1 /* 2 * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. 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 (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 * SOFTWARE. 22 * 23 * Authors: 24 * Eddie Dong <eddie.dong@intel.com> 25 * Kevin Tian <kevin.tian@intel.com> 26 * 27 * Contributors: 28 * Zhi Wang <zhi.a.wang@intel.com> 29 * Changbin Du <changbin.du@intel.com> 30 * Zhenyu Wang <zhenyuw@linux.intel.com> 31 * Tina Zhang <tina.zhang@intel.com> 32 * Bing Niu <bing.niu@intel.com> 33 * 34 */ 35 36 #include "i915_drv.h" 37 #include "gvt.h" 38 #include "trace.h" 39 40 #define GEN9_MOCS_SIZE 64 41 42 /* Raw offset is appened to each line for convenience. */ 43 static struct engine_mmio gen8_engine_mmio_list[] __cacheline_aligned = { 44 {RCS0, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */ 45 {RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */ 46 {RCS0, HWSTAM, 0x0, false}, /* 0x2098 */ 47 {RCS0, INSTPM, 0xffff, true}, /* 0x20c0 */ 48 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */ 49 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */ 50 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */ 51 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */ 52 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */ 53 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */ 54 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */ 55 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */ 56 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */ 57 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */ 58 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */ 59 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */ 60 {RCS0, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */ 61 {RCS0, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */ 62 {RCS0, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */ 63 {RCS0, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */ 64 {RCS0, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */ 65 {RCS0, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */ 66 67 {BCS0, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */ 68 {BCS0, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */ 69 {BCS0, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */ 70 {BCS0, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */ 71 {BCS0, RING_EXCC(BLT_RING_BASE), 0x0, false}, /* 0x22028 */ 72 {RCS0, INVALID_MMIO_REG, 0, false } /* Terminated */ 73 }; 74 75 static struct engine_mmio gen9_engine_mmio_list[] __cacheline_aligned = { 76 {RCS0, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */ 77 {RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */ 78 {RCS0, HWSTAM, 0x0, false}, /* 0x2098 */ 79 {RCS0, INSTPM, 0xffff, true}, /* 0x20c0 */ 80 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 0), 0, false}, /* 0x24d0 */ 81 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 1), 0, false}, /* 0x24d4 */ 82 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 2), 0, false}, /* 0x24d8 */ 83 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 3), 0, false}, /* 0x24dc */ 84 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 4), 0, false}, /* 0x24e0 */ 85 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 5), 0, false}, /* 0x24e4 */ 86 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 6), 0, false}, /* 0x24e8 */ 87 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 7), 0, false}, /* 0x24ec */ 88 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 8), 0, false}, /* 0x24f0 */ 89 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 9), 0, false}, /* 0x24f4 */ 90 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 10), 0, false}, /* 0x24f8 */ 91 {RCS0, RING_FORCE_TO_NONPRIV(RENDER_RING_BASE, 11), 0, false}, /* 0x24fc */ 92 {RCS0, CACHE_MODE_1, 0xffff, true}, /* 0x7004 */ 93 {RCS0, GEN7_GT_MODE, 0xffff, true}, /* 0x7008 */ 94 {RCS0, CACHE_MODE_0_GEN7, 0xffff, true}, /* 0x7000 */ 95 {RCS0, GEN7_COMMON_SLICE_CHICKEN1, 0xffff, true}, /* 0x7010 */ 96 {RCS0, HDC_CHICKEN0, 0xffff, true}, /* 0x7300 */ 97 {RCS0, VF_GUARDBAND, 0xffff, true}, /* 0x83a4 */ 98 99 {RCS0, GEN8_PRIVATE_PAT_LO, 0, false}, /* 0x40e0 */ 100 {RCS0, GEN8_PRIVATE_PAT_HI, 0, false}, /* 0x40e4 */ 101 {RCS0, GEN8_CS_CHICKEN1, 0xffff, true}, /* 0x2580 */ 102 {RCS0, COMMON_SLICE_CHICKEN2, 0xffff, true}, /* 0x7014 */ 103 {RCS0, GEN9_CS_DEBUG_MODE1, 0xffff, false}, /* 0x20ec */ 104 {RCS0, GEN8_L3SQCREG4, 0, false}, /* 0xb118 */ 105 {RCS0, GEN7_HALF_SLICE_CHICKEN1, 0xffff, true}, /* 0xe100 */ 106 {RCS0, HALF_SLICE_CHICKEN2, 0xffff, true}, /* 0xe180 */ 107 {RCS0, HALF_SLICE_CHICKEN3, 0xffff, true}, /* 0xe184 */ 108 {RCS0, GEN9_HALF_SLICE_CHICKEN5, 0xffff, true}, /* 0xe188 */ 109 {RCS0, GEN9_HALF_SLICE_CHICKEN7, 0xffff, true}, /* 0xe194 */ 110 {RCS0, GEN8_ROW_CHICKEN, 0xffff, true}, /* 0xe4f0 */ 111 {RCS0, TRVATTL3PTRDW(0), 0, false}, /* 0x4de0 */ 112 {RCS0, TRVATTL3PTRDW(1), 0, false}, /* 0x4de4 */ 113 {RCS0, TRNULLDETCT, 0, false}, /* 0x4de8 */ 114 {RCS0, TRINVTILEDETCT, 0, false}, /* 0x4dec */ 115 {RCS0, TRVADR, 0, false}, /* 0x4df0 */ 116 {RCS0, TRTTE, 0, false}, /* 0x4df4 */ 117 118 {BCS0, RING_GFX_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2229c */ 119 {BCS0, RING_MI_MODE(BLT_RING_BASE), 0xffff, false}, /* 0x2209c */ 120 {BCS0, RING_INSTPM(BLT_RING_BASE), 0xffff, false}, /* 0x220c0 */ 121 {BCS0, RING_HWSTAM(BLT_RING_BASE), 0x0, false}, /* 0x22098 */ 122 {BCS0, RING_EXCC(BLT_RING_BASE), 0x0, false}, /* 0x22028 */ 123 124 {VCS1, RING_EXCC(GEN8_BSD2_RING_BASE), 0xffff, false}, /* 0x1c028 */ 125 126 {VECS0, RING_EXCC(VEBOX_RING_BASE), 0xffff, false}, /* 0x1a028 */ 127 128 {RCS0, GEN8_HDC_CHICKEN1, 0xffff, true}, /* 0x7304 */ 129 {RCS0, GEN9_CTX_PREEMPT_REG, 0x0, false}, /* 0x2248 */ 130 {RCS0, GEN7_UCGCTL4, 0x0, false}, /* 0x940c */ 131 {RCS0, GAMT_CHKN_BIT_REG, 0x0, false}, /* 0x4ab8 */ 132 133 {RCS0, GEN9_GAMT_ECO_REG_RW_IA, 0x0, false}, /* 0x4ab0 */ 134 {RCS0, GEN9_CSFE_CHICKEN1_RCS, 0xffff, false}, /* 0x20d4 */ 135 136 {RCS0, GEN8_GARBCNTL, 0x0, false}, /* 0xb004 */ 137 {RCS0, GEN7_FF_THREAD_MODE, 0x0, false}, /* 0x20a0 */ 138 {RCS0, FF_SLICE_CS_CHICKEN2, 0xffff, false}, /* 0x20e4 */ 139 {RCS0, INVALID_MMIO_REG, 0, false } /* Terminated */ 140 }; 141 142 static struct { 143 bool initialized; 144 u32 control_table[I915_NUM_ENGINES][GEN9_MOCS_SIZE]; 145 u32 l3cc_table[GEN9_MOCS_SIZE / 2]; 146 } gen9_render_mocs; 147 148 static void load_render_mocs(struct drm_i915_private *dev_priv) 149 { 150 i915_reg_t offset; 151 u32 regs[] = { 152 [RCS0] = 0xc800, 153 [VCS0] = 0xc900, 154 [VCS1] = 0xca00, 155 [BCS0] = 0xcc00, 156 [VECS0] = 0xcb00, 157 }; 158 int ring_id, i; 159 160 for (ring_id = 0; ring_id < ARRAY_SIZE(regs); ring_id++) { 161 if (!HAS_ENGINE(dev_priv, ring_id)) 162 continue; 163 offset.reg = regs[ring_id]; 164 for (i = 0; i < GEN9_MOCS_SIZE; i++) { 165 gen9_render_mocs.control_table[ring_id][i] = 166 I915_READ_FW(offset); 167 offset.reg += 4; 168 } 169 } 170 171 offset.reg = 0xb020; 172 for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) { 173 gen9_render_mocs.l3cc_table[i] = 174 I915_READ_FW(offset); 175 offset.reg += 4; 176 } 177 gen9_render_mocs.initialized = true; 178 } 179 180 static int 181 restore_context_mmio_for_inhibit(struct intel_vgpu *vgpu, 182 struct i915_request *req) 183 { 184 u32 *cs; 185 int ret; 186 struct engine_mmio *mmio; 187 struct intel_gvt *gvt = vgpu->gvt; 188 int ring_id = req->engine->id; 189 int count = gvt->engine_mmio_list.ctx_mmio_count[ring_id]; 190 191 if (count == 0) 192 return 0; 193 194 ret = req->engine->emit_flush(req, EMIT_BARRIER); 195 if (ret) 196 return ret; 197 198 cs = intel_ring_begin(req, count * 2 + 2); 199 if (IS_ERR(cs)) 200 return PTR_ERR(cs); 201 202 *cs++ = MI_LOAD_REGISTER_IMM(count); 203 for (mmio = gvt->engine_mmio_list.mmio; 204 i915_mmio_reg_valid(mmio->reg); mmio++) { 205 if (mmio->ring_id != ring_id || 206 !mmio->in_context) 207 continue; 208 209 *cs++ = i915_mmio_reg_offset(mmio->reg); 210 *cs++ = vgpu_vreg_t(vgpu, mmio->reg) | 211 (mmio->mask << 16); 212 gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n", 213 *(cs-2), *(cs-1), vgpu->id, ring_id); 214 } 215 216 *cs++ = MI_NOOP; 217 intel_ring_advance(req, cs); 218 219 ret = req->engine->emit_flush(req, EMIT_BARRIER); 220 if (ret) 221 return ret; 222 223 return 0; 224 } 225 226 static int 227 restore_render_mocs_control_for_inhibit(struct intel_vgpu *vgpu, 228 struct i915_request *req) 229 { 230 unsigned int index; 231 u32 *cs; 232 233 cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE + 2); 234 if (IS_ERR(cs)) 235 return PTR_ERR(cs); 236 237 *cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE); 238 239 for (index = 0; index < GEN9_MOCS_SIZE; index++) { 240 *cs++ = i915_mmio_reg_offset(GEN9_GFX_MOCS(index)); 241 *cs++ = vgpu_vreg_t(vgpu, GEN9_GFX_MOCS(index)); 242 gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n", 243 *(cs-2), *(cs-1), vgpu->id, req->engine->id); 244 245 } 246 247 *cs++ = MI_NOOP; 248 intel_ring_advance(req, cs); 249 250 return 0; 251 } 252 253 static int 254 restore_render_mocs_l3cc_for_inhibit(struct intel_vgpu *vgpu, 255 struct i915_request *req) 256 { 257 unsigned int index; 258 u32 *cs; 259 260 cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE / 2 + 2); 261 if (IS_ERR(cs)) 262 return PTR_ERR(cs); 263 264 *cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE / 2); 265 266 for (index = 0; index < GEN9_MOCS_SIZE / 2; index++) { 267 *cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(index)); 268 *cs++ = vgpu_vreg_t(vgpu, GEN9_LNCFCMOCS(index)); 269 gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n", 270 *(cs-2), *(cs-1), vgpu->id, req->engine->id); 271 272 } 273 274 *cs++ = MI_NOOP; 275 intel_ring_advance(req, cs); 276 277 return 0; 278 } 279 280 /* 281 * Use lri command to initialize the mmio which is in context state image for 282 * inhibit context, it contains tracked engine mmio, render_mocs and 283 * render_mocs_l3cc. 284 */ 285 int intel_vgpu_restore_inhibit_context(struct intel_vgpu *vgpu, 286 struct i915_request *req) 287 { 288 int ret; 289 u32 *cs; 290 291 cs = intel_ring_begin(req, 2); 292 if (IS_ERR(cs)) 293 return PTR_ERR(cs); 294 295 *cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE; 296 *cs++ = MI_NOOP; 297 intel_ring_advance(req, cs); 298 299 ret = restore_context_mmio_for_inhibit(vgpu, req); 300 if (ret) 301 goto out; 302 303 /* no MOCS register in context except render engine */ 304 if (req->engine->id != RCS0) 305 goto out; 306 307 ret = restore_render_mocs_control_for_inhibit(vgpu, req); 308 if (ret) 309 goto out; 310 311 ret = restore_render_mocs_l3cc_for_inhibit(vgpu, req); 312 if (ret) 313 goto out; 314 315 out: 316 cs = intel_ring_begin(req, 2); 317 if (IS_ERR(cs)) 318 return PTR_ERR(cs); 319 320 *cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE; 321 *cs++ = MI_NOOP; 322 intel_ring_advance(req, cs); 323 324 return ret; 325 } 326 327 static void handle_tlb_pending_event(struct intel_vgpu *vgpu, int ring_id) 328 { 329 struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; 330 struct intel_uncore *uncore = &dev_priv->uncore; 331 struct intel_vgpu_submission *s = &vgpu->submission; 332 enum forcewake_domains fw; 333 i915_reg_t reg; 334 u32 regs[] = { 335 [RCS0] = 0x4260, 336 [VCS0] = 0x4264, 337 [VCS1] = 0x4268, 338 [BCS0] = 0x426c, 339 [VECS0] = 0x4270, 340 }; 341 342 if (WARN_ON(ring_id >= ARRAY_SIZE(regs))) 343 return; 344 345 if (!test_and_clear_bit(ring_id, (void *)s->tlb_handle_pending)) 346 return; 347 348 reg = _MMIO(regs[ring_id]); 349 350 /* WaForceWakeRenderDuringMmioTLBInvalidate:skl 351 * we need to put a forcewake when invalidating RCS TLB caches, 352 * otherwise device can go to RC6 state and interrupt invalidation 353 * process 354 */ 355 fw = intel_uncore_forcewake_for_reg(uncore, reg, 356 FW_REG_READ | FW_REG_WRITE); 357 if (ring_id == RCS0 && INTEL_GEN(dev_priv) >= 9) 358 fw |= FORCEWAKE_RENDER; 359 360 intel_uncore_forcewake_get(uncore, fw); 361 362 intel_uncore_write_fw(uncore, reg, 0x1); 363 364 if (wait_for_atomic((intel_uncore_read_fw(uncore, reg) == 0), 50)) 365 gvt_vgpu_err("timeout in invalidate ring (%d) tlb\n", ring_id); 366 else 367 vgpu_vreg_t(vgpu, reg) = 0; 368 369 intel_uncore_forcewake_put(uncore, fw); 370 371 gvt_dbg_core("invalidate TLB for ring %d\n", ring_id); 372 } 373 374 static void switch_mocs(struct intel_vgpu *pre, struct intel_vgpu *next, 375 int ring_id) 376 { 377 struct drm_i915_private *dev_priv; 378 i915_reg_t offset, l3_offset; 379 u32 old_v, new_v; 380 381 u32 regs[] = { 382 [RCS0] = 0xc800, 383 [VCS0] = 0xc900, 384 [VCS1] = 0xca00, 385 [BCS0] = 0xcc00, 386 [VECS0] = 0xcb00, 387 }; 388 int i; 389 390 dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv; 391 if (WARN_ON(ring_id >= ARRAY_SIZE(regs))) 392 return; 393 394 if (ring_id == RCS0 && 395 (IS_KABYLAKE(dev_priv) || 396 IS_BROXTON(dev_priv) || 397 IS_COFFEELAKE(dev_priv))) 398 return; 399 400 if (!pre && !gen9_render_mocs.initialized) 401 load_render_mocs(dev_priv); 402 403 offset.reg = regs[ring_id]; 404 for (i = 0; i < GEN9_MOCS_SIZE; i++) { 405 if (pre) 406 old_v = vgpu_vreg_t(pre, offset); 407 else 408 old_v = gen9_render_mocs.control_table[ring_id][i]; 409 if (next) 410 new_v = vgpu_vreg_t(next, offset); 411 else 412 new_v = gen9_render_mocs.control_table[ring_id][i]; 413 414 if (old_v != new_v) 415 I915_WRITE_FW(offset, new_v); 416 417 offset.reg += 4; 418 } 419 420 if (ring_id == RCS0) { 421 l3_offset.reg = 0xb020; 422 for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) { 423 if (pre) 424 old_v = vgpu_vreg_t(pre, l3_offset); 425 else 426 old_v = gen9_render_mocs.l3cc_table[i]; 427 if (next) 428 new_v = vgpu_vreg_t(next, l3_offset); 429 else 430 new_v = gen9_render_mocs.l3cc_table[i]; 431 432 if (old_v != new_v) 433 I915_WRITE_FW(l3_offset, new_v); 434 435 l3_offset.reg += 4; 436 } 437 } 438 } 439 440 #define CTX_CONTEXT_CONTROL_VAL 0x03 441 442 bool is_inhibit_context(struct intel_context *ce) 443 { 444 const u32 *reg_state = ce->lrc_reg_state; 445 u32 inhibit_mask = 446 _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT); 447 448 return inhibit_mask == 449 (reg_state[CTX_CONTEXT_CONTROL_VAL] & inhibit_mask); 450 } 451 452 /* Switch ring mmio values (context). */ 453 static void switch_mmio(struct intel_vgpu *pre, 454 struct intel_vgpu *next, 455 int ring_id) 456 { 457 struct drm_i915_private *dev_priv; 458 struct intel_vgpu_submission *s; 459 struct engine_mmio *mmio; 460 u32 old_v, new_v; 461 462 dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv; 463 if (INTEL_GEN(dev_priv) >= 9) 464 switch_mocs(pre, next, ring_id); 465 466 for (mmio = dev_priv->gvt->engine_mmio_list.mmio; 467 i915_mmio_reg_valid(mmio->reg); mmio++) { 468 if (mmio->ring_id != ring_id) 469 continue; 470 /* 471 * No need to do save or restore of the mmio which is in context 472 * state image on kabylake, it's initialized by lri command and 473 * save or restore with context together. 474 */ 475 if ((IS_KABYLAKE(dev_priv) || IS_BROXTON(dev_priv) 476 || IS_COFFEELAKE(dev_priv)) && mmio->in_context) 477 continue; 478 479 // save 480 if (pre) { 481 vgpu_vreg_t(pre, mmio->reg) = I915_READ_FW(mmio->reg); 482 if (mmio->mask) 483 vgpu_vreg_t(pre, mmio->reg) &= 484 ~(mmio->mask << 16); 485 old_v = vgpu_vreg_t(pre, mmio->reg); 486 } else 487 old_v = mmio->value = I915_READ_FW(mmio->reg); 488 489 // restore 490 if (next) { 491 s = &next->submission; 492 /* 493 * No need to restore the mmio which is in context state 494 * image if it's not inhibit context, it will restore 495 * itself. 496 */ 497 if (mmio->in_context && 498 !is_inhibit_context(intel_context_lookup(s->shadow_ctx, 499 dev_priv->engine[ring_id]))) 500 continue; 501 502 if (mmio->mask) 503 new_v = vgpu_vreg_t(next, mmio->reg) | 504 (mmio->mask << 16); 505 else 506 new_v = vgpu_vreg_t(next, mmio->reg); 507 } else { 508 if (mmio->in_context) 509 continue; 510 if (mmio->mask) 511 new_v = mmio->value | (mmio->mask << 16); 512 else 513 new_v = mmio->value; 514 } 515 516 I915_WRITE_FW(mmio->reg, new_v); 517 518 trace_render_mmio(pre ? pre->id : 0, 519 next ? next->id : 0, 520 "switch", 521 i915_mmio_reg_offset(mmio->reg), 522 old_v, new_v); 523 } 524 525 if (next) 526 handle_tlb_pending_event(next, ring_id); 527 } 528 529 /** 530 * intel_gvt_switch_render_mmio - switch mmio context of specific engine 531 * @pre: the last vGPU that own the engine 532 * @next: the vGPU to switch to 533 * @ring_id: specify the engine 534 * 535 * If pre is null indicates that host own the engine. If next is null 536 * indicates that we are switching to host workload. 537 */ 538 void intel_gvt_switch_mmio(struct intel_vgpu *pre, 539 struct intel_vgpu *next, int ring_id) 540 { 541 struct drm_i915_private *dev_priv; 542 543 if (WARN_ON(!pre && !next)) 544 return; 545 546 gvt_dbg_render("switch ring %d from %s to %s\n", ring_id, 547 pre ? "vGPU" : "host", next ? "vGPU" : "HOST"); 548 549 dev_priv = pre ? pre->gvt->dev_priv : next->gvt->dev_priv; 550 551 /** 552 * We are using raw mmio access wrapper to improve the 553 * performace for batch mmio read/write, so we need 554 * handle forcewake mannually. 555 */ 556 intel_uncore_forcewake_get(&dev_priv->uncore, FORCEWAKE_ALL); 557 switch_mmio(pre, next, ring_id); 558 intel_uncore_forcewake_put(&dev_priv->uncore, FORCEWAKE_ALL); 559 } 560 561 /** 562 * intel_gvt_init_engine_mmio_context - Initiate the engine mmio list 563 * @gvt: GVT device 564 * 565 */ 566 void intel_gvt_init_engine_mmio_context(struct intel_gvt *gvt) 567 { 568 struct engine_mmio *mmio; 569 570 if (INTEL_GEN(gvt->dev_priv) >= 9) 571 gvt->engine_mmio_list.mmio = gen9_engine_mmio_list; 572 else 573 gvt->engine_mmio_list.mmio = gen8_engine_mmio_list; 574 575 for (mmio = gvt->engine_mmio_list.mmio; 576 i915_mmio_reg_valid(mmio->reg); mmio++) { 577 if (mmio->in_context) { 578 gvt->engine_mmio_list.ctx_mmio_count[mmio->ring_id]++; 579 intel_gvt_mmio_set_in_ctx(gvt, mmio->reg.reg); 580 } 581 } 582 } 583