drivers/gpu/drm/i915/intel_mocs.c

   1 /*
   2  * Copyright (c) 2015 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  * The above copyright notice and this permission notice (including the next
  11  * paragraph) shall be included in all copies or substantial portions of the
  12  * 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 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  18  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  19  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  20  * SOFTWARE.
  21  */
  22
  23 #include "intel_mocs.h"
  24 #include "intel_lrc.h"
  25 #include "intel_ringbuffer.h"
  26
  27 /* structures required */
  28 struct drm_i915_mocs_entry {
  29         u32 control_value;
  30         u16 l3cc_value;
  31 };
  32
  33 struct drm_i915_mocs_table {
  34         u32 size;
  35         const struct drm_i915_mocs_entry *table;
  36 };
  37
  38 /* Defines for the tables (XXX_MOCS_0 - XXX_MOCS_63) */
  39 #define LE_CACHEABILITY(value)  ((value) << 0)
  40 #define LE_TGT_CACHE(value)     ((value) << 2)
  41 #define LE_LRUM(value)          ((value) << 4)
  42 #define LE_AOM(value)           ((value) << 6)
  43 #define LE_RSC(value)           ((value) << 7)
  44 #define LE_SCC(value)           ((value) << 8)
  45 #define LE_PFM(value)           ((value) << 11)
  46 #define LE_SCF(value)           ((value) << 14)
  47
  48 /* Defines for the tables (LNCFMOCS0 - LNCFMOCS31) - two entries per word */
  49 #define L3_ESC(value)           ((value) << 0)
  50 #define L3_SCC(value)           ((value) << 1)
  51 #define L3_CACHEABILITY(value)  ((value) << 4)
  52
  53 /* Helper defines */
  54 #define GEN9_NUM_MOCS_ENTRIES   62  /* 62 out of 64 - 63 & 64 are reserved. */
  55
  56 /* (e)LLC caching options */
  57 #define LE_PAGETABLE            0
  58 #define LE_UC                   1
  59 #define LE_WT                   2
  60 #define LE_WB                   3
  61
  62 /* L3 caching options */
  63 #define L3_DIRECT               0
  64 #define L3_UC                   1
  65 #define L3_RESERVED             2
  66 #define L3_WB                   3
  67
  68 /* Target cache */
  69 #define ELLC                    0
  70 #define LLC                     1
  71 #define LLC_ELLC                2
  72
  73 /*
  74  * MOCS tables
  75  *
  76  * These are the MOCS tables that are programmed across all the rings.
  77  * The control value is programmed to all the rings that support the
  78  * MOCS registers. While the l3cc_values are only programmed to the
  79  * LNCFCMOCS0 - LNCFCMOCS32 registers.
  80  *
  81  * These tables are intended to be kept reasonably consistent across
  82  * platforms. However some of the fields are not applicable to all of
  83  * them.
  84  *
  85  * Entries not part of the following tables are undefined as far as
  86  * userspace is concerned and shouldn't be relied upon.  For the time
  87  * being they will be implicitly initialized to the strictest caching
  88  * configuration (uncached) to guarantee forwards compatibility with
  89  * userspace programs written against more recent kernels providing
  90  * additional MOCS entries.
  91  *
  92  * NOTE: These tables MUST start with being uncached and the length
  93  *       MUST be less than 63 as the last two registers are reserved
  94  *       by the hardware.  These tables are part of the kernel ABI and
  95  *       may only be updated incrementally by adding entries at the
  96  *       end.
  97  */
  98 static const struct drm_i915_mocs_entry skylake_mocs_table[] = {
  99         /* { 0x00000009, 0x0010 } */
 100         { (LE_CACHEABILITY(LE_UC) | LE_TGT_CACHE(LLC_ELLC) | LE_LRUM(0) |
 101            LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | LE_PFM(0) | LE_SCF(0)),
 102           (L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_UC)) },
 103         /* { 0x00000038, 0x0030 } */
 104         { (LE_CACHEABILITY(LE_PAGETABLE) | LE_TGT_CACHE(LLC_ELLC) | LE_LRUM(3) |
 105            LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | LE_PFM(0) | LE_SCF(0)),
 106           (L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_WB)) },
 107         /* { 0x0000003b, 0x0030 } */
 108         { (LE_CACHEABILITY(LE_WB) | LE_TGT_CACHE(LLC_ELLC) | LE_LRUM(3) |
 109            LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | LE_PFM(0) | LE_SCF(0)),
 110           (L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_WB)) }
 111 };
 112
 113 /* NOTE: the LE_TGT_CACHE is not used on Broxton */
 114 static const struct drm_i915_mocs_entry broxton_mocs_table[] = {
 115         /* { 0x00000009, 0x0010 } */
 116         { (LE_CACHEABILITY(LE_UC) | LE_TGT_CACHE(LLC_ELLC) | LE_LRUM(0) |
 117            LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | LE_PFM(0) | LE_SCF(0)),
 118           (L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_UC)) },
 119         /* { 0x00000038, 0x0030 } */
 120         { (LE_CACHEABILITY(LE_PAGETABLE) | LE_TGT_CACHE(LLC_ELLC) | LE_LRUM(3) |
 121            LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | LE_PFM(0) | LE_SCF(0)),
 122           (L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_WB)) },
 123         /* { 0x0000003b, 0x0030 } */
 124         { (LE_CACHEABILITY(LE_WB) | LE_TGT_CACHE(LLC_ELLC) | LE_LRUM(3) |
 125            LE_AOM(0) | LE_RSC(0) | LE_SCC(0) | LE_PFM(0) | LE_SCF(0)),
 126           (L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_WB)) }
 127 };
 128
 129 /**
 130  * get_mocs_settings()
 131  * @dev:        DRM device.
 132  * @table:      Output table that will be made to point at appropriate
 133  *              MOCS values for the device.
 134  *
 135  * This function will return the values of the MOCS table that needs to
 136  * be programmed for the platform. It will return the values that need
 137  * to be programmed and if they need to be programmed.
 138  *
 139  * Return: true if there are applicable MOCS settings for the device.
 140  */
 141 static bool get_mocs_settings(struct drm_device *dev,
 142                               struct drm_i915_mocs_table *table)
 143 {
 144         bool result = false;
 145
 146         if (IS_SKYLAKE(dev)) {
 147                 table->size  = ARRAY_SIZE(skylake_mocs_table);
 148                 table->table = skylake_mocs_table;
 149                 result = true;
 150         } else if (IS_BROXTON(dev)) {
 151                 table->size  = ARRAY_SIZE(broxton_mocs_table);
 152                 table->table = broxton_mocs_table;
 153                 result = true;
 154         } else {
 155                 WARN_ONCE(INTEL_INFO(dev)->gen >= 9,
 156                           "Platform that should have a MOCS table does not.\n");
 157         }
 158
 159         return result;
 160 }
 161
 162 /**
 163  * emit_mocs_control_table() - emit the mocs control table
 164  * @req:        Request to set up the MOCS table for.
 165  * @table:      The values to program into the control regs.
 166  * @reg_base:   The base for the engine that needs to be programmed.
 167  *
 168  * This function simply emits a MI_LOAD_REGISTER_IMM command for the
 169  * given table starting at the given address.
 170  *
 171  * Return: 0 on success, otherwise the error status.
 172  */
 173 static int emit_mocs_control_table(struct drm_i915_gem_request *req,
 174                                    const struct drm_i915_mocs_table *table,
 175                                    u32 reg_base)
 176 {
 177         struct intel_ringbuffer *ringbuf = req->ringbuf;
 178         unsigned int index;
 179         int ret;
 180
 181         if (WARN_ON(table->size > GEN9_NUM_MOCS_ENTRIES))
 182                 return -ENODEV;
 183
 184         ret = intel_logical_ring_begin(req, 2 + 2 * GEN9_NUM_MOCS_ENTRIES);
 185         if (ret) {
 186                 DRM_DEBUG("intel_logical_ring_begin failed %d\n", ret);
 187                 return ret;
 188         }
 189
 190         intel_logical_ring_emit(ringbuf,
 191                                 MI_LOAD_REGISTER_IMM(GEN9_NUM_MOCS_ENTRIES));
 192
 193         for (index = 0; index < table->size; index++) {
 194                 intel_logical_ring_emit(ringbuf, reg_base + index * 4);
 195                 intel_logical_ring_emit(ringbuf,
 196                                         table->table[index].control_value);
 197         }
 198
 199         /*
 200          * Ok, now set the unused entries to uncached. These entries
 201          * are officially undefined and no contract for the contents
 202          * and settings is given for these entries.
 203          *
 204          * Entry 0 in the table is uncached - so we are just writing
 205          * that value to all the used entries.
 206          */
 207         for (; index < GEN9_NUM_MOCS_ENTRIES; index++) {
 208                 intel_logical_ring_emit(ringbuf, reg_base + index * 4);
 209                 intel_logical_ring_emit(ringbuf, table->table[0].control_value);
 210         }
 211
 212         intel_logical_ring_emit(ringbuf, MI_NOOP);
 213         intel_logical_ring_advance(ringbuf);
 214
 215         return 0;
 216 }
 217
 218 /**
 219  * emit_mocs_l3cc_table() - emit the mocs control table
 220  * @req:        Request to set up the MOCS table for.
 221  * @table:      The values to program into the control regs.
 222  *
 223  * This function simply emits a MI_LOAD_REGISTER_IMM command for the
 224  * given table starting at the given address. This register set is
 225  * programmed in pairs.
 226  *
 227  * Return: 0 on success, otherwise the error status.
 228  */
 229 static int emit_mocs_l3cc_table(struct drm_i915_gem_request *req,
 230                                 const struct drm_i915_mocs_table *table)
 231 {
 232         struct intel_ringbuffer *ringbuf = req->ringbuf;
 233         unsigned int count;
 234         unsigned int i;
 235         u32 value;
 236         u32 filler = (table->table[0].l3cc_value & 0xffff) |
 237                         ((table->table[0].l3cc_value & 0xffff) << 16);
 238         int ret;
 239
 240         if (WARN_ON(table->size > GEN9_NUM_MOCS_ENTRIES))
 241                 return -ENODEV;
 242
 243         ret = intel_logical_ring_begin(req, 2 + GEN9_NUM_MOCS_ENTRIES);
 244         if (ret) {
 245                 DRM_DEBUG("intel_logical_ring_begin failed %d\n", ret);
 246                 return ret;
 247         }
 248
 249         intel_logical_ring_emit(ringbuf,
 250                         MI_LOAD_REGISTER_IMM(GEN9_NUM_MOCS_ENTRIES / 2));
 251
 252         for (i = 0, count = 0; i < table->size / 2; i++, count += 2) {
 253                 value = (table->table[count].l3cc_value & 0xffff) |
 254                         ((table->table[count + 1].l3cc_value & 0xffff) << 16);
 255
 256                 intel_logical_ring_emit(ringbuf, GEN9_LNCFCMOCS0 + i * 4);
 257                 intel_logical_ring_emit(ringbuf, value);
 258         }
 259
 260         if (table->size & 0x01) {
 261                 /* Odd table size - 1 left over */
 262                 value = (table->table[count].l3cc_value & 0xffff) |
 263                         ((table->table[0].l3cc_value & 0xffff) << 16);
 264         } else
 265                 value = filler;
 266
 267         /*
 268          * Now set the rest of the table to uncached - use entry 0 as
 269          * this will be uncached. Leave the last pair uninitialised as
 270          * they are reserved by the hardware.
 271          */
 272         for (; i < GEN9_NUM_MOCS_ENTRIES / 2; i++) {
 273                 intel_logical_ring_emit(ringbuf, GEN9_LNCFCMOCS0 + i * 4);
 274                 intel_logical_ring_emit(ringbuf, value);
 275
 276                 value = filler;
 277         }
 278
 279         intel_logical_ring_emit(ringbuf, MI_NOOP);
 280         intel_logical_ring_advance(ringbuf);
 281
 282         return 0;
 283 }
 284
 285 /**
 286  * intel_rcs_context_init_mocs() - program the MOCS register.
 287  * @req:        Request to set up the MOCS tables for.
 288  *
 289  * This function will emit a batch buffer with the values required for
 290  * programming the MOCS register values for all the currently supported
 291  * rings.
 292  *
 293  * These registers are partially stored in the RCS context, so they are
 294  * emitted at the same time so that when a context is created these registers
 295  * are set up. These registers have to be emitted into the start of the
 296  * context as setting the ELSP will re-init some of these registers back
 297  * to the hw values.
 298  *
 299  * Return: 0 on success, otherwise the error status.
 300  */
 301 int intel_rcs_context_init_mocs(struct drm_i915_gem_request *req)
 302 {
 303         struct drm_i915_mocs_table t;
 304         int ret;
 305
 306         if (get_mocs_settings(req->ring->dev, &t)) {
 307                 /* Program the control registers */
 308                 ret = emit_mocs_control_table(req, &t, GEN9_GFX_MOCS_0);
 309                 if (ret)
 310                         return ret;
 311
 312                 ret = emit_mocs_control_table(req, &t, GEN9_MFX0_MOCS_0);
 313                 if (ret)
 314                         return ret;
 315
 316                 ret = emit_mocs_control_table(req, &t, GEN9_MFX1_MOCS_0);
 317                 if (ret)
 318                         return ret;
 319
 320                 ret = emit_mocs_control_table(req, &t, GEN9_VEBOX_MOCS_0);
 321                 if (ret)
 322                         return ret;
 323
 324                 ret = emit_mocs_control_table(req, &t, GEN9_BLT_MOCS_0);
 325                 if (ret)
 326                         return ret;
 327
 328                 /* Now program the l3cc registers */
 329                 ret = emit_mocs_l3cc_table(req, &t);
 330                 if (ret)
 331                         return ret;
 332         }
 333
 334         return 0;
 335 }