drivers/gpu/drm/i915/i915_gem_tiling.c

   1 /*
   2  * Copyright © 2008 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  * Authors:
  24  *    Eric Anholt <eric@anholt.net>
  25  *
  26  */
  27
  28 #include <linux/string.h>
  29 #include <linux/bitops.h>
  30 #include <drm/drmP.h>
  31 #include <drm/i915_drm.h>
  32 #include "i915_drv.h"
  33
  34 /**
  35  * DOC: buffer object tiling
  36  *
  37  * i915_gem_set_tiling() and i915_gem_get_tiling() is the userspace interface to
  38  * declare fence register requirements.
  39  *
  40  * In principle GEM doesn't care at all about the internal data layout of an
  41  * object, and hence it also doesn't care about tiling or swizzling. There's two
  42  * exceptions:
  43  *
  44  * - For X and Y tiling the hardware provides detilers for CPU access, so called
  45  *   fences. Since there's only a limited amount of them the kernel must manage
  46  *   these, and therefore userspace must tell the kernel the object tiling if it
  47  *   wants to use fences for detiling.
  48  * - On gen3 and gen4 platforms have a swizzling pattern for tiled objects which
  49  *   depends upon the physical page frame number. When swapping such objects the
  50  *   page frame number might change and the kernel must be able to fix this up
  51  *   and hence now the tiling. Note that on a subset of platforms with
  52  *   asymmetric memory channel population the swizzling pattern changes in an
  53  *   unknown way, and for those the kernel simply forbids swapping completely.
  54  *
  55  * Since neither of this applies for new tiling layouts on modern platforms like
  56  * W, Ys and Yf tiling GEM only allows object tiling to be set to X or Y tiled.
  57  * Anything else can be handled in userspace entirely without the kernel's
  58  * invovlement.
  59  */
  60
  61 /* Check pitch constriants for all chips & tiling formats */
  62 static bool
  63 i915_tiling_ok(struct drm_device *dev, int stride, int size, int tiling_mode)
  64 {
  65         int tile_width;
  66
  67         /* Linear is always fine */
  68         if (tiling_mode == I915_TILING_NONE)
  69                 return true;
  70
  71         if (tiling_mode > I915_TILING_LAST)
  72                 return false;
  73
  74         if (IS_GEN2(dev) ||
  75             (tiling_mode == I915_TILING_Y && HAS_128_BYTE_Y_TILING(dev)))
  76                 tile_width = 128;
  77         else
  78                 tile_width = 512;
  79
  80         /* check maximum stride & object size */
  81         /* i965+ stores the end address of the gtt mapping in the fence
  82          * reg, so dont bother to check the size */
  83         if (INTEL_INFO(dev)->gen >= 7) {
  84                 if (stride / 128 > GEN7_FENCE_MAX_PITCH_VAL)
  85                         return false;
  86         } else if (INTEL_INFO(dev)->gen >= 4) {
  87                 if (stride / 128 > I965_FENCE_MAX_PITCH_VAL)
  88                         return false;
  89         } else {
  90                 if (stride > 8192)
  91                         return false;
  92
  93                 if (IS_GEN3(dev)) {
  94                         if (size > I830_FENCE_MAX_SIZE_VAL << 20)
  95                                 return false;
  96                 } else {
  97                         if (size > I830_FENCE_MAX_SIZE_VAL << 19)
  98                                 return false;
  99                 }
 100         }
 101
 102         if (stride < tile_width)
 103                 return false;
 104
 105         /* 965+ just needs multiples of tile width */
 106         if (INTEL_INFO(dev)->gen >= 4) {
 107                 if (stride & (tile_width - 1))
 108                         return false;
 109                 return true;
 110         }
 111
 112         /* Pre-965 needs power of two tile widths */
 113         if (stride & (stride - 1))
 114                 return false;
 115
 116         return true;
 117 }
 118
 119 /* Make the current GTT allocation valid for the change in tiling. */
 120 static int
 121 i915_gem_object_fence_prepare(struct drm_i915_gem_object *obj, int tiling_mode)
 122 {
 123         struct drm_i915_private *dev_priv = to_i915(obj->base.dev);
 124         struct i915_vma *vma;
 125         u32 size;
 126
 127         if (tiling_mode == I915_TILING_NONE)
 128                 return 0;
 129
 130         if (INTEL_GEN(dev_priv) >= 4)
 131                 return 0;
 132
 133         vma = i915_gem_obj_to_ggtt(obj);
 134         if (!vma)
 135                 return 0;
 136
 137         if (!obj->map_and_fenceable)
 138                 return 0;
 139
 140         if (IS_GEN3(dev_priv)) {
 141                 if (vma->node.start & ~I915_FENCE_START_MASK)
 142                         goto bad;
 143         } else {
 144                 if (vma->node.start & ~I830_FENCE_START_MASK)
 145                         goto bad;
 146         }
 147
 148         size = i915_gem_get_ggtt_size(dev_priv, obj->base.size, tiling_mode);
 149         if (vma->node.size < size)
 150                 goto bad;
 151
 152         if (vma->node.start & (size - 1))
 153                 goto bad;
 154
 155         return 0;
 156
 157 bad:
 158         return i915_vma_unbind(vma);
 159 }
 160
 161 /**
 162  * i915_gem_set_tiling - IOCTL handler to set tiling mode
 163  * @dev: DRM device
 164  * @data: data pointer for the ioctl
 165  * @file: DRM file for the ioctl call
 166  *
 167  * Sets the tiling mode of an object, returning the required swizzling of
 168  * bit 6 of addresses in the object.
 169  *
 170  * Called by the user via ioctl.
 171  *
 172  * Returns:
 173  * Zero on success, negative errno on failure.
 174  */
 175 int
 176 i915_gem_set_tiling(struct drm_device *dev, void *data,
 177                    struct drm_file *file)
 178 {
 179         struct drm_i915_gem_set_tiling *args = data;
 180         struct drm_i915_private *dev_priv = to_i915(dev);
 181         struct drm_i915_gem_object *obj;
 182         int err = 0;
 183
 184         /* Make sure we don't cross-contaminate obj->tiling_and_stride */
 185         BUILD_BUG_ON(I915_TILING_LAST & STRIDE_MASK);
 186
 187         obj = i915_gem_object_lookup(file, args->handle);
 188         if (!obj)
 189                 return -ENOENT;
 190
 191         if (!i915_tiling_ok(dev,
 192                             args->stride, obj->base.size, args->tiling_mode)) {
 193                 i915_gem_object_put_unlocked(obj);
 194                 return -EINVAL;
 195         }
 196
 197         intel_runtime_pm_get(dev_priv);
 198
 199         mutex_lock(&dev->struct_mutex);
 200         if (obj->pin_display || obj->framebuffer_references) {
 201                 err = -EBUSY;
 202                 goto err;
 203         }
 204
 205         if (args->tiling_mode == I915_TILING_NONE) {
 206                 args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
 207                 args->stride = 0;
 208         } else {
 209                 if (args->tiling_mode == I915_TILING_X)
 210                         args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x;
 211                 else
 212                         args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y;
 213
 214                 /* Hide bit 17 swizzling from the user.  This prevents old Mesa
 215                  * from aborting the application on sw fallbacks to bit 17,
 216                  * and we use the pread/pwrite bit17 paths to swizzle for it.
 217                  * If there was a user that was relying on the swizzle
 218                  * information for drm_intel_bo_map()ed reads/writes this would
 219                  * break it, but we don't have any of those.
 220                  */
 221                 if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_17)
 222                         args->swizzle_mode = I915_BIT_6_SWIZZLE_9;
 223                 if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_10_17)
 224                         args->swizzle_mode = I915_BIT_6_SWIZZLE_9_10;
 225
 226                 /* If we can't handle the swizzling, make it untiled. */
 227                 if (args->swizzle_mode == I915_BIT_6_SWIZZLE_UNKNOWN) {
 228                         args->tiling_mode = I915_TILING_NONE;
 229                         args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
 230                         args->stride = 0;
 231                 }
 232         }
 233
 234         if (args->tiling_mode != i915_gem_object_get_tiling(obj) ||
 235             args->stride != i915_gem_object_get_stride(obj)) {
 236                 /* We need to rebind the object if its current allocation
 237                  * no longer meets the alignment restrictions for its new
 238                  * tiling mode. Otherwise we can just leave it alone, but
 239                  * need to ensure that any fence register is updated before
 240                  * the next fenced (either through the GTT or by the BLT unit
 241                  * on older GPUs) access.
 242                  *
 243                  * After updating the tiling parameters, we then flag whether
 244                  * we need to update an associated fence register. Note this
 245                  * has to also include the unfenced register the GPU uses
 246                  * whilst executing a fenced command for an untiled object.
 247                  */
 248
 249                 err = i915_gem_object_fence_prepare(obj, args->tiling_mode);
 250                 if (!err) {
 251                         if (obj->pages &&
 252                             obj->madv == I915_MADV_WILLNEED &&
 253                             dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES) {
 254                                 if (args->tiling_mode == I915_TILING_NONE)
 255                                         i915_gem_object_unpin_pages(obj);
 256                                 if (!i915_gem_object_is_tiled(obj))
 257                                         i915_gem_object_pin_pages(obj);
 258                         }
 259
 260                         obj->fence_dirty =
 261                                 !i915_gem_active_is_idle(&obj->last_fence,
 262                                                          &dev->struct_mutex) ||
 263                                 obj->fence_reg != I915_FENCE_REG_NONE;
 264
 265                         obj->tiling_and_stride =
 266                                 args->stride | args->tiling_mode;
 267
 268                         /* Force the fence to be reacquired for GTT access */
 269                         i915_gem_release_mmap(obj);
 270                 }
 271         }
 272         /* we have to maintain this existing ABI... */
 273         args->stride = i915_gem_object_get_stride(obj);
 274         args->tiling_mode = i915_gem_object_get_tiling(obj);
 275
 276         /* Try to preallocate memory required to save swizzling on put-pages */
 277         if (i915_gem_object_needs_bit17_swizzle(obj)) {
 278                 if (obj->bit_17 == NULL) {
 279                         obj->bit_17 = kcalloc(BITS_TO_LONGS(obj->base.size >> PAGE_SHIFT),
 280                                               sizeof(long), GFP_KERNEL);
 281                 }
 282         } else {
 283                 kfree(obj->bit_17);
 284                 obj->bit_17 = NULL;
 285         }
 286
 287 err:
 288         i915_gem_object_put(obj);
 289         mutex_unlock(&dev->struct_mutex);
 290
 291         intel_runtime_pm_put(dev_priv);
 292
 293         return err;
 294 }
 295
 296 /**
 297  * i915_gem_get_tiling - IOCTL handler to get tiling mode
 298  * @dev: DRM device
 299  * @data: data pointer for the ioctl
 300  * @file: DRM file for the ioctl call
 301  *
 302  * Returns the current tiling mode and required bit 6 swizzling for the object.
 303  *
 304  * Called by the user via ioctl.
 305  *
 306  * Returns:
 307  * Zero on success, negative errno on failure.
 308  */
 309 int
 310 i915_gem_get_tiling(struct drm_device *dev, void *data,
 311                    struct drm_file *file)
 312 {
 313         struct drm_i915_gem_get_tiling *args = data;
 314         struct drm_i915_private *dev_priv = to_i915(dev);
 315         struct drm_i915_gem_object *obj;
 316
 317         obj = i915_gem_object_lookup(file, args->handle);
 318         if (!obj)
 319                 return -ENOENT;
 320
 321         args->tiling_mode = READ_ONCE(obj->tiling_and_stride) & TILING_MASK;
 322         switch (args->tiling_mode) {
 323         case I915_TILING_X:
 324                 args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x;
 325                 break;
 326         case I915_TILING_Y:
 327                 args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y;
 328                 break;
 329         case I915_TILING_NONE:
 330                 args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
 331                 break;
 332         default:
 333                 DRM_ERROR("unknown tiling mode\n");
 334         }
 335
 336         /* Hide bit 17 from the user -- see comment in i915_gem_set_tiling */
 337         if (dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES)
 338                 args->phys_swizzle_mode = I915_BIT_6_SWIZZLE_UNKNOWN;
 339         else
 340                 args->phys_swizzle_mode = args->swizzle_mode;
 341         if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_17)
 342                 args->swizzle_mode = I915_BIT_6_SWIZZLE_9;
 343         if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_10_17)
 344                 args->swizzle_mode = I915_BIT_6_SWIZZLE_9_10;
 345
 346         i915_gem_object_put_unlocked(obj);
 347         return 0;
 348 }