xref: /linux/drivers/gpu/drm/msm/adreno/a6xx_gpu_state.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2018-2019 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 static const unsigned int *gen7_0_0_external_core_regs[] __always_unused;
12 static const unsigned int *gen7_2_0_external_core_regs[] __always_unused;
13 static const unsigned int *gen7_9_0_external_core_regs[] __always_unused;
14 static struct gen7_sptp_cluster_registers gen7_9_0_sptp_clusters[] __always_unused;
15 static const u32 gen7_9_0_cx_debugbus_blocks[] __always_unused;
16 
17 #include "adreno_gen7_0_0_snapshot.h"
18 #include "adreno_gen7_2_0_snapshot.h"
19 #include "adreno_gen7_9_0_snapshot.h"
20 
21 struct a6xx_gpu_state_obj {
22 	const void *handle;
23 	u32 *data;
24 	u32 count;	/* optional, used when count potentially read from hw */
25 };
26 
27 struct a6xx_gpu_state {
28 	struct msm_gpu_state base;
29 
30 	struct a6xx_gpu_state_obj *gmu_registers;
31 	int nr_gmu_registers;
32 
33 	struct a6xx_gpu_state_obj *registers;
34 	int nr_registers;
35 
36 	struct a6xx_gpu_state_obj *shaders;
37 	int nr_shaders;
38 
39 	struct a6xx_gpu_state_obj *clusters;
40 	int nr_clusters;
41 
42 	struct a6xx_gpu_state_obj *dbgahb_clusters;
43 	int nr_dbgahb_clusters;
44 
45 	struct a6xx_gpu_state_obj *indexed_regs;
46 	int nr_indexed_regs;
47 
48 	struct a6xx_gpu_state_obj *debugbus;
49 	int nr_debugbus;
50 
51 	struct a6xx_gpu_state_obj *vbif_debugbus;
52 
53 	struct a6xx_gpu_state_obj *cx_debugbus;
54 	int nr_cx_debugbus;
55 
56 	struct msm_gpu_state_bo *gmu_log;
57 	struct msm_gpu_state_bo *gmu_hfi;
58 	struct msm_gpu_state_bo *gmu_debug;
59 
60 	s32 hfi_queue_history[2][HFI_HISTORY_SZ];
61 
62 	struct list_head objs;
63 
64 	bool gpu_initialized;
65 };
66 
67 static inline int CRASHDUMP_WRITE(u64 *in, u32 reg, u32 val)
68 {
69 	in[0] = val;
70 	in[1] = (((u64) reg) << 44 | (1 << 21) | 1);
71 
72 	return 2;
73 }
74 
75 static inline int CRASHDUMP_READ(u64 *in, u32 reg, u32 dwords, u64 target)
76 {
77 	in[0] = target;
78 	in[1] = (((u64) reg) << 44 | dwords);
79 
80 	return 2;
81 }
82 
83 static inline int CRASHDUMP_FINI(u64 *in)
84 {
85 	in[0] = 0;
86 	in[1] = 0;
87 
88 	return 2;
89 }
90 
91 struct a6xx_crashdumper {
92 	void *ptr;
93 	struct drm_gem_object *bo;
94 	u64 iova;
95 };
96 
97 struct a6xx_state_memobj {
98 	struct list_head node;
99 	unsigned long long data[];
100 };
101 
102 static void *state_kcalloc(struct a6xx_gpu_state *a6xx_state, int nr, size_t objsize)
103 {
104 	struct a6xx_state_memobj *obj =
105 		kvzalloc((nr * objsize) + sizeof(*obj), GFP_KERNEL);
106 
107 	if (!obj)
108 		return NULL;
109 
110 	list_add_tail(&obj->node, &a6xx_state->objs);
111 	return &obj->data;
112 }
113 
114 static void *state_kmemdup(struct a6xx_gpu_state *a6xx_state, void *src,
115 		size_t size)
116 {
117 	void *dst = state_kcalloc(a6xx_state, 1, size);
118 
119 	if (dst)
120 		memcpy(dst, src, size);
121 	return dst;
122 }
123 
124 /*
125  * Allocate 1MB for the crashdumper scratch region - 8k for the script and
126  * the rest for the data
127  */
128 #define A6XX_CD_DATA_OFFSET 8192
129 #define A6XX_CD_DATA_SIZE  (SZ_1M - 8192)
130 
131 static int a6xx_crashdumper_init(struct msm_gpu *gpu,
132 		struct a6xx_crashdumper *dumper)
133 {
134 	dumper->ptr = msm_gem_kernel_new(gpu->dev,
135 		SZ_1M, MSM_BO_WC, gpu->aspace,
136 		&dumper->bo, &dumper->iova);
137 
138 	if (!IS_ERR(dumper->ptr))
139 		msm_gem_object_set_name(dumper->bo, "crashdump");
140 
141 	return PTR_ERR_OR_ZERO(dumper->ptr);
142 }
143 
144 static int a6xx_crashdumper_run(struct msm_gpu *gpu,
145 		struct a6xx_crashdumper *dumper)
146 {
147 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
148 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
149 	u32 val;
150 	int ret;
151 
152 	if (IS_ERR_OR_NULL(dumper->ptr))
153 		return -EINVAL;
154 
155 	if (!a6xx_gmu_sptprac_is_on(&a6xx_gpu->gmu))
156 		return -EINVAL;
157 
158 	/* Make sure all pending memory writes are posted */
159 	wmb();
160 
161 	gpu_write64(gpu, REG_A6XX_CP_CRASH_SCRIPT_BASE, dumper->iova);
162 
163 	gpu_write(gpu, REG_A6XX_CP_CRASH_DUMP_CNTL, 1);
164 
165 	ret = gpu_poll_timeout(gpu, REG_A6XX_CP_CRASH_DUMP_STATUS, val,
166 		val & 0x02, 100, 10000);
167 
168 	gpu_write(gpu, REG_A6XX_CP_CRASH_DUMP_CNTL, 0);
169 
170 	return ret;
171 }
172 
173 /* read a value from the GX debug bus */
174 static int debugbus_read(struct msm_gpu *gpu, u32 block, u32 offset,
175 		u32 *data)
176 {
177 	u32 reg = A6XX_DBGC_CFG_DBGBUS_SEL_D_PING_INDEX(offset) |
178 		A6XX_DBGC_CFG_DBGBUS_SEL_D_PING_BLK_SEL(block);
179 
180 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_SEL_A, reg);
181 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_SEL_B, reg);
182 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_SEL_C, reg);
183 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_SEL_D, reg);
184 
185 	/* Wait 1 us to make sure the data is flowing */
186 	udelay(1);
187 
188 	data[0] = gpu_read(gpu, REG_A6XX_DBGC_CFG_DBGBUS_TRACE_BUF2);
189 	data[1] = gpu_read(gpu, REG_A6XX_DBGC_CFG_DBGBUS_TRACE_BUF1);
190 
191 	return 2;
192 }
193 
194 #define cxdbg_write(ptr, offset, val) \
195 	writel((val), (ptr) + ((offset) << 2))
196 
197 #define cxdbg_read(ptr, offset) \
198 	readl((ptr) + ((offset) << 2))
199 
200 /* read a value from the CX debug bus */
201 static int cx_debugbus_read(void __iomem *cxdbg, u32 block, u32 offset,
202 		u32 *data)
203 {
204 	u32 reg = A6XX_CX_DBGC_CFG_DBGBUS_SEL_A_PING_INDEX(offset) |
205 		A6XX_CX_DBGC_CFG_DBGBUS_SEL_A_PING_BLK_SEL(block);
206 
207 	cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_SEL_A, reg);
208 	cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_SEL_B, reg);
209 	cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_SEL_C, reg);
210 	cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_SEL_D, reg);
211 
212 	/* Wait 1 us to make sure the data is flowing */
213 	udelay(1);
214 
215 	data[0] = cxdbg_read(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_TRACE_BUF2);
216 	data[1] = cxdbg_read(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_TRACE_BUF1);
217 
218 	return 2;
219 }
220 
221 /* Read a chunk of data from the VBIF debug bus */
222 static int vbif_debugbus_read(struct msm_gpu *gpu, u32 ctrl0, u32 ctrl1,
223 		u32 reg, int count, u32 *data)
224 {
225 	int i;
226 
227 	gpu_write(gpu, ctrl0, reg);
228 
229 	for (i = 0; i < count; i++) {
230 		gpu_write(gpu, ctrl1, i);
231 		data[i] = gpu_read(gpu, REG_A6XX_VBIF_TEST_BUS_OUT);
232 	}
233 
234 	return count;
235 }
236 
237 #define AXI_ARB_BLOCKS 2
238 #define XIN_AXI_BLOCKS 5
239 #define XIN_CORE_BLOCKS 4
240 
241 #define VBIF_DEBUGBUS_BLOCK_SIZE \
242 	((16 * AXI_ARB_BLOCKS) + \
243 	 (18 * XIN_AXI_BLOCKS) + \
244 	 (12 * XIN_CORE_BLOCKS))
245 
246 static void a6xx_get_vbif_debugbus_block(struct msm_gpu *gpu,
247 		struct a6xx_gpu_state *a6xx_state,
248 		struct a6xx_gpu_state_obj *obj)
249 {
250 	u32 clk, *ptr;
251 	int i;
252 
253 	obj->data = state_kcalloc(a6xx_state, VBIF_DEBUGBUS_BLOCK_SIZE,
254 		sizeof(u32));
255 	if (!obj->data)
256 		return;
257 
258 	obj->handle = NULL;
259 
260 	/* Get the current clock setting */
261 	clk = gpu_read(gpu, REG_A6XX_VBIF_CLKON);
262 
263 	/* Force on the bus so we can read it */
264 	gpu_write(gpu, REG_A6XX_VBIF_CLKON,
265 		clk | A6XX_VBIF_CLKON_FORCE_ON_TESTBUS);
266 
267 	/* We will read from BUS2 first, so disable BUS1 */
268 	gpu_write(gpu, REG_A6XX_VBIF_TEST_BUS1_CTRL0, 0);
269 
270 	/* Enable the VBIF bus for reading */
271 	gpu_write(gpu, REG_A6XX_VBIF_TEST_BUS_OUT_CTRL, 1);
272 
273 	ptr = obj->data;
274 
275 	for (i = 0; i < AXI_ARB_BLOCKS; i++)
276 		ptr += vbif_debugbus_read(gpu,
277 			REG_A6XX_VBIF_TEST_BUS2_CTRL0,
278 			REG_A6XX_VBIF_TEST_BUS2_CTRL1,
279 			1 << (i + 16), 16, ptr);
280 
281 	for (i = 0; i < XIN_AXI_BLOCKS; i++)
282 		ptr += vbif_debugbus_read(gpu,
283 			REG_A6XX_VBIF_TEST_BUS2_CTRL0,
284 			REG_A6XX_VBIF_TEST_BUS2_CTRL1,
285 			1 << i, 18, ptr);
286 
287 	/* Stop BUS2 so we can turn on BUS1 */
288 	gpu_write(gpu, REG_A6XX_VBIF_TEST_BUS2_CTRL0, 0);
289 
290 	for (i = 0; i < XIN_CORE_BLOCKS; i++)
291 		ptr += vbif_debugbus_read(gpu,
292 			REG_A6XX_VBIF_TEST_BUS1_CTRL0,
293 			REG_A6XX_VBIF_TEST_BUS1_CTRL1,
294 			1 << i, 12, ptr);
295 
296 	/* Restore the VBIF clock setting */
297 	gpu_write(gpu, REG_A6XX_VBIF_CLKON, clk);
298 }
299 
300 static void a6xx_get_debugbus_block(struct msm_gpu *gpu,
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 += debugbus_read(gpu, block->id, i, ptr);
316 }
317 
318 static void a6xx_get_cx_debugbus_block(void __iomem *cxdbg,
319 		struct a6xx_gpu_state *a6xx_state,
320 		const struct a6xx_debugbus_block *block,
321 		struct a6xx_gpu_state_obj *obj)
322 {
323 	int i;
324 	u32 *ptr;
325 
326 	obj->data = state_kcalloc(a6xx_state, block->count, sizeof(u64));
327 	if (!obj->data)
328 		return;
329 
330 	obj->handle = block;
331 
332 	for (ptr = obj->data, i = 0; i < block->count; i++)
333 		ptr += cx_debugbus_read(cxdbg, block->id, i, ptr);
334 }
335 
336 static void a6xx_get_debugbus_blocks(struct msm_gpu *gpu,
337 		struct a6xx_gpu_state *a6xx_state)
338 {
339 	int nr_debugbus_blocks = ARRAY_SIZE(a6xx_debugbus_blocks) +
340 		(a6xx_has_gbif(to_adreno_gpu(gpu)) ? 1 : 0);
341 
342 	if (adreno_is_a650_family(to_adreno_gpu(gpu)))
343 		nr_debugbus_blocks += ARRAY_SIZE(a650_debugbus_blocks);
344 
345 	a6xx_state->debugbus = state_kcalloc(a6xx_state, nr_debugbus_blocks,
346 			sizeof(*a6xx_state->debugbus));
347 
348 	if (a6xx_state->debugbus) {
349 		int i;
350 
351 		for (i = 0; i < ARRAY_SIZE(a6xx_debugbus_blocks); i++)
352 			a6xx_get_debugbus_block(gpu,
353 				a6xx_state,
354 				&a6xx_debugbus_blocks[i],
355 				&a6xx_state->debugbus[i]);
356 
357 		a6xx_state->nr_debugbus = ARRAY_SIZE(a6xx_debugbus_blocks);
358 
359 		/*
360 		 * GBIF has same debugbus as of other GPU blocks, fall back to
361 		 * default path if GPU uses GBIF, also GBIF uses exactly same
362 		 * ID as of VBIF.
363 		 */
364 		if (a6xx_has_gbif(to_adreno_gpu(gpu))) {
365 			a6xx_get_debugbus_block(gpu, a6xx_state,
366 				&a6xx_gbif_debugbus_block,
367 				&a6xx_state->debugbus[i]);
368 
369 			a6xx_state->nr_debugbus += 1;
370 		}
371 
372 
373 		if (adreno_is_a650_family(to_adreno_gpu(gpu))) {
374 			for (i = 0; i < ARRAY_SIZE(a650_debugbus_blocks); i++)
375 				a6xx_get_debugbus_block(gpu,
376 					a6xx_state,
377 					&a650_debugbus_blocks[i],
378 					&a6xx_state->debugbus[i]);
379 		}
380 	}
381 }
382 
383 static void a7xx_get_debugbus_blocks(struct msm_gpu *gpu,
384 		struct a6xx_gpu_state *a6xx_state)
385 {
386 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
387 	int debugbus_blocks_count, gbif_debugbus_blocks_count, total_debugbus_blocks;
388 	const u32 *debugbus_blocks, *gbif_debugbus_blocks;
389 	int i;
390 
391 	if (adreno_gpu->info->family == ADRENO_7XX_GEN1) {
392 		debugbus_blocks = gen7_0_0_debugbus_blocks;
393 		debugbus_blocks_count = ARRAY_SIZE(gen7_0_0_debugbus_blocks);
394 		gbif_debugbus_blocks = a7xx_gbif_debugbus_blocks;
395 		gbif_debugbus_blocks_count = ARRAY_SIZE(a7xx_gbif_debugbus_blocks);
396 	} else if (adreno_gpu->info->family == ADRENO_7XX_GEN2) {
397 		debugbus_blocks = gen7_2_0_debugbus_blocks;
398 		debugbus_blocks_count = ARRAY_SIZE(gen7_2_0_debugbus_blocks);
399 		gbif_debugbus_blocks = a7xx_gbif_debugbus_blocks;
400 		gbif_debugbus_blocks_count = ARRAY_SIZE(a7xx_gbif_debugbus_blocks);
401 	} else {
402 		BUG_ON(adreno_gpu->info->family != ADRENO_7XX_GEN3);
403 		debugbus_blocks = gen7_9_0_debugbus_blocks;
404 		debugbus_blocks_count = ARRAY_SIZE(gen7_9_0_debugbus_blocks);
405 		gbif_debugbus_blocks = gen7_9_0_gbif_debugbus_blocks;
406 		gbif_debugbus_blocks_count = ARRAY_SIZE(gen7_9_0_gbif_debugbus_blocks);
407 	}
408 
409 	total_debugbus_blocks = debugbus_blocks_count + gbif_debugbus_blocks_count;
410 
411 	a6xx_state->debugbus = state_kcalloc(a6xx_state, total_debugbus_blocks,
412 			sizeof(*a6xx_state->debugbus));
413 
414 	if (a6xx_state->debugbus) {
415 		for (i = 0; i < debugbus_blocks_count; i++) {
416 			a6xx_get_debugbus_block(gpu,
417 				a6xx_state, &a7xx_debugbus_blocks[debugbus_blocks[i]],
418 				&a6xx_state->debugbus[i]);
419 		}
420 
421 		for (i = 0; i < gbif_debugbus_blocks_count; i++) {
422 			a6xx_get_debugbus_block(gpu,
423 				a6xx_state, &a7xx_debugbus_blocks[gbif_debugbus_blocks[i]],
424 				&a6xx_state->debugbus[i + debugbus_blocks_count]);
425 		}
426 	}
427 
428 }
429 
430 static void a6xx_get_debugbus(struct msm_gpu *gpu,
431 		struct a6xx_gpu_state *a6xx_state)
432 {
433 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
434 	struct resource *res;
435 	void __iomem *cxdbg = NULL;
436 
437 	/* Set up the GX debug bus */
438 
439 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_CNTLT,
440 		A6XX_DBGC_CFG_DBGBUS_CNTLT_SEGT(0xf));
441 
442 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_CNTLM,
443 		A6XX_DBGC_CFG_DBGBUS_CNTLM_ENABLE(0xf));
444 
445 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_0, 0);
446 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_1, 0);
447 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_2, 0);
448 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_IVTL_3, 0);
449 
450 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_BYTEL_0, 0x76543210);
451 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_BYTEL_1, 0xFEDCBA98);
452 
453 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_0, 0);
454 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_1, 0);
455 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_2, 0);
456 	gpu_write(gpu, REG_A6XX_DBGC_CFG_DBGBUS_MASKL_3, 0);
457 
458 	/* Set up the CX debug bus - it lives elsewhere in the system so do a
459 	 * temporary ioremap for the registers
460 	 */
461 	res = platform_get_resource_byname(gpu->pdev, IORESOURCE_MEM,
462 			"cx_dbgc");
463 
464 	if (res)
465 		cxdbg = ioremap(res->start, resource_size(res));
466 
467 	if (cxdbg) {
468 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_CNTLT,
469 			A6XX_DBGC_CFG_DBGBUS_CNTLT_SEGT(0xf));
470 
471 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_CNTLM,
472 			A6XX_DBGC_CFG_DBGBUS_CNTLM_ENABLE(0xf));
473 
474 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_IVTL_0, 0);
475 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_IVTL_1, 0);
476 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_IVTL_2, 0);
477 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_IVTL_3, 0);
478 
479 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_BYTEL_0,
480 			0x76543210);
481 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_BYTEL_1,
482 			0xFEDCBA98);
483 
484 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_MASKL_0, 0);
485 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_MASKL_1, 0);
486 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_MASKL_2, 0);
487 		cxdbg_write(cxdbg, REG_A6XX_CX_DBGC_CFG_DBGBUS_MASKL_3, 0);
488 	}
489 
490 	if (adreno_is_a7xx(adreno_gpu)) {
491 		a7xx_get_debugbus_blocks(gpu, a6xx_state);
492 	} else {
493 		a6xx_get_debugbus_blocks(gpu, a6xx_state);
494 	}
495 
496 	/*  Dump the VBIF debugbus on applicable targets */
497 	if (!a6xx_has_gbif(adreno_gpu)) {
498 		a6xx_state->vbif_debugbus =
499 			state_kcalloc(a6xx_state, 1,
500 					sizeof(*a6xx_state->vbif_debugbus));
501 
502 		if (a6xx_state->vbif_debugbus)
503 			a6xx_get_vbif_debugbus_block(gpu, a6xx_state,
504 					a6xx_state->vbif_debugbus);
505 	}
506 
507 	if (cxdbg) {
508 		unsigned nr_cx_debugbus_blocks;
509 		const struct a6xx_debugbus_block *cx_debugbus_blocks;
510 
511 		if (adreno_is_a7xx(adreno_gpu)) {
512 			BUG_ON(adreno_gpu->info->family > ADRENO_7XX_GEN3);
513 			cx_debugbus_blocks = a7xx_cx_debugbus_blocks;
514 			nr_cx_debugbus_blocks = ARRAY_SIZE(a7xx_cx_debugbus_blocks);
515 		} else {
516 			cx_debugbus_blocks = a6xx_cx_debugbus_blocks;
517 			nr_cx_debugbus_blocks = ARRAY_SIZE(a6xx_cx_debugbus_blocks);
518 		}
519 
520 		a6xx_state->cx_debugbus =
521 			state_kcalloc(a6xx_state,
522 			nr_cx_debugbus_blocks,
523 			sizeof(*a6xx_state->cx_debugbus));
524 
525 		if (a6xx_state->cx_debugbus) {
526 			int i;
527 
528 			for (i = 0; i < nr_cx_debugbus_blocks; i++)
529 				a6xx_get_cx_debugbus_block(cxdbg,
530 					a6xx_state,
531 					&cx_debugbus_blocks[i],
532 					&a6xx_state->cx_debugbus[i]);
533 
534 			a6xx_state->nr_cx_debugbus =
535 				nr_cx_debugbus_blocks;
536 		}
537 
538 		iounmap(cxdbg);
539 	}
540 }
541 
542 #define RANGE(reg, a) ((reg)[(a) + 1] - (reg)[(a)] + 1)
543 
544 /* Read a data cluster from behind the AHB aperture */
545 static void a6xx_get_dbgahb_cluster(struct msm_gpu *gpu,
546 		struct a6xx_gpu_state *a6xx_state,
547 		const struct a6xx_dbgahb_cluster *dbgahb,
548 		struct a6xx_gpu_state_obj *obj,
549 		struct a6xx_crashdumper *dumper)
550 {
551 	u64 *in = dumper->ptr;
552 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
553 	size_t datasize;
554 	int i, regcount = 0;
555 
556 	for (i = 0; i < A6XX_NUM_CONTEXTS; i++) {
557 		int j;
558 
559 		in += CRASHDUMP_WRITE(in, REG_A6XX_HLSQ_DBG_READ_SEL,
560 			(dbgahb->statetype + i * 2) << 8);
561 
562 		for (j = 0; j < dbgahb->count; j += 2) {
563 			int count = RANGE(dbgahb->registers, j);
564 			u32 offset = REG_A6XX_HLSQ_DBG_AHB_READ_APERTURE +
565 				dbgahb->registers[j] - (dbgahb->base >> 2);
566 
567 			in += CRASHDUMP_READ(in, offset, count, out);
568 
569 			out += count * sizeof(u32);
570 
571 			if (i == 0)
572 				regcount += count;
573 		}
574 	}
575 
576 	CRASHDUMP_FINI(in);
577 
578 	datasize = regcount * A6XX_NUM_CONTEXTS * sizeof(u32);
579 
580 	if (WARN_ON(datasize > A6XX_CD_DATA_SIZE))
581 		return;
582 
583 	if (a6xx_crashdumper_run(gpu, dumper))
584 		return;
585 
586 	obj->handle = dbgahb;
587 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
588 		datasize);
589 }
590 
591 static void a7xx_get_dbgahb_cluster(struct msm_gpu *gpu,
592 		struct a6xx_gpu_state *a6xx_state,
593 		const struct gen7_sptp_cluster_registers *dbgahb,
594 		struct a6xx_gpu_state_obj *obj,
595 		struct a6xx_crashdumper *dumper)
596 {
597 	u64 *in = dumper->ptr;
598 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
599 	size_t datasize;
600 	int i, regcount = 0;
601 
602 	in += CRASHDUMP_WRITE(in, REG_A7XX_SP_READ_SEL,
603 		A7XX_SP_READ_SEL_LOCATION(dbgahb->location_id) |
604 		A7XX_SP_READ_SEL_PIPE(dbgahb->pipe_id) |
605 		A7XX_SP_READ_SEL_STATETYPE(dbgahb->statetype));
606 
607 	for (i = 0; dbgahb->regs[i] != UINT_MAX; i += 2) {
608 		int count = RANGE(dbgahb->regs, i);
609 		u32 offset = REG_A7XX_SP_AHB_READ_APERTURE +
610 			dbgahb->regs[i] - dbgahb->regbase;
611 
612 		in += CRASHDUMP_READ(in, offset, count, out);
613 
614 		out += count * sizeof(u32);
615 		regcount += count;
616 	}
617 
618 	CRASHDUMP_FINI(in);
619 
620 	datasize = regcount * sizeof(u32);
621 
622 	if (WARN_ON(datasize > A6XX_CD_DATA_SIZE))
623 		return;
624 
625 	if (a6xx_crashdumper_run(gpu, dumper))
626 		return;
627 
628 	obj->handle = dbgahb;
629 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
630 		datasize);
631 }
632 
633 static void a6xx_get_dbgahb_clusters(struct msm_gpu *gpu,
634 		struct a6xx_gpu_state *a6xx_state,
635 		struct a6xx_crashdumper *dumper)
636 {
637 	int i;
638 
639 	a6xx_state->dbgahb_clusters = state_kcalloc(a6xx_state,
640 		ARRAY_SIZE(a6xx_dbgahb_clusters),
641 		sizeof(*a6xx_state->dbgahb_clusters));
642 
643 	if (!a6xx_state->dbgahb_clusters)
644 		return;
645 
646 	a6xx_state->nr_dbgahb_clusters = ARRAY_SIZE(a6xx_dbgahb_clusters);
647 
648 	for (i = 0; i < ARRAY_SIZE(a6xx_dbgahb_clusters); i++)
649 		a6xx_get_dbgahb_cluster(gpu, a6xx_state,
650 			&a6xx_dbgahb_clusters[i],
651 			&a6xx_state->dbgahb_clusters[i], dumper);
652 }
653 
654 static void a7xx_get_dbgahb_clusters(struct msm_gpu *gpu,
655 		struct a6xx_gpu_state *a6xx_state,
656 		struct a6xx_crashdumper *dumper)
657 {
658 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
659 	int i;
660 	const struct gen7_sptp_cluster_registers *dbgahb_clusters;
661 	unsigned dbgahb_clusters_size;
662 
663 	if (adreno_gpu->info->family == ADRENO_7XX_GEN1) {
664 		dbgahb_clusters = gen7_0_0_sptp_clusters;
665 		dbgahb_clusters_size = ARRAY_SIZE(gen7_0_0_sptp_clusters);
666 	} else if (adreno_gpu->info->family == ADRENO_7XX_GEN2) {
667 		dbgahb_clusters = gen7_2_0_sptp_clusters;
668 		dbgahb_clusters_size = ARRAY_SIZE(gen7_2_0_sptp_clusters);
669 	} else {
670 		BUG_ON(adreno_gpu->info->family != ADRENO_7XX_GEN3);
671 		dbgahb_clusters = gen7_9_0_sptp_clusters;
672 		dbgahb_clusters_size = ARRAY_SIZE(gen7_9_0_sptp_clusters);
673 	}
674 
675 	a6xx_state->dbgahb_clusters = state_kcalloc(a6xx_state,
676 		dbgahb_clusters_size,
677 		sizeof(*a6xx_state->dbgahb_clusters));
678 
679 	if (!a6xx_state->dbgahb_clusters)
680 		return;
681 
682 	a6xx_state->nr_dbgahb_clusters = dbgahb_clusters_size;
683 
684 	for (i = 0; i < dbgahb_clusters_size; i++)
685 		a7xx_get_dbgahb_cluster(gpu, a6xx_state,
686 			&dbgahb_clusters[i],
687 			&a6xx_state->dbgahb_clusters[i], dumper);
688 }
689 
690 /* Read a data cluster from the CP aperture with the crashdumper */
691 static void a6xx_get_cluster(struct msm_gpu *gpu,
692 		struct a6xx_gpu_state *a6xx_state,
693 		const struct a6xx_cluster *cluster,
694 		struct a6xx_gpu_state_obj *obj,
695 		struct a6xx_crashdumper *dumper)
696 {
697 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
698 	u64 *in = dumper->ptr;
699 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
700 	size_t datasize;
701 	int i, regcount = 0;
702 	u32 id = cluster->id;
703 
704 	/* Skip registers that are not present on older generation */
705 	if (!adreno_is_a660_family(adreno_gpu) &&
706 			cluster->registers == a660_fe_cluster)
707 		return;
708 
709 	if (adreno_is_a650_family(adreno_gpu) &&
710 			cluster->registers == a6xx_ps_cluster)
711 		id = CLUSTER_VPC_PS;
712 
713 	/* Some clusters need a selector register to be programmed too */
714 	if (cluster->sel_reg)
715 		in += CRASHDUMP_WRITE(in, cluster->sel_reg, cluster->sel_val);
716 
717 	for (i = 0; i < A6XX_NUM_CONTEXTS; i++) {
718 		int j;
719 
720 		in += CRASHDUMP_WRITE(in, REG_A6XX_CP_APERTURE_CNTL_CD,
721 			(id << 8) | (i << 4) | i);
722 
723 		for (j = 0; j < cluster->count; j += 2) {
724 			int count = RANGE(cluster->registers, j);
725 
726 			in += CRASHDUMP_READ(in, cluster->registers[j],
727 				count, out);
728 
729 			out += count * sizeof(u32);
730 
731 			if (i == 0)
732 				regcount += count;
733 		}
734 	}
735 
736 	CRASHDUMP_FINI(in);
737 
738 	datasize = regcount * A6XX_NUM_CONTEXTS * sizeof(u32);
739 
740 	if (WARN_ON(datasize > A6XX_CD_DATA_SIZE))
741 		return;
742 
743 	if (a6xx_crashdumper_run(gpu, dumper))
744 		return;
745 
746 	obj->handle = cluster;
747 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
748 		datasize);
749 }
750 
751 static void a7xx_get_cluster(struct msm_gpu *gpu,
752 		struct a6xx_gpu_state *a6xx_state,
753 		const struct gen7_cluster_registers *cluster,
754 		struct a6xx_gpu_state_obj *obj,
755 		struct a6xx_crashdumper *dumper)
756 {
757 	u64 *in = dumper->ptr;
758 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
759 	size_t datasize;
760 	int i, regcount = 0;
761 
762 	/* Some clusters need a selector register to be programmed too */
763 	if (cluster->sel)
764 		in += CRASHDUMP_WRITE(in, cluster->sel->cd_reg, cluster->sel->val);
765 
766 	in += CRASHDUMP_WRITE(in, REG_A7XX_CP_APERTURE_CNTL_CD,
767 		A7XX_CP_APERTURE_CNTL_CD_PIPE(cluster->pipe_id) |
768 		A7XX_CP_APERTURE_CNTL_CD_CLUSTER(cluster->cluster_id) |
769 		A7XX_CP_APERTURE_CNTL_CD_CONTEXT(cluster->context_id));
770 
771 	for (i = 0; cluster->regs[i] != UINT_MAX; i += 2) {
772 		int count = RANGE(cluster->regs, i);
773 
774 		in += CRASHDUMP_READ(in, cluster->regs[i],
775 			count, out);
776 
777 		out += count * sizeof(u32);
778 		regcount += count;
779 	}
780 
781 	CRASHDUMP_FINI(in);
782 
783 	datasize = regcount * sizeof(u32);
784 
785 	if (WARN_ON(datasize > A6XX_CD_DATA_SIZE))
786 		return;
787 
788 	if (a6xx_crashdumper_run(gpu, dumper))
789 		return;
790 
791 	obj->handle = cluster;
792 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
793 		datasize);
794 }
795 
796 static void a6xx_get_clusters(struct msm_gpu *gpu,
797 		struct a6xx_gpu_state *a6xx_state,
798 		struct a6xx_crashdumper *dumper)
799 {
800 	int i;
801 
802 	a6xx_state->clusters = state_kcalloc(a6xx_state,
803 		ARRAY_SIZE(a6xx_clusters), sizeof(*a6xx_state->clusters));
804 
805 	if (!a6xx_state->clusters)
806 		return;
807 
808 	a6xx_state->nr_clusters = ARRAY_SIZE(a6xx_clusters);
809 
810 	for (i = 0; i < ARRAY_SIZE(a6xx_clusters); i++)
811 		a6xx_get_cluster(gpu, a6xx_state, &a6xx_clusters[i],
812 			&a6xx_state->clusters[i], dumper);
813 }
814 
815 static void a7xx_get_clusters(struct msm_gpu *gpu,
816 		struct a6xx_gpu_state *a6xx_state,
817 		struct a6xx_crashdumper *dumper)
818 {
819 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
820 	int i;
821 	const struct gen7_cluster_registers *clusters;
822 	unsigned clusters_size;
823 
824 	if (adreno_gpu->info->family == ADRENO_7XX_GEN1) {
825 		clusters = gen7_0_0_clusters;
826 		clusters_size = ARRAY_SIZE(gen7_0_0_clusters);
827 	} else if (adreno_gpu->info->family == ADRENO_7XX_GEN2) {
828 		clusters = gen7_2_0_clusters;
829 		clusters_size = ARRAY_SIZE(gen7_2_0_clusters);
830 	} else {
831 		BUG_ON(adreno_gpu->info->family != ADRENO_7XX_GEN3);
832 		clusters = gen7_9_0_clusters;
833 		clusters_size = ARRAY_SIZE(gen7_9_0_clusters);
834 	}
835 
836 	a6xx_state->clusters = state_kcalloc(a6xx_state,
837 		clusters_size, sizeof(*a6xx_state->clusters));
838 
839 	if (!a6xx_state->clusters)
840 		return;
841 
842 	a6xx_state->nr_clusters = clusters_size;
843 
844 	for (i = 0; i < clusters_size; i++)
845 		a7xx_get_cluster(gpu, a6xx_state, &clusters[i],
846 			&a6xx_state->clusters[i], dumper);
847 }
848 
849 /* Read a shader / debug block from the HLSQ aperture with the crashdumper */
850 static void a6xx_get_shader_block(struct msm_gpu *gpu,
851 		struct a6xx_gpu_state *a6xx_state,
852 		const struct a6xx_shader_block *block,
853 		struct a6xx_gpu_state_obj *obj,
854 		struct a6xx_crashdumper *dumper)
855 {
856 	u64 *in = dumper->ptr;
857 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
858 	size_t datasize = block->size * A6XX_NUM_SHADER_BANKS * sizeof(u32);
859 	int i;
860 
861 	if (WARN_ON(datasize > A6XX_CD_DATA_SIZE))
862 		return;
863 
864 	for (i = 0; i < A6XX_NUM_SHADER_BANKS; i++) {
865 		in += CRASHDUMP_WRITE(in, REG_A6XX_HLSQ_DBG_READ_SEL,
866 			(block->type << 8) | i);
867 
868 		in += CRASHDUMP_READ(in, REG_A6XX_HLSQ_DBG_AHB_READ_APERTURE,
869 			block->size, out);
870 
871 		out += block->size * sizeof(u32);
872 	}
873 
874 	CRASHDUMP_FINI(in);
875 
876 	if (a6xx_crashdumper_run(gpu, dumper))
877 		return;
878 
879 	obj->handle = block;
880 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
881 		datasize);
882 }
883 
884 static void a7xx_get_shader_block(struct msm_gpu *gpu,
885 		struct a6xx_gpu_state *a6xx_state,
886 		const struct gen7_shader_block *block,
887 		struct a6xx_gpu_state_obj *obj,
888 		struct a6xx_crashdumper *dumper)
889 {
890 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
891 	u64 *in = dumper->ptr;
892 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
893 	size_t datasize = block->size * block->num_sps * block->num_usptps * sizeof(u32);
894 	int i, j;
895 
896 	if (WARN_ON(datasize > A6XX_CD_DATA_SIZE))
897 		return;
898 
899 	if (adreno_gpu->info->family == ADRENO_7XX_GEN1) {
900 		gpu_rmw(gpu, REG_A7XX_SP_DBG_CNTL, GENMASK(1, 0), 3);
901 	}
902 
903 	for (i = 0; i < block->num_sps; i++) {
904 		for (j = 0; j < block->num_usptps; j++) {
905 			in += CRASHDUMP_WRITE(in, REG_A7XX_SP_READ_SEL,
906 				A7XX_SP_READ_SEL_LOCATION(block->location) |
907 				A7XX_SP_READ_SEL_PIPE(block->pipeid) |
908 				A7XX_SP_READ_SEL_STATETYPE(block->statetype) |
909 				A7XX_SP_READ_SEL_USPTP(j) |
910 				A7XX_SP_READ_SEL_SPTP(i));
911 
912 			in += CRASHDUMP_READ(in, REG_A7XX_SP_AHB_READ_APERTURE,
913 				block->size, out);
914 
915 			out += block->size * sizeof(u32);
916 		}
917 	}
918 
919 	CRASHDUMP_FINI(in);
920 
921 	if (a6xx_crashdumper_run(gpu, dumper))
922 		goto out;
923 
924 	obj->handle = block;
925 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
926 		datasize);
927 
928 out:
929 	if (adreno_gpu->info->family == ADRENO_7XX_GEN1) {
930 		gpu_rmw(gpu, REG_A7XX_SP_DBG_CNTL, GENMASK(1, 0), 0);
931 	}
932 }
933 
934 static void a6xx_get_shaders(struct msm_gpu *gpu,
935 		struct a6xx_gpu_state *a6xx_state,
936 		struct a6xx_crashdumper *dumper)
937 {
938 	int i;
939 
940 	a6xx_state->shaders = state_kcalloc(a6xx_state,
941 		ARRAY_SIZE(a6xx_shader_blocks), sizeof(*a6xx_state->shaders));
942 
943 	if (!a6xx_state->shaders)
944 		return;
945 
946 	a6xx_state->nr_shaders = ARRAY_SIZE(a6xx_shader_blocks);
947 
948 	for (i = 0; i < ARRAY_SIZE(a6xx_shader_blocks); i++)
949 		a6xx_get_shader_block(gpu, a6xx_state, &a6xx_shader_blocks[i],
950 			&a6xx_state->shaders[i], dumper);
951 }
952 
953 static void a7xx_get_shaders(struct msm_gpu *gpu,
954 		struct a6xx_gpu_state *a6xx_state,
955 		struct a6xx_crashdumper *dumper)
956 {
957 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
958 	const struct gen7_shader_block *shader_blocks;
959 	unsigned num_shader_blocks;
960 	int i;
961 
962 	if (adreno_gpu->info->family == ADRENO_7XX_GEN1) {
963 		shader_blocks = gen7_0_0_shader_blocks;
964 		num_shader_blocks = ARRAY_SIZE(gen7_0_0_shader_blocks);
965 	} else if (adreno_gpu->info->family == ADRENO_7XX_GEN2) {
966 		shader_blocks = gen7_2_0_shader_blocks;
967 		num_shader_blocks = ARRAY_SIZE(gen7_2_0_shader_blocks);
968 	} else {
969 		BUG_ON(adreno_gpu->info->family != ADRENO_7XX_GEN3);
970 		shader_blocks = gen7_9_0_shader_blocks;
971 		num_shader_blocks = ARRAY_SIZE(gen7_9_0_shader_blocks);
972 	}
973 
974 	a6xx_state->shaders = state_kcalloc(a6xx_state,
975 		num_shader_blocks, sizeof(*a6xx_state->shaders));
976 
977 	if (!a6xx_state->shaders)
978 		return;
979 
980 	a6xx_state->nr_shaders = num_shader_blocks;
981 
982 	for (i = 0; i < num_shader_blocks; i++)
983 		a7xx_get_shader_block(gpu, a6xx_state, &shader_blocks[i],
984 			&a6xx_state->shaders[i], dumper);
985 }
986 
987 /* Read registers from behind the HLSQ aperture with the crashdumper */
988 static void a6xx_get_crashdumper_hlsq_registers(struct msm_gpu *gpu,
989 		struct a6xx_gpu_state *a6xx_state,
990 		const struct a6xx_registers *regs,
991 		struct a6xx_gpu_state_obj *obj,
992 		struct a6xx_crashdumper *dumper)
993 
994 {
995 	u64 *in = dumper->ptr;
996 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
997 	int i, regcount = 0;
998 
999 	in += CRASHDUMP_WRITE(in, REG_A6XX_HLSQ_DBG_READ_SEL, regs->val1);
1000 
1001 	for (i = 0; i < regs->count; i += 2) {
1002 		u32 count = RANGE(regs->registers, i);
1003 		u32 offset = REG_A6XX_HLSQ_DBG_AHB_READ_APERTURE +
1004 			regs->registers[i] - (regs->val0 >> 2);
1005 
1006 		in += CRASHDUMP_READ(in, offset, count, out);
1007 
1008 		out += count * sizeof(u32);
1009 		regcount += count;
1010 	}
1011 
1012 	CRASHDUMP_FINI(in);
1013 
1014 	if (WARN_ON((regcount * sizeof(u32)) > A6XX_CD_DATA_SIZE))
1015 		return;
1016 
1017 	if (a6xx_crashdumper_run(gpu, dumper))
1018 		return;
1019 
1020 	obj->handle = regs;
1021 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
1022 		regcount * sizeof(u32));
1023 }
1024 
1025 /* Read a block of registers using the crashdumper */
1026 static void a6xx_get_crashdumper_registers(struct msm_gpu *gpu,
1027 		struct a6xx_gpu_state *a6xx_state,
1028 		const struct a6xx_registers *regs,
1029 		struct a6xx_gpu_state_obj *obj,
1030 		struct a6xx_crashdumper *dumper)
1031 
1032 {
1033 	u64 *in = dumper->ptr;
1034 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
1035 	int i, regcount = 0;
1036 
1037 	/* Skip unsupported registers on older generations */
1038 	if (!adreno_is_a660_family(to_adreno_gpu(gpu)) &&
1039 			(regs->registers == a660_registers))
1040 		return;
1041 
1042 	/* Some blocks might need to program a selector register first */
1043 	if (regs->val0)
1044 		in += CRASHDUMP_WRITE(in, regs->val0, regs->val1);
1045 
1046 	for (i = 0; i < regs->count; i += 2) {
1047 		u32 count = RANGE(regs->registers, i);
1048 
1049 		in += CRASHDUMP_READ(in, regs->registers[i], count, out);
1050 
1051 		out += count * sizeof(u32);
1052 		regcount += count;
1053 	}
1054 
1055 	CRASHDUMP_FINI(in);
1056 
1057 	if (WARN_ON((regcount * sizeof(u32)) > A6XX_CD_DATA_SIZE))
1058 		return;
1059 
1060 	if (a6xx_crashdumper_run(gpu, dumper))
1061 		return;
1062 
1063 	obj->handle = regs;
1064 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
1065 		regcount * sizeof(u32));
1066 }
1067 
1068 static void a7xx_get_crashdumper_registers(struct msm_gpu *gpu,
1069 		struct a6xx_gpu_state *a6xx_state,
1070 		const struct gen7_reg_list *regs,
1071 		struct a6xx_gpu_state_obj *obj,
1072 		struct a6xx_crashdumper *dumper)
1073 
1074 {
1075 	u64 *in = dumper->ptr;
1076 	u64 out = dumper->iova + A6XX_CD_DATA_OFFSET;
1077 	int i, regcount = 0;
1078 
1079 	/* Some blocks might need to program a selector register first */
1080 	if (regs->sel)
1081 		in += CRASHDUMP_WRITE(in, regs->sel->cd_reg, regs->sel->val);
1082 
1083 	for (i = 0; regs->regs[i] != UINT_MAX; i += 2) {
1084 		u32 count = RANGE(regs->regs, i);
1085 
1086 		in += CRASHDUMP_READ(in, regs->regs[i], count, out);
1087 
1088 		out += count * sizeof(u32);
1089 		regcount += count;
1090 	}
1091 
1092 	CRASHDUMP_FINI(in);
1093 
1094 	if (WARN_ON((regcount * sizeof(u32)) > A6XX_CD_DATA_SIZE))
1095 		return;
1096 
1097 	if (a6xx_crashdumper_run(gpu, dumper))
1098 		return;
1099 
1100 	obj->handle = regs->regs;
1101 	obj->data = state_kmemdup(a6xx_state, dumper->ptr + A6XX_CD_DATA_OFFSET,
1102 		regcount * sizeof(u32));
1103 }
1104 
1105 
1106 /* Read a block of registers via AHB */
1107 static void a6xx_get_ahb_gpu_registers(struct msm_gpu *gpu,
1108 		struct a6xx_gpu_state *a6xx_state,
1109 		const struct a6xx_registers *regs,
1110 		struct a6xx_gpu_state_obj *obj)
1111 {
1112 	int i, regcount = 0, index = 0;
1113 
1114 	/* Skip unsupported registers on older generations */
1115 	if (!adreno_is_a660_family(to_adreno_gpu(gpu)) &&
1116 			(regs->registers == a660_registers))
1117 		return;
1118 
1119 	for (i = 0; i < regs->count; i += 2)
1120 		regcount += RANGE(regs->registers, i);
1121 
1122 	obj->handle = (const void *) regs;
1123 	obj->data = state_kcalloc(a6xx_state, regcount, sizeof(u32));
1124 	if (!obj->data)
1125 		return;
1126 
1127 	for (i = 0; i < regs->count; i += 2) {
1128 		u32 count = RANGE(regs->registers, i);
1129 		int j;
1130 
1131 		for (j = 0; j < count; j++)
1132 			obj->data[index++] = gpu_read(gpu,
1133 				regs->registers[i] + j);
1134 	}
1135 }
1136 
1137 static void a7xx_get_ahb_gpu_registers(struct msm_gpu *gpu,
1138 		struct a6xx_gpu_state *a6xx_state,
1139 		const u32 *regs,
1140 		struct a6xx_gpu_state_obj *obj)
1141 {
1142 	int i, regcount = 0, index = 0;
1143 
1144 	for (i = 0; regs[i] != UINT_MAX; i += 2)
1145 		regcount += RANGE(regs, i);
1146 
1147 	obj->handle = (const void *) regs;
1148 	obj->data = state_kcalloc(a6xx_state, regcount, sizeof(u32));
1149 	if (!obj->data)
1150 		return;
1151 
1152 	for (i = 0; regs[i] != UINT_MAX; i += 2) {
1153 		u32 count = RANGE(regs, i);
1154 		int j;
1155 
1156 		for (j = 0; j < count; j++)
1157 			obj->data[index++] = gpu_read(gpu, regs[i] + j);
1158 	}
1159 }
1160 
1161 static void a7xx_get_ahb_gpu_reglist(struct msm_gpu *gpu,
1162 		struct a6xx_gpu_state *a6xx_state,
1163 		const struct gen7_reg_list *regs,
1164 		struct a6xx_gpu_state_obj *obj)
1165 {
1166 	if (regs->sel)
1167 		gpu_write(gpu, regs->sel->host_reg, regs->sel->val);
1168 
1169 	a7xx_get_ahb_gpu_registers(gpu, a6xx_state, regs->regs, obj);
1170 }
1171 
1172 /* Read a block of GMU registers */
1173 static void _a6xx_get_gmu_registers(struct msm_gpu *gpu,
1174 		struct a6xx_gpu_state *a6xx_state,
1175 		const struct a6xx_registers *regs,
1176 		struct a6xx_gpu_state_obj *obj,
1177 		bool rscc)
1178 {
1179 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1180 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1181 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
1182 	int i, regcount = 0, index = 0;
1183 
1184 	for (i = 0; i < regs->count; i += 2)
1185 		regcount += RANGE(regs->registers, i);
1186 
1187 	obj->handle = (const void *) regs;
1188 	obj->data = state_kcalloc(a6xx_state, regcount, sizeof(u32));
1189 	if (!obj->data)
1190 		return;
1191 
1192 	for (i = 0; i < regs->count; i += 2) {
1193 		u32 count = RANGE(regs->registers, i);
1194 		int j;
1195 
1196 		for (j = 0; j < count; j++) {
1197 			u32 offset = regs->registers[i] + j;
1198 			u32 val;
1199 
1200 			if (rscc)
1201 				val = gmu_read_rscc(gmu, offset);
1202 			else
1203 				val = gmu_read(gmu, offset);
1204 
1205 			obj->data[index++] = val;
1206 		}
1207 	}
1208 }
1209 
1210 static void a6xx_get_gmu_registers(struct msm_gpu *gpu,
1211 		struct a6xx_gpu_state *a6xx_state)
1212 {
1213 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1214 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1215 
1216 	a6xx_state->gmu_registers = state_kcalloc(a6xx_state,
1217 		3, sizeof(*a6xx_state->gmu_registers));
1218 
1219 	if (!a6xx_state->gmu_registers)
1220 		return;
1221 
1222 	a6xx_state->nr_gmu_registers = 3;
1223 
1224 	/* Get the CX GMU registers from AHB */
1225 	_a6xx_get_gmu_registers(gpu, a6xx_state, &a6xx_gmu_reglist[0],
1226 		&a6xx_state->gmu_registers[0], false);
1227 	_a6xx_get_gmu_registers(gpu, a6xx_state, &a6xx_gmu_reglist[1],
1228 		&a6xx_state->gmu_registers[1], true);
1229 
1230 	if (!a6xx_gmu_gx_is_on(&a6xx_gpu->gmu))
1231 		return;
1232 
1233 	/* Set the fence to ALLOW mode so we can access the registers */
1234 	gpu_write(gpu, REG_A6XX_GMU_AO_AHB_FENCE_CTRL, 0);
1235 
1236 	_a6xx_get_gmu_registers(gpu, a6xx_state, &a6xx_gmu_reglist[2],
1237 		&a6xx_state->gmu_registers[2], false);
1238 }
1239 
1240 static struct msm_gpu_state_bo *a6xx_snapshot_gmu_bo(
1241 		struct a6xx_gpu_state *a6xx_state, struct a6xx_gmu_bo *bo)
1242 {
1243 	struct msm_gpu_state_bo *snapshot;
1244 
1245 	if (!bo->size)
1246 		return NULL;
1247 
1248 	snapshot = state_kcalloc(a6xx_state, 1, sizeof(*snapshot));
1249 	if (!snapshot)
1250 		return NULL;
1251 
1252 	snapshot->iova = bo->iova;
1253 	snapshot->size = bo->size;
1254 	snapshot->data = kvzalloc(snapshot->size, GFP_KERNEL);
1255 	if (!snapshot->data)
1256 		return NULL;
1257 
1258 	memcpy(snapshot->data, bo->virt, bo->size);
1259 
1260 	return snapshot;
1261 }
1262 
1263 static void a6xx_snapshot_gmu_hfi_history(struct msm_gpu *gpu,
1264 					  struct a6xx_gpu_state *a6xx_state)
1265 {
1266 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1267 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1268 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
1269 	unsigned i, j;
1270 
1271 	BUILD_BUG_ON(ARRAY_SIZE(gmu->queues) != ARRAY_SIZE(a6xx_state->hfi_queue_history));
1272 
1273 	for (i = 0; i < ARRAY_SIZE(gmu->queues); i++) {
1274 		struct a6xx_hfi_queue *queue = &gmu->queues[i];
1275 		for (j = 0; j < HFI_HISTORY_SZ; j++) {
1276 			unsigned idx = (j + queue->history_idx) % HFI_HISTORY_SZ;
1277 			a6xx_state->hfi_queue_history[i][j] = queue->history[idx];
1278 		}
1279 	}
1280 }
1281 
1282 #define A6XX_REGLIST_SIZE        1
1283 #define A6XX_GBIF_REGLIST_SIZE   1
1284 static void a6xx_get_registers(struct msm_gpu *gpu,
1285 		struct a6xx_gpu_state *a6xx_state,
1286 		struct a6xx_crashdumper *dumper)
1287 {
1288 	int i, count = A6XX_REGLIST_SIZE +
1289 		ARRAY_SIZE(a6xx_reglist) +
1290 		ARRAY_SIZE(a6xx_hlsq_reglist) + A6XX_GBIF_REGLIST_SIZE;
1291 	int index = 0;
1292 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1293 
1294 	a6xx_state->registers = state_kcalloc(a6xx_state,
1295 		count, sizeof(*a6xx_state->registers));
1296 
1297 	if (!a6xx_state->registers)
1298 		return;
1299 
1300 	a6xx_state->nr_registers = count;
1301 
1302 	a6xx_get_ahb_gpu_registers(gpu,
1303 		a6xx_state, &a6xx_ahb_reglist,
1304 		&a6xx_state->registers[index++]);
1305 
1306 	if (a6xx_has_gbif(adreno_gpu))
1307 		a6xx_get_ahb_gpu_registers(gpu,
1308 				a6xx_state, &a6xx_gbif_reglist,
1309 				&a6xx_state->registers[index++]);
1310 	else
1311 		a6xx_get_ahb_gpu_registers(gpu,
1312 				a6xx_state, &a6xx_vbif_reglist,
1313 				&a6xx_state->registers[index++]);
1314 	if (!dumper) {
1315 		/*
1316 		 * We can't use the crashdumper when the SMMU is stalled,
1317 		 * because the GPU has no memory access until we resume
1318 		 * translation (but we don't want to do that until after
1319 		 * we have captured as much useful GPU state as possible).
1320 		 * So instead collect registers via the CPU:
1321 		 */
1322 		for (i = 0; i < ARRAY_SIZE(a6xx_reglist); i++)
1323 			a6xx_get_ahb_gpu_registers(gpu,
1324 				a6xx_state, &a6xx_reglist[i],
1325 				&a6xx_state->registers[index++]);
1326 		return;
1327 	}
1328 
1329 	for (i = 0; i < ARRAY_SIZE(a6xx_reglist); i++)
1330 		a6xx_get_crashdumper_registers(gpu,
1331 			a6xx_state, &a6xx_reglist[i],
1332 			&a6xx_state->registers[index++],
1333 			dumper);
1334 
1335 	for (i = 0; i < ARRAY_SIZE(a6xx_hlsq_reglist); i++)
1336 		a6xx_get_crashdumper_hlsq_registers(gpu,
1337 			a6xx_state, &a6xx_hlsq_reglist[i],
1338 			&a6xx_state->registers[index++],
1339 			dumper);
1340 }
1341 
1342 #define A7XX_PRE_CRASHDUMPER_SIZE    1
1343 #define A7XX_POST_CRASHDUMPER_SIZE   1
1344 static void a7xx_get_registers(struct msm_gpu *gpu,
1345 		struct a6xx_gpu_state *a6xx_state,
1346 		struct a6xx_crashdumper *dumper)
1347 {
1348 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1349 	int i, count;
1350 	int index = 0;
1351 	const u32 *pre_crashdumper_regs;
1352 	const struct gen7_reg_list *reglist;
1353 
1354 	if (adreno_gpu->info->family == ADRENO_7XX_GEN1) {
1355 		reglist = gen7_0_0_reg_list;
1356 		pre_crashdumper_regs = gen7_0_0_pre_crashdumper_gpu_registers;
1357 	} else if (adreno_gpu->info->family == ADRENO_7XX_GEN2) {
1358 		reglist = gen7_2_0_reg_list;
1359 		pre_crashdumper_regs = gen7_0_0_pre_crashdumper_gpu_registers;
1360 	} else {
1361 		BUG_ON(adreno_gpu->info->family != ADRENO_7XX_GEN3);
1362 		reglist = gen7_9_0_reg_list;
1363 		pre_crashdumper_regs = gen7_9_0_pre_crashdumper_gpu_registers;
1364 	}
1365 
1366 	count = A7XX_PRE_CRASHDUMPER_SIZE + A7XX_POST_CRASHDUMPER_SIZE;
1367 
1368 	/* The downstream reglist contains registers in other memory regions
1369 	 * (cx_misc/cx_mem and cx_dbgc) and we need to plumb through their
1370 	 * offsets and map them to read them on the CPU. For now only read the
1371 	 * first region which is the main one.
1372 	 */
1373 	if (dumper) {
1374 		for (i = 0; reglist[i].regs; i++)
1375 			count++;
1376 	} else {
1377 		count++;
1378 	}
1379 
1380 	a6xx_state->registers = state_kcalloc(a6xx_state,
1381 		count, sizeof(*a6xx_state->registers));
1382 
1383 	if (!a6xx_state->registers)
1384 		return;
1385 
1386 	a6xx_state->nr_registers = count;
1387 
1388 	a7xx_get_ahb_gpu_registers(gpu, a6xx_state, pre_crashdumper_regs,
1389 		&a6xx_state->registers[index++]);
1390 
1391 	if (!dumper) {
1392 		a7xx_get_ahb_gpu_reglist(gpu,
1393 			a6xx_state, &reglist[0],
1394 			&a6xx_state->registers[index++]);
1395 		return;
1396 	}
1397 
1398 	for (i = 0; reglist[i].regs; i++)
1399 		a7xx_get_crashdumper_registers(gpu,
1400 			a6xx_state, &reglist[i],
1401 			&a6xx_state->registers[index++],
1402 			dumper);
1403 }
1404 
1405 static void a7xx_get_post_crashdumper_registers(struct msm_gpu *gpu,
1406 		struct a6xx_gpu_state *a6xx_state)
1407 {
1408 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1409 	const u32 *regs;
1410 
1411 	BUG_ON(adreno_gpu->info->family > ADRENO_7XX_GEN3);
1412 	regs = gen7_0_0_post_crashdumper_registers;
1413 
1414 	a7xx_get_ahb_gpu_registers(gpu,
1415 		a6xx_state, regs,
1416 		&a6xx_state->registers[a6xx_state->nr_registers - 1]);
1417 }
1418 
1419 static u32 a6xx_get_cp_roq_size(struct msm_gpu *gpu)
1420 {
1421 	/* The value at [16:31] is in 4dword units. Convert it to dwords */
1422 	return gpu_read(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2) >> 14;
1423 }
1424 
1425 static u32 a7xx_get_cp_roq_size(struct msm_gpu *gpu)
1426 {
1427 	/*
1428 	 * The value at CP_ROQ_THRESHOLDS_2[20:31] is in 4dword units.
1429 	 * That register however is not directly accessible from APSS on A7xx.
1430 	 * Program the SQE_UCODE_DBG_ADDR with offset=0x70d3 and read the value.
1431 	 */
1432 	gpu_write(gpu, REG_A6XX_CP_SQE_UCODE_DBG_ADDR, 0x70d3);
1433 
1434 	return 4 * (gpu_read(gpu, REG_A6XX_CP_SQE_UCODE_DBG_DATA) >> 20);
1435 }
1436 
1437 /* Read a block of data from an indexed register pair */
1438 static void a6xx_get_indexed_regs(struct msm_gpu *gpu,
1439 		struct a6xx_gpu_state *a6xx_state,
1440 		const struct a6xx_indexed_registers *indexed,
1441 		struct a6xx_gpu_state_obj *obj)
1442 {
1443 	u32 count = indexed->count;
1444 	int i;
1445 
1446 	obj->handle = (const void *) indexed;
1447 	if (indexed->count_fn)
1448 		count = indexed->count_fn(gpu);
1449 
1450 	obj->data = state_kcalloc(a6xx_state, count, sizeof(u32));
1451 	obj->count = count;
1452 	if (!obj->data)
1453 		return;
1454 
1455 	/* All the indexed banks start at address 0 */
1456 	gpu_write(gpu, indexed->addr, 0);
1457 
1458 	/* Read the data - each read increments the internal address by 1 */
1459 	for (i = 0; i < count; i++)
1460 		obj->data[i] = gpu_read(gpu, indexed->data);
1461 }
1462 
1463 static void a6xx_get_indexed_registers(struct msm_gpu *gpu,
1464 		struct a6xx_gpu_state *a6xx_state)
1465 {
1466 	u32 mempool_size;
1467 	int count = ARRAY_SIZE(a6xx_indexed_reglist) + 1;
1468 	int i;
1469 
1470 	a6xx_state->indexed_regs = state_kcalloc(a6xx_state, count,
1471 		sizeof(*a6xx_state->indexed_regs));
1472 	if (!a6xx_state->indexed_regs)
1473 		return;
1474 
1475 	for (i = 0; i < ARRAY_SIZE(a6xx_indexed_reglist); i++)
1476 		a6xx_get_indexed_regs(gpu, a6xx_state, &a6xx_indexed_reglist[i],
1477 			&a6xx_state->indexed_regs[i]);
1478 
1479 	if (adreno_is_a650_family(to_adreno_gpu(gpu))) {
1480 		u32 val;
1481 
1482 		val = gpu_read(gpu, REG_A6XX_CP_CHICKEN_DBG);
1483 		gpu_write(gpu, REG_A6XX_CP_CHICKEN_DBG, val | 4);
1484 
1485 		/* Get the contents of the CP mempool */
1486 		a6xx_get_indexed_regs(gpu, a6xx_state, &a6xx_cp_mempool_indexed,
1487 			&a6xx_state->indexed_regs[i]);
1488 
1489 		gpu_write(gpu, REG_A6XX_CP_CHICKEN_DBG, val);
1490 		a6xx_state->nr_indexed_regs = count;
1491 		return;
1492 	}
1493 
1494 	/* Set the CP mempool size to 0 to stabilize it while dumping */
1495 	mempool_size = gpu_read(gpu, REG_A6XX_CP_MEM_POOL_SIZE);
1496 	gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 0);
1497 
1498 	/* Get the contents of the CP mempool */
1499 	a6xx_get_indexed_regs(gpu, a6xx_state, &a6xx_cp_mempool_indexed,
1500 		&a6xx_state->indexed_regs[i]);
1501 
1502 	/*
1503 	 * Offset 0x2000 in the mempool is the size - copy the saved size over
1504 	 * so the data is consistent
1505 	 */
1506 	a6xx_state->indexed_regs[i].data[0x2000] = mempool_size;
1507 
1508 	/* Restore the size in the hardware */
1509 	gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, mempool_size);
1510 }
1511 
1512 static void a7xx_get_indexed_registers(struct msm_gpu *gpu,
1513 		struct a6xx_gpu_state *a6xx_state)
1514 {
1515 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1516 	const struct a6xx_indexed_registers *indexed_regs;
1517 	int i, indexed_count, mempool_count;
1518 
1519 	if (adreno_gpu->info->family <= ADRENO_7XX_GEN2) {
1520 		indexed_regs = a7xx_indexed_reglist;
1521 		indexed_count = ARRAY_SIZE(a7xx_indexed_reglist);
1522 	} else {
1523 		BUG_ON(adreno_gpu->info->family != ADRENO_7XX_GEN3);
1524 		indexed_regs = gen7_9_0_cp_indexed_reg_list;
1525 		indexed_count = ARRAY_SIZE(gen7_9_0_cp_indexed_reg_list);
1526 	}
1527 
1528 	mempool_count = ARRAY_SIZE(a7xx_cp_bv_mempool_indexed);
1529 
1530 	a6xx_state->indexed_regs = state_kcalloc(a6xx_state,
1531 					indexed_count + mempool_count,
1532 					sizeof(*a6xx_state->indexed_regs));
1533 	if (!a6xx_state->indexed_regs)
1534 		return;
1535 
1536 	a6xx_state->nr_indexed_regs = indexed_count + mempool_count;
1537 
1538 	/* First read the common regs */
1539 	for (i = 0; i < indexed_count; i++)
1540 		a6xx_get_indexed_regs(gpu, a6xx_state, &indexed_regs[i],
1541 			&a6xx_state->indexed_regs[i]);
1542 
1543 	gpu_rmw(gpu, REG_A6XX_CP_CHICKEN_DBG, 0, BIT(2));
1544 	gpu_rmw(gpu, REG_A7XX_CP_BV_CHICKEN_DBG, 0, BIT(2));
1545 
1546 	/* Get the contents of the CP_BV mempool */
1547 	for (i = 0; i < mempool_count; i++)
1548 		a6xx_get_indexed_regs(gpu, a6xx_state, &a7xx_cp_bv_mempool_indexed[i],
1549 			&a6xx_state->indexed_regs[indexed_count + i]);
1550 
1551 	gpu_rmw(gpu, REG_A6XX_CP_CHICKEN_DBG, BIT(2), 0);
1552 	gpu_rmw(gpu, REG_A7XX_CP_BV_CHICKEN_DBG, BIT(2), 0);
1553 	return;
1554 }
1555 
1556 struct msm_gpu_state *a6xx_gpu_state_get(struct msm_gpu *gpu)
1557 {
1558 	struct a6xx_crashdumper _dumper = { 0 }, *dumper = NULL;
1559 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1560 	struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
1561 	struct a6xx_gpu_state *a6xx_state = kzalloc(sizeof(*a6xx_state),
1562 		GFP_KERNEL);
1563 	bool stalled = !!(gpu_read(gpu, REG_A6XX_RBBM_STATUS3) &
1564 			A6XX_RBBM_STATUS3_SMMU_STALLED_ON_FAULT);
1565 
1566 	if (!a6xx_state)
1567 		return ERR_PTR(-ENOMEM);
1568 
1569 	INIT_LIST_HEAD(&a6xx_state->objs);
1570 
1571 	/* Get the generic state from the adreno core */
1572 	adreno_gpu_state_get(gpu, &a6xx_state->base);
1573 
1574 	if (!adreno_has_gmu_wrapper(adreno_gpu)) {
1575 		a6xx_get_gmu_registers(gpu, a6xx_state);
1576 
1577 		a6xx_state->gmu_log = a6xx_snapshot_gmu_bo(a6xx_state, &a6xx_gpu->gmu.log);
1578 		a6xx_state->gmu_hfi = a6xx_snapshot_gmu_bo(a6xx_state, &a6xx_gpu->gmu.hfi);
1579 		a6xx_state->gmu_debug = a6xx_snapshot_gmu_bo(a6xx_state, &a6xx_gpu->gmu.debug);
1580 
1581 		a6xx_snapshot_gmu_hfi_history(gpu, a6xx_state);
1582 	}
1583 
1584 	/* If GX isn't on the rest of the data isn't going to be accessible */
1585 	if (!adreno_has_gmu_wrapper(adreno_gpu) && !a6xx_gmu_gx_is_on(&a6xx_gpu->gmu))
1586 		return &a6xx_state->base;
1587 
1588 	/* Get the banks of indexed registers */
1589 	if (adreno_is_a7xx(adreno_gpu))
1590 		a7xx_get_indexed_registers(gpu, a6xx_state);
1591 	else
1592 		a6xx_get_indexed_registers(gpu, a6xx_state);
1593 
1594 	/*
1595 	 * Try to initialize the crashdumper, if we are not dumping state
1596 	 * with the SMMU stalled.  The crashdumper needs memory access to
1597 	 * write out GPU state, so we need to skip this when the SMMU is
1598 	 * stalled in response to an iova fault
1599 	 */
1600 	if (!stalled && !gpu->needs_hw_init &&
1601 	    !a6xx_crashdumper_init(gpu, &_dumper)) {
1602 		dumper = &_dumper;
1603 	}
1604 
1605 	if (adreno_is_a7xx(adreno_gpu)) {
1606 		a7xx_get_registers(gpu, a6xx_state, dumper);
1607 
1608 		if (dumper) {
1609 			a7xx_get_shaders(gpu, a6xx_state, dumper);
1610 			a7xx_get_clusters(gpu, a6xx_state, dumper);
1611 			a7xx_get_dbgahb_clusters(gpu, a6xx_state, dumper);
1612 
1613 			msm_gem_kernel_put(dumper->bo, gpu->aspace);
1614 		}
1615 
1616 		a7xx_get_post_crashdumper_registers(gpu, a6xx_state);
1617 	} else {
1618 		a6xx_get_registers(gpu, a6xx_state, dumper);
1619 
1620 		if (dumper) {
1621 			a6xx_get_shaders(gpu, a6xx_state, dumper);
1622 			a6xx_get_clusters(gpu, a6xx_state, dumper);
1623 			a6xx_get_dbgahb_clusters(gpu, a6xx_state, dumper);
1624 
1625 			msm_gem_kernel_put(dumper->bo, gpu->aspace);
1626 		}
1627 	}
1628 
1629 	if (snapshot_debugbus)
1630 		a6xx_get_debugbus(gpu, a6xx_state);
1631 
1632 	a6xx_state->gpu_initialized = !gpu->needs_hw_init;
1633 
1634 	return  &a6xx_state->base;
1635 }
1636 
1637 static void a6xx_gpu_state_destroy(struct kref *kref)
1638 {
1639 	struct a6xx_state_memobj *obj, *tmp;
1640 	struct msm_gpu_state *state = container_of(kref,
1641 			struct msm_gpu_state, ref);
1642 	struct a6xx_gpu_state *a6xx_state = container_of(state,
1643 			struct a6xx_gpu_state, base);
1644 
1645 	if (a6xx_state->gmu_log)
1646 		kvfree(a6xx_state->gmu_log->data);
1647 
1648 	if (a6xx_state->gmu_hfi)
1649 		kvfree(a6xx_state->gmu_hfi->data);
1650 
1651 	if (a6xx_state->gmu_debug)
1652 		kvfree(a6xx_state->gmu_debug->data);
1653 
1654 	list_for_each_entry_safe(obj, tmp, &a6xx_state->objs, node) {
1655 		list_del(&obj->node);
1656 		kvfree(obj);
1657 	}
1658 
1659 	adreno_gpu_state_destroy(state);
1660 	kfree(a6xx_state);
1661 }
1662 
1663 int a6xx_gpu_state_put(struct msm_gpu_state *state)
1664 {
1665 	if (IS_ERR_OR_NULL(state))
1666 		return 1;
1667 
1668 	return kref_put(&state->ref, a6xx_gpu_state_destroy);
1669 }
1670 
1671 static void a6xx_show_registers(const u32 *registers, u32 *data, size_t count,
1672 		struct drm_printer *p)
1673 {
1674 	int i, index = 0;
1675 
1676 	if (!data)
1677 		return;
1678 
1679 	for (i = 0; i < count; i += 2) {
1680 		u32 count = RANGE(registers, i);
1681 		u32 offset = registers[i];
1682 		int j;
1683 
1684 		for (j = 0; j < count; index++, offset++, j++) {
1685 			if (data[index] == 0xdeafbead)
1686 				continue;
1687 
1688 			drm_printf(p, "  - { offset: 0x%06x, value: 0x%08x }\n",
1689 				offset << 2, data[index]);
1690 		}
1691 	}
1692 }
1693 
1694 static void a7xx_show_registers_indented(const u32 *registers, u32 *data,
1695 		struct drm_printer *p, unsigned indent)
1696 {
1697 	int i, index = 0;
1698 
1699 	for (i = 0; registers[i] != UINT_MAX; i += 2) {
1700 		u32 count = RANGE(registers, i);
1701 		u32 offset = registers[i];
1702 		int j;
1703 
1704 		for (j = 0; j < count; index++, offset++, j++) {
1705 			int k;
1706 
1707 			if (data[index] == 0xdeafbead)
1708 				continue;
1709 
1710 			for (k = 0; k < indent; k++)
1711 				drm_printf(p, "  ");
1712 			drm_printf(p, "- { offset: 0x%06x, value: 0x%08x }\n",
1713 				offset << 2, data[index]);
1714 		}
1715 	}
1716 }
1717 
1718 static void a7xx_show_registers(const u32 *registers, u32 *data, struct drm_printer *p)
1719 {
1720 	a7xx_show_registers_indented(registers, data, p, 1);
1721 }
1722 
1723 static void print_ascii85(struct drm_printer *p, size_t len, u32 *data)
1724 {
1725 	char out[ASCII85_BUFSZ];
1726 	long i, l, datalen = 0;
1727 
1728 	for (i = 0; i < len >> 2; i++) {
1729 		if (data[i])
1730 			datalen = (i + 1) << 2;
1731 	}
1732 
1733 	if (datalen == 0)
1734 		return;
1735 
1736 	drm_puts(p, "    data: !!ascii85 |\n");
1737 	drm_puts(p, "      ");
1738 
1739 
1740 	l = ascii85_encode_len(datalen);
1741 
1742 	for (i = 0; i < l; i++)
1743 		drm_puts(p, ascii85_encode(data[i], out));
1744 
1745 	drm_puts(p, "\n");
1746 }
1747 
1748 static void print_name(struct drm_printer *p, const char *fmt, const char *name)
1749 {
1750 	drm_puts(p, fmt);
1751 	drm_puts(p, name);
1752 	drm_puts(p, "\n");
1753 }
1754 
1755 static void a6xx_show_shader(struct a6xx_gpu_state_obj *obj,
1756 		struct drm_printer *p)
1757 {
1758 	const struct a6xx_shader_block *block = obj->handle;
1759 	int i;
1760 
1761 	if (!obj->handle)
1762 		return;
1763 
1764 	print_name(p, "  - type: ", block->name);
1765 
1766 	for (i = 0; i < A6XX_NUM_SHADER_BANKS; i++) {
1767 		drm_printf(p, "    - bank: %d\n", i);
1768 		drm_printf(p, "      size: %d\n", block->size);
1769 
1770 		if (!obj->data)
1771 			continue;
1772 
1773 		print_ascii85(p, block->size << 2,
1774 			obj->data + (block->size * i));
1775 	}
1776 }
1777 
1778 static void a7xx_show_shader(struct a6xx_gpu_state_obj *obj,
1779 		struct drm_printer *p)
1780 {
1781 	const struct gen7_shader_block *block = obj->handle;
1782 	int i, j;
1783 	u32 *data = obj->data;
1784 
1785 	if (!obj->handle)
1786 		return;
1787 
1788 	print_name(p, "  - type: ", a7xx_statetype_names[block->statetype]);
1789 	print_name(p, "    - pipe: ", a7xx_pipe_names[block->pipeid]);
1790 
1791 	for (i = 0; i < block->num_sps; i++) {
1792 		drm_printf(p, "      - sp: %d\n", i);
1793 
1794 		for (j = 0; j < block->num_usptps; j++) {
1795 			drm_printf(p, "        - usptp: %d\n", j);
1796 			drm_printf(p, "          size: %d\n", block->size);
1797 
1798 			if (!obj->data)
1799 				continue;
1800 
1801 			print_ascii85(p, block->size << 2, data);
1802 
1803 			data += block->size;
1804 		}
1805 	}
1806 }
1807 
1808 static void a6xx_show_cluster_data(const u32 *registers, int size, u32 *data,
1809 		struct drm_printer *p)
1810 {
1811 	int ctx, index = 0;
1812 
1813 	for (ctx = 0; ctx < A6XX_NUM_CONTEXTS; ctx++) {
1814 		int j;
1815 
1816 		drm_printf(p, "    - context: %d\n", ctx);
1817 
1818 		for (j = 0; j < size; j += 2) {
1819 			u32 count = RANGE(registers, j);
1820 			u32 offset = registers[j];
1821 			int k;
1822 
1823 			for (k = 0; k < count; index++, offset++, k++) {
1824 				if (data[index] == 0xdeafbead)
1825 					continue;
1826 
1827 				drm_printf(p, "      - { offset: 0x%06x, value: 0x%08x }\n",
1828 					offset << 2, data[index]);
1829 			}
1830 		}
1831 	}
1832 }
1833 
1834 static void a6xx_show_dbgahb_cluster(struct a6xx_gpu_state_obj *obj,
1835 		struct drm_printer *p)
1836 {
1837 	const struct a6xx_dbgahb_cluster *dbgahb = obj->handle;
1838 
1839 	if (dbgahb) {
1840 		print_name(p, "  - cluster-name: ", dbgahb->name);
1841 		a6xx_show_cluster_data(dbgahb->registers, dbgahb->count,
1842 			obj->data, p);
1843 	}
1844 }
1845 
1846 static void a6xx_show_cluster(struct a6xx_gpu_state_obj *obj,
1847 		struct drm_printer *p)
1848 {
1849 	const struct a6xx_cluster *cluster = obj->handle;
1850 
1851 	if (cluster) {
1852 		print_name(p, "  - cluster-name: ", cluster->name);
1853 		a6xx_show_cluster_data(cluster->registers, cluster->count,
1854 			obj->data, p);
1855 	}
1856 }
1857 
1858 static void a7xx_show_dbgahb_cluster(struct a6xx_gpu_state_obj *obj,
1859 		struct drm_printer *p)
1860 {
1861 	const struct gen7_sptp_cluster_registers *dbgahb = obj->handle;
1862 
1863 	if (dbgahb) {
1864 		print_name(p, "  - pipe: ", a7xx_pipe_names[dbgahb->pipe_id]);
1865 		print_name(p, "    - cluster-name: ", a7xx_cluster_names[dbgahb->cluster_id]);
1866 		drm_printf(p, "      - context: %d\n", dbgahb->context_id);
1867 		a7xx_show_registers_indented(dbgahb->regs, obj->data, p, 4);
1868 	}
1869 }
1870 
1871 static void a7xx_show_cluster(struct a6xx_gpu_state_obj *obj,
1872 		struct drm_printer *p)
1873 {
1874 	const struct gen7_cluster_registers *cluster = obj->handle;
1875 
1876 	if (cluster) {
1877 		int context = (cluster->context_id == STATE_FORCE_CTXT_1) ? 1 : 0;
1878 
1879 		print_name(p, "  - pipe: ", a7xx_pipe_names[cluster->pipe_id]);
1880 		print_name(p, "    - cluster-name: ", a7xx_cluster_names[cluster->cluster_id]);
1881 		drm_printf(p, "      - context: %d\n", context);
1882 		a7xx_show_registers_indented(cluster->regs, obj->data, p, 4);
1883 	}
1884 }
1885 
1886 static void a6xx_show_indexed_regs(struct a6xx_gpu_state_obj *obj,
1887 		struct drm_printer *p)
1888 {
1889 	const struct a6xx_indexed_registers *indexed = obj->handle;
1890 
1891 	if (!indexed)
1892 		return;
1893 
1894 	print_name(p, "  - regs-name: ", indexed->name);
1895 	drm_printf(p, "    dwords: %d\n", obj->count);
1896 
1897 	print_ascii85(p, obj->count << 2, obj->data);
1898 }
1899 
1900 static void a6xx_show_debugbus_block(const struct a6xx_debugbus_block *block,
1901 		u32 *data, struct drm_printer *p)
1902 {
1903 	if (block) {
1904 		print_name(p, "  - debugbus-block: ", block->name);
1905 
1906 		/*
1907 		 * count for regular debugbus data is in quadwords,
1908 		 * but print the size in dwords for consistency
1909 		 */
1910 		drm_printf(p, "    count: %d\n", block->count << 1);
1911 
1912 		print_ascii85(p, block->count << 3, data);
1913 	}
1914 }
1915 
1916 static void a6xx_show_debugbus(struct a6xx_gpu_state *a6xx_state,
1917 		struct drm_printer *p)
1918 {
1919 	int i;
1920 
1921 	for (i = 0; i < a6xx_state->nr_debugbus; i++) {
1922 		struct a6xx_gpu_state_obj *obj = &a6xx_state->debugbus[i];
1923 
1924 		a6xx_show_debugbus_block(obj->handle, obj->data, p);
1925 	}
1926 
1927 	if (a6xx_state->vbif_debugbus) {
1928 		struct a6xx_gpu_state_obj *obj = a6xx_state->vbif_debugbus;
1929 
1930 		drm_puts(p, "  - debugbus-block: A6XX_DBGBUS_VBIF\n");
1931 		drm_printf(p, "    count: %d\n", VBIF_DEBUGBUS_BLOCK_SIZE);
1932 
1933 		/* vbif debugbus data is in dwords.  Confusing, huh? */
1934 		print_ascii85(p, VBIF_DEBUGBUS_BLOCK_SIZE << 2, obj->data);
1935 	}
1936 
1937 	for (i = 0; i < a6xx_state->nr_cx_debugbus; i++) {
1938 		struct a6xx_gpu_state_obj *obj = &a6xx_state->cx_debugbus[i];
1939 
1940 		a6xx_show_debugbus_block(obj->handle, obj->data, p);
1941 	}
1942 }
1943 
1944 void a6xx_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
1945 		struct drm_printer *p)
1946 {
1947 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
1948 	struct a6xx_gpu_state *a6xx_state = container_of(state,
1949 			struct a6xx_gpu_state, base);
1950 	int i;
1951 
1952 	if (IS_ERR_OR_NULL(state))
1953 		return;
1954 
1955 	drm_printf(p, "gpu-initialized: %d\n", a6xx_state->gpu_initialized);
1956 
1957 	adreno_show(gpu, state, p);
1958 
1959 	drm_puts(p, "gmu-log:\n");
1960 	if (a6xx_state->gmu_log) {
1961 		struct msm_gpu_state_bo *gmu_log = a6xx_state->gmu_log;
1962 
1963 		drm_printf(p, "    iova: 0x%016llx\n", gmu_log->iova);
1964 		drm_printf(p, "    size: %zu\n", gmu_log->size);
1965 		adreno_show_object(p, &gmu_log->data, gmu_log->size,
1966 				&gmu_log->encoded);
1967 	}
1968 
1969 	drm_puts(p, "gmu-hfi:\n");
1970 	if (a6xx_state->gmu_hfi) {
1971 		struct msm_gpu_state_bo *gmu_hfi = a6xx_state->gmu_hfi;
1972 		unsigned i, j;
1973 
1974 		drm_printf(p, "    iova: 0x%016llx\n", gmu_hfi->iova);
1975 		drm_printf(p, "    size: %zu\n", gmu_hfi->size);
1976 		for (i = 0; i < ARRAY_SIZE(a6xx_state->hfi_queue_history); i++) {
1977 			drm_printf(p, "    queue-history[%u]:", i);
1978 			for (j = 0; j < HFI_HISTORY_SZ; j++) {
1979 				drm_printf(p, " %d", a6xx_state->hfi_queue_history[i][j]);
1980 			}
1981 			drm_printf(p, "\n");
1982 		}
1983 		adreno_show_object(p, &gmu_hfi->data, gmu_hfi->size,
1984 				&gmu_hfi->encoded);
1985 	}
1986 
1987 	drm_puts(p, "gmu-debug:\n");
1988 	if (a6xx_state->gmu_debug) {
1989 		struct msm_gpu_state_bo *gmu_debug = a6xx_state->gmu_debug;
1990 
1991 		drm_printf(p, "    iova: 0x%016llx\n", gmu_debug->iova);
1992 		drm_printf(p, "    size: %zu\n", gmu_debug->size);
1993 		adreno_show_object(p, &gmu_debug->data, gmu_debug->size,
1994 				&gmu_debug->encoded);
1995 	}
1996 
1997 	drm_puts(p, "registers:\n");
1998 	for (i = 0; i < a6xx_state->nr_registers; i++) {
1999 		struct a6xx_gpu_state_obj *obj = &a6xx_state->registers[i];
2000 
2001 		if (!obj->handle)
2002 			continue;
2003 
2004 		if (adreno_is_a7xx(adreno_gpu)) {
2005 			a7xx_show_registers(obj->handle, obj->data, p);
2006 		} else {
2007 			const struct a6xx_registers *regs = obj->handle;
2008 
2009 			a6xx_show_registers(regs->registers, obj->data, regs->count, p);
2010 		}
2011 	}
2012 
2013 	drm_puts(p, "registers-gmu:\n");
2014 	for (i = 0; i < a6xx_state->nr_gmu_registers; i++) {
2015 		struct a6xx_gpu_state_obj *obj = &a6xx_state->gmu_registers[i];
2016 		const struct a6xx_registers *regs = obj->handle;
2017 
2018 		if (!obj->handle)
2019 			continue;
2020 
2021 		a6xx_show_registers(regs->registers, obj->data, regs->count, p);
2022 	}
2023 
2024 	drm_puts(p, "indexed-registers:\n");
2025 	for (i = 0; i < a6xx_state->nr_indexed_regs; i++)
2026 		a6xx_show_indexed_regs(&a6xx_state->indexed_regs[i], p);
2027 
2028 	drm_puts(p, "shader-blocks:\n");
2029 	for (i = 0; i < a6xx_state->nr_shaders; i++) {
2030 		if (adreno_is_a7xx(adreno_gpu))
2031 			a7xx_show_shader(&a6xx_state->shaders[i], p);
2032 		else
2033 			a6xx_show_shader(&a6xx_state->shaders[i], p);
2034 	}
2035 
2036 	drm_puts(p, "clusters:\n");
2037 	for (i = 0; i < a6xx_state->nr_clusters; i++) {
2038 		if (adreno_is_a7xx(adreno_gpu))
2039 			a7xx_show_cluster(&a6xx_state->clusters[i], p);
2040 		else
2041 			a6xx_show_cluster(&a6xx_state->clusters[i], p);
2042 	}
2043 
2044 	for (i = 0; i < a6xx_state->nr_dbgahb_clusters; i++) {
2045 		if (adreno_is_a7xx(adreno_gpu))
2046 			a7xx_show_dbgahb_cluster(&a6xx_state->dbgahb_clusters[i], p);
2047 		else
2048 			a6xx_show_dbgahb_cluster(&a6xx_state->dbgahb_clusters[i], p);
2049 	}
2050 
2051 	drm_puts(p, "debugbus:\n");
2052 	a6xx_show_debugbus(a6xx_state, p);
2053 }
2054