src/gallium/drivers/vc4/vc4_simulator.c

   1 /*
   2  * Copyright © 2014 Broadcom
   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 #ifdef USE_VC4_SIMULATOR
  25
  26 #include <stdio.h>
  27
  28 #include "util/u_memory.h"
  29
  30 #include "vc4_screen.h"
  31 #include "vc4_context.h"
  32 #include "vc4_simulator_validate.h"
  33 #include "simpenrose/simpenrose.h"
  34
  35 static struct drm_gem_cma_object *
  36 vc4_wrap_bo_with_cma(struct drm_device *dev, struct vc4_bo *bo)
  37 {
  38         struct vc4_context *vc4 = dev->vc4;
  39         struct vc4_screen *screen = vc4->screen;
  40         struct drm_gem_cma_object *obj = CALLOC_STRUCT(drm_gem_cma_object);
  41         uint32_t size = align(bo->size, 4096);
  42
  43         obj->bo = bo;
  44         obj->base.size = size;
  45         obj->vaddr = screen->simulator_mem_base + dev->simulator_mem_next;
  46         obj->paddr = simpenrose_hw_addr(obj->vaddr);
  47
  48         dev->simulator_mem_next += size;
  49         dev->simulator_mem_next = align(dev->simulator_mem_next, 4096);
  50         assert(dev->simulator_mem_next <= screen->simulator_mem_size);
  51
  52         return obj;
  53 }
  54
  55 static struct drm_gem_cma_object *
  56 drm_gem_cma_create(struct drm_device *dev, size_t size)
  57 {
  58         struct vc4_context *vc4 = dev->vc4;
  59         struct vc4_screen *screen = vc4->screen;
  60
  61         struct vc4_bo *bo = vc4_bo_alloc(screen, size, "simulator validate");
  62         return vc4_wrap_bo_with_cma(dev, bo);
  63 }
  64
  65 static int
  66 vc4_simulator_pin_bos(struct drm_device *dev, struct exec_info *exec)
  67 {
  68         struct drm_vc4_submit_cl *args = exec->args;
  69         struct vc4_context *vc4 = dev->vc4;
  70         struct vc4_bo **bos = vc4->bo_pointers.base;
  71
  72         exec->bo_count = args->bo_handle_count;
  73         exec->bo = calloc(exec->bo_count, sizeof(void *));
  74         for (int i = 0; i < exec->bo_count; i++) {
  75                 struct vc4_bo *bo = bos[i];
  76                 struct drm_gem_cma_object *obj = vc4_wrap_bo_with_cma(dev, bo);
  77
  78                 memcpy(obj->vaddr, bo->map, bo->size);
  79
  80                 exec->bo[i] = obj;
  81         }
  82
  83         return 0;
  84 }
  85
  86 static int
  87 vc4_simulator_unpin_bos(struct exec_info *exec)
  88 {
  89         for (int i = 0; i < exec->bo_count; i++) {
  90                 struct drm_gem_cma_object *obj = exec->bo[i];
  91                 struct vc4_bo *bo = obj->bo;
  92
  93                 memcpy(bo->map, obj->vaddr, bo->size);
  94
  95                 free(obj);
  96         }
  97
  98         free(exec->bo);
  99
 100         return 0;
 101 }
 102
 103 static int
 104 vc4_cl_validate(struct drm_device *dev, struct exec_info *exec)
 105 {
 106         struct drm_vc4_submit_cl *args = exec->args;
 107         void *temp = NULL;
 108         void *bin, *render;
 109         int ret = 0;
 110         uint32_t bin_offset = 0;
 111         uint32_t render_offset = bin_offset + args->bin_cl_size;
 112         uint32_t shader_rec_offset = roundup(render_offset +
 113                                              args->render_cl_size, 16);
 114         uint32_t uniforms_offset = shader_rec_offset + args->shader_rec_size;
 115         uint32_t exec_size = uniforms_offset + args->uniforms_size;
 116         uint32_t temp_size = exec_size + (sizeof(struct vc4_shader_state) *
 117                                           args->shader_rec_count);
 118
 119         if (shader_rec_offset < render_offset ||
 120             uniforms_offset < shader_rec_offset ||
 121             exec_size < uniforms_offset ||
 122             args->shader_rec_count >= (UINT_MAX /
 123                                           sizeof(struct vc4_shader_state)) ||
 124             temp_size < exec_size) {
 125                 DRM_ERROR("overflow in exec arguments\n");
 126                 goto fail;
 127         }
 128
 129         /* Allocate space where we'll store the copied in user command lists
 130          * and shader records.
 131          *
 132          * We don't just copy directly into the BOs because we need to
 133          * read the contents back for validation, and I think the
 134          * bo->vaddr is uncached access.
 135          */
 136         temp = kmalloc(temp_size, GFP_KERNEL);
 137         if (!temp) {
 138                 DRM_ERROR("Failed to allocate storage for copying "
 139                           "in bin/render CLs.\n");
 140                 ret = -ENOMEM;
 141                 goto fail;
 142         }
 143         bin = temp + bin_offset;
 144         render = temp + render_offset;
 145         exec->shader_rec_u = temp + shader_rec_offset;
 146         exec->uniforms_u = temp + uniforms_offset;
 147         exec->shader_state = temp + exec_size;
 148         exec->shader_state_size = args->shader_rec_count;
 149
 150         ret = copy_from_user(bin, args->bin_cl, args->bin_cl_size);
 151         if (ret) {
 152                 DRM_ERROR("Failed to copy in bin cl\n");
 153                 goto fail;
 154         }
 155
 156         ret = copy_from_user(render, args->render_cl, args->render_cl_size);
 157         if (ret) {
 158                 DRM_ERROR("Failed to copy in render cl\n");
 159                 goto fail;
 160         }
 161
 162         ret = copy_from_user(exec->shader_rec_u, args->shader_rec,
 163                              args->shader_rec_size);
 164         if (ret) {
 165                 DRM_ERROR("Failed to copy in shader recs\n");
 166                 goto fail;
 167         }
 168
 169         ret = copy_from_user(exec->uniforms_u, args->uniforms,
 170                              args->uniforms_size);
 171         if (ret) {
 172                 DRM_ERROR("Failed to copy in uniforms cl\n");
 173                 goto fail;
 174         }
 175
 176         exec->exec_bo = drm_gem_cma_create(dev, exec_size);
 177 #if 0
 178         if (IS_ERR(exec->exec_bo)) {
 179                 DRM_ERROR("Couldn't allocate BO for exec\n");
 180                 ret = PTR_ERR(exec->exec_bo);
 181                 exec->exec_bo = NULL;
 182                 goto fail;
 183         }
 184 #endif
 185
 186         exec->ct0ca = exec->exec_bo->paddr + bin_offset;
 187         exec->ct1ca = exec->exec_bo->paddr + render_offset;
 188
 189         exec->shader_rec_v = exec->exec_bo->vaddr + shader_rec_offset;
 190         exec->shader_rec_p = exec->exec_bo->paddr + shader_rec_offset;
 191         exec->shader_rec_size = args->shader_rec_size;
 192
 193         exec->uniforms_v = exec->exec_bo->vaddr + uniforms_offset;
 194         exec->uniforms_p = exec->exec_bo->paddr + uniforms_offset;
 195         exec->uniforms_size = args->uniforms_size;
 196
 197         ret = vc4_validate_cl(dev,
 198                               exec->exec_bo->vaddr + bin_offset,
 199                               bin,
 200                               args->bin_cl_size,
 201                               true,
 202                               exec);
 203         if (ret)
 204                 goto fail;
 205
 206         ret = vc4_validate_cl(dev,
 207                               exec->exec_bo->vaddr + render_offset,
 208                               render,
 209                               args->render_cl_size,
 210                               false,
 211                               exec);
 212         if (ret)
 213                 goto fail;
 214
 215         ret = vc4_validate_shader_recs(dev, exec);
 216
 217 fail:
 218         kfree(temp);
 219         return ret;
 220 }
 221
 222 int
 223 vc4_simulator_flush(struct vc4_context *vc4, struct drm_vc4_submit_cl *args,
 224                     struct vc4_surface *csurf)
 225 {
 226         struct vc4_resource *ctex = vc4_resource(csurf->base.texture);
 227         uint32_t winsys_stride = ctex->bo->simulator_winsys_stride;
 228         uint32_t sim_stride = ctex->slices[0].stride;
 229         uint32_t row_len = MIN2(sim_stride, winsys_stride);
 230         struct exec_info exec;
 231         struct drm_device local_dev = {
 232                 .vc4 = vc4,
 233                 .simulator_mem_next = 0,
 234         };
 235         struct drm_device *dev = &local_dev;
 236         int ret;
 237
 238         memset(&exec, 0, sizeof(exec));
 239
 240         if (ctex->bo->simulator_winsys_map) {
 241 #if 0
 242                 fprintf(stderr, "%dx%d %d %d %d\n",
 243                         ctex->base.b.width0, ctex->base.b.height0,
 244                         winsys_stride,
 245                         sim_stride,
 246                         ctex->bo->size);
 247 #endif
 248
 249                 for (int y = 0; y < ctex->base.b.height0; y++) {
 250                         memcpy(ctex->bo->map + y * sim_stride,
 251                                ctex->bo->simulator_winsys_map + y * winsys_stride,
 252                                row_len);
 253                 }
 254         }
 255
 256         exec.args = args;
 257
 258         ret = vc4_simulator_pin_bos(dev, &exec);
 259         if (ret)
 260                 return ret;
 261
 262         ret = vc4_cl_validate(dev, &exec);
 263         if (ret)
 264                 return ret;
 265
 266         simpenrose_do_binning(exec.ct0ca, exec.ct0ea);
 267         simpenrose_do_rendering(exec.ct1ca, exec.ct1ea);
 268
 269         ret = vc4_simulator_unpin_bos(&exec);
 270         if (ret)
 271                 return ret;
 272
 273         free(exec.exec_bo);
 274
 275         if (ctex->bo->simulator_winsys_map) {
 276                 for (int y = 0; y < ctex->base.b.height0; y++) {
 277                         memcpy(ctex->bo->simulator_winsys_map + y * winsys_stride,
 278                                ctex->bo->map + y * sim_stride,
 279                                row_len);
 280                 }
 281         }
 282
 283         return 0;
 284 }
 285
 286 void
 287 vc4_simulator_init(struct vc4_screen *screen)
 288 {
 289         simpenrose_init_hardware();
 290         screen->simulator_mem_base = simpenrose_get_mem_start();
 291         screen->simulator_mem_size = simpenrose_get_mem_size();
 292 }
 293
 294 #endif /* USE_VC4_SIMULATOR */