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 drm_vc4_submit_cl *args,
  67                       struct exec_info *exec)
  68 {
  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 drm_vc4_submit_cl *args,
  88                         struct exec_info *exec)
  89 {
  90         for (int i = 0; i < exec->bo_count; i++) {
  91                 struct drm_gem_cma_object *obj = exec->bo[i];
  92                 struct vc4_bo *bo = obj->bo;
  93
  94                 memcpy(bo->map, obj->vaddr, bo->size);
  95
  96                 free(obj);
  97         }
  98
  99         free(exec->bo);
 100
 101         return 0;
 102 }
 103
 104 static int
 105 vc4_cl_validate(struct drm_device *dev, struct drm_vc4_submit_cl *args,
 106                 struct exec_info *exec)
 107 {
 108         void *temp = NULL;
 109         void *bin, *render, *shader_rec;
 110         int ret = 0;
 111         uint32_t bin_offset = 0;
 112         uint32_t render_offset = bin_offset + args->bin_cl_len;
 113         uint32_t shader_rec_offset = roundup(render_offset +
 114                                              args->render_cl_len, 16);
 115         uint32_t exec_size = shader_rec_offset + args->shader_record_len;
 116         uint32_t temp_size = exec_size + (sizeof(struct vc4_shader_state) *
 117                                           args->shader_record_count);
 118
 119         if (shader_rec_offset < render_offset ||
 120             exec_size < shader_rec_offset ||
 121             args->shader_record_count >= (UINT_MAX /
 122                                           sizeof(struct vc4_shader_state)) ||
 123             temp_size < exec_size) {
 124                 DRM_ERROR("overflow in exec arguments\n");
 125                 goto fail;
 126         }
 127
 128         /* Allocate space where we'll store the copied in user command lists
 129          * and shader records.
 130          *
 131          * We don't just copy directly into the BOs because we need to
 132          * read the contents back for validation, and I think the
 133          * bo->vaddr is uncached access.
 134          */
 135         temp = kmalloc(temp_size, GFP_KERNEL);
 136         if (!temp) {
 137                 DRM_ERROR("Failed to allocate storage for copying "
 138                           "in bin/render CLs.\n");
 139                 ret = -ENOMEM;
 140                 goto fail;
 141         }
 142         bin = temp + bin_offset;
 143         render = temp + render_offset;
 144         shader_rec = temp + shader_rec_offset;
 145         exec->shader_state = temp + exec_size;
 146         exec->shader_state_size = args->shader_record_count;
 147
 148         ret = copy_from_user(bin, args->bin_cl, args->bin_cl_len);
 149         if (ret) {
 150                 DRM_ERROR("Failed to copy in bin cl\n");
 151                 goto fail;
 152         }
 153
 154         ret = copy_from_user(render, args->render_cl, args->render_cl_len);
 155         if (ret) {
 156                 DRM_ERROR("Failed to copy in render cl\n");
 157                 goto fail;
 158         }
 159
 160         ret = copy_from_user(shader_rec, args->shader_records,
 161                              args->shader_record_len);
 162         if (ret) {
 163                 DRM_ERROR("Failed to copy in shader recs\n");
 164                 goto fail;
 165         }
 166
 167         exec->exec_bo = drm_gem_cma_create(dev, exec_size);
 168 #if 0
 169         if (IS_ERR(exec->exec_bo)) {
 170                 DRM_ERROR("Couldn't allocate BO for exec\n");
 171                 ret = PTR_ERR(exec->exec_bo);
 172                 exec->exec_bo = NULL;
 173                 goto fail;
 174         }
 175 #endif
 176
 177         exec->ct0ca = exec->exec_bo->paddr + bin_offset;
 178         exec->ct0ea = exec->ct0ca + args->bin_cl_len;
 179         exec->ct1ca = exec->exec_bo->paddr + render_offset;
 180         exec->ct1ea = exec->ct1ca + args->render_cl_len;
 181         exec->shader_paddr = exec->exec_bo->paddr + shader_rec_offset;
 182
 183         ret = vc4_validate_cl(dev,
 184                               exec->exec_bo->vaddr + bin_offset,
 185                               bin,
 186                               args->bin_cl_len,
 187                               true,
 188                               exec);
 189         if (ret)
 190                 goto fail;
 191
 192         ret = vc4_validate_cl(dev,
 193                               exec->exec_bo->vaddr + render_offset,
 194                               render,
 195                               args->render_cl_len,
 196                               false,
 197                               exec);
 198         if (ret)
 199                 goto fail;
 200
 201         ret = vc4_validate_shader_recs(dev,
 202                                        exec->exec_bo->vaddr + shader_rec_offset,
 203                                        shader_rec,
 204                                        args->shader_record_len,
 205                                        exec);
 206
 207 fail:
 208         kfree(temp);
 209         return ret;
 210 }
 211
 212 int
 213 vc4_simulator_flush(struct vc4_context *vc4, struct drm_vc4_submit_cl *args,
 214                     struct vc4_surface *csurf)
 215 {
 216         struct vc4_resource *ctex = vc4_resource(csurf->base.texture);
 217         uint32_t winsys_stride = ctex->bo->simulator_winsys_stride;
 218         uint32_t sim_stride = ctex->slices[0].stride;
 219         uint32_t row_len = MIN2(sim_stride, winsys_stride);
 220         struct exec_info exec;
 221         struct drm_device local_dev = {
 222                 .vc4 = vc4,
 223                 .simulator_mem_next = 0,
 224         };
 225         struct drm_device *dev = &local_dev;
 226         int ret;
 227
 228         memset(&exec, 0, sizeof(exec));
 229
 230         if (ctex->bo->simulator_winsys_map) {
 231 #if 0
 232                 fprintf(stderr, "%dx%d %d %d %d\n",
 233                         ctex->base.b.width0, ctex->base.b.height0,
 234                         winsys_stride,
 235                         sim_stride,
 236                         ctex->bo->size);
 237 #endif
 238
 239                 for (int y = 0; y < ctex->base.b.height0; y++) {
 240                         memcpy(ctex->bo->map + y * sim_stride,
 241                                ctex->bo->simulator_winsys_map + y * winsys_stride,
 242                                row_len);
 243                 }
 244         }
 245
 246         ret = vc4_simulator_pin_bos(dev, args, &exec);
 247         if (ret)
 248                 return ret;
 249
 250         ret = vc4_cl_validate(dev, args, &exec);
 251         if (ret)
 252                 return ret;
 253
 254         simpenrose_do_binning(exec.ct0ca, exec.ct0ea);
 255         simpenrose_do_rendering(exec.ct1ca, exec.ct1ea);
 256
 257         ret = vc4_simulator_unpin_bos(args, &exec);
 258         if (ret)
 259                 return ret;
 260
 261         free(exec.exec_bo);
 262
 263         if (ctex->bo->simulator_winsys_map) {
 264                 for (int y = 0; y < ctex->base.b.height0; y++) {
 265                         memcpy(ctex->bo->simulator_winsys_map + y * winsys_stride,
 266                                ctex->bo->map + y * sim_stride,
 267                                row_len);
 268                 }
 269         }
 270
 271         return 0;
 272 }
 273
 274 void
 275 vc4_simulator_init(struct vc4_screen *screen)
 276 {
 277         simpenrose_init_hardware();
 278         screen->simulator_mem_base = simpenrose_get_mem_start();
 279         screen->simulator_mem_size = simpenrose_get_mem_size();
 280 }
 281
 282 #endif /* USE_VC4_SIMULATOR */