Chris Wilson | 94b4f3b | 2016-07-05 10:40:20 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright © 2016 Intel Corporation |
| 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 |
| 20 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS |
| 21 | * IN THE SOFTWARE. |
| 22 | * |
| 23 | */ |
| 24 | |
| 25 | #include "i915_drv.h" |
| 26 | |
| 27 | void intel_device_info_dump(struct drm_i915_private *dev_priv) |
| 28 | { |
| 29 | const struct intel_device_info *info = &dev_priv->info; |
| 30 | |
| 31 | #define PRINT_S(name) "%s" |
| 32 | #define SEP_EMPTY |
| 33 | #define PRINT_FLAG(name) info->name ? #name "," : "" |
| 34 | #define SEP_COMMA , |
| 35 | DRM_DEBUG_DRIVER("i915 device info: gen=%i, pciid=0x%04x rev=0x%02x flags=" |
| 36 | DEV_INFO_FOR_EACH_FLAG(PRINT_S, SEP_EMPTY), |
| 37 | info->gen, |
| 38 | dev_priv->drm.pdev->device, |
| 39 | dev_priv->drm.pdev->revision, |
| 40 | DEV_INFO_FOR_EACH_FLAG(PRINT_FLAG, SEP_COMMA)); |
| 41 | #undef PRINT_S |
| 42 | #undef SEP_EMPTY |
| 43 | #undef PRINT_FLAG |
| 44 | #undef SEP_COMMA |
| 45 | } |
| 46 | |
| 47 | static void cherryview_sseu_info_init(struct drm_i915_private *dev_priv) |
| 48 | { |
| 49 | struct intel_device_info *info = mkwrite_device_info(dev_priv); |
| 50 | u32 fuse, eu_dis; |
| 51 | |
| 52 | fuse = I915_READ(CHV_FUSE_GT); |
| 53 | |
| 54 | info->slice_total = 1; |
| 55 | |
| 56 | if (!(fuse & CHV_FGT_DISABLE_SS0)) { |
| 57 | info->subslice_per_slice++; |
| 58 | eu_dis = fuse & (CHV_FGT_EU_DIS_SS0_R0_MASK | |
| 59 | CHV_FGT_EU_DIS_SS0_R1_MASK); |
| 60 | info->eu_total += 8 - hweight32(eu_dis); |
| 61 | } |
| 62 | |
| 63 | if (!(fuse & CHV_FGT_DISABLE_SS1)) { |
| 64 | info->subslice_per_slice++; |
| 65 | eu_dis = fuse & (CHV_FGT_EU_DIS_SS1_R0_MASK | |
| 66 | CHV_FGT_EU_DIS_SS1_R1_MASK); |
| 67 | info->eu_total += 8 - hweight32(eu_dis); |
| 68 | } |
| 69 | |
| 70 | info->subslice_total = info->subslice_per_slice; |
| 71 | /* |
| 72 | * CHV expected to always have a uniform distribution of EU |
| 73 | * across subslices. |
| 74 | */ |
| 75 | info->eu_per_subslice = info->subslice_total ? |
| 76 | info->eu_total / info->subslice_total : |
| 77 | 0; |
| 78 | /* |
| 79 | * CHV supports subslice power gating on devices with more than |
| 80 | * one subslice, and supports EU power gating on devices with |
| 81 | * more than one EU pair per subslice. |
| 82 | */ |
| 83 | info->has_slice_pg = 0; |
| 84 | info->has_subslice_pg = (info->subslice_total > 1); |
| 85 | info->has_eu_pg = (info->eu_per_subslice > 2); |
| 86 | } |
| 87 | |
| 88 | static void gen9_sseu_info_init(struct drm_i915_private *dev_priv) |
| 89 | { |
| 90 | struct intel_device_info *info = mkwrite_device_info(dev_priv); |
| 91 | int s_max = 3, ss_max = 4, eu_max = 8; |
| 92 | int s, ss; |
| 93 | u32 fuse2, s_enable, ss_disable, eu_disable; |
| 94 | u8 eu_mask = 0xff; |
| 95 | |
| 96 | fuse2 = I915_READ(GEN8_FUSE2); |
| 97 | s_enable = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT; |
| 98 | ss_disable = (fuse2 & GEN9_F2_SS_DIS_MASK) >> GEN9_F2_SS_DIS_SHIFT; |
| 99 | |
| 100 | info->slice_total = hweight32(s_enable); |
| 101 | /* |
| 102 | * The subslice disable field is global, i.e. it applies |
| 103 | * to each of the enabled slices. |
| 104 | */ |
| 105 | info->subslice_per_slice = ss_max - hweight32(ss_disable); |
| 106 | info->subslice_total = info->slice_total * info->subslice_per_slice; |
| 107 | |
| 108 | /* |
| 109 | * Iterate through enabled slices and subslices to |
| 110 | * count the total enabled EU. |
| 111 | */ |
| 112 | for (s = 0; s < s_max; s++) { |
| 113 | if (!(s_enable & BIT(s))) |
| 114 | /* skip disabled slice */ |
| 115 | continue; |
| 116 | |
| 117 | eu_disable = I915_READ(GEN9_EU_DISABLE(s)); |
| 118 | for (ss = 0; ss < ss_max; ss++) { |
| 119 | int eu_per_ss; |
| 120 | |
| 121 | if (ss_disable & BIT(ss)) |
| 122 | /* skip disabled subslice */ |
| 123 | continue; |
| 124 | |
| 125 | eu_per_ss = eu_max - hweight8((eu_disable >> (ss*8)) & |
| 126 | eu_mask); |
| 127 | |
| 128 | /* |
| 129 | * Record which subslice(s) has(have) 7 EUs. we |
| 130 | * can tune the hash used to spread work among |
| 131 | * subslices if they are unbalanced. |
| 132 | */ |
| 133 | if (eu_per_ss == 7) |
| 134 | info->subslice_7eu[s] |= BIT(ss); |
| 135 | |
| 136 | info->eu_total += eu_per_ss; |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | /* |
| 141 | * SKL is expected to always have a uniform distribution |
| 142 | * of EU across subslices with the exception that any one |
| 143 | * EU in any one subslice may be fused off for die |
| 144 | * recovery. BXT is expected to be perfectly uniform in EU |
| 145 | * distribution. |
| 146 | */ |
| 147 | info->eu_per_subslice = info->subslice_total ? |
| 148 | DIV_ROUND_UP(info->eu_total, |
| 149 | info->subslice_total) : 0; |
| 150 | /* |
| 151 | * SKL supports slice power gating on devices with more than |
| 152 | * one slice, and supports EU power gating on devices with |
| 153 | * more than one EU pair per subslice. BXT supports subslice |
| 154 | * power gating on devices with more than one subslice, and |
| 155 | * supports EU power gating on devices with more than one EU |
| 156 | * pair per subslice. |
| 157 | */ |
| 158 | info->has_slice_pg = |
| 159 | (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) && |
| 160 | info->slice_total > 1; |
| 161 | info->has_subslice_pg = |
| 162 | IS_BROXTON(dev_priv) && info->subslice_total > 1; |
| 163 | info->has_eu_pg = info->eu_per_subslice > 2; |
| 164 | |
| 165 | if (IS_BROXTON(dev_priv)) { |
| 166 | #define IS_SS_DISABLED(_ss_disable, ss) (_ss_disable & BIT(ss)) |
| 167 | /* |
| 168 | * There is a HW issue in 2x6 fused down parts that requires |
| 169 | * Pooled EU to be enabled as a WA. The pool configuration |
| 170 | * changes depending upon which subslice is fused down. This |
| 171 | * doesn't affect if the device has all 3 subslices enabled. |
| 172 | */ |
| 173 | /* WaEnablePooledEuFor2x6:bxt */ |
| 174 | info->has_pooled_eu = ((info->subslice_per_slice == 3) || |
| 175 | (info->subslice_per_slice == 2 && |
| 176 | INTEL_REVID(dev_priv) < BXT_REVID_C0)); |
| 177 | |
| 178 | info->min_eu_in_pool = 0; |
| 179 | if (info->has_pooled_eu) { |
| 180 | if (IS_SS_DISABLED(ss_disable, 0) || |
| 181 | IS_SS_DISABLED(ss_disable, 2)) |
| 182 | info->min_eu_in_pool = 3; |
| 183 | else if (IS_SS_DISABLED(ss_disable, 1)) |
| 184 | info->min_eu_in_pool = 6; |
| 185 | else |
| 186 | info->min_eu_in_pool = 9; |
| 187 | } |
| 188 | #undef IS_SS_DISABLED |
| 189 | } |
| 190 | } |
| 191 | |
| 192 | static void broadwell_sseu_info_init(struct drm_i915_private *dev_priv) |
| 193 | { |
| 194 | struct intel_device_info *info = mkwrite_device_info(dev_priv); |
| 195 | const int s_max = 3, ss_max = 3, eu_max = 8; |
| 196 | int s, ss; |
| 197 | u32 fuse2, eu_disable[s_max], s_enable, ss_disable; |
| 198 | |
| 199 | fuse2 = I915_READ(GEN8_FUSE2); |
| 200 | s_enable = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT; |
| 201 | ss_disable = (fuse2 & GEN8_F2_SS_DIS_MASK) >> GEN8_F2_SS_DIS_SHIFT; |
| 202 | |
| 203 | eu_disable[0] = I915_READ(GEN8_EU_DISABLE0) & GEN8_EU_DIS0_S0_MASK; |
| 204 | eu_disable[1] = (I915_READ(GEN8_EU_DISABLE0) >> GEN8_EU_DIS0_S1_SHIFT) | |
| 205 | ((I915_READ(GEN8_EU_DISABLE1) & GEN8_EU_DIS1_S1_MASK) << |
| 206 | (32 - GEN8_EU_DIS0_S1_SHIFT)); |
| 207 | eu_disable[2] = (I915_READ(GEN8_EU_DISABLE1) >> GEN8_EU_DIS1_S2_SHIFT) | |
| 208 | ((I915_READ(GEN8_EU_DISABLE2) & GEN8_EU_DIS2_S2_MASK) << |
| 209 | (32 - GEN8_EU_DIS1_S2_SHIFT)); |
| 210 | |
| 211 | info->slice_total = hweight32(s_enable); |
| 212 | |
| 213 | /* |
| 214 | * The subslice disable field is global, i.e. it applies |
| 215 | * to each of the enabled slices. |
| 216 | */ |
| 217 | info->subslice_per_slice = ss_max - hweight32(ss_disable); |
| 218 | info->subslice_total = info->slice_total * info->subslice_per_slice; |
| 219 | |
| 220 | /* |
| 221 | * Iterate through enabled slices and subslices to |
| 222 | * count the total enabled EU. |
| 223 | */ |
| 224 | for (s = 0; s < s_max; s++) { |
| 225 | if (!(s_enable & (0x1 << s))) |
| 226 | /* skip disabled slice */ |
| 227 | continue; |
| 228 | |
| 229 | for (ss = 0; ss < ss_max; ss++) { |
| 230 | u32 n_disabled; |
| 231 | |
| 232 | if (ss_disable & (0x1 << ss)) |
| 233 | /* skip disabled subslice */ |
| 234 | continue; |
| 235 | |
| 236 | n_disabled = hweight8(eu_disable[s] >> (ss * eu_max)); |
| 237 | |
| 238 | /* |
| 239 | * Record which subslices have 7 EUs. |
| 240 | */ |
| 241 | if (eu_max - n_disabled == 7) |
| 242 | info->subslice_7eu[s] |= 1 << ss; |
| 243 | |
| 244 | info->eu_total += eu_max - n_disabled; |
| 245 | } |
| 246 | } |
| 247 | |
| 248 | /* |
| 249 | * BDW is expected to always have a uniform distribution of EU across |
| 250 | * subslices with the exception that any one EU in any one subslice may |
| 251 | * be fused off for die recovery. |
| 252 | */ |
| 253 | info->eu_per_subslice = info->subslice_total ? |
| 254 | DIV_ROUND_UP(info->eu_total, info->subslice_total) : 0; |
| 255 | |
| 256 | /* |
| 257 | * BDW supports slice power gating on devices with more than |
| 258 | * one slice. |
| 259 | */ |
| 260 | info->has_slice_pg = (info->slice_total > 1); |
| 261 | info->has_subslice_pg = 0; |
| 262 | info->has_eu_pg = 0; |
| 263 | } |
| 264 | |
| 265 | /* |
| 266 | * Determine various intel_device_info fields at runtime. |
| 267 | * |
| 268 | * Use it when either: |
| 269 | * - it's judged too laborious to fill n static structures with the limit |
| 270 | * when a simple if statement does the job, |
| 271 | * - run-time checks (eg read fuse/strap registers) are needed. |
| 272 | * |
| 273 | * This function needs to be called: |
| 274 | * - after the MMIO has been setup as we are reading registers, |
| 275 | * - after the PCH has been detected, |
| 276 | * - before the first usage of the fields it can tweak. |
| 277 | */ |
| 278 | void intel_device_info_runtime_init(struct drm_i915_private *dev_priv) |
| 279 | { |
| 280 | struct intel_device_info *info = mkwrite_device_info(dev_priv); |
| 281 | enum pipe pipe; |
| 282 | |
| 283 | /* |
| 284 | * Skylake and Broxton currently don't expose the topmost plane as its |
| 285 | * use is exclusive with the legacy cursor and we only want to expose |
| 286 | * one of those, not both. Until we can safely expose the topmost plane |
| 287 | * as a DRM_PLANE_TYPE_CURSOR with all the features exposed/supported, |
| 288 | * we don't expose the topmost plane at all to prevent ABI breakage |
| 289 | * down the line. |
| 290 | */ |
| 291 | if (IS_BROXTON(dev_priv)) { |
| 292 | info->num_sprites[PIPE_A] = 2; |
| 293 | info->num_sprites[PIPE_B] = 2; |
| 294 | info->num_sprites[PIPE_C] = 1; |
| 295 | } else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) |
| 296 | for_each_pipe(dev_priv, pipe) |
| 297 | info->num_sprites[pipe] = 2; |
| 298 | else |
| 299 | for_each_pipe(dev_priv, pipe) |
| 300 | info->num_sprites[pipe] = 1; |
| 301 | |
| 302 | if (i915.disable_display) { |
| 303 | DRM_INFO("Display disabled (module parameter)\n"); |
| 304 | info->num_pipes = 0; |
| 305 | } else if (info->num_pipes > 0 && |
| 306 | (IS_GEN7(dev_priv) || IS_GEN8(dev_priv)) && |
| 307 | HAS_PCH_SPLIT(dev_priv)) { |
| 308 | u32 fuse_strap = I915_READ(FUSE_STRAP); |
| 309 | u32 sfuse_strap = I915_READ(SFUSE_STRAP); |
| 310 | |
| 311 | /* |
| 312 | * SFUSE_STRAP is supposed to have a bit signalling the display |
| 313 | * is fused off. Unfortunately it seems that, at least in |
| 314 | * certain cases, fused off display means that PCH display |
| 315 | * reads don't land anywhere. In that case, we read 0s. |
| 316 | * |
| 317 | * On CPT/PPT, we can detect this case as SFUSE_STRAP_FUSE_LOCK |
| 318 | * should be set when taking over after the firmware. |
| 319 | */ |
| 320 | if (fuse_strap & ILK_INTERNAL_DISPLAY_DISABLE || |
| 321 | sfuse_strap & SFUSE_STRAP_DISPLAY_DISABLED || |
| 322 | (dev_priv->pch_type == PCH_CPT && |
| 323 | !(sfuse_strap & SFUSE_STRAP_FUSE_LOCK))) { |
| 324 | DRM_INFO("Display fused off, disabling\n"); |
| 325 | info->num_pipes = 0; |
| 326 | } else if (fuse_strap & IVB_PIPE_C_DISABLE) { |
| 327 | DRM_INFO("PipeC fused off\n"); |
| 328 | info->num_pipes -= 1; |
| 329 | } |
| 330 | } else if (info->num_pipes > 0 && IS_GEN9(dev_priv)) { |
| 331 | u32 dfsm = I915_READ(SKL_DFSM); |
| 332 | u8 disabled_mask = 0; |
| 333 | bool invalid; |
| 334 | int num_bits; |
| 335 | |
| 336 | if (dfsm & SKL_DFSM_PIPE_A_DISABLE) |
| 337 | disabled_mask |= BIT(PIPE_A); |
| 338 | if (dfsm & SKL_DFSM_PIPE_B_DISABLE) |
| 339 | disabled_mask |= BIT(PIPE_B); |
| 340 | if (dfsm & SKL_DFSM_PIPE_C_DISABLE) |
| 341 | disabled_mask |= BIT(PIPE_C); |
| 342 | |
| 343 | num_bits = hweight8(disabled_mask); |
| 344 | |
| 345 | switch (disabled_mask) { |
| 346 | case BIT(PIPE_A): |
| 347 | case BIT(PIPE_B): |
| 348 | case BIT(PIPE_A) | BIT(PIPE_B): |
| 349 | case BIT(PIPE_A) | BIT(PIPE_C): |
| 350 | invalid = true; |
| 351 | break; |
| 352 | default: |
| 353 | invalid = false; |
| 354 | } |
| 355 | |
| 356 | if (num_bits > info->num_pipes || invalid) |
| 357 | DRM_ERROR("invalid pipe fuse configuration: 0x%x\n", |
| 358 | disabled_mask); |
| 359 | else |
| 360 | info->num_pipes -= num_bits; |
| 361 | } |
| 362 | |
| 363 | /* Initialize slice/subslice/EU info */ |
| 364 | if (IS_CHERRYVIEW(dev_priv)) |
| 365 | cherryview_sseu_info_init(dev_priv); |
| 366 | else if (IS_BROADWELL(dev_priv)) |
| 367 | broadwell_sseu_info_init(dev_priv); |
| 368 | else if (INTEL_INFO(dev_priv)->gen >= 9) |
| 369 | gen9_sseu_info_init(dev_priv); |
| 370 | |
| 371 | info->has_snoop = !info->has_llc; |
| 372 | |
| 373 | /* Snooping is broken on BXT A stepping. */ |
| 374 | if (IS_BXT_REVID(dev_priv, 0, BXT_REVID_A1)) |
| 375 | info->has_snoop = false; |
| 376 | |
| 377 | DRM_DEBUG_DRIVER("slice total: %u\n", info->slice_total); |
| 378 | DRM_DEBUG_DRIVER("subslice total: %u\n", info->subslice_total); |
| 379 | DRM_DEBUG_DRIVER("subslice per slice: %u\n", info->subslice_per_slice); |
| 380 | DRM_DEBUG_DRIVER("EU total: %u\n", info->eu_total); |
| 381 | DRM_DEBUG_DRIVER("EU per subslice: %u\n", info->eu_per_subslice); |
| 382 | DRM_DEBUG_DRIVER("has slice power gating: %s\n", |
| 383 | info->has_slice_pg ? "y" : "n"); |
| 384 | DRM_DEBUG_DRIVER("has subslice power gating: %s\n", |
| 385 | info->has_subslice_pg ? "y" : "n"); |
| 386 | DRM_DEBUG_DRIVER("has EU power gating: %s\n", |
| 387 | info->has_eu_pg ? "y" : "n"); |
| 388 | } |