Loading...
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
12struct 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 bin_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 in
66 * the binner. The first job in the list is the one currently
67 * programmed into ct0ca for execution.
68 */
69 struct list_head bin_job_list;
70
71 /* List of all struct vc4_exec_info for jobs that have
72 * completed binning and are ready for rendering. The first
73 * job in the list is the one currently programmed into ct1ca
74 * for execution.
75 */
76 struct list_head render_job_list;
77
78 /* List of the finished vc4_exec_infos waiting to be freed by
79 * job_done_work.
80 */
81 struct list_head job_done_list;
82 /* Spinlock used to synchronize the job_list and seqno
83 * accesses between the IRQ handler and GEM ioctls.
84 */
85 spinlock_t job_lock;
86 wait_queue_head_t job_wait_queue;
87 struct work_struct job_done_work;
88
89 /* List of struct vc4_seqno_cb for callbacks to be made from a
90 * workqueue when the given seqno is passed.
91 */
92 struct list_head seqno_cb_list;
93
94 /* The binner overflow memory that's currently set up in
95 * BPOA/BPOS registers. When overflow occurs and a new one is
96 * allocated, the previous one will be moved to
97 * vc4->current_exec's free list.
98 */
99 struct vc4_bo *overflow_mem;
100 struct work_struct overflow_mem_work;
101
102 int power_refcount;
103
104 /* Mutex controlling the power refcount. */
105 struct mutex power_lock;
106
107 struct {
108 struct timer_list timer;
109 struct work_struct reset_work;
110 } hangcheck;
111
112 struct semaphore async_modeset;
113};
114
115static inline struct vc4_dev *
116to_vc4_dev(struct drm_device *dev)
117{
118 return (struct vc4_dev *)dev->dev_private;
119}
120
121struct vc4_bo {
122 struct drm_gem_cma_object base;
123
124 /* seqno of the last job to render to this BO. */
125 uint64_t seqno;
126
127 /* List entry for the BO's position in either
128 * vc4_exec_info->unref_list or vc4_dev->bo_cache.time_list
129 */
130 struct list_head unref_head;
131
132 /* Time in jiffies when the BO was put in vc4->bo_cache. */
133 unsigned long free_time;
134
135 /* List entry for the BO's position in vc4_dev->bo_cache.size_list */
136 struct list_head size_head;
137
138 /* Struct for shader validation state, if created by
139 * DRM_IOCTL_VC4_CREATE_SHADER_BO.
140 */
141 struct vc4_validated_shader_info *validated_shader;
142};
143
144static inline struct vc4_bo *
145to_vc4_bo(struct drm_gem_object *bo)
146{
147 return (struct vc4_bo *)bo;
148}
149
150struct vc4_seqno_cb {
151 struct work_struct work;
152 uint64_t seqno;
153 void (*func)(struct vc4_seqno_cb *cb);
154};
155
156struct vc4_v3d {
157 struct vc4_dev *vc4;
158 struct platform_device *pdev;
159 void __iomem *regs;
160};
161
162struct vc4_hvs {
163 struct platform_device *pdev;
164 void __iomem *regs;
165 u32 __iomem *dlist;
166
167 /* Memory manager for CRTCs to allocate space in the display
168 * list. Units are dwords.
169 */
170 struct drm_mm dlist_mm;
171 /* Memory manager for the LBM memory used by HVS scaling. */
172 struct drm_mm lbm_mm;
173 spinlock_t mm_lock;
174
175 struct drm_mm_node mitchell_netravali_filter;
176};
177
178struct vc4_plane {
179 struct drm_plane base;
180};
181
182static inline struct vc4_plane *
183to_vc4_plane(struct drm_plane *plane)
184{
185 return (struct vc4_plane *)plane;
186}
187
188enum vc4_encoder_type {
189 VC4_ENCODER_TYPE_HDMI,
190 VC4_ENCODER_TYPE_VEC,
191 VC4_ENCODER_TYPE_DSI0,
192 VC4_ENCODER_TYPE_DSI1,
193 VC4_ENCODER_TYPE_SMI,
194 VC4_ENCODER_TYPE_DPI,
195};
196
197struct vc4_encoder {
198 struct drm_encoder base;
199 enum vc4_encoder_type type;
200 u32 clock_select;
201};
202
203static inline struct vc4_encoder *
204to_vc4_encoder(struct drm_encoder *encoder)
205{
206 return container_of(encoder, struct vc4_encoder, base);
207}
208
209#define V3D_READ(offset) readl(vc4->v3d->regs + offset)
210#define V3D_WRITE(offset, val) writel(val, vc4->v3d->regs + offset)
211#define HVS_READ(offset) readl(vc4->hvs->regs + offset)
212#define HVS_WRITE(offset, val) writel(val, vc4->hvs->regs + offset)
213
214struct vc4_exec_info {
215 /* Sequence number for this bin/render job. */
216 uint64_t seqno;
217
218 /* Last current addresses the hardware was processing when the
219 * hangcheck timer checked on us.
220 */
221 uint32_t last_ct0ca, last_ct1ca;
222
223 /* Kernel-space copy of the ioctl arguments */
224 struct drm_vc4_submit_cl *args;
225
226 /* This is the array of BOs that were looked up at the start of exec.
227 * Command validation will use indices into this array.
228 */
229 struct drm_gem_cma_object **bo;
230 uint32_t bo_count;
231
232 /* Pointers for our position in vc4->job_list */
233 struct list_head head;
234
235 /* List of other BOs used in the job that need to be released
236 * once the job is complete.
237 */
238 struct list_head unref_list;
239
240 /* Current unvalidated indices into @bo loaded by the non-hardware
241 * VC4_PACKET_GEM_HANDLES.
242 */
243 uint32_t bo_index[2];
244
245 /* This is the BO where we store the validated command lists, shader
246 * records, and uniforms.
247 */
248 struct drm_gem_cma_object *exec_bo;
249
250 /**
251 * This tracks the per-shader-record state (packet 64) that
252 * determines the length of the shader record and the offset
253 * it's expected to be found at. It gets read in from the
254 * command lists.
255 */
256 struct vc4_shader_state {
257 uint32_t addr;
258 /* Maximum vertex index referenced by any primitive using this
259 * shader state.
260 */
261 uint32_t max_index;
262 } *shader_state;
263
264 /** How many shader states the user declared they were using. */
265 uint32_t shader_state_size;
266 /** How many shader state records the validator has seen. */
267 uint32_t shader_state_count;
268
269 bool found_tile_binning_mode_config_packet;
270 bool found_start_tile_binning_packet;
271 bool found_increment_semaphore_packet;
272 bool found_flush;
273 uint8_t bin_tiles_x, bin_tiles_y;
274 struct drm_gem_cma_object *tile_bo;
275 uint32_t tile_alloc_offset;
276
277 /**
278 * Computed addresses pointing into exec_bo where we start the
279 * bin thread (ct0) and render thread (ct1).
280 */
281 uint32_t ct0ca, ct0ea;
282 uint32_t ct1ca, ct1ea;
283
284 /* Pointer to the unvalidated bin CL (if present). */
285 void *bin_u;
286
287 /* Pointers to the shader recs. These paddr gets incremented as CL
288 * packets are relocated in validate_gl_shader_state, and the vaddrs
289 * (u and v) get incremented and size decremented as the shader recs
290 * themselves are validated.
291 */
292 void *shader_rec_u;
293 void *shader_rec_v;
294 uint32_t shader_rec_p;
295 uint32_t shader_rec_size;
296
297 /* Pointers to the uniform data. These pointers are incremented, and
298 * size decremented, as each batch of uniforms is uploaded.
299 */
300 void *uniforms_u;
301 void *uniforms_v;
302 uint32_t uniforms_p;
303 uint32_t uniforms_size;
304};
305
306static inline struct vc4_exec_info *
307vc4_first_bin_job(struct vc4_dev *vc4)
308{
309 if (list_empty(&vc4->bin_job_list))
310 return NULL;
311 return list_first_entry(&vc4->bin_job_list, struct vc4_exec_info, head);
312}
313
314static inline struct vc4_exec_info *
315vc4_first_render_job(struct vc4_dev *vc4)
316{
317 if (list_empty(&vc4->render_job_list))
318 return NULL;
319 return list_first_entry(&vc4->render_job_list,
320 struct vc4_exec_info, head);
321}
322
323/**
324 * struct vc4_texture_sample_info - saves the offsets into the UBO for texture
325 * setup parameters.
326 *
327 * This will be used at draw time to relocate the reference to the texture
328 * contents in p0, and validate that the offset combined with
329 * width/height/stride/etc. from p1 and p2/p3 doesn't sample outside the BO.
330 * Note that the hardware treats unprovided config parameters as 0, so not all
331 * of them need to be set up for every texure sample, and we'll store ~0 as
332 * the offset to mark the unused ones.
333 *
334 * See the VC4 3D architecture guide page 41 ("Texture and Memory Lookup Unit
335 * Setup") for definitions of the texture parameters.
336 */
337struct vc4_texture_sample_info {
338 bool is_direct;
339 uint32_t p_offset[4];
340};
341
342/**
343 * struct vc4_validated_shader_info - information about validated shaders that
344 * needs to be used from command list validation.
345 *
346 * For a given shader, each time a shader state record references it, we need
347 * to verify that the shader doesn't read more uniforms than the shader state
348 * record's uniform BO pointer can provide, and we need to apply relocations
349 * and validate the shader state record's uniforms that define the texture
350 * samples.
351 */
352struct vc4_validated_shader_info {
353 uint32_t uniforms_size;
354 uint32_t uniforms_src_size;
355 uint32_t num_texture_samples;
356 struct vc4_texture_sample_info *texture_samples;
357};
358
359/**
360 * _wait_for - magic (register) wait macro
361 *
362 * Does the right thing for modeset paths when run under kdgb or similar atomic
363 * contexts. Note that it's important that we check the condition again after
364 * having timed out, since the timeout could be due to preemption or similar and
365 * we've never had a chance to check the condition before the timeout.
366 */
367#define _wait_for(COND, MS, W) ({ \
368 unsigned long timeout__ = jiffies + msecs_to_jiffies(MS) + 1; \
369 int ret__ = 0; \
370 while (!(COND)) { \
371 if (time_after(jiffies, timeout__)) { \
372 if (!(COND)) \
373 ret__ = -ETIMEDOUT; \
374 break; \
375 } \
376 if (W && drm_can_sleep()) { \
377 msleep(W); \
378 } else { \
379 cpu_relax(); \
380 } \
381 } \
382 ret__; \
383})
384
385#define wait_for(COND, MS) _wait_for(COND, MS, 1)
386
387/* vc4_bo.c */
388struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size);
389void vc4_free_object(struct drm_gem_object *gem_obj);
390struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t size,
391 bool from_cache);
392int vc4_dumb_create(struct drm_file *file_priv,
393 struct drm_device *dev,
394 struct drm_mode_create_dumb *args);
395struct dma_buf *vc4_prime_export(struct drm_device *dev,
396 struct drm_gem_object *obj, int flags);
397int vc4_create_bo_ioctl(struct drm_device *dev, void *data,
398 struct drm_file *file_priv);
399int vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data,
400 struct drm_file *file_priv);
401int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
402 struct drm_file *file_priv);
403int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
404 struct drm_file *file_priv);
405int vc4_mmap(struct file *filp, struct vm_area_struct *vma);
406int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma);
407void *vc4_prime_vmap(struct drm_gem_object *obj);
408void vc4_bo_cache_init(struct drm_device *dev);
409void vc4_bo_cache_destroy(struct drm_device *dev);
410int vc4_bo_stats_debugfs(struct seq_file *m, void *arg);
411
412/* vc4_crtc.c */
413extern struct platform_driver vc4_crtc_driver;
414int vc4_enable_vblank(struct drm_device *dev, unsigned int crtc_id);
415void vc4_disable_vblank(struct drm_device *dev, unsigned int crtc_id);
416int vc4_crtc_debugfs_regs(struct seq_file *m, void *arg);
417
418/* vc4_debugfs.c */
419int vc4_debugfs_init(struct drm_minor *minor);
420void vc4_debugfs_cleanup(struct drm_minor *minor);
421
422/* vc4_drv.c */
423void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index);
424
425/* vc4_gem.c */
426void vc4_gem_init(struct drm_device *dev);
427void vc4_gem_destroy(struct drm_device *dev);
428int vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
429 struct drm_file *file_priv);
430int vc4_wait_seqno_ioctl(struct drm_device *dev, void *data,
431 struct drm_file *file_priv);
432int vc4_wait_bo_ioctl(struct drm_device *dev, void *data,
433 struct drm_file *file_priv);
434void vc4_submit_next_bin_job(struct drm_device *dev);
435void vc4_submit_next_render_job(struct drm_device *dev);
436void vc4_move_job_to_render(struct drm_device *dev, struct vc4_exec_info *exec);
437int vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno,
438 uint64_t timeout_ns, bool interruptible);
439void vc4_job_handle_completed(struct vc4_dev *vc4);
440int vc4_queue_seqno_cb(struct drm_device *dev,
441 struct vc4_seqno_cb *cb, uint64_t seqno,
442 void (*func)(struct vc4_seqno_cb *cb));
443
444/* vc4_hdmi.c */
445extern struct platform_driver vc4_hdmi_driver;
446int vc4_hdmi_debugfs_regs(struct seq_file *m, void *unused);
447
448/* vc4_irq.c */
449irqreturn_t vc4_irq(int irq, void *arg);
450void vc4_irq_preinstall(struct drm_device *dev);
451int vc4_irq_postinstall(struct drm_device *dev);
452void vc4_irq_uninstall(struct drm_device *dev);
453void vc4_irq_reset(struct drm_device *dev);
454
455/* vc4_hvs.c */
456extern struct platform_driver vc4_hvs_driver;
457void vc4_hvs_dump_state(struct drm_device *dev);
458int vc4_hvs_debugfs_regs(struct seq_file *m, void *unused);
459
460/* vc4_kms.c */
461int vc4_kms_load(struct drm_device *dev);
462
463/* vc4_plane.c */
464struct drm_plane *vc4_plane_init(struct drm_device *dev,
465 enum drm_plane_type type);
466u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist);
467u32 vc4_plane_dlist_size(struct drm_plane_state *state);
468void vc4_plane_async_set_fb(struct drm_plane *plane,
469 struct drm_framebuffer *fb);
470
471/* vc4_v3d.c */
472extern struct platform_driver vc4_v3d_driver;
473int vc4_v3d_debugfs_ident(struct seq_file *m, void *unused);
474int vc4_v3d_debugfs_regs(struct seq_file *m, void *unused);
475
476/* vc4_validate.c */
477int
478vc4_validate_bin_cl(struct drm_device *dev,
479 void *validated,
480 void *unvalidated,
481 struct vc4_exec_info *exec);
482
483int
484vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec);
485
486struct drm_gem_cma_object *vc4_use_bo(struct vc4_exec_info *exec,
487 uint32_t hindex);
488
489int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec);
490
491bool vc4_check_tex_size(struct vc4_exec_info *exec,
492 struct drm_gem_cma_object *fbo,
493 uint32_t offset, uint8_t tiling_format,
494 uint32_t width, uint32_t height, uint8_t cpp);
495
496/* vc4_validate_shader.c */
497struct vc4_validated_shader_info *
498vc4_validate_shader(struct drm_gem_cma_object *shader_obj);
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * Copyright (C) 2015 Broadcom
4 */
5
6#include <linux/delay.h>
7#include <linux/refcount.h>
8#include <linux/uaccess.h>
9
10#include <drm/drm_atomic.h>
11#include <drm/drm_debugfs.h>
12#include <drm/drm_device.h>
13#include <drm/drm_encoder.h>
14#include <drm/drm_gem_cma_helper.h>
15#include <drm/drm_mm.h>
16#include <drm/drm_modeset_lock.h>
17
18#include "uapi/drm/vc4_drm.h"
19
20struct drm_device;
21struct drm_gem_object;
22
23/* Don't forget to update vc4_bo.c: bo_type_names[] when adding to
24 * this.
25 */
26enum vc4_kernel_bo_type {
27 /* Any kernel allocation (gem_create_object hook) before it
28 * gets another type set.
29 */
30 VC4_BO_TYPE_KERNEL,
31 VC4_BO_TYPE_V3D,
32 VC4_BO_TYPE_V3D_SHADER,
33 VC4_BO_TYPE_DUMB,
34 VC4_BO_TYPE_BIN,
35 VC4_BO_TYPE_RCL,
36 VC4_BO_TYPE_BCL,
37 VC4_BO_TYPE_KERNEL_CACHE,
38 VC4_BO_TYPE_COUNT
39};
40
41/* Performance monitor object. The perform lifetime is controlled by userspace
42 * using perfmon related ioctls. A perfmon can be attached to a submit_cl
43 * request, and when this is the case, HW perf counters will be activated just
44 * before the submit_cl is submitted to the GPU and disabled when the job is
45 * done. This way, only events related to a specific job will be counted.
46 */
47struct vc4_perfmon {
48 /* Tracks the number of users of the perfmon, when this counter reaches
49 * zero the perfmon is destroyed.
50 */
51 refcount_t refcnt;
52
53 /* Number of counters activated in this perfmon instance
54 * (should be less than DRM_VC4_MAX_PERF_COUNTERS).
55 */
56 u8 ncounters;
57
58 /* Events counted by the HW perf counters. */
59 u8 events[DRM_VC4_MAX_PERF_COUNTERS];
60
61 /* Storage for counter values. Counters are incremented by the HW
62 * perf counter values every time the perfmon is attached to a GPU job.
63 * This way, perfmon users don't have to retrieve the results after
64 * each job if they want to track events covering several submissions.
65 * Note that counter values can't be reset, but you can fake a reset by
66 * destroying the perfmon and creating a new one.
67 */
68 u64 counters[0];
69};
70
71struct vc4_dev {
72 struct drm_device *dev;
73
74 struct vc4_hdmi *hdmi;
75 struct vc4_hvs *hvs;
76 struct vc4_v3d *v3d;
77 struct vc4_dpi *dpi;
78 struct vc4_dsi *dsi1;
79 struct vc4_vec *vec;
80 struct vc4_txp *txp;
81
82 struct vc4_hang_state *hang_state;
83
84 /* The kernel-space BO cache. Tracks buffers that have been
85 * unreferenced by all other users (refcounts of 0!) but not
86 * yet freed, so we can do cheap allocations.
87 */
88 struct vc4_bo_cache {
89 /* Array of list heads for entries in the BO cache,
90 * based on number of pages, so we can do O(1) lookups
91 * in the cache when allocating.
92 */
93 struct list_head *size_list;
94 uint32_t size_list_size;
95
96 /* List of all BOs in the cache, ordered by age, so we
97 * can do O(1) lookups when trying to free old
98 * buffers.
99 */
100 struct list_head time_list;
101 struct work_struct time_work;
102 struct timer_list time_timer;
103 } bo_cache;
104
105 u32 num_labels;
106 struct vc4_label {
107 const char *name;
108 u32 num_allocated;
109 u32 size_allocated;
110 } *bo_labels;
111
112 /* Protects bo_cache and bo_labels. */
113 struct mutex bo_lock;
114
115 /* Purgeable BO pool. All BOs in this pool can have their memory
116 * reclaimed if the driver is unable to allocate new BOs. We also
117 * keep stats related to the purge mechanism here.
118 */
119 struct {
120 struct list_head list;
121 unsigned int num;
122 size_t size;
123 unsigned int purged_num;
124 size_t purged_size;
125 struct mutex lock;
126 } purgeable;
127
128 uint64_t dma_fence_context;
129
130 /* Sequence number for the last job queued in bin_job_list.
131 * Starts at 0 (no jobs emitted).
132 */
133 uint64_t emit_seqno;
134
135 /* Sequence number for the last completed job on the GPU.
136 * Starts at 0 (no jobs completed).
137 */
138 uint64_t finished_seqno;
139
140 /* List of all struct vc4_exec_info for jobs to be executed in
141 * the binner. The first job in the list is the one currently
142 * programmed into ct0ca for execution.
143 */
144 struct list_head bin_job_list;
145
146 /* List of all struct vc4_exec_info for jobs that have
147 * completed binning and are ready for rendering. The first
148 * job in the list is the one currently programmed into ct1ca
149 * for execution.
150 */
151 struct list_head render_job_list;
152
153 /* List of the finished vc4_exec_infos waiting to be freed by
154 * job_done_work.
155 */
156 struct list_head job_done_list;
157 /* Spinlock used to synchronize the job_list and seqno
158 * accesses between the IRQ handler and GEM ioctls.
159 */
160 spinlock_t job_lock;
161 wait_queue_head_t job_wait_queue;
162 struct work_struct job_done_work;
163
164 /* Used to track the active perfmon if any. Access to this field is
165 * protected by job_lock.
166 */
167 struct vc4_perfmon *active_perfmon;
168
169 /* List of struct vc4_seqno_cb for callbacks to be made from a
170 * workqueue when the given seqno is passed.
171 */
172 struct list_head seqno_cb_list;
173
174 /* The memory used for storing binner tile alloc, tile state,
175 * and overflow memory allocations. This is freed when V3D
176 * powers down.
177 */
178 struct vc4_bo *bin_bo;
179
180 /* Size of blocks allocated within bin_bo. */
181 uint32_t bin_alloc_size;
182
183 /* Bitmask of the bin_alloc_size chunks in bin_bo that are
184 * used.
185 */
186 uint32_t bin_alloc_used;
187
188 /* Bitmask of the current bin_alloc used for overflow memory. */
189 uint32_t bin_alloc_overflow;
190
191 /* Incremented when an underrun error happened after an atomic commit.
192 * This is particularly useful to detect when a specific modeset is too
193 * demanding in term of memory or HVS bandwidth which is hard to guess
194 * at atomic check time.
195 */
196 atomic_t underrun;
197
198 struct work_struct overflow_mem_work;
199
200 int power_refcount;
201
202 /* Set to true when the load tracker is active. */
203 bool load_tracker_enabled;
204
205 /* Mutex controlling the power refcount. */
206 struct mutex power_lock;
207
208 struct {
209 struct timer_list timer;
210 struct work_struct reset_work;
211 } hangcheck;
212
213 struct semaphore async_modeset;
214
215 struct drm_modeset_lock ctm_state_lock;
216 struct drm_private_obj ctm_manager;
217 struct drm_private_obj load_tracker;
218
219 /* List of vc4_debugfs_info_entry for adding to debugfs once
220 * the minor is available (after drm_dev_register()).
221 */
222 struct list_head debugfs_list;
223
224 /* Mutex for binner bo allocation. */
225 struct mutex bin_bo_lock;
226 /* Reference count for our binner bo. */
227 struct kref bin_bo_kref;
228};
229
230static inline struct vc4_dev *
231to_vc4_dev(struct drm_device *dev)
232{
233 return (struct vc4_dev *)dev->dev_private;
234}
235
236struct vc4_bo {
237 struct drm_gem_cma_object base;
238
239 /* seqno of the last job to render using this BO. */
240 uint64_t seqno;
241
242 /* seqno of the last job to use the RCL to write to this BO.
243 *
244 * Note that this doesn't include binner overflow memory
245 * writes.
246 */
247 uint64_t write_seqno;
248
249 bool t_format;
250
251 /* List entry for the BO's position in either
252 * vc4_exec_info->unref_list or vc4_dev->bo_cache.time_list
253 */
254 struct list_head unref_head;
255
256 /* Time in jiffies when the BO was put in vc4->bo_cache. */
257 unsigned long free_time;
258
259 /* List entry for the BO's position in vc4_dev->bo_cache.size_list */
260 struct list_head size_head;
261
262 /* Struct for shader validation state, if created by
263 * DRM_IOCTL_VC4_CREATE_SHADER_BO.
264 */
265 struct vc4_validated_shader_info *validated_shader;
266
267 /* One of enum vc4_kernel_bo_type, or VC4_BO_TYPE_COUNT + i
268 * for user-allocated labels.
269 */
270 int label;
271
272 /* Count the number of active users. This is needed to determine
273 * whether we can move the BO to the purgeable list or not (when the BO
274 * is used by the GPU or the display engine we can't purge it).
275 */
276 refcount_t usecnt;
277
278 /* Store purgeable/purged state here */
279 u32 madv;
280 struct mutex madv_lock;
281};
282
283static inline struct vc4_bo *
284to_vc4_bo(struct drm_gem_object *bo)
285{
286 return (struct vc4_bo *)bo;
287}
288
289struct vc4_fence {
290 struct dma_fence base;
291 struct drm_device *dev;
292 /* vc4 seqno for signaled() test */
293 uint64_t seqno;
294};
295
296static inline struct vc4_fence *
297to_vc4_fence(struct dma_fence *fence)
298{
299 return (struct vc4_fence *)fence;
300}
301
302struct vc4_seqno_cb {
303 struct work_struct work;
304 uint64_t seqno;
305 void (*func)(struct vc4_seqno_cb *cb);
306};
307
308struct vc4_v3d {
309 struct vc4_dev *vc4;
310 struct platform_device *pdev;
311 void __iomem *regs;
312 struct clk *clk;
313 struct debugfs_regset32 regset;
314};
315
316struct vc4_hvs {
317 struct platform_device *pdev;
318 void __iomem *regs;
319 u32 __iomem *dlist;
320
321 /* Memory manager for CRTCs to allocate space in the display
322 * list. Units are dwords.
323 */
324 struct drm_mm dlist_mm;
325 /* Memory manager for the LBM memory used by HVS scaling. */
326 struct drm_mm lbm_mm;
327 spinlock_t mm_lock;
328
329 struct drm_mm_node mitchell_netravali_filter;
330 struct debugfs_regset32 regset;
331};
332
333struct vc4_plane {
334 struct drm_plane base;
335};
336
337static inline struct vc4_plane *
338to_vc4_plane(struct drm_plane *plane)
339{
340 return (struct vc4_plane *)plane;
341}
342
343enum vc4_scaling_mode {
344 VC4_SCALING_NONE,
345 VC4_SCALING_TPZ,
346 VC4_SCALING_PPF,
347};
348
349struct vc4_plane_state {
350 struct drm_plane_state base;
351 /* System memory copy of the display list for this element, computed
352 * at atomic_check time.
353 */
354 u32 *dlist;
355 u32 dlist_size; /* Number of dwords allocated for the display list */
356 u32 dlist_count; /* Number of used dwords in the display list. */
357
358 /* Offset in the dlist to various words, for pageflip or
359 * cursor updates.
360 */
361 u32 pos0_offset;
362 u32 pos2_offset;
363 u32 ptr0_offset;
364 u32 lbm_offset;
365
366 /* Offset where the plane's dlist was last stored in the
367 * hardware at vc4_crtc_atomic_flush() time.
368 */
369 u32 __iomem *hw_dlist;
370
371 /* Clipped coordinates of the plane on the display. */
372 int crtc_x, crtc_y, crtc_w, crtc_h;
373 /* Clipped area being scanned from in the FB. */
374 u32 src_x, src_y;
375
376 u32 src_w[2], src_h[2];
377
378 /* Scaling selection for the RGB/Y plane and the Cb/Cr planes. */
379 enum vc4_scaling_mode x_scaling[2], y_scaling[2];
380 bool is_unity;
381 bool is_yuv;
382
383 /* Offset to start scanning out from the start of the plane's
384 * BO.
385 */
386 u32 offsets[3];
387
388 /* Our allocation in LBM for temporary storage during scaling. */
389 struct drm_mm_node lbm;
390
391 /* Set when the plane has per-pixel alpha content or does not cover
392 * the entire screen. This is a hint to the CRTC that it might need
393 * to enable background color fill.
394 */
395 bool needs_bg_fill;
396
397 /* Mark the dlist as initialized. Useful to avoid initializing it twice
398 * when async update is not possible.
399 */
400 bool dlist_initialized;
401
402 /* Load of this plane on the HVS block. The load is expressed in HVS
403 * cycles/sec.
404 */
405 u64 hvs_load;
406
407 /* Memory bandwidth needed for this plane. This is expressed in
408 * bytes/sec.
409 */
410 u64 membus_load;
411};
412
413static inline struct vc4_plane_state *
414to_vc4_plane_state(struct drm_plane_state *state)
415{
416 return (struct vc4_plane_state *)state;
417}
418
419enum vc4_encoder_type {
420 VC4_ENCODER_TYPE_NONE,
421 VC4_ENCODER_TYPE_HDMI,
422 VC4_ENCODER_TYPE_VEC,
423 VC4_ENCODER_TYPE_DSI0,
424 VC4_ENCODER_TYPE_DSI1,
425 VC4_ENCODER_TYPE_SMI,
426 VC4_ENCODER_TYPE_DPI,
427};
428
429struct vc4_encoder {
430 struct drm_encoder base;
431 enum vc4_encoder_type type;
432 u32 clock_select;
433};
434
435static inline struct vc4_encoder *
436to_vc4_encoder(struct drm_encoder *encoder)
437{
438 return container_of(encoder, struct vc4_encoder, base);
439}
440
441struct vc4_crtc_data {
442 /* Which channel of the HVS this pixelvalve sources from. */
443 int hvs_channel;
444
445 enum vc4_encoder_type encoder_types[4];
446 const char *debugfs_name;
447};
448
449struct vc4_crtc {
450 struct drm_crtc base;
451 struct platform_device *pdev;
452 const struct vc4_crtc_data *data;
453 void __iomem *regs;
454
455 /* Timestamp at start of vblank irq - unaffected by lock delays. */
456 ktime_t t_vblank;
457
458 /* Which HVS channel we're using for our CRTC. */
459 int channel;
460
461 u8 lut_r[256];
462 u8 lut_g[256];
463 u8 lut_b[256];
464 /* Size in pixels of the COB memory allocated to this CRTC. */
465 u32 cob_size;
466
467 struct drm_pending_vblank_event *event;
468
469 struct debugfs_regset32 regset;
470};
471
472static inline struct vc4_crtc *
473to_vc4_crtc(struct drm_crtc *crtc)
474{
475 return (struct vc4_crtc *)crtc;
476}
477
478#define V3D_READ(offset) readl(vc4->v3d->regs + offset)
479#define V3D_WRITE(offset, val) writel(val, vc4->v3d->regs + offset)
480#define HVS_READ(offset) readl(vc4->hvs->regs + offset)
481#define HVS_WRITE(offset, val) writel(val, vc4->hvs->regs + offset)
482
483#define VC4_REG32(reg) { .name = #reg, .offset = reg }
484
485struct vc4_exec_info {
486 /* Sequence number for this bin/render job. */
487 uint64_t seqno;
488
489 /* Latest write_seqno of any BO that binning depends on. */
490 uint64_t bin_dep_seqno;
491
492 struct dma_fence *fence;
493
494 /* Last current addresses the hardware was processing when the
495 * hangcheck timer checked on us.
496 */
497 uint32_t last_ct0ca, last_ct1ca;
498
499 /* Kernel-space copy of the ioctl arguments */
500 struct drm_vc4_submit_cl *args;
501
502 /* This is the array of BOs that were looked up at the start of exec.
503 * Command validation will use indices into this array.
504 */
505 struct drm_gem_cma_object **bo;
506 uint32_t bo_count;
507
508 /* List of BOs that are being written by the RCL. Other than
509 * the binner temporary storage, this is all the BOs written
510 * by the job.
511 */
512 struct drm_gem_cma_object *rcl_write_bo[4];
513 uint32_t rcl_write_bo_count;
514
515 /* Pointers for our position in vc4->job_list */
516 struct list_head head;
517
518 /* List of other BOs used in the job that need to be released
519 * once the job is complete.
520 */
521 struct list_head unref_list;
522
523 /* Current unvalidated indices into @bo loaded by the non-hardware
524 * VC4_PACKET_GEM_HANDLES.
525 */
526 uint32_t bo_index[2];
527
528 /* This is the BO where we store the validated command lists, shader
529 * records, and uniforms.
530 */
531 struct drm_gem_cma_object *exec_bo;
532
533 /**
534 * This tracks the per-shader-record state (packet 64) that
535 * determines the length of the shader record and the offset
536 * it's expected to be found at. It gets read in from the
537 * command lists.
538 */
539 struct vc4_shader_state {
540 uint32_t addr;
541 /* Maximum vertex index referenced by any primitive using this
542 * shader state.
543 */
544 uint32_t max_index;
545 } *shader_state;
546
547 /** How many shader states the user declared they were using. */
548 uint32_t shader_state_size;
549 /** How many shader state records the validator has seen. */
550 uint32_t shader_state_count;
551
552 bool found_tile_binning_mode_config_packet;
553 bool found_start_tile_binning_packet;
554 bool found_increment_semaphore_packet;
555 bool found_flush;
556 uint8_t bin_tiles_x, bin_tiles_y;
557 /* Physical address of the start of the tile alloc array
558 * (where each tile's binned CL will start)
559 */
560 uint32_t tile_alloc_offset;
561 /* Bitmask of which binner slots are freed when this job completes. */
562 uint32_t bin_slots;
563
564 /**
565 * Computed addresses pointing into exec_bo where we start the
566 * bin thread (ct0) and render thread (ct1).
567 */
568 uint32_t ct0ca, ct0ea;
569 uint32_t ct1ca, ct1ea;
570
571 /* Pointer to the unvalidated bin CL (if present). */
572 void *bin_u;
573
574 /* Pointers to the shader recs. These paddr gets incremented as CL
575 * packets are relocated in validate_gl_shader_state, and the vaddrs
576 * (u and v) get incremented and size decremented as the shader recs
577 * themselves are validated.
578 */
579 void *shader_rec_u;
580 void *shader_rec_v;
581 uint32_t shader_rec_p;
582 uint32_t shader_rec_size;
583
584 /* Pointers to the uniform data. These pointers are incremented, and
585 * size decremented, as each batch of uniforms is uploaded.
586 */
587 void *uniforms_u;
588 void *uniforms_v;
589 uint32_t uniforms_p;
590 uint32_t uniforms_size;
591
592 /* Pointer to a performance monitor object if the user requested it,
593 * NULL otherwise.
594 */
595 struct vc4_perfmon *perfmon;
596
597 /* Whether the exec has taken a reference to the binner BO, which should
598 * happen with a VC4_PACKET_TILE_BINNING_MODE_CONFIG packet.
599 */
600 bool bin_bo_used;
601};
602
603/* Per-open file private data. Any driver-specific resource that has to be
604 * released when the DRM file is closed should be placed here.
605 */
606struct vc4_file {
607 struct {
608 struct idr idr;
609 struct mutex lock;
610 } perfmon;
611
612 bool bin_bo_used;
613};
614
615static inline struct vc4_exec_info *
616vc4_first_bin_job(struct vc4_dev *vc4)
617{
618 return list_first_entry_or_null(&vc4->bin_job_list,
619 struct vc4_exec_info, head);
620}
621
622static inline struct vc4_exec_info *
623vc4_first_render_job(struct vc4_dev *vc4)
624{
625 return list_first_entry_or_null(&vc4->render_job_list,
626 struct vc4_exec_info, head);
627}
628
629static inline struct vc4_exec_info *
630vc4_last_render_job(struct vc4_dev *vc4)
631{
632 if (list_empty(&vc4->render_job_list))
633 return NULL;
634 return list_last_entry(&vc4->render_job_list,
635 struct vc4_exec_info, head);
636}
637
638/**
639 * struct vc4_texture_sample_info - saves the offsets into the UBO for texture
640 * setup parameters.
641 *
642 * This will be used at draw time to relocate the reference to the texture
643 * contents in p0, and validate that the offset combined with
644 * width/height/stride/etc. from p1 and p2/p3 doesn't sample outside the BO.
645 * Note that the hardware treats unprovided config parameters as 0, so not all
646 * of them need to be set up for every texure sample, and we'll store ~0 as
647 * the offset to mark the unused ones.
648 *
649 * See the VC4 3D architecture guide page 41 ("Texture and Memory Lookup Unit
650 * Setup") for definitions of the texture parameters.
651 */
652struct vc4_texture_sample_info {
653 bool is_direct;
654 uint32_t p_offset[4];
655};
656
657/**
658 * struct vc4_validated_shader_info - information about validated shaders that
659 * needs to be used from command list validation.
660 *
661 * For a given shader, each time a shader state record references it, we need
662 * to verify that the shader doesn't read more uniforms than the shader state
663 * record's uniform BO pointer can provide, and we need to apply relocations
664 * and validate the shader state record's uniforms that define the texture
665 * samples.
666 */
667struct vc4_validated_shader_info {
668 uint32_t uniforms_size;
669 uint32_t uniforms_src_size;
670 uint32_t num_texture_samples;
671 struct vc4_texture_sample_info *texture_samples;
672
673 uint32_t num_uniform_addr_offsets;
674 uint32_t *uniform_addr_offsets;
675
676 bool is_threaded;
677};
678
679/**
680 * _wait_for - magic (register) wait macro
681 *
682 * Does the right thing for modeset paths when run under kdgb or similar atomic
683 * contexts. Note that it's important that we check the condition again after
684 * having timed out, since the timeout could be due to preemption or similar and
685 * we've never had a chance to check the condition before the timeout.
686 */
687#define _wait_for(COND, MS, W) ({ \
688 unsigned long timeout__ = jiffies + msecs_to_jiffies(MS) + 1; \
689 int ret__ = 0; \
690 while (!(COND)) { \
691 if (time_after(jiffies, timeout__)) { \
692 if (!(COND)) \
693 ret__ = -ETIMEDOUT; \
694 break; \
695 } \
696 if (W && drm_can_sleep()) { \
697 msleep(W); \
698 } else { \
699 cpu_relax(); \
700 } \
701 } \
702 ret__; \
703})
704
705#define wait_for(COND, MS) _wait_for(COND, MS, 1)
706
707/* vc4_bo.c */
708struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size);
709void vc4_free_object(struct drm_gem_object *gem_obj);
710struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t size,
711 bool from_cache, enum vc4_kernel_bo_type type);
712int vc4_dumb_create(struct drm_file *file_priv,
713 struct drm_device *dev,
714 struct drm_mode_create_dumb *args);
715struct dma_buf *vc4_prime_export(struct drm_gem_object *obj, int flags);
716int vc4_create_bo_ioctl(struct drm_device *dev, void *data,
717 struct drm_file *file_priv);
718int vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data,
719 struct drm_file *file_priv);
720int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
721 struct drm_file *file_priv);
722int vc4_set_tiling_ioctl(struct drm_device *dev, void *data,
723 struct drm_file *file_priv);
724int vc4_get_tiling_ioctl(struct drm_device *dev, void *data,
725 struct drm_file *file_priv);
726int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
727 struct drm_file *file_priv);
728int vc4_label_bo_ioctl(struct drm_device *dev, void *data,
729 struct drm_file *file_priv);
730vm_fault_t vc4_fault(struct vm_fault *vmf);
731int vc4_mmap(struct file *filp, struct vm_area_struct *vma);
732int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma);
733struct drm_gem_object *vc4_prime_import_sg_table(struct drm_device *dev,
734 struct dma_buf_attachment *attach,
735 struct sg_table *sgt);
736void *vc4_prime_vmap(struct drm_gem_object *obj);
737int vc4_bo_cache_init(struct drm_device *dev);
738void vc4_bo_cache_destroy(struct drm_device *dev);
739int vc4_bo_inc_usecnt(struct vc4_bo *bo);
740void vc4_bo_dec_usecnt(struct vc4_bo *bo);
741void vc4_bo_add_to_purgeable_pool(struct vc4_bo *bo);
742void vc4_bo_remove_from_purgeable_pool(struct vc4_bo *bo);
743
744/* vc4_crtc.c */
745extern struct platform_driver vc4_crtc_driver;
746bool vc4_crtc_get_scanoutpos(struct drm_device *dev, unsigned int crtc_id,
747 bool in_vblank_irq, int *vpos, int *hpos,
748 ktime_t *stime, ktime_t *etime,
749 const struct drm_display_mode *mode);
750void vc4_crtc_handle_vblank(struct vc4_crtc *crtc);
751void vc4_crtc_txp_armed(struct drm_crtc_state *state);
752void vc4_crtc_get_margins(struct drm_crtc_state *state,
753 unsigned int *right, unsigned int *left,
754 unsigned int *top, unsigned int *bottom);
755
756/* vc4_debugfs.c */
757int vc4_debugfs_init(struct drm_minor *minor);
758#ifdef CONFIG_DEBUG_FS
759void vc4_debugfs_add_file(struct drm_device *drm,
760 const char *filename,
761 int (*show)(struct seq_file*, void*),
762 void *data);
763void vc4_debugfs_add_regset32(struct drm_device *drm,
764 const char *filename,
765 struct debugfs_regset32 *regset);
766#else
767static inline void vc4_debugfs_add_file(struct drm_device *drm,
768 const char *filename,
769 int (*show)(struct seq_file*, void*),
770 void *data)
771{
772}
773
774static inline void vc4_debugfs_add_regset32(struct drm_device *drm,
775 const char *filename,
776 struct debugfs_regset32 *regset)
777{
778}
779#endif
780
781/* vc4_drv.c */
782void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index);
783
784/* vc4_dpi.c */
785extern struct platform_driver vc4_dpi_driver;
786
787/* vc4_dsi.c */
788extern struct platform_driver vc4_dsi_driver;
789
790/* vc4_fence.c */
791extern const struct dma_fence_ops vc4_fence_ops;
792
793/* vc4_gem.c */
794void vc4_gem_init(struct drm_device *dev);
795void vc4_gem_destroy(struct drm_device *dev);
796int vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
797 struct drm_file *file_priv);
798int vc4_wait_seqno_ioctl(struct drm_device *dev, void *data,
799 struct drm_file *file_priv);
800int vc4_wait_bo_ioctl(struct drm_device *dev, void *data,
801 struct drm_file *file_priv);
802void vc4_submit_next_bin_job(struct drm_device *dev);
803void vc4_submit_next_render_job(struct drm_device *dev);
804void vc4_move_job_to_render(struct drm_device *dev, struct vc4_exec_info *exec);
805int vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno,
806 uint64_t timeout_ns, bool interruptible);
807void vc4_job_handle_completed(struct vc4_dev *vc4);
808int vc4_queue_seqno_cb(struct drm_device *dev,
809 struct vc4_seqno_cb *cb, uint64_t seqno,
810 void (*func)(struct vc4_seqno_cb *cb));
811int vc4_gem_madvise_ioctl(struct drm_device *dev, void *data,
812 struct drm_file *file_priv);
813
814/* vc4_hdmi.c */
815extern struct platform_driver vc4_hdmi_driver;
816
817/* vc4_vec.c */
818extern struct platform_driver vc4_vec_driver;
819
820/* vc4_txp.c */
821extern struct platform_driver vc4_txp_driver;
822
823/* vc4_irq.c */
824irqreturn_t vc4_irq(int irq, void *arg);
825void vc4_irq_preinstall(struct drm_device *dev);
826int vc4_irq_postinstall(struct drm_device *dev);
827void vc4_irq_uninstall(struct drm_device *dev);
828void vc4_irq_reset(struct drm_device *dev);
829
830/* vc4_hvs.c */
831extern struct platform_driver vc4_hvs_driver;
832void vc4_hvs_dump_state(struct drm_device *dev);
833void vc4_hvs_unmask_underrun(struct drm_device *dev, int channel);
834void vc4_hvs_mask_underrun(struct drm_device *dev, int channel);
835
836/* vc4_kms.c */
837int vc4_kms_load(struct drm_device *dev);
838
839/* vc4_plane.c */
840struct drm_plane *vc4_plane_init(struct drm_device *dev,
841 enum drm_plane_type type);
842u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist);
843u32 vc4_plane_dlist_size(const struct drm_plane_state *state);
844void vc4_plane_async_set_fb(struct drm_plane *plane,
845 struct drm_framebuffer *fb);
846
847/* vc4_v3d.c */
848extern struct platform_driver vc4_v3d_driver;
849extern const struct of_device_id vc4_v3d_dt_match[];
850int vc4_v3d_get_bin_slot(struct vc4_dev *vc4);
851int vc4_v3d_bin_bo_get(struct vc4_dev *vc4, bool *used);
852void vc4_v3d_bin_bo_put(struct vc4_dev *vc4);
853int vc4_v3d_pm_get(struct vc4_dev *vc4);
854void vc4_v3d_pm_put(struct vc4_dev *vc4);
855
856/* vc4_validate.c */
857int
858vc4_validate_bin_cl(struct drm_device *dev,
859 void *validated,
860 void *unvalidated,
861 struct vc4_exec_info *exec);
862
863int
864vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec);
865
866struct drm_gem_cma_object *vc4_use_bo(struct vc4_exec_info *exec,
867 uint32_t hindex);
868
869int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec);
870
871bool vc4_check_tex_size(struct vc4_exec_info *exec,
872 struct drm_gem_cma_object *fbo,
873 uint32_t offset, uint8_t tiling_format,
874 uint32_t width, uint32_t height, uint8_t cpp);
875
876/* vc4_validate_shader.c */
877struct vc4_validated_shader_info *
878vc4_validate_shader(struct drm_gem_cma_object *shader_obj);
879
880/* vc4_perfmon.c */
881void vc4_perfmon_get(struct vc4_perfmon *perfmon);
882void vc4_perfmon_put(struct vc4_perfmon *perfmon);
883void vc4_perfmon_start(struct vc4_dev *vc4, struct vc4_perfmon *perfmon);
884void vc4_perfmon_stop(struct vc4_dev *vc4, struct vc4_perfmon *perfmon,
885 bool capture);
886struct vc4_perfmon *vc4_perfmon_find(struct vc4_file *vc4file, int id);
887void vc4_perfmon_open_file(struct vc4_file *vc4file);
888void vc4_perfmon_close_file(struct vc4_file *vc4file);
889int vc4_perfmon_create_ioctl(struct drm_device *dev, void *data,
890 struct drm_file *file_priv);
891int vc4_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
892 struct drm_file *file_priv);
893int vc4_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
894 struct drm_file *file_priv);