| 1 |
/* |
| 2 |
* Copyright (c) 2009-2024 Broadcom. All Rights Reserved. |
| 3 |
* The term “Broadcom” refers to Broadcom Inc. |
| 4 |
* and/or its subsidiaries. |
| 5 |
* SPDX-License-Identifier: MIT |
| 6 |
*/ |
| 7 |
|
| 8 |
/** |
| 9 |
* @file |
| 10 |
* |
| 11 |
* Wrappers for DRM ioctl functionlaity used by the rest of the vmw |
| 12 |
* drm winsys. |
| 13 |
* |
| 14 |
* Based on svgaicd_escape.c |
| 15 |
*/ |
| 16 |
|
| 17 |
|
| 18 |
#include "svga_cmd.h" |
| 19 |
#include "util/u_memory.h" |
| 20 |
#include "util/u_math.h" |
| 21 |
#include "svgadump/svga_dump.h" |
| 22 |
#include "frontend/drm_driver.h" |
| 23 |
#include "vmw_screen.h" |
| 24 |
#include "vmw_context.h" |
| 25 |
#include "vmw_fence.h" |
| 26 |
#include "xf86drm.h" |
| 27 |
#include "vmwgfx_drm.h" |
| 28 |
#include "svga3d_devcaps.h" |
| 29 |
#include "svga3d_reg.h" |
| 30 |
|
| 31 |
#include "util/os_mman.h" |
| 32 |
|
| 33 |
#include <errno.h> |
| 34 |
#include <unistd.h> |
| 35 |
|
| 36 |
#define VMW_MAX_DEFAULT_TEXTURE_SIZE (128 * 1024 * 1024) |
| 37 |
#define VMW_FENCE_TIMEOUT_SECONDS 3600UL |
| 38 |
|
| 39 |
#define SVGA3D_FLAGS_64(upper32, lower32) (((uint64_t)upper32 << 32) | lower32) |
| 40 |
#define SVGA3D_FLAGS_UPPER_32(svga3d_flags) (svga3d_flags >> 32) |
| 41 |
#define SVGA3D_FLAGS_LOWER_32(svga3d_flags) \ |
| 42 |
(svga3d_flags & ((uint64_t)UINT32_MAX)) |
| 43 |
|
| 44 |
struct vmw_region |
| 45 |
{ |
| 46 |
uint32_t handle; |
| 47 |
uint64_t map_handle; |
| 48 |
void *data; |
| 49 |
uint32_t map_count; |
| 50 |
int drm_fd; |
| 51 |
uint32_t size; |
| 52 |
}; |
| 53 |
|
| 54 |
uint32_t |
| 55 |
vmw_region_size(struct vmw_region *region) |
| 56 |
{ |
| 57 |
return region->size; |
| 58 |
} |
| 59 |
|
| 60 |
#if defined(__DragonFly__) || defined(__FreeBSD__) || \ |
| 61 |
defined(__NetBSD__) || defined(__OpenBSD__) |
| 62 |
#define ERESTART EINTR |
| 63 |
#endif |
| 64 |
|
| 65 |
uint32 |
| 66 |
vmw_ioctl_context_create(struct vmw_winsys_screen *vws) |
| 67 |
{ |
| 68 |
struct drm_vmw_context_arg c_arg; |
| 69 |
int ret; |
| 70 |
|
| 71 |
VMW_FUNC; |
| 72 |
|
| 73 |
ret = drmCommandRead(vws->ioctl.drm_fd, DRM_VMW_CREATE_CONTEXT, |
| 74 |
&c_arg, sizeof(c_arg)); |
| 75 |
|
| 76 |
if (ret) |
| 77 |
return -1; |
| 78 |
|
| 79 |
vmw_printf("Context id is %d\n", c_arg.cid); |
| 80 |
return c_arg.cid; |
| 81 |
} |
| 82 |
|
| 83 |
uint32 |
| 84 |
vmw_ioctl_extended_context_create(struct vmw_winsys_screen *vws, |
| 85 |
bool vgpu10) |
| 86 |
{ |
| 87 |
union drm_vmw_extended_context_arg c_arg; |
| 88 |
int ret; |
| 89 |
|
| 90 |
VMW_FUNC; |
| 91 |
memset(&c_arg, 0, sizeof(c_arg)); |
| 92 |
c_arg.req = (vgpu10 ? drm_vmw_context_dx : drm_vmw_context_legacy); |
| 93 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, |
| 94 |
DRM_VMW_CREATE_EXTENDED_CONTEXT, |
| 95 |
&c_arg, sizeof(c_arg)); |
| 96 |
|
| 97 |
if (ret) |
| 98 |
return -1; |
| 99 |
|
| 100 |
vmw_printf("Context id is %d\n", c_arg.cid); |
| 101 |
return c_arg.rep.cid; |
| 102 |
} |
| 103 |
|
| 104 |
void |
| 105 |
vmw_ioctl_context_destroy(struct vmw_winsys_screen *vws, uint32 cid) |
| 106 |
{ |
| 107 |
struct drm_vmw_context_arg c_arg; |
| 108 |
|
| 109 |
VMW_FUNC; |
| 110 |
|
| 111 |
memset(&c_arg, 0, sizeof(c_arg)); |
| 112 |
c_arg.cid = cid; |
| 113 |
|
| 114 |
(void)drmCommandWrite(vws->ioctl.drm_fd, DRM_VMW_UNREF_CONTEXT, |
| 115 |
&c_arg, sizeof(c_arg)); |
| 116 |
|
| 117 |
} |
| 118 |
|
| 119 |
uint32 |
| 120 |
vmw_ioctl_surface_create(struct vmw_winsys_screen *vws, |
| 121 |
SVGA3dSurface1Flags flags, |
| 122 |
SVGA3dSurfaceFormat format, |
| 123 |
unsigned usage, |
| 124 |
SVGA3dSize size, |
| 125 |
uint32_t numFaces, uint32_t numMipLevels, |
| 126 |
unsigned sampleCount) |
| 127 |
{ |
| 128 |
union drm_vmw_surface_create_arg s_arg; |
| 129 |
struct drm_vmw_surface_create_req *req = &s_arg.req; |
| 130 |
struct drm_vmw_surface_arg *rep = &s_arg.rep; |
| 131 |
struct drm_vmw_size sizes[DRM_VMW_MAX_SURFACE_FACES* |
| 132 |
DRM_VMW_MAX_MIP_LEVELS]; |
| 133 |
struct drm_vmw_size *cur_size; |
| 134 |
uint32_t iFace; |
| 135 |
uint32_t iMipLevel; |
| 136 |
int ret; |
| 137 |
|
| 138 |
vmw_printf("%s flags %d format %d\n", __func__, flags, format); |
| 139 |
|
| 140 |
memset(&s_arg, 0, sizeof(s_arg)); |
| 141 |
req->flags = (uint32_t) flags; |
| 142 |
req->scanout = !!(usage & SVGA_SURFACE_USAGE_SCANOUT); |
| 143 |
req->format = (uint32_t) format; |
| 144 |
req->shareable = true; |
| 145 |
|
| 146 |
assert(numFaces * numMipLevels < DRM_VMW_MAX_SURFACE_FACES* |
| 147 |
DRM_VMW_MAX_MIP_LEVELS); |
| 148 |
cur_size = sizes; |
| 149 |
for (iFace = 0; iFace < numFaces; ++iFace) { |
| 150 |
SVGA3dSize mipSize = size; |
| 151 |
|
| 152 |
req->mip_levels[iFace] = numMipLevels; |
| 153 |
for (iMipLevel = 0; iMipLevel < numMipLevels; ++iMipLevel) { |
| 154 |
cur_size->width = mipSize.width; |
| 155 |
cur_size->height = mipSize.height; |
| 156 |
cur_size->depth = mipSize.depth; |
| 157 |
mipSize.width = MAX2(mipSize.width >> 1, 1); |
| 158 |
mipSize.height = MAX2(mipSize.height >> 1, 1); |
| 159 |
mipSize.depth = MAX2(mipSize.depth >> 1, 1); |
| 160 |
cur_size++; |
| 161 |
} |
| 162 |
} |
| 163 |
for (iFace = numFaces; iFace < SVGA3D_MAX_SURFACE_FACES; ++iFace) { |
| 164 |
req->mip_levels[iFace] = 0; |
| 165 |
} |
| 166 |
|
| 167 |
req->size_addr = (unsigned long)&sizes; |
| 168 |
|
| 169 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_CREATE_SURFACE, |
| 170 |
&s_arg, sizeof(s_arg)); |
| 171 |
|
| 172 |
if (ret) |
| 173 |
return -1; |
| 174 |
|
| 175 |
vmw_printf("Surface id is %d\n", rep->sid); |
| 176 |
|
| 177 |
return rep->sid; |
| 178 |
} |
| 179 |
|
| 180 |
|
| 181 |
uint32 |
| 182 |
vmw_ioctl_gb_surface_create(struct vmw_winsys_screen *vws, |
| 183 |
SVGA3dSurfaceAllFlags flags, |
| 184 |
SVGA3dSurfaceFormat format, |
| 185 |
unsigned usage, |
| 186 |
SVGA3dSize size, |
| 187 |
uint32_t numFaces, |
| 188 |
uint32_t numMipLevels, |
| 189 |
unsigned sampleCount, |
| 190 |
uint32_t buffer_handle, |
| 191 |
SVGA3dMSPattern multisamplePattern, |
| 192 |
SVGA3dMSQualityLevel qualityLevel, |
| 193 |
struct vmw_region **p_region) |
| 194 |
{ |
| 195 |
union { |
| 196 |
union drm_vmw_gb_surface_create_ext_arg ext_arg; |
| 197 |
union drm_vmw_gb_surface_create_arg arg; |
| 198 |
} s_arg; |
| 199 |
struct drm_vmw_gb_surface_create_rep *rep; |
| 200 |
struct vmw_region *region = NULL; |
| 201 |
int ret; |
| 202 |
|
| 203 |
vmw_printf("%s flags %d format %d\n", __func__, flags, format); |
| 204 |
|
| 205 |
if (p_region) { |
| 206 |
region = CALLOC_STRUCT(vmw_region); |
| 207 |
if (!region) |
| 208 |
return SVGA3D_INVALID_ID; |
| 209 |
} |
| 210 |
|
| 211 |
memset(&s_arg, 0, sizeof(s_arg)); |
| 212 |
|
| 213 |
if (vws->ioctl.have_drm_2_15) { |
| 214 |
struct drm_vmw_gb_surface_create_ext_req *req = &s_arg.ext_arg.req; |
| 215 |
rep = &s_arg.ext_arg.rep; |
| 216 |
|
| 217 |
req->version = drm_vmw_gb_surface_v1; |
| 218 |
req->multisample_pattern = multisamplePattern; |
| 219 |
req->quality_level = qualityLevel; |
| 220 |
req->buffer_byte_stride = 0; |
| 221 |
req->must_be_zero = 0; |
| 222 |
req->base.svga3d_flags = SVGA3D_FLAGS_LOWER_32(flags); |
| 223 |
req->svga3d_flags_upper_32_bits = SVGA3D_FLAGS_UPPER_32(flags); |
| 224 |
req->base.format = (uint32_t) format; |
| 225 |
|
| 226 |
if (usage & SVGA_SURFACE_USAGE_SCANOUT) |
| 227 |
req->base.drm_surface_flags |= drm_vmw_surface_flag_scanout; |
| 228 |
|
| 229 |
if ((usage & SVGA_SURFACE_USAGE_COHERENT) || vws->force_coherent) |
| 230 |
req->base.drm_surface_flags |= drm_vmw_surface_flag_coherent; |
| 231 |
|
| 232 |
req->base.drm_surface_flags |= drm_vmw_surface_flag_shareable; |
| 233 |
req->base.drm_surface_flags |= drm_vmw_surface_flag_create_buffer; |
| 234 |
req->base.base_size.width = size.width; |
| 235 |
req->base.base_size.height = size.height; |
| 236 |
req->base.base_size.depth = size.depth; |
| 237 |
req->base.mip_levels = numMipLevels; |
| 238 |
req->base.multisample_count = 0; |
| 239 |
req->base.autogen_filter = SVGA3D_TEX_FILTER_NONE; |
| 240 |
|
| 241 |
if (vws->base.have_vgpu10) { |
| 242 |
req->base.array_size = numFaces; |
| 243 |
req->base.multisample_count = sampleCount; |
| 244 |
} else { |
| 245 |
assert(numFaces * numMipLevels < DRM_VMW_MAX_SURFACE_FACES* |
| 246 |
DRM_VMW_MAX_MIP_LEVELS); |
| 247 |
req->base.array_size = 0; |
| 248 |
} |
| 249 |
|
| 250 |
req->base.buffer_handle = buffer_handle ? |
| 251 |
buffer_handle : SVGA3D_INVALID_ID; |
| 252 |
|
| 253 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, |
| 254 |
DRM_VMW_GB_SURFACE_CREATE_EXT, &s_arg.ext_arg, |
| 255 |
sizeof(s_arg.ext_arg)); |
| 256 |
|
| 257 |
if (ret) |
| 258 |
goto out_fail_create; |
| 259 |
} else { |
| 260 |
struct drm_vmw_gb_surface_create_req *req = &s_arg.arg.req; |
| 261 |
rep = &s_arg.arg.rep; |
| 262 |
|
| 263 |
req->svga3d_flags = (uint32_t) flags; |
| 264 |
req->format = (uint32_t) format; |
| 265 |
|
| 266 |
if (usage & SVGA_SURFACE_USAGE_SCANOUT) |
| 267 |
req->drm_surface_flags |= drm_vmw_surface_flag_scanout; |
| 268 |
|
| 269 |
req->drm_surface_flags |= drm_vmw_surface_flag_shareable; |
| 270 |
|
| 271 |
req->drm_surface_flags |= drm_vmw_surface_flag_create_buffer; |
| 272 |
req->base_size.width = size.width; |
| 273 |
req->base_size.height = size.height; |
| 274 |
req->base_size.depth = size.depth; |
| 275 |
req->mip_levels = numMipLevels; |
| 276 |
req->multisample_count = 0; |
| 277 |
req->autogen_filter = SVGA3D_TEX_FILTER_NONE; |
| 278 |
|
| 279 |
if (vws->base.have_vgpu10) { |
| 280 |
req->array_size = numFaces; |
| 281 |
req->multisample_count = sampleCount; |
| 282 |
} else { |
| 283 |
assert(numFaces * numMipLevels < DRM_VMW_MAX_SURFACE_FACES* |
| 284 |
DRM_VMW_MAX_MIP_LEVELS); |
| 285 |
req->array_size = 0; |
| 286 |
} |
| 287 |
|
| 288 |
req->buffer_handle = buffer_handle ? |
| 289 |
buffer_handle : SVGA3D_INVALID_ID; |
| 290 |
|
| 291 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GB_SURFACE_CREATE, |
| 292 |
&s_arg.arg, sizeof(s_arg.arg)); |
| 293 |
|
| 294 |
if (ret) |
| 295 |
goto out_fail_create; |
| 296 |
} |
| 297 |
|
| 298 |
if (p_region) { |
| 299 |
region->handle = rep->buffer_handle; |
| 300 |
region->map_handle = rep->buffer_map_handle; |
| 301 |
region->drm_fd = vws->ioctl.drm_fd; |
| 302 |
region->size = rep->backup_size; |
| 303 |
*p_region = region; |
| 304 |
} |
| 305 |
|
| 306 |
vmw_printf("Surface id is %d\n", rep->sid); |
| 307 |
return rep->handle; |
| 308 |
|
| 309 |
out_fail_create: |
| 310 |
FREE(region); |
| 311 |
return SVGA3D_INVALID_ID; |
| 312 |
} |
| 313 |
|
| 314 |
/** |
| 315 |
* vmw_ioctl_surface_req - Fill in a struct surface_req |
| 316 |
* |
| 317 |
* @vws: Winsys screen |
| 318 |
* @whandle: Surface handle |
| 319 |
* @req: The struct surface req to fill in |
| 320 |
* @needs_unref: This call takes a kernel surface reference that needs to |
| 321 |
* be unreferenced. |
| 322 |
* |
| 323 |
* Returns 0 on success, negative error type otherwise. |
| 324 |
* Fills in the surface_req structure according to handle type and kernel |
| 325 |
* capabilities. |
| 326 |
*/ |
| 327 |
static int |
| 328 |
vmw_ioctl_surface_req(const struct vmw_winsys_screen *vws, |
| 329 |
const struct winsys_handle *whandle, |
| 330 |
struct drm_vmw_surface_arg *req, |
| 331 |
bool *needs_unref) |
| 332 |
{ |
| 333 |
int ret; |
| 334 |
|
| 335 |
switch (whandle->type) { |
| 336 |
case WINSYS_HANDLE_TYPE_SHARED: |
| 337 |
case WINSYS_HANDLE_TYPE_KMS: |
| 338 |
*needs_unref = false; |
| 339 |
req->handle_type = DRM_VMW_HANDLE_LEGACY; |
| 340 |
req->sid = whandle->handle; |
| 341 |
break; |
| 342 |
case WINSYS_HANDLE_TYPE_FD: |
| 343 |
if (!vws->ioctl.have_drm_2_6) { |
| 344 |
uint32_t handle; |
| 345 |
|
| 346 |
ret = drmPrimeFDToHandle(vws->ioctl.drm_fd, whandle->handle, &handle); |
| 347 |
if (ret) { |
| 348 |
vmw_error("Failed to get handle from prime fd %d.\n", |
| 349 |
(int) whandle->handle); |
| 350 |
return -EINVAL; |
| 351 |
} |
| 352 |
|
| 353 |
*needs_unref = true; |
| 354 |
req->handle_type = DRM_VMW_HANDLE_LEGACY; |
| 355 |
req->sid = handle; |
| 356 |
} else { |
| 357 |
*needs_unref = false; |
| 358 |
req->handle_type = DRM_VMW_HANDLE_PRIME; |
| 359 |
req->sid = whandle->handle; |
| 360 |
} |
| 361 |
break; |
| 362 |
default: |
| 363 |
vmw_error("Attempt to import unsupported handle type %d.\n", |
| 364 |
whandle->type); |
| 365 |
return -EINVAL; |
| 366 |
} |
| 367 |
|
| 368 |
return 0; |
| 369 |
} |
| 370 |
|
| 371 |
/** |
| 372 |
* vmw_ioctl_gb_surface_ref - Put a reference on a guest-backed surface and |
| 373 |
* get surface information |
| 374 |
* |
| 375 |
* @vws: Screen to register the reference on |
| 376 |
* @handle: Kernel handle of the guest-backed surface |
| 377 |
* @flags: flags used when the surface was created |
| 378 |
* @format: Format used when the surface was created |
| 379 |
* @numMipLevels: Number of mipmap levels of the surface |
| 380 |
* @p_region: On successful return points to a newly allocated |
| 381 |
* struct vmw_region holding a reference to the surface backup buffer. |
| 382 |
* |
| 383 |
* Returns 0 on success, a system error on failure. |
| 384 |
*/ |
| 385 |
int |
| 386 |
vmw_ioctl_gb_surface_ref(struct vmw_winsys_screen *vws, |
| 387 |
const struct winsys_handle *whandle, |
| 388 |
SVGA3dSurfaceAllFlags *flags, |
| 389 |
SVGA3dSurfaceFormat *format, |
| 390 |
uint32_t *numMipLevels, |
| 391 |
uint32_t *handle, |
| 392 |
struct vmw_region **p_region) |
| 393 |
{ |
| 394 |
struct vmw_region *region = NULL; |
| 395 |
bool needs_unref = false; |
| 396 |
int ret; |
| 397 |
|
| 398 |
assert(p_region != NULL); |
| 399 |
region = CALLOC_STRUCT(vmw_region); |
| 400 |
if (!region) |
| 401 |
return -ENOMEM; |
| 402 |
|
| 403 |
if (vws->ioctl.have_drm_2_15) { |
| 404 |
union drm_vmw_gb_surface_reference_ext_arg s_arg; |
| 405 |
struct drm_vmw_surface_arg *req = &s_arg.req; |
| 406 |
struct drm_vmw_gb_surface_ref_ext_rep *rep = &s_arg.rep; |
| 407 |
|
| 408 |
memset(&s_arg, 0, sizeof(s_arg)); |
| 409 |
ret = vmw_ioctl_surface_req(vws, whandle, req, &needs_unref); |
| 410 |
if (ret) |
| 411 |
goto out_fail_req; |
| 412 |
|
| 413 |
*handle = req->sid; |
| 414 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GB_SURFACE_REF_EXT, |
| 415 |
&s_arg, sizeof(s_arg)); |
| 416 |
|
| 417 |
if (ret) |
| 418 |
goto out_fail_ref; |
| 419 |
|
| 420 |
region->handle = rep->crep.buffer_handle; |
| 421 |
region->map_handle = rep->crep.buffer_map_handle; |
| 422 |
region->drm_fd = vws->ioctl.drm_fd; |
| 423 |
region->size = rep->crep.backup_size; |
| 424 |
*p_region = region; |
| 425 |
|
| 426 |
*handle = rep->crep.handle; |
| 427 |
*flags = SVGA3D_FLAGS_64(rep->creq.svga3d_flags_upper_32_bits, |
| 428 |
rep->creq.base.svga3d_flags); |
| 429 |
*format = rep->creq.base.format; |
| 430 |
*numMipLevels = rep->creq.base.mip_levels; |
| 431 |
} else { |
| 432 |
union drm_vmw_gb_surface_reference_arg s_arg; |
| 433 |
struct drm_vmw_surface_arg *req = &s_arg.req; |
| 434 |
struct drm_vmw_gb_surface_ref_rep *rep = &s_arg.rep; |
| 435 |
|
| 436 |
memset(&s_arg, 0, sizeof(s_arg)); |
| 437 |
ret = vmw_ioctl_surface_req(vws, whandle, req, &needs_unref); |
| 438 |
if (ret) |
| 439 |
goto out_fail_req; |
| 440 |
|
| 441 |
*handle = req->sid; |
| 442 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GB_SURFACE_REF, |
| 443 |
&s_arg, sizeof(s_arg)); |
| 444 |
|
| 445 |
if (ret) |
| 446 |
goto out_fail_ref; |
| 447 |
|
| 448 |
region->handle = rep->crep.buffer_handle; |
| 449 |
region->map_handle = rep->crep.buffer_map_handle; |
| 450 |
region->drm_fd = vws->ioctl.drm_fd; |
| 451 |
region->size = rep->crep.backup_size; |
| 452 |
*p_region = region; |
| 453 |
|
| 454 |
*handle = rep->crep.handle; |
| 455 |
*flags = rep->creq.svga3d_flags; |
| 456 |
*format = rep->creq.format; |
| 457 |
*numMipLevels = rep->creq.mip_levels; |
| 458 |
} |
| 459 |
|
| 460 |
vmw_printf("%s flags %d format %d\n", __func__, *flags, *format); |
| 461 |
|
| 462 |
if (needs_unref) |
| 463 |
vmw_ioctl_surface_destroy(vws, *handle); |
| 464 |
|
| 465 |
return 0; |
| 466 |
out_fail_ref: |
| 467 |
if (needs_unref) |
| 468 |
vmw_ioctl_surface_destroy(vws, *handle); |
| 469 |
out_fail_req: |
| 470 |
FREE(region); |
| 471 |
return ret; |
| 472 |
} |
| 473 |
|
| 474 |
void |
| 475 |
vmw_ioctl_surface_destroy(struct vmw_winsys_screen *vws, uint32 sid) |
| 476 |
{ |
| 477 |
struct drm_vmw_surface_arg s_arg; |
| 478 |
|
| 479 |
VMW_FUNC; |
| 480 |
|
| 481 |
memset(&s_arg, 0, sizeof(s_arg)); |
| 482 |
s_arg.sid = sid; |
| 483 |
|
| 484 |
(void)drmCommandWrite(vws->ioctl.drm_fd, DRM_VMW_UNREF_SURFACE, |
| 485 |
&s_arg, sizeof(s_arg)); |
| 486 |
} |
| 487 |
|
| 488 |
void |
| 489 |
vmw_ioctl_command(struct vmw_winsys_screen *vws, int32_t cid, |
| 490 |
uint32_t throttle_us, void *commands, uint32_t size, |
| 491 |
struct pipe_fence_handle **pfence, int32_t imported_fence_fd, |
| 492 |
uint32_t flags) |
| 493 |
{ |
| 494 |
struct drm_vmw_execbuf_arg arg; |
| 495 |
struct drm_vmw_fence_rep rep; |
| 496 |
int ret; |
| 497 |
int argsize; |
| 498 |
|
| 499 |
#if MESA_DEBUG |
| 500 |
{ |
| 501 |
static bool firsttime = true; |
| 502 |
static bool debug = false; |
| 503 |
static bool skip = false; |
| 504 |
if (firsttime) { |
| 505 |
debug = debug_get_bool_option("SVGA_DUMP_CMD", false); |
| 506 |
skip = debug_get_bool_option("SVGA_SKIP_CMD", false); |
| 507 |
} |
| 508 |
if (debug) { |
| 509 |
VMW_FUNC; |
| 510 |
svga_dump_commands(commands, size); |
| 511 |
} |
| 512 |
firsttime = false; |
| 513 |
if (skip) { |
| 514 |
size = 0; |
| 515 |
} |
| 516 |
} |
| 517 |
#endif |
| 518 |
|
| 519 |
memset(&arg, 0, sizeof(arg)); |
| 520 |
memset(&rep, 0, sizeof(rep)); |
| 521 |
|
| 522 |
if (flags & SVGA_HINT_FLAG_EXPORT_FENCE_FD) { |
| 523 |
arg.flags |= DRM_VMW_EXECBUF_FLAG_EXPORT_FENCE_FD; |
| 524 |
} |
| 525 |
|
| 526 |
if (imported_fence_fd != -1) { |
| 527 |
arg.flags |= DRM_VMW_EXECBUF_FLAG_IMPORT_FENCE_FD; |
| 528 |
} |
| 529 |
|
| 530 |
rep.error = -EFAULT; |
| 531 |
if (pfence) |
| 532 |
arg.fence_rep = (unsigned long)&rep; |
| 533 |
arg.commands = (unsigned long)commands; |
| 534 |
arg.command_size = size; |
| 535 |
arg.throttle_us = throttle_us; |
| 536 |
arg.version = vws->ioctl.drm_execbuf_version; |
| 537 |
arg.context_handle = (vws->base.have_vgpu10 ? cid : SVGA3D_INVALID_ID); |
| 538 |
|
| 539 |
/* Older DRM module requires this to be zero */ |
| 540 |
if (vws->base.have_fence_fd) |
| 541 |
arg.imported_fence_fd = imported_fence_fd; |
| 542 |
|
| 543 |
/* In DRM_VMW_EXECBUF_VERSION 1, the drm_vmw_execbuf_arg structure ends with |
| 544 |
* the flags field. The structure size sent to drmCommandWrite must match |
| 545 |
* the drm_execbuf_version. Otherwise, an invalid value will be returned. |
| 546 |
*/ |
| 547 |
argsize = vws->ioctl.drm_execbuf_version > 1 ? sizeof(arg) : |
| 548 |
offsetof(struct drm_vmw_execbuf_arg, context_handle); |
| 549 |
do { |
| 550 |
ret = drmCommandWrite(vws->ioctl.drm_fd, DRM_VMW_EXECBUF, &arg, argsize); |
| 551 |
if (ret == -EBUSY) |
| 552 |
usleep(1000); |
| 553 |
} while (ret == -ERESTART || ret == -EBUSY); |
| 554 |
if (ret) { |
| 555 |
vmw_error("%s error %s.\n", __func__, strerror(-ret)); |
| 556 |
abort(); |
| 557 |
} |
| 558 |
|
| 559 |
if (rep.error) { |
| 560 |
|
| 561 |
/* |
| 562 |
* Kernel has already synced, or caller requested no fence. |
| 563 |
*/ |
| 564 |
if (pfence) |
| 565 |
*pfence = NULL; |
| 566 |
} else { |
| 567 |
if (pfence) { |
| 568 |
vmw_fences_signal(vws->fence_ops, rep.passed_seqno, rep.seqno, |
| 569 |
true); |
| 570 |
|
| 571 |
/* Older DRM module will set this to zero, but -1 is the proper FD |
| 572 |
* to use for no Fence FD support */ |
| 573 |
if (!vws->base.have_fence_fd) |
| 574 |
rep.fd = -1; |
| 575 |
|
| 576 |
*pfence = vmw_fence_create(vws->fence_ops, rep.handle, |
| 577 |
rep.seqno, rep.mask, rep.fd); |
| 578 |
if (*pfence == NULL) { |
| 579 |
/* |
| 580 |
* Fence creation failed. Need to sync. |
| 581 |
*/ |
| 582 |
(void) vmw_ioctl_fence_finish(vws, rep.handle, rep.mask); |
| 583 |
vmw_ioctl_fence_unref(vws, rep.handle); |
| 584 |
} |
| 585 |
} |
| 586 |
} |
| 587 |
} |
| 588 |
|
| 589 |
|
| 590 |
struct vmw_region * |
| 591 |
vmw_ioctl_region_create(struct vmw_winsys_screen *vws, uint32_t size) |
| 592 |
{ |
| 593 |
struct vmw_region *region; |
| 594 |
union drm_vmw_alloc_dmabuf_arg arg; |
| 595 |
struct drm_vmw_alloc_dmabuf_req *req = &arg.req; |
| 596 |
struct drm_vmw_dmabuf_rep *rep = &arg.rep; |
| 597 |
int ret; |
| 598 |
|
| 599 |
vmw_printf("%s: size = %u\n", __func__, size); |
| 600 |
|
| 601 |
region = CALLOC_STRUCT(vmw_region); |
| 602 |
if (!region) |
| 603 |
goto out_err1; |
| 604 |
|
| 605 |
memset(&arg, 0, sizeof(arg)); |
| 606 |
req->size = size; |
| 607 |
do { |
| 608 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_ALLOC_DMABUF, &arg, |
| 609 |
sizeof(arg)); |
| 610 |
} while (ret == -ERESTART); |
| 611 |
|
| 612 |
if (ret) { |
| 613 |
vmw_error("IOCTL failed %d: %s\n", ret, strerror(-ret)); |
| 614 |
goto out_err1; |
| 615 |
} |
| 616 |
|
| 617 |
region->data = NULL; |
| 618 |
region->handle = rep->handle; |
| 619 |
region->map_handle = rep->map_handle; |
| 620 |
region->map_count = 0; |
| 621 |
region->size = size; |
| 622 |
region->drm_fd = vws->ioctl.drm_fd; |
| 623 |
|
| 624 |
vmw_printf(" gmrId = %u, offset = %u\n", |
| 625 |
region->ptr.gmrId, region->ptr.offset); |
| 626 |
|
| 627 |
return region; |
| 628 |
|
| 629 |
out_err1: |
| 630 |
FREE(region); |
| 631 |
return NULL; |
| 632 |
} |
| 633 |
|
| 634 |
void |
| 635 |
vmw_ioctl_region_destroy(struct vmw_region *region) |
| 636 |
{ |
| 637 |
struct drm_vmw_unref_dmabuf_arg arg; |
| 638 |
|
| 639 |
vmw_printf("%s: gmrId = %u, offset = %u\n", __func__, |
| 640 |
region->ptr.gmrId, region->ptr.offset); |
| 641 |
|
| 642 |
if (region->data) { |
| 643 |
os_munmap(region->data, region->size); |
| 644 |
region->data = NULL; |
| 645 |
} |
| 646 |
|
| 647 |
memset(&arg, 0, sizeof(arg)); |
| 648 |
arg.handle = region->handle; |
| 649 |
drmCommandWrite(region->drm_fd, DRM_VMW_UNREF_DMABUF, &arg, sizeof(arg)); |
| 650 |
|
| 651 |
FREE(region); |
| 652 |
} |
| 653 |
|
| 654 |
SVGAGuestPtr |
| 655 |
vmw_ioctl_region_ptr(struct vmw_region *region) |
| 656 |
{ |
| 657 |
SVGAGuestPtr ptr = {region->handle, 0}; |
| 658 |
return ptr; |
| 659 |
} |
| 660 |
|
| 661 |
void * |
| 662 |
vmw_ioctl_region_map(struct vmw_region *region) |
| 663 |
{ |
| 664 |
void *map; |
| 665 |
|
| 666 |
vmw_printf("%s: gmrId = %u, offset = %u\n", __func__, |
| 667 |
region->ptr.gmrId, region->ptr.offset); |
| 668 |
|
| 669 |
if (region->data == NULL) { |
| 670 |
map = os_mmap(NULL, region->size, PROT_READ | PROT_WRITE, MAP_SHARED, |
| 671 |
region->drm_fd, region->map_handle); |
| 672 |
if (map == MAP_FAILED) { |
| 673 |
vmw_error("%s: Map failed.\n", __func__); |
| 674 |
return NULL; |
| 675 |
} |
| 676 |
|
| 677 |
// MADV_HUGEPAGE only exists on Linux |
| 678 |
#ifdef MADV_HUGEPAGE |
| 679 |
(void) madvise(map, region->size, MADV_HUGEPAGE); |
| 680 |
#endif |
| 681 |
region->data = map; |
| 682 |
} |
| 683 |
|
| 684 |
++region->map_count; |
| 685 |
|
| 686 |
return region->data; |
| 687 |
} |
| 688 |
|
| 689 |
void |
| 690 |
vmw_ioctl_region_unmap(struct vmw_region *region) |
| 691 |
{ |
| 692 |
vmw_printf("%s: gmrId = %u, offset = %u\n", __func__, |
| 693 |
region->ptr.gmrId, region->ptr.offset); |
| 694 |
|
| 695 |
--region->map_count; |
| 696 |
os_munmap(region->data, region->size); |
| 697 |
region->data = NULL; |
| 698 |
} |
| 699 |
|
| 700 |
/** |
| 701 |
* vmw_ioctl_syncforcpu - Synchronize a buffer object for CPU usage |
| 702 |
* |
| 703 |
* @region: Pointer to a struct vmw_region representing the buffer object. |
| 704 |
* @dont_block: Dont wait for GPU idle, but rather return -EBUSY if the |
| 705 |
* GPU is busy with the buffer object. |
| 706 |
* @readonly: Hint that the CPU access is read-only. |
| 707 |
* @allow_cs: Allow concurrent command submission while the buffer is |
| 708 |
* synchronized for CPU. If FALSE command submissions referencing the |
| 709 |
* buffer will block until a corresponding call to vmw_ioctl_releasefromcpu. |
| 710 |
* |
| 711 |
* This function idles any GPU activities touching the buffer and blocks |
| 712 |
* command submission of commands referencing the buffer, even from |
| 713 |
* other processes. |
| 714 |
*/ |
| 715 |
int |
| 716 |
vmw_ioctl_syncforcpu(struct vmw_region *region, |
| 717 |
bool dont_block, |
| 718 |
bool readonly, |
| 719 |
bool allow_cs) |
| 720 |
{ |
| 721 |
struct drm_vmw_synccpu_arg arg; |
| 722 |
int ret; |
| 723 |
|
| 724 |
memset(&arg, 0, sizeof(arg)); |
| 725 |
arg.op = drm_vmw_synccpu_grab; |
| 726 |
arg.handle = region->handle; |
| 727 |
arg.flags = drm_vmw_synccpu_read; |
| 728 |
if (!readonly) |
| 729 |
arg.flags |= drm_vmw_synccpu_write; |
| 730 |
if (dont_block) |
| 731 |
arg.flags |= drm_vmw_synccpu_dontblock; |
| 732 |
if (allow_cs) |
| 733 |
arg.flags |= drm_vmw_synccpu_allow_cs; |
| 734 |
|
| 735 |
do { |
| 736 |
ret = drmCommandWrite(region->drm_fd, DRM_VMW_SYNCCPU, &arg, sizeof(arg)); |
| 737 |
if (ret == -EBUSY) |
| 738 |
usleep(1000); |
| 739 |
} while (ret == -ERESTART || ret == -EBUSY); |
| 740 |
|
| 741 |
if (ret) |
| 742 |
vmw_error("%s Failed synccpu with error %s.\n", __func__, strerror(-ret)); |
| 743 |
|
| 744 |
return ret; |
| 745 |
} |
| 746 |
|
| 747 |
/** |
| 748 |
* vmw_ioctl_releasefromcpu - Undo a previous syncforcpu. |
| 749 |
* |
| 750 |
* @region: Pointer to a struct vmw_region representing the buffer object. |
| 751 |
* @readonly: Should hold the same value as the matching syncforcpu call. |
| 752 |
* @allow_cs: Should hold the same value as the matching syncforcpu call. |
| 753 |
*/ |
| 754 |
void |
| 755 |
vmw_ioctl_releasefromcpu(struct vmw_region *region, |
| 756 |
bool readonly, |
| 757 |
bool allow_cs) |
| 758 |
{ |
| 759 |
struct drm_vmw_synccpu_arg arg; |
| 760 |
|
| 761 |
memset(&arg, 0, sizeof(arg)); |
| 762 |
arg.op = drm_vmw_synccpu_release; |
| 763 |
arg.handle = region->handle; |
| 764 |
arg.flags = drm_vmw_synccpu_read; |
| 765 |
if (!readonly) |
| 766 |
arg.flags |= drm_vmw_synccpu_write; |
| 767 |
if (allow_cs) |
| 768 |
arg.flags |= drm_vmw_synccpu_allow_cs; |
| 769 |
|
| 770 |
(void) drmCommandWrite(region->drm_fd, DRM_VMW_SYNCCPU, &arg, sizeof(arg)); |
| 771 |
} |
| 772 |
|
| 773 |
void |
| 774 |
vmw_ioctl_fence_unref(struct vmw_winsys_screen *vws, |
| 775 |
uint32_t handle) |
| 776 |
{ |
| 777 |
struct drm_vmw_fence_arg arg; |
| 778 |
int ret; |
| 779 |
|
| 780 |
memset(&arg, 0, sizeof(arg)); |
| 781 |
arg.handle = handle; |
| 782 |
|
| 783 |
ret = drmCommandWrite(vws->ioctl.drm_fd, DRM_VMW_FENCE_UNREF, |
| 784 |
&arg, sizeof(arg)); |
| 785 |
if (ret != 0) |
| 786 |
vmw_error("%s Failed\n", __func__); |
| 787 |
} |
| 788 |
|
| 789 |
static inline uint32_t |
| 790 |
vmw_drm_fence_flags(uint32_t flags) |
| 791 |
{ |
| 792 |
uint32_t dflags = 0; |
| 793 |
|
| 794 |
if (flags & SVGA_FENCE_FLAG_EXEC) |
| 795 |
dflags |= DRM_VMW_FENCE_FLAG_EXEC; |
| 796 |
if (flags & SVGA_FENCE_FLAG_QUERY) |
| 797 |
dflags |= DRM_VMW_FENCE_FLAG_QUERY; |
| 798 |
|
| 799 |
return dflags; |
| 800 |
} |
| 801 |
|
| 802 |
|
| 803 |
int |
| 804 |
vmw_ioctl_fence_signalled(struct vmw_winsys_screen *vws, |
| 805 |
uint32_t handle, |
| 806 |
uint32_t flags) |
| 807 |
{ |
| 808 |
struct drm_vmw_fence_signaled_arg arg; |
| 809 |
uint32_t vflags = vmw_drm_fence_flags(flags); |
| 810 |
int ret; |
| 811 |
|
| 812 |
memset(&arg, 0, sizeof(arg)); |
| 813 |
arg.handle = handle; |
| 814 |
arg.flags = vflags; |
| 815 |
|
| 816 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_FENCE_SIGNALED, |
| 817 |
&arg, sizeof(arg)); |
| 818 |
|
| 819 |
if (ret != 0) |
| 820 |
return ret; |
| 821 |
|
| 822 |
vmw_fences_signal(vws->fence_ops, arg.passed_seqno, 0, false); |
| 823 |
|
| 824 |
return (arg.signaled) ? 0 : -1; |
| 825 |
} |
| 826 |
|
| 827 |
|
| 828 |
|
| 829 |
int |
| 830 |
vmw_ioctl_fence_finish(struct vmw_winsys_screen *vws, |
| 831 |
uint32_t handle, |
| 832 |
uint32_t flags) |
| 833 |
{ |
| 834 |
struct drm_vmw_fence_wait_arg arg; |
| 835 |
uint32_t vflags = vmw_drm_fence_flags(flags); |
| 836 |
int ret; |
| 837 |
|
| 838 |
memset(&arg, 0, sizeof(arg)); |
| 839 |
|
| 840 |
arg.handle = handle; |
| 841 |
arg.timeout_us = VMW_FENCE_TIMEOUT_SECONDS*1000000; |
| 842 |
arg.lazy = 0; |
| 843 |
arg.flags = vflags; |
| 844 |
|
| 845 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_FENCE_WAIT, |
| 846 |
&arg, sizeof(arg)); |
| 847 |
|
| 848 |
if (ret != 0) |
| 849 |
vmw_error("%s Failed\n", __func__); |
| 850 |
|
| 851 |
return 0; |
| 852 |
} |
| 853 |
|
| 854 |
uint32 |
| 855 |
vmw_ioctl_shader_create(struct vmw_winsys_screen *vws, |
| 856 |
SVGA3dShaderType type, |
| 857 |
uint32 code_len) |
| 858 |
{ |
| 859 |
struct drm_vmw_shader_create_arg sh_arg; |
| 860 |
int ret; |
| 861 |
|
| 862 |
VMW_FUNC; |
| 863 |
|
| 864 |
memset(&sh_arg, 0, sizeof(sh_arg)); |
| 865 |
|
| 866 |
sh_arg.size = code_len; |
| 867 |
sh_arg.buffer_handle = SVGA3D_INVALID_ID; |
| 868 |
sh_arg.shader_handle = SVGA3D_INVALID_ID; |
| 869 |
switch (type) { |
| 870 |
case SVGA3D_SHADERTYPE_VS: |
| 871 |
sh_arg.shader_type = drm_vmw_shader_type_vs; |
| 872 |
break; |
| 873 |
case SVGA3D_SHADERTYPE_PS: |
| 874 |
sh_arg.shader_type = drm_vmw_shader_type_ps; |
| 875 |
break; |
| 876 |
default: |
| 877 |
assert(!"Invalid shader type."); |
| 878 |
break; |
| 879 |
} |
| 880 |
|
| 881 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_CREATE_SHADER, |
| 882 |
&sh_arg, sizeof(sh_arg)); |
| 883 |
|
| 884 |
if (ret) |
| 885 |
return SVGA3D_INVALID_ID; |
| 886 |
|
| 887 |
return sh_arg.shader_handle; |
| 888 |
} |
| 889 |
|
| 890 |
void |
| 891 |
vmw_ioctl_shader_destroy(struct vmw_winsys_screen *vws, uint32 shid) |
| 892 |
{ |
| 893 |
struct drm_vmw_shader_arg sh_arg; |
| 894 |
|
| 895 |
VMW_FUNC; |
| 896 |
|
| 897 |
memset(&sh_arg, 0, sizeof(sh_arg)); |
| 898 |
sh_arg.handle = shid; |
| 899 |
|
| 900 |
(void)drmCommandWrite(vws->ioctl.drm_fd, DRM_VMW_UNREF_SHADER, |
| 901 |
&sh_arg, sizeof(sh_arg)); |
| 902 |
|
| 903 |
} |
| 904 |
|
| 905 |
struct svga_3d_compat_cap { |
| 906 |
SVGA3dFifoCapsRecordHeader header; |
| 907 |
SVGA3dFifoCapPair pairs[SVGA3D_DEVCAP_MAX]; |
| 908 |
}; |
| 909 |
|
| 910 |
static int |
| 911 |
vmw_ioctl_parse_caps(struct vmw_winsys_screen *vws, |
| 912 |
const uint32_t *cap_buffer) |
| 913 |
{ |
| 914 |
int i; |
| 915 |
|
| 916 |
if (vws->base.have_gb_objects) { |
| 917 |
for (i = 0; i < vws->ioctl.num_cap_3d; ++i) { |
| 918 |
vws->ioctl.cap_3d[i].has_cap = true; |
| 919 |
vws->ioctl.cap_3d[i].result.u = cap_buffer[i]; |
| 920 |
} |
| 921 |
return 0; |
| 922 |
} else { |
| 923 |
const uint32 *capsBlock; |
| 924 |
const struct svga_3d_compat_cap *capsRecord = NULL; |
| 925 |
uint32 offset; |
| 926 |
const SVGA3dFifoCapPair *capArray; |
| 927 |
int numCaps, index; |
| 928 |
|
| 929 |
/* |
| 930 |
* Search linearly through the caps block records for the specified type. |
| 931 |
*/ |
| 932 |
capsBlock = cap_buffer; |
| 933 |
for (offset = 0; capsBlock[offset] != 0; offset += capsBlock[offset]) { |
| 934 |
const struct svga_3d_compat_cap *record; |
| 935 |
assert(offset < SVGA_FIFO_3D_CAPS_SIZE); |
| 936 |
record = (const struct svga_3d_compat_cap *) (capsBlock + offset); |
| 937 |
if ((record->header.type >= 0) && |
| 938 |
(record->header.type <= SVGA3D_DEVCAP_MAX) && |
| 939 |
(!capsRecord || (record->header.type > capsRecord->header.type))) { |
| 940 |
capsRecord = record; |
| 941 |
} |
| 942 |
} |
| 943 |
|
| 944 |
if (!capsRecord) |
| 945 |
return -1; |
| 946 |
|
| 947 |
/* |
| 948 |
* Calculate the number of caps from the size of the record. |
| 949 |
*/ |
| 950 |
capArray = (const SVGA3dFifoCapPair *) capsRecord->pairs; |
| 951 |
numCaps = (int) ((capsRecord->header.length * sizeof(uint32) - |
| 952 |
sizeof capsRecord->header) / (2 * sizeof(uint32))); |
| 953 |
|
| 954 |
for (i = 0; i < numCaps; i++) { |
| 955 |
index = capArray[i][0]; |
| 956 |
if (index < vws->ioctl.num_cap_3d) { |
| 957 |
vws->ioctl.cap_3d[index].has_cap = true; |
| 958 |
vws->ioctl.cap_3d[index].result.u = capArray[i][1]; |
| 959 |
} else { |
| 960 |
debug_printf("Unknown devcaps seen: %d\n", index); |
| 961 |
} |
| 962 |
} |
| 963 |
} |
| 964 |
return 0; |
| 965 |
} |
| 966 |
|
| 967 |
bool |
| 968 |
vmw_ioctl_init(struct vmw_winsys_screen *vws) |
| 969 |
{ |
| 970 |
struct drm_vmw_getparam_arg gp_arg; |
| 971 |
struct drm_vmw_get_3d_cap_arg cap_arg; |
| 972 |
unsigned int size; |
| 973 |
int ret; |
| 974 |
uint32_t *cap_buffer; |
| 975 |
drmVersionPtr version; |
| 976 |
bool drm_gb_capable; |
| 977 |
bool have_drm_2_5; |
| 978 |
const char *getenv_val; |
| 979 |
|
| 980 |
VMW_FUNC; |
| 981 |
|
| 982 |
version = drmGetVersion(vws->ioctl.drm_fd); |
| 983 |
if (!version) |
| 984 |
goto out_no_version; |
| 985 |
|
| 986 |
have_drm_2_5 = version->version_major > 2 || |
| 987 |
(version->version_major == 2 && version->version_minor > 4); |
| 988 |
vws->ioctl.have_drm_2_6 = version->version_major > 2 || |
| 989 |
(version->version_major == 2 && version->version_minor > 5); |
| 990 |
vws->ioctl.have_drm_2_9 = version->version_major > 2 || |
| 991 |
(version->version_major == 2 && version->version_minor > 8); |
| 992 |
vws->ioctl.have_drm_2_15 = version->version_major > 2 || |
| 993 |
(version->version_major == 2 && version->version_minor > 14); |
| 994 |
vws->ioctl.have_drm_2_16 = version->version_major > 2 || |
| 995 |
(version->version_major == 2 && version->version_minor > 15); |
| 996 |
vws->ioctl.have_drm_2_17 = version->version_major > 2 || |
| 997 |
(version->version_major == 2 && version->version_minor > 16); |
| 998 |
vws->ioctl.have_drm_2_18 = version->version_major > 2 || |
| 999 |
(version->version_major == 2 && version->version_minor > 17); |
| 1000 |
vws->ioctl.have_drm_2_19 = version->version_major > 2 || |
| 1001 |
(version->version_major == 2 && version->version_minor > 18); |
| 1002 |
vws->ioctl.have_drm_2_20 = version->version_major > 2 || |
| 1003 |
(version->version_major == 2 && version->version_minor > 19); |
| 1004 |
|
| 1005 |
vws->ioctl.drm_execbuf_version = vws->ioctl.have_drm_2_9 ? 2 : 1; |
| 1006 |
|
| 1007 |
drm_gb_capable = have_drm_2_5; |
| 1008 |
|
| 1009 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1010 |
gp_arg.param = DRM_VMW_PARAM_3D; |
| 1011 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1012 |
&gp_arg, sizeof(gp_arg)); |
| 1013 |
if (ret || gp_arg.value == 0) { |
| 1014 |
vmw_error("No 3D enabled (%i, %s).\n", ret, strerror(-ret)); |
| 1015 |
goto out_no_3d; |
| 1016 |
} |
| 1017 |
|
| 1018 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1019 |
gp_arg.param = DRM_VMW_PARAM_FIFO_HW_VERSION; |
| 1020 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1021 |
&gp_arg, sizeof(gp_arg)); |
| 1022 |
if (ret) { |
| 1023 |
vmw_error("Failed to get fifo hw version (%i, %s).\n", |
| 1024 |
ret, strerror(-ret)); |
| 1025 |
goto out_no_3d; |
| 1026 |
} |
| 1027 |
vws->ioctl.hwversion = gp_arg.value; |
| 1028 |
getenv_val = os_get_option("SVGA_FORCE_HOST_BACKED"); |
| 1029 |
if (!getenv_val || strcmp(getenv_val, "0") == 0) { |
| 1030 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1031 |
gp_arg.param = DRM_VMW_PARAM_HW_CAPS; |
| 1032 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1033 |
&gp_arg, sizeof(gp_arg)); |
| 1034 |
} else { |
| 1035 |
ret = -EINVAL; |
| 1036 |
} |
| 1037 |
if (ret) |
| 1038 |
vws->base.have_gb_objects = false; |
| 1039 |
else |
| 1040 |
vws->base.have_gb_objects = |
| 1041 |
!!(gp_arg.value & (uint64_t) SVGA_CAP_GBOBJECTS); |
| 1042 |
|
| 1043 |
if (vws->base.have_gb_objects && !drm_gb_capable) |
| 1044 |
goto out_no_3d; |
| 1045 |
|
| 1046 |
vws->base.have_vgpu10 = false; |
| 1047 |
vws->base.have_sm4_1 = false; |
| 1048 |
vws->base.have_intra_surface_copy = false; |
| 1049 |
|
| 1050 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1051 |
gp_arg.param = DRM_VMW_PARAM_DEVICE_ID; |
| 1052 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1053 |
&gp_arg, sizeof(gp_arg)); |
| 1054 |
if (ret || gp_arg.value == 0) { |
| 1055 |
vws->base.device_id = 0x0405; /* assume SVGA II */ |
| 1056 |
} else { |
| 1057 |
vws->base.device_id = gp_arg.value; |
| 1058 |
} |
| 1059 |
|
| 1060 |
if (vws->base.have_gb_objects) { |
| 1061 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1062 |
gp_arg.param = DRM_VMW_PARAM_MAX_MOB_MEMORY; |
| 1063 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1064 |
&gp_arg, sizeof(gp_arg)); |
| 1065 |
if (ret) { |
| 1066 |
/* Just guess a large enough value. */ |
| 1067 |
vws->ioctl.max_mob_memory = 256*1024*1024; |
| 1068 |
} else { |
| 1069 |
vws->ioctl.max_mob_memory = gp_arg.value; |
| 1070 |
} |
| 1071 |
|
| 1072 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1073 |
gp_arg.param = DRM_VMW_PARAM_MAX_MOB_SIZE; |
| 1074 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1075 |
&gp_arg, sizeof(gp_arg)); |
| 1076 |
|
| 1077 |
if (ret || gp_arg.value == 0) { |
| 1078 |
vws->ioctl.max_texture_size = VMW_MAX_DEFAULT_TEXTURE_SIZE; |
| 1079 |
} else { |
| 1080 |
vws->ioctl.max_texture_size = gp_arg.value; |
| 1081 |
} |
| 1082 |
|
| 1083 |
/* Never early flush surfaces, mobs do accounting. */ |
| 1084 |
vws->ioctl.max_surface_memory = -1; |
| 1085 |
|
| 1086 |
if (vws->ioctl.have_drm_2_9) { |
| 1087 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1088 |
gp_arg.param = DRM_VMW_PARAM_DX; |
| 1089 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1090 |
&gp_arg, sizeof(gp_arg)); |
| 1091 |
if (ret == 0 && gp_arg.value != 0) { |
| 1092 |
const char *vgpu10_val; |
| 1093 |
|
| 1094 |
debug_printf("Have VGPU10 interface and hardware.\n"); |
| 1095 |
vws->base.have_vgpu10 = true; |
| 1096 |
vgpu10_val = os_get_option("SVGA_VGPU10"); |
| 1097 |
if (vgpu10_val && strcmp(vgpu10_val, "0") == 0) { |
| 1098 |
debug_printf("Disabling VGPU10 interface.\n"); |
| 1099 |
vws->base.have_vgpu10 = false; |
| 1100 |
} else { |
| 1101 |
debug_printf("Enabling VGPU10 interface.\n"); |
| 1102 |
} |
| 1103 |
} |
| 1104 |
} |
| 1105 |
|
| 1106 |
if (vws->ioctl.have_drm_2_15 && vws->base.have_vgpu10) { |
| 1107 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1108 |
gp_arg.param = DRM_VMW_PARAM_HW_CAPS2; |
| 1109 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1110 |
&gp_arg, sizeof(gp_arg)); |
| 1111 |
if (ret == 0 && gp_arg.value != 0) { |
| 1112 |
vws->base.have_intra_surface_copy = true; |
| 1113 |
} |
| 1114 |
|
| 1115 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1116 |
gp_arg.param = DRM_VMW_PARAM_SM4_1; |
| 1117 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1118 |
&gp_arg, sizeof(gp_arg)); |
| 1119 |
if (ret == 0 && gp_arg.value != 0) { |
| 1120 |
vws->base.have_sm4_1 = true; |
| 1121 |
} |
| 1122 |
} |
| 1123 |
|
| 1124 |
if (vws->ioctl.have_drm_2_18 && vws->base.have_sm4_1) { |
| 1125 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1126 |
gp_arg.param = DRM_VMW_PARAM_SM5; |
| 1127 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1128 |
&gp_arg, sizeof(gp_arg)); |
| 1129 |
if (ret == 0 && gp_arg.value != 0) { |
| 1130 |
vws->base.have_sm5 = true; |
| 1131 |
} |
| 1132 |
} |
| 1133 |
|
| 1134 |
if (vws->ioctl.have_drm_2_20 && vws->base.have_sm5) { |
| 1135 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1136 |
gp_arg.param = DRM_VMW_PARAM_GL43; |
| 1137 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1138 |
&gp_arg, sizeof(gp_arg)); |
| 1139 |
if (ret == 0 && gp_arg.value != 0) { |
| 1140 |
vws->base.have_gl43 = true; |
| 1141 |
} |
| 1142 |
} |
| 1143 |
|
| 1144 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1145 |
gp_arg.param = DRM_VMW_PARAM_3D_CAPS_SIZE; |
| 1146 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1147 |
&gp_arg, sizeof(gp_arg)); |
| 1148 |
if (ret) |
| 1149 |
size = SVGA_FIFO_3D_CAPS_SIZE * sizeof(uint32_t); |
| 1150 |
else |
| 1151 |
size = gp_arg.value; |
| 1152 |
|
| 1153 |
if (vws->base.have_gb_objects) |
| 1154 |
vws->ioctl.num_cap_3d = size / sizeof(uint32_t); |
| 1155 |
else |
| 1156 |
vws->ioctl.num_cap_3d = SVGA3D_DEVCAP_MAX; |
| 1157 |
|
| 1158 |
if (vws->ioctl.have_drm_2_16) { |
| 1159 |
vws->base.have_coherent = true; |
| 1160 |
getenv_val = os_get_option("SVGA_FORCE_COHERENT"); |
| 1161 |
if (getenv_val && strcmp(getenv_val, "0") != 0) |
| 1162 |
vws->force_coherent = true; |
| 1163 |
} |
| 1164 |
} else { |
| 1165 |
vws->ioctl.num_cap_3d = SVGA3D_DEVCAP_MAX; |
| 1166 |
|
| 1167 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1168 |
gp_arg.param = DRM_VMW_PARAM_MAX_SURF_MEMORY; |
| 1169 |
if (have_drm_2_5) |
| 1170 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, |
| 1171 |
&gp_arg, sizeof(gp_arg)); |
| 1172 |
if (!have_drm_2_5 || ret) { |
| 1173 |
/* Just guess a large enough value, around 800mb. */ |
| 1174 |
vws->ioctl.max_surface_memory = 0x30000000; |
| 1175 |
} else { |
| 1176 |
vws->ioctl.max_surface_memory = gp_arg.value; |
| 1177 |
} |
| 1178 |
|
| 1179 |
vws->ioctl.max_texture_size = VMW_MAX_DEFAULT_TEXTURE_SIZE; |
| 1180 |
|
| 1181 |
size = SVGA_FIFO_3D_CAPS_SIZE * sizeof(uint32_t); |
| 1182 |
} |
| 1183 |
|
| 1184 |
/* Userspace surfaces are only supported on guest-backed hardware */ |
| 1185 |
vws->userspace_surface = false; |
| 1186 |
getenv_val = os_get_option("VMW_SVGA_USERSPACE_SURFACE"); |
| 1187 |
if (getenv_val && atoi(getenv_val)) { |
| 1188 |
assert(vws->base.have_gb_objects); |
| 1189 |
assert(vws->base.have_vgpu10); |
| 1190 |
memset(&gp_arg, 0, sizeof(gp_arg)); |
| 1191 |
gp_arg.param = DRM_VMW_PARAM_USER_SRF; |
| 1192 |
ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_GET_PARAM, &gp_arg, |
| 1193 |
sizeof(gp_arg)); |
| 1194 |
if (!ret && gp_arg.value == true) { |
| 1195 |
vws->userspace_surface = true; |
| 1196 |
debug_printf("Using userspace managed surfaces\n"); |
| 1197 |
} |
| 1198 |
} |
| 1199 |
|
| 1200 |
debug_printf("VGPU10 interface is %s.\n", |
| 1201 |
vws->base.have_vgpu10 ? "on" : "off"); |
| 1202 |
|
| 1203 |
cap_buffer = calloc(1, size); |
| 1204 |
if (!cap_buffer) { |
| 1205 |
debug_printf("Failed alloc fifo 3D caps buffer.\n"); |
| 1206 |
goto out_no_3d; |
| 1207 |
} |
| 1208 |
|
| 1209 |
vws->ioctl.cap_3d = calloc(vws->ioctl.num_cap_3d, |
| 1210 |
sizeof(*vws->ioctl.cap_3d)); |
| 1211 |
if (!vws->ioctl.cap_3d) { |
| 1212 |
debug_printf("Failed alloc fifo 3D caps buffer.\n"); |
| 1213 |
goto out_no_caparray; |
| 1214 |
} |
| 1215 |
|
| 1216 |
memset(&cap_arg, 0, sizeof(cap_arg)); |
| 1217 |
cap_arg.buffer = (uint64_t) (unsigned long) (cap_buffer); |
| 1218 |
cap_arg.max_size = size; |
| 1219 |
|
| 1220 |
/* |
| 1221 |
* This call must always be after DRM_VMW_PARAM_MAX_MOB_MEMORY and |
| 1222 |
* DRM_VMW_PARAM_SM4_1. This is because, based on these calls, kernel |
| 1223 |
* driver sends the supported cap. |
| 1224 |
*/ |
| 1225 |
ret = drmCommandWrite(vws->ioctl.drm_fd, DRM_VMW_GET_3D_CAP, |
| 1226 |
&cap_arg, sizeof(cap_arg)); |
| 1227 |
|
| 1228 |
if (ret) { |
| 1229 |
debug_printf("Failed to get 3D capabilities" |
| 1230 |
" (%i, %s).\n", ret, strerror(-ret)); |
| 1231 |
goto out_no_caps; |
| 1232 |
} |
| 1233 |
|
| 1234 |
ret = vmw_ioctl_parse_caps(vws, cap_buffer); |
| 1235 |
if (ret) { |
| 1236 |
debug_printf("Failed to parse 3D capabilities" |
| 1237 |
" (%i, %s).\n", ret, strerror(-ret)); |
| 1238 |
goto out_no_caps; |
| 1239 |
} |
| 1240 |
|
| 1241 |
if (((version->version_major == 2 && version->version_minor >= 10) |
| 1242 |
|| version->version_major > 2) && vws->base.have_vgpu10) { |
| 1243 |
|
| 1244 |
/* support for these commands didn't make it into vmwgfx kernel |
| 1245 |
* modules before 2.10. |
| 1246 |
*/ |
| 1247 |
vws->base.have_generate_mipmap_cmd = true; |
| 1248 |
vws->base.have_set_predication_cmd = true; |
| 1249 |
} |
| 1250 |
|
| 1251 |
if (version->version_major == 2 && version->version_minor >= 14) { |
| 1252 |
vws->base.have_fence_fd = true; |
| 1253 |
} |
| 1254 |
|
| 1255 |
free(cap_buffer); |
| 1256 |
drmFreeVersion(version); |
| 1257 |
vmw_printf("%s OK\n", __func__); |
| 1258 |
return true; |
| 1259 |
out_no_caps: |
| 1260 |
free(vws->ioctl.cap_3d); |
| 1261 |
out_no_caparray: |
| 1262 |
free(cap_buffer); |
| 1263 |
out_no_3d: |
| 1264 |
drmFreeVersion(version); |
| 1265 |
out_no_version: |
| 1266 |
vws->ioctl.num_cap_3d = 0; |
| 1267 |
debug_printf("%s Failed\n", __func__); |
| 1268 |
return false; |
| 1269 |
} |
| 1270 |
|
| 1271 |
|
| 1272 |
|
| 1273 |
void |
| 1274 |
vmw_ioctl_cleanup(struct vmw_winsys_screen *vws) |
| 1275 |
{ |
| 1276 |
VMW_FUNC; |
| 1277 |
|
| 1278 |
free(vws->ioctl.cap_3d); |
| 1279 |
} |