xref: /openbmc/linux/include/uapi/drm/ivpu_accel.h (revision 18afb028)
1 /* SPDX-License-Identifier: GPL-2.0-only WITH Linux-syscall-note */
2 /*
3  * Copyright (C) 2020-2023 Intel Corporation
4  */
5 
6 #ifndef __UAPI_IVPU_DRM_H__
7 #define __UAPI_IVPU_DRM_H__
8 
9 #include "drm.h"
10 
11 #if defined(__cplusplus)
12 extern "C" {
13 #endif
14 
15 #define DRM_IVPU_DRIVER_MAJOR 1
16 #define DRM_IVPU_DRIVER_MINOR 0
17 
18 #define DRM_IVPU_GET_PARAM		  0x00
19 #define DRM_IVPU_SET_PARAM		  0x01
20 #define DRM_IVPU_BO_CREATE		  0x02
21 #define DRM_IVPU_BO_INFO		  0x03
22 #define DRM_IVPU_SUBMIT			  0x05
23 #define DRM_IVPU_BO_WAIT		  0x06
24 
25 #define DRM_IOCTL_IVPU_GET_PARAM                                               \
26 	DRM_IOWR(DRM_COMMAND_BASE + DRM_IVPU_GET_PARAM, struct drm_ivpu_param)
27 
28 #define DRM_IOCTL_IVPU_SET_PARAM                                               \
29 	DRM_IOW(DRM_COMMAND_BASE + DRM_IVPU_SET_PARAM, struct drm_ivpu_param)
30 
31 #define DRM_IOCTL_IVPU_BO_CREATE                                               \
32 	DRM_IOWR(DRM_COMMAND_BASE + DRM_IVPU_BO_CREATE, struct drm_ivpu_bo_create)
33 
34 #define DRM_IOCTL_IVPU_BO_INFO                                                 \
35 	DRM_IOWR(DRM_COMMAND_BASE + DRM_IVPU_BO_INFO, struct drm_ivpu_bo_info)
36 
37 #define DRM_IOCTL_IVPU_SUBMIT                                                  \
38 	DRM_IOW(DRM_COMMAND_BASE + DRM_IVPU_SUBMIT, struct drm_ivpu_submit)
39 
40 #define DRM_IOCTL_IVPU_BO_WAIT                                                 \
41 	DRM_IOWR(DRM_COMMAND_BASE + DRM_IVPU_BO_WAIT, struct drm_ivpu_bo_wait)
42 
43 /**
44  * DOC: contexts
45  *
46  * VPU contexts have private virtual address space, job queues and priority.
47  * Each context is identified by an unique ID. Context is created on open().
48  */
49 
50 #define DRM_IVPU_PARAM_DEVICE_ID	    0
51 #define DRM_IVPU_PARAM_DEVICE_REVISION	    1
52 #define DRM_IVPU_PARAM_PLATFORM_TYPE	    2
53 #define DRM_IVPU_PARAM_CORE_CLOCK_RATE	    3
54 #define DRM_IVPU_PARAM_NUM_CONTEXTS	    4
55 #define DRM_IVPU_PARAM_CONTEXT_BASE_ADDRESS 5
56 #define DRM_IVPU_PARAM_CONTEXT_PRIORITY	    6
57 #define DRM_IVPU_PARAM_CONTEXT_ID	    7
58 #define DRM_IVPU_PARAM_FW_API_VERSION	    8
59 #define DRM_IVPU_PARAM_ENGINE_HEARTBEAT	    9
60 #define DRM_IVPU_PARAM_UNIQUE_INFERENCE_ID  10
61 #define DRM_IVPU_PARAM_TILE_CONFIG	    11
62 #define DRM_IVPU_PARAM_SKU		    12
63 #define DRM_IVPU_PARAM_CAPABILITIES	    13
64 
65 #define DRM_IVPU_PLATFORM_TYPE_SILICON	    0
66 
67 #define DRM_IVPU_CONTEXT_PRIORITY_IDLE	    0
68 #define DRM_IVPU_CONTEXT_PRIORITY_NORMAL    1
69 #define DRM_IVPU_CONTEXT_PRIORITY_FOCUS	    2
70 #define DRM_IVPU_CONTEXT_PRIORITY_REALTIME  3
71 
72 #define DRM_IVPU_CAP_METRIC_STREAMER	    1
73 #define DRM_IVPU_CAP_DMA_MEMORY_RANGE       2
74 
75 /**
76  * struct drm_ivpu_param - Get/Set VPU parameters
77  */
78 struct drm_ivpu_param {
79 	/**
80 	 * @param:
81 	 *
82 	 * Supported params:
83 	 *
84 	 * %DRM_IVPU_PARAM_DEVICE_ID:
85 	 * PCI Device ID of the VPU device (read-only)
86 	 *
87 	 * %DRM_IVPU_PARAM_DEVICE_REVISION:
88 	 * VPU device revision (read-only)
89 	 *
90 	 * %DRM_IVPU_PARAM_PLATFORM_TYPE:
91 	 * Returns %DRM_IVPU_PLATFORM_TYPE_SILICON on real hardware or device specific
92 	 * platform type when executing on a simulator or emulator (read-only)
93 	 *
94 	 * %DRM_IVPU_PARAM_CORE_CLOCK_RATE:
95 	 * Current PLL frequency (read-only)
96 	 *
97 	 * %DRM_IVPU_PARAM_NUM_CONTEXTS:
98 	 * Maximum number of simultaneously existing contexts (read-only)
99 	 *
100 	 * %DRM_IVPU_PARAM_CONTEXT_BASE_ADDRESS:
101 	 * Lowest VPU virtual address available in the current context (read-only)
102 	 *
103 	 * %DRM_IVPU_PARAM_CONTEXT_PRIORITY:
104 	 * Value of current context scheduling priority (read-write).
105 	 * See DRM_IVPU_CONTEXT_PRIORITY_* for possible values.
106 	 *
107 	 * %DRM_IVPU_PARAM_CONTEXT_ID:
108 	 * Current context ID, always greater than 0 (read-only)
109 	 *
110 	 * %DRM_IVPU_PARAM_FW_API_VERSION:
111 	 * Firmware API version array (read-only)
112 	 *
113 	 * %DRM_IVPU_PARAM_ENGINE_HEARTBEAT:
114 	 * Heartbeat value from an engine (read-only).
115 	 * Engine ID (i.e. DRM_IVPU_ENGINE_COMPUTE) is given via index.
116 	 *
117 	 * %DRM_IVPU_PARAM_UNIQUE_INFERENCE_ID:
118 	 * Device-unique inference ID (read-only)
119 	 *
120 	 * %DRM_IVPU_PARAM_TILE_CONFIG:
121 	 * VPU tile configuration  (read-only)
122 	 *
123 	 * %DRM_IVPU_PARAM_SKU:
124 	 * VPU SKU ID (read-only)
125 	 *
126 	 */
127 	__u32 param;
128 
129 	/** @index: Index for params that have multiple instances */
130 	__u32 index;
131 
132 	/** @value: Param value */
133 	__u64 value;
134 };
135 
136 #define DRM_IVPU_BO_SHAVE_MEM  0x00000001
137 #define DRM_IVPU_BO_HIGH_MEM   DRM_IVPU_BO_SHAVE_MEM
138 #define DRM_IVPU_BO_MAPPABLE   0x00000002
139 #define DRM_IVPU_BO_DMA_MEM    0x00000004
140 
141 #define DRM_IVPU_BO_CACHED     0x00000000
142 #define DRM_IVPU_BO_UNCACHED   0x00010000
143 #define DRM_IVPU_BO_WC	       0x00020000
144 #define DRM_IVPU_BO_CACHE_MASK 0x00030000
145 
146 #define DRM_IVPU_BO_FLAGS \
147 	(DRM_IVPU_BO_HIGH_MEM | \
148 	 DRM_IVPU_BO_MAPPABLE | \
149 	 DRM_IVPU_BO_DMA_MEM | \
150 	 DRM_IVPU_BO_CACHE_MASK)
151 
152 /**
153  * struct drm_ivpu_bo_create - Create BO backed by SHMEM
154  *
155  * Create GEM buffer object allocated in SHMEM memory.
156  */
157 struct drm_ivpu_bo_create {
158 	/** @size: The size in bytes of the allocated memory */
159 	__u64 size;
160 
161 	/**
162 	 * @flags:
163 	 *
164 	 * Supported flags:
165 	 *
166 	 * %DRM_IVPU_BO_HIGH_MEM:
167 	 *
168 	 * Allocate VPU address from >4GB range.
169 	 * Buffer object with vpu address >4GB can be always accessed by the
170 	 * VPU DMA engine, but some HW generation may not be able to access
171 	 * this memory from then firmware running on the VPU management processor.
172 	 * Suitable for input, output and some scratch buffers.
173 	 *
174 	 * %DRM_IVPU_BO_MAPPABLE:
175 	 *
176 	 * Buffer object can be mapped using mmap().
177 	 *
178 	 * %DRM_IVPU_BO_CACHED:
179 	 *
180 	 * Allocated BO will be cached on host side (WB) and snooped on the VPU side.
181 	 * This is the default caching mode.
182 	 *
183 	 * %DRM_IVPU_BO_UNCACHED:
184 	 *
185 	 * Allocated BO will not be cached on host side nor snooped on the VPU side.
186 	 *
187 	 * %DRM_IVPU_BO_WC:
188 	 *
189 	 * Allocated BO will use write combining buffer for writes but reads will be
190 	 * uncached.
191 	 */
192 	__u32 flags;
193 
194 	/** @handle: Returned GEM object handle */
195 	__u32 handle;
196 
197 	/** @vpu_addr: Returned VPU virtual address */
198 	__u64 vpu_addr;
199 };
200 
201 /**
202  * struct drm_ivpu_bo_info - Query buffer object info
203  */
204 struct drm_ivpu_bo_info {
205 	/** @handle: Handle of the queried BO */
206 	__u32 handle;
207 
208 	/** @flags: Returned flags used to create the BO */
209 	__u32 flags;
210 
211 	/** @vpu_addr: Returned VPU virtual address */
212 	__u64 vpu_addr;
213 
214 	/**
215 	 * @mmap_offset:
216 	 *
217 	 * Returned offset to be used in mmap(). 0 in case the BO is not mappable.
218 	 */
219 	__u64 mmap_offset;
220 
221 	/** @size: Returned GEM object size, aligned to PAGE_SIZE */
222 	__u64 size;
223 };
224 
225 /* drm_ivpu_submit engines */
226 #define DRM_IVPU_ENGINE_COMPUTE 0
227 #define DRM_IVPU_ENGINE_COPY    1
228 
229 /**
230  * struct drm_ivpu_submit - Submit commands to the VPU
231  *
232  * Execute a single command buffer on a given VPU engine.
233  * Handles to all referenced buffer objects have to be provided in @buffers_ptr.
234  *
235  * User space may wait on job completion using %DRM_IVPU_BO_WAIT ioctl.
236  */
237 struct drm_ivpu_submit {
238 	/**
239 	 * @buffers_ptr:
240 	 *
241 	 * A pointer to an u32 array of GEM handles of the BOs required for this job.
242 	 * The number of elements in the array must be equal to the value given by @buffer_count.
243 	 *
244 	 * The first BO is the command buffer. The rest of array has to contain all
245 	 * BOs referenced from the command buffer.
246 	 */
247 	__u64 buffers_ptr;
248 
249 	/** @buffer_count: Number of elements in the @buffers_ptr */
250 	__u32 buffer_count;
251 
252 	/**
253 	 * @engine: Select the engine this job should be executed on
254 	 *
255 	 * %DRM_IVPU_ENGINE_COMPUTE:
256 	 *
257 	 * Performs Deep Learning Neural Compute Inference Operations
258 	 *
259 	 * %DRM_IVPU_ENGINE_COPY:
260 	 *
261 	 * Performs memory copy operations to/from system memory allocated for VPU
262 	 */
263 	__u32 engine;
264 
265 	/** @flags: Reserved for future use - must be zero */
266 	__u32 flags;
267 
268 	/**
269 	 * @commands_offset:
270 	 *
271 	 * Offset inside the first buffer in @buffers_ptr containing commands
272 	 * to be executed. The offset has to be 8-byte aligned.
273 	 */
274 	__u32 commands_offset;
275 };
276 
277 /* drm_ivpu_bo_wait job status codes */
278 #define DRM_IVPU_JOB_STATUS_SUCCESS 0
279 
280 /**
281  * struct drm_ivpu_bo_wait - Wait for BO to become inactive
282  *
283  * Blocks until a given buffer object becomes inactive.
284  * With @timeout_ms set to 0 returns immediately.
285  */
286 struct drm_ivpu_bo_wait {
287 	/** @handle: Handle to the buffer object to be waited on */
288 	__u32 handle;
289 
290 	/** @flags: Reserved for future use - must be zero */
291 	__u32 flags;
292 
293 	/** @timeout_ns: Absolute timeout in nanoseconds (may be zero) */
294 	__s64 timeout_ns;
295 
296 	/**
297 	 * @job_status:
298 	 *
299 	 * Job status code which is updated after the job is completed.
300 	 * &DRM_IVPU_JOB_STATUS_SUCCESS or device specific error otherwise.
301 	 * Valid only if @handle points to a command buffer.
302 	 */
303 	__u32 job_status;
304 
305 	/** @pad: Padding - must be zero */
306 	__u32 pad;
307 };
308 
309 #if defined(__cplusplus)
310 }
311 #endif
312 
313 #endif /* __UAPI_IVPU_DRM_H__ */
314