i915_cmd_parser.c 31.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
/*
 * Copyright © 2013 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 * Authors:
 *    Brad Volkin <bradley.d.volkin@intel.com>
 *
 */

#include "i915_drv.h"

/**
31
 * DOC: batch buffer command parser
32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75
 *
 * Motivation:
 * Certain OpenGL features (e.g. transform feedback, performance monitoring)
 * require userspace code to submit batches containing commands such as
 * MI_LOAD_REGISTER_IMM to access various registers. Unfortunately, some
 * generations of the hardware will noop these commands in "unsecure" batches
 * (which includes all userspace batches submitted via i915) even though the
 * commands may be safe and represent the intended programming model of the
 * device.
 *
 * The software command parser is similar in operation to the command parsing
 * done in hardware for unsecure batches. However, the software parser allows
 * some operations that would be noop'd by hardware, if the parser determines
 * the operation is safe, and submits the batch as "secure" to prevent hardware
 * parsing.
 *
 * Threats:
 * At a high level, the hardware (and software) checks attempt to prevent
 * granting userspace undue privileges. There are three categories of privilege.
 *
 * First, commands which are explicitly defined as privileged or which should
 * only be used by the kernel driver. The parser generally rejects such
 * commands, though it may allow some from the drm master process.
 *
 * Second, commands which access registers. To support correct/enhanced
 * userspace functionality, particularly certain OpenGL extensions, the parser
 * provides a whitelist of registers which userspace may safely access (for both
 * normal and drm master processes).
 *
 * Third, commands which access privileged memory (i.e. GGTT, HWS page, etc).
 * The parser always rejects such commands.
 *
 * The majority of the problematic commands fall in the MI_* range, with only a
 * few specific commands on each ring (e.g. PIPE_CONTROL and MI_FLUSH_DW).
 *
 * Implementation:
 * Each ring maintains tables of commands and registers which the parser uses in
 * scanning batch buffers submitted to that ring.
 *
 * Since the set of commands that the parser must check for is significantly
 * smaller than the number of commands supported, the parser tables contain only
 * those commands required by the parser. This generally works because command
 * opcode ranges have standard command length encodings. So for commands that
 * the parser does not need to check, it can easily skip them. This is
76
 * implemented via a per-ring length decoding vfunc.
77 78 79 80 81 82 83 84 85 86 87 88
 *
 * Unfortunately, there are a number of commands that do not follow the standard
 * length encoding for their opcode range, primarily amongst the MI_* commands.
 * To handle this, the parser provides a way to define explicit "skip" entries
 * in the per-ring command tables.
 *
 * Other command table entries map fairly directly to high level categories
 * mentioned above: rejected, master-only, register whitelist. The parser
 * implements a number of checks, including the privileged memory checks, via a
 * general bitmasking mechanism.
 */

89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117
#define STD_MI_OPCODE_MASK  0xFF800000
#define STD_3D_OPCODE_MASK  0xFFFF0000
#define STD_2D_OPCODE_MASK  0xFFC00000
#define STD_MFX_OPCODE_MASK 0xFFFF0000

#define CMD(op, opm, f, lm, fl, ...)				\
	{							\
		.flags = (fl) | ((f) ? CMD_DESC_FIXED : 0),	\
		.cmd = { (op), (opm) }, 			\
		.length = { (lm) },				\
		__VA_ARGS__					\
	}

/* Convenience macros to compress the tables */
#define SMI STD_MI_OPCODE_MASK
#define S3D STD_3D_OPCODE_MASK
#define S2D STD_2D_OPCODE_MASK
#define SMFX STD_MFX_OPCODE_MASK
#define F true
#define S CMD_DESC_SKIP
#define R CMD_DESC_REJECT
#define W CMD_DESC_REGISTER
#define B CMD_DESC_BITMASK
#define M CMD_DESC_MASTER

/*            Command                          Mask   Fixed Len   Action
	      ---------------------------------------------------------- */
static const struct drm_i915_cmd_descriptor common_cmds[] = {
	CMD(  MI_NOOP,                          SMI,    F,  1,      S  ),
118
	CMD(  MI_USER_INTERRUPT,                SMI,    F,  1,      R  ),
119
	CMD(  MI_WAIT_FOR_EVENT,                SMI,    F,  1,      M  ),
120 121 122
	CMD(  MI_ARB_CHECK,                     SMI,    F,  1,      S  ),
	CMD(  MI_REPORT_HEAD,                   SMI,    F,  1,      S  ),
	CMD(  MI_SUSPEND_FLUSH,                 SMI,    F,  1,      S  ),
123 124
	CMD(  MI_SEMAPHORE_MBOX,                SMI,   !F,  0xFF,   R  ),
	CMD(  MI_STORE_DWORD_INDEX,             SMI,   !F,  0xFF,   R  ),
125 126
	CMD(  MI_LOAD_REGISTER_IMM(1),          SMI,   !F,  0xFF,   W,
	      .reg = { .offset = 1, .mask = 0x007FFFFC }               ),
127 128 129 130 131 132 133 134 135 136 137 138 139 140
	CMD(  MI_STORE_REGISTER_MEM(1),         SMI,   !F,  0xFF,   W | B,
	      .reg = { .offset = 1, .mask = 0x007FFFFC },
	      .bits = {{
			.offset = 0,
			.mask = MI_GLOBAL_GTT,
			.expected = 0,
	      }},						       ),
	CMD(  MI_LOAD_REGISTER_MEM,             SMI,   !F,  0xFF,   W | B,
	      .reg = { .offset = 1, .mask = 0x007FFFFC },
	      .bits = {{
			.offset = 0,
			.mask = MI_GLOBAL_GTT,
			.expected = 0,
	      }},						       ),
141 142 143 144 145
	/*
	 * MI_BATCH_BUFFER_START requires some special handling. It's not
	 * really a 'skip' action but it doesn't seem like it's worth adding
	 * a new action. See i915_parse_cmds().
	 */
146 147 148 149 150
	CMD(  MI_BATCH_BUFFER_START,            SMI,   !F,  0xFF,   S  ),
};

static const struct drm_i915_cmd_descriptor render_cmds[] = {
	CMD(  MI_FLUSH,                         SMI,    F,  1,      S  ),
151
	CMD(  MI_ARB_ON_OFF,                    SMI,    F,  1,      R  ),
152 153
	CMD(  MI_PREDICATE,                     SMI,    F,  1,      S  ),
	CMD(  MI_TOPOLOGY_FILTER,               SMI,    F,  1,      S  ),
154 155
	CMD(  MI_DISPLAY_FLIP,                  SMI,   !F,  0xFF,   R  ),
	CMD(  MI_SET_CONTEXT,                   SMI,   !F,  0xFF,   R  ),
156
	CMD(  MI_URB_CLEAR,                     SMI,   !F,  0xFF,   S  ),
157 158 159 160 161 162
	CMD(  MI_STORE_DWORD_IMM,               SMI,   !F,  0x3F,   B,
	      .bits = {{
			.offset = 0,
			.mask = MI_GLOBAL_GTT,
			.expected = 0,
	      }},						       ),
163
	CMD(  MI_UPDATE_GTT,                    SMI,   !F,  0xFF,   R  ),
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181
	CMD(  MI_CLFLUSH,                       SMI,   !F,  0x3FF,  B,
	      .bits = {{
			.offset = 0,
			.mask = MI_GLOBAL_GTT,
			.expected = 0,
	      }},						       ),
	CMD(  MI_REPORT_PERF_COUNT,             SMI,   !F,  0x3F,   B,
	      .bits = {{
			.offset = 1,
			.mask = MI_REPORT_PERF_COUNT_GGTT,
			.expected = 0,
	      }},						       ),
	CMD(  MI_CONDITIONAL_BATCH_BUFFER_END,  SMI,   !F,  0xFF,   B,
	      .bits = {{
			.offset = 0,
			.mask = MI_GLOBAL_GTT,
			.expected = 0,
	      }},						       ),
182 183
	CMD(  GFX_OP_3DSTATE_VF_STATISTICS,     S3D,    F,  1,      S  ),
	CMD(  PIPELINE_SELECT,                  S3D,    F,  1,      S  ),
184 185 186 187 188 189
	CMD(  MEDIA_VFE_STATE,			S3D,   !F,  0xFFFF, B,
	      .bits = {{
			.offset = 2,
			.mask = MEDIA_VFE_STATE_MMIO_ACCESS_MASK,
			.expected = 0,
	      }},						       ),
190 191 192
	CMD(  GPGPU_OBJECT,                     S3D,   !F,  0xFF,   S  ),
	CMD(  GPGPU_WALKER,                     S3D,   !F,  0xFF,   S  ),
	CMD(  GFX_OP_3DSTATE_SO_DECL_LIST,      S3D,   !F,  0x1FF,  S  ),
193 194 195
	CMD(  GFX_OP_PIPE_CONTROL(5),           S3D,   !F,  0xFF,   B,
	      .bits = {{
			.offset = 1,
196
			.mask = (PIPE_CONTROL_MMIO_WRITE | PIPE_CONTROL_NOTIFY),
197
			.expected = 0,
198 199 200
	      },
	      {
			.offset = 1,
201 202
		        .mask = (PIPE_CONTROL_GLOBAL_GTT_IVB |
				 PIPE_CONTROL_STORE_DATA_INDEX),
203 204 205
			.expected = 0,
			.condition_offset = 1,
			.condition_mask = PIPE_CONTROL_POST_SYNC_OP_MASK,
206
	      }},						       ),
207 208 209 210 211 212 213
};

static const struct drm_i915_cmd_descriptor hsw_render_cmds[] = {
	CMD(  MI_SET_PREDICATE,                 SMI,    F,  1,      S  ),
	CMD(  MI_RS_CONTROL,                    SMI,    F,  1,      S  ),
	CMD(  MI_URB_ATOMIC_ALLOC,              SMI,    F,  1,      S  ),
	CMD(  MI_RS_CONTEXT,                    SMI,    F,  1,      S  ),
214
	CMD(  MI_LOAD_SCAN_LINES_INCL,          SMI,   !F,  0x3F,   M  ),
215 216
	CMD(  MI_LOAD_SCAN_LINES_EXCL,          SMI,   !F,  0x3F,   R  ),
	CMD(  MI_LOAD_REGISTER_REG,             SMI,   !F,  0xFF,   R  ),
217 218 219 220 221 222 223 224 225 226 227 228 229 230
	CMD(  MI_RS_STORE_DATA_IMM,             SMI,   !F,  0xFF,   S  ),
	CMD(  MI_LOAD_URB_MEM,                  SMI,   !F,  0xFF,   S  ),
	CMD(  MI_STORE_URB_MEM,                 SMI,   !F,  0xFF,   S  ),
	CMD(  GFX_OP_3DSTATE_DX9_CONSTANTF_VS,  S3D,   !F,  0x7FF,  S  ),
	CMD(  GFX_OP_3DSTATE_DX9_CONSTANTF_PS,  S3D,   !F,  0x7FF,  S  ),

	CMD(  GFX_OP_3DSTATE_BINDING_TABLE_EDIT_VS,  S3D,   !F,  0x1FF,  S  ),
	CMD(  GFX_OP_3DSTATE_BINDING_TABLE_EDIT_GS,  S3D,   !F,  0x1FF,  S  ),
	CMD(  GFX_OP_3DSTATE_BINDING_TABLE_EDIT_HS,  S3D,   !F,  0x1FF,  S  ),
	CMD(  GFX_OP_3DSTATE_BINDING_TABLE_EDIT_DS,  S3D,   !F,  0x1FF,  S  ),
	CMD(  GFX_OP_3DSTATE_BINDING_TABLE_EDIT_PS,  S3D,   !F,  0x1FF,  S  ),
};

static const struct drm_i915_cmd_descriptor video_cmds[] = {
231
	CMD(  MI_ARB_ON_OFF,                    SMI,    F,  1,      R  ),
232 233 234 235 236 237
	CMD(  MI_STORE_DWORD_IMM,               SMI,   !F,  0xFF,   B,
	      .bits = {{
			.offset = 0,
			.mask = MI_GLOBAL_GTT,
			.expected = 0,
	      }},						       ),
238
	CMD(  MI_UPDATE_GTT,                    SMI,   !F,  0x3F,   R  ),
239 240 241 242 243
	CMD(  MI_FLUSH_DW,                      SMI,   !F,  0x3F,   B,
	      .bits = {{
			.offset = 0,
			.mask = MI_FLUSH_DW_NOTIFY,
			.expected = 0,
244 245 246 247 248 249 250
	      },
	      {
			.offset = 1,
			.mask = MI_FLUSH_DW_USE_GTT,
			.expected = 0,
			.condition_offset = 0,
			.condition_mask = MI_FLUSH_DW_OP_MASK,
251 252 253 254 255 256 257
	      },
	      {
			.offset = 0,
			.mask = MI_FLUSH_DW_STORE_INDEX,
			.expected = 0,
			.condition_offset = 0,
			.condition_mask = MI_FLUSH_DW_OP_MASK,
258 259 260 261 262 263
	      }},						       ),
	CMD(  MI_CONDITIONAL_BATCH_BUFFER_END,  SMI,   !F,  0xFF,   B,
	      .bits = {{
			.offset = 0,
			.mask = MI_GLOBAL_GTT,
			.expected = 0,
264
	      }},						       ),
265 266 267 268 269 270 271 272 273
	/*
	 * MFX_WAIT doesn't fit the way we handle length for most commands.
	 * It has a length field but it uses a non-standard length bias.
	 * It is always 1 dword though, so just treat it as fixed length.
	 */
	CMD(  MFX_WAIT,                         SMFX,   F,  1,      S  ),
};

static const struct drm_i915_cmd_descriptor vecs_cmds[] = {
274
	CMD(  MI_ARB_ON_OFF,                    SMI,    F,  1,      R  ),
275 276 277 278 279 280
	CMD(  MI_STORE_DWORD_IMM,               SMI,   !F,  0xFF,   B,
	      .bits = {{
			.offset = 0,
			.mask = MI_GLOBAL_GTT,
			.expected = 0,
	      }},						       ),
281
	CMD(  MI_UPDATE_GTT,                    SMI,   !F,  0x3F,   R  ),
282 283 284 285 286
	CMD(  MI_FLUSH_DW,                      SMI,   !F,  0x3F,   B,
	      .bits = {{
			.offset = 0,
			.mask = MI_FLUSH_DW_NOTIFY,
			.expected = 0,
287 288 289 290 291 292 293
	      },
	      {
			.offset = 1,
			.mask = MI_FLUSH_DW_USE_GTT,
			.expected = 0,
			.condition_offset = 0,
			.condition_mask = MI_FLUSH_DW_OP_MASK,
294 295 296 297 298 299 300
	      },
	      {
			.offset = 0,
			.mask = MI_FLUSH_DW_STORE_INDEX,
			.expected = 0,
			.condition_offset = 0,
			.condition_mask = MI_FLUSH_DW_OP_MASK,
301 302 303 304 305 306
	      }},						       ),
	CMD(  MI_CONDITIONAL_BATCH_BUFFER_END,  SMI,   !F,  0xFF,   B,
	      .bits = {{
			.offset = 0,
			.mask = MI_GLOBAL_GTT,
			.expected = 0,
307
	      }},						       ),
308 309 310
};

static const struct drm_i915_cmd_descriptor blt_cmds[] = {
311
	CMD(  MI_DISPLAY_FLIP,                  SMI,   !F,  0xFF,   R  ),
312 313 314 315 316 317
	CMD(  MI_STORE_DWORD_IMM,               SMI,   !F,  0x3FF,  B,
	      .bits = {{
			.offset = 0,
			.mask = MI_GLOBAL_GTT,
			.expected = 0,
	      }},						       ),
318
	CMD(  MI_UPDATE_GTT,                    SMI,   !F,  0x3F,   R  ),
319 320 321 322 323
	CMD(  MI_FLUSH_DW,                      SMI,   !F,  0x3F,   B,
	      .bits = {{
			.offset = 0,
			.mask = MI_FLUSH_DW_NOTIFY,
			.expected = 0,
324 325 326 327 328 329 330
	      },
	      {
			.offset = 1,
			.mask = MI_FLUSH_DW_USE_GTT,
			.expected = 0,
			.condition_offset = 0,
			.condition_mask = MI_FLUSH_DW_OP_MASK,
331 332 333 334 335 336 337
	      },
	      {
			.offset = 0,
			.mask = MI_FLUSH_DW_STORE_INDEX,
			.expected = 0,
			.condition_offset = 0,
			.condition_mask = MI_FLUSH_DW_OP_MASK,
338
	      }},						       ),
339 340 341 342
	CMD(  COLOR_BLT,                        S2D,   !F,  0x3F,   S  ),
	CMD(  SRC_COPY_BLT,                     S2D,   !F,  0x3F,   S  ),
};

343
static const struct drm_i915_cmd_descriptor hsw_blt_cmds[] = {
344
	CMD(  MI_LOAD_SCAN_LINES_INCL,          SMI,   !F,  0x3F,   M  ),
345 346 347
	CMD(  MI_LOAD_SCAN_LINES_EXCL,          SMI,   !F,  0x3F,   R  ),
};

348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385
#undef CMD
#undef SMI
#undef S3D
#undef S2D
#undef SMFX
#undef F
#undef S
#undef R
#undef W
#undef B
#undef M

static const struct drm_i915_cmd_table gen7_render_cmds[] = {
	{ common_cmds, ARRAY_SIZE(common_cmds) },
	{ render_cmds, ARRAY_SIZE(render_cmds) },
};

static const struct drm_i915_cmd_table hsw_render_ring_cmds[] = {
	{ common_cmds, ARRAY_SIZE(common_cmds) },
	{ render_cmds, ARRAY_SIZE(render_cmds) },
	{ hsw_render_cmds, ARRAY_SIZE(hsw_render_cmds) },
};

static const struct drm_i915_cmd_table gen7_video_cmds[] = {
	{ common_cmds, ARRAY_SIZE(common_cmds) },
	{ video_cmds, ARRAY_SIZE(video_cmds) },
};

static const struct drm_i915_cmd_table hsw_vebox_cmds[] = {
	{ common_cmds, ARRAY_SIZE(common_cmds) },
	{ vecs_cmds, ARRAY_SIZE(vecs_cmds) },
};

static const struct drm_i915_cmd_table gen7_blt_cmds[] = {
	{ common_cmds, ARRAY_SIZE(common_cmds) },
	{ blt_cmds, ARRAY_SIZE(blt_cmds) },
};

386 387 388 389 390 391
static const struct drm_i915_cmd_table hsw_blt_ring_cmds[] = {
	{ common_cmds, ARRAY_SIZE(common_cmds) },
	{ blt_cmds, ARRAY_SIZE(blt_cmds) },
	{ hsw_blt_cmds, ARRAY_SIZE(hsw_blt_cmds) },
};

392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414
/*
 * Register whitelists, sorted by increasing register offset.
 *
 * Some registers that userspace accesses are 64 bits. The register
 * access commands only allow 32-bit accesses. Hence, we have to include
 * entries for both halves of the 64-bit registers.
 */

/* Convenience macro for adding 64-bit registers */
#define REG64(addr) (addr), (addr + sizeof(u32))

static const u32 gen7_render_regs[] = {
	REG64(HS_INVOCATION_COUNT),
	REG64(DS_INVOCATION_COUNT),
	REG64(IA_VERTICES_COUNT),
	REG64(IA_PRIMITIVES_COUNT),
	REG64(VS_INVOCATION_COUNT),
	REG64(GS_INVOCATION_COUNT),
	REG64(GS_PRIMITIVES_COUNT),
	REG64(CL_INVOCATION_COUNT),
	REG64(CL_PRIMITIVES_COUNT),
	REG64(PS_INVOCATION_COUNT),
	REG64(PS_DEPTH_COUNT),
415
	OACONTROL, /* Only allowed for LRI and SRM. See below. */
416 417
	REG64(MI_PREDICATE_SRC0),
	REG64(MI_PREDICATE_SRC1),
418 419 420 421 422 423
	GEN7_3DPRIM_END_OFFSET,
	GEN7_3DPRIM_START_VERTEX,
	GEN7_3DPRIM_VERTEX_COUNT,
	GEN7_3DPRIM_INSTANCE_COUNT,
	GEN7_3DPRIM_START_INSTANCE,
	GEN7_3DPRIM_BASE_VERTEX,
424 425 426 427
	REG64(GEN7_SO_NUM_PRIMS_WRITTEN(0)),
	REG64(GEN7_SO_NUM_PRIMS_WRITTEN(1)),
	REG64(GEN7_SO_NUM_PRIMS_WRITTEN(2)),
	REG64(GEN7_SO_NUM_PRIMS_WRITTEN(3)),
428 429 430 431
	REG64(GEN7_SO_PRIM_STORAGE_NEEDED(0)),
	REG64(GEN7_SO_PRIM_STORAGE_NEEDED(1)),
	REG64(GEN7_SO_PRIM_STORAGE_NEEDED(2)),
	REG64(GEN7_SO_PRIM_STORAGE_NEEDED(3)),
432 433 434 435
	GEN7_SO_WRITE_OFFSET(0),
	GEN7_SO_WRITE_OFFSET(1),
	GEN7_SO_WRITE_OFFSET(2),
	GEN7_SO_WRITE_OFFSET(3),
436 437 438
	GEN7_L3SQCREG1,
	GEN7_L3CNTLREG2,
	GEN7_L3CNTLREG3,
439 440 441 442 443 444
};

static const u32 gen7_blt_regs[] = {
	BCS_SWCTRL,
};

445 446 447 448 449 450 451 452 453 454 455 456 457
static const u32 ivb_master_regs[] = {
	FORCEWAKE_MT,
	DERRMR,
	GEN7_PIPE_DE_LOAD_SL(PIPE_A),
	GEN7_PIPE_DE_LOAD_SL(PIPE_B),
	GEN7_PIPE_DE_LOAD_SL(PIPE_C),
};

static const u32 hsw_master_regs[] = {
	FORCEWAKE_MT,
	DERRMR,
};

458 459
#undef REG64

460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510
static u32 gen7_render_get_cmd_length_mask(u32 cmd_header)
{
	u32 client = (cmd_header & INSTR_CLIENT_MASK) >> INSTR_CLIENT_SHIFT;
	u32 subclient =
		(cmd_header & INSTR_SUBCLIENT_MASK) >> INSTR_SUBCLIENT_SHIFT;

	if (client == INSTR_MI_CLIENT)
		return 0x3F;
	else if (client == INSTR_RC_CLIENT) {
		if (subclient == INSTR_MEDIA_SUBCLIENT)
			return 0xFFFF;
		else
			return 0xFF;
	}

	DRM_DEBUG_DRIVER("CMD: Abnormal rcs cmd length! 0x%08X\n", cmd_header);
	return 0;
}

static u32 gen7_bsd_get_cmd_length_mask(u32 cmd_header)
{
	u32 client = (cmd_header & INSTR_CLIENT_MASK) >> INSTR_CLIENT_SHIFT;
	u32 subclient =
		(cmd_header & INSTR_SUBCLIENT_MASK) >> INSTR_SUBCLIENT_SHIFT;

	if (client == INSTR_MI_CLIENT)
		return 0x3F;
	else if (client == INSTR_RC_CLIENT) {
		if (subclient == INSTR_MEDIA_SUBCLIENT)
			return 0xFFF;
		else
			return 0xFF;
	}

	DRM_DEBUG_DRIVER("CMD: Abnormal bsd cmd length! 0x%08X\n", cmd_header);
	return 0;
}

static u32 gen7_blt_get_cmd_length_mask(u32 cmd_header)
{
	u32 client = (cmd_header & INSTR_CLIENT_MASK) >> INSTR_CLIENT_SHIFT;

	if (client == INSTR_MI_CLIENT)
		return 0x3F;
	else if (client == INSTR_BC_CLIENT)
		return 0xFF;

	DRM_DEBUG_DRIVER("CMD: Abnormal blt cmd length! 0x%08X\n", cmd_header);
	return 0;
}

511
static bool validate_cmds_sorted(struct intel_engine_cs *ring,
512 513
				 const struct drm_i915_cmd_table *cmd_tables,
				 int cmd_table_count)
514 515
{
	int i;
516
	bool ret = true;
517

518
	if (!cmd_tables || cmd_table_count == 0)
519
		return true;
520

521 522
	for (i = 0; i < cmd_table_count; i++) {
		const struct drm_i915_cmd_table *table = &cmd_tables[i];
523 524 525 526 527 528 529 530
		u32 previous = 0;
		int j;

		for (j = 0; j < table->count; j++) {
			const struct drm_i915_cmd_descriptor *desc =
				&table->table[i];
			u32 curr = desc->cmd.value & desc->cmd.mask;

531
			if (curr < previous) {
532 533
				DRM_ERROR("CMD: table not sorted ring=%d table=%d entry=%d cmd=0x%08X prev=0x%08X\n",
					  ring->id, i, j, curr, previous);
534 535
				ret = false;
			}
536 537 538 539

			previous = curr;
		}
	}
540 541

	return ret;
542 543
}

544
static bool check_sorted(int ring_id, const u32 *reg_table, int reg_count)
545 546 547
{
	int i;
	u32 previous = 0;
548
	bool ret = true;
549 550 551 552

	for (i = 0; i < reg_count; i++) {
		u32 curr = reg_table[i];

553
		if (curr < previous) {
554 555
			DRM_ERROR("CMD: table not sorted ring=%d entry=%d reg=0x%08X prev=0x%08X\n",
				  ring_id, i, curr, previous);
556 557
			ret = false;
		}
558 559 560

		previous = curr;
	}
561 562

	return ret;
563 564
}

565
static bool validate_regs_sorted(struct intel_engine_cs *ring)
566
{
567 568 569
	return check_sorted(ring->id, ring->reg_table, ring->reg_count) &&
		check_sorted(ring->id, ring->master_reg_table,
			     ring->master_reg_count);
570 571
}

572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592
struct cmd_node {
	const struct drm_i915_cmd_descriptor *desc;
	struct hlist_node node;
};

/*
 * Different command ranges have different numbers of bits for the opcode. For
 * example, MI commands use bits 31:23 while 3D commands use bits 31:16. The
 * problem is that, for example, MI commands use bits 22:16 for other fields
 * such as GGTT vs PPGTT bits. If we include those bits in the mask then when
 * we mask a command from a batch it could hash to the wrong bucket due to
 * non-opcode bits being set. But if we don't include those bits, some 3D
 * commands may hash to the same bucket due to not including opcode bits that
 * make the command unique. For now, we will risk hashing to the same bucket.
 *
 * If we attempt to generate a perfect hash, we should be able to look at bits
 * 31:29 of a command from a batch buffer and use the full mask for that
 * client. The existing INSTR_CLIENT_MASK/SHIFT defines can be used for this.
 */
#define CMD_HASH_MASK STD_MI_OPCODE_MASK

593
static int init_hash_table(struct intel_engine_cs *ring,
594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621
			   const struct drm_i915_cmd_table *cmd_tables,
			   int cmd_table_count)
{
	int i, j;

	hash_init(ring->cmd_hash);

	for (i = 0; i < cmd_table_count; i++) {
		const struct drm_i915_cmd_table *table = &cmd_tables[i];

		for (j = 0; j < table->count; j++) {
			const struct drm_i915_cmd_descriptor *desc =
				&table->table[j];
			struct cmd_node *desc_node =
				kmalloc(sizeof(*desc_node), GFP_KERNEL);

			if (!desc_node)
				return -ENOMEM;

			desc_node->desc = desc;
			hash_add(ring->cmd_hash, &desc_node->node,
				 desc->cmd.value & CMD_HASH_MASK);
		}
	}

	return 0;
}

622
static void fini_hash_table(struct intel_engine_cs *ring)
623 624 625 626 627 628 629 630 631 632 633
{
	struct hlist_node *tmp;
	struct cmd_node *desc_node;
	int i;

	hash_for_each_safe(ring->cmd_hash, i, tmp, desc_node, node) {
		hash_del(&desc_node->node);
		kfree(desc_node);
	}
}

634 635 636 637 638
/**
 * i915_cmd_parser_init_ring() - set cmd parser related fields for a ringbuffer
 * @ring: the ringbuffer to initialize
 *
 * Optionally initializes fields related to batch buffer command parsing in the
639
 * struct intel_engine_cs based on whether the platform requires software
640
 * command parsing.
641 642
 *
 * Return: non-zero if initialization fails
643
 */
644
int i915_cmd_parser_init_ring(struct intel_engine_cs *ring)
645
{
646 647 648 649
	const struct drm_i915_cmd_table *cmd_tables;
	int cmd_table_count;
	int ret;

650
	if (!IS_GEN7(ring->dev))
651
		return 0;
652 653 654

	switch (ring->id) {
	case RCS:
655
		if (IS_HASWELL(ring->dev)) {
656 657
			cmd_tables = hsw_render_ring_cmds;
			cmd_table_count =
658 659
				ARRAY_SIZE(hsw_render_ring_cmds);
		} else {
660 661
			cmd_tables = gen7_render_cmds;
			cmd_table_count = ARRAY_SIZE(gen7_render_cmds);
662 663
		}

664 665 666
		ring->reg_table = gen7_render_regs;
		ring->reg_count = ARRAY_SIZE(gen7_render_regs);

667 668 669 670 671 672 673 674
		if (IS_HASWELL(ring->dev)) {
			ring->master_reg_table = hsw_master_regs;
			ring->master_reg_count = ARRAY_SIZE(hsw_master_regs);
		} else {
			ring->master_reg_table = ivb_master_regs;
			ring->master_reg_count = ARRAY_SIZE(ivb_master_regs);
		}

675 676 677
		ring->get_cmd_length_mask = gen7_render_get_cmd_length_mask;
		break;
	case VCS:
678 679
		cmd_tables = gen7_video_cmds;
		cmd_table_count = ARRAY_SIZE(gen7_video_cmds);
680 681 682
		ring->get_cmd_length_mask = gen7_bsd_get_cmd_length_mask;
		break;
	case BCS:
683
		if (IS_HASWELL(ring->dev)) {
684 685
			cmd_tables = hsw_blt_ring_cmds;
			cmd_table_count = ARRAY_SIZE(hsw_blt_ring_cmds);
686
		} else {
687 688
			cmd_tables = gen7_blt_cmds;
			cmd_table_count = ARRAY_SIZE(gen7_blt_cmds);
689 690
		}

691 692 693
		ring->reg_table = gen7_blt_regs;
		ring->reg_count = ARRAY_SIZE(gen7_blt_regs);

694 695 696 697 698 699 700 701
		if (IS_HASWELL(ring->dev)) {
			ring->master_reg_table = hsw_master_regs;
			ring->master_reg_count = ARRAY_SIZE(hsw_master_regs);
		} else {
			ring->master_reg_table = ivb_master_regs;
			ring->master_reg_count = ARRAY_SIZE(ivb_master_regs);
		}

702 703 704
		ring->get_cmd_length_mask = gen7_blt_get_cmd_length_mask;
		break;
	case VECS:
705 706
		cmd_tables = hsw_vebox_cmds;
		cmd_table_count = ARRAY_SIZE(hsw_vebox_cmds);
707 708 709 710 711 712 713 714 715
		/* VECS can use the same length_mask function as VCS */
		ring->get_cmd_length_mask = gen7_bsd_get_cmd_length_mask;
		break;
	default:
		DRM_ERROR("CMD: cmd_parser_init with unknown ring: %d\n",
			  ring->id);
		BUG();
	}

716
	BUG_ON(!validate_cmds_sorted(ring, cmd_tables, cmd_table_count));
717
	BUG_ON(!validate_regs_sorted(ring));
718

719 720 721 722 723 724 725
	WARN_ON(!hash_empty(ring->cmd_hash));

	ret = init_hash_table(ring, cmd_tables, cmd_table_count);
	if (ret) {
		DRM_ERROR("CMD: cmd_parser_init failed!\n");
		fini_hash_table(ring);
		return ret;
726 727 728 729 730 731 732 733 734 735 736 737 738 739
	}

	ring->needs_cmd_parser = true;

	return 0;
}

/**
 * i915_cmd_parser_fini_ring() - clean up cmd parser related fields
 * @ring: the ringbuffer to clean up
 *
 * Releases any resources related to command parsing that may have been
 * initialized for the specified ring.
 */
740
void i915_cmd_parser_fini_ring(struct intel_engine_cs *ring)
741 742 743 744 745
{
	if (!ring->needs_cmd_parser)
		return;

	fini_hash_table(ring);
746 747 748
}

static const struct drm_i915_cmd_descriptor*
749
find_cmd_in_table(struct intel_engine_cs *ring,
750 751
		  u32 cmd_header)
{
752
	struct cmd_node *desc_node;
753

754 755 756
	hash_for_each_possible(ring->cmd_hash, desc_node, node,
			       cmd_header & CMD_HASH_MASK) {
		const struct drm_i915_cmd_descriptor *desc = desc_node->desc;
757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775
		u32 masked_cmd = desc->cmd.mask & cmd_header;
		u32 masked_value = desc->cmd.value & desc->cmd.mask;

		if (masked_cmd == masked_value)
			return desc;
	}

	return NULL;
}

/*
 * Returns a pointer to a descriptor for the command specified by cmd_header.
 *
 * The caller must supply space for a default descriptor via the default_desc
 * parameter. If no descriptor for the specified command exists in the ring's
 * command parser tables, this function fills in default_desc based on the
 * ring's default length encoding and returns default_desc.
 */
static const struct drm_i915_cmd_descriptor*
776
find_cmd(struct intel_engine_cs *ring,
777 778 779
	 u32 cmd_header,
	 struct drm_i915_cmd_descriptor *default_desc)
{
780
	const struct drm_i915_cmd_descriptor *desc;
781 782
	u32 mask;

783 784 785
	desc = find_cmd_in_table(ring, cmd_header);
	if (desc)
		return desc;
786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847

	mask = ring->get_cmd_length_mask(cmd_header);
	if (!mask)
		return NULL;

	BUG_ON(!default_desc);
	default_desc->flags = CMD_DESC_SKIP;
	default_desc->length.mask = mask;

	return default_desc;
}

static bool valid_reg(const u32 *table, int count, u32 addr)
{
	if (table && count != 0) {
		int i;

		for (i = 0; i < count; i++) {
			if (table[i] == addr)
				return true;
		}
	}

	return false;
}

static u32 *vmap_batch(struct drm_i915_gem_object *obj)
{
	int i;
	void *addr = NULL;
	struct sg_page_iter sg_iter;
	struct page **pages;

	pages = drm_malloc_ab(obj->base.size >> PAGE_SHIFT, sizeof(*pages));
	if (pages == NULL) {
		DRM_DEBUG_DRIVER("Failed to get space for pages\n");
		goto finish;
	}

	i = 0;
	for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, 0) {
		pages[i] = sg_page_iter_page(&sg_iter);
		i++;
	}

	addr = vmap(pages, i, 0, PAGE_KERNEL);
	if (addr == NULL) {
		DRM_DEBUG_DRIVER("Failed to vmap pages\n");
		goto finish;
	}

finish:
	if (pages)
		drm_free_large(pages);
	return (u32*)addr;
}

/**
 * i915_needs_cmd_parser() - should a given ring use software command parsing?
 * @ring: the ring in question
 *
 * Only certain platforms require software batch buffer command parsing, and
848
 * only when enabled via module parameter.
849 850 851
 *
 * Return: true if the ring requires software command parsing
 */
852
bool i915_needs_cmd_parser(struct intel_engine_cs *ring)
853
{
854
	if (!ring->needs_cmd_parser)
855 856
		return false;

857
	if (!USES_PPGTT(ring->dev))
858 859
		return false;

860 861 862
	return (i915.enable_cmd_parser == 1);
}

863
static bool check_cmd(const struct intel_engine_cs *ring,
864 865
		      const struct drm_i915_cmd_descriptor *desc,
		      const u32 *cmd,
866 867
		      const bool is_master,
		      bool *oacontrol_set)
868 869 870 871 872 873 874 875 876 877 878 879 880 881 882
{
	if (desc->flags & CMD_DESC_REJECT) {
		DRM_DEBUG_DRIVER("CMD: Rejected command: 0x%08X\n", *cmd);
		return false;
	}

	if ((desc->flags & CMD_DESC_MASTER) && !is_master) {
		DRM_DEBUG_DRIVER("CMD: Rejected master-only command: 0x%08X\n",
				 *cmd);
		return false;
	}

	if (desc->flags & CMD_DESC_REGISTER) {
		u32 reg_addr = cmd[desc->reg.offset] & desc->reg.mask;

883 884 885 886 887 888 889 890 891 892
		/*
		 * OACONTROL requires some special handling for writes. We
		 * want to make sure that any batch which enables OA also
		 * disables it before the end of the batch. The goal is to
		 * prevent one process from snooping on the perf data from
		 * another process. To do that, we need to check the value
		 * that will be written to the register. Hence, limit
		 * OACONTROL writes to only MI_LOAD_REGISTER_IMM commands.
		 */
		if (reg_addr == OACONTROL) {
893 894
			if (desc->cmd.value == MI_LOAD_REGISTER_MEM) {
				DRM_DEBUG_DRIVER("CMD: Rejected LRM to OACONTROL\n");
895
				return false;
896
			}
897 898 899 900 901

			if (desc->cmd.value == MI_LOAD_REGISTER_IMM(1))
				*oacontrol_set = (cmd[2] != 0);
		}

902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952
		if (!valid_reg(ring->reg_table,
			       ring->reg_count, reg_addr)) {
			if (!is_master ||
			    !valid_reg(ring->master_reg_table,
				       ring->master_reg_count,
				       reg_addr)) {
				DRM_DEBUG_DRIVER("CMD: Rejected register 0x%08X in command: 0x%08X (ring=%d)\n",
						 reg_addr,
						 *cmd,
						 ring->id);
				return false;
			}
		}
	}

	if (desc->flags & CMD_DESC_BITMASK) {
		int i;

		for (i = 0; i < MAX_CMD_DESC_BITMASKS; i++) {
			u32 dword;

			if (desc->bits[i].mask == 0)
				break;

			if (desc->bits[i].condition_mask != 0) {
				u32 offset =
					desc->bits[i].condition_offset;
				u32 condition = cmd[offset] &
					desc->bits[i].condition_mask;

				if (condition == 0)
					continue;
			}

			dword = cmd[desc->bits[i].offset] &
				desc->bits[i].mask;

			if (dword != desc->bits[i].expected) {
				DRM_DEBUG_DRIVER("CMD: Rejected command 0x%08X for bitmask 0x%08X (exp=0x%08X act=0x%08X) (ring=%d)\n",
						 *cmd,
						 desc->bits[i].mask,
						 desc->bits[i].expected,
						 dword, ring->id);
				return false;
			}
		}
	}

	return true;
}

953 954 955 956 957 958 959 960 961 962 963 964
#define LENGTH_BIAS 2

/**
 * i915_parse_cmds() - parse a submitted batch buffer for privilege violations
 * @ring: the ring on which the batch is to execute
 * @batch_obj: the batch buffer in question
 * @batch_start_offset: byte offset in the batch at which execution starts
 * @is_master: is the submitting process the drm master?
 *
 * Parses the specified batch buffer looking for privilege violations as
 * described in the overview.
 *
965 966
 * Return: non-zero if the parser finds violations or otherwise fails; -EACCES
 * if the batch appears legal but should use hardware parsing
967
 */
968
int i915_parse_cmds(struct intel_engine_cs *ring,
969 970 971 972 973 974 975 976
		    struct drm_i915_gem_object *batch_obj,
		    u32 batch_start_offset,
		    bool is_master)
{
	int ret = 0;
	u32 *cmd, *batch_base, *batch_end;
	struct drm_i915_cmd_descriptor default_desc = { 0 };
	int needs_clflush = 0;
977
	bool oacontrol_set = false; /* OACONTROL tracking. See check_cmd() */
978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012

	ret = i915_gem_obj_prepare_shmem_read(batch_obj, &needs_clflush);
	if (ret) {
		DRM_DEBUG_DRIVER("CMD: failed to prep read\n");
		return ret;
	}

	batch_base = vmap_batch(batch_obj);
	if (!batch_base) {
		DRM_DEBUG_DRIVER("CMD: Failed to vmap batch\n");
		i915_gem_object_unpin_pages(batch_obj);
		return -ENOMEM;
	}

	if (needs_clflush)
		drm_clflush_virt_range((char *)batch_base, batch_obj->base.size);

	cmd = batch_base + (batch_start_offset / sizeof(*cmd));
	batch_end = cmd + (batch_obj->base.size / sizeof(*batch_end));

	while (cmd < batch_end) {
		const struct drm_i915_cmd_descriptor *desc;
		u32 length;

		if (*cmd == MI_BATCH_BUFFER_END)
			break;

		desc = find_cmd(ring, *cmd, &default_desc);
		if (!desc) {
			DRM_DEBUG_DRIVER("CMD: Unrecognized command: 0x%08X\n",
					 *cmd);
			ret = -EINVAL;
			break;
		}

1013 1014 1015 1016 1017 1018 1019 1020 1021 1022
		/*
		 * If the batch buffer contains a chained batch, return an
		 * error that tells the caller to abort and dispatch the
		 * workload as a non-secure batch.
		 */
		if (desc->cmd.value == MI_BATCH_BUFFER_START) {
			ret = -EACCES;
			break;
		}

1023 1024 1025 1026 1027 1028
		if (desc->flags & CMD_DESC_FIXED)
			length = desc->length.fixed;
		else
			length = ((*cmd & desc->length.mask) + LENGTH_BIAS);

		if ((batch_end - cmd) < length) {
1029
			DRM_DEBUG_DRIVER("CMD: Command length exceeds batch length: 0x%08X length=%u batchlen=%td\n",
1030 1031
					 *cmd,
					 length,
1032
					 batch_end - cmd);
1033 1034 1035 1036
			ret = -EINVAL;
			break;
		}

1037
		if (!check_cmd(ring, desc, cmd, is_master, &oacontrol_set)) {
1038 1039 1040 1041 1042 1043 1044
			ret = -EINVAL;
			break;
		}

		cmd += length;
	}

1045 1046 1047 1048 1049
	if (oacontrol_set) {
		DRM_DEBUG_DRIVER("CMD: batch set OACONTROL but did not clear it\n");
		ret = -EINVAL;
	}

1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060
	if (cmd >= batch_end) {
		DRM_DEBUG_DRIVER("CMD: Got to the end of the buffer w/o a BBE cmd!\n");
		ret = -EINVAL;
	}

	vunmap(batch_base);

	i915_gem_object_unpin_pages(batch_obj);

	return ret;
}
1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076

/**
 * i915_cmd_parser_get_version() - get the cmd parser version number
 *
 * The cmd parser maintains a simple increasing integer version number suitable
 * for passing to userspace clients to determine what operations are permitted.
 *
 * Return: the current version number of the cmd parser
 */
int i915_cmd_parser_get_version(void)
{
	/*
	 * Command parser version history
	 *
	 * 1. Initial version. Checks batches and reports violations, but leaves
	 *    hardware parsing enabled (so does not allow new use cases).
1077 1078
	 * 2. Allow access to the MI_PREDICATE_SRC0 and
	 *    MI_PREDICATE_SRC1 registers.
1079
	 */
1080
	return 2;
1081
}