xref: /linux/drivers/gpu/drm/i915/gt/intel_gt_types.h (revision c6fbb759)
1 /* SPDX-License-Identifier: MIT */
2 /*
3  * Copyright © 2019 Intel Corporation
4  */
5 
6 #ifndef __INTEL_GT_TYPES__
7 #define __INTEL_GT_TYPES__
8 
9 #include <linux/ktime.h>
10 #include <linux/list.h>
11 #include <linux/llist.h>
12 #include <linux/mutex.h>
13 #include <linux/notifier.h>
14 #include <linux/seqlock.h>
15 #include <linux/spinlock.h>
16 #include <linux/types.h>
17 #include <linux/workqueue.h>
18 
19 #include "uc/intel_uc.h"
20 #include "intel_gsc.h"
21 
22 #include "i915_vma.h"
23 #include "intel_engine_types.h"
24 #include "intel_gt_buffer_pool_types.h"
25 #include "intel_hwconfig.h"
26 #include "intel_llc_types.h"
27 #include "intel_reset_types.h"
28 #include "intel_rc6_types.h"
29 #include "intel_rps_types.h"
30 #include "intel_migrate_types.h"
31 #include "intel_wakeref.h"
32 #include "pxp/intel_pxp_types.h"
33 
34 struct drm_i915_private;
35 struct i915_ggtt;
36 struct intel_engine_cs;
37 struct intel_uncore;
38 
39 struct intel_mmio_range {
40 	u32 start;
41 	u32 end;
42 };
43 
44 /*
45  * The hardware has multiple kinds of multicast register ranges that need
46  * special register steering (and future platforms are expected to add
47  * additional types).
48  *
49  * During driver startup, we initialize the steering control register to
50  * direct reads to a slice/subslice that are valid for the 'subslice' class
51  * of multicast registers.  If another type of steering does not have any
52  * overlap in valid steering targets with 'subslice' style registers, we will
53  * need to explicitly re-steer reads of registers of the other type.
54  *
55  * Only the replication types that may need additional non-default steering
56  * are listed here.
57  */
58 enum intel_steering_type {
59 	L3BANK,
60 	MSLICE,
61 	LNCF,
62 
63 	/*
64 	 * On some platforms there are multiple types of MCR registers that
65 	 * will always return a non-terminated value at instance (0, 0).  We'll
66 	 * lump those all into a single category to keep things simple.
67 	 */
68 	INSTANCE0,
69 
70 	NUM_STEERING_TYPES
71 };
72 
73 enum intel_submission_method {
74 	INTEL_SUBMISSION_RING,
75 	INTEL_SUBMISSION_ELSP,
76 	INTEL_SUBMISSION_GUC,
77 };
78 
79 struct gt_defaults {
80 	u32 min_freq;
81 	u32 max_freq;
82 };
83 
84 enum intel_gt_type {
85 	GT_PRIMARY,
86 	GT_TILE,
87 	GT_MEDIA,
88 };
89 
90 struct intel_gt {
91 	struct drm_i915_private *i915;
92 	const char *name;
93 	enum intel_gt_type type;
94 
95 	struct intel_uncore *uncore;
96 	struct i915_ggtt *ggtt;
97 
98 	struct intel_uc uc;
99 	struct intel_gsc gsc;
100 
101 	struct {
102 		/* Serialize global tlb invalidations */
103 		struct mutex invalidate_lock;
104 
105 		/*
106 		 * Batch TLB invalidations
107 		 *
108 		 * After unbinding the PTE, we need to ensure the TLB
109 		 * are invalidated prior to releasing the physical pages.
110 		 * But we only need one such invalidation for all unbinds,
111 		 * so we track how many TLB invalidations have been
112 		 * performed since unbind the PTE and only emit an extra
113 		 * invalidate if no full barrier has been passed.
114 		 */
115 		seqcount_mutex_t seqno;
116 	} tlb;
117 
118 	struct i915_wa_list wa_list;
119 
120 	struct intel_gt_timelines {
121 		spinlock_t lock; /* protects active_list */
122 		struct list_head active_list;
123 	} timelines;
124 
125 	struct intel_gt_requests {
126 		/**
127 		 * We leave the user IRQ off as much as possible,
128 		 * but this means that requests will finish and never
129 		 * be retired once the system goes idle. Set a timer to
130 		 * fire periodically while the ring is running. When it
131 		 * fires, go retire requests.
132 		 */
133 		struct delayed_work retire_work;
134 	} requests;
135 
136 	struct {
137 		struct llist_head list;
138 		struct work_struct work;
139 	} watchdog;
140 
141 	struct intel_wakeref wakeref;
142 	atomic_t user_wakeref;
143 
144 	/**
145 	 *  Protects access to lmem usefault list.
146 	 *  It is required, if we are outside of the runtime suspend path,
147 	 *  access to @lmem_userfault_list requires always first grabbing the
148 	 *  runtime pm, to ensure we can't race against runtime suspend.
149 	 *  Once we have that we also need to grab @lmem_userfault_lock,
150 	 *  at which point we have exclusive access.
151 	 *  The runtime suspend path is special since it doesn't really hold any locks,
152 	 *  but instead has exclusive access by virtue of all other accesses requiring
153 	 *  holding the runtime pm wakeref.
154 	 */
155 	struct mutex lmem_userfault_lock;
156 	struct list_head lmem_userfault_list;
157 
158 	struct list_head closed_vma;
159 	spinlock_t closed_lock; /* guards the list of closed_vma */
160 
161 	ktime_t last_init_time;
162 	struct intel_reset reset;
163 
164 	/**
165 	 * Is the GPU currently considered idle, or busy executing
166 	 * userspace requests? Whilst idle, we allow runtime power
167 	 * management to power down the hardware and display clocks.
168 	 * In order to reduce the effect on performance, there
169 	 * is a slight delay before we do so.
170 	 */
171 	intel_wakeref_t awake;
172 
173 	/* Manual runtime pm autosuspend delay for user GGTT/lmem mmaps */
174 	struct intel_wakeref_auto userfault_wakeref;
175 
176 	u32 clock_frequency;
177 	u32 clock_period_ns;
178 
179 	struct intel_llc llc;
180 	struct intel_rc6 rc6;
181 	struct intel_rps rps;
182 
183 	spinlock_t *irq_lock;
184 	u32 gt_imr;
185 	u32 pm_ier;
186 	u32 pm_imr;
187 
188 	u32 pm_guc_events;
189 
190 	struct {
191 		bool active;
192 
193 		/**
194 		 * @lock: Lock protecting the below fields.
195 		 */
196 		seqcount_mutex_t lock;
197 
198 		/**
199 		 * @total: Total time this engine was busy.
200 		 *
201 		 * Accumulated time not counting the most recent block in cases
202 		 * where engine is currently busy (active > 0).
203 		 */
204 		ktime_t total;
205 
206 		/**
207 		 * @start: Timestamp of the last idle to active transition.
208 		 *
209 		 * Idle is defined as active == 0, active is active > 0.
210 		 */
211 		ktime_t start;
212 	} stats;
213 
214 	struct intel_engine_cs *engine[I915_NUM_ENGINES];
215 	struct intel_engine_cs *engine_class[MAX_ENGINE_CLASS + 1]
216 					    [MAX_ENGINE_INSTANCE + 1];
217 	enum intel_submission_method submission_method;
218 
219 	/*
220 	 * Default address space (either GGTT or ppGTT depending on arch).
221 	 *
222 	 * Reserved for exclusive use by the kernel.
223 	 */
224 	struct i915_address_space *vm;
225 
226 	/*
227 	 * A pool of objects to use as shadow copies of client batch buffers
228 	 * when the command parser is enabled. Prevents the client from
229 	 * modifying the batch contents after software parsing.
230 	 *
231 	 * Buffers older than 1s are periodically reaped from the pool,
232 	 * or may be reclaimed by the shrinker before then.
233 	 */
234 	struct intel_gt_buffer_pool buffer_pool;
235 
236 	struct i915_vma *scratch;
237 
238 	struct intel_migrate migrate;
239 
240 	const struct intel_mmio_range *steering_table[NUM_STEERING_TYPES];
241 
242 	struct {
243 		u8 groupid;
244 		u8 instanceid;
245 	} default_steering;
246 
247 	/*
248 	 * Base of per-tile GTTMMADR where we can derive the MMIO and the GGTT.
249 	 */
250 	phys_addr_t phys_addr;
251 
252 	struct intel_gt_info {
253 		unsigned int id;
254 
255 		intel_engine_mask_t engine_mask;
256 
257 		u32 l3bank_mask;
258 
259 		u8 num_engines;
260 
261 		/* General presence of SFC units */
262 		u8 sfc_mask;
263 
264 		/* Media engine access to SFC per instance */
265 		u8 vdbox_sfc_access;
266 
267 		/* Slice/subslice/EU info */
268 		struct sseu_dev_info sseu;
269 
270 		unsigned long mslice_mask;
271 
272 		/** @hwconfig: hardware configuration data */
273 		struct intel_hwconfig hwconfig;
274 	} info;
275 
276 	struct {
277 		u8 uc_index;
278 		u8 wb_index; /* Only used on HAS_L3_CCS_READ() platforms */
279 	} mocs;
280 
281 	struct intel_pxp pxp;
282 
283 	/* gt/gtN sysfs */
284 	struct kobject sysfs_gt;
285 
286 	/* sysfs defaults per gt */
287 	struct gt_defaults defaults;
288 	struct kobject *sysfs_defaults;
289 };
290 
291 struct intel_gt_definition {
292 	enum intel_gt_type type;
293 	char *name;
294 	u32 mapping_base;
295 	u32 gsi_offset;
296 	intel_engine_mask_t engine_mask;
297 };
298 
299 enum intel_gt_scratch_field {
300 	/* 8 bytes */
301 	INTEL_GT_SCRATCH_FIELD_DEFAULT = 0,
302 
303 	/* 8 bytes */
304 	INTEL_GT_SCRATCH_FIELD_RENDER_FLUSH = 128,
305 
306 	/* 8 bytes */
307 	INTEL_GT_SCRATCH_FIELD_COHERENTL3_WA = 256,
308 
309 	/* 6 * 8 bytes */
310 	INTEL_GT_SCRATCH_FIELD_PERF_CS_GPR = 2048,
311 
312 	/* 4 bytes */
313 	INTEL_GT_SCRATCH_FIELD_PERF_PREDICATE_RESULT_1 = 2096,
314 };
315 
316 #endif /* __INTEL_GT_TYPES_H__ */
317