xref: /openbmc/linux/include/uapi/drm/vc4_drm.h (revision 260ea95c)
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 
44 #define DRM_IOCTL_VC4_SUBMIT_CL           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SUBMIT_CL, struct drm_vc4_submit_cl)
45 #define DRM_IOCTL_VC4_WAIT_SEQNO          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_SEQNO, struct drm_vc4_wait_seqno)
46 #define DRM_IOCTL_VC4_WAIT_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_BO, struct drm_vc4_wait_bo)
47 #define DRM_IOCTL_VC4_CREATE_BO           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_BO, struct drm_vc4_create_bo)
48 #define DRM_IOCTL_VC4_MMAP_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_MMAP_BO, struct drm_vc4_mmap_bo)
49 #define DRM_IOCTL_VC4_CREATE_SHADER_BO    DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_SHADER_BO, struct drm_vc4_create_shader_bo)
50 #define DRM_IOCTL_VC4_GET_HANG_STATE      DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_HANG_STATE, struct drm_vc4_get_hang_state)
51 #define DRM_IOCTL_VC4_GET_PARAM           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_PARAM, struct drm_vc4_get_param)
52 #define DRM_IOCTL_VC4_SET_TILING          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SET_TILING, struct drm_vc4_set_tiling)
53 #define DRM_IOCTL_VC4_GET_TILING          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_TILING, struct drm_vc4_get_tiling)
54 
55 struct drm_vc4_submit_rcl_surface {
56 	__u32 hindex; /* Handle index, or ~0 if not present. */
57 	__u32 offset; /* Offset to start of buffer. */
58 	/*
59 	 * Bits for either render config (color_write) or load/store packet.
60 	 * Bits should all be 0 for MSAA load/stores.
61 	 */
62 	__u16 bits;
63 
64 #define VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES		(1 << 0)
65 	__u16 flags;
66 };
67 
68 /**
69  * struct drm_vc4_submit_cl - ioctl argument for submitting commands to the 3D
70  * engine.
71  *
72  * Drivers typically use GPU BOs to store batchbuffers / command lists and
73  * their associated state.  However, because the VC4 lacks an MMU, we have to
74  * do validation of memory accesses by the GPU commands.  If we were to store
75  * our commands in BOs, we'd need to do uncached readback from them to do the
76  * validation process, which is too expensive.  Instead, userspace accumulates
77  * commands and associated state in plain memory, then the kernel copies the
78  * data to its own address space, and then validates and stores it in a GPU
79  * BO.
80  */
81 struct drm_vc4_submit_cl {
82 	/* Pointer to the binner command list.
83 	 *
84 	 * This is the first set of commands executed, which runs the
85 	 * coordinate shader to determine where primitives land on the screen,
86 	 * then writes out the state updates and draw calls necessary per tile
87 	 * to the tile allocation BO.
88 	 */
89 	__u64 bin_cl;
90 
91 	/* Pointer to the shader records.
92 	 *
93 	 * Shader records are the structures read by the hardware that contain
94 	 * pointers to uniforms, shaders, and vertex attributes.  The
95 	 * reference to the shader record has enough information to determine
96 	 * how many pointers are necessary (fixed number for shaders/uniforms,
97 	 * and an attribute count), so those BO indices into bo_handles are
98 	 * just stored as __u32s before each shader record passed in.
99 	 */
100 	__u64 shader_rec;
101 
102 	/* Pointer to uniform data and texture handles for the textures
103 	 * referenced by the shader.
104 	 *
105 	 * For each shader state record, there is a set of uniform data in the
106 	 * order referenced by the record (FS, VS, then CS).  Each set of
107 	 * uniform data has a __u32 index into bo_handles per texture
108 	 * sample operation, in the order the QPU_W_TMUn_S writes appear in
109 	 * the program.  Following the texture BO handle indices is the actual
110 	 * uniform data.
111 	 *
112 	 * The individual uniform state blocks don't have sizes passed in,
113 	 * because the kernel has to determine the sizes anyway during shader
114 	 * code validation.
115 	 */
116 	__u64 uniforms;
117 	__u64 bo_handles;
118 
119 	/* Size in bytes of the binner command list. */
120 	__u32 bin_cl_size;
121 	/* Size in bytes of the set of shader records. */
122 	__u32 shader_rec_size;
123 	/* Number of shader records.
124 	 *
125 	 * This could just be computed from the contents of shader_records and
126 	 * the address bits of references to them from the bin CL, but it
127 	 * keeps the kernel from having to resize some allocations it makes.
128 	 */
129 	__u32 shader_rec_count;
130 	/* Size in bytes of the uniform state. */
131 	__u32 uniforms_size;
132 
133 	/* Number of BO handles passed in (size is that times 4). */
134 	__u32 bo_handle_count;
135 
136 	/* RCL setup: */
137 	__u16 width;
138 	__u16 height;
139 	__u8 min_x_tile;
140 	__u8 min_y_tile;
141 	__u8 max_x_tile;
142 	__u8 max_y_tile;
143 	struct drm_vc4_submit_rcl_surface color_read;
144 	struct drm_vc4_submit_rcl_surface color_write;
145 	struct drm_vc4_submit_rcl_surface zs_read;
146 	struct drm_vc4_submit_rcl_surface zs_write;
147 	struct drm_vc4_submit_rcl_surface msaa_color_write;
148 	struct drm_vc4_submit_rcl_surface msaa_zs_write;
149 	__u32 clear_color[2];
150 	__u32 clear_z;
151 	__u8 clear_s;
152 
153 	__u32 pad:24;
154 
155 #define VC4_SUBMIT_CL_USE_CLEAR_COLOR			(1 << 0)
156 	__u32 flags;
157 
158 	/* Returned value of the seqno of this render job (for the
159 	 * wait ioctl).
160 	 */
161 	__u64 seqno;
162 };
163 
164 /**
165  * struct drm_vc4_wait_seqno - ioctl argument for waiting for
166  * DRM_VC4_SUBMIT_CL completion using its returned seqno.
167  *
168  * timeout_ns is the timeout in nanoseconds, where "0" means "don't
169  * block, just return the status."
170  */
171 struct drm_vc4_wait_seqno {
172 	__u64 seqno;
173 	__u64 timeout_ns;
174 };
175 
176 /**
177  * struct drm_vc4_wait_bo - ioctl argument for waiting for
178  * completion of the last DRM_VC4_SUBMIT_CL on a BO.
179  *
180  * This is useful for cases where multiple processes might be
181  * rendering to a BO and you want to wait for all rendering to be
182  * completed.
183  */
184 struct drm_vc4_wait_bo {
185 	__u32 handle;
186 	__u32 pad;
187 	__u64 timeout_ns;
188 };
189 
190 /**
191  * struct drm_vc4_create_bo - ioctl argument for creating VC4 BOs.
192  *
193  * There are currently no values for the flags argument, but it may be
194  * used in a future extension.
195  */
196 struct drm_vc4_create_bo {
197 	__u32 size;
198 	__u32 flags;
199 	/** Returned GEM handle for the BO. */
200 	__u32 handle;
201 	__u32 pad;
202 };
203 
204 /**
205  * struct drm_vc4_mmap_bo - ioctl argument for mapping VC4 BOs.
206  *
207  * This doesn't actually perform an mmap.  Instead, it returns the
208  * offset you need to use in an mmap on the DRM device node.  This
209  * means that tools like valgrind end up knowing about the mapped
210  * memory.
211  *
212  * There are currently no values for the flags argument, but it may be
213  * used in a future extension.
214  */
215 struct drm_vc4_mmap_bo {
216 	/** Handle for the object being mapped. */
217 	__u32 handle;
218 	__u32 flags;
219 	/** offset into the drm node to use for subsequent mmap call. */
220 	__u64 offset;
221 };
222 
223 /**
224  * struct drm_vc4_create_shader_bo - ioctl argument for creating VC4
225  * shader BOs.
226  *
227  * Since allowing a shader to be overwritten while it's also being
228  * executed from would allow privlege escalation, shaders must be
229  * created using this ioctl, and they can't be mmapped later.
230  */
231 struct drm_vc4_create_shader_bo {
232 	/* Size of the data argument. */
233 	__u32 size;
234 	/* Flags, currently must be 0. */
235 	__u32 flags;
236 
237 	/* Pointer to the data. */
238 	__u64 data;
239 
240 	/** Returned GEM handle for the BO. */
241 	__u32 handle;
242 	/* Pad, must be 0. */
243 	__u32 pad;
244 };
245 
246 struct drm_vc4_get_hang_state_bo {
247 	__u32 handle;
248 	__u32 paddr;
249 	__u32 size;
250 	__u32 pad;
251 };
252 
253 /**
254  * struct drm_vc4_hang_state - ioctl argument for collecting state
255  * from a GPU hang for analysis.
256 */
257 struct drm_vc4_get_hang_state {
258 	/** Pointer to array of struct drm_vc4_get_hang_state_bo. */
259 	__u64 bo;
260 	/**
261 	 * On input, the size of the bo array.  Output is the number
262 	 * of bos to be returned.
263 	 */
264 	__u32 bo_count;
265 
266 	__u32 start_bin, start_render;
267 
268 	__u32 ct0ca, ct0ea;
269 	__u32 ct1ca, ct1ea;
270 	__u32 ct0cs, ct1cs;
271 	__u32 ct0ra0, ct1ra0;
272 
273 	__u32 bpca, bpcs;
274 	__u32 bpoa, bpos;
275 
276 	__u32 vpmbase;
277 
278 	__u32 dbge;
279 	__u32 fdbgo;
280 	__u32 fdbgb;
281 	__u32 fdbgr;
282 	__u32 fdbgs;
283 	__u32 errstat;
284 
285 	/* Pad that we may save more registers into in the future. */
286 	__u32 pad[16];
287 };
288 
289 #define DRM_VC4_PARAM_V3D_IDENT0		0
290 #define DRM_VC4_PARAM_V3D_IDENT1		1
291 #define DRM_VC4_PARAM_V3D_IDENT2		2
292 #define DRM_VC4_PARAM_SUPPORTS_BRANCHES		3
293 #define DRM_VC4_PARAM_SUPPORTS_ETC1		4
294 #define DRM_VC4_PARAM_SUPPORTS_THREADED_FS	5
295 
296 struct drm_vc4_get_param {
297 	__u32 param;
298 	__u32 pad;
299 	__u64 value;
300 };
301 
302 struct drm_vc4_get_tiling {
303 	__u32 handle;
304 	__u32 flags;
305 	__u64 modifier;
306 };
307 
308 struct drm_vc4_set_tiling {
309 	__u32 handle;
310 	__u32 flags;
311 	__u64 modifier;
312 };
313 
314 #if defined(__cplusplus)
315 }
316 #endif
317 
318 #endif /* _UAPI_VC4_DRM_H_ */
319