xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_drv.h (revision 2596e07a)
1 /*
2  * Copyright (C) 2015 Broadcom
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  */
8 
9 #include "drmP.h"
10 #include "drm_gem_cma_helper.h"
11 
12 struct vc4_dev {
13 	struct drm_device *dev;
14 
15 	struct vc4_hdmi *hdmi;
16 	struct vc4_hvs *hvs;
17 	struct vc4_crtc *crtc[3];
18 	struct vc4_v3d *v3d;
19 
20 	struct drm_fbdev_cma *fbdev;
21 
22 	struct vc4_hang_state *hang_state;
23 
24 	/* The kernel-space BO cache.  Tracks buffers that have been
25 	 * unreferenced by all other users (refcounts of 0!) but not
26 	 * yet freed, so we can do cheap allocations.
27 	 */
28 	struct vc4_bo_cache {
29 		/* Array of list heads for entries in the BO cache,
30 		 * based on number of pages, so we can do O(1) lookups
31 		 * in the cache when allocating.
32 		 */
33 		struct list_head *size_list;
34 		uint32_t size_list_size;
35 
36 		/* List of all BOs in the cache, ordered by age, so we
37 		 * can do O(1) lookups when trying to free old
38 		 * buffers.
39 		 */
40 		struct list_head time_list;
41 		struct work_struct time_work;
42 		struct timer_list time_timer;
43 	} bo_cache;
44 
45 	struct vc4_bo_stats {
46 		u32 num_allocated;
47 		u32 size_allocated;
48 		u32 num_cached;
49 		u32 size_cached;
50 	} bo_stats;
51 
52 	/* Protects bo_cache and the BO stats. */
53 	struct mutex bo_lock;
54 
55 	/* Sequence number for the last job queued in job_list.
56 	 * Starts at 0 (no jobs emitted).
57 	 */
58 	uint64_t emit_seqno;
59 
60 	/* Sequence number for the last completed job on the GPU.
61 	 * Starts at 0 (no jobs completed).
62 	 */
63 	uint64_t finished_seqno;
64 
65 	/* List of all struct vc4_exec_info for jobs to be executed.
66 	 * The first job in the list is the one currently programmed
67 	 * into ct0ca/ct1ca for execution.
68 	 */
69 	struct list_head job_list;
70 	/* List of the finished vc4_exec_infos waiting to be freed by
71 	 * job_done_work.
72 	 */
73 	struct list_head job_done_list;
74 	/* Spinlock used to synchronize the job_list and seqno
75 	 * accesses between the IRQ handler and GEM ioctls.
76 	 */
77 	spinlock_t job_lock;
78 	wait_queue_head_t job_wait_queue;
79 	struct work_struct job_done_work;
80 
81 	/* List of struct vc4_seqno_cb for callbacks to be made from a
82 	 * workqueue when the given seqno is passed.
83 	 */
84 	struct list_head seqno_cb_list;
85 
86 	/* The binner overflow memory that's currently set up in
87 	 * BPOA/BPOS registers.  When overflow occurs and a new one is
88 	 * allocated, the previous one will be moved to
89 	 * vc4->current_exec's free list.
90 	 */
91 	struct vc4_bo *overflow_mem;
92 	struct work_struct overflow_mem_work;
93 
94 	int power_refcount;
95 
96 	/* Mutex controlling the power refcount. */
97 	struct mutex power_lock;
98 
99 	struct {
100 		struct timer_list timer;
101 		struct work_struct reset_work;
102 	} hangcheck;
103 
104 	struct semaphore async_modeset;
105 };
106 
107 static inline struct vc4_dev *
108 to_vc4_dev(struct drm_device *dev)
109 {
110 	return (struct vc4_dev *)dev->dev_private;
111 }
112 
113 struct vc4_bo {
114 	struct drm_gem_cma_object base;
115 
116 	/* seqno of the last job to render to this BO. */
117 	uint64_t seqno;
118 
119 	/* List entry for the BO's position in either
120 	 * vc4_exec_info->unref_list or vc4_dev->bo_cache.time_list
121 	 */
122 	struct list_head unref_head;
123 
124 	/* Time in jiffies when the BO was put in vc4->bo_cache. */
125 	unsigned long free_time;
126 
127 	/* List entry for the BO's position in vc4_dev->bo_cache.size_list */
128 	struct list_head size_head;
129 
130 	/* Struct for shader validation state, if created by
131 	 * DRM_IOCTL_VC4_CREATE_SHADER_BO.
132 	 */
133 	struct vc4_validated_shader_info *validated_shader;
134 };
135 
136 static inline struct vc4_bo *
137 to_vc4_bo(struct drm_gem_object *bo)
138 {
139 	return (struct vc4_bo *)bo;
140 }
141 
142 struct vc4_seqno_cb {
143 	struct work_struct work;
144 	uint64_t seqno;
145 	void (*func)(struct vc4_seqno_cb *cb);
146 };
147 
148 struct vc4_v3d {
149 	struct vc4_dev *vc4;
150 	struct platform_device *pdev;
151 	void __iomem *regs;
152 };
153 
154 struct vc4_hvs {
155 	struct platform_device *pdev;
156 	void __iomem *regs;
157 	void __iomem *dlist;
158 };
159 
160 struct vc4_plane {
161 	struct drm_plane base;
162 };
163 
164 static inline struct vc4_plane *
165 to_vc4_plane(struct drm_plane *plane)
166 {
167 	return (struct vc4_plane *)plane;
168 }
169 
170 enum vc4_encoder_type {
171 	VC4_ENCODER_TYPE_HDMI,
172 	VC4_ENCODER_TYPE_VEC,
173 	VC4_ENCODER_TYPE_DSI0,
174 	VC4_ENCODER_TYPE_DSI1,
175 	VC4_ENCODER_TYPE_SMI,
176 	VC4_ENCODER_TYPE_DPI,
177 };
178 
179 struct vc4_encoder {
180 	struct drm_encoder base;
181 	enum vc4_encoder_type type;
182 	u32 clock_select;
183 };
184 
185 static inline struct vc4_encoder *
186 to_vc4_encoder(struct drm_encoder *encoder)
187 {
188 	return container_of(encoder, struct vc4_encoder, base);
189 }
190 
191 #define V3D_READ(offset) readl(vc4->v3d->regs + offset)
192 #define V3D_WRITE(offset, val) writel(val, vc4->v3d->regs + offset)
193 #define HVS_READ(offset) readl(vc4->hvs->regs + offset)
194 #define HVS_WRITE(offset, val) writel(val, vc4->hvs->regs + offset)
195 
196 struct vc4_exec_info {
197 	/* Sequence number for this bin/render job. */
198 	uint64_t seqno;
199 
200 	/* Last current addresses the hardware was processing when the
201 	 * hangcheck timer checked on us.
202 	 */
203 	uint32_t last_ct0ca, last_ct1ca;
204 
205 	/* Kernel-space copy of the ioctl arguments */
206 	struct drm_vc4_submit_cl *args;
207 
208 	/* This is the array of BOs that were looked up at the start of exec.
209 	 * Command validation will use indices into this array.
210 	 */
211 	struct drm_gem_cma_object **bo;
212 	uint32_t bo_count;
213 
214 	/* Pointers for our position in vc4->job_list */
215 	struct list_head head;
216 
217 	/* List of other BOs used in the job that need to be released
218 	 * once the job is complete.
219 	 */
220 	struct list_head unref_list;
221 
222 	/* Current unvalidated indices into @bo loaded by the non-hardware
223 	 * VC4_PACKET_GEM_HANDLES.
224 	 */
225 	uint32_t bo_index[2];
226 
227 	/* This is the BO where we store the validated command lists, shader
228 	 * records, and uniforms.
229 	 */
230 	struct drm_gem_cma_object *exec_bo;
231 
232 	/**
233 	 * This tracks the per-shader-record state (packet 64) that
234 	 * determines the length of the shader record and the offset
235 	 * it's expected to be found at.  It gets read in from the
236 	 * command lists.
237 	 */
238 	struct vc4_shader_state {
239 		uint32_t addr;
240 		/* Maximum vertex index referenced by any primitive using this
241 		 * shader state.
242 		 */
243 		uint32_t max_index;
244 	} *shader_state;
245 
246 	/** How many shader states the user declared they were using. */
247 	uint32_t shader_state_size;
248 	/** How many shader state records the validator has seen. */
249 	uint32_t shader_state_count;
250 
251 	bool found_tile_binning_mode_config_packet;
252 	bool found_start_tile_binning_packet;
253 	bool found_increment_semaphore_packet;
254 	bool found_flush;
255 	uint8_t bin_tiles_x, bin_tiles_y;
256 	struct drm_gem_cma_object *tile_bo;
257 	uint32_t tile_alloc_offset;
258 
259 	/**
260 	 * Computed addresses pointing into exec_bo where we start the
261 	 * bin thread (ct0) and render thread (ct1).
262 	 */
263 	uint32_t ct0ca, ct0ea;
264 	uint32_t ct1ca, ct1ea;
265 
266 	/* Pointer to the unvalidated bin CL (if present). */
267 	void *bin_u;
268 
269 	/* Pointers to the shader recs.  These paddr gets incremented as CL
270 	 * packets are relocated in validate_gl_shader_state, and the vaddrs
271 	 * (u and v) get incremented and size decremented as the shader recs
272 	 * themselves are validated.
273 	 */
274 	void *shader_rec_u;
275 	void *shader_rec_v;
276 	uint32_t shader_rec_p;
277 	uint32_t shader_rec_size;
278 
279 	/* Pointers to the uniform data.  These pointers are incremented, and
280 	 * size decremented, as each batch of uniforms is uploaded.
281 	 */
282 	void *uniforms_u;
283 	void *uniforms_v;
284 	uint32_t uniforms_p;
285 	uint32_t uniforms_size;
286 };
287 
288 static inline struct vc4_exec_info *
289 vc4_first_job(struct vc4_dev *vc4)
290 {
291 	if (list_empty(&vc4->job_list))
292 		return NULL;
293 	return list_first_entry(&vc4->job_list, struct vc4_exec_info, head);
294 }
295 
296 /**
297  * struct vc4_texture_sample_info - saves the offsets into the UBO for texture
298  * setup parameters.
299  *
300  * This will be used at draw time to relocate the reference to the texture
301  * contents in p0, and validate that the offset combined with
302  * width/height/stride/etc. from p1 and p2/p3 doesn't sample outside the BO.
303  * Note that the hardware treats unprovided config parameters as 0, so not all
304  * of them need to be set up for every texure sample, and we'll store ~0 as
305  * the offset to mark the unused ones.
306  *
307  * See the VC4 3D architecture guide page 41 ("Texture and Memory Lookup Unit
308  * Setup") for definitions of the texture parameters.
309  */
310 struct vc4_texture_sample_info {
311 	bool is_direct;
312 	uint32_t p_offset[4];
313 };
314 
315 /**
316  * struct vc4_validated_shader_info - information about validated shaders that
317  * needs to be used from command list validation.
318  *
319  * For a given shader, each time a shader state record references it, we need
320  * to verify that the shader doesn't read more uniforms than the shader state
321  * record's uniform BO pointer can provide, and we need to apply relocations
322  * and validate the shader state record's uniforms that define the texture
323  * samples.
324  */
325 struct vc4_validated_shader_info {
326 	uint32_t uniforms_size;
327 	uint32_t uniforms_src_size;
328 	uint32_t num_texture_samples;
329 	struct vc4_texture_sample_info *texture_samples;
330 };
331 
332 /**
333  * _wait_for - magic (register) wait macro
334  *
335  * Does the right thing for modeset paths when run under kdgb or similar atomic
336  * contexts. Note that it's important that we check the condition again after
337  * having timed out, since the timeout could be due to preemption or similar and
338  * we've never had a chance to check the condition before the timeout.
339  */
340 #define _wait_for(COND, MS, W) ({ \
341 	unsigned long timeout__ = jiffies + msecs_to_jiffies(MS) + 1;	\
342 	int ret__ = 0;							\
343 	while (!(COND)) {						\
344 		if (time_after(jiffies, timeout__)) {			\
345 			if (!(COND))					\
346 				ret__ = -ETIMEDOUT;			\
347 			break;						\
348 		}							\
349 		if (W && drm_can_sleep())  {				\
350 			msleep(W);					\
351 		} else {						\
352 			cpu_relax();					\
353 		}							\
354 	}								\
355 	ret__;								\
356 })
357 
358 #define wait_for(COND, MS) _wait_for(COND, MS, 1)
359 
360 /* vc4_bo.c */
361 struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size);
362 void vc4_free_object(struct drm_gem_object *gem_obj);
363 struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t size,
364 			     bool from_cache);
365 int vc4_dumb_create(struct drm_file *file_priv,
366 		    struct drm_device *dev,
367 		    struct drm_mode_create_dumb *args);
368 struct dma_buf *vc4_prime_export(struct drm_device *dev,
369 				 struct drm_gem_object *obj, int flags);
370 int vc4_create_bo_ioctl(struct drm_device *dev, void *data,
371 			struct drm_file *file_priv);
372 int vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data,
373 			       struct drm_file *file_priv);
374 int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
375 		      struct drm_file *file_priv);
376 int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
377 			     struct drm_file *file_priv);
378 int vc4_mmap(struct file *filp, struct vm_area_struct *vma);
379 int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma);
380 void *vc4_prime_vmap(struct drm_gem_object *obj);
381 void vc4_bo_cache_init(struct drm_device *dev);
382 void vc4_bo_cache_destroy(struct drm_device *dev);
383 int vc4_bo_stats_debugfs(struct seq_file *m, void *arg);
384 
385 /* vc4_crtc.c */
386 extern struct platform_driver vc4_crtc_driver;
387 int vc4_enable_vblank(struct drm_device *dev, unsigned int crtc_id);
388 void vc4_disable_vblank(struct drm_device *dev, unsigned int crtc_id);
389 void vc4_cancel_page_flip(struct drm_crtc *crtc, struct drm_file *file);
390 int vc4_crtc_debugfs_regs(struct seq_file *m, void *arg);
391 
392 /* vc4_debugfs.c */
393 int vc4_debugfs_init(struct drm_minor *minor);
394 void vc4_debugfs_cleanup(struct drm_minor *minor);
395 
396 /* vc4_drv.c */
397 void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index);
398 
399 /* vc4_gem.c */
400 void vc4_gem_init(struct drm_device *dev);
401 void vc4_gem_destroy(struct drm_device *dev);
402 int vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
403 			struct drm_file *file_priv);
404 int vc4_wait_seqno_ioctl(struct drm_device *dev, void *data,
405 			 struct drm_file *file_priv);
406 int vc4_wait_bo_ioctl(struct drm_device *dev, void *data,
407 		      struct drm_file *file_priv);
408 void vc4_submit_next_job(struct drm_device *dev);
409 int vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno,
410 		       uint64_t timeout_ns, bool interruptible);
411 void vc4_job_handle_completed(struct vc4_dev *vc4);
412 int vc4_queue_seqno_cb(struct drm_device *dev,
413 		       struct vc4_seqno_cb *cb, uint64_t seqno,
414 		       void (*func)(struct vc4_seqno_cb *cb));
415 
416 /* vc4_hdmi.c */
417 extern struct platform_driver vc4_hdmi_driver;
418 int vc4_hdmi_debugfs_regs(struct seq_file *m, void *unused);
419 
420 /* vc4_irq.c */
421 irqreturn_t vc4_irq(int irq, void *arg);
422 void vc4_irq_preinstall(struct drm_device *dev);
423 int vc4_irq_postinstall(struct drm_device *dev);
424 void vc4_irq_uninstall(struct drm_device *dev);
425 void vc4_irq_reset(struct drm_device *dev);
426 
427 /* vc4_hvs.c */
428 extern struct platform_driver vc4_hvs_driver;
429 void vc4_hvs_dump_state(struct drm_device *dev);
430 int vc4_hvs_debugfs_regs(struct seq_file *m, void *unused);
431 
432 /* vc4_kms.c */
433 int vc4_kms_load(struct drm_device *dev);
434 
435 /* vc4_plane.c */
436 struct drm_plane *vc4_plane_init(struct drm_device *dev,
437 				 enum drm_plane_type type);
438 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist);
439 u32 vc4_plane_dlist_size(struct drm_plane_state *state);
440 void vc4_plane_async_set_fb(struct drm_plane *plane,
441 			    struct drm_framebuffer *fb);
442 
443 /* vc4_v3d.c */
444 extern struct platform_driver vc4_v3d_driver;
445 int vc4_v3d_debugfs_ident(struct seq_file *m, void *unused);
446 int vc4_v3d_debugfs_regs(struct seq_file *m, void *unused);
447 
448 /* vc4_validate.c */
449 int
450 vc4_validate_bin_cl(struct drm_device *dev,
451 		    void *validated,
452 		    void *unvalidated,
453 		    struct vc4_exec_info *exec);
454 
455 int
456 vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec);
457 
458 struct drm_gem_cma_object *vc4_use_bo(struct vc4_exec_info *exec,
459 				      uint32_t hindex);
460 
461 int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec);
462 
463 bool vc4_check_tex_size(struct vc4_exec_info *exec,
464 			struct drm_gem_cma_object *fbo,
465 			uint32_t offset, uint8_t tiling_format,
466 			uint32_t width, uint32_t height, uint8_t cpp);
467 
468 /* vc4_validate_shader.c */
469 struct vc4_validated_shader_info *
470 vc4_validate_shader(struct drm_gem_cma_object *shader_obj);
471