; i < ARRAY_SIZE(ih); i++) { |
| 148 | if (ih[i]->ring_size) { |
| 149 | r = vega10_ih_toggle_ring_interrupts(adev, ih: ih[i], enable); |
| 150 | if (r) |
| 151 | return r; |
| 152 | } |
| 153 | } |
| 154 | |
| 155 | return 0; |
| 156 | } |
| 157 | |
| 158 | static uint32_t vega10_ih_rb_cntl(struct amdgpu_ih_ring *ih, uint32_t ih_rb_cntl) |
| 159 | { |
| 160 | int rb_bufsz = order_base_2(ih->ring_size / 4); |
| 161 | |
| 162 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, |
| 163 | MC_SPACE, ih->use_bus_addr ? 1 : 4); |
| 164 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, |
| 165 | WPTR_OVERFLOW_CLEAR, 1); |
| 166 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, |
| 167 | WPTR_OVERFLOW_ENABLE, 1); |
| 168 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_SIZE, rb_bufsz); |
| 169 | /* Ring Buffer write pointer writeback. If enabled, IH_RB_WPTR register |
| 170 | * value is written to memory |
| 171 | */ |
| 172 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, |
| 173 | WPTR_WRITEBACK_ENABLE, 1); |
| 174 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SNOOP, 1); |
| 175 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_RO, 0); |
| 176 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_VMID, 0); |
| 177 | |
| 178 | return ih_rb_cntl; |
| 179 | } |
| 180 | |
| 181 | static uint32_t vega10_ih_doorbell_rptr(struct amdgpu_ih_ring *ih) |
| 182 | { |
| 183 | u32 ih_doorbell_rtpr = 0; |
| 184 | |
| 185 | if (ih->use_doorbell) { |
| 186 | ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, |
| 187 | IH_DOORBELL_RPTR, OFFSET, |
| 188 | ih->doorbell_index); |
| 189 | ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, |
| 190 | IH_DOORBELL_RPTR, |
| 191 | ENABLE, 1); |
| 192 | } else { |
| 193 | ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, |
| 194 | IH_DOORBELL_RPTR, |
| 195 | ENABLE, 0); |
| 196 | } |
| 197 | return ih_doorbell_rtpr; |
| 198 | } |
| 199 | |
| 200 | /** |
| 201 | * vega10_ih_enable_ring - enable an ih ring buffer |
| 202 | * |
| 203 | * @adev: amdgpu_device pointer |
| 204 | * @ih: amdgpu_ih_ring pointer |
| 205 | * |
| 206 | * Enable an ih ring buffer (VEGA10) |
| 207 | */ |
| 208 | static int vega10_ih_enable_ring(struct amdgpu_device *adev, |
| 209 | struct amdgpu_ih_ring *ih) |
| 210 | { |
| 211 | struct amdgpu_ih_regs *ih_regs; |
| 212 | uint32_t tmp; |
| 213 | |
| 214 | ih_regs = &ih->ih_regs; |
| 215 | |
| 216 | /* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/ |
| 217 | WREG32(ih_regs->ih_rb_base, ih->gpu_addr >> 8); |
| 218 | WREG32(ih_regs->ih_rb_base_hi, (ih->gpu_addr >> 40) & 0xff); |
| 219 | |
| 220 | tmp = RREG32(ih_regs->ih_rb_cntl); |
| 221 | tmp = vega10_ih_rb_cntl(ih, ih_rb_cntl: tmp); |
| 222 | if (ih == &adev->irq.ih) |
| 223 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RPTR_REARM, !!adev->irq.msi_enabled); |
| 224 | if (ih == &adev->irq.ih1) |
| 225 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RB_FULL_DRAIN_ENABLE, 1); |
| 226 | if (amdgpu_sriov_vf(adev)) { |
| 227 | if (psp_reg_program(psp: &adev->psp, reg: ih_regs->psp_reg_id, value: tmp)) { |
| 228 | dev_err(adev->dev, "PSP program IH_RB_CNTL failed!\n" ); |
| 229 | return -ETIMEDOUT; |
| 230 | } |
| 231 | } else { |
| 232 | WREG32(ih_regs->ih_rb_cntl, tmp); |
| 233 | } |
| 234 | |
| 235 | if (ih == &adev->irq.ih) { |
| 236 | /* set the ih ring 0 writeback address whether it's enabled or not */ |
| 237 | WREG32(ih_regs->ih_rb_wptr_addr_lo, lower_32_bits(ih->wptr_addr)); |
| 238 | WREG32(ih_regs->ih_rb_wptr_addr_hi, upper_32_bits(ih->wptr_addr) & 0xFFFF); |
| 239 | } |
| 240 | |
| 241 | /* set rptr, wptr to 0 */ |
| 242 | WREG32(ih_regs->ih_rb_wptr, 0); |
| 243 | WREG32(ih_regs->ih_rb_rptr, 0); |
| 244 | |
| 245 | WREG32(ih_regs->ih_doorbell_rptr, vega10_ih_doorbell_rptr(ih)); |
| 246 | |
| 247 | return 0; |
| 248 | } |
| 249 | |
| 250 | /** |
| 251 | * vega10_ih_irq_init - init and enable the interrupt ring |
| 252 | * |
| 253 | * @adev: amdgpu_device pointer |
| 254 | * |
| 255 | * Allocate a ring buffer for the interrupt controller, |
| 256 | * enable the RLC, disable interrupts, enable the IH |
| 257 | * ring buffer and enable it (VI). |
| 258 | * Called at device load and reume. |
| 259 | * Returns 0 for success, errors for failure. |
| 260 | */ |
| 261 | static int vega10_ih_irq_init(struct amdgpu_device *adev) |
| 262 | { |
| 263 | struct amdgpu_ih_ring *ih[] = {&adev->irq.ih, &adev->irq.ih1, &adev->irq.ih2}; |
| 264 | u32 ih_chicken; |
| 265 | int ret; |
| 266 | int i; |
| 267 | |
| 268 | /* disable irqs */ |
| 269 | ret = vega10_ih_toggle_interrupts(adev, enable: false); |
| 270 | if (ret) |
| 271 | return ret; |
| 272 | |
| 273 | adev->nbio.funcs->ih_control(adev); |
| 274 | |
| 275 | if (adev->asic_type == CHIP_RENOIR) { |
| 276 | ih_chicken = RREG32_SOC15(OSSSYS, 0, mmIH_CHICKEN); |
| 277 | if (adev->irq.ih.use_bus_addr) { |
| 278 | ih_chicken = REG_SET_FIELD(ih_chicken, IH_CHICKEN, |
| 279 | MC_SPACE_GPA_ENABLE, 1); |
| 280 | } |
| 281 | WREG32_SOC15(OSSSYS, 0, mmIH_CHICKEN, ih_chicken); |
| 282 | } |
| 283 | |
| 284 | for (i = 0; i < ARRAY_SIZE(ih); i++) { |
| 285 | if (ih[i]->ring_size) { |
| 286 | ret = vega10_ih_enable_ring(adev, ih: ih[i]); |
| 287 | if (ret) |
| 288 | return ret; |
| 289 | } |
| 290 | } |
| 291 | |
| 292 | if (!amdgpu_sriov_vf(adev)) |
| 293 | adev->nbio.funcs->ih_doorbell_range(adev, adev->irq.ih.use_doorbell, |
| 294 | adev->irq.ih.doorbell_index); |
| 295 | |
| 296 | pci_set_master(dev: adev->pdev); |
| 297 | |
| 298 | /* enable interrupts */ |
| 299 | ret = vega10_ih_toggle_interrupts(adev, enable: true); |
| 300 | if (ret) |
| 301 | return ret; |
| 302 | |
| 303 | if (adev->irq.ih_soft.ring_size) |
| 304 | adev->irq.ih_soft.enabled = true; |
| 305 | |
| 306 | return 0; |
| 307 | } |
| 308 | |
| 309 | /** |
| 310 | * vega10_ih_irq_disable - disable interrupts |
| 311 | * |
| 312 | * @adev: amdgpu_device pointer |
| 313 | * |
| 314 | * Disable interrupts on the hw (VEGA10). |
| 315 | */ |
| 316 | static void vega10_ih_irq_disable(struct amdgpu_device *adev) |
| 317 | { |
| 318 | vega10_ih_toggle_interrupts(adev, enable: false); |
| 319 | |
| 320 | /* Wait and acknowledge irq */ |
| 321 | mdelay(1); |
| 322 | } |
| 323 | |
| 324 | /** |
| 325 | * vega10_ih_get_wptr - get the IH ring buffer wptr |
| 326 | * |
| 327 | * @adev: amdgpu_device pointer |
| 328 | * @ih: IH ring buffer to fetch wptr |
| 329 | * |
| 330 | * Get the IH ring buffer wptr from either the register |
| 331 | * or the writeback memory buffer (VEGA10). Also check for |
| 332 | * ring buffer overflow and deal with it. |
| 333 | * Returns the value of the wptr. |
| 334 | */ |
| 335 | static u32 vega10_ih_get_wptr(struct amdgpu_device *adev, |
| 336 | struct amdgpu_ih_ring *ih) |
| 337 | { |
| 338 | u32 wptr, tmp; |
| 339 | struct amdgpu_ih_regs *ih_regs; |
| 340 | |
| 341 | if (ih == &adev->irq.ih || ih == &adev->irq.ih_soft) { |
| 342 | /* Only ring0 supports writeback. On other rings fall back |
| 343 | * to register-based code with overflow checking below. |
| 344 | * ih_soft ring doesn't have any backing hardware registers, |
| 345 | * update wptr and return. |
| 346 | */ |
| 347 | wptr = le32_to_cpu(*ih->wptr_cpu); |
| 348 | |
| 349 | if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) |
| 350 | goto out; |
| 351 | } |
| 352 | |
| 353 | ih_regs = &ih->ih_regs; |
| 354 | |
| 355 | /* Double check that the overflow wasn't already cleared. */ |
| 356 | wptr = RREG32_NO_KIQ(ih_regs->ih_rb_wptr); |
| 357 | if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) |
| 358 | goto out; |
| 359 | |
| 360 | wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0); |
| 361 | |
| 362 | /* When a ring buffer overflow happen start parsing interrupt |
| 363 | * from the last not overwritten vector (wptr + 32). Hopefully |
| 364 | * this should allow us to catchup. |
| 365 | */ |
| 366 | tmp = (wptr + 32) & ih->ptr_mask; |
| 367 | dev_warn_ratelimited(adev->dev, "%s ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n" , |
| 368 | amdgpu_ih_ring_name(adev, ih), wptr, ih->rptr, tmp); |
| 369 | ih->rptr = tmp; |
| 370 | |
| 371 | tmp = RREG32_NO_KIQ(ih_regs->ih_rb_cntl); |
| 372 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); |
| 373 | WREG32_NO_KIQ(ih_regs->ih_rb_cntl, tmp); |
| 374 | |
| 375 | /* Unset the CLEAR_OVERFLOW bit immediately so new overflows |
| 376 | * can be detected. |
| 377 | */ |
| 378 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 0); |
| 379 | WREG32_NO_KIQ(ih_regs->ih_rb_cntl, tmp); |
| 380 | |
| 381 | out: |
| 382 | return (wptr & ih->ptr_mask); |
| 383 | } |
| 384 | |
| 385 | /** |
| 386 | * vega10_ih_irq_rearm - rearm IRQ if lost |
| 387 | * |
| 388 | * @adev: amdgpu_device pointer |
| 389 | * @ih: IH ring to match |
| 390 | * |
| 391 | */ |
| 392 | static void vega10_ih_irq_rearm(struct amdgpu_device *adev, |
| 393 | struct amdgpu_ih_ring *ih) |
| 394 | { |
| 395 | uint32_t v = 0; |
| 396 | uint32_t i = 0; |
| 397 | struct amdgpu_ih_regs *ih_regs; |
| 398 | |
| 399 | ih_regs = &ih->ih_regs; |
| 400 | /* Rearm IRQ / re-wwrite doorbell if doorbell write is lost */ |
| 401 | for (i = 0; i < MAX_REARM_RETRY; i++) { |
| 402 | v = RREG32_NO_KIQ(ih_regs->ih_rb_rptr); |
| 403 | if ((v < ih->ring_size) && (v != ih->rptr)) |
| 404 | WDOORBELL32(ih->doorbell_index, ih->rptr); |
| 405 | else |
| 406 | break; |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | /** |
| 411 | * vega10_ih_set_rptr - set the IH ring buffer rptr |
| 412 | * |
| 413 | * @adev: amdgpu_device pointer |
| 414 | * @ih: IH ring buffer to set rptr |
| 415 | * |
| 416 | * Set the IH ring buffer rptr. |
| 417 | */ |
| 418 | static void vega10_ih_set_rptr(struct amdgpu_device *adev, |
| 419 | struct amdgpu_ih_ring *ih) |
| 420 | { |
| 421 | struct amdgpu_ih_regs *ih_regs; |
| 422 | |
| 423 | if (ih == &adev->irq.ih_soft) |
| 424 | return; |
| 425 | |
| 426 | if (ih->use_doorbell) { |
| 427 | /* XXX check if swapping is necessary on BE */ |
| 428 | *ih->rptr_cpu = ih->rptr; |
| 429 | WDOORBELL32(ih->doorbell_index, ih->rptr); |
| 430 | |
| 431 | if (amdgpu_sriov_vf(adev)) |
| 432 | vega10_ih_irq_rearm(adev, ih); |
| 433 | } else { |
| 434 | ih_regs = &ih->ih_regs; |
| 435 | WREG32(ih_regs->ih_rb_rptr, ih->rptr); |
| 436 | } |
| 437 | } |
| 438 | |
| 439 | /** |
| 440 | * vega10_ih_self_irq - dispatch work for ring 1 and 2 |
| 441 | * |
| 442 | * @adev: amdgpu_device pointer |
| 443 | * @source: irq source |
| 444 | * @entry: IV with WPTR update |
| 445 | * |
| 446 | * Update the WPTR from the IV and schedule work to handle the entries. |
| 447 | */ |
| 448 | static int vega10_ih_self_irq(struct amdgpu_device *adev, |
| 449 | struct amdgpu_irq_src *source, |
| 450 | struct amdgpu_iv_entry *entry) |
| 451 | { |
| 452 | switch (entry->ring_id) { |
| 453 | case 1: |
| 454 | schedule_work(work: &adev->irq.ih1_work); |
| 455 | break; |
| 456 | case 2: |
| 457 | schedule_work(work: &adev->irq.ih2_work); |
| 458 | break; |
| 459 | default: break; |
| 460 | } |
| 461 | return 0; |
| 462 | } |
| 463 | |
| 464 | static const struct amdgpu_irq_src_funcs vega10_ih_self_irq_funcs = { |
| 465 | .process = vega10_ih_self_irq, |
| 466 | }; |
| 467 | |
| 468 | static void vega10_ih_set_self_irq_funcs(struct amdgpu_device *adev) |
| 469 | { |
| 470 | adev->irq.self_irq.num_types = 0; |
| 471 | adev->irq.self_irq.funcs = &vega10_ih_self_irq_funcs; |
| 472 | } |
| 473 | |
| 474 | static int vega10_ih_early_init(struct amdgpu_ip_block *ip_block) |
| 475 | { |
| 476 | struct amdgpu_device *adev = ip_block->adev; |
| 477 | |
| 478 | vega10_ih_set_interrupt_funcs(adev); |
| 479 | vega10_ih_set_self_irq_funcs(adev); |
| 480 | return 0; |
| 481 | } |
| 482 | |
| 483 | static int vega10_ih_sw_init(struct amdgpu_ip_block *ip_block) |
| 484 | { |
| 485 | struct amdgpu_device *adev = ip_block->adev; |
| 486 | int r; |
| 487 | |
| 488 | r = amdgpu_irq_add_id(adev, client_id: SOC15_IH_CLIENTID_IH, src_id: 0, |
| 489 | source: &adev->irq.self_irq); |
| 490 | if (r) |
| 491 | return r; |
| 492 | |
| 493 | r = amdgpu_ih_ring_init(adev, ih: &adev->irq.ih, IH_RING_SIZE, use_bus_addr: true); |
| 494 | if (r) |
| 495 | return r; |
| 496 | |
| 497 | adev->irq.ih.use_doorbell = true; |
| 498 | adev->irq.ih.doorbell_index = adev->doorbell_index.ih << 1; |
| 499 | |
| 500 | if (!(adev->flags & AMD_IS_APU)) { |
| 501 | r = amdgpu_ih_ring_init(adev, ih: &adev->irq.ih1, PAGE_SIZE, use_bus_addr: true); |
| 502 | if (r) |
| 503 | return r; |
| 504 | |
| 505 | adev->irq.ih1.use_doorbell = true; |
| 506 | adev->irq.ih1.doorbell_index = (adev->doorbell_index.ih + 1) << 1; |
| 507 | |
| 508 | r = amdgpu_ih_ring_init(adev, ih: &adev->irq.ih2, PAGE_SIZE, use_bus_addr: true); |
| 509 | if (r) |
| 510 | return r; |
| 511 | |
| 512 | adev->irq.ih2.use_doorbell = true; |
| 513 | adev->irq.ih2.doorbell_index = (adev->doorbell_index.ih + 2) << 1; |
| 514 | } |
| 515 | /* initialize ih control registers offset */ |
| 516 | vega10_ih_init_register_offset(adev); |
| 517 | |
| 518 | r = amdgpu_ih_ring_init(adev, ih: &adev->irq.ih_soft, IH_SW_RING_SIZE, use_bus_addr: true); |
| 519 | if (r) |
| 520 | return r; |
| 521 | |
| 522 | r = amdgpu_irq_init(adev); |
| 523 | |
| 524 | return r; |
| 525 | } |
| 526 | |
| 527 | static int vega10_ih_sw_fini(struct amdgpu_ip_block *ip_block) |
| 528 | { |
| 529 | struct amdgpu_device *adev = ip_block->adev; |
| 530 | |
| 531 | amdgpu_irq_fini_sw(adev); |
| 532 | |
| 533 | return 0; |
| 534 | } |
| 535 | |
| 536 | static int vega10_ih_hw_init(struct amdgpu_ip_block *ip_block) |
| 537 | { |
| 538 | return vega10_ih_irq_init(adev: ip_block->adev); |
| 539 | } |
| 540 | |
| 541 | static int vega10_ih_hw_fini(struct amdgpu_ip_block *ip_block) |
| 542 | { |
| 543 | vega10_ih_irq_disable(adev: ip_block->adev); |
| 544 | |
| 545 | return 0; |
| 546 | } |
| 547 | |
| 548 | static int vega10_ih_suspend(struct amdgpu_ip_block *ip_block) |
| 549 | { |
| 550 | return vega10_ih_hw_fini(ip_block); |
| 551 | } |
| 552 | |
| 553 | static int vega10_ih_resume(struct amdgpu_ip_block *ip_block) |
| 554 | { |
| 555 | return vega10_ih_hw_init(ip_block); |
| 556 | } |
| 557 | |
| 558 | static bool vega10_ih_is_idle(struct amdgpu_ip_block *ip_block) |
| 559 | { |
| 560 | /* todo */ |
| 561 | return true; |
| 562 | } |
| 563 | |
| 564 | static int vega10_ih_wait_for_idle(struct amdgpu_ip_block *ip_block) |
| 565 | { |
| 566 | /* todo */ |
| 567 | return -ETIMEDOUT; |
| 568 | } |
| 569 | |
| 570 | static int vega10_ih_soft_reset(struct amdgpu_ip_block *ip_block) |
| 571 | { |
| 572 | /* todo */ |
| 573 | |
| 574 | return 0; |
| 575 | } |
| 576 | |
| 577 | static void vega10_ih_update_clockgating_state(struct amdgpu_device *adev, |
| 578 | bool enable) |
| 579 | { |
| 580 | uint32_t data, def, field_val; |
| 581 | |
| 582 | if (adev->cg_flags & AMD_CG_SUPPORT_IH_CG) { |
| 583 | def = data = RREG32_SOC15(OSSSYS, 0, mmIH_CLK_CTRL); |
| 584 | field_val = enable ? 0 : 1; |
| 585 | /** |
| 586 | * Vega10/12 and RAVEN don't have IH_BUFFER_MEM_CLK_SOFT_OVERRIDE field. |
| 587 | */ |
| 588 | if (adev->asic_type == CHIP_RENOIR) |
| 589 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 590 | IH_BUFFER_MEM_CLK_SOFT_OVERRIDE, field_val); |
| 591 | |
| 592 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 593 | DBUS_MUX_CLK_SOFT_OVERRIDE, field_val); |
| 594 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 595 | OSSSYS_SHARE_CLK_SOFT_OVERRIDE, field_val); |
| 596 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 597 | LIMIT_SMN_CLK_SOFT_OVERRIDE, field_val); |
| 598 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 599 | DYN_CLK_SOFT_OVERRIDE, field_val); |
| 600 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 601 | REG_CLK_SOFT_OVERRIDE, field_val); |
| 602 | if (def != data) |
| 603 | WREG32_SOC15(OSSSYS, 0, mmIH_CLK_CTRL, data); |
| 604 | } |
| 605 | } |
| 606 | |
| 607 | static int vega10_ih_set_clockgating_state(struct amdgpu_ip_block *ip_block, |
| 608 | enum amd_clockgating_state state) |
| 609 | { |
| 610 | struct amdgpu_device *adev = ip_block->adev; |
| 611 | |
| 612 | vega10_ih_update_clockgating_state(adev, |
| 613 | enable: state == AMD_CG_STATE_GATE); |
| 614 | return 0; |
| 615 | |
| 616 | } |
| 617 | |
| 618 | static int vega10_ih_set_powergating_state(struct amdgpu_ip_block *ip_block, |
| 619 | enum amd_powergating_state state) |
| 620 | { |
| 621 | return 0; |
| 622 | } |
| 623 | |
| 624 | const struct amd_ip_funcs vega10_ih_ip_funcs = { |
| 625 | .name = "vega10_ih" , |
| 626 | .early_init = vega10_ih_early_init, |
| 627 | .sw_init = vega10_ih_sw_init, |
| 628 | .sw_fini = vega10_ih_sw_fini, |
| 629 | .hw_init = vega10_ih_hw_init, |
| 630 | .hw_fini = vega10_ih_hw_fini, |
| 631 | .suspend = vega10_ih_suspend, |
| 632 | .resume = vega10_ih_resume, |
| 633 | .is_idle = vega10_ih_is_idle, |
| 634 | .wait_for_idle = vega10_ih_wait_for_idle, |
| 635 | .soft_reset = vega10_ih_soft_reset, |
| 636 | .set_clockgating_state = vega10_ih_set_clockgating_state, |
| 637 | .set_powergating_state = vega10_ih_set_powergating_state, |
| 638 | }; |
| 639 | |
| 640 | static const struct amdgpu_ih_funcs vega10_ih_funcs = { |
| 641 | .get_wptr = vega10_ih_get_wptr, |
| 642 | .decode_iv = amdgpu_ih_decode_iv_helper, |
| 643 | .decode_iv_ts = amdgpu_ih_decode_iv_ts_helper, |
| 644 | .set_rptr = vega10_ih_set_rptr |
| 645 | }; |
| 646 | |
| 647 | static void vega10_ih_set_interrupt_funcs(struct amdgpu_device *adev) |
| 648 | { |
| 649 | adev->irq.ih_funcs = &vega10_ih_funcs; |
| 650 | } |
| 651 | |
| 652 | const struct amdgpu_ip_block_version vega10_ih_ip_block = |
| 653 | { |
| 654 | .type = AMD_IP_BLOCK_TYPE_IH, |
| 655 | .major = 4, |
| 656 | .minor = 0, |
| 657 | .rev = 0, |
| 658 | .funcs = &vega10_ih_ip_funcs, |
| 659 | }; |
| 660 | |