| 1 | /* |
| 2 | * Copyright 2019 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 | |
| 29 | #include "oss/osssys_5_0_0_offset.h" |
| 30 | #include "oss/osssys_5_0_0_sh_mask.h" |
| 31 | |
| 32 | #include "soc15_common.h" |
| 33 | #include "navi10_ih.h" |
| 34 | |
| 35 | #define MAX_REARM_RETRY 10 |
| 36 | |
| 37 | #define mmIH_CHICKEN_Sienna_Cichlid 0x018d |
| 38 | #define mmIH_CHICKEN_Sienna_Cichlid_BASE_IDX 0 |
| 39 | |
| 40 | static void navi10_ih_set_interrupt_funcs(struct amdgpu_device *adev); |
| 41 | |
| 42 | /** |
| 43 | * navi10_ih_init_register_offset - Initialize register offset for ih rings |
| 44 | * |
| 45 | * @adev: amdgpu_device pointer |
| 46 | * |
| 47 | * Initialize register offset ih rings (NAVI10). |
| 48 | */ |
| 49 | static void navi10_ih_init_register_offset(struct amdgpu_device *adev) |
| 50 | { |
| 51 | struct amdgpu_ih_regs *ih_regs; |
| 52 | |
| 53 | if (adev->irq.ih.ring_size) { |
| 54 | ih_regs = &adev->irq.ih.ih_regs; |
| 55 | ih_regs->ih_rb_base = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE); |
| 56 | ih_regs->ih_rb_base_hi = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_HI); |
| 57 | ih_regs->ih_rb_cntl = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL); |
| 58 | ih_regs->ih_rb_wptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR); |
| 59 | ih_regs->ih_rb_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR); |
| 60 | ih_regs->ih_doorbell_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_DOORBELL_RPTR); |
| 61 | ih_regs->ih_rb_wptr_addr_lo = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_ADDR_LO); |
| 62 | ih_regs->ih_rb_wptr_addr_hi = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_ADDR_HI); |
| 63 | ih_regs->psp_reg_id = PSP_REG_IH_RB_CNTL; |
| 64 | } |
| 65 | |
| 66 | if (adev->irq.ih1.ring_size) { |
| 67 | ih_regs = &adev->irq.ih1.ih_regs; |
| 68 | ih_regs->ih_rb_base = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_RING1); |
| 69 | ih_regs->ih_rb_base_hi = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_HI_RING1); |
| 70 | ih_regs->ih_rb_cntl = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL_RING1); |
| 71 | ih_regs->ih_rb_wptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_RING1); |
| 72 | ih_regs->ih_rb_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR_RING1); |
| 73 | ih_regs->ih_doorbell_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_DOORBELL_RPTR_RING1); |
| 74 | ih_regs->psp_reg_id = PSP_REG_IH_RB_CNTL_RING1; |
| 75 | } |
| 76 | |
| 77 | if (adev->irq.ih2.ring_size) { |
| 78 | ih_regs = &adev->irq.ih2.ih_regs; |
| 79 | ih_regs->ih_rb_base = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_RING2); |
| 80 | ih_regs->ih_rb_base_hi = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_HI_RING2); |
| 81 | ih_regs->ih_rb_cntl = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL_RING2); |
| 82 | ih_regs->ih_rb_wptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_RING2); |
| 83 | ih_regs->ih_rb_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR_RING2); |
| 84 | ih_regs->ih_doorbell_rptr = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_DOORBELL_RPTR_RING2); |
| 85 | ih_regs->psp_reg_id = PSP_REG_IH_RB_CNTL_RING2; |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | /** |
| 90 | * force_update_wptr_for_self_int - Force update the wptr for self interrupt |
| 91 | * |
| 92 | * @adev: amdgpu_device pointer |
| 93 | * @threshold: threshold to trigger the wptr reporting |
| 94 | * @timeout: timeout to trigger the wptr reporting |
| 95 | * @enabled: Enable/disable timeout flush mechanism |
| 96 | * |
| 97 | * threshold input range: 0 ~ 15, default 0, |
| 98 | * real_threshold = 2^threshold |
| 99 | * timeout input range: 0 ~ 20, default 8, |
| 100 | * real_timeout = (2^timeout) * 1024 / (socclk_freq) |
| 101 | * |
| 102 | * Force update wptr for self interrupt ( >= SIENNA_CICHLID). |
| 103 | */ |
| 104 | static void |
| 105 | force_update_wptr_for_self_int(struct amdgpu_device *adev, |
| 106 | u32 threshold, u32 timeout, bool enabled) |
| 107 | { |
| 108 | u32 ih_cntl, ih_rb_cntl; |
| 109 | |
| 110 | if (amdgpu_ip_version(adev, ip: OSSSYS_HWIP, inst: 0) < IP_VERSION(5, 0, 3)) |
| 111 | return; |
| 112 | |
| 113 | ih_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_CNTL2); |
| 114 | ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING1); |
| 115 | |
| 116 | ih_cntl = REG_SET_FIELD(ih_cntl, IH_CNTL2, |
| 117 | SELF_IV_FORCE_WPTR_UPDATE_TIMEOUT, timeout); |
| 118 | ih_cntl = REG_SET_FIELD(ih_cntl, IH_CNTL2, |
| 119 | SELF_IV_FORCE_WPTR_UPDATE_ENABLE, enabled); |
| 120 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL_RING1, |
| 121 | RB_USED_INT_THRESHOLD, threshold); |
| 122 | |
| 123 | if (amdgpu_sriov_vf(adev) && amdgpu_sriov_reg_indirect_ih(adev)) { |
| 124 | if (psp_reg_program(psp: &adev->psp, reg: PSP_REG_IH_RB_CNTL_RING1, value: ih_rb_cntl)) |
| 125 | return; |
| 126 | } else { |
| 127 | WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING1, ih_rb_cntl); |
| 128 | } |
| 129 | |
| 130 | ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING2); |
| 131 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL_RING2, |
| 132 | RB_USED_INT_THRESHOLD, threshold); |
| 133 | if (amdgpu_sriov_vf(adev) && amdgpu_sriov_reg_indirect_ih(adev)) { |
| 134 | if (psp_reg_program(psp: &adev->psp, reg: PSP_REG_IH_RB_CNTL_RING2, value: ih_rb_cntl)) |
| 135 | return; |
| 136 | } else { |
| 137 | WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL_RING2, ih_rb_cntl); |
| 138 | } |
| 139 | |
| 140 | WREG32_SOC15(OSSSYS, 0, mmIH_CNTL2, ih_cntl); |
| 141 | } |
| 142 | |
| 143 | /** |
| 144 | * navi10_ih_toggle_ring_interrupts - toggle the interrupt ring buffer |
| 145 | * |
| 146 | * @adev: amdgpu_device pointer |
| 147 | * @ih: amdgpu_ih_ring pointet |
| 148 | * @enable: true - enable the interrupts, false - disable the interrupts |
| 149 | * |
| 150 | * Toggle the interrupt ring buffer (NAVI10) |
| 151 | */ |
| 152 | static int navi10_ih_toggle_ring_interrupts(struct amdgpu_device *adev, |
| 153 | struct amdgpu_ih_ring *ih, |
| 154 | bool enable) |
| 155 | { |
| 156 | struct amdgpu_ih_regs *ih_regs; |
| 157 | uint32_t tmp; |
| 158 | |
| 159 | ih_regs = &ih->ih_regs; |
| 160 | |
| 161 | tmp = RREG32(ih_regs->ih_rb_cntl); |
| 162 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RB_ENABLE, (enable ? 1 : 0)); |
| 163 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RB_GPU_TS_ENABLE, 1); |
| 164 | /* enable_intr field is only valid in ring0 */ |
| 165 | if (ih == &adev->irq.ih) |
| 166 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, ENABLE_INTR, (enable ? 1 : 0)); |
| 167 | |
| 168 | if (amdgpu_sriov_vf(adev) && amdgpu_sriov_reg_indirect_ih(adev)) { |
| 169 | if (psp_reg_program(psp: &adev->psp, reg: ih_regs->psp_reg_id, value: tmp)) |
| 170 | return -ETIMEDOUT; |
| 171 | } else { |
| 172 | WREG32(ih_regs->ih_rb_cntl, tmp); |
| 173 | } |
| 174 | |
| 175 | if (enable) { |
| 176 | ih->enabled = true; |
| 177 | } else { |
| 178 | /* set rptr, wptr to 0 */ |
| 179 | WREG32(ih_regs->ih_rb_rptr, 0); |
| 180 | WREG32(ih_regs->ih_rb_wptr, 0); |
| 181 | ih->enabled = false; |
| 182 | ih->rptr = 0; |
| 183 | } |
| 184 | |
| 185 | return 0; |
| 186 | } |
| 187 | |
| 188 | /** |
| 189 | * navi10_ih_toggle_interrupts - Toggle all the available interrupt ring buffers |
| 190 | * |
| 191 | * @adev: amdgpu_device pointer |
| 192 | * @enable: enable or disable interrupt ring buffers |
| 193 | * |
| 194 | * Toggle all the available interrupt ring buffers (NAVI10). |
| 195 | */ |
| 196 | static int navi10_ih_toggle_interrupts(struct amdgpu_device *adev, bool enable) |
| 197 | { |
| 198 | struct amdgpu_ih_ring *ih[] = {&adev->irq.ih, &adev->irq.ih1, &adev->irq.ih2}; |
| 199 | int i; |
| 200 | int r; |
| 201 | |
| 202 | for (i = 0; i < ARRAY_SIZE(ih); i++) { |
| 203 | if (ih[i]->ring_size) { |
| 204 | r = navi10_ih_toggle_ring_interrupts(adev, ih: ih[i], enable); |
| 205 | if (r) |
| 206 | return r; |
| 207 | } |
| 208 | } |
| 209 | |
| 210 | return 0; |
| 211 | } |
| 212 | |
| 213 | static uint32_t navi10_ih_rb_cntl(struct amdgpu_ih_ring *ih, uint32_t ih_rb_cntl) |
| 214 | { |
| 215 | int rb_bufsz = order_base_2(ih->ring_size / 4); |
| 216 | |
| 217 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, |
| 218 | MC_SPACE, ih->use_bus_addr ? 1 : 4); |
| 219 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, |
| 220 | WPTR_OVERFLOW_CLEAR, 1); |
| 221 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, |
| 222 | WPTR_OVERFLOW_ENABLE, 1); |
| 223 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_SIZE, rb_bufsz); |
| 224 | /* Ring Buffer write pointer writeback. If enabled, IH_RB_WPTR register |
| 225 | * value is written to memory |
| 226 | */ |
| 227 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, |
| 228 | WPTR_WRITEBACK_ENABLE, 1); |
| 229 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SNOOP, 1); |
| 230 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_RO, 0); |
| 231 | ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_VMID, 0); |
| 232 | |
| 233 | return ih_rb_cntl; |
| 234 | } |
| 235 | |
| 236 | static uint32_t navi10_ih_doorbell_rptr(struct amdgpu_ih_ring *ih) |
| 237 | { |
| 238 | u32 ih_doorbell_rtpr = 0; |
| 239 | |
| 240 | if (ih->use_doorbell) { |
| 241 | ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, |
| 242 | IH_DOORBELL_RPTR, OFFSET, |
| 243 | ih->doorbell_index); |
| 244 | ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, |
| 245 | IH_DOORBELL_RPTR, |
| 246 | ENABLE, 1); |
| 247 | } else { |
| 248 | ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, |
| 249 | IH_DOORBELL_RPTR, |
| 250 | ENABLE, 0); |
| 251 | } |
| 252 | return ih_doorbell_rtpr; |
| 253 | } |
| 254 | |
| 255 | /** |
| 256 | * navi10_ih_enable_ring - enable an ih ring buffer |
| 257 | * |
| 258 | * @adev: amdgpu_device pointer |
| 259 | * @ih: amdgpu_ih_ring pointer |
| 260 | * |
| 261 | * Enable an ih ring buffer (NAVI10) |
| 262 | */ |
| 263 | static int navi10_ih_enable_ring(struct amdgpu_device *adev, |
| 264 | struct amdgpu_ih_ring *ih) |
| 265 | { |
| 266 | struct amdgpu_ih_regs *ih_regs; |
| 267 | uint32_t tmp; |
| 268 | |
| 269 | ih_regs = &ih->ih_regs; |
| 270 | |
| 271 | /* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/ |
| 272 | WREG32(ih_regs->ih_rb_base, ih->gpu_addr >> 8); |
| 273 | WREG32(ih_regs->ih_rb_base_hi, (ih->gpu_addr >> 40) & 0xff); |
| 274 | |
| 275 | tmp = RREG32(ih_regs->ih_rb_cntl); |
| 276 | tmp = navi10_ih_rb_cntl(ih, ih_rb_cntl: tmp); |
| 277 | if (ih == &adev->irq.ih) |
| 278 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RPTR_REARM, !!adev->irq.msi_enabled); |
| 279 | if (ih == &adev->irq.ih1) |
| 280 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, RB_FULL_DRAIN_ENABLE, 1); |
| 281 | |
| 282 | if (amdgpu_sriov_vf(adev) && amdgpu_sriov_reg_indirect_ih(adev)) { |
| 283 | if (psp_reg_program(psp: &adev->psp, reg: ih_regs->psp_reg_id, value: tmp)) { |
| 284 | DRM_ERROR("PSP program IH_RB_CNTL failed!\n" ); |
| 285 | return -ETIMEDOUT; |
| 286 | } |
| 287 | } else { |
| 288 | WREG32(ih_regs->ih_rb_cntl, tmp); |
| 289 | } |
| 290 | |
| 291 | if (ih == &adev->irq.ih) { |
| 292 | /* set the ih ring 0 writeback address whether it's enabled or not */ |
| 293 | WREG32(ih_regs->ih_rb_wptr_addr_lo, lower_32_bits(ih->wptr_addr)); |
| 294 | WREG32(ih_regs->ih_rb_wptr_addr_hi, upper_32_bits(ih->wptr_addr) & 0xFFFF); |
| 295 | } |
| 296 | |
| 297 | /* set rptr, wptr to 0 */ |
| 298 | WREG32(ih_regs->ih_rb_wptr, 0); |
| 299 | WREG32(ih_regs->ih_rb_rptr, 0); |
| 300 | |
| 301 | WREG32(ih_regs->ih_doorbell_rptr, navi10_ih_doorbell_rptr(ih)); |
| 302 | |
| 303 | return 0; |
| 304 | } |
| 305 | |
| 306 | /** |
| 307 | * navi10_ih_irq_init - init and enable the interrupt ring |
| 308 | * |
| 309 | * @adev: amdgpu_device pointer |
| 310 | * |
| 311 | * Allocate a ring buffer for the interrupt controller, |
| 312 | * enable the RLC, disable interrupts, enable the IH |
| 313 | * ring buffer and enable it (NAVI). |
| 314 | * Called at device load and reume. |
| 315 | * Returns 0 for success, errors for failure. |
| 316 | */ |
| 317 | static int navi10_ih_irq_init(struct amdgpu_device *adev) |
| 318 | { |
| 319 | struct amdgpu_ih_ring *ih[] = {&adev->irq.ih, &adev->irq.ih1, &adev->irq.ih2}; |
| 320 | u32 ih_chicken; |
| 321 | int ret; |
| 322 | int i; |
| 323 | |
| 324 | /* disable irqs */ |
| 325 | ret = navi10_ih_toggle_interrupts(adev, enable: false); |
| 326 | if (ret) |
| 327 | return ret; |
| 328 | |
| 329 | adev->nbio.funcs->ih_control(adev); |
| 330 | |
| 331 | if (unlikely(adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT)) { |
| 332 | if (ih[0]->use_bus_addr) { |
| 333 | switch (amdgpu_ip_version(adev, ip: OSSSYS_HWIP, inst: 0)) { |
| 334 | case IP_VERSION(5, 0, 3): |
| 335 | case IP_VERSION(5, 2, 0): |
| 336 | case IP_VERSION(5, 2, 1): |
| 337 | ih_chicken = RREG32_SOC15(OSSSYS, 0, mmIH_CHICKEN_Sienna_Cichlid); |
| 338 | ih_chicken = REG_SET_FIELD(ih_chicken, |
| 339 | IH_CHICKEN, MC_SPACE_GPA_ENABLE, 1); |
| 340 | WREG32_SOC15(OSSSYS, 0, mmIH_CHICKEN_Sienna_Cichlid, ih_chicken); |
| 341 | break; |
| 342 | default: |
| 343 | ih_chicken = RREG32_SOC15(OSSSYS, 0, mmIH_CHICKEN); |
| 344 | ih_chicken = REG_SET_FIELD(ih_chicken, |
| 345 | IH_CHICKEN, MC_SPACE_GPA_ENABLE, 1); |
| 346 | WREG32_SOC15(OSSSYS, 0, mmIH_CHICKEN, ih_chicken); |
| 347 | break; |
| 348 | } |
| 349 | } |
| 350 | } |
| 351 | |
| 352 | for (i = 0; i < ARRAY_SIZE(ih); i++) { |
| 353 | if (ih[i]->ring_size) { |
| 354 | ret = navi10_ih_enable_ring(adev, ih: ih[i]); |
| 355 | if (ret) |
| 356 | return ret; |
| 357 | } |
| 358 | } |
| 359 | |
| 360 | /* update doorbell range for ih ring 0*/ |
| 361 | adev->nbio.funcs->ih_doorbell_range(adev, ih[0]->use_doorbell, |
| 362 | ih[0]->doorbell_index); |
| 363 | |
| 364 | pci_set_master(dev: adev->pdev); |
| 365 | |
| 366 | /* enable interrupts */ |
| 367 | ret = navi10_ih_toggle_interrupts(adev, enable: true); |
| 368 | if (ret) |
| 369 | return ret; |
| 370 | /* enable wptr force update for self int */ |
| 371 | force_update_wptr_for_self_int(adev, threshold: 0, timeout: 8, enabled: true); |
| 372 | |
| 373 | if (adev->irq.ih_soft.ring_size) |
| 374 | adev->irq.ih_soft.enabled = true; |
| 375 | |
| 376 | return 0; |
| 377 | } |
| 378 | |
| 379 | /** |
| 380 | * navi10_ih_irq_disable - disable interrupts |
| 381 | * |
| 382 | * @adev: amdgpu_device pointer |
| 383 | * |
| 384 | * Disable interrupts on the hw (NAVI10). |
| 385 | */ |
| 386 | static void navi10_ih_irq_disable(struct amdgpu_device *adev) |
| 387 | { |
| 388 | force_update_wptr_for_self_int(adev, threshold: 0, timeout: 8, enabled: false); |
| 389 | navi10_ih_toggle_interrupts(adev, enable: false); |
| 390 | |
| 391 | /* Wait and acknowledge irq */ |
| 392 | mdelay(1); |
| 393 | } |
| 394 | |
| 395 | /** |
| 396 | * navi10_ih_get_wptr - get the IH ring buffer wptr |
| 397 | * |
| 398 | * @adev: amdgpu_device pointer |
| 399 | * @ih: IH ring buffer to fetch wptr |
| 400 | * |
| 401 | * Get the IH ring buffer wptr from either the register |
| 402 | * or the writeback memory buffer (NAVI10). Also check for |
| 403 | * ring buffer overflow and deal with it. |
| 404 | * Returns the value of the wptr. |
| 405 | */ |
| 406 | static u32 navi10_ih_get_wptr(struct amdgpu_device *adev, |
| 407 | struct amdgpu_ih_ring *ih) |
| 408 | { |
| 409 | u32 wptr, tmp; |
| 410 | struct amdgpu_ih_regs *ih_regs; |
| 411 | |
| 412 | if (ih == &adev->irq.ih || ih == &adev->irq.ih_soft) { |
| 413 | /* Only ring0 supports writeback. On other rings fall back |
| 414 | * to register-based code with overflow checking below. |
| 415 | * ih_soft ring doesn't have any backing hardware registers, |
| 416 | * update wptr and return. |
| 417 | */ |
| 418 | wptr = le32_to_cpu(*ih->wptr_cpu); |
| 419 | |
| 420 | if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) |
| 421 | goto out; |
| 422 | } |
| 423 | |
| 424 | ih_regs = &ih->ih_regs; |
| 425 | |
| 426 | /* Double check that the overflow wasn't already cleared. */ |
| 427 | wptr = RREG32_NO_KIQ(ih_regs->ih_rb_wptr); |
| 428 | if (!REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) |
| 429 | goto out; |
| 430 | wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0); |
| 431 | |
| 432 | /* When a ring buffer overflow happen start parsing interrupt |
| 433 | * from the last not overwritten vector (wptr + 32). Hopefully |
| 434 | * this should allow us to catch up. |
| 435 | */ |
| 436 | tmp = (wptr + 32) & ih->ptr_mask; |
| 437 | dev_warn(adev->dev, "%s ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n" , |
| 438 | amdgpu_ih_ring_name(adev, ih), wptr, ih->rptr, tmp); |
| 439 | ih->rptr = tmp; |
| 440 | |
| 441 | tmp = RREG32_NO_KIQ(ih_regs->ih_rb_cntl); |
| 442 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1); |
| 443 | WREG32_NO_KIQ(ih_regs->ih_rb_cntl, tmp); |
| 444 | |
| 445 | /* Unset the CLEAR_OVERFLOW bit immediately so new overflows |
| 446 | * can be detected. |
| 447 | */ |
| 448 | tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 0); |
| 449 | WREG32_NO_KIQ(ih_regs->ih_rb_cntl, tmp); |
| 450 | out: |
| 451 | return (wptr & ih->ptr_mask); |
| 452 | } |
| 453 | |
| 454 | /** |
| 455 | * navi10_ih_irq_rearm - rearm IRQ if lost |
| 456 | * |
| 457 | * @adev: amdgpu_device pointer |
| 458 | * @ih: IH ring to match |
| 459 | * |
| 460 | */ |
| 461 | static void navi10_ih_irq_rearm(struct amdgpu_device *adev, |
| 462 | struct amdgpu_ih_ring *ih) |
| 463 | { |
| 464 | uint32_t v = 0; |
| 465 | uint32_t i = 0; |
| 466 | struct amdgpu_ih_regs *ih_regs; |
| 467 | |
| 468 | ih_regs = &ih->ih_regs; |
| 469 | |
| 470 | /* Rearm IRQ / re-write doorbell if doorbell write is lost */ |
| 471 | for (i = 0; i < MAX_REARM_RETRY; i++) { |
| 472 | v = RREG32_NO_KIQ(ih_regs->ih_rb_rptr); |
| 473 | if ((v < ih->ring_size) && (v != ih->rptr)) |
| 474 | WDOORBELL32(ih->doorbell_index, ih->rptr); |
| 475 | else |
| 476 | break; |
| 477 | } |
| 478 | } |
| 479 | |
| 480 | /** |
| 481 | * navi10_ih_set_rptr - set the IH ring buffer rptr |
| 482 | * |
| 483 | * @adev: amdgpu_device pointer |
| 484 | * |
| 485 | * @ih: IH ring buffer to set rptr |
| 486 | * Set the IH ring buffer rptr. |
| 487 | */ |
| 488 | static void navi10_ih_set_rptr(struct amdgpu_device *adev, |
| 489 | struct amdgpu_ih_ring *ih) |
| 490 | { |
| 491 | struct amdgpu_ih_regs *ih_regs; |
| 492 | |
| 493 | if (ih == &adev->irq.ih_soft) |
| 494 | return; |
| 495 | |
| 496 | if (ih->use_doorbell) { |
| 497 | /* XXX check if swapping is necessary on BE */ |
| 498 | *ih->rptr_cpu = ih->rptr; |
| 499 | WDOORBELL32(ih->doorbell_index, ih->rptr); |
| 500 | |
| 501 | if (amdgpu_sriov_vf(adev)) |
| 502 | navi10_ih_irq_rearm(adev, ih); |
| 503 | } else { |
| 504 | ih_regs = &ih->ih_regs; |
| 505 | WREG32(ih_regs->ih_rb_rptr, ih->rptr); |
| 506 | } |
| 507 | } |
| 508 | |
| 509 | /** |
| 510 | * navi10_ih_self_irq - dispatch work for ring 1 and 2 |
| 511 | * |
| 512 | * @adev: amdgpu_device pointer |
| 513 | * @source: irq source |
| 514 | * @entry: IV with WPTR update |
| 515 | * |
| 516 | * Update the WPTR from the IV and schedule work to handle the entries. |
| 517 | */ |
| 518 | static int navi10_ih_self_irq(struct amdgpu_device *adev, |
| 519 | struct amdgpu_irq_src *source, |
| 520 | struct amdgpu_iv_entry *entry) |
| 521 | { |
| 522 | switch (entry->ring_id) { |
| 523 | case 1: |
| 524 | schedule_work(work: &adev->irq.ih1_work); |
| 525 | break; |
| 526 | case 2: |
| 527 | schedule_work(work: &adev->irq.ih2_work); |
| 528 | break; |
| 529 | default: break; |
| 530 | } |
| 531 | return 0; |
| 532 | } |
| 533 | |
| 534 | static const struct amdgpu_irq_src_funcs navi10_ih_self_irq_funcs = { |
| 535 | .process = navi10_ih_self_irq, |
| 536 | }; |
| 537 | |
| 538 | static void navi10_ih_set_self_irq_funcs(struct amdgpu_device *adev) |
| 539 | { |
| 540 | adev->irq.self_irq.num_types = 0; |
| 541 | adev->irq.self_irq.funcs = &navi10_ih_self_irq_funcs; |
| 542 | } |
| 543 | |
| 544 | static int navi10_ih_early_init(struct amdgpu_ip_block *ip_block) |
| 545 | { |
| 546 | struct amdgpu_device *adev = ip_block->adev; |
| 547 | |
| 548 | navi10_ih_set_interrupt_funcs(adev); |
| 549 | navi10_ih_set_self_irq_funcs(adev); |
| 550 | return 0; |
| 551 | } |
| 552 | |
| 553 | static int navi10_ih_sw_init(struct amdgpu_ip_block *ip_block) |
| 554 | { |
| 555 | int r; |
| 556 | struct amdgpu_device *adev = ip_block->adev; |
| 557 | bool use_bus_addr; |
| 558 | |
| 559 | r = amdgpu_irq_add_id(adev, client_id: SOC15_IH_CLIENTID_IH, src_id: 0, |
| 560 | source: &adev->irq.self_irq); |
| 561 | |
| 562 | if (r) |
| 563 | return r; |
| 564 | |
| 565 | /* use gpu virtual address for ih ring |
| 566 | * until ih_checken is programmed to allow |
| 567 | * use bus address for ih ring by psp bl */ |
| 568 | if ((adev->flags & AMD_IS_APU) || |
| 569 | (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP)) |
| 570 | use_bus_addr = false; |
| 571 | else |
| 572 | use_bus_addr = true; |
| 573 | r = amdgpu_ih_ring_init(adev, ih: &adev->irq.ih, IH_RING_SIZE, use_bus_addr); |
| 574 | if (r) |
| 575 | return r; |
| 576 | |
| 577 | adev->irq.ih.use_doorbell = true; |
| 578 | adev->irq.ih.doorbell_index = adev->doorbell_index.ih << 1; |
| 579 | |
| 580 | adev->irq.ih1.ring_size = 0; |
| 581 | adev->irq.ih2.ring_size = 0; |
| 582 | |
| 583 | /* initialize ih control registers offset */ |
| 584 | navi10_ih_init_register_offset(adev); |
| 585 | |
| 586 | r = amdgpu_ih_ring_init(adev, ih: &adev->irq.ih_soft, IH_SW_RING_SIZE, use_bus_addr: true); |
| 587 | if (r) |
| 588 | return r; |
| 589 | |
| 590 | r = amdgpu_irq_init(adev); |
| 591 | |
| 592 | return r; |
| 593 | } |
| 594 | |
| 595 | static int navi10_ih_sw_fini(struct amdgpu_ip_block *ip_block) |
| 596 | { |
| 597 | struct amdgpu_device *adev = ip_block->adev; |
| 598 | |
| 599 | amdgpu_irq_fini_sw(adev); |
| 600 | |
| 601 | return 0; |
| 602 | } |
| 603 | |
| 604 | static int navi10_ih_hw_init(struct amdgpu_ip_block *ip_block) |
| 605 | { |
| 606 | struct amdgpu_device *adev = ip_block->adev; |
| 607 | |
| 608 | return navi10_ih_irq_init(adev); |
| 609 | } |
| 610 | |
| 611 | static int navi10_ih_hw_fini(struct amdgpu_ip_block *ip_block) |
| 612 | { |
| 613 | navi10_ih_irq_disable(adev: ip_block->adev); |
| 614 | |
| 615 | return 0; |
| 616 | } |
| 617 | |
| 618 | static int navi10_ih_suspend(struct amdgpu_ip_block *ip_block) |
| 619 | { |
| 620 | return navi10_ih_hw_fini(ip_block); |
| 621 | } |
| 622 | |
| 623 | static int navi10_ih_resume(struct amdgpu_ip_block *ip_block) |
| 624 | { |
| 625 | return navi10_ih_hw_init(ip_block); |
| 626 | } |
| 627 | |
| 628 | static bool navi10_ih_is_idle(struct amdgpu_ip_block *ip_block) |
| 629 | { |
| 630 | /* todo */ |
| 631 | return true; |
| 632 | } |
| 633 | |
| 634 | static int navi10_ih_wait_for_idle(struct amdgpu_ip_block *ip_block) |
| 635 | { |
| 636 | /* todo */ |
| 637 | return -ETIMEDOUT; |
| 638 | } |
| 639 | |
| 640 | static int navi10_ih_soft_reset(struct amdgpu_ip_block *ip_block) |
| 641 | { |
| 642 | /* todo */ |
| 643 | return 0; |
| 644 | } |
| 645 | |
| 646 | static void navi10_ih_update_clockgating_state(struct amdgpu_device *adev, |
| 647 | bool enable) |
| 648 | { |
| 649 | uint32_t data, def, field_val; |
| 650 | |
| 651 | if (adev->cg_flags & AMD_CG_SUPPORT_IH_CG) { |
| 652 | def = data = RREG32_SOC15(OSSSYS, 0, mmIH_CLK_CTRL); |
| 653 | field_val = enable ? 0 : 1; |
| 654 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 655 | DBUS_MUX_CLK_SOFT_OVERRIDE, field_val); |
| 656 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 657 | OSSSYS_SHARE_CLK_SOFT_OVERRIDE, field_val); |
| 658 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 659 | LIMIT_SMN_CLK_SOFT_OVERRIDE, field_val); |
| 660 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 661 | DYN_CLK_SOFT_OVERRIDE, field_val); |
| 662 | data = REG_SET_FIELD(data, IH_CLK_CTRL, |
| 663 | REG_CLK_SOFT_OVERRIDE, field_val); |
| 664 | if (def != data) |
| 665 | WREG32_SOC15(OSSSYS, 0, mmIH_CLK_CTRL, data); |
| 666 | } |
| 667 | } |
| 668 | |
| 669 | static int navi10_ih_set_clockgating_state(struct amdgpu_ip_block *ip_block, |
| 670 | enum amd_clockgating_state state) |
| 671 | { |
| 672 | struct amdgpu_device *adev = ip_block->adev; |
| 673 | |
| 674 | navi10_ih_update_clockgating_state(adev, |
| 675 | enable: state == AMD_CG_STATE_GATE); |
| 676 | return 0; |
| 677 | } |
| 678 | |
| 679 | static int navi10_ih_set_powergating_state(struct amdgpu_ip_block *ip_block, |
| 680 | enum amd_powergating_state state) |
| 681 | { |
| 682 | return 0; |
| 683 | } |
| 684 | |
| 685 | static void navi10_ih_get_clockgating_state(struct amdgpu_ip_block *ip_block, u64 *flags) |
| 686 | { |
| 687 | struct amdgpu_device *adev = ip_block->adev; |
| 688 | |
| 689 | if (!RREG32_SOC15(OSSSYS, 0, mmIH_CLK_CTRL)) |
| 690 | *flags |= AMD_CG_SUPPORT_IH_CG; |
| 691 | } |
| 692 | |
| 693 | static const struct amd_ip_funcs navi10_ih_ip_funcs = { |
| 694 | .name = "navi10_ih" , |
| 695 | .early_init = navi10_ih_early_init, |
| 696 | .sw_init = navi10_ih_sw_init, |
| 697 | .sw_fini = navi10_ih_sw_fini, |
| 698 | .hw_init = navi10_ih_hw_init, |
| 699 | .hw_fini = navi10_ih_hw_fini, |
| 700 | .suspend = navi10_ih_suspend, |
| 701 | .resume = navi10_ih_resume, |
| 702 | .is_idle = navi10_ih_is_idle, |
| 703 | .wait_for_idle = navi10_ih_wait_for_idle, |
| 704 | .soft_reset = navi10_ih_soft_reset, |
| 705 | .set_clockgating_state = navi10_ih_set_clockgating_state, |
| 706 | .set_powergating_state = navi10_ih_set_powergating_state, |
| 707 | .get_clockgating_state = navi10_ih_get_clockgating_state, |
| 708 | }; |
| 709 | |
| 710 | static const struct amdgpu_ih_funcs navi10_ih_funcs = { |
| 711 | .get_wptr = navi10_ih_get_wptr, |
| 712 | .decode_iv = amdgpu_ih_decode_iv_helper, |
| 713 | .decode_iv_ts = amdgpu_ih_decode_iv_ts_helper, |
| 714 | .set_rptr = navi10_ih_set_rptr |
| 715 | }; |
| 716 | |
| 717 | static void navi10_ih_set_interrupt_funcs(struct amdgpu_device *adev) |
| 718 | { |
| 719 | if (adev->irq.ih_funcs == NULL) |
| 720 | adev->irq.ih_funcs = &navi10_ih_funcs; |
| 721 | } |
| 722 | |
| 723 | const struct amdgpu_ip_block_version navi10_ih_ip_block = { |
| 724 | .type = AMD_IP_BLOCK_TYPE_IH, |
| 725 | .major = 5, |
| 726 | .minor = 0, |
| 727 | .rev = 0, |
| 728 | .funcs = &navi10_ih_ip_funcs, |
| 729 | }; |
| 730 | |