1 /*
2  * Copyright © 2014-2015 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 #ifndef _UAPI_VC4_DRM_H_
25 #define _UAPI_VC4_DRM_H_
26 
27 #include "drm.h"
28 
29 #if defined(__cplusplus)
30 extern "C" {
31 #endif
32 
33 #define DRM_VC4_SUBMIT_CL                         0x00
34 #define DRM_VC4_WAIT_SEQNO                        0x01
35 #define DRM_VC4_WAIT_BO                           0x02
36 #define DRM_VC4_CREATE_BO                         0x03
37 #define DRM_VC4_MMAP_BO                           0x04
38 #define DRM_VC4_CREATE_SHADER_BO                  0x05
39 #define DRM_VC4_GET_HANG_STATE                    0x06
40 #define DRM_VC4_GET_PARAM                         0x07
41 #define DRM_VC4_SET_TILING                        0x08
42 #define DRM_VC4_GET_TILING                        0x09
43 #define DRM_VC4_LABEL_BO                          0x0a
44 #define DRM_VC4_GEM_MADVISE                       0x0b
45 #define DRM_VC4_PERFMON_CREATE                    0x0c
46 #define DRM_VC4_PERFMON_DESTROY                   0x0d
47 #define DRM_VC4_PERFMON_GET_VALUES                0x0e
48 
49 #define DRM_IOCTL_VC4_SUBMIT_CL           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SUBMIT_CL, struct drm_vc4_submit_cl)
50 #define DRM_IOCTL_VC4_WAIT_SEQNO          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_SEQNO, struct drm_vc4_wait_seqno)
51 #define DRM_IOCTL_VC4_WAIT_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_BO, struct drm_vc4_wait_bo)
52 #define DRM_IOCTL_VC4_CREATE_BO           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_BO, struct drm_vc4_create_bo)
53 #define DRM_IOCTL_VC4_MMAP_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_MMAP_BO, struct drm_vc4_mmap_bo)
54 #define DRM_IOCTL_VC4_CREATE_SHADER_BO    DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_SHADER_BO, struct drm_vc4_create_shader_bo)
55 #define DRM_IOCTL_VC4_GET_HANG_STATE      DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_HANG_STATE, struct drm_vc4_get_hang_state)
56 #define DRM_IOCTL_VC4_GET_PARAM           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_PARAM, struct drm_vc4_get_param)
57 #define DRM_IOCTL_VC4_SET_TILING          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SET_TILING, struct drm_vc4_set_tiling)
58 #define DRM_IOCTL_VC4_GET_TILING          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_TILING, struct drm_vc4_get_tiling)
59 #define DRM_IOCTL_VC4_LABEL_BO            DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_LABEL_BO, struct drm_vc4_label_bo)
60 #define DRM_IOCTL_VC4_GEM_MADVISE         DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GEM_MADVISE, struct drm_vc4_gem_madvise)
61 #define DRM_IOCTL_VC4_PERFMON_CREATE      DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_PERFMON_CREATE, struct drm_vc4_perfmon_create)
62 #define DRM_IOCTL_VC4_PERFMON_DESTROY     DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_PERFMON_DESTROY, struct drm_vc4_perfmon_destroy)
63 #define DRM_IOCTL_VC4_PERFMON_GET_VALUES  DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_PERFMON_GET_VALUES, struct drm_vc4_perfmon_get_values)
64 
65 struct drm_vc4_submit_rcl_surface {
66 	__u32 hindex; /* Handle index, or ~0 if not present. */
67 	__u32 offset; /* Offset to start of buffer. */
68 	/*
69 	 * Bits for either render config (color_write) or load/store packet.
70 	 * Bits should all be 0 for MSAA load/stores.
71 	 */
72 	__u16 bits;
73 
74 #define VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES		(1 << 0)
75 	__u16 flags;
76 };
77 
78 /**
79  * struct drm_vc4_submit_cl - ioctl argument for submitting commands to the 3D
80  * engine.
81  *
82  * Drivers typically use GPU BOs to store batchbuffers / command lists and
83  * their associated state.  However, because the VC4 lacks an MMU, we have to
84  * do validation of memory accesses by the GPU commands.  If we were to store
85  * our commands in BOs, we'd need to do uncached readback from them to do the
86  * validation process, which is too expensive.  Instead, userspace accumulates
87  * commands and associated state in plain memory, then the kernel copies the
88  * data to its own address space, and then validates and stores it in a GPU
89  * BO.
90  */
91 struct drm_vc4_submit_cl {
92 	/* Pointer to the binner command list.
93 	 *
94 	 * This is the first set of commands executed, which runs the
95 	 * coordinate shader to determine where primitives land on the screen,
96 	 * then writes out the state updates and draw calls necessary per tile
97 	 * to the tile allocation BO.
98 	 */
99 	__u64 bin_cl;
100 
101 	/* Pointer to the shader records.
102 	 *
103 	 * Shader records are the structures read by the hardware that contain
104 	 * pointers to uniforms, shaders, and vertex attributes.  The
105 	 * reference to the shader record has enough information to determine
106 	 * how many pointers are necessary (fixed number for shaders/uniforms,
107 	 * and an attribute count), so those BO indices into bo_handles are
108 	 * just stored as __u32s before each shader record passed in.
109 	 */
110 	__u64 shader_rec;
111 
112 	/* Pointer to uniform data and texture handles for the textures
113 	 * referenced by the shader.
114 	 *
115 	 * For each shader state record, there is a set of uniform data in the
116 	 * order referenced by the record (FS, VS, then CS).  Each set of
117 	 * uniform data has a __u32 index into bo_handles per texture
118 	 * sample operation, in the order the QPU_W_TMUn_S writes appear in
119 	 * the program.  Following the texture BO handle indices is the actual
120 	 * uniform data.
121 	 *
122 	 * The individual uniform state blocks don't have sizes passed in,
123 	 * because the kernel has to determine the sizes anyway during shader
124 	 * code validation.
125 	 */
126 	__u64 uniforms;
127 	__u64 bo_handles;
128 
129 	/* Size in bytes of the binner command list. */
130 	__u32 bin_cl_size;
131 	/* Size in bytes of the set of shader records. */
132 	__u32 shader_rec_size;
133 	/* Number of shader records.
134 	 *
135 	 * This could just be computed from the contents of shader_records and
136 	 * the address bits of references to them from the bin CL, but it
137 	 * keeps the kernel from having to resize some allocations it makes.
138 	 */
139 	__u32 shader_rec_count;
140 	/* Size in bytes of the uniform state. */
141 	__u32 uniforms_size;
142 
143 	/* Number of BO handles passed in (size is that times 4). */
144 	__u32 bo_handle_count;
145 
146 	/* RCL setup: */
147 	__u16 width;
148 	__u16 height;
149 	__u8 min_x_tile;
150 	__u8 min_y_tile;
151 	__u8 max_x_tile;
152 	__u8 max_y_tile;
153 	struct drm_vc4_submit_rcl_surface color_read;
154 	struct drm_vc4_submit_rcl_surface color_write;
155 	struct drm_vc4_submit_rcl_surface zs_read;
156 	struct drm_vc4_submit_rcl_surface zs_write;
157 	struct drm_vc4_submit_rcl_surface msaa_color_write;
158 	struct drm_vc4_submit_rcl_surface msaa_zs_write;
159 	__u32 clear_color[2];
160 	__u32 clear_z;
161 	__u8 clear_s;
162 
163 	__u32 pad:24;
164 
165 #define VC4_SUBMIT_CL_USE_CLEAR_COLOR			(1 << 0)
166 /* By default, the kernel gets to choose the order that the tiles are
167  * rendered in.  If this is set, then the tiles will be rendered in a
168  * raster order, with the right-to-left vs left-to-right and
169  * top-to-bottom vs bottom-to-top dictated by
170  * VC4_SUBMIT_CL_RCL_ORDER_INCREASING_*.  This allows overlapping
171  * blits to be implemented using the 3D engine.
172  */
173 #define VC4_SUBMIT_CL_FIXED_RCL_ORDER			(1 << 1)
174 #define VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X		(1 << 2)
175 #define VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y		(1 << 3)
176 	__u32 flags;
177 
178 	/* Returned value of the seqno of this render job (for the
179 	 * wait ioctl).
180 	 */
181 	__u64 seqno;
182 
183 	/* ID of the perfmon to attach to this job. 0 means no perfmon. */
184 	__u32 perfmonid;
185 
186 	/* Syncobj handle to wait on. If set, processing of this render job
187 	 * will not start until the syncobj is signaled. 0 means ignore.
188 	 */
189 	__u32 in_sync;
190 
191 	/* Syncobj handle to export fence to. If set, the fence in the syncobj
192 	 * will be replaced with a fence that signals upon completion of this
193 	 * render job. 0 means ignore.
194 	 */
195 	__u32 out_sync;
196 
197 	__u32 pad2;
198 };
199 
200 /**
201  * struct drm_vc4_wait_seqno - ioctl argument for waiting for
202  * DRM_VC4_SUBMIT_CL completion using its returned seqno.
203  *
204  * timeout_ns is the timeout in nanoseconds, where "0" means "don't
205  * block, just return the status."
206  */
207 struct drm_vc4_wait_seqno {
208 	__u64 seqno;
209 	__u64 timeout_ns;
210 };
211 
212 /**
213  * struct drm_vc4_wait_bo - ioctl argument for waiting for
214  * completion of the last DRM_VC4_SUBMIT_CL on a BO.
215  *
216  * This is useful for cases where multiple processes might be
217  * rendering to a BO and you want to wait for all rendering to be
218  * completed.
219  */
220 struct drm_vc4_wait_bo {
221 	__u32 handle;
222 	__u32 pad;
223 	__u64 timeout_ns;
224 };
225 
226 /**
227  * struct drm_vc4_create_bo - ioctl argument for creating VC4 BOs.
228  *
229  * There are currently no values for the flags argument, but it may be
230  * used in a future extension.
231  */
232 struct drm_vc4_create_bo {
233 	__u32 size;
234 	__u32 flags;
235 	/** Returned GEM handle for the BO. */
236 	__u32 handle;
237 	__u32 pad;
238 };
239 
240 /**
241  * struct drm_vc4_mmap_bo - ioctl argument for mapping VC4 BOs.
242  *
243  * This doesn't actually perform an mmap.  Instead, it returns the
244  * offset you need to use in an mmap on the DRM device node.  This
245  * means that tools like valgrind end up knowing about the mapped
246  * memory.
247  *
248  * There are currently no values for the flags argument, but it may be
249  * used in a future extension.
250  */
251 struct drm_vc4_mmap_bo {
252 	/** Handle for the object being mapped. */
253 	__u32 handle;
254 	__u32 flags;
255 	/** offset into the drm node to use for subsequent mmap call. */
256 	__u64 offset;
257 };
258 
259 /**
260  * struct drm_vc4_create_shader_bo - ioctl argument for creating VC4
261  * shader BOs.
262  *
263  * Since allowing a shader to be overwritten while it's also being
264  * executed from would allow privlege escalation, shaders must be
265  * created using this ioctl, and they can't be mmapped later.
266  */
267 struct drm_vc4_create_shader_bo {
268 	/* Size of the data argument. */
269 	__u32 size;
270 	/* Flags, currently must be 0. */
271 	__u32 flags;
272 
273 	/* Pointer to the data. */
274 	__u64 data;
275 
276 	/** Returned GEM handle for the BO. */
277 	__u32 handle;
278 	/* Pad, must be 0. */
279 	__u32 pad;
280 };
281 
282 struct drm_vc4_get_hang_state_bo {
283 	__u32 handle;
284 	__u32 paddr;
285 	__u32 size;
286 	__u32 pad;
287 };
288 
289 /**
290  * struct drm_vc4_hang_state - ioctl argument for collecting state
291  * from a GPU hang for analysis.
292 */
293 struct drm_vc4_get_hang_state {
294 	/** Pointer to array of struct drm_vc4_get_hang_state_bo. */
295 	__u64 bo;
296 	/**
297 	 * On input, the size of the bo array.  Output is the number
298 	 * of bos to be returned.
299 	 */
300 	__u32 bo_count;
301 
302 	__u32 start_bin, start_render;
303 
304 	__u32 ct0ca, ct0ea;
305 	__u32 ct1ca, ct1ea;
306 	__u32 ct0cs, ct1cs;
307 	__u32 ct0ra0, ct1ra0;
308 
309 	__u32 bpca, bpcs;
310 	__u32 bpoa, bpos;
311 
312 	__u32 vpmbase;
313 
314 	__u32 dbge;
315 	__u32 fdbgo;
316 	__u32 fdbgb;
317 	__u32 fdbgr;
318 	__u32 fdbgs;
319 	__u32 errstat;
320 
321 	/* Pad that we may save more registers into in the future. */
322 	__u32 pad[16];
323 };
324 
325 #define DRM_VC4_PARAM_V3D_IDENT0		0
326 #define DRM_VC4_PARAM_V3D_IDENT1		1
327 #define DRM_VC4_PARAM_V3D_IDENT2		2
328 #define DRM_VC4_PARAM_SUPPORTS_BRANCHES		3
329 #define DRM_VC4_PARAM_SUPPORTS_ETC1		4
330 #define DRM_VC4_PARAM_SUPPORTS_THREADED_FS	5
331 #define DRM_VC4_PARAM_SUPPORTS_FIXED_RCL_ORDER	6
332 #define DRM_VC4_PARAM_SUPPORTS_MADVISE		7
333 #define DRM_VC4_PARAM_SUPPORTS_PERFMON		8
334 
335 struct drm_vc4_get_param {
336 	__u32 param;
337 	__u32 pad;
338 	__u64 value;
339 };
340 
341 struct drm_vc4_get_tiling {
342 	__u32 handle;
343 	__u32 flags;
344 	__u64 modifier;
345 };
346 
347 struct drm_vc4_set_tiling {
348 	__u32 handle;
349 	__u32 flags;
350 	__u64 modifier;
351 };
352 
353 /**
354  * struct drm_vc4_label_bo - Attach a name to a BO for debug purposes.
355  */
356 struct drm_vc4_label_bo {
357 	__u32 handle;
358 	__u32 len;
359 	__u64 name;
360 };
361 
362 /*
363  * States prefixed with '__' are internal states and cannot be passed to the
364  * DRM_IOCTL_VC4_GEM_MADVISE ioctl.
365  */
366 #define VC4_MADV_WILLNEED			0
367 #define VC4_MADV_DONTNEED			1
368 #define __VC4_MADV_PURGED			2
369 #define __VC4_MADV_NOTSUPP			3
370 
371 struct drm_vc4_gem_madvise {
372 	__u32 handle;
373 	__u32 madv;
374 	__u32 retained;
375 	__u32 pad;
376 };
377 
378 enum {
379 	VC4_PERFCNT_FEP_VALID_PRIMS_NO_RENDER,
380 	VC4_PERFCNT_FEP_VALID_PRIMS_RENDER,
381 	VC4_PERFCNT_FEP_CLIPPED_QUADS,
382 	VC4_PERFCNT_FEP_VALID_QUADS,
383 	VC4_PERFCNT_TLB_QUADS_NOT_PASSING_STENCIL,
384 	VC4_PERFCNT_TLB_QUADS_NOT_PASSING_Z_AND_STENCIL,
385 	VC4_PERFCNT_TLB_QUADS_PASSING_Z_AND_STENCIL,
386 	VC4_PERFCNT_TLB_QUADS_ZERO_COVERAGE,
387 	VC4_PERFCNT_TLB_QUADS_NON_ZERO_COVERAGE,
388 	VC4_PERFCNT_TLB_QUADS_WRITTEN_TO_COLOR_BUF,
389 	VC4_PERFCNT_PLB_PRIMS_OUTSIDE_VIEWPORT,
390 	VC4_PERFCNT_PLB_PRIMS_NEED_CLIPPING,
391 	VC4_PERFCNT_PSE_PRIMS_REVERSED,
392 	VC4_PERFCNT_QPU_TOTAL_IDLE_CYCLES,
393 	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_VERTEX_COORD_SHADING,
394 	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_FRAGMENT_SHADING,
395 	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_EXEC_VALID_INST,
396 	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_WAITING_TMUS,
397 	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_WAITING_SCOREBOARD,
398 	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_WAITING_VARYINGS,
399 	VC4_PERFCNT_QPU_TOTAL_INST_CACHE_HIT,
400 	VC4_PERFCNT_QPU_TOTAL_INST_CACHE_MISS,
401 	VC4_PERFCNT_QPU_TOTAL_UNIFORM_CACHE_HIT,
402 	VC4_PERFCNT_QPU_TOTAL_UNIFORM_CACHE_MISS,
403 	VC4_PERFCNT_TMU_TOTAL_TEXT_QUADS_PROCESSED,
404 	VC4_PERFCNT_TMU_TOTAL_TEXT_CACHE_MISS,
405 	VC4_PERFCNT_VPM_TOTAL_CLK_CYCLES_VDW_STALLED,
406 	VC4_PERFCNT_VPM_TOTAL_CLK_CYCLES_VCD_STALLED,
407 	VC4_PERFCNT_L2C_TOTAL_L2_CACHE_HIT,
408 	VC4_PERFCNT_L2C_TOTAL_L2_CACHE_MISS,
409 	VC4_PERFCNT_NUM_EVENTS,
410 };
411 
412 #define DRM_VC4_MAX_PERF_COUNTERS	16
413 
414 struct drm_vc4_perfmon_create {
415 	__u32 id;
416 	__u32 ncounters;
417 	__u8 events[DRM_VC4_MAX_PERF_COUNTERS];
418 };
419 
420 struct drm_vc4_perfmon_destroy {
421 	__u32 id;
422 };
423 
424 /*
425  * Returns the values of the performance counters tracked by this
426  * perfmon (as an array of ncounters u64 values).
427  *
428  * No implicit synchronization is performed, so the user has to
429  * guarantee that any jobs using this perfmon have already been
430  * completed  (probably by blocking on the seqno returned by the
431  * last exec that used the perfmon).
432  */
433 struct drm_vc4_perfmon_get_values {
434 	__u32 id;
435 	__u64 values_ptr;
436 };
437 
438 #if defined(__cplusplus)
439 }
440 #endif
441 
442 #endif /* _UAPI_VC4_DRM_H_ */
443