xref: /linux/drivers/gpu/drm/v3d/v3d_perfmon.c (revision 4b99990cdf9560e8a071640baf19f312e6ae02f4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2021 Raspberry Pi
4  */
5 
6 #include "v3d_drv.h"
7 #include "v3d_regs.h"
8 
9 static const struct v3d_perf_counter_desc v3d_v42_performance_counters[] = {
10 	{"FEP", "FEP-valid-primitives-no-rendered-pixels", "[FEP] Valid primitives that result in no rendered pixels, for all rendered tiles"},
11 	{"FEP", "FEP-valid-primitives-rendered-pixels", "[FEP] Valid primitives for all rendered tiles (primitives may be counted in more than one tile)"},
12 	{"FEP", "FEP-clipped-quads", "[FEP] Early-Z/Near/Far clipped quads"},
13 	{"FEP", "FEP-valid-quads", "[FEP] Valid quads"},
14 	{"TLB", "TLB-quads-not-passing-stencil-test", "[TLB] Quads with no pixels passing the stencil test"},
15 	{"TLB", "TLB-quads-not-passing-z-and-stencil-test", "[TLB] Quads with no pixels passing the Z and stencil tests"},
16 	{"TLB", "TLB-quads-passing-z-and-stencil-test", "[TLB] Quads with any pixels passing the Z and stencil tests"},
17 	{"TLB", "TLB-quads-with-zero-coverage", "[TLB] Quads with all pixels having zero coverage"},
18 	{"TLB", "TLB-quads-with-non-zero-coverage", "[TLB] Quads with any pixels having non-zero coverage"},
19 	{"TLB", "TLB-quads-written-to-color-buffer", "[TLB] Quads with valid pixels written to colour buffer"},
20 	{"PTB", "PTB-primitives-discarded-outside-viewport", "[PTB] Primitives discarded by being outside the viewport"},
21 	{"PTB", "PTB-primitives-need-clipping", "[PTB] Primitives that need clipping"},
22 	{"PTB", "PTB-primitives-discarded-reversed", "[PTB] Primitives that are discarded because they are reversed"},
23 	{"QPU", "QPU-total-idle-clk-cycles", "[QPU] Total idle clock cycles for all QPUs"},
24 	{"QPU", "QPU-total-active-clk-cycles-vertex-coord-shading", "[QPU] Total active clock cycles for all QPUs doing vertex/coordinate/user shading (counts only when QPU is not stalled)"},
25 	{"QPU", "QPU-total-active-clk-cycles-fragment-shading", "[QPU] Total active clock cycles for all QPUs doing fragment shading (counts only when QPU is not stalled)"},
26 	{"QPU", "QPU-total-clk-cycles-executing-valid-instr", "[QPU] Total clock cycles for all QPUs executing valid instructions"},
27 	{"QPU", "QPU-total-clk-cycles-waiting-TMU", "[QPU] Total clock cycles for all QPUs stalled waiting for TMUs only (counter won't increment if QPU also stalling for another reason)"},
28 	{"QPU", "QPU-total-clk-cycles-waiting-scoreboard", "[QPU] Total clock cycles for all QPUs stalled waiting for Scoreboard only (counter won't increment if QPU also stalling for another reason)"},
29 	{"QPU", "QPU-total-clk-cycles-waiting-varyings", "[QPU] Total clock cycles for all QPUs stalled waiting for Varyings only (counter won't increment if QPU also stalling for another reason)"},
30 	{"QPU", "QPU-total-instr-cache-hit", "[QPU] Total instruction cache hits for all slices"},
31 	{"QPU", "QPU-total-instr-cache-miss", "[QPU] Total instruction cache misses for all slices"},
32 	{"QPU", "QPU-total-uniform-cache-hit", "[QPU] Total uniforms cache hits for all slices"},
33 	{"QPU", "QPU-total-uniform-cache-miss", "[QPU] Total uniforms cache misses for all slices"},
34 	{"TMU", "TMU-total-text-quads-access", "[TMU] Total texture cache accesses"},
35 	{"TMU", "TMU-total-text-cache-miss", "[TMU] Total texture cache misses (number of fetches from memory/L2cache)"},
36 	{"VPM", "VPM-total-clk-cycles-VDW-stalled", "[VPM] Total clock cycles VDW is stalled waiting for VPM access"},
37 	{"VPM", "VPM-total-clk-cycles-VCD-stalled", "[VPM] Total clock cycles VCD is stalled waiting for VPM access"},
38 	{"CLE", "CLE-bin-thread-active-cycles", "[CLE] Bin thread active cycles"},
39 	{"CLE", "CLE-render-thread-active-cycles", "[CLE] Render thread active cycles"},
40 	{"L2T", "L2T-total-cache-hit", "[L2T] Total Level 2 cache hits"},
41 	{"L2T", "L2T-total-cache-miss", "[L2T] Total Level 2 cache misses"},
42 	{"CORE", "cycle-count", "[CORE] Cycle counter"},
43 	{"QPU", "QPU-total-clk-cycles-waiting-vertex-coord-shading", "[QPU] Total stalled clock cycles for all QPUs doing vertex/coordinate/user shading"},
44 	{"QPU", "QPU-total-clk-cycles-waiting-fragment-shading", "[QPU] Total stalled clock cycles for all QPUs doing fragment shading"},
45 	{"PTB", "PTB-primitives-binned", "[PTB] Total primitives binned"},
46 	{"AXI", "AXI-writes-seen-watch-0", "[AXI] Writes seen by watch 0"},
47 	{"AXI", "AXI-reads-seen-watch-0", "[AXI] Reads seen by watch 0"},
48 	{"AXI", "AXI-writes-stalled-seen-watch-0", "[AXI] Write stalls seen by watch 0"},
49 	{"AXI", "AXI-reads-stalled-seen-watch-0", "[AXI] Read stalls seen by watch 0"},
50 	{"AXI", "AXI-write-bytes-seen-watch-0", "[AXI] Total bytes written seen by watch 0"},
51 	{"AXI", "AXI-read-bytes-seen-watch-0", "[AXI] Total bytes read seen by watch 0"},
52 	{"AXI", "AXI-writes-seen-watch-1", "[AXI] Writes seen by watch 1"},
53 	{"AXI", "AXI-reads-seen-watch-1", "[AXI] Reads seen by watch 1"},
54 	{"AXI", "AXI-writes-stalled-seen-watch-1", "[AXI] Write stalls seen by watch 1"},
55 	{"AXI", "AXI-reads-stalled-seen-watch-1", "[AXI] Read stalls seen by watch 1"},
56 	{"AXI", "AXI-write-bytes-seen-watch-1", "[AXI] Total bytes written seen by watch 1"},
57 	{"AXI", "AXI-read-bytes-seen-watch-1", "[AXI] Total bytes read seen by watch 1"},
58 	{"TLB", "TLB-partial-quads-written-to-color-buffer", "[TLB] Partial quads written to the colour buffer"},
59 	{"TMU", "TMU-total-config-access", "[TMU] Total config accesses"},
60 	{"L2T", "L2T-no-id-stalled", "[L2T] No ID stall"},
61 	{"L2T", "L2T-command-queue-stalled", "[L2T] Command queue full stall"},
62 	{"L2T", "L2T-TMU-writes", "[L2T] TMU write accesses"},
63 	{"TMU", "TMU-active-cycles", "[TMU] Active cycles"},
64 	{"TMU", "TMU-stalled-cycles", "[TMU] Stalled cycles"},
65 	{"CLE", "CLE-thread-active-cycles", "[CLE] Bin or render thread active cycles"},
66 	{"L2T", "L2T-TMU-reads", "[L2T] TMU read accesses"},
67 	{"L2T", "L2T-CLE-reads", "[L2T] CLE read accesses"},
68 	{"L2T", "L2T-VCD-reads", "[L2T] VCD read accesses"},
69 	{"L2T", "L2T-TMU-config-reads", "[L2T] TMU CFG read accesses"},
70 	{"L2T", "L2T-SLC0-reads", "[L2T] SLC0 read accesses"},
71 	{"L2T", "L2T-SLC1-reads", "[L2T] SLC1 read accesses"},
72 	{"L2T", "L2T-SLC2-reads", "[L2T] SLC2 read accesses"},
73 	{"L2T", "L2T-TMU-write-miss", "[L2T] TMU write misses"},
74 	{"L2T", "L2T-TMU-read-miss", "[L2T] TMU read misses"},
75 	{"L2T", "L2T-CLE-read-miss", "[L2T] CLE read misses"},
76 	{"L2T", "L2T-VCD-read-miss", "[L2T] VCD read misses"},
77 	{"L2T", "L2T-TMU-config-read-miss", "[L2T] TMU CFG read misses"},
78 	{"L2T", "L2T-SLC0-read-miss", "[L2T] SLC0 read misses"},
79 	{"L2T", "L2T-SLC1-read-miss", "[L2T] SLC1 read misses"},
80 	{"L2T", "L2T-SLC2-read-miss", "[L2T] SLC2 read misses"},
81 	{"CORE", "core-memory-writes", "[CORE] Total memory writes"},
82 	{"L2T", "L2T-memory-writes", "[L2T] Total memory writes"},
83 	{"PTB", "PTB-memory-writes", "[PTB] Total memory writes"},
84 	{"TLB", "TLB-memory-writes", "[TLB] Total memory writes"},
85 	{"CORE", "core-memory-reads", "[CORE] Total memory reads"},
86 	{"L2T", "L2T-memory-reads", "[L2T] Total memory reads"},
87 	{"PTB", "PTB-memory-reads", "[PTB] Total memory reads"},
88 	{"PSE", "PSE-memory-reads", "[PSE] Total memory reads"},
89 	{"TLB", "TLB-memory-reads", "[TLB] Total memory reads"},
90 	{"GMP", "GMP-memory-reads", "[GMP] Total memory reads"},
91 	{"PTB", "PTB-memory-words-writes", "[PTB] Total memory words written"},
92 	{"TLB", "TLB-memory-words-writes", "[TLB] Total memory words written"},
93 	{"PSE", "PSE-memory-words-reads", "[PSE] Total memory words read"},
94 	{"TLB", "TLB-memory-words-reads", "[TLB] Total memory words read"},
95 	{"TMU", "TMU-MRU-hits", "[TMU] Total MRU hits"},
96 	{"CORE", "compute-active-cycles", "[CORE] Compute active cycles"},
97 };
98 
99 static const struct v3d_perf_counter_desc v3d_v71_performance_counters[] = {
100 	{"CORE", "cycle-count", "[CORE] Cycle counter"},
101 	{"CORE", "core-active", "[CORE] Bin/Render/Compute active cycles"},
102 	{"CLE", "CLE-bin-thread-active-cycles", "[CLE] Bin thread active cycles"},
103 	{"CLE", "CLE-render-thread-active-cycles", "[CLE] Render thread active cycles"},
104 	{"CORE", "compute-active-cycles", "[CORE] Compute active cycles"},
105 	{"FEP", "FEP-valid-primitives-no-rendered-pixels", "[FEP] Valid primitives that result in no rendered pixels, for all rendered tiles"},
106 	{"FEP", "FEP-valid-primitives-rendered-pixels", "[FEP] Valid primitives for all rendered tiles (primitives may be counted in more than one tile)"},
107 	{"FEP", "FEP-clipped-quads", "[FEP] Early-Z/Near/Far clipped quads"},
108 	{"FEP", "FEP-valid-quads", "[FEP] Valid quads"},
109 	{"TLB", "TLB-quads-not-passing-stencil-test", "[TLB] Quads with no pixels passing the stencil test"},
110 	{"TLB", "TLB-quads-not-passing-z-and-stencil-test", "[TLB] Quads with no pixels passing the Z and stencil tests"},
111 	{"TLB", "TLB-quads-passing-z-and-stencil-test", "[TLB] Quads with any pixels passing the Z and stencil tests"},
112 	{"TLB", "TLB-quads-written-to-color-buffer", "[TLB] Quads with valid pixels written to colour buffer"},
113 	{"TLB", "TLB-partial-quads-written-to-color-buffer", "[TLB] Partial quads written to the colour buffer"},
114 	{"PTB", "PTB-primitives-need-clipping", "[PTB] Primitives that need clipping"},
115 	{"PTB", "PTB-primitives-discarded-outside-viewport", "[PTB] Primitives discarded by being outside the viewport"},
116 	{"PTB", "PTB-primitives-binned", "[PTB] Total primitives binned"},
117 	{"PTB", "PTB-primitives-discarded-reversed", "[PTB] Primitives that are discarded because they are reversed"},
118 	{"QPU", "QPU-total-instr-cache-hit", "[QPU] Total instruction cache hits for all slices"},
119 	{"QPU", "QPU-total-instr-cache-miss", "[QPU] Total instruction cache misses for all slices"},
120 	{"QPU", "QPU-total-uniform-cache-hit", "[QPU] Total uniforms cache hits for all slices"},
121 	{"QPU", "QPU-total-uniform-cache-miss", "[QPU] Total uniforms cache misses for all slices"},
122 	{"TMU", "TMU-active-cycles", "[TMU] Active cycles"},
123 	{"TMU", "TMU-stalled-cycles", "[TMU] Stalled cycles"},
124 	{"TMU", "TMU-total-text-quads-access", "[TMU] Total texture cache accesses"},
125 	{"TMU", "TMU-cache-x4-active-cycles", "[TMU] Cache active cycles for x4 access"},
126 	{"TMU", "TMU-cache-x4-stalled-cycles", "[TMU] Cache stalled cycles for x4 access"},
127 	{"TMU", "TMU-total-text-quads-x4-access", "[TMU] Total texture cache x4 access"},
128 	{"L2T", "L2T-total-cache-hit", "[L2T] Total Level 2 cache hits"},
129 	{"L2T", "L2T-total-cache-miss", "[L2T] Total Level 2 cache misses"},
130 	{"L2T", "L2T-local", "[L2T] Local mode access"},
131 	{"L2T", "L2T-writeback", "[L2T] Writeback"},
132 	{"L2T", "L2T-zero", "[L2T] Zero"},
133 	{"L2T", "L2T-merge", "[L2T] Merge"},
134 	{"L2T", "L2T-fill", "[L2T] Fill"},
135 	{"L2T", "L2T-stalls-no-wid", "[L2T] Stalls because no WID available"},
136 	{"L2T", "L2T-stalls-no-rid", "[L2T] Stalls because no RID available"},
137 	{"L2T", "L2T-stalls-queue-full", "[L2T] Stalls because internal queue full"},
138 	{"L2T", "L2T-stalls-wrightback", "[L2T] Stalls because writeback in flight"},
139 	{"L2T", "L2T-stalls-mem", "[L2T] Stalls because AXI blocks read"},
140 	{"L2T", "L2T-stalls-fill", "[L2T] Stalls because fill pending for victim cache-line"},
141 	{"L2T", "L2T-hitq", "[L2T] Sent request via hit queue"},
142 	{"L2T", "L2T-hitq-full", "[L2T] Sent request via main queue because hit queue is full"},
143 	{"L2T", "L2T-stalls-read-data", "[L2T] Stalls because waiting for data from SDRAM"},
144 	{"L2T", "L2T-TMU-read-hits", "[L2T] TMU read hits"},
145 	{"L2T", "L2T-TMU-read-miss", "[L2T] TMU read misses"},
146 	{"L2T", "L2T-VCD-read-hits", "[L2T] VCD read hits"},
147 	{"L2T", "L2T-VCD-read-miss", "[L2T] VCD read misses"},
148 	{"L2T", "L2T-SLC-read-hits", "[L2T] SLC read hits (all slices)"},
149 	{"L2T", "L2T-SLC-read-miss", "[L2T] SLC read misses (all slices)"},
150 	{"AXI", "AXI-writes-seen-watch-0", "[AXI] Writes seen by watch 0"},
151 	{"AXI", "AXI-reads-seen-watch-0", "[AXI] Reads seen by watch 0"},
152 	{"AXI", "AXI-writes-stalled-seen-watch-0", "[AXI] Write stalls seen by watch 0"},
153 	{"AXI", "AXI-reads-stalled-seen-watch-0", "[AXI] Read stalls seen by watch 0"},
154 	{"AXI", "AXI-write-bytes-seen-watch-0", "[AXI] Total bytes written seen by watch 0"},
155 	{"AXI", "AXI-read-bytes-seen-watch-0", "[AXI] Total bytes read seen by watch 0"},
156 	{"AXI", "AXI-writes-seen-watch-1", "[AXI] Writes seen by watch 1"},
157 	{"AXI", "AXI-reads-seen-watch-1", "[AXI] Reads seen by watch 1"},
158 	{"AXI", "AXI-writes-stalled-seen-watch-1", "[AXI] Write stalls seen by watch 1"},
159 	{"AXI", "AXI-reads-stalled-seen-watch-1", "[AXI] Read stalls seen by watch 1"},
160 	{"AXI", "AXI-write-bytes-seen-watch-1", "[AXI] Total bytes written seen by watch 1"},
161 	{"AXI", "AXI-read-bytes-seen-watch-1", "[AXI] Total bytes read seen by watch 1"},
162 	{"CORE", "core-memory-writes", "[CORE] Total memory writes"},
163 	{"L2T", "L2T-memory-writes", "[L2T] Total memory writes"},
164 	{"PTB", "PTB-memory-writes", "[PTB] Total memory writes"},
165 	{"TLB", "TLB-memory-writes", "[TLB] Total memory writes"},
166 	{"CORE", "core-memory-reads", "[CORE] Total memory reads"},
167 	{"L2T", "L2T-memory-reads", "[L2T] Total memory reads"},
168 	{"PTB", "PTB-memory-reads", "[PTB] Total memory reads"},
169 	{"PSE", "PSE-memory-reads", "[PSE] Total memory reads"},
170 	{"TLB", "TLB-memory-reads", "[TLB] Total memory reads"},
171 	{"PTB", "PTB-memory-words-writes", "[PTB] Total memory words written"},
172 	{"TLB", "TLB-memory-words-writes", "[TLB] Total memory words written"},
173 	{"PSE", "PSE-memory-words-reads", "[PSE] Total memory words read"},
174 	{"TLB", "TLB-memory-words-reads", "[TLB] Total memory words read"},
175 	{"AXI", "AXI-read-trans", "[AXI] Read transaction count"},
176 	{"AXI", "AXI-write-trans", "[AXI] Write transaction count"},
177 	{"AXI", "AXI-read-wait-cycles", "[AXI] Read total wait cycles"},
178 	{"AXI", "AXI-write-wait-cycles", "[AXI] Write total wait cycles"},
179 	{"AXI", "AXI-max-outstanding-reads", "[AXI] Maximum outstanding read transactions"},
180 	{"AXI", "AXI-max-outstanding-writes", "[AXI] Maximum outstanding write transactions"},
181 	{"QPU", "QPU-wait-bubble", "[QPU] Pipeline bubble in qcycles due all threads waiting"},
182 	{"QPU", "QPU-ic-miss-bubble", "[QPU] Pipeline bubble in qcycles due instruction-cache miss"},
183 	{"QPU", "QPU-active", "[QPU] Executed shader instruction"},
184 	{"QPU", "QPU-total-active-clk-cycles-fragment-shading", "[QPU] Total active clock cycles for all QPUs doing fragment shading (counts only when QPU is not stalled)"},
185 	{"QPU", "QPU-stalls", "[QPU] Stalled qcycles executing shader instruction"},
186 	{"QPU", "QPU-total-clk-cycles-waiting-fragment-shading", "[QPU] Total stalled clock cycles for all QPUs doing fragment shading"},
187 	{"QPU", "QPU-stalls-TMU", "[QPU] Stalled qcycles waiting for TMU"},
188 	{"QPU", "QPU-stalls-TLB", "[QPU] Stalled qcycles waiting for TLB"},
189 	{"QPU", "QPU-stalls-VPM", "[QPU] Stalled qcycles waiting for VPM"},
190 	{"QPU", "QPU-stalls-uniforms", "[QPU] Stalled qcycles waiting for uniforms"},
191 	{"QPU", "QPU-stalls-SFU", "[QPU] Stalled qcycles waiting for SFU"},
192 	{"QPU", "QPU-stalls-other", "[QPU] Stalled qcycles waiting for any other reason (vary/W/Z)"},
193 };
194 
195 void v3d_perfmon_init(struct v3d_dev *v3d)
196 {
197 	const struct v3d_perf_counter_desc *counters = NULL;
198 	unsigned int max = 0;
199 
200 	if (v3d->ver >= V3D_GEN_71) {
201 		counters = v3d_v71_performance_counters;
202 		max = ARRAY_SIZE(v3d_v71_performance_counters);
203 	} else if (v3d->ver >= V3D_GEN_42) {
204 		counters = v3d_v42_performance_counters;
205 		max = ARRAY_SIZE(v3d_v42_performance_counters);
206 	}
207 
208 	v3d->perfmon_info.max_counters = max;
209 	v3d->perfmon_info.counters = counters;
210 }
211 
212 void v3d_perfmon_get(struct v3d_perfmon *perfmon)
213 {
214 	if (perfmon)
215 		refcount_inc(&perfmon->refcnt);
216 }
217 
218 void v3d_perfmon_put(struct v3d_perfmon *perfmon)
219 {
220 	if (perfmon && refcount_dec_and_test(&perfmon->refcnt)) {
221 		mutex_destroy(&perfmon->lock);
222 		kfree(perfmon);
223 	}
224 }
225 
226 void v3d_perfmon_start(struct v3d_dev *v3d, struct v3d_perfmon *perfmon)
227 {
228 	unsigned int i;
229 	u32 mask;
230 	u8 ncounters;
231 
232 	if (WARN_ON_ONCE(!perfmon || v3d->active_perfmon))
233 		return;
234 
235 	if (!pm_runtime_get_if_active(v3d->drm.dev))
236 		return;
237 
238 	ncounters = perfmon->ncounters;
239 	mask = GENMASK(ncounters - 1, 0);
240 
241 	for (i = 0; i < ncounters; i++) {
242 		u32 source = i / 4;
243 		u32 channel = V3D_SET_FIELD_VER(perfmon->counters[i], V3D_PCTR_S0,
244 						v3d->ver);
245 
246 		i++;
247 		channel |= V3D_SET_FIELD_VER(i < ncounters ? perfmon->counters[i] : 0,
248 					     V3D_PCTR_S1, v3d->ver);
249 		i++;
250 		channel |= V3D_SET_FIELD_VER(i < ncounters ? perfmon->counters[i] : 0,
251 					     V3D_PCTR_S2, v3d->ver);
252 		i++;
253 		channel |= V3D_SET_FIELD_VER(i < ncounters ? perfmon->counters[i] : 0,
254 					     V3D_PCTR_S3, v3d->ver);
255 		V3D_CORE_WRITE(0, V3D_V4_PCTR_0_SRC_X(source), channel);
256 	}
257 
258 	V3D_CORE_WRITE(0, V3D_V4_PCTR_0_EN, mask);
259 	V3D_CORE_WRITE(0, V3D_V4_PCTR_0_CLR, mask);
260 	V3D_CORE_WRITE(0, V3D_PCTR_0_OVERFLOW, mask);
261 
262 	v3d->active_perfmon = perfmon;
263 
264 	v3d_pm_runtime_put(v3d);
265 }
266 
267 void v3d_perfmon_stop(struct v3d_dev *v3d, struct v3d_perfmon *perfmon,
268 		      bool capture)
269 {
270 	unsigned int i;
271 
272 	if (!perfmon || !v3d->active_perfmon)
273 		return;
274 
275 	mutex_lock(&perfmon->lock);
276 	if (perfmon != v3d->active_perfmon)
277 		goto out;
278 
279 	if (!pm_runtime_get_if_active(v3d->drm.dev))
280 		goto out_clear;
281 
282 	if (capture)
283 		for (i = 0; i < perfmon->ncounters; i++)
284 			perfmon->values[i] += V3D_CORE_READ(0, V3D_PCTR_0_PCTRX(i));
285 
286 	V3D_CORE_WRITE(0, V3D_V4_PCTR_0_EN, 0);
287 
288 	v3d_pm_runtime_put(v3d);
289 
290 out_clear:
291 	v3d->active_perfmon = NULL;
292 out:
293 	mutex_unlock(&perfmon->lock);
294 }
295 
296 struct v3d_perfmon *v3d_perfmon_find(struct v3d_file_priv *v3d_priv, int id)
297 {
298 	struct v3d_perfmon *perfmon;
299 
300 	xa_lock(&v3d_priv->perfmons);
301 	perfmon = xa_load(&v3d_priv->perfmons, id);
302 	v3d_perfmon_get(perfmon);
303 	xa_unlock(&v3d_priv->perfmons);
304 
305 	return perfmon;
306 }
307 
308 void v3d_perfmon_open_file(struct v3d_file_priv *v3d_priv)
309 {
310 	xa_init_flags(&v3d_priv->perfmons, XA_FLAGS_ALLOC1);
311 }
312 
313 static void v3d_perfmon_delete(struct v3d_file_priv *v3d_priv,
314 			       struct v3d_perfmon *perfmon)
315 {
316 	struct v3d_dev *v3d = v3d_priv->v3d;
317 
318 	/* If the active perfmon is being destroyed, stop it first */
319 	if (perfmon == v3d->active_perfmon)
320 		v3d_perfmon_stop(v3d, perfmon, false);
321 
322 	/* If the global perfmon is being destroyed, clean it and release
323 	 * the reference stashed in v3d_perfmon_set_global_ioctl().
324 	 */
325 	if (cmpxchg(&v3d->global_perfmon, perfmon, NULL) == perfmon)
326 		v3d_perfmon_put(perfmon);
327 
328 	v3d_perfmon_put(perfmon);
329 }
330 
331 void v3d_perfmon_close_file(struct v3d_file_priv *v3d_priv)
332 {
333 	struct v3d_perfmon *perfmon;
334 	unsigned long id;
335 
336 	xa_for_each(&v3d_priv->perfmons, id, perfmon)
337 		v3d_perfmon_delete(v3d_priv, perfmon);
338 
339 	xa_destroy(&v3d_priv->perfmons);
340 }
341 
342 int v3d_perfmon_create_ioctl(struct drm_device *dev, void *data,
343 			     struct drm_file *file_priv)
344 {
345 	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
346 	struct drm_v3d_perfmon_create *req = data;
347 	struct v3d_dev *v3d = v3d_priv->v3d;
348 	struct v3d_perfmon *perfmon;
349 	unsigned int i;
350 	int ret;
351 	u32 id;
352 
353 	/* Number of monitored counters cannot exceed HW limits. */
354 	if (req->ncounters > DRM_V3D_MAX_PERF_COUNTERS ||
355 	    !req->ncounters)
356 		return -EINVAL;
357 
358 	/* Make sure all counters are valid. */
359 	for (i = 0; i < req->ncounters; i++) {
360 		if (req->counters[i] >= v3d->perfmon_info.max_counters)
361 			return -EINVAL;
362 	}
363 
364 	perfmon = kzalloc_flex(*perfmon, values, req->ncounters);
365 	if (!perfmon)
366 		return -ENOMEM;
367 
368 	for (i = 0; i < req->ncounters; i++)
369 		perfmon->counters[i] = req->counters[i];
370 
371 	perfmon->ncounters = req->ncounters;
372 
373 	refcount_set(&perfmon->refcnt, 1);
374 	mutex_init(&perfmon->lock);
375 
376 	ret = xa_alloc(&v3d_priv->perfmons, &id, perfmon, xa_limit_32b,
377 		       GFP_KERNEL);
378 	if (ret < 0) {
379 		mutex_destroy(&perfmon->lock);
380 		kfree(perfmon);
381 		return ret;
382 	}
383 
384 	req->id = id;
385 
386 	return 0;
387 }
388 
389 int v3d_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
390 			      struct drm_file *file_priv)
391 {
392 	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
393 	struct drm_v3d_perfmon_destroy *req = data;
394 	struct v3d_perfmon *perfmon;
395 
396 	perfmon = xa_erase(&v3d_priv->perfmons, req->id);
397 	if (!perfmon)
398 		return -EINVAL;
399 
400 	v3d_perfmon_delete(v3d_priv, perfmon);
401 
402 	return 0;
403 }
404 
405 int v3d_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
406 				 struct drm_file *file_priv)
407 {
408 	struct v3d_dev *v3d = to_v3d_dev(dev);
409 	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
410 	struct drm_v3d_perfmon_get_values *req = data;
411 	struct v3d_perfmon *perfmon;
412 	int ret = 0;
413 
414 	if (req->pad != 0)
415 		return -EINVAL;
416 
417 	perfmon = v3d_perfmon_find(v3d_priv, req->id);
418 	if (!perfmon)
419 		return -EINVAL;
420 
421 	v3d_perfmon_stop(v3d, perfmon, true);
422 
423 	if (copy_to_user(u64_to_user_ptr(req->values_ptr), perfmon->values,
424 			 perfmon->ncounters * sizeof(u64)))
425 		ret = -EFAULT;
426 
427 	v3d_perfmon_put(perfmon);
428 
429 	return ret;
430 }
431 
432 int v3d_perfmon_get_counter_ioctl(struct drm_device *dev, void *data,
433 				  struct drm_file *file_priv)
434 {
435 	struct drm_v3d_perfmon_get_counter *req = data;
436 	struct v3d_dev *v3d = to_v3d_dev(dev);
437 	const struct v3d_perf_counter_desc *counter;
438 
439 	for (int i = 0; i < ARRAY_SIZE(req->reserved); i++) {
440 		if (req->reserved[i] != 0)
441 			return -EINVAL;
442 	}
443 
444 	if (!v3d->perfmon_info.max_counters)
445 		return -EOPNOTSUPP;
446 
447 	/* Make sure that the counter ID is valid */
448 	if (req->counter >= v3d->perfmon_info.max_counters)
449 		return -EINVAL;
450 
451 	counter = &v3d->perfmon_info.counters[req->counter];
452 
453 	strscpy(req->name, counter->name, sizeof(req->name));
454 	strscpy(req->category, counter->category, sizeof(req->category));
455 	strscpy(req->description, counter->description, sizeof(req->description));
456 
457 	return 0;
458 }
459 
460 int v3d_perfmon_set_global_ioctl(struct drm_device *dev, void *data,
461 				 struct drm_file *file_priv)
462 {
463 	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
464 	struct drm_v3d_perfmon_set_global *req = data;
465 	struct v3d_dev *v3d = to_v3d_dev(dev);
466 	struct v3d_perfmon *perfmon;
467 
468 	if (req->flags & ~DRM_V3D_PERFMON_CLEAR_GLOBAL)
469 		return -EINVAL;
470 
471 	perfmon = v3d_perfmon_find(v3d_priv, req->id);
472 	if (!perfmon)
473 		return -EINVAL;
474 
475 	/* If the request is to clear the global performance monitor */
476 	if (req->flags & DRM_V3D_PERFMON_CLEAR_GLOBAL) {
477 		struct v3d_perfmon *old;
478 
479 		/* DRM_V3D_PERFMON_CLEAR_GLOBAL doesn't check if
480 		 * v3d->global_perfmon == perfmon. Therefore, there
481 		 * is no need to keep perfmon's reference.
482 		 */
483 		v3d_perfmon_put(perfmon);
484 
485 		old = xchg(&v3d->global_perfmon, NULL);
486 		if (!old)
487 			return -EINVAL;
488 
489 		v3d_perfmon_put(old);
490 
491 		return 0;
492 	}
493 
494 	if (cmpxchg(&v3d->global_perfmon, NULL, perfmon)) {
495 		v3d_perfmon_put(perfmon);
496 		return -EBUSY;
497 	}
498 
499 	return 0;
500 }
501