1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2018 The Linux Foundation. All rights reserved. */
3 
4 #include <linux/ascii85.h>
5 #include "msm_gem.h"
6 #include "a6xx_gpu.h"
7 #include "a6xx_gmu.h"
8 #include "a6xx_gpu_state.h"
9 #include "a6xx_gmu.xml.h"
10 
11 struct a6xx_gpu_state_obj {
12 	const void *handle;
13 	u32 *data;
14 };
15 
16 struct a6xx_gpu_state {
17 	struct msm_gpu_state base;
18 
19 	struct a6xx_gpu_state_obj *gmu_registers;
20 	int nr_gmu_registers;
21 
22 	struct a6xx_gpu_state_obj *registers;
23 	int nr_registers;
24 
25 	struct a6xx_gpu_state_obj *shaders;
26 	int nr_shaders;
27 
28 	struct a6xx_gpu_state_obj *clusters;
29 	int nr_clusters;
30 
31 	struct a6xx_gpu_state_obj *dbgahb_clusters;
32 	int nr_dbgahb_clusters;
33 
34 	struct a6xx_gpu_state_obj *indexed_regs;
35 	int nr_indexed_regs;
36 
37 	struct a6xx_gpu_state_obj *debugbus;
38 	int nr_debugbus;
39 
40 	struct a6xx_gpu_state_obj *vbif_debugbus;
41 
42 	struct a6xx_gpu_state_obj *cx_debugbus;
43 	int nr_cx_debugbus;
44 
45 	struct list_head objs;
46 };
47 
48 static inline int CRASHDUMP_WRITE(u64 *in, u32 reg, u32 val)
49 {
50 	in[0] = val;
51 	in[1] = (((u64) reg) << 44 | (1 << 21) | 1);
52 
53 	return 2;
54 }
55 
56 static inline int CRASHDUMP_READ(u64 *in, u32 reg, u32 dwords, u64 target)
57 {
58 	in[0] = target;
59 	in[1] = (((u64) reg) << 44 | dwords);
60 
61 	return 2;
62 }
63 
64 static inline int CRASHDUMP_FINI(u64 *in)
65 {
66 	in[0] = 0;
67 	in[1] = 0;
68 
69 	return 2;
70 }
71 
72 struct a6xx_crashdumper {
73 	void *ptr;
74 	struct drm_gem_object *bo;
75 	u64 iova;
76 };
77 
78 struct a6xx_state_memobj {
79 	struct list_head node;
80 	unsigned long long data[];
81 };
82 
83 void *state_kcalloc(struct a6xx_gpu_state *a6xx_state, int nr, size_t objsize)
84 {
85 	struct a6xx_state_memobj *obj =
86 		kzalloc((nr * objsize) + sizeof(*obj), GFP_KERNEL);
87 
88 	if (!obj)
89 		return NULL;
90 
91 	list_add_tail(&obj->node, &a6xx_state->objs);
92 	return &obj->data;
93 }
94 
95 void *state_kmemdup(struct a6xx_gpu_state *a6xx_state, void *src,
96 		size_t size)
97 {
98 	void *dst = state_kcalloc(a6xx_state, 1, size);
99 
100 	if (dst)
101 		memcpy(dst, src, size);
102 	return dst;
103 }
104 
105 /*
106  * Allocate 1MB for the crashdumper scratch region - 8k for the script and
107  * the rest for the data
108  */
109 #define A6XX_CD_DATA_OFFSET 8192
110 #define A6XX_CD_DATA_SIZE  (SZ_1M - 8192)
111 
112 static int a6xx_crashdumper_init(struct msm_gpu *gpu,
113 		struct a6xx_crashdumper *dumper)
114 {
115 	dumper->ptr = msm_gem_kernel_new_locked(gpu->dev,
116 		SZ_1M, MSM_BO_UNCACHED, gpu->aspace,
117 		&dumper->bo, &dumper->iova);
118 
119 	if (!IS_ERR(dumper->ptr))
120 		msm_gem_object_set_name(dumper->bo, "crashdump");
121 
122 	return PTR_ERR_OR_ZERO(dumper->ptr);
123 }
124 
125 static int a6xx_crashdumper_run(struct msm_gpu *gpu,
126 		struct a6xx_crashdumper *dumper)
127 {
128 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
129 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
130 	u32 val;
131 	int ret;
132 
133 	if (IS_ERR_OR_NULL(dumper->ptr))
134 		return -EINVAL;
135 
136 	if (!a6xx_gmu_sptprac_is_on(&a6xx_gpu->gmu))
137 		return -EINVAL;
138 
139 	/* Make sure all pending memory writes are posted */
140 	wmb();
141 
142 	gpu_write64(gpu, REG_A6XX_CP_CRASH_SCRIPT_BASE_LO,
143 		REG_A6XX_CP_CRASH_SCRIPT_BASE_HI, dumper->iova);
144 
145 	gpu_write(gpu, REG_A6XX_CP_CRASH_DUMP_CNTL, 1);
146 
147 	ret = gpu_poll_timeout(gpu, REG_A6XX_CP_CRASH_DUMP_STATUS, val,
148 		val & 0x02, 100, 10000);
149 
150 	gpu_write(gpu, REG_A6XX_CP_CRASH_DUMP_CNTL, 0);
151 
152 	return ret;
153 }
154 
155 /* read a value from the GX debug bus */
156 static int debugbus_read(struct msm_gpu *gpu, u32 block, u32 offset,
157 		u32 *data)
158 {
159 	u32 reg = A6XX_DBGC_CFG_DBGBUS_SEL_D_PING_INDEX(offset) |
160 		A6XX_DBGC_CFG_DBGBUS_SEL_D_PING_BLK_SEL(block);
161 
162 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_SEL_A, reg);
163 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_SEL_B, reg);
164 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_SEL_C, reg);
165 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_SEL_D, reg);
166 
167 	/* Wait 1 us to make sure the data is flowing */
168 	udelay(1);
169 
170 	data[0] = gpu_read(gpu, REG_A6XX_DBGC_CFG_DBGBUS_TRACE_BUF2);
171 	data[1] = gpu_read(gpu, REG_A6XX_DBGC_CFG_DBGBUS_TRACE_BUF1);
172 
173 	return 2;
174 }
175 
176 #define cxdbg_write(ptr, offset, val) \
177 	msm_writel((val), (ptr) + ((offset) << 2))
178 
179 #define cxdbg_read(ptr, offset) \
180 	msm_readl((ptr) + ((offset) << 2))
181 
182 /* read a value from the CX debug bus */
183 static int cx_debugbus_read(void *__iomem cxdbg, u32 block, u32 offset,
184 		u32 *data)
185 {
186 	u32 reg = A6XX_CX_DBGC_CFG_DBGBUS_SEL_A_PING_INDEX(offset) |
187 		A6XX_CX_DBGC_CFG_DBGBUS_SEL_A_PING_BLK_SEL(block);
188 
189 	cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_SEL_A, reg);
190 	cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_SEL_B, reg);
191 	cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_SEL_C, reg);
192 	cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_SEL_D, reg);
193 
194 	/* Wait 1 us to make sure the data is flowing */
195 	udelay(1);
196 
197 	data[0] = cxdbg_read(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_TRACE_BUF2);
198 	data[1] = cxdbg_read(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_TRACE_BUF1);
199 
200 	return 2;
201 }
202 
203 /* Read a chunk of data from the VBIF debug bus */
204 static int vbif_debugbus_read(struct msm_gpu *gpu, u32 ctrl0, u32 ctrl1,
205 		u32 reg, int count, u32 *data)
206 {
207 	int i;
208 
209 	gpu_write(gpu, ctrl0, reg);
210 
211 	for (i = 0; i < count; i++) {
212 		gpu_write(gpu, ctrl1, i);
213 		data[i] = gpu_read(gpu, REG_A6XX_VBIF_TEST_BUS_OUT);
214 	}
215 
216 	return count;
217 }
218 
219 #define AXI_ARB_BLOCKS 2
220 #define XIN_AXI_BLOCKS 5
221 #define XIN_CORE_BLOCKS 4
222 
223 #define VBIF_DEBUGBUS_BLOCK_SIZE \
224 	((16 * AXI_ARB_BLOCKS) + \
225 	 (18 * XIN_AXI_BLOCKS) + \
226 	 (12 * XIN_CORE_BLOCKS))
227 
228 static void a6xx_get_vbif_debugbus_block(struct msm_gpu *gpu,
229 		struct a6xx_gpu_state *a6xx_state,
230 		struct a6xx_gpu_state_obj *obj)
231 {
232 	u32 clk, *ptr;
233 	int i;
234 
235 	obj->data = state_kcalloc(a6xx_state, VBIF_DEBUGBUS_BLOCK_SIZE,
236 		sizeof(u32));
237 	if (!obj->data)
238 		return;
239 
240 	obj->handle = NULL;
241 
242 	/* Get the current clock setting */
243 	clk = gpu_read(gpu, REG_A6XX_VBIF_CLKON);
244 
245 	/* Force on the bus so we can read it */
246 	gpu_write(gpu, REG_A6XX_VBIF_CLKON,
247 		clk | A6XX_VBIF_CLKON_FORCE_ON_TESTBUS);
248 
249 	/* We will read from BUS2 first, so disable BUS1 */
250 	gpu_write(gpu, REG_A6XX_VBIF_TEST_BUS1_CTRL0, 0);
251 
252 	/* Enable the VBIF bus for reading */
253 	gpu_write(gpu, REG_A6XX_VBIF_TEST_BUS_OUT_CTRL, 1);
254 
255 	ptr = obj->data;
256 
257 	for (i = 0; i < AXI_ARB_BLOCKS; i++)
258 		ptr += vbif_debugbus_read(gpu,
259 			REG_A6XX_VBIF_TEST_BUS2_CTRL0,
260 			REG_A6XX_VBIF_TEST_BUS2_CTRL1,
261 			1 << (i + 16), 16, ptr);
262 
263 	for (i = 0; i < XIN_AXI_BLOCKS; i++)
264 		ptr += vbif_debugbus_read(gpu,
265 			REG_A6XX_VBIF_TEST_BUS2_CTRL0,
266 			REG_A6XX_VBIF_TEST_BUS2_CTRL1,
267 			1 << i, 18, ptr);
268 
269 	/* Stop BUS2 so we can turn on BUS1 */
270 	gpu_write(gpu, REG_A6XX_VBIF_TEST_BUS2_CTRL0, 0);
271 
272 	for (i = 0; i < XIN_CORE_BLOCKS; i++)
273 		ptr += vbif_debugbus_read(gpu,
274 			REG_A6XX_VBIF_TEST_BUS1_CTRL0,
275 			REG_A6XX_VBIF_TEST_BUS1_CTRL1,
276 			1 << i, 12, ptr);
277 
278 	/* Restore the VBIF clock setting */
279 	gpu_write(gpu, REG_A6XX_VBIF_CLKON, clk);
280 }
281 
282 static void a6xx_get_debugbus_block(struct msm_gpu *gpu,
283 		struct a6xx_gpu_state *a6xx_state,
284 		const struct a6xx_debugbus_block *block,
285 		struct a6xx_gpu_state_obj *obj)
286 {
287 	int i;
288 	u32 *ptr;
289 
290 	obj->data = state_kcalloc(a6xx_state, block->count, sizeof(u64));
291 	if (!obj->data)
292 		return;
293 
294 	obj->handle = block;
295 
296 	for (ptr = obj->data, i = 0; i < block->count; i++)
297 		ptr += debugbus_read(gpu, block->id, i, ptr);
298 }
299 
300 static void a6xx_get_cx_debugbus_block(void __iomem *cxdbg,
301 		struct a6xx_gpu_state *a6xx_state,
302 		const struct a6xx_debugbus_block *block,
303 		struct a6xx_gpu_state_obj *obj)
304 {
305 	int i;
306 	u32 *ptr;
307 
308 	obj->data = state_kcalloc(a6xx_state, block->count, sizeof(u64));
309 	if (!obj->data)
310 		return;
311 
312 	obj->handle = block;
313 
314 	for (ptr = obj->data, i = 0; i < block->count; i++)
315 		ptr += cx_debugbus_read(cxdbg, block->id, i, ptr);
316 }
317 
318 static void a6xx_get_debugbus(struct msm_gpu *gpu,
319 		struct a6xx_gpu_state *a6xx_state)
320 {
321 	struct resource *res;
322 	void __iomem *cxdbg = NULL;
323 
324 	/* Set up the GX debug bus */
325 
326 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_CNTLT,
327 		A6XX_DBGC_CFG_DBGBUS_CNTLT_SEGT(0xf));
328 
329 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_CNTLM,
330 		A6XX_DBGC_CFG_DBGBUS_CNTLM_ENABLE(0xf));
331 
332 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_0, 0);
333 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_1, 0);
334 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_2, 0);
335 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_3, 0);
336 
337 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_BYTEL_0, 0x76543210);
338 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_BYTEL_1, 0xFEDCBA98);
339 
340 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_0, 0);
341 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_1, 0);
342 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_2, 0);
343 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_3, 0);
344 
345 	/* Set up the CX debug bus - it lives elsewhere in the system so do a
346 	 * temporary ioremap for the registers
347 	 */
348 	res = platform_get_resource_byname(gpu->pdev, IORESOURCE_MEM,
349 			"cx_dbgc");
350 
351 	if (res)
352 		cxdbg = ioremap(res->start, resource_size(res));
353 
354 	if (cxdbg) {
355 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_CNTLT,
356 			A6XX_DBGC_CFG_DBGBUS_CNTLT_SEGT(0xf));
357 
358 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_CNTLM,
359 			A6XX_DBGC_CFG_DBGBUS_CNTLM_ENABLE(0xf));
360 
361 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_0, 0);
362 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_1, 0);
363 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_2, 0);
364 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_3, 0);
365 
366 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_BYTEL_0,
367 			0x76543210);
368 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_BYTEL_1,
369 			0xFEDCBA98);
370 
371 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_0, 0);
372 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_1, 0);
373 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_2, 0);
374 		cxdbg_write(cxdbg, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_3, 0);
375 	}
376 
377 	a6xx_state->debugbus = state_kcalloc(a6xx_state,
378 		ARRAY_SIZE(a6xx_debugbus_blocks),
379 		sizeof(*a6xx_state->debugbus));
380 
381 	if (a6xx_state->debugbus) {
382 		int i;
383 
384 		for (i = 0; i < ARRAY_SIZE(a6xx_debugbus_blocks); i++)
385 			a6xx_get_debugbus_block(gpu,
386 				a6xx_state,
387 				&a6xx_debugbus_blocks[i],
388 				&a6xx_state->debugbus[i]);
389 
390 		a6xx_state->nr_debugbus = ARRAY_SIZE(a6xx_debugbus_blocks);
391 	}
392 
393 	a6xx_state->vbif_debugbus =
394 		state_kcalloc(a6xx_state, 1,
395 			sizeof(*a6xx_state->vbif_debugbus));
396 
397 	if (a6xx_state->vbif_debugbus)
398 		a6xx_get_vbif_debugbus_block(gpu, a6xx_state,
399 			a6xx_state->vbif_debugbus);
400 
401 	if (cxdbg) {
402 		a6xx_state->cx_debugbus =
403 			state_kcalloc(a6xx_state,
404 			ARRAY_SIZE(a6xx_cx_debugbus_blocks),
405 			sizeof(*a6xx_state->cx_debugbus));
406 
407 		if (a6xx_state->cx_debugbus) {
408 			int i;
409 
410 			for (i = 0; i < ARRAY_SIZE(a6xx_cx_debugbus_blocks); i++)
411 				a6xx_get_cx_debugbus_block(cxdbg,
412 					a6xx_state,
413 					&a6xx_cx_debugbus_blocks[i],
414 					&a6xx_state->cx_debugbus[i]);
415 
416 			a6xx_state->nr_cx_debugbus =
417 				ARRAY_SIZE(a6xx_cx_debugbus_blocks);
418 		}
419 
420 		iounmap(cxdbg);
421 	}
422 }
423 
424 #define RANGE(reg, a) ((reg)[(a) + 1] - (reg)[(a)] + 1)
425 
426 /* Read a data cluster from behind the AHB aperture */
427 static void a6xx_get_dbgahb_cluster(struct msm_gpu *gpu,
428 		struct a6xx_gpu_state *a6xx_state,
429 		const struct a6xx_dbgahb_cluster *dbgahb,
430 		struct a6xx_gpu_state_obj *obj,
431 		struct a6xx_crashdumper *dumper)
432 {
433 	u64 *in = dumper->ptr;
434 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
435 	size_t datasize;
436 	int i, regcount = 0;
437 
438 	for (i = 0; i < A6XX_NUM_CONTEXTS; i++) {
439 		int j;
440 
441 		in += CRASHDUMP_WRITE(in, REG_A6XX_HLSQ_DBG_READ_SEL,
442 			(dbgahb->statetype + i * 2) << 8);
443 
444 		for (j = 0; j < dbgahb->count; j += 2) {
445 			int count = RANGE(dbgahb->registers, j);
446 			u32 offset = REG_A6XX_HLSQ_DBG_AHB_READ_APERTURE +
447 				dbgahb->registers[j] - (dbgahb->base >> 2);
448 
449 			in += CRASHDUMP_READ(in, offset, count, out);
450 
451 			out += count * sizeof(u32);
452 
453 			if (i == 0)
454 				regcount += count;
455 		}
456 	}
457 
458 	CRASHDUMP_FINI(in);
459 
460 	datasize = regcount * A6XX_NUM_CONTEXTS * sizeof(u32);
461 
462 	if (WARN_ON(datasize > A6XX_CD_DATA_SIZE))
463 		return;
464 
465 	if (a6xx_crashdumper_run(gpu, dumper))
466 		return;
467 
468 	obj->handle = dbgahb;
469 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
470 		datasize);
471 }
472 
473 static void a6xx_get_dbgahb_clusters(struct msm_gpu *gpu,
474 		struct a6xx_gpu_state *a6xx_state,
475 		struct a6xx_crashdumper *dumper)
476 {
477 	int i;
478 
479 	a6xx_state->dbgahb_clusters = state_kcalloc(a6xx_state,
480 		ARRAY_SIZE(a6xx_dbgahb_clusters),
481 		sizeof(*a6xx_state->dbgahb_clusters));
482 
483 	if (!a6xx_state->dbgahb_clusters)
484 		return;
485 
486 	a6xx_state->nr_dbgahb_clusters = ARRAY_SIZE(a6xx_dbgahb_clusters);
487 
488 	for (i = 0; i < ARRAY_SIZE(a6xx_dbgahb_clusters); i++)
489 		a6xx_get_dbgahb_cluster(gpu, a6xx_state,
490 			&a6xx_dbgahb_clusters[i],
491 			&a6xx_state->dbgahb_clusters[i], dumper);
492 }
493 
494 /* Read a data cluster from the CP aperture with the crashdumper */
495 static void a6xx_get_cluster(struct msm_gpu *gpu,
496 		struct a6xx_gpu_state *a6xx_state,
497 		const struct a6xx_cluster *cluster,
498 		struct a6xx_gpu_state_obj *obj,
499 		struct a6xx_crashdumper *dumper)
500 {
501 	u64 *in = dumper->ptr;
502 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
503 	size_t datasize;
504 	int i, regcount = 0;
505 
506 	/* Some clusters need a selector register to be programmed too */
507 	if (cluster->sel_reg)
508 		in += CRASHDUMP_WRITE(in, cluster->sel_reg, cluster->sel_val);
509 
510 	for (i = 0; i < A6XX_NUM_CONTEXTS; i++) {
511 		int j;
512 
513 		in += CRASHDUMP_WRITE(in, REG_A6XX_CP_APERTURE_CNTL_CD,
514 			(cluster->id << 8) | (i << 4) | i);
515 
516 		for (j = 0; j < cluster->count; j += 2) {
517 			int count = RANGE(cluster->registers, j);
518 
519 			in += CRASHDUMP_READ(in, cluster->registers[j],
520 				count, out);
521 
522 			out += count * sizeof(u32);
523 
524 			if (i == 0)
525 				regcount += count;
526 		}
527 	}
528 
529 	CRASHDUMP_FINI(in);
530 
531 	datasize = regcount * A6XX_NUM_CONTEXTS * sizeof(u32);
532 
533 	if (WARN_ON(datasize > A6XX_CD_DATA_SIZE))
534 		return;
535 
536 	if (a6xx_crashdumper_run(gpu, dumper))
537 		return;
538 
539 	obj->handle = cluster;
540 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
541 		datasize);
542 }
543 
544 static void a6xx_get_clusters(struct msm_gpu *gpu,
545 		struct a6xx_gpu_state *a6xx_state,
546 		struct a6xx_crashdumper *dumper)
547 {
548 	int i;
549 
550 	a6xx_state->clusters = state_kcalloc(a6xx_state,
551 		ARRAY_SIZE(a6xx_clusters), sizeof(*a6xx_state->clusters));
552 
553 	if (!a6xx_state->clusters)
554 		return;
555 
556 	a6xx_state->nr_clusters = ARRAY_SIZE(a6xx_clusters);
557 
558 	for (i = 0; i < ARRAY_SIZE(a6xx_clusters); i++)
559 		a6xx_get_cluster(gpu, a6xx_state, &a6xx_clusters[i],
560 			&a6xx_state->clusters[i], dumper);
561 }
562 
563 /* Read a shader / debug block from the HLSQ aperture with the crashdumper */
564 static void a6xx_get_shader_block(struct msm_gpu *gpu,
565 		struct a6xx_gpu_state *a6xx_state,
566 		const struct a6xx_shader_block *block,
567 		struct a6xx_gpu_state_obj *obj,
568 		struct a6xx_crashdumper *dumper)
569 {
570 	u64 *in = dumper->ptr;
571 	size_t datasize = block->size * A6XX_NUM_SHADER_BANKS * sizeof(u32);
572 	int i;
573 
574 	if (WARN_ON(datasize > A6XX_CD_DATA_SIZE))
575 		return;
576 
577 	for (i = 0; i < A6XX_NUM_SHADER_BANKS; i++) {
578 		in += CRASHDUMP_WRITE(in, REG_A6XX_HLSQ_DBG_READ_SEL,
579 			(block->type << 8) | i);
580 
581 		in += CRASHDUMP_READ(in, REG_A6XX_HLSQ_DBG_AHB_READ_APERTURE,
582 			block->size, dumper->iova + A6XX_CD_DATA_OFFSET);
583 	}
584 
585 	CRASHDUMP_FINI(in);
586 
587 	if (a6xx_crashdumper_run(gpu, dumper))
588 		return;
589 
590 	obj->handle = block;
591 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
592 		datasize);
593 }
594 
595 static void a6xx_get_shaders(struct msm_gpu *gpu,
596 		struct a6xx_gpu_state *a6xx_state,
597 		struct a6xx_crashdumper *dumper)
598 {
599 	int i;
600 
601 	a6xx_state->shaders = state_kcalloc(a6xx_state,
602 		ARRAY_SIZE(a6xx_shader_blocks), sizeof(*a6xx_state->shaders));
603 
604 	if (!a6xx_state->shaders)
605 		return;
606 
607 	a6xx_state->nr_shaders = ARRAY_SIZE(a6xx_shader_blocks);
608 
609 	for (i = 0; i < ARRAY_SIZE(a6xx_shader_blocks); i++)
610 		a6xx_get_shader_block(gpu, a6xx_state, &a6xx_shader_blocks[i],
611 			&a6xx_state->shaders[i], dumper);
612 }
613 
614 /* Read registers from behind the HLSQ aperture with the crashdumper */
615 static void a6xx_get_crashdumper_hlsq_registers(struct msm_gpu *gpu,
616 		struct a6xx_gpu_state *a6xx_state,
617 		const struct a6xx_registers *regs,
618 		struct a6xx_gpu_state_obj *obj,
619 		struct a6xx_crashdumper *dumper)
620 
621 {
622 	u64 *in = dumper->ptr;
623 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
624 	int i, regcount = 0;
625 
626 	in += CRASHDUMP_WRITE(in, REG_A6XX_HLSQ_DBG_READ_SEL, regs->val1);
627 
628 	for (i = 0; i < regs->count; i += 2) {
629 		u32 count = RANGE(regs->registers, i);
630 		u32 offset = REG_A6XX_HLSQ_DBG_AHB_READ_APERTURE +
631 			regs->registers[i] - (regs->val0 >> 2);
632 
633 		in += CRASHDUMP_READ(in, offset, count, out);
634 
635 		out += count * sizeof(u32);
636 		regcount += count;
637 	}
638 
639 	CRASHDUMP_FINI(in);
640 
641 	if (WARN_ON((regcount * sizeof(u32)) > A6XX_CD_DATA_SIZE))
642 		return;
643 
644 	if (a6xx_crashdumper_run(gpu, dumper))
645 		return;
646 
647 	obj->handle = regs;
648 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
649 		regcount * sizeof(u32));
650 }
651 
652 /* Read a block of registers using the crashdumper */
653 static void a6xx_get_crashdumper_registers(struct msm_gpu *gpu,
654 		struct a6xx_gpu_state *a6xx_state,
655 		const struct a6xx_registers *regs,
656 		struct a6xx_gpu_state_obj *obj,
657 		struct a6xx_crashdumper *dumper)
658 
659 {
660 	u64 *in = dumper->ptr;
661 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
662 	int i, regcount = 0;
663 
664 	/* Some blocks might need to program a selector register first */
665 	if (regs->val0)
666 		in += CRASHDUMP_WRITE(in, regs->val0, regs->val1);
667 
668 	for (i = 0; i < regs->count; i += 2) {
669 		u32 count = RANGE(regs->registers, i);
670 
671 		in += CRASHDUMP_READ(in, regs->registers[i], count, out);
672 
673 		out += count * sizeof(u32);
674 		regcount += count;
675 	}
676 
677 	CRASHDUMP_FINI(in);
678 
679 	if (WARN_ON((regcount * sizeof(u32)) > A6XX_CD_DATA_SIZE))
680 		return;
681 
682 	if (a6xx_crashdumper_run(gpu, dumper))
683 		return;
684 
685 	obj->handle = regs;
686 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
687 		regcount * sizeof(u32));
688 }
689 
690 /* Read a block of registers via AHB */
691 static void a6xx_get_ahb_gpu_registers(struct msm_gpu *gpu,
692 		struct a6xx_gpu_state *a6xx_state,
693 		const struct a6xx_registers *regs,
694 		struct a6xx_gpu_state_obj *obj)
695 {
696 	int i, regcount = 0, index = 0;
697 
698 	for (i = 0; i < regs->count; i += 2)
699 		regcount += RANGE(regs->registers, i);
700 
701 	obj->handle = (const void *) regs;
702 	obj->data = state_kcalloc(a6xx_state, regcount, sizeof(u32));
703 	if (!obj->data)
704 		return;
705 
706 	for (i = 0; i < regs->count; i += 2) {
707 		u32 count = RANGE(regs->registers, i);
708 		int j;
709 
710 		for (j = 0; j < count; j++)
711 			obj->data[index++] = gpu_read(gpu,
712 				regs->registers[i] + j);
713 	}
714 }
715 
716 /* Read a block of GMU registers */
717 static void _a6xx_get_gmu_registers(struct msm_gpu *gpu,
718 		struct a6xx_gpu_state *a6xx_state,
719 		const struct a6xx_registers *regs,
720 		struct a6xx_gpu_state_obj *obj)
721 {
722 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
723 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
724 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
725 	int i, regcount = 0, index = 0;
726 
727 	for (i = 0; i < regs->count; i += 2)
728 		regcount += RANGE(regs->registers, i);
729 
730 	obj->handle = (const void *) regs;
731 	obj->data = state_kcalloc(a6xx_state, regcount, sizeof(u32));
732 	if (!obj->data)
733 		return;
734 
735 	for (i = 0; i < regs->count; i += 2) {
736 		u32 count = RANGE(regs->registers, i);
737 		int j;
738 
739 		for (j = 0; j < count; j++)
740 			obj->data[index++] = gmu_read(gmu,
741 				regs->registers[i] + j);
742 	}
743 }
744 
745 static void a6xx_get_gmu_registers(struct msm_gpu *gpu,
746 		struct a6xx_gpu_state *a6xx_state)
747 {
748 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
749 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
750 
751 	a6xx_state->gmu_registers = state_kcalloc(a6xx_state,
752 		2, sizeof(*a6xx_state->gmu_registers));
753 
754 	if (!a6xx_state->gmu_registers)
755 		return;
756 
757 	a6xx_state->nr_gmu_registers = 2;
758 
759 	/* Get the CX GMU registers from AHB */
760 	_a6xx_get_gmu_registers(gpu, a6xx_state, &a6xx_gmu_reglist[0],
761 		&a6xx_state->gmu_registers[0]);
762 
763 	if (!a6xx_gmu_gx_is_on(&a6xx_gpu->gmu))
764 		return;
765 
766 	/* Set the fence to ALLOW mode so we can access the registers */
767 	gpu_write(gpu, REG_A6XX_GMU_AO_AHB_FENCE_CTRL, 0);
768 
769 	_a6xx_get_gmu_registers(gpu, a6xx_state, &a6xx_gmu_reglist[1],
770 		&a6xx_state->gmu_registers[1]);
771 }
772 
773 static void a6xx_get_registers(struct msm_gpu *gpu,
774 		struct a6xx_gpu_state *a6xx_state,
775 		struct a6xx_crashdumper *dumper)
776 {
777 	int i, count = ARRAY_SIZE(a6xx_ahb_reglist) +
778 		ARRAY_SIZE(a6xx_reglist) +
779 		ARRAY_SIZE(a6xx_hlsq_reglist);
780 	int index = 0;
781 
782 	a6xx_state->registers = state_kcalloc(a6xx_state,
783 		count, sizeof(*a6xx_state->registers));
784 
785 	if (!a6xx_state->registers)
786 		return;
787 
788 	a6xx_state->nr_registers = count;
789 
790 	for (i = 0; i < ARRAY_SIZE(a6xx_ahb_reglist); i++)
791 		a6xx_get_ahb_gpu_registers(gpu,
792 			a6xx_state, &a6xx_ahb_reglist[i],
793 			&a6xx_state->registers[index++]);
794 
795 	for (i = 0; i < ARRAY_SIZE(a6xx_reglist); i++)
796 		a6xx_get_crashdumper_registers(gpu,
797 			a6xx_state, &a6xx_reglist[i],
798 			&a6xx_state->registers[index++],
799 			dumper);
800 
801 	for (i = 0; i < ARRAY_SIZE(a6xx_hlsq_reglist); i++)
802 		a6xx_get_crashdumper_hlsq_registers(gpu,
803 			a6xx_state, &a6xx_hlsq_reglist[i],
804 			&a6xx_state->registers[index++],
805 			dumper);
806 }
807 
808 /* Read a block of data from an indexed register pair */
809 static void a6xx_get_indexed_regs(struct msm_gpu *gpu,
810 		struct a6xx_gpu_state *a6xx_state,
811 		const struct a6xx_indexed_registers *indexed,
812 		struct a6xx_gpu_state_obj *obj)
813 {
814 	int i;
815 
816 	obj->handle = (const void *) indexed;
817 	obj->data = state_kcalloc(a6xx_state, indexed->count, sizeof(u32));
818 	if (!obj->data)
819 		return;
820 
821 	/* All the indexed banks start at address 0 */
822 	gpu_write(gpu, indexed->addr, 0);
823 
824 	/* Read the data - each read increments the internal address by 1 */
825 	for (i = 0; i < indexed->count; i++)
826 		obj->data[i] = gpu_read(gpu, indexed->data);
827 }
828 
829 static void a6xx_get_indexed_registers(struct msm_gpu *gpu,
830 		struct a6xx_gpu_state *a6xx_state)
831 {
832 	u32 mempool_size;
833 	int count = ARRAY_SIZE(a6xx_indexed_reglist) + 1;
834 	int i;
835 
836 	a6xx_state->indexed_regs = state_kcalloc(a6xx_state, count,
837 		sizeof(a6xx_state->indexed_regs));
838 	if (!a6xx_state->indexed_regs)
839 		return;
840 
841 	for (i = 0; i < ARRAY_SIZE(a6xx_indexed_reglist); i++)
842 		a6xx_get_indexed_regs(gpu, a6xx_state, &a6xx_indexed_reglist[i],
843 			&a6xx_state->indexed_regs[i]);
844 
845 	/* Set the CP mempool size to 0 to stabilize it while dumping */
846 	mempool_size = gpu_read(gpu, REG_A6XX_CP_MEM_POOL_SIZE);
847 	gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 0);
848 
849 	/* Get the contents of the CP mempool */
850 	a6xx_get_indexed_regs(gpu, a6xx_state, &a6xx_cp_mempool_indexed,
851 		&a6xx_state->indexed_regs[i]);
852 
853 	/*
854 	 * Offset 0x2000 in the mempool is the size - copy the saved size over
855 	 * so the data is consistent
856 	 */
857 	a6xx_state->indexed_regs[i].data[0x2000] = mempool_size;
858 
859 	/* Restore the size in the hardware */
860 	gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, mempool_size);
861 
862 	a6xx_state->nr_indexed_regs = count;
863 }
864 
865 struct msm_gpu_state *a6xx_gpu_state_get(struct msm_gpu *gpu)
866 {
867 	struct a6xx_crashdumper dumper = { 0 };
868 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
869 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
870 	struct a6xx_gpu_state *a6xx_state = kzalloc(sizeof(*a6xx_state),
871 		GFP_KERNEL);
872 
873 	if (!a6xx_state)
874 		return ERR_PTR(-ENOMEM);
875 
876 	INIT_LIST_HEAD(&a6xx_state->objs);
877 
878 	/* Get the generic state from the adreno core */
879 	adreno_gpu_state_get(gpu, &a6xx_state->base);
880 
881 	a6xx_get_gmu_registers(gpu, a6xx_state);
882 
883 	/* If GX isn't on the rest of the data isn't going to be accessible */
884 	if (!a6xx_gmu_gx_is_on(&a6xx_gpu->gmu))
885 		return &a6xx_state->base;
886 
887 	/* Get the banks of indexed registers */
888 	a6xx_get_indexed_registers(gpu, a6xx_state);
889 
890 	/* Try to initialize the crashdumper */
891 	if (!a6xx_crashdumper_init(gpu, &dumper)) {
892 		a6xx_get_registers(gpu, a6xx_state, &dumper);
893 		a6xx_get_shaders(gpu, a6xx_state, &dumper);
894 		a6xx_get_clusters(gpu, a6xx_state, &dumper);
895 		a6xx_get_dbgahb_clusters(gpu, a6xx_state, &dumper);
896 
897 		msm_gem_kernel_put(dumper.bo, gpu->aspace, true);
898 	}
899 
900 	a6xx_get_debugbus(gpu, a6xx_state);
901 
902 	return  &a6xx_state->base;
903 }
904 
905 void a6xx_gpu_state_destroy(struct kref *kref)
906 {
907 	struct a6xx_state_memobj *obj, *tmp;
908 	struct msm_gpu_state *state = container_of(kref,
909 			struct msm_gpu_state, ref);
910 	struct a6xx_gpu_state *a6xx_state = container_of(state,
911 			struct a6xx_gpu_state, base);
912 
913 	list_for_each_entry_safe(obj, tmp, &a6xx_state->objs, node)
914 		kfree(obj);
915 
916 	adreno_gpu_state_destroy(state);
917 	kfree(a6xx_state);
918 }
919 
920 int a6xx_gpu_state_put(struct msm_gpu_state *state)
921 {
922 	if (IS_ERR_OR_NULL(state))
923 		return 1;
924 
925 	return kref_put(&state->ref, a6xx_gpu_state_destroy);
926 }
927 
928 static void a6xx_show_registers(const u32 *registers, u32 *data, size_t count,
929 		struct drm_printer *p)
930 {
931 	int i, index = 0;
932 
933 	if (!data)
934 		return;
935 
936 	for (i = 0; i < count; i += 2) {
937 		u32 count = RANGE(registers, i);
938 		u32 offset = registers[i];
939 		int j;
940 
941 		for (j = 0; j < count; index++, offset++, j++) {
942 			if (data[index] == 0xdeafbead)
943 				continue;
944 
945 			drm_printf(p, "  - { offset: 0x%06x, value: 0x%08x }\n",
946 				offset << 2, data[index]);
947 		}
948 	}
949 }
950 
951 static void print_ascii85(struct drm_printer *p, size_t len, u32 *data)
952 {
953 	char out[ASCII85_BUFSZ];
954 	long i, l, datalen = 0;
955 
956 	for (i = 0; i < len >> 2; i++) {
957 		if (data[i])
958 			datalen = (i + 1) << 2;
959 	}
960 
961 	if (datalen == 0)
962 		return;
963 
964 	drm_puts(p, "    data: !!ascii85 |\n");
965 	drm_puts(p, "      ");
966 
967 
968 	l = ascii85_encode_len(datalen);
969 
970 	for (i = 0; i < l; i++)
971 		drm_puts(p, ascii85_encode(data[i], out));
972 
973 	drm_puts(p, "\n");
974 }
975 
976 static void print_name(struct drm_printer *p, const char *fmt, const char *name)
977 {
978 	drm_puts(p, fmt);
979 	drm_puts(p, name);
980 	drm_puts(p, "\n");
981 }
982 
983 static void a6xx_show_shader(struct a6xx_gpu_state_obj *obj,
984 		struct drm_printer *p)
985 {
986 	const struct a6xx_shader_block *block = obj->handle;
987 	int i;
988 
989 	if (!obj->handle)
990 		return;
991 
992 	print_name(p, "  - type: ", block->name);
993 
994 	for (i = 0; i < A6XX_NUM_SHADER_BANKS; i++) {
995 		drm_printf(p, "    - bank: %d\n", i);
996 		drm_printf(p, "      size: %d\n", block->size);
997 
998 		if (!obj->data)
999 			continue;
1000 
1001 		print_ascii85(p, block->size << 2,
1002 			obj->data + (block->size * i));
1003 	}
1004 }
1005 
1006 static void a6xx_show_cluster_data(const u32 *registers, int size, u32 *data,
1007 		struct drm_printer *p)
1008 {
1009 	int ctx, index = 0;
1010 
1011 	for (ctx = 0; ctx < A6XX_NUM_CONTEXTS; ctx++) {
1012 		int j;
1013 
1014 		drm_printf(p, "    - context: %d\n", ctx);
1015 
1016 		for (j = 0; j < size; j += 2) {
1017 			u32 count = RANGE(registers, j);
1018 			u32 offset = registers[j];
1019 			int k;
1020 
1021 			for (k = 0; k < count; index++, offset++, k++) {
1022 				if (data[index] == 0xdeafbead)
1023 					continue;
1024 
1025 				drm_printf(p, "      - { offset: 0x%06x, value: 0x%08x }\n",
1026 					offset << 2, data[index]);
1027 			}
1028 		}
1029 	}
1030 }
1031 
1032 static void a6xx_show_dbgahb_cluster(struct a6xx_gpu_state_obj *obj,
1033 		struct drm_printer *p)
1034 {
1035 	const struct a6xx_dbgahb_cluster *dbgahb = obj->handle;
1036 
1037 	if (dbgahb) {
1038 		print_name(p, "  - cluster-name: ", dbgahb->name);
1039 		a6xx_show_cluster_data(dbgahb->registers, dbgahb->count,
1040 			obj->data, p);
1041 	}
1042 }
1043 
1044 static void a6xx_show_cluster(struct a6xx_gpu_state_obj *obj,
1045 		struct drm_printer *p)
1046 {
1047 	const struct a6xx_cluster *cluster = obj->handle;
1048 
1049 	if (cluster) {
1050 		print_name(p, "  - cluster-name: ", cluster->name);
1051 		a6xx_show_cluster_data(cluster->registers, cluster->count,
1052 			obj->data, p);
1053 	}
1054 }
1055 
1056 static void a6xx_show_indexed_regs(struct a6xx_gpu_state_obj *obj,
1057 		struct drm_printer *p)
1058 {
1059 	const struct a6xx_indexed_registers *indexed = obj->handle;
1060 
1061 	if (!indexed)
1062 		return;
1063 
1064 	print_name(p, "  - regs-name: ", indexed->name);
1065 	drm_printf(p, "    dwords: %d\n", indexed->count);
1066 
1067 	print_ascii85(p, indexed->count << 2, obj->data);
1068 }
1069 
1070 static void a6xx_show_debugbus_block(const struct a6xx_debugbus_block *block,
1071 		u32 *data, struct drm_printer *p)
1072 {
1073 	if (block) {
1074 		print_name(p, "  - debugbus-block: ", block->name);
1075 
1076 		/*
1077 		 * count for regular debugbus data is in quadwords,
1078 		 * but print the size in dwords for consistency
1079 		 */
1080 		drm_printf(p, "    count: %d\n", block->count << 1);
1081 
1082 		print_ascii85(p, block->count << 3, data);
1083 	}
1084 }
1085 
1086 static void a6xx_show_debugbus(struct a6xx_gpu_state *a6xx_state,
1087 		struct drm_printer *p)
1088 {
1089 	int i;
1090 
1091 	for (i = 0; i < a6xx_state->nr_debugbus; i++) {
1092 		struct a6xx_gpu_state_obj *obj = &a6xx_state->debugbus[i];
1093 
1094 		a6xx_show_debugbus_block(obj->handle, obj->data, p);
1095 	}
1096 
1097 	if (a6xx_state->vbif_debugbus) {
1098 		struct a6xx_gpu_state_obj *obj = a6xx_state->vbif_debugbus;
1099 
1100 		drm_puts(p, "  - debugbus-block: A6XX_DBGBUS_VBIF\n");
1101 		drm_printf(p, "    count: %d\n", VBIF_DEBUGBUS_BLOCK_SIZE);
1102 
1103 		/* vbif debugbus data is in dwords.  Confusing, huh? */
1104 		print_ascii85(p, VBIF_DEBUGBUS_BLOCK_SIZE << 2, obj->data);
1105 	}
1106 
1107 	for (i = 0; i < a6xx_state->nr_cx_debugbus; i++) {
1108 		struct a6xx_gpu_state_obj *obj = &a6xx_state->cx_debugbus[i];
1109 
1110 		a6xx_show_debugbus_block(obj->handle, obj->data, p);
1111 	}
1112 }
1113 
1114 void a6xx_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
1115 		struct drm_printer *p)
1116 {
1117 	struct a6xx_gpu_state *a6xx_state = container_of(state,
1118 			struct a6xx_gpu_state, base);
1119 	int i;
1120 
1121 	if (IS_ERR_OR_NULL(state))
1122 		return;
1123 
1124 	adreno_show(gpu, state, p);
1125 
1126 	drm_puts(p, "registers:\n");
1127 	for (i = 0; i < a6xx_state->nr_registers; i++) {
1128 		struct a6xx_gpu_state_obj *obj = &a6xx_state->registers[i];
1129 		const struct a6xx_registers *regs = obj->handle;
1130 
1131 		if (!obj->handle)
1132 			continue;
1133 
1134 		a6xx_show_registers(regs->registers, obj->data, regs->count, p);
1135 	}
1136 
1137 	drm_puts(p, "registers-gmu:\n");
1138 	for (i = 0; i < a6xx_state->nr_gmu_registers; i++) {
1139 		struct a6xx_gpu_state_obj *obj = &a6xx_state->gmu_registers[i];
1140 		const struct a6xx_registers *regs = obj->handle;
1141 
1142 		if (!obj->handle)
1143 			continue;
1144 
1145 		a6xx_show_registers(regs->registers, obj->data, regs->count, p);
1146 	}
1147 
1148 	drm_puts(p, "indexed-registers:\n");
1149 	for (i = 0; i < a6xx_state->nr_indexed_regs; i++)
1150 		a6xx_show_indexed_regs(&a6xx_state->indexed_regs[i], p);
1151 
1152 	drm_puts(p, "shader-blocks:\n");
1153 	for (i = 0; i < a6xx_state->nr_shaders; i++)
1154 		a6xx_show_shader(&a6xx_state->shaders[i], p);
1155 
1156 	drm_puts(p, "clusters:\n");
1157 	for (i = 0; i < a6xx_state->nr_clusters; i++)
1158 		a6xx_show_cluster(&a6xx_state->clusters[i], p);
1159 
1160 	for (i = 0; i < a6xx_state->nr_dbgahb_clusters; i++)
1161 		a6xx_show_dbgahb_cluster(&a6xx_state->dbgahb_clusters[i], p);
1162 
1163 	drm_puts(p, "debugbus:\n");
1164 	a6xx_show_debugbus(a6xx_state, p);
1165 }
1166