xref: /linux/include/uapi/drm/vc4_drm.h (revision d257f9bf06129613de539ea71ecea60848b662cd)
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 
45 #define DRM_IOCTL_VC4_SUBMIT_CL           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SUBMIT_CL, struct drm_vc4_submit_cl)
46 #define DRM_IOCTL_VC4_WAIT_SEQNO          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_SEQNO, struct drm_vc4_wait_seqno)
47 #define DRM_IOCTL_VC4_WAIT_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_BO, struct drm_vc4_wait_bo)
48 #define DRM_IOCTL_VC4_CREATE_BO           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_BO, struct drm_vc4_create_bo)
49 #define DRM_IOCTL_VC4_MMAP_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_MMAP_BO, struct drm_vc4_mmap_bo)
50 #define DRM_IOCTL_VC4_CREATE_SHADER_BO    DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_SHADER_BO, struct drm_vc4_create_shader_bo)
51 #define DRM_IOCTL_VC4_GET_HANG_STATE      DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_HANG_STATE, struct drm_vc4_get_hang_state)
52 #define DRM_IOCTL_VC4_GET_PARAM           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_PARAM, struct drm_vc4_get_param)
53 #define DRM_IOCTL_VC4_SET_TILING          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SET_TILING, struct drm_vc4_set_tiling)
54 #define DRM_IOCTL_VC4_GET_TILING          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_TILING, struct drm_vc4_get_tiling)
55 #define DRM_IOCTL_VC4_LABEL_BO            DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_LABEL_BO, struct drm_vc4_label_bo)
56 
57 struct drm_vc4_submit_rcl_surface {
58 	__u32 hindex; /* Handle index, or ~0 if not present. */
59 	__u32 offset; /* Offset to start of buffer. */
60 	/*
61 	 * Bits for either render config (color_write) or load/store packet.
62 	 * Bits should all be 0 for MSAA load/stores.
63 	 */
64 	__u16 bits;
65 
66 #define VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES		(1 << 0)
67 	__u16 flags;
68 };
69 
70 /**
71  * struct drm_vc4_submit_cl - ioctl argument for submitting commands to the 3D
72  * engine.
73  *
74  * Drivers typically use GPU BOs to store batchbuffers / command lists and
75  * their associated state.  However, because the VC4 lacks an MMU, we have to
76  * do validation of memory accesses by the GPU commands.  If we were to store
77  * our commands in BOs, we'd need to do uncached readback from them to do the
78  * validation process, which is too expensive.  Instead, userspace accumulates
79  * commands and associated state in plain memory, then the kernel copies the
80  * data to its own address space, and then validates and stores it in a GPU
81  * BO.
82  */
83 struct drm_vc4_submit_cl {
84 	/* Pointer to the binner command list.
85 	 *
86 	 * This is the first set of commands executed, which runs the
87 	 * coordinate shader to determine where primitives land on the screen,
88 	 * then writes out the state updates and draw calls necessary per tile
89 	 * to the tile allocation BO.
90 	 */
91 	__u64 bin_cl;
92 
93 	/* Pointer to the shader records.
94 	 *
95 	 * Shader records are the structures read by the hardware that contain
96 	 * pointers to uniforms, shaders, and vertex attributes.  The
97 	 * reference to the shader record has enough information to determine
98 	 * how many pointers are necessary (fixed number for shaders/uniforms,
99 	 * and an attribute count), so those BO indices into bo_handles are
100 	 * just stored as __u32s before each shader record passed in.
101 	 */
102 	__u64 shader_rec;
103 
104 	/* Pointer to uniform data and texture handles for the textures
105 	 * referenced by the shader.
106 	 *
107 	 * For each shader state record, there is a set of uniform data in the
108 	 * order referenced by the record (FS, VS, then CS).  Each set of
109 	 * uniform data has a __u32 index into bo_handles per texture
110 	 * sample operation, in the order the QPU_W_TMUn_S writes appear in
111 	 * the program.  Following the texture BO handle indices is the actual
112 	 * uniform data.
113 	 *
114 	 * The individual uniform state blocks don't have sizes passed in,
115 	 * because the kernel has to determine the sizes anyway during shader
116 	 * code validation.
117 	 */
118 	__u64 uniforms;
119 	__u64 bo_handles;
120 
121 	/* Size in bytes of the binner command list. */
122 	__u32 bin_cl_size;
123 	/* Size in bytes of the set of shader records. */
124 	__u32 shader_rec_size;
125 	/* Number of shader records.
126 	 *
127 	 * This could just be computed from the contents of shader_records and
128 	 * the address bits of references to them from the bin CL, but it
129 	 * keeps the kernel from having to resize some allocations it makes.
130 	 */
131 	__u32 shader_rec_count;
132 	/* Size in bytes of the uniform state. */
133 	__u32 uniforms_size;
134 
135 	/* Number of BO handles passed in (size is that times 4). */
136 	__u32 bo_handle_count;
137 
138 	/* RCL setup: */
139 	__u16 width;
140 	__u16 height;
141 	__u8 min_x_tile;
142 	__u8 min_y_tile;
143 	__u8 max_x_tile;
144 	__u8 max_y_tile;
145 	struct drm_vc4_submit_rcl_surface color_read;
146 	struct drm_vc4_submit_rcl_surface color_write;
147 	struct drm_vc4_submit_rcl_surface zs_read;
148 	struct drm_vc4_submit_rcl_surface zs_write;
149 	struct drm_vc4_submit_rcl_surface msaa_color_write;
150 	struct drm_vc4_submit_rcl_surface msaa_zs_write;
151 	__u32 clear_color[2];
152 	__u32 clear_z;
153 	__u8 clear_s;
154 
155 	__u32 pad:24;
156 
157 #define VC4_SUBMIT_CL_USE_CLEAR_COLOR			(1 << 0)
158 /* By default, the kernel gets to choose the order that the tiles are
159  * rendered in.  If this is set, then the tiles will be rendered in a
160  * raster order, with the right-to-left vs left-to-right and
161  * top-to-bottom vs bottom-to-top dictated by
162  * VC4_SUBMIT_CL_RCL_ORDER_INCREASING_*.  This allows overlapping
163  * blits to be implemented using the 3D engine.
164  */
165 #define VC4_SUBMIT_CL_FIXED_RCL_ORDER			(1 << 1)
166 #define VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X		(1 << 2)
167 #define VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y		(1 << 3)
168 	__u32 flags;
169 
170 	/* Returned value of the seqno of this render job (for the
171 	 * wait ioctl).
172 	 */
173 	__u64 seqno;
174 };
175 
176 /**
177  * struct drm_vc4_wait_seqno - ioctl argument for waiting for
178  * DRM_VC4_SUBMIT_CL completion using its returned seqno.
179  *
180  * timeout_ns is the timeout in nanoseconds, where "0" means "don't
181  * block, just return the status."
182  */
183 struct drm_vc4_wait_seqno {
184 	__u64 seqno;
185 	__u64 timeout_ns;
186 };
187 
188 /**
189  * struct drm_vc4_wait_bo - ioctl argument for waiting for
190  * completion of the last DRM_VC4_SUBMIT_CL on a BO.
191  *
192  * This is useful for cases where multiple processes might be
193  * rendering to a BO and you want to wait for all rendering to be
194  * completed.
195  */
196 struct drm_vc4_wait_bo {
197 	__u32 handle;
198 	__u32 pad;
199 	__u64 timeout_ns;
200 };
201 
202 /**
203  * struct drm_vc4_create_bo - ioctl argument for creating VC4 BOs.
204  *
205  * There are currently no values for the flags argument, but it may be
206  * used in a future extension.
207  */
208 struct drm_vc4_create_bo {
209 	__u32 size;
210 	__u32 flags;
211 	/** Returned GEM handle for the BO. */
212 	__u32 handle;
213 	__u32 pad;
214 };
215 
216 /**
217  * struct drm_vc4_mmap_bo - ioctl argument for mapping VC4 BOs.
218  *
219  * This doesn't actually perform an mmap.  Instead, it returns the
220  * offset you need to use in an mmap on the DRM device node.  This
221  * means that tools like valgrind end up knowing about the mapped
222  * memory.
223  *
224  * There are currently no values for the flags argument, but it may be
225  * used in a future extension.
226  */
227 struct drm_vc4_mmap_bo {
228 	/** Handle for the object being mapped. */
229 	__u32 handle;
230 	__u32 flags;
231 	/** offset into the drm node to use for subsequent mmap call. */
232 	__u64 offset;
233 };
234 
235 /**
236  * struct drm_vc4_create_shader_bo - ioctl argument for creating VC4
237  * shader BOs.
238  *
239  * Since allowing a shader to be overwritten while it's also being
240  * executed from would allow privlege escalation, shaders must be
241  * created using this ioctl, and they can't be mmapped later.
242  */
243 struct drm_vc4_create_shader_bo {
244 	/* Size of the data argument. */
245 	__u32 size;
246 	/* Flags, currently must be 0. */
247 	__u32 flags;
248 
249 	/* Pointer to the data. */
250 	__u64 data;
251 
252 	/** Returned GEM handle for the BO. */
253 	__u32 handle;
254 	/* Pad, must be 0. */
255 	__u32 pad;
256 };
257 
258 struct drm_vc4_get_hang_state_bo {
259 	__u32 handle;
260 	__u32 paddr;
261 	__u32 size;
262 	__u32 pad;
263 };
264 
265 /**
266  * struct drm_vc4_hang_state - ioctl argument for collecting state
267  * from a GPU hang for analysis.
268 */
269 struct drm_vc4_get_hang_state {
270 	/** Pointer to array of struct drm_vc4_get_hang_state_bo. */
271 	__u64 bo;
272 	/**
273 	 * On input, the size of the bo array.  Output is the number
274 	 * of bos to be returned.
275 	 */
276 	__u32 bo_count;
277 
278 	__u32 start_bin, start_render;
279 
280 	__u32 ct0ca, ct0ea;
281 	__u32 ct1ca, ct1ea;
282 	__u32 ct0cs, ct1cs;
283 	__u32 ct0ra0, ct1ra0;
284 
285 	__u32 bpca, bpcs;
286 	__u32 bpoa, bpos;
287 
288 	__u32 vpmbase;
289 
290 	__u32 dbge;
291 	__u32 fdbgo;
292 	__u32 fdbgb;
293 	__u32 fdbgr;
294 	__u32 fdbgs;
295 	__u32 errstat;
296 
297 	/* Pad that we may save more registers into in the future. */
298 	__u32 pad[16];
299 };
300 
301 #define DRM_VC4_PARAM_V3D_IDENT0		0
302 #define DRM_VC4_PARAM_V3D_IDENT1		1
303 #define DRM_VC4_PARAM_V3D_IDENT2		2
304 #define DRM_VC4_PARAM_SUPPORTS_BRANCHES		3
305 #define DRM_VC4_PARAM_SUPPORTS_ETC1		4
306 #define DRM_VC4_PARAM_SUPPORTS_THREADED_FS	5
307 #define DRM_VC4_PARAM_SUPPORTS_FIXED_RCL_ORDER	6
308 
309 struct drm_vc4_get_param {
310 	__u32 param;
311 	__u32 pad;
312 	__u64 value;
313 };
314 
315 struct drm_vc4_get_tiling {
316 	__u32 handle;
317 	__u32 flags;
318 	__u64 modifier;
319 };
320 
321 struct drm_vc4_set_tiling {
322 	__u32 handle;
323 	__u32 flags;
324 	__u64 modifier;
325 };
326 
327 /**
328  * struct drm_vc4_label_bo - Attach a name to a BO for debug purposes.
329  */
330 struct drm_vc4_label_bo {
331 	__u32 handle;
332 	__u32 len;
333 	__u64 name;
334 };
335 
336 #if defined(__cplusplus)
337 }
338 #endif
339 
340 #endif /* _UAPI_VC4_DRM_H_ */
341