xref: /linux/drivers/accel/habanalabs/goya/goya.c (revision e7d759f31ca295d589f7420719c311870bb3166f)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 /*
4  * Copyright 2016-2022 HabanaLabs, Ltd.
5  * All Rights Reserved.
6  */
7 
8 #include "goyaP.h"
9 #include "../include/hw_ip/mmu/mmu_general.h"
10 #include "../include/hw_ip/mmu/mmu_v1_0.h"
11 #include "../include/goya/asic_reg/goya_masks.h"
12 #include "../include/goya/goya_reg_map.h"
13 
14 #include <linux/pci.h>
15 #include <linux/hwmon.h>
16 #include <linux/iommu.h>
17 #include <linux/seq_file.h>
18 
19 /*
20  * GOYA security scheme:
21  *
22  * 1. Host is protected by:
23  *        - Range registers (When MMU is enabled, DMA RR does NOT protect host)
24  *        - MMU
25  *
26  * 2. DRAM is protected by:
27  *        - Range registers (protect the first 512MB)
28  *        - MMU (isolation between users)
29  *
30  * 3. Configuration is protected by:
31  *        - Range registers
32  *        - Protection bits
33  *
34  * When MMU is disabled:
35  *
36  * QMAN DMA: PQ, CQ, CP, DMA are secured.
37  * PQ, CB and the data are on the host.
38  *
39  * QMAN TPC/MME:
40  * PQ, CQ and CP are not secured.
41  * PQ, CB and the data are on the SRAM/DRAM.
42  *
43  * Since QMAN DMA is secured, the driver is parsing the DMA CB:
44  *     - checks DMA pointer
45  *     - WREG, MSG_PROT are not allowed.
46  *     - MSG_LONG/SHORT are allowed.
47  *
48  * A read/write transaction by the QMAN to a protected area will succeed if
49  * and only if the QMAN's CP is secured and MSG_PROT is used
50  *
51  *
52  * When MMU is enabled:
53  *
54  * QMAN DMA: PQ, CQ and CP are secured.
55  * MMU is set to bypass on the Secure props register of the QMAN.
56  * The reasons we don't enable MMU for PQ, CQ and CP are:
57  *     - PQ entry is in kernel address space and the driver doesn't map it.
58  *     - CP writes to MSIX register and to kernel address space (completion
59  *       queue).
60  *
61  * DMA is not secured but because CP is secured, the driver still needs to parse
62  * the CB, but doesn't need to check the DMA addresses.
63  *
64  * For QMAN DMA 0, DMA is also secured because only the driver uses this DMA and
65  * the driver doesn't map memory in MMU.
66  *
67  * QMAN TPC/MME: PQ, CQ and CP aren't secured (no change from MMU disabled mode)
68  *
69  * DMA RR does NOT protect host because DMA is not secured
70  *
71  */
72 
73 #define GOYA_BOOT_FIT_FILE	"habanalabs/goya/goya-boot-fit.itb"
74 #define GOYA_LINUX_FW_FILE	"habanalabs/goya/goya-fit.itb"
75 
76 #define GOYA_MMU_REGS_NUM		63
77 
78 #define GOYA_DMA_POOL_BLK_SIZE		0x100		/* 256 bytes */
79 
80 #define GOYA_RESET_TIMEOUT_MSEC		500		/* 500ms */
81 #define GOYA_PLDM_RESET_TIMEOUT_MSEC	20000		/* 20s */
82 #define GOYA_RESET_WAIT_MSEC		1		/* 1ms */
83 #define GOYA_CPU_RESET_WAIT_MSEC	100		/* 100ms */
84 #define GOYA_PLDM_RESET_WAIT_MSEC	1000		/* 1s */
85 #define GOYA_TEST_QUEUE_WAIT_USEC	100000		/* 100ms */
86 #define GOYA_PLDM_MMU_TIMEOUT_USEC	(MMU_CONFIG_TIMEOUT_USEC * 100)
87 #define GOYA_PLDM_QMAN0_TIMEOUT_USEC	(HL_DEVICE_TIMEOUT_USEC * 30)
88 #define GOYA_BOOT_FIT_REQ_TIMEOUT_USEC	1000000		/* 1s */
89 #define GOYA_MSG_TO_CPU_TIMEOUT_USEC	4000000		/* 4s */
90 #define GOYA_WAIT_FOR_BL_TIMEOUT_USEC	15000000	/* 15s */
91 
92 #define GOYA_QMAN0_FENCE_VAL		0xD169B243
93 
94 #define GOYA_MAX_STRING_LEN		20
95 
96 #define GOYA_CB_POOL_CB_CNT		512
97 #define GOYA_CB_POOL_CB_SIZE		0x20000		/* 128KB */
98 
99 #define IS_QM_IDLE(engine, qm_glbl_sts0) \
100 	(((qm_glbl_sts0) & engine##_QM_IDLE_MASK) == engine##_QM_IDLE_MASK)
101 #define IS_DMA_QM_IDLE(qm_glbl_sts0)	IS_QM_IDLE(DMA, qm_glbl_sts0)
102 #define IS_TPC_QM_IDLE(qm_glbl_sts0)	IS_QM_IDLE(TPC, qm_glbl_sts0)
103 #define IS_MME_QM_IDLE(qm_glbl_sts0)	IS_QM_IDLE(MME, qm_glbl_sts0)
104 
105 #define IS_CMDQ_IDLE(engine, cmdq_glbl_sts0) \
106 	(((cmdq_glbl_sts0) & engine##_CMDQ_IDLE_MASK) == \
107 			engine##_CMDQ_IDLE_MASK)
108 #define IS_TPC_CMDQ_IDLE(cmdq_glbl_sts0) \
109 	IS_CMDQ_IDLE(TPC, cmdq_glbl_sts0)
110 #define IS_MME_CMDQ_IDLE(cmdq_glbl_sts0) \
111 	IS_CMDQ_IDLE(MME, cmdq_glbl_sts0)
112 
113 #define IS_DMA_IDLE(dma_core_sts0) \
114 	!((dma_core_sts0) & DMA_CH_0_STS0_DMA_BUSY_MASK)
115 
116 #define IS_TPC_IDLE(tpc_cfg_sts) \
117 	(((tpc_cfg_sts) & TPC_CFG_IDLE_MASK) == TPC_CFG_IDLE_MASK)
118 
119 #define IS_MME_IDLE(mme_arch_sts) \
120 	(((mme_arch_sts) & MME_ARCH_IDLE_MASK) == MME_ARCH_IDLE_MASK)
121 
122 static const char goya_irq_name[GOYA_MSIX_ENTRIES][GOYA_MAX_STRING_LEN] = {
123 		"goya cq 0", "goya cq 1", "goya cq 2", "goya cq 3",
124 		"goya cq 4", "goya cpu eq"
125 };
126 
127 static u16 goya_packet_sizes[MAX_PACKET_ID] = {
128 	[PACKET_WREG_32]	= sizeof(struct packet_wreg32),
129 	[PACKET_WREG_BULK]	= sizeof(struct packet_wreg_bulk),
130 	[PACKET_MSG_LONG]	= sizeof(struct packet_msg_long),
131 	[PACKET_MSG_SHORT]	= sizeof(struct packet_msg_short),
132 	[PACKET_CP_DMA]		= sizeof(struct packet_cp_dma),
133 	[PACKET_MSG_PROT]	= sizeof(struct packet_msg_prot),
134 	[PACKET_FENCE]		= sizeof(struct packet_fence),
135 	[PACKET_LIN_DMA]	= sizeof(struct packet_lin_dma),
136 	[PACKET_NOP]		= sizeof(struct packet_nop),
137 	[PACKET_STOP]		= sizeof(struct packet_stop)
138 };
139 
140 static inline bool validate_packet_id(enum packet_id id)
141 {
142 	switch (id) {
143 	case PACKET_WREG_32:
144 	case PACKET_WREG_BULK:
145 	case PACKET_MSG_LONG:
146 	case PACKET_MSG_SHORT:
147 	case PACKET_CP_DMA:
148 	case PACKET_MSG_PROT:
149 	case PACKET_FENCE:
150 	case PACKET_LIN_DMA:
151 	case PACKET_NOP:
152 	case PACKET_STOP:
153 		return true;
154 	default:
155 		return false;
156 	}
157 }
158 
159 static u64 goya_mmu_regs[GOYA_MMU_REGS_NUM] = {
160 	mmDMA_QM_0_GLBL_NON_SECURE_PROPS,
161 	mmDMA_QM_1_GLBL_NON_SECURE_PROPS,
162 	mmDMA_QM_2_GLBL_NON_SECURE_PROPS,
163 	mmDMA_QM_3_GLBL_NON_SECURE_PROPS,
164 	mmDMA_QM_4_GLBL_NON_SECURE_PROPS,
165 	mmTPC0_QM_GLBL_SECURE_PROPS,
166 	mmTPC0_QM_GLBL_NON_SECURE_PROPS,
167 	mmTPC0_CMDQ_GLBL_SECURE_PROPS,
168 	mmTPC0_CMDQ_GLBL_NON_SECURE_PROPS,
169 	mmTPC0_CFG_ARUSER,
170 	mmTPC0_CFG_AWUSER,
171 	mmTPC1_QM_GLBL_SECURE_PROPS,
172 	mmTPC1_QM_GLBL_NON_SECURE_PROPS,
173 	mmTPC1_CMDQ_GLBL_SECURE_PROPS,
174 	mmTPC1_CMDQ_GLBL_NON_SECURE_PROPS,
175 	mmTPC1_CFG_ARUSER,
176 	mmTPC1_CFG_AWUSER,
177 	mmTPC2_QM_GLBL_SECURE_PROPS,
178 	mmTPC2_QM_GLBL_NON_SECURE_PROPS,
179 	mmTPC2_CMDQ_GLBL_SECURE_PROPS,
180 	mmTPC2_CMDQ_GLBL_NON_SECURE_PROPS,
181 	mmTPC2_CFG_ARUSER,
182 	mmTPC2_CFG_AWUSER,
183 	mmTPC3_QM_GLBL_SECURE_PROPS,
184 	mmTPC3_QM_GLBL_NON_SECURE_PROPS,
185 	mmTPC3_CMDQ_GLBL_SECURE_PROPS,
186 	mmTPC3_CMDQ_GLBL_NON_SECURE_PROPS,
187 	mmTPC3_CFG_ARUSER,
188 	mmTPC3_CFG_AWUSER,
189 	mmTPC4_QM_GLBL_SECURE_PROPS,
190 	mmTPC4_QM_GLBL_NON_SECURE_PROPS,
191 	mmTPC4_CMDQ_GLBL_SECURE_PROPS,
192 	mmTPC4_CMDQ_GLBL_NON_SECURE_PROPS,
193 	mmTPC4_CFG_ARUSER,
194 	mmTPC4_CFG_AWUSER,
195 	mmTPC5_QM_GLBL_SECURE_PROPS,
196 	mmTPC5_QM_GLBL_NON_SECURE_PROPS,
197 	mmTPC5_CMDQ_GLBL_SECURE_PROPS,
198 	mmTPC5_CMDQ_GLBL_NON_SECURE_PROPS,
199 	mmTPC5_CFG_ARUSER,
200 	mmTPC5_CFG_AWUSER,
201 	mmTPC6_QM_GLBL_SECURE_PROPS,
202 	mmTPC6_QM_GLBL_NON_SECURE_PROPS,
203 	mmTPC6_CMDQ_GLBL_SECURE_PROPS,
204 	mmTPC6_CMDQ_GLBL_NON_SECURE_PROPS,
205 	mmTPC6_CFG_ARUSER,
206 	mmTPC6_CFG_AWUSER,
207 	mmTPC7_QM_GLBL_SECURE_PROPS,
208 	mmTPC7_QM_GLBL_NON_SECURE_PROPS,
209 	mmTPC7_CMDQ_GLBL_SECURE_PROPS,
210 	mmTPC7_CMDQ_GLBL_NON_SECURE_PROPS,
211 	mmTPC7_CFG_ARUSER,
212 	mmTPC7_CFG_AWUSER,
213 	mmMME_QM_GLBL_SECURE_PROPS,
214 	mmMME_QM_GLBL_NON_SECURE_PROPS,
215 	mmMME_CMDQ_GLBL_SECURE_PROPS,
216 	mmMME_CMDQ_GLBL_NON_SECURE_PROPS,
217 	mmMME_SBA_CONTROL_DATA,
218 	mmMME_SBB_CONTROL_DATA,
219 	mmMME_SBC_CONTROL_DATA,
220 	mmMME_WBC_CONTROL_DATA,
221 	mmPCIE_WRAP_PSOC_ARUSER,
222 	mmPCIE_WRAP_PSOC_AWUSER
223 };
224 
225 static u32 goya_all_events[] = {
226 	GOYA_ASYNC_EVENT_ID_PCIE_IF,
227 	GOYA_ASYNC_EVENT_ID_TPC0_ECC,
228 	GOYA_ASYNC_EVENT_ID_TPC1_ECC,
229 	GOYA_ASYNC_EVENT_ID_TPC2_ECC,
230 	GOYA_ASYNC_EVENT_ID_TPC3_ECC,
231 	GOYA_ASYNC_EVENT_ID_TPC4_ECC,
232 	GOYA_ASYNC_EVENT_ID_TPC5_ECC,
233 	GOYA_ASYNC_EVENT_ID_TPC6_ECC,
234 	GOYA_ASYNC_EVENT_ID_TPC7_ECC,
235 	GOYA_ASYNC_EVENT_ID_MME_ECC,
236 	GOYA_ASYNC_EVENT_ID_MME_ECC_EXT,
237 	GOYA_ASYNC_EVENT_ID_MMU_ECC,
238 	GOYA_ASYNC_EVENT_ID_DMA_MACRO,
239 	GOYA_ASYNC_EVENT_ID_DMA_ECC,
240 	GOYA_ASYNC_EVENT_ID_CPU_IF_ECC,
241 	GOYA_ASYNC_EVENT_ID_PSOC_MEM,
242 	GOYA_ASYNC_EVENT_ID_PSOC_CORESIGHT,
243 	GOYA_ASYNC_EVENT_ID_SRAM0,
244 	GOYA_ASYNC_EVENT_ID_SRAM1,
245 	GOYA_ASYNC_EVENT_ID_SRAM2,
246 	GOYA_ASYNC_EVENT_ID_SRAM3,
247 	GOYA_ASYNC_EVENT_ID_SRAM4,
248 	GOYA_ASYNC_EVENT_ID_SRAM5,
249 	GOYA_ASYNC_EVENT_ID_SRAM6,
250 	GOYA_ASYNC_EVENT_ID_SRAM7,
251 	GOYA_ASYNC_EVENT_ID_SRAM8,
252 	GOYA_ASYNC_EVENT_ID_SRAM9,
253 	GOYA_ASYNC_EVENT_ID_SRAM10,
254 	GOYA_ASYNC_EVENT_ID_SRAM11,
255 	GOYA_ASYNC_EVENT_ID_SRAM12,
256 	GOYA_ASYNC_EVENT_ID_SRAM13,
257 	GOYA_ASYNC_EVENT_ID_SRAM14,
258 	GOYA_ASYNC_EVENT_ID_SRAM15,
259 	GOYA_ASYNC_EVENT_ID_SRAM16,
260 	GOYA_ASYNC_EVENT_ID_SRAM17,
261 	GOYA_ASYNC_EVENT_ID_SRAM18,
262 	GOYA_ASYNC_EVENT_ID_SRAM19,
263 	GOYA_ASYNC_EVENT_ID_SRAM20,
264 	GOYA_ASYNC_EVENT_ID_SRAM21,
265 	GOYA_ASYNC_EVENT_ID_SRAM22,
266 	GOYA_ASYNC_EVENT_ID_SRAM23,
267 	GOYA_ASYNC_EVENT_ID_SRAM24,
268 	GOYA_ASYNC_EVENT_ID_SRAM25,
269 	GOYA_ASYNC_EVENT_ID_SRAM26,
270 	GOYA_ASYNC_EVENT_ID_SRAM27,
271 	GOYA_ASYNC_EVENT_ID_SRAM28,
272 	GOYA_ASYNC_EVENT_ID_SRAM29,
273 	GOYA_ASYNC_EVENT_ID_GIC500,
274 	GOYA_ASYNC_EVENT_ID_PLL0,
275 	GOYA_ASYNC_EVENT_ID_PLL1,
276 	GOYA_ASYNC_EVENT_ID_PLL3,
277 	GOYA_ASYNC_EVENT_ID_PLL4,
278 	GOYA_ASYNC_EVENT_ID_PLL5,
279 	GOYA_ASYNC_EVENT_ID_PLL6,
280 	GOYA_ASYNC_EVENT_ID_AXI_ECC,
281 	GOYA_ASYNC_EVENT_ID_L2_RAM_ECC,
282 	GOYA_ASYNC_EVENT_ID_PSOC_GPIO_05_SW_RESET,
283 	GOYA_ASYNC_EVENT_ID_PSOC_GPIO_10_VRHOT_ICRIT,
284 	GOYA_ASYNC_EVENT_ID_PCIE_DEC,
285 	GOYA_ASYNC_EVENT_ID_TPC0_DEC,
286 	GOYA_ASYNC_EVENT_ID_TPC1_DEC,
287 	GOYA_ASYNC_EVENT_ID_TPC2_DEC,
288 	GOYA_ASYNC_EVENT_ID_TPC3_DEC,
289 	GOYA_ASYNC_EVENT_ID_TPC4_DEC,
290 	GOYA_ASYNC_EVENT_ID_TPC5_DEC,
291 	GOYA_ASYNC_EVENT_ID_TPC6_DEC,
292 	GOYA_ASYNC_EVENT_ID_TPC7_DEC,
293 	GOYA_ASYNC_EVENT_ID_MME_WACS,
294 	GOYA_ASYNC_EVENT_ID_MME_WACSD,
295 	GOYA_ASYNC_EVENT_ID_CPU_AXI_SPLITTER,
296 	GOYA_ASYNC_EVENT_ID_PSOC_AXI_DEC,
297 	GOYA_ASYNC_EVENT_ID_PSOC,
298 	GOYA_ASYNC_EVENT_ID_TPC0_KRN_ERR,
299 	GOYA_ASYNC_EVENT_ID_TPC1_KRN_ERR,
300 	GOYA_ASYNC_EVENT_ID_TPC2_KRN_ERR,
301 	GOYA_ASYNC_EVENT_ID_TPC3_KRN_ERR,
302 	GOYA_ASYNC_EVENT_ID_TPC4_KRN_ERR,
303 	GOYA_ASYNC_EVENT_ID_TPC5_KRN_ERR,
304 	GOYA_ASYNC_EVENT_ID_TPC6_KRN_ERR,
305 	GOYA_ASYNC_EVENT_ID_TPC7_KRN_ERR,
306 	GOYA_ASYNC_EVENT_ID_TPC0_CMDQ,
307 	GOYA_ASYNC_EVENT_ID_TPC1_CMDQ,
308 	GOYA_ASYNC_EVENT_ID_TPC2_CMDQ,
309 	GOYA_ASYNC_EVENT_ID_TPC3_CMDQ,
310 	GOYA_ASYNC_EVENT_ID_TPC4_CMDQ,
311 	GOYA_ASYNC_EVENT_ID_TPC5_CMDQ,
312 	GOYA_ASYNC_EVENT_ID_TPC6_CMDQ,
313 	GOYA_ASYNC_EVENT_ID_TPC7_CMDQ,
314 	GOYA_ASYNC_EVENT_ID_TPC0_QM,
315 	GOYA_ASYNC_EVENT_ID_TPC1_QM,
316 	GOYA_ASYNC_EVENT_ID_TPC2_QM,
317 	GOYA_ASYNC_EVENT_ID_TPC3_QM,
318 	GOYA_ASYNC_EVENT_ID_TPC4_QM,
319 	GOYA_ASYNC_EVENT_ID_TPC5_QM,
320 	GOYA_ASYNC_EVENT_ID_TPC6_QM,
321 	GOYA_ASYNC_EVENT_ID_TPC7_QM,
322 	GOYA_ASYNC_EVENT_ID_MME_QM,
323 	GOYA_ASYNC_EVENT_ID_MME_CMDQ,
324 	GOYA_ASYNC_EVENT_ID_DMA0_QM,
325 	GOYA_ASYNC_EVENT_ID_DMA1_QM,
326 	GOYA_ASYNC_EVENT_ID_DMA2_QM,
327 	GOYA_ASYNC_EVENT_ID_DMA3_QM,
328 	GOYA_ASYNC_EVENT_ID_DMA4_QM,
329 	GOYA_ASYNC_EVENT_ID_DMA0_CH,
330 	GOYA_ASYNC_EVENT_ID_DMA1_CH,
331 	GOYA_ASYNC_EVENT_ID_DMA2_CH,
332 	GOYA_ASYNC_EVENT_ID_DMA3_CH,
333 	GOYA_ASYNC_EVENT_ID_DMA4_CH,
334 	GOYA_ASYNC_EVENT_ID_TPC0_BMON_SPMU,
335 	GOYA_ASYNC_EVENT_ID_TPC1_BMON_SPMU,
336 	GOYA_ASYNC_EVENT_ID_TPC2_BMON_SPMU,
337 	GOYA_ASYNC_EVENT_ID_TPC3_BMON_SPMU,
338 	GOYA_ASYNC_EVENT_ID_TPC4_BMON_SPMU,
339 	GOYA_ASYNC_EVENT_ID_TPC5_BMON_SPMU,
340 	GOYA_ASYNC_EVENT_ID_TPC6_BMON_SPMU,
341 	GOYA_ASYNC_EVENT_ID_TPC7_BMON_SPMU,
342 	GOYA_ASYNC_EVENT_ID_DMA_BM_CH0,
343 	GOYA_ASYNC_EVENT_ID_DMA_BM_CH1,
344 	GOYA_ASYNC_EVENT_ID_DMA_BM_CH2,
345 	GOYA_ASYNC_EVENT_ID_DMA_BM_CH3,
346 	GOYA_ASYNC_EVENT_ID_DMA_BM_CH4,
347 	GOYA_ASYNC_EVENT_ID_FIX_POWER_ENV_S,
348 	GOYA_ASYNC_EVENT_ID_FIX_POWER_ENV_E,
349 	GOYA_ASYNC_EVENT_ID_FIX_THERMAL_ENV_S,
350 	GOYA_ASYNC_EVENT_ID_FIX_THERMAL_ENV_E
351 };
352 
353 static s64 goya_state_dump_specs_props[SP_MAX] = {0};
354 
355 static int goya_mmu_clear_pgt_range(struct hl_device *hdev);
356 static int goya_mmu_set_dram_default_page(struct hl_device *hdev);
357 static int goya_mmu_add_mappings_for_device_cpu(struct hl_device *hdev);
358 static void goya_mmu_prepare(struct hl_device *hdev, u32 asid);
359 
360 int goya_set_fixed_properties(struct hl_device *hdev)
361 {
362 	struct asic_fixed_properties *prop = &hdev->asic_prop;
363 	int i;
364 
365 	prop->max_queues = GOYA_QUEUE_ID_SIZE;
366 	prop->hw_queues_props = kcalloc(prop->max_queues,
367 			sizeof(struct hw_queue_properties),
368 			GFP_KERNEL);
369 
370 	if (!prop->hw_queues_props)
371 		return -ENOMEM;
372 
373 	for (i = 0 ; i < NUMBER_OF_EXT_HW_QUEUES ; i++) {
374 		prop->hw_queues_props[i].type = QUEUE_TYPE_EXT;
375 		prop->hw_queues_props[i].driver_only = 0;
376 		prop->hw_queues_props[i].cb_alloc_flags = CB_ALLOC_KERNEL;
377 	}
378 
379 	for (; i < NUMBER_OF_EXT_HW_QUEUES + NUMBER_OF_CPU_HW_QUEUES ; i++) {
380 		prop->hw_queues_props[i].type = QUEUE_TYPE_CPU;
381 		prop->hw_queues_props[i].driver_only = 1;
382 		prop->hw_queues_props[i].cb_alloc_flags = CB_ALLOC_KERNEL;
383 	}
384 
385 	for (; i < NUMBER_OF_EXT_HW_QUEUES + NUMBER_OF_CPU_HW_QUEUES +
386 			NUMBER_OF_INT_HW_QUEUES; i++) {
387 		prop->hw_queues_props[i].type = QUEUE_TYPE_INT;
388 		prop->hw_queues_props[i].driver_only = 0;
389 		prop->hw_queues_props[i].cb_alloc_flags = CB_ALLOC_USER;
390 	}
391 
392 	prop->cfg_base_address = CFG_BASE;
393 	prop->device_dma_offset_for_host_access = HOST_PHYS_BASE;
394 	prop->host_base_address = HOST_PHYS_BASE;
395 	prop->host_end_address = prop->host_base_address + HOST_PHYS_SIZE;
396 	prop->completion_queues_count = NUMBER_OF_CMPLT_QUEUES;
397 	prop->completion_mode = HL_COMPLETION_MODE_JOB;
398 	prop->dram_base_address = DRAM_PHYS_BASE;
399 	prop->dram_size = DRAM_PHYS_DEFAULT_SIZE;
400 	prop->dram_end_address = prop->dram_base_address + prop->dram_size;
401 	prop->dram_user_base_address = DRAM_BASE_ADDR_USER;
402 
403 	prop->sram_base_address = SRAM_BASE_ADDR;
404 	prop->sram_size = SRAM_SIZE;
405 	prop->sram_end_address = prop->sram_base_address + prop->sram_size;
406 	prop->sram_user_base_address = prop->sram_base_address +
407 						SRAM_USER_BASE_OFFSET;
408 
409 	prop->mmu_pgt_addr = MMU_PAGE_TABLES_ADDR;
410 	prop->mmu_dram_default_page_addr = MMU_DRAM_DEFAULT_PAGE_ADDR;
411 	if (hdev->pldm)
412 		prop->mmu_pgt_size = 0x800000; /* 8MB */
413 	else
414 		prop->mmu_pgt_size = MMU_PAGE_TABLES_SIZE;
415 	prop->mmu_pte_size = HL_PTE_SIZE;
416 	prop->mmu_hop_table_size = HOP_TABLE_SIZE_512_PTE;
417 	prop->mmu_hop0_tables_total_size = HOP0_512_PTE_TABLES_TOTAL_SIZE;
418 	prop->dram_page_size = PAGE_SIZE_2MB;
419 	prop->device_mem_alloc_default_page_size = prop->dram_page_size;
420 	prop->dram_supports_virtual_memory = true;
421 
422 	prop->dmmu.hop_shifts[MMU_HOP0] = MMU_V1_0_HOP0_SHIFT;
423 	prop->dmmu.hop_shifts[MMU_HOP1] = MMU_V1_0_HOP1_SHIFT;
424 	prop->dmmu.hop_shifts[MMU_HOP2] = MMU_V1_0_HOP2_SHIFT;
425 	prop->dmmu.hop_shifts[MMU_HOP3] = MMU_V1_0_HOP3_SHIFT;
426 	prop->dmmu.hop_shifts[MMU_HOP4] = MMU_V1_0_HOP4_SHIFT;
427 	prop->dmmu.hop_masks[MMU_HOP0] = MMU_V1_0_HOP0_MASK;
428 	prop->dmmu.hop_masks[MMU_HOP1] = MMU_V1_0_HOP1_MASK;
429 	prop->dmmu.hop_masks[MMU_HOP2] = MMU_V1_0_HOP2_MASK;
430 	prop->dmmu.hop_masks[MMU_HOP3] = MMU_V1_0_HOP3_MASK;
431 	prop->dmmu.hop_masks[MMU_HOP4] = MMU_V1_0_HOP4_MASK;
432 	prop->dmmu.start_addr = VA_DDR_SPACE_START;
433 	prop->dmmu.end_addr = VA_DDR_SPACE_END;
434 	prop->dmmu.page_size = PAGE_SIZE_2MB;
435 	prop->dmmu.num_hops = MMU_ARCH_5_HOPS;
436 	prop->dmmu.last_mask = LAST_MASK;
437 	/* TODO: will be duplicated until implementing per-MMU props */
438 	prop->dmmu.hop_table_size = prop->mmu_hop_table_size;
439 	prop->dmmu.hop0_tables_total_size = prop->mmu_hop0_tables_total_size;
440 
441 	/* shifts and masks are the same in PMMU and DMMU */
442 	memcpy(&prop->pmmu, &prop->dmmu, sizeof(prop->dmmu));
443 	prop->pmmu.start_addr = VA_HOST_SPACE_START;
444 	prop->pmmu.end_addr = VA_HOST_SPACE_END;
445 	prop->pmmu.page_size = PAGE_SIZE_4KB;
446 	prop->pmmu.num_hops = MMU_ARCH_5_HOPS;
447 	prop->pmmu.last_mask = LAST_MASK;
448 	/* TODO: will be duplicated until implementing per-MMU props */
449 	prop->pmmu.hop_table_size = prop->mmu_hop_table_size;
450 	prop->pmmu.hop0_tables_total_size = prop->mmu_hop0_tables_total_size;
451 
452 	/* PMMU and HPMMU are the same except of page size */
453 	memcpy(&prop->pmmu_huge, &prop->pmmu, sizeof(prop->pmmu));
454 	prop->pmmu_huge.page_size = PAGE_SIZE_2MB;
455 
456 	prop->dram_size_for_default_page_mapping = VA_DDR_SPACE_END;
457 	prop->cfg_size = CFG_SIZE;
458 	prop->max_asid = MAX_ASID;
459 	prop->num_of_events = GOYA_ASYNC_EVENT_ID_SIZE;
460 	prop->high_pll = PLL_HIGH_DEFAULT;
461 	prop->cb_pool_cb_cnt = GOYA_CB_POOL_CB_CNT;
462 	prop->cb_pool_cb_size = GOYA_CB_POOL_CB_SIZE;
463 	prop->max_power_default = MAX_POWER_DEFAULT;
464 	prop->dc_power_default = DC_POWER_DEFAULT;
465 	prop->tpc_enabled_mask = TPC_ENABLED_MASK;
466 	prop->pcie_dbi_base_address = mmPCIE_DBI_BASE;
467 	prop->pcie_aux_dbi_reg_addr = CFG_BASE + mmPCIE_AUX_DBI;
468 
469 	strscpy_pad(prop->cpucp_info.card_name, GOYA_DEFAULT_CARD_NAME,
470 		CARD_NAME_MAX_LEN);
471 
472 	prop->max_pending_cs = GOYA_MAX_PENDING_CS;
473 
474 	prop->first_available_user_interrupt = USHRT_MAX;
475 	prop->tpc_interrupt_id = USHRT_MAX;
476 	prop->eq_interrupt_id = GOYA_EVENT_QUEUE_MSIX_IDX;
477 
478 	for (i = 0 ; i < HL_MAX_DCORES ; i++)
479 		prop->first_available_cq[i] = USHRT_MAX;
480 
481 	prop->fw_cpu_boot_dev_sts0_valid = false;
482 	prop->fw_cpu_boot_dev_sts1_valid = false;
483 	prop->hard_reset_done_by_fw = false;
484 	prop->gic_interrupts_enable = true;
485 
486 	prop->server_type = HL_SERVER_TYPE_UNKNOWN;
487 
488 	prop->clk_pll_index = HL_GOYA_MME_PLL;
489 
490 	prop->use_get_power_for_reset_history = true;
491 
492 	prop->configurable_stop_on_err = true;
493 
494 	prop->set_max_power_on_device_init = true;
495 
496 	prop->dma_mask = 48;
497 
498 	return 0;
499 }
500 
501 /*
502  * goya_pci_bars_map - Map PCI BARS of Goya device
503  *
504  * @hdev: pointer to hl_device structure
505  *
506  * Request PCI regions and map them to kernel virtual addresses.
507  * Returns 0 on success
508  *
509  */
510 static int goya_pci_bars_map(struct hl_device *hdev)
511 {
512 	static const char * const name[] = {"SRAM_CFG", "MSIX", "DDR"};
513 	bool is_wc[3] = {false, false, true};
514 	int rc;
515 
516 	rc = hl_pci_bars_map(hdev, name, is_wc);
517 	if (rc)
518 		return rc;
519 
520 	hdev->rmmio = hdev->pcie_bar[SRAM_CFG_BAR_ID] +
521 			(CFG_BASE - SRAM_BASE_ADDR);
522 
523 	return 0;
524 }
525 
526 static u64 goya_set_ddr_bar_base(struct hl_device *hdev, u64 addr)
527 {
528 	struct goya_device *goya = hdev->asic_specific;
529 	struct hl_inbound_pci_region pci_region;
530 	u64 old_addr = addr;
531 	int rc;
532 
533 	if ((goya) && (goya->ddr_bar_cur_addr == addr))
534 		return old_addr;
535 
536 	/* Inbound Region 1 - Bar 4 - Point to DDR */
537 	pci_region.mode = PCI_BAR_MATCH_MODE;
538 	pci_region.bar = DDR_BAR_ID;
539 	pci_region.addr = addr;
540 	rc = hl_pci_set_inbound_region(hdev, 1, &pci_region);
541 	if (rc)
542 		return U64_MAX;
543 
544 	if (goya) {
545 		old_addr = goya->ddr_bar_cur_addr;
546 		goya->ddr_bar_cur_addr = addr;
547 	}
548 
549 	return old_addr;
550 }
551 
552 /*
553  * goya_init_iatu - Initialize the iATU unit inside the PCI controller
554  *
555  * @hdev: pointer to hl_device structure
556  *
557  * This is needed in case the firmware doesn't initialize the iATU
558  *
559  */
560 static int goya_init_iatu(struct hl_device *hdev)
561 {
562 	struct hl_inbound_pci_region inbound_region;
563 	struct hl_outbound_pci_region outbound_region;
564 	int rc;
565 
566 	if (hdev->asic_prop.iatu_done_by_fw)
567 		return 0;
568 
569 	/* Inbound Region 0 - Bar 0 - Point to SRAM and CFG */
570 	inbound_region.mode = PCI_BAR_MATCH_MODE;
571 	inbound_region.bar = SRAM_CFG_BAR_ID;
572 	inbound_region.addr = SRAM_BASE_ADDR;
573 	rc = hl_pci_set_inbound_region(hdev, 0, &inbound_region);
574 	if (rc)
575 		goto done;
576 
577 	/* Inbound Region 1 - Bar 4 - Point to DDR */
578 	inbound_region.mode = PCI_BAR_MATCH_MODE;
579 	inbound_region.bar = DDR_BAR_ID;
580 	inbound_region.addr = DRAM_PHYS_BASE;
581 	rc = hl_pci_set_inbound_region(hdev, 1, &inbound_region);
582 	if (rc)
583 		goto done;
584 
585 	/* Outbound Region 0 - Point to Host  */
586 	outbound_region.addr = HOST_PHYS_BASE;
587 	outbound_region.size = HOST_PHYS_SIZE;
588 	rc = hl_pci_set_outbound_region(hdev, &outbound_region);
589 
590 done:
591 	return rc;
592 }
593 
594 static enum hl_device_hw_state goya_get_hw_state(struct hl_device *hdev)
595 {
596 	return RREG32(mmHW_STATE);
597 }
598 
599 /*
600  * goya_early_init - GOYA early initialization code
601  *
602  * @hdev: pointer to hl_device structure
603  *
604  * Verify PCI bars
605  * Set DMA masks
606  * PCI controller initialization
607  * Map PCI bars
608  *
609  */
610 static int goya_early_init(struct hl_device *hdev)
611 {
612 	struct asic_fixed_properties *prop = &hdev->asic_prop;
613 	struct pci_dev *pdev = hdev->pdev;
614 	resource_size_t pci_bar_size;
615 	u32 fw_boot_status, val;
616 	int rc;
617 
618 	rc = goya_set_fixed_properties(hdev);
619 	if (rc) {
620 		dev_err(hdev->dev, "Failed to get fixed properties\n");
621 		return rc;
622 	}
623 
624 	/* Check BAR sizes */
625 	pci_bar_size = pci_resource_len(pdev, SRAM_CFG_BAR_ID);
626 
627 	if (pci_bar_size != CFG_BAR_SIZE) {
628 		dev_err(hdev->dev, "Not " HL_NAME "? BAR %d size %pa, expecting %llu\n",
629 			SRAM_CFG_BAR_ID, &pci_bar_size, CFG_BAR_SIZE);
630 		rc = -ENODEV;
631 		goto free_queue_props;
632 	}
633 
634 	pci_bar_size = pci_resource_len(pdev, MSIX_BAR_ID);
635 
636 	if (pci_bar_size != MSIX_BAR_SIZE) {
637 		dev_err(hdev->dev, "Not " HL_NAME "? BAR %d size %pa, expecting %llu\n",
638 			MSIX_BAR_ID, &pci_bar_size, MSIX_BAR_SIZE);
639 		rc = -ENODEV;
640 		goto free_queue_props;
641 	}
642 
643 	prop->dram_pci_bar_size = pci_resource_len(pdev, DDR_BAR_ID);
644 	hdev->dram_pci_bar_start = pci_resource_start(pdev, DDR_BAR_ID);
645 
646 	/* If FW security is enabled at this point it means no access to ELBI */
647 	if (hdev->asic_prop.fw_security_enabled) {
648 		hdev->asic_prop.iatu_done_by_fw = true;
649 		goto pci_init;
650 	}
651 
652 	rc = hl_pci_elbi_read(hdev, CFG_BASE + mmCPU_BOOT_DEV_STS0,
653 				&fw_boot_status);
654 	if (rc)
655 		goto free_queue_props;
656 
657 	/* Check whether FW is configuring iATU */
658 	if ((fw_boot_status & CPU_BOOT_DEV_STS0_ENABLED) &&
659 			(fw_boot_status & CPU_BOOT_DEV_STS0_FW_IATU_CONF_EN))
660 		hdev->asic_prop.iatu_done_by_fw = true;
661 
662 pci_init:
663 	rc = hl_pci_init(hdev);
664 	if (rc)
665 		goto free_queue_props;
666 
667 	/* Before continuing in the initialization, we need to read the preboot
668 	 * version to determine whether we run with a security-enabled firmware
669 	 */
670 	rc = hl_fw_read_preboot_status(hdev);
671 	if (rc) {
672 		if (hdev->reset_on_preboot_fail)
673 			/* we are already on failure flow, so don't check if hw_fini fails. */
674 			hdev->asic_funcs->hw_fini(hdev, true, false);
675 		goto pci_fini;
676 	}
677 
678 	if (goya_get_hw_state(hdev) == HL_DEVICE_HW_STATE_DIRTY) {
679 		dev_dbg(hdev->dev, "H/W state is dirty, must reset before initializing\n");
680 		rc = hdev->asic_funcs->hw_fini(hdev, true, false);
681 		if (rc) {
682 			dev_err(hdev->dev, "failed to reset HW in dirty state (%d)\n", rc);
683 			goto pci_fini;
684 		}
685 	}
686 
687 	if (!hdev->pldm) {
688 		val = RREG32(mmPSOC_GLOBAL_CONF_BOOT_STRAP_PINS);
689 		if (val & PSOC_GLOBAL_CONF_BOOT_STRAP_PINS_SRIOV_EN_MASK)
690 			dev_warn(hdev->dev,
691 				"PCI strap is not configured correctly, PCI bus errors may occur\n");
692 	}
693 
694 	return 0;
695 
696 pci_fini:
697 	hl_pci_fini(hdev);
698 free_queue_props:
699 	kfree(hdev->asic_prop.hw_queues_props);
700 	return rc;
701 }
702 
703 /*
704  * goya_early_fini - GOYA early finalization code
705  *
706  * @hdev: pointer to hl_device structure
707  *
708  * Unmap PCI bars
709  *
710  */
711 static int goya_early_fini(struct hl_device *hdev)
712 {
713 	kfree(hdev->asic_prop.hw_queues_props);
714 	hl_pci_fini(hdev);
715 
716 	return 0;
717 }
718 
719 static void goya_mmu_prepare_reg(struct hl_device *hdev, u64 reg, u32 asid)
720 {
721 	/* mask to zero the MMBP and ASID bits */
722 	WREG32_AND(reg, ~0x7FF);
723 	WREG32_OR(reg, asid);
724 }
725 
726 static void goya_qman0_set_security(struct hl_device *hdev, bool secure)
727 {
728 	struct goya_device *goya = hdev->asic_specific;
729 
730 	if (!(goya->hw_cap_initialized & HW_CAP_MMU))
731 		return;
732 
733 	if (secure)
734 		WREG32(mmDMA_QM_0_GLBL_PROT, QMAN_DMA_FULLY_TRUSTED);
735 	else
736 		WREG32(mmDMA_QM_0_GLBL_PROT, QMAN_DMA_PARTLY_TRUSTED);
737 
738 	RREG32(mmDMA_QM_0_GLBL_PROT);
739 }
740 
741 /*
742  * goya_fetch_psoc_frequency - Fetch PSOC frequency values
743  *
744  * @hdev: pointer to hl_device structure
745  *
746  */
747 static void goya_fetch_psoc_frequency(struct hl_device *hdev)
748 {
749 	struct asic_fixed_properties *prop = &hdev->asic_prop;
750 	u32 nr = 0, nf = 0, od = 0, div_fctr = 0, pll_clk, div_sel;
751 	u16 pll_freq_arr[HL_PLL_NUM_OUTPUTS], freq;
752 	int rc;
753 
754 	if (hdev->asic_prop.fw_security_enabled) {
755 		struct goya_device *goya = hdev->asic_specific;
756 
757 		if (!(goya->hw_cap_initialized & HW_CAP_CPU_Q))
758 			return;
759 
760 		rc = hl_fw_cpucp_pll_info_get(hdev, HL_GOYA_PCI_PLL,
761 				pll_freq_arr);
762 
763 		if (rc)
764 			return;
765 
766 		freq = pll_freq_arr[1];
767 	} else {
768 		div_fctr = RREG32(mmPSOC_PCI_PLL_DIV_FACTOR_1);
769 		div_sel = RREG32(mmPSOC_PCI_PLL_DIV_SEL_1);
770 		nr = RREG32(mmPSOC_PCI_PLL_NR);
771 		nf = RREG32(mmPSOC_PCI_PLL_NF);
772 		od = RREG32(mmPSOC_PCI_PLL_OD);
773 
774 		if (div_sel == DIV_SEL_REF_CLK ||
775 				div_sel == DIV_SEL_DIVIDED_REF) {
776 			if (div_sel == DIV_SEL_REF_CLK)
777 				freq = PLL_REF_CLK;
778 			else
779 				freq = PLL_REF_CLK / (div_fctr + 1);
780 		} else if (div_sel == DIV_SEL_PLL_CLK ||
781 				div_sel == DIV_SEL_DIVIDED_PLL) {
782 			pll_clk = PLL_REF_CLK * (nf + 1) /
783 					((nr + 1) * (od + 1));
784 			if (div_sel == DIV_SEL_PLL_CLK)
785 				freq = pll_clk;
786 			else
787 				freq = pll_clk / (div_fctr + 1);
788 		} else {
789 			dev_warn(hdev->dev,
790 				"Received invalid div select value: %d",
791 				div_sel);
792 			freq = 0;
793 		}
794 	}
795 
796 	prop->psoc_timestamp_frequency = freq;
797 	prop->psoc_pci_pll_nr = nr;
798 	prop->psoc_pci_pll_nf = nf;
799 	prop->psoc_pci_pll_od = od;
800 	prop->psoc_pci_pll_div_factor = div_fctr;
801 }
802 
803 /*
804  * goya_set_frequency - set the frequency of the device
805  *
806  * @hdev: pointer to habanalabs device structure
807  * @freq: the new frequency value
808  *
809  * Change the frequency if needed. This function has no protection against
810  * concurrency, therefore it is assumed that the calling function has protected
811  * itself against the case of calling this function from multiple threads with
812  * different values
813  *
814  * Returns 0 if no change was done, otherwise returns 1
815  */
816 int goya_set_frequency(struct hl_device *hdev, enum hl_pll_frequency freq)
817 {
818 	struct goya_device *goya = hdev->asic_specific;
819 
820 	if ((goya->pm_mng_profile == PM_MANUAL) ||
821 			(goya->curr_pll_profile == freq))
822 		return 0;
823 
824 	dev_dbg(hdev->dev, "Changing device frequency to %s\n",
825 		freq == PLL_HIGH ? "high" : "low");
826 
827 	goya_set_pll_profile(hdev, freq);
828 
829 	goya->curr_pll_profile = freq;
830 
831 	return 1;
832 }
833 
834 static void goya_set_freq_to_low_job(struct work_struct *work)
835 {
836 	struct goya_work_freq *goya_work = container_of(work,
837 						struct goya_work_freq,
838 						work_freq.work);
839 	struct hl_device *hdev = goya_work->hdev;
840 
841 	mutex_lock(&hdev->fpriv_list_lock);
842 
843 	if (!hdev->is_compute_ctx_active)
844 		goya_set_frequency(hdev, PLL_LOW);
845 
846 	mutex_unlock(&hdev->fpriv_list_lock);
847 
848 	schedule_delayed_work(&goya_work->work_freq,
849 			usecs_to_jiffies(HL_PLL_LOW_JOB_FREQ_USEC));
850 }
851 
852 int goya_late_init(struct hl_device *hdev)
853 {
854 	struct asic_fixed_properties *prop = &hdev->asic_prop;
855 	struct goya_device *goya = hdev->asic_specific;
856 	int rc;
857 
858 	goya_fetch_psoc_frequency(hdev);
859 
860 	rc = goya_mmu_clear_pgt_range(hdev);
861 	if (rc) {
862 		dev_err(hdev->dev,
863 			"Failed to clear MMU page tables range %d\n", rc);
864 		return rc;
865 	}
866 
867 	rc = goya_mmu_set_dram_default_page(hdev);
868 	if (rc) {
869 		dev_err(hdev->dev, "Failed to set DRAM default page %d\n", rc);
870 		return rc;
871 	}
872 
873 	rc = goya_mmu_add_mappings_for_device_cpu(hdev);
874 	if (rc)
875 		return rc;
876 
877 	rc = goya_init_cpu_queues(hdev);
878 	if (rc)
879 		return rc;
880 
881 	rc = goya_test_cpu_queue(hdev);
882 	if (rc)
883 		return rc;
884 
885 	rc = goya_cpucp_info_get(hdev);
886 	if (rc) {
887 		dev_err(hdev->dev, "Failed to get cpucp info %d\n", rc);
888 		return rc;
889 	}
890 
891 	/* Now that we have the DRAM size in ASIC prop, we need to check
892 	 * its size and configure the DMA_IF DDR wrap protection (which is in
893 	 * the MMU block) accordingly. The value is the log2 of the DRAM size
894 	 */
895 	WREG32(mmMMU_LOG2_DDR_SIZE, ilog2(prop->dram_size));
896 
897 	rc = hl_fw_send_pci_access_msg(hdev, CPUCP_PACKET_ENABLE_PCI_ACCESS, 0x0);
898 	if (rc) {
899 		dev_err(hdev->dev,
900 			"Failed to enable PCI access from CPU %d\n", rc);
901 		return rc;
902 	}
903 
904 	/* force setting to low frequency */
905 	goya->curr_pll_profile = PLL_LOW;
906 
907 	goya->pm_mng_profile = PM_AUTO;
908 
909 	goya_set_pll_profile(hdev, PLL_LOW);
910 
911 	schedule_delayed_work(&goya->goya_work->work_freq,
912 		usecs_to_jiffies(HL_PLL_LOW_JOB_FREQ_USEC));
913 
914 	return 0;
915 }
916 
917 /*
918  * goya_late_fini - GOYA late tear-down code
919  *
920  * @hdev: pointer to hl_device structure
921  *
922  * Free sensors allocated structures
923  */
924 void goya_late_fini(struct hl_device *hdev)
925 {
926 	struct goya_device *goya = hdev->asic_specific;
927 
928 	cancel_delayed_work_sync(&goya->goya_work->work_freq);
929 
930 	hl_hwmon_release_resources(hdev);
931 }
932 
933 static void goya_set_pci_memory_regions(struct hl_device *hdev)
934 {
935 	struct asic_fixed_properties *prop = &hdev->asic_prop;
936 	struct pci_mem_region *region;
937 
938 	/* CFG */
939 	region = &hdev->pci_mem_region[PCI_REGION_CFG];
940 	region->region_base = CFG_BASE;
941 	region->region_size = CFG_SIZE;
942 	region->offset_in_bar = CFG_BASE - SRAM_BASE_ADDR;
943 	region->bar_size = CFG_BAR_SIZE;
944 	region->bar_id = SRAM_CFG_BAR_ID;
945 	region->used = 1;
946 
947 	/* SRAM */
948 	region = &hdev->pci_mem_region[PCI_REGION_SRAM];
949 	region->region_base = SRAM_BASE_ADDR;
950 	region->region_size = SRAM_SIZE;
951 	region->offset_in_bar = 0;
952 	region->bar_size = CFG_BAR_SIZE;
953 	region->bar_id = SRAM_CFG_BAR_ID;
954 	region->used = 1;
955 
956 	/* DRAM */
957 	region = &hdev->pci_mem_region[PCI_REGION_DRAM];
958 	region->region_base = DRAM_PHYS_BASE;
959 	region->region_size = hdev->asic_prop.dram_size;
960 	region->offset_in_bar = 0;
961 	region->bar_size = prop->dram_pci_bar_size;
962 	region->bar_id = DDR_BAR_ID;
963 	region->used = 1;
964 }
965 
966 /*
967  * goya_sw_init - Goya software initialization code
968  *
969  * @hdev: pointer to hl_device structure
970  *
971  */
972 static int goya_sw_init(struct hl_device *hdev)
973 {
974 	struct goya_device *goya;
975 	int rc;
976 
977 	/* Allocate device structure */
978 	goya = kzalloc(sizeof(*goya), GFP_KERNEL);
979 	if (!goya)
980 		return -ENOMEM;
981 
982 	/* according to goya_init_iatu */
983 	goya->ddr_bar_cur_addr = DRAM_PHYS_BASE;
984 
985 	goya->mme_clk = GOYA_PLL_FREQ_LOW;
986 	goya->tpc_clk = GOYA_PLL_FREQ_LOW;
987 	goya->ic_clk = GOYA_PLL_FREQ_LOW;
988 
989 	hdev->asic_specific = goya;
990 
991 	/* Create DMA pool for small allocations */
992 	hdev->dma_pool = dma_pool_create(dev_name(hdev->dev),
993 			&hdev->pdev->dev, GOYA_DMA_POOL_BLK_SIZE, 8, 0);
994 	if (!hdev->dma_pool) {
995 		dev_err(hdev->dev, "failed to create DMA pool\n");
996 		rc = -ENOMEM;
997 		goto free_goya_device;
998 	}
999 
1000 	hdev->cpu_accessible_dma_mem = hl_asic_dma_alloc_coherent(hdev, HL_CPU_ACCESSIBLE_MEM_SIZE,
1001 							&hdev->cpu_accessible_dma_address,
1002 							GFP_KERNEL | __GFP_ZERO);
1003 
1004 	if (!hdev->cpu_accessible_dma_mem) {
1005 		rc = -ENOMEM;
1006 		goto free_dma_pool;
1007 	}
1008 
1009 	dev_dbg(hdev->dev, "cpu accessible memory at bus address %pad\n",
1010 		&hdev->cpu_accessible_dma_address);
1011 
1012 	hdev->cpu_accessible_dma_pool = gen_pool_create(ilog2(32), -1);
1013 	if (!hdev->cpu_accessible_dma_pool) {
1014 		dev_err(hdev->dev,
1015 			"Failed to create CPU accessible DMA pool\n");
1016 		rc = -ENOMEM;
1017 		goto free_cpu_dma_mem;
1018 	}
1019 
1020 	rc = gen_pool_add(hdev->cpu_accessible_dma_pool,
1021 				(uintptr_t) hdev->cpu_accessible_dma_mem,
1022 				HL_CPU_ACCESSIBLE_MEM_SIZE, -1);
1023 	if (rc) {
1024 		dev_err(hdev->dev,
1025 			"Failed to add memory to CPU accessible DMA pool\n");
1026 		rc = -EFAULT;
1027 		goto free_cpu_accessible_dma_pool;
1028 	}
1029 
1030 	spin_lock_init(&goya->hw_queues_lock);
1031 	hdev->supports_coresight = true;
1032 	hdev->asic_prop.supports_compute_reset = true;
1033 	hdev->asic_prop.allow_inference_soft_reset = true;
1034 	hdev->supports_wait_for_multi_cs = false;
1035 	hdev->supports_ctx_switch = true;
1036 
1037 	hdev->asic_funcs->set_pci_memory_regions(hdev);
1038 
1039 	goya->goya_work = kmalloc(sizeof(struct goya_work_freq), GFP_KERNEL);
1040 	if (!goya->goya_work) {
1041 		rc = -ENOMEM;
1042 		goto free_cpu_accessible_dma_pool;
1043 	}
1044 
1045 	goya->goya_work->hdev = hdev;
1046 	INIT_DELAYED_WORK(&goya->goya_work->work_freq, goya_set_freq_to_low_job);
1047 
1048 	return 0;
1049 
1050 free_cpu_accessible_dma_pool:
1051 	gen_pool_destroy(hdev->cpu_accessible_dma_pool);
1052 free_cpu_dma_mem:
1053 	hl_asic_dma_free_coherent(hdev, HL_CPU_ACCESSIBLE_MEM_SIZE, hdev->cpu_accessible_dma_mem,
1054 					hdev->cpu_accessible_dma_address);
1055 free_dma_pool:
1056 	dma_pool_destroy(hdev->dma_pool);
1057 free_goya_device:
1058 	kfree(goya);
1059 
1060 	return rc;
1061 }
1062 
1063 /*
1064  * goya_sw_fini - Goya software tear-down code
1065  *
1066  * @hdev: pointer to hl_device structure
1067  *
1068  */
1069 static int goya_sw_fini(struct hl_device *hdev)
1070 {
1071 	struct goya_device *goya = hdev->asic_specific;
1072 
1073 	gen_pool_destroy(hdev->cpu_accessible_dma_pool);
1074 
1075 	hl_asic_dma_free_coherent(hdev, HL_CPU_ACCESSIBLE_MEM_SIZE, hdev->cpu_accessible_dma_mem,
1076 					hdev->cpu_accessible_dma_address);
1077 
1078 	dma_pool_destroy(hdev->dma_pool);
1079 
1080 	kfree(goya->goya_work);
1081 	kfree(goya);
1082 
1083 	return 0;
1084 }
1085 
1086 static void goya_init_dma_qman(struct hl_device *hdev, int dma_id,
1087 		dma_addr_t bus_address)
1088 {
1089 	struct goya_device *goya = hdev->asic_specific;
1090 	u32 mtr_base_lo, mtr_base_hi;
1091 	u32 so_base_lo, so_base_hi;
1092 	u32 gic_base_lo, gic_base_hi;
1093 	u32 reg_off = dma_id * (mmDMA_QM_1_PQ_PI - mmDMA_QM_0_PQ_PI);
1094 	u32 dma_err_cfg = QMAN_DMA_ERR_MSG_EN;
1095 
1096 	mtr_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_MON_PAY_ADDRL_0);
1097 	mtr_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_MON_PAY_ADDRL_0);
1098 	so_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1099 	so_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1100 
1101 	gic_base_lo =
1102 		lower_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
1103 	gic_base_hi =
1104 		upper_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
1105 
1106 	WREG32(mmDMA_QM_0_PQ_BASE_LO + reg_off, lower_32_bits(bus_address));
1107 	WREG32(mmDMA_QM_0_PQ_BASE_HI + reg_off, upper_32_bits(bus_address));
1108 
1109 	WREG32(mmDMA_QM_0_PQ_SIZE + reg_off, ilog2(HL_QUEUE_LENGTH));
1110 	WREG32(mmDMA_QM_0_PQ_PI + reg_off, 0);
1111 	WREG32(mmDMA_QM_0_PQ_CI + reg_off, 0);
1112 
1113 	WREG32(mmDMA_QM_0_CP_MSG_BASE0_ADDR_LO + reg_off, mtr_base_lo);
1114 	WREG32(mmDMA_QM_0_CP_MSG_BASE0_ADDR_HI + reg_off, mtr_base_hi);
1115 	WREG32(mmDMA_QM_0_CP_MSG_BASE1_ADDR_LO + reg_off, so_base_lo);
1116 	WREG32(mmDMA_QM_0_CP_MSG_BASE1_ADDR_HI + reg_off, so_base_hi);
1117 	WREG32(mmDMA_QM_0_GLBL_ERR_ADDR_LO + reg_off, gic_base_lo);
1118 	WREG32(mmDMA_QM_0_GLBL_ERR_ADDR_HI + reg_off, gic_base_hi);
1119 	WREG32(mmDMA_QM_0_GLBL_ERR_WDATA + reg_off,
1120 			GOYA_ASYNC_EVENT_ID_DMA0_QM + dma_id);
1121 
1122 	/* PQ has buffer of 2 cache lines, while CQ has 8 lines */
1123 	WREG32(mmDMA_QM_0_PQ_CFG1 + reg_off, 0x00020002);
1124 	WREG32(mmDMA_QM_0_CQ_CFG1 + reg_off, 0x00080008);
1125 
1126 	if (goya->hw_cap_initialized & HW_CAP_MMU)
1127 		WREG32(mmDMA_QM_0_GLBL_PROT + reg_off, QMAN_DMA_PARTLY_TRUSTED);
1128 	else
1129 		WREG32(mmDMA_QM_0_GLBL_PROT + reg_off, QMAN_DMA_FULLY_TRUSTED);
1130 
1131 	if (hdev->stop_on_err)
1132 		dma_err_cfg |= 1 << DMA_QM_0_GLBL_ERR_CFG_DMA_STOP_ON_ERR_SHIFT;
1133 
1134 	WREG32(mmDMA_QM_0_GLBL_ERR_CFG + reg_off, dma_err_cfg);
1135 	WREG32(mmDMA_QM_0_GLBL_CFG0 + reg_off, QMAN_DMA_ENABLE);
1136 }
1137 
1138 static void goya_init_dma_ch(struct hl_device *hdev, int dma_id)
1139 {
1140 	u32 gic_base_lo, gic_base_hi;
1141 	u64 sob_addr;
1142 	u32 reg_off = dma_id * (mmDMA_CH_1_CFG1 - mmDMA_CH_0_CFG1);
1143 
1144 	gic_base_lo =
1145 		lower_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
1146 	gic_base_hi =
1147 		upper_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
1148 
1149 	WREG32(mmDMA_CH_0_ERRMSG_ADDR_LO + reg_off, gic_base_lo);
1150 	WREG32(mmDMA_CH_0_ERRMSG_ADDR_HI + reg_off, gic_base_hi);
1151 	WREG32(mmDMA_CH_0_ERRMSG_WDATA + reg_off,
1152 			GOYA_ASYNC_EVENT_ID_DMA0_CH + dma_id);
1153 
1154 	if (dma_id)
1155 		sob_addr = CFG_BASE + mmSYNC_MNGR_SOB_OBJ_1000 +
1156 				(dma_id - 1) * 4;
1157 	else
1158 		sob_addr = CFG_BASE + mmSYNC_MNGR_SOB_OBJ_1007;
1159 
1160 	WREG32(mmDMA_CH_0_WR_COMP_ADDR_HI + reg_off, upper_32_bits(sob_addr));
1161 	WREG32(mmDMA_CH_0_WR_COMP_WDATA + reg_off, 0x80000001);
1162 }
1163 
1164 /*
1165  * goya_init_dma_qmans - Initialize QMAN DMA registers
1166  *
1167  * @hdev: pointer to hl_device structure
1168  *
1169  * Initialize the H/W registers of the QMAN DMA channels
1170  *
1171  */
1172 void goya_init_dma_qmans(struct hl_device *hdev)
1173 {
1174 	struct goya_device *goya = hdev->asic_specific;
1175 	struct hl_hw_queue *q;
1176 	int i;
1177 
1178 	if (goya->hw_cap_initialized & HW_CAP_DMA)
1179 		return;
1180 
1181 	q = &hdev->kernel_queues[0];
1182 
1183 	for (i = 0 ; i < NUMBER_OF_EXT_HW_QUEUES ; i++, q++) {
1184 		q->cq_id = q->msi_vec = i;
1185 		goya_init_dma_qman(hdev, i, q->bus_address);
1186 		goya_init_dma_ch(hdev, i);
1187 	}
1188 
1189 	goya->hw_cap_initialized |= HW_CAP_DMA;
1190 }
1191 
1192 /*
1193  * goya_disable_external_queues - Disable external queues
1194  *
1195  * @hdev: pointer to hl_device structure
1196  *
1197  */
1198 static void goya_disable_external_queues(struct hl_device *hdev)
1199 {
1200 	struct goya_device *goya = hdev->asic_specific;
1201 
1202 	if (!(goya->hw_cap_initialized & HW_CAP_DMA))
1203 		return;
1204 
1205 	WREG32(mmDMA_QM_0_GLBL_CFG0, 0);
1206 	WREG32(mmDMA_QM_1_GLBL_CFG0, 0);
1207 	WREG32(mmDMA_QM_2_GLBL_CFG0, 0);
1208 	WREG32(mmDMA_QM_3_GLBL_CFG0, 0);
1209 	WREG32(mmDMA_QM_4_GLBL_CFG0, 0);
1210 }
1211 
1212 static int goya_stop_queue(struct hl_device *hdev, u32 cfg_reg,
1213 				u32 cp_sts_reg, u32 glbl_sts0_reg)
1214 {
1215 	int rc;
1216 	u32 status;
1217 
1218 	/* use the values of TPC0 as they are all the same*/
1219 
1220 	WREG32(cfg_reg, 1 << TPC0_QM_GLBL_CFG1_CP_STOP_SHIFT);
1221 
1222 	status = RREG32(cp_sts_reg);
1223 	if (status & TPC0_QM_CP_STS_FENCE_IN_PROGRESS_MASK) {
1224 		rc = hl_poll_timeout(
1225 			hdev,
1226 			cp_sts_reg,
1227 			status,
1228 			!(status & TPC0_QM_CP_STS_FENCE_IN_PROGRESS_MASK),
1229 			1000,
1230 			QMAN_FENCE_TIMEOUT_USEC);
1231 
1232 		/* if QMAN is stuck in fence no need to check for stop */
1233 		if (rc)
1234 			return 0;
1235 	}
1236 
1237 	rc = hl_poll_timeout(
1238 		hdev,
1239 		glbl_sts0_reg,
1240 		status,
1241 		(status & TPC0_QM_GLBL_STS0_CP_IS_STOP_MASK),
1242 		1000,
1243 		QMAN_STOP_TIMEOUT_USEC);
1244 
1245 	if (rc) {
1246 		dev_err(hdev->dev,
1247 			"Timeout while waiting for QMAN to stop\n");
1248 		return -EINVAL;
1249 	}
1250 
1251 	return 0;
1252 }
1253 
1254 /*
1255  * goya_stop_external_queues - Stop external queues
1256  *
1257  * @hdev: pointer to hl_device structure
1258  *
1259  * Returns 0 on success
1260  *
1261  */
1262 static int goya_stop_external_queues(struct hl_device *hdev)
1263 {
1264 	int rc, retval = 0;
1265 
1266 	struct goya_device *goya = hdev->asic_specific;
1267 
1268 	if (!(goya->hw_cap_initialized & HW_CAP_DMA))
1269 		return retval;
1270 
1271 	rc = goya_stop_queue(hdev,
1272 			mmDMA_QM_0_GLBL_CFG1,
1273 			mmDMA_QM_0_CP_STS,
1274 			mmDMA_QM_0_GLBL_STS0);
1275 
1276 	if (rc) {
1277 		dev_err(hdev->dev, "failed to stop DMA QMAN 0\n");
1278 		retval = -EIO;
1279 	}
1280 
1281 	rc = goya_stop_queue(hdev,
1282 			mmDMA_QM_1_GLBL_CFG1,
1283 			mmDMA_QM_1_CP_STS,
1284 			mmDMA_QM_1_GLBL_STS0);
1285 
1286 	if (rc) {
1287 		dev_err(hdev->dev, "failed to stop DMA QMAN 1\n");
1288 		retval = -EIO;
1289 	}
1290 
1291 	rc = goya_stop_queue(hdev,
1292 			mmDMA_QM_2_GLBL_CFG1,
1293 			mmDMA_QM_2_CP_STS,
1294 			mmDMA_QM_2_GLBL_STS0);
1295 
1296 	if (rc) {
1297 		dev_err(hdev->dev, "failed to stop DMA QMAN 2\n");
1298 		retval = -EIO;
1299 	}
1300 
1301 	rc = goya_stop_queue(hdev,
1302 			mmDMA_QM_3_GLBL_CFG1,
1303 			mmDMA_QM_3_CP_STS,
1304 			mmDMA_QM_3_GLBL_STS0);
1305 
1306 	if (rc) {
1307 		dev_err(hdev->dev, "failed to stop DMA QMAN 3\n");
1308 		retval = -EIO;
1309 	}
1310 
1311 	rc = goya_stop_queue(hdev,
1312 			mmDMA_QM_4_GLBL_CFG1,
1313 			mmDMA_QM_4_CP_STS,
1314 			mmDMA_QM_4_GLBL_STS0);
1315 
1316 	if (rc) {
1317 		dev_err(hdev->dev, "failed to stop DMA QMAN 4\n");
1318 		retval = -EIO;
1319 	}
1320 
1321 	return retval;
1322 }
1323 
1324 /*
1325  * goya_init_cpu_queues - Initialize PQ/CQ/EQ of CPU
1326  *
1327  * @hdev: pointer to hl_device structure
1328  *
1329  * Returns 0 on success
1330  *
1331  */
1332 int goya_init_cpu_queues(struct hl_device *hdev)
1333 {
1334 	struct goya_device *goya = hdev->asic_specific;
1335 	struct asic_fixed_properties *prop = &hdev->asic_prop;
1336 	struct hl_eq *eq;
1337 	u32 status;
1338 	struct hl_hw_queue *cpu_pq = &hdev->kernel_queues[GOYA_QUEUE_ID_CPU_PQ];
1339 	int err;
1340 
1341 	if (!hdev->cpu_queues_enable)
1342 		return 0;
1343 
1344 	if (goya->hw_cap_initialized & HW_CAP_CPU_Q)
1345 		return 0;
1346 
1347 	eq = &hdev->event_queue;
1348 
1349 	WREG32(mmCPU_PQ_BASE_ADDR_LOW, lower_32_bits(cpu_pq->bus_address));
1350 	WREG32(mmCPU_PQ_BASE_ADDR_HIGH, upper_32_bits(cpu_pq->bus_address));
1351 
1352 	WREG32(mmCPU_EQ_BASE_ADDR_LOW, lower_32_bits(eq->bus_address));
1353 	WREG32(mmCPU_EQ_BASE_ADDR_HIGH, upper_32_bits(eq->bus_address));
1354 
1355 	WREG32(mmCPU_CQ_BASE_ADDR_LOW,
1356 			lower_32_bits(VA_CPU_ACCESSIBLE_MEM_ADDR));
1357 	WREG32(mmCPU_CQ_BASE_ADDR_HIGH,
1358 			upper_32_bits(VA_CPU_ACCESSIBLE_MEM_ADDR));
1359 
1360 	WREG32(mmCPU_PQ_LENGTH, HL_QUEUE_SIZE_IN_BYTES);
1361 	WREG32(mmCPU_EQ_LENGTH, HL_EQ_SIZE_IN_BYTES);
1362 	WREG32(mmCPU_CQ_LENGTH, HL_CPU_ACCESSIBLE_MEM_SIZE);
1363 
1364 	/* Used for EQ CI */
1365 	WREG32(mmCPU_EQ_CI, 0);
1366 
1367 	WREG32(mmCPU_IF_PF_PQ_PI, 0);
1368 
1369 	WREG32(mmCPU_PQ_INIT_STATUS, PQ_INIT_STATUS_READY_FOR_CP);
1370 
1371 	WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR,
1372 			GOYA_ASYNC_EVENT_ID_PI_UPDATE);
1373 
1374 	err = hl_poll_timeout(
1375 		hdev,
1376 		mmCPU_PQ_INIT_STATUS,
1377 		status,
1378 		(status == PQ_INIT_STATUS_READY_FOR_HOST),
1379 		1000,
1380 		GOYA_CPU_TIMEOUT_USEC);
1381 
1382 	if (err) {
1383 		dev_err(hdev->dev,
1384 			"Failed to setup communication with device CPU\n");
1385 		return -EIO;
1386 	}
1387 
1388 	/* update FW application security bits */
1389 	if (prop->fw_cpu_boot_dev_sts0_valid)
1390 		prop->fw_app_cpu_boot_dev_sts0 = RREG32(mmCPU_BOOT_DEV_STS0);
1391 
1392 	if (prop->fw_cpu_boot_dev_sts1_valid)
1393 		prop->fw_app_cpu_boot_dev_sts1 = RREG32(mmCPU_BOOT_DEV_STS1);
1394 
1395 	goya->hw_cap_initialized |= HW_CAP_CPU_Q;
1396 	return 0;
1397 }
1398 
1399 static void goya_set_pll_refclk(struct hl_device *hdev)
1400 {
1401 	WREG32(mmCPU_PLL_DIV_SEL_0, 0x0);
1402 	WREG32(mmCPU_PLL_DIV_SEL_1, 0x0);
1403 	WREG32(mmCPU_PLL_DIV_SEL_2, 0x0);
1404 	WREG32(mmCPU_PLL_DIV_SEL_3, 0x0);
1405 
1406 	WREG32(mmIC_PLL_DIV_SEL_0, 0x0);
1407 	WREG32(mmIC_PLL_DIV_SEL_1, 0x0);
1408 	WREG32(mmIC_PLL_DIV_SEL_2, 0x0);
1409 	WREG32(mmIC_PLL_DIV_SEL_3, 0x0);
1410 
1411 	WREG32(mmMC_PLL_DIV_SEL_0, 0x0);
1412 	WREG32(mmMC_PLL_DIV_SEL_1, 0x0);
1413 	WREG32(mmMC_PLL_DIV_SEL_2, 0x0);
1414 	WREG32(mmMC_PLL_DIV_SEL_3, 0x0);
1415 
1416 	WREG32(mmPSOC_MME_PLL_DIV_SEL_0, 0x0);
1417 	WREG32(mmPSOC_MME_PLL_DIV_SEL_1, 0x0);
1418 	WREG32(mmPSOC_MME_PLL_DIV_SEL_2, 0x0);
1419 	WREG32(mmPSOC_MME_PLL_DIV_SEL_3, 0x0);
1420 
1421 	WREG32(mmPSOC_PCI_PLL_DIV_SEL_0, 0x0);
1422 	WREG32(mmPSOC_PCI_PLL_DIV_SEL_1, 0x0);
1423 	WREG32(mmPSOC_PCI_PLL_DIV_SEL_2, 0x0);
1424 	WREG32(mmPSOC_PCI_PLL_DIV_SEL_3, 0x0);
1425 
1426 	WREG32(mmPSOC_EMMC_PLL_DIV_SEL_0, 0x0);
1427 	WREG32(mmPSOC_EMMC_PLL_DIV_SEL_1, 0x0);
1428 	WREG32(mmPSOC_EMMC_PLL_DIV_SEL_2, 0x0);
1429 	WREG32(mmPSOC_EMMC_PLL_DIV_SEL_3, 0x0);
1430 
1431 	WREG32(mmTPC_PLL_DIV_SEL_0, 0x0);
1432 	WREG32(mmTPC_PLL_DIV_SEL_1, 0x0);
1433 	WREG32(mmTPC_PLL_DIV_SEL_2, 0x0);
1434 	WREG32(mmTPC_PLL_DIV_SEL_3, 0x0);
1435 }
1436 
1437 static void goya_disable_clk_rlx(struct hl_device *hdev)
1438 {
1439 	WREG32(mmPSOC_MME_PLL_CLK_RLX_0, 0x100010);
1440 	WREG32(mmIC_PLL_CLK_RLX_0, 0x100010);
1441 }
1442 
1443 static void _goya_tpc_mbist_workaround(struct hl_device *hdev, u8 tpc_id)
1444 {
1445 	u64 tpc_eml_address;
1446 	u32 val, tpc_offset, tpc_eml_offset, tpc_slm_offset;
1447 	int err, slm_index;
1448 
1449 	tpc_offset = tpc_id * 0x40000;
1450 	tpc_eml_offset = tpc_id * 0x200000;
1451 	tpc_eml_address = (mmTPC0_EML_CFG_BASE + tpc_eml_offset - CFG_BASE);
1452 	tpc_slm_offset = tpc_eml_address + 0x100000;
1453 
1454 	/*
1455 	 * Workaround for Bug H2 #2443 :
1456 	 * "TPC SB is not initialized on chip reset"
1457 	 */
1458 
1459 	val = RREG32(mmTPC0_CFG_FUNC_MBIST_CNTRL + tpc_offset);
1460 	if (val & TPC0_CFG_FUNC_MBIST_CNTRL_MBIST_ACTIVE_MASK)
1461 		dev_warn(hdev->dev, "TPC%d MBIST ACTIVE is not cleared\n",
1462 			tpc_id);
1463 
1464 	WREG32(mmTPC0_CFG_FUNC_MBIST_PAT + tpc_offset, val & 0xFFFFF000);
1465 
1466 	WREG32(mmTPC0_CFG_FUNC_MBIST_MEM_0 + tpc_offset, 0x37FF);
1467 	WREG32(mmTPC0_CFG_FUNC_MBIST_MEM_1 + tpc_offset, 0x303F);
1468 	WREG32(mmTPC0_CFG_FUNC_MBIST_MEM_2 + tpc_offset, 0x71FF);
1469 	WREG32(mmTPC0_CFG_FUNC_MBIST_MEM_3 + tpc_offset, 0x71FF);
1470 	WREG32(mmTPC0_CFG_FUNC_MBIST_MEM_4 + tpc_offset, 0x70FF);
1471 	WREG32(mmTPC0_CFG_FUNC_MBIST_MEM_5 + tpc_offset, 0x70FF);
1472 	WREG32(mmTPC0_CFG_FUNC_MBIST_MEM_6 + tpc_offset, 0x70FF);
1473 	WREG32(mmTPC0_CFG_FUNC_MBIST_MEM_7 + tpc_offset, 0x70FF);
1474 	WREG32(mmTPC0_CFG_FUNC_MBIST_MEM_8 + tpc_offset, 0x70FF);
1475 	WREG32(mmTPC0_CFG_FUNC_MBIST_MEM_9 + tpc_offset, 0x70FF);
1476 
1477 	WREG32_OR(mmTPC0_CFG_FUNC_MBIST_CNTRL + tpc_offset,
1478 		1 << TPC0_CFG_FUNC_MBIST_CNTRL_MBIST_START_SHIFT);
1479 
1480 	err = hl_poll_timeout(
1481 		hdev,
1482 		mmTPC0_CFG_FUNC_MBIST_CNTRL + tpc_offset,
1483 		val,
1484 		(val & TPC0_CFG_FUNC_MBIST_CNTRL_MBIST_DONE_MASK),
1485 		1000,
1486 		HL_DEVICE_TIMEOUT_USEC);
1487 
1488 	if (err)
1489 		dev_err(hdev->dev,
1490 			"Timeout while waiting for TPC%d MBIST DONE\n", tpc_id);
1491 
1492 	WREG32_OR(mmTPC0_EML_CFG_DBG_CNT + tpc_eml_offset,
1493 		1 << TPC0_EML_CFG_DBG_CNT_CORE_RST_SHIFT);
1494 
1495 	msleep(GOYA_RESET_WAIT_MSEC);
1496 
1497 	WREG32_AND(mmTPC0_EML_CFG_DBG_CNT + tpc_eml_offset,
1498 		~(1 << TPC0_EML_CFG_DBG_CNT_CORE_RST_SHIFT));
1499 
1500 	msleep(GOYA_RESET_WAIT_MSEC);
1501 
1502 	for (slm_index = 0 ; slm_index < 256 ; slm_index++)
1503 		WREG32(tpc_slm_offset + (slm_index << 2), 0);
1504 
1505 	val = RREG32(tpc_slm_offset);
1506 }
1507 
1508 static void goya_tpc_mbist_workaround(struct hl_device *hdev)
1509 {
1510 	struct goya_device *goya = hdev->asic_specific;
1511 	int i;
1512 
1513 	if (hdev->pldm)
1514 		return;
1515 
1516 	if (goya->hw_cap_initialized & HW_CAP_TPC_MBIST)
1517 		return;
1518 
1519 	/* Workaround for H2 #2443 */
1520 
1521 	for (i = 0 ; i < TPC_MAX_NUM ; i++)
1522 		_goya_tpc_mbist_workaround(hdev, i);
1523 
1524 	goya->hw_cap_initialized |= HW_CAP_TPC_MBIST;
1525 }
1526 
1527 /*
1528  * goya_init_golden_registers - Initialize golden registers
1529  *
1530  * @hdev: pointer to hl_device structure
1531  *
1532  * Initialize the H/W registers of the device
1533  *
1534  */
1535 static void goya_init_golden_registers(struct hl_device *hdev)
1536 {
1537 	struct goya_device *goya = hdev->asic_specific;
1538 	u32 polynom[10], tpc_intr_mask, offset;
1539 	int i;
1540 
1541 	if (goya->hw_cap_initialized & HW_CAP_GOLDEN)
1542 		return;
1543 
1544 	polynom[0] = 0x00020080;
1545 	polynom[1] = 0x00401000;
1546 	polynom[2] = 0x00200800;
1547 	polynom[3] = 0x00002000;
1548 	polynom[4] = 0x00080200;
1549 	polynom[5] = 0x00040100;
1550 	polynom[6] = 0x00100400;
1551 	polynom[7] = 0x00004000;
1552 	polynom[8] = 0x00010000;
1553 	polynom[9] = 0x00008000;
1554 
1555 	/* Mask all arithmetic interrupts from TPC */
1556 	tpc_intr_mask = 0x7FFF;
1557 
1558 	for (i = 0, offset = 0 ; i < 6 ; i++, offset += 0x20000) {
1559 		WREG32(mmSRAM_Y0_X0_RTR_HBW_RD_RQ_L_ARB + offset, 0x302);
1560 		WREG32(mmSRAM_Y0_X1_RTR_HBW_RD_RQ_L_ARB + offset, 0x302);
1561 		WREG32(mmSRAM_Y0_X2_RTR_HBW_RD_RQ_L_ARB + offset, 0x302);
1562 		WREG32(mmSRAM_Y0_X3_RTR_HBW_RD_RQ_L_ARB + offset, 0x302);
1563 		WREG32(mmSRAM_Y0_X4_RTR_HBW_RD_RQ_L_ARB + offset, 0x302);
1564 
1565 		WREG32(mmSRAM_Y0_X0_RTR_HBW_DATA_L_ARB + offset, 0x204);
1566 		WREG32(mmSRAM_Y0_X1_RTR_HBW_DATA_L_ARB + offset, 0x204);
1567 		WREG32(mmSRAM_Y0_X2_RTR_HBW_DATA_L_ARB + offset, 0x204);
1568 		WREG32(mmSRAM_Y0_X3_RTR_HBW_DATA_L_ARB + offset, 0x204);
1569 		WREG32(mmSRAM_Y0_X4_RTR_HBW_DATA_L_ARB + offset, 0x204);
1570 
1571 
1572 		WREG32(mmSRAM_Y0_X0_RTR_HBW_DATA_E_ARB + offset, 0x206);
1573 		WREG32(mmSRAM_Y0_X1_RTR_HBW_DATA_E_ARB + offset, 0x206);
1574 		WREG32(mmSRAM_Y0_X2_RTR_HBW_DATA_E_ARB + offset, 0x206);
1575 		WREG32(mmSRAM_Y0_X3_RTR_HBW_DATA_E_ARB + offset, 0x207);
1576 		WREG32(mmSRAM_Y0_X4_RTR_HBW_DATA_E_ARB + offset, 0x207);
1577 
1578 		WREG32(mmSRAM_Y0_X0_RTR_HBW_DATA_W_ARB + offset, 0x207);
1579 		WREG32(mmSRAM_Y0_X1_RTR_HBW_DATA_W_ARB + offset, 0x207);
1580 		WREG32(mmSRAM_Y0_X2_RTR_HBW_DATA_W_ARB + offset, 0x206);
1581 		WREG32(mmSRAM_Y0_X3_RTR_HBW_DATA_W_ARB + offset, 0x206);
1582 		WREG32(mmSRAM_Y0_X4_RTR_HBW_DATA_W_ARB + offset, 0x206);
1583 
1584 		WREG32(mmSRAM_Y0_X0_RTR_HBW_WR_RS_E_ARB + offset, 0x101);
1585 		WREG32(mmSRAM_Y0_X1_RTR_HBW_WR_RS_E_ARB + offset, 0x102);
1586 		WREG32(mmSRAM_Y0_X2_RTR_HBW_WR_RS_E_ARB + offset, 0x103);
1587 		WREG32(mmSRAM_Y0_X3_RTR_HBW_WR_RS_E_ARB + offset, 0x104);
1588 		WREG32(mmSRAM_Y0_X4_RTR_HBW_WR_RS_E_ARB + offset, 0x105);
1589 
1590 		WREG32(mmSRAM_Y0_X0_RTR_HBW_WR_RS_W_ARB + offset, 0x105);
1591 		WREG32(mmSRAM_Y0_X1_RTR_HBW_WR_RS_W_ARB + offset, 0x104);
1592 		WREG32(mmSRAM_Y0_X2_RTR_HBW_WR_RS_W_ARB + offset, 0x103);
1593 		WREG32(mmSRAM_Y0_X3_RTR_HBW_WR_RS_W_ARB + offset, 0x102);
1594 		WREG32(mmSRAM_Y0_X4_RTR_HBW_WR_RS_W_ARB + offset, 0x101);
1595 	}
1596 
1597 	WREG32(mmMME_STORE_MAX_CREDIT, 0x21);
1598 	WREG32(mmMME_AGU, 0x0f0f0f10);
1599 	WREG32(mmMME_SEI_MASK, ~0x0);
1600 
1601 	WREG32(mmMME6_RTR_HBW_RD_RQ_N_ARB, 0x01010101);
1602 	WREG32(mmMME5_RTR_HBW_RD_RQ_N_ARB, 0x01040101);
1603 	WREG32(mmMME4_RTR_HBW_RD_RQ_N_ARB, 0x01030101);
1604 	WREG32(mmMME3_RTR_HBW_RD_RQ_N_ARB, 0x01020101);
1605 	WREG32(mmMME2_RTR_HBW_RD_RQ_N_ARB, 0x01010101);
1606 	WREG32(mmMME1_RTR_HBW_RD_RQ_N_ARB, 0x07010701);
1607 	WREG32(mmMME6_RTR_HBW_RD_RQ_S_ARB, 0x04010401);
1608 	WREG32(mmMME5_RTR_HBW_RD_RQ_S_ARB, 0x04050401);
1609 	WREG32(mmMME4_RTR_HBW_RD_RQ_S_ARB, 0x03070301);
1610 	WREG32(mmMME3_RTR_HBW_RD_RQ_S_ARB, 0x01030101);
1611 	WREG32(mmMME2_RTR_HBW_RD_RQ_S_ARB, 0x01040101);
1612 	WREG32(mmMME1_RTR_HBW_RD_RQ_S_ARB, 0x01050105);
1613 	WREG32(mmMME6_RTR_HBW_RD_RQ_W_ARB, 0x01010501);
1614 	WREG32(mmMME5_RTR_HBW_RD_RQ_W_ARB, 0x01010501);
1615 	WREG32(mmMME4_RTR_HBW_RD_RQ_W_ARB, 0x01040301);
1616 	WREG32(mmMME3_RTR_HBW_RD_RQ_W_ARB, 0x01030401);
1617 	WREG32(mmMME2_RTR_HBW_RD_RQ_W_ARB, 0x01040101);
1618 	WREG32(mmMME1_RTR_HBW_RD_RQ_W_ARB, 0x01050101);
1619 	WREG32(mmMME6_RTR_HBW_WR_RQ_N_ARB, 0x02020202);
1620 	WREG32(mmMME5_RTR_HBW_WR_RQ_N_ARB, 0x01070101);
1621 	WREG32(mmMME4_RTR_HBW_WR_RQ_N_ARB, 0x02020201);
1622 	WREG32(mmMME3_RTR_HBW_WR_RQ_N_ARB, 0x07020701);
1623 	WREG32(mmMME2_RTR_HBW_WR_RQ_N_ARB, 0x01020101);
1624 	WREG32(mmMME1_RTR_HBW_WR_RQ_S_ARB, 0x01010101);
1625 	WREG32(mmMME6_RTR_HBW_WR_RQ_S_ARB, 0x01070101);
1626 	WREG32(mmMME5_RTR_HBW_WR_RQ_S_ARB, 0x01070101);
1627 	WREG32(mmMME4_RTR_HBW_WR_RQ_S_ARB, 0x07020701);
1628 	WREG32(mmMME3_RTR_HBW_WR_RQ_S_ARB, 0x02020201);
1629 	WREG32(mmMME2_RTR_HBW_WR_RQ_S_ARB, 0x01070101);
1630 	WREG32(mmMME1_RTR_HBW_WR_RQ_S_ARB, 0x01020102);
1631 	WREG32(mmMME6_RTR_HBW_WR_RQ_W_ARB, 0x01020701);
1632 	WREG32(mmMME5_RTR_HBW_WR_RQ_W_ARB, 0x01020701);
1633 	WREG32(mmMME4_RTR_HBW_WR_RQ_W_ARB, 0x07020707);
1634 	WREG32(mmMME3_RTR_HBW_WR_RQ_W_ARB, 0x01020201);
1635 	WREG32(mmMME2_RTR_HBW_WR_RQ_W_ARB, 0x01070201);
1636 	WREG32(mmMME1_RTR_HBW_WR_RQ_W_ARB, 0x01070201);
1637 	WREG32(mmMME6_RTR_HBW_RD_RS_N_ARB, 0x01070102);
1638 	WREG32(mmMME5_RTR_HBW_RD_RS_N_ARB, 0x01070102);
1639 	WREG32(mmMME4_RTR_HBW_RD_RS_N_ARB, 0x01060102);
1640 	WREG32(mmMME3_RTR_HBW_RD_RS_N_ARB, 0x01040102);
1641 	WREG32(mmMME2_RTR_HBW_RD_RS_N_ARB, 0x01020102);
1642 	WREG32(mmMME1_RTR_HBW_RD_RS_N_ARB, 0x01020107);
1643 	WREG32(mmMME6_RTR_HBW_RD_RS_S_ARB, 0x01020106);
1644 	WREG32(mmMME5_RTR_HBW_RD_RS_S_ARB, 0x01020102);
1645 	WREG32(mmMME4_RTR_HBW_RD_RS_S_ARB, 0x01040102);
1646 	WREG32(mmMME3_RTR_HBW_RD_RS_S_ARB, 0x01060102);
1647 	WREG32(mmMME2_RTR_HBW_RD_RS_S_ARB, 0x01070102);
1648 	WREG32(mmMME1_RTR_HBW_RD_RS_S_ARB, 0x01070102);
1649 	WREG32(mmMME6_RTR_HBW_RD_RS_E_ARB, 0x01020702);
1650 	WREG32(mmMME5_RTR_HBW_RD_RS_E_ARB, 0x01020702);
1651 	WREG32(mmMME4_RTR_HBW_RD_RS_E_ARB, 0x01040602);
1652 	WREG32(mmMME3_RTR_HBW_RD_RS_E_ARB, 0x01060402);
1653 	WREG32(mmMME2_RTR_HBW_RD_RS_E_ARB, 0x01070202);
1654 	WREG32(mmMME1_RTR_HBW_RD_RS_E_ARB, 0x01070102);
1655 	WREG32(mmMME6_RTR_HBW_RD_RS_W_ARB, 0x01060401);
1656 	WREG32(mmMME5_RTR_HBW_RD_RS_W_ARB, 0x01060401);
1657 	WREG32(mmMME4_RTR_HBW_RD_RS_W_ARB, 0x01060401);
1658 	WREG32(mmMME3_RTR_HBW_RD_RS_W_ARB, 0x01060401);
1659 	WREG32(mmMME2_RTR_HBW_RD_RS_W_ARB, 0x01060401);
1660 	WREG32(mmMME1_RTR_HBW_RD_RS_W_ARB, 0x01060401);
1661 	WREG32(mmMME6_RTR_HBW_WR_RS_N_ARB, 0x01050101);
1662 	WREG32(mmMME5_RTR_HBW_WR_RS_N_ARB, 0x01040101);
1663 	WREG32(mmMME4_RTR_HBW_WR_RS_N_ARB, 0x01030101);
1664 	WREG32(mmMME3_RTR_HBW_WR_RS_N_ARB, 0x01020101);
1665 	WREG32(mmMME2_RTR_HBW_WR_RS_N_ARB, 0x01010101);
1666 	WREG32(mmMME1_RTR_HBW_WR_RS_N_ARB, 0x01010107);
1667 	WREG32(mmMME6_RTR_HBW_WR_RS_S_ARB, 0x01010107);
1668 	WREG32(mmMME5_RTR_HBW_WR_RS_S_ARB, 0x01010101);
1669 	WREG32(mmMME4_RTR_HBW_WR_RS_S_ARB, 0x01020101);
1670 	WREG32(mmMME3_RTR_HBW_WR_RS_S_ARB, 0x01030101);
1671 	WREG32(mmMME2_RTR_HBW_WR_RS_S_ARB, 0x01040101);
1672 	WREG32(mmMME1_RTR_HBW_WR_RS_S_ARB, 0x01050101);
1673 	WREG32(mmMME6_RTR_HBW_WR_RS_E_ARB, 0x01010501);
1674 	WREG32(mmMME5_RTR_HBW_WR_RS_E_ARB, 0x01010501);
1675 	WREG32(mmMME4_RTR_HBW_WR_RS_E_ARB, 0x01040301);
1676 	WREG32(mmMME3_RTR_HBW_WR_RS_E_ARB, 0x01030401);
1677 	WREG32(mmMME2_RTR_HBW_WR_RS_E_ARB, 0x01040101);
1678 	WREG32(mmMME1_RTR_HBW_WR_RS_E_ARB, 0x01050101);
1679 	WREG32(mmMME6_RTR_HBW_WR_RS_W_ARB, 0x01010101);
1680 	WREG32(mmMME5_RTR_HBW_WR_RS_W_ARB, 0x01010101);
1681 	WREG32(mmMME4_RTR_HBW_WR_RS_W_ARB, 0x01010101);
1682 	WREG32(mmMME3_RTR_HBW_WR_RS_W_ARB, 0x01010101);
1683 	WREG32(mmMME2_RTR_HBW_WR_RS_W_ARB, 0x01010101);
1684 	WREG32(mmMME1_RTR_HBW_WR_RS_W_ARB, 0x01010101);
1685 
1686 	WREG32(mmTPC1_RTR_HBW_RD_RQ_N_ARB, 0x01010101);
1687 	WREG32(mmTPC1_RTR_HBW_RD_RQ_S_ARB, 0x01010101);
1688 	WREG32(mmTPC1_RTR_HBW_RD_RQ_E_ARB, 0x01060101);
1689 	WREG32(mmTPC1_RTR_HBW_WR_RQ_N_ARB, 0x02020102);
1690 	WREG32(mmTPC1_RTR_HBW_WR_RQ_S_ARB, 0x01010101);
1691 	WREG32(mmTPC1_RTR_HBW_WR_RQ_E_ARB, 0x02070202);
1692 	WREG32(mmTPC1_RTR_HBW_RD_RS_N_ARB, 0x01020201);
1693 	WREG32(mmTPC1_RTR_HBW_RD_RS_S_ARB, 0x01070201);
1694 	WREG32(mmTPC1_RTR_HBW_RD_RS_W_ARB, 0x01070202);
1695 	WREG32(mmTPC1_RTR_HBW_WR_RS_N_ARB, 0x01010101);
1696 	WREG32(mmTPC1_RTR_HBW_WR_RS_S_ARB, 0x01050101);
1697 	WREG32(mmTPC1_RTR_HBW_WR_RS_W_ARB, 0x01050101);
1698 
1699 	WREG32(mmTPC2_RTR_HBW_RD_RQ_N_ARB, 0x01020101);
1700 	WREG32(mmTPC2_RTR_HBW_RD_RQ_S_ARB, 0x01050101);
1701 	WREG32(mmTPC2_RTR_HBW_RD_RQ_E_ARB, 0x01010201);
1702 	WREG32(mmTPC2_RTR_HBW_WR_RQ_N_ARB, 0x02040102);
1703 	WREG32(mmTPC2_RTR_HBW_WR_RQ_S_ARB, 0x01050101);
1704 	WREG32(mmTPC2_RTR_HBW_WR_RQ_E_ARB, 0x02060202);
1705 	WREG32(mmTPC2_RTR_HBW_RD_RS_N_ARB, 0x01020201);
1706 	WREG32(mmTPC2_RTR_HBW_RD_RS_S_ARB, 0x01070201);
1707 	WREG32(mmTPC2_RTR_HBW_RD_RS_W_ARB, 0x01070202);
1708 	WREG32(mmTPC2_RTR_HBW_WR_RS_N_ARB, 0x01010101);
1709 	WREG32(mmTPC2_RTR_HBW_WR_RS_S_ARB, 0x01040101);
1710 	WREG32(mmTPC2_RTR_HBW_WR_RS_W_ARB, 0x01040101);
1711 
1712 	WREG32(mmTPC3_RTR_HBW_RD_RQ_N_ARB, 0x01030101);
1713 	WREG32(mmTPC3_RTR_HBW_RD_RQ_S_ARB, 0x01040101);
1714 	WREG32(mmTPC3_RTR_HBW_RD_RQ_E_ARB, 0x01040301);
1715 	WREG32(mmTPC3_RTR_HBW_WR_RQ_N_ARB, 0x02060102);
1716 	WREG32(mmTPC3_RTR_HBW_WR_RQ_S_ARB, 0x01040101);
1717 	WREG32(mmTPC3_RTR_HBW_WR_RQ_E_ARB, 0x01040301);
1718 	WREG32(mmTPC3_RTR_HBW_RD_RS_N_ARB, 0x01040201);
1719 	WREG32(mmTPC3_RTR_HBW_RD_RS_S_ARB, 0x01060201);
1720 	WREG32(mmTPC3_RTR_HBW_RD_RS_W_ARB, 0x01060402);
1721 	WREG32(mmTPC3_RTR_HBW_WR_RS_N_ARB, 0x01020101);
1722 	WREG32(mmTPC3_RTR_HBW_WR_RS_S_ARB, 0x01030101);
1723 	WREG32(mmTPC3_RTR_HBW_WR_RS_W_ARB, 0x01030401);
1724 
1725 	WREG32(mmTPC4_RTR_HBW_RD_RQ_N_ARB, 0x01040101);
1726 	WREG32(mmTPC4_RTR_HBW_RD_RQ_S_ARB, 0x01030101);
1727 	WREG32(mmTPC4_RTR_HBW_RD_RQ_E_ARB, 0x01030401);
1728 	WREG32(mmTPC4_RTR_HBW_WR_RQ_N_ARB, 0x02070102);
1729 	WREG32(mmTPC4_RTR_HBW_WR_RQ_S_ARB, 0x01030101);
1730 	WREG32(mmTPC4_RTR_HBW_WR_RQ_E_ARB, 0x02060702);
1731 	WREG32(mmTPC4_RTR_HBW_RD_RS_N_ARB, 0x01060201);
1732 	WREG32(mmTPC4_RTR_HBW_RD_RS_S_ARB, 0x01040201);
1733 	WREG32(mmTPC4_RTR_HBW_RD_RS_W_ARB, 0x01040602);
1734 	WREG32(mmTPC4_RTR_HBW_WR_RS_N_ARB, 0x01030101);
1735 	WREG32(mmTPC4_RTR_HBW_WR_RS_S_ARB, 0x01020101);
1736 	WREG32(mmTPC4_RTR_HBW_WR_RS_W_ARB, 0x01040301);
1737 
1738 	WREG32(mmTPC5_RTR_HBW_RD_RQ_N_ARB, 0x01050101);
1739 	WREG32(mmTPC5_RTR_HBW_RD_RQ_S_ARB, 0x01020101);
1740 	WREG32(mmTPC5_RTR_HBW_RD_RQ_E_ARB, 0x01200501);
1741 	WREG32(mmTPC5_RTR_HBW_WR_RQ_N_ARB, 0x02070102);
1742 	WREG32(mmTPC5_RTR_HBW_WR_RQ_S_ARB, 0x01020101);
1743 	WREG32(mmTPC5_RTR_HBW_WR_RQ_E_ARB, 0x02020602);
1744 	WREG32(mmTPC5_RTR_HBW_RD_RS_N_ARB, 0x01070201);
1745 	WREG32(mmTPC5_RTR_HBW_RD_RS_S_ARB, 0x01020201);
1746 	WREG32(mmTPC5_RTR_HBW_RD_RS_W_ARB, 0x01020702);
1747 	WREG32(mmTPC5_RTR_HBW_WR_RS_N_ARB, 0x01040101);
1748 	WREG32(mmTPC5_RTR_HBW_WR_RS_S_ARB, 0x01010101);
1749 	WREG32(mmTPC5_RTR_HBW_WR_RS_W_ARB, 0x01010501);
1750 
1751 	WREG32(mmTPC6_RTR_HBW_RD_RQ_N_ARB, 0x01010101);
1752 	WREG32(mmTPC6_RTR_HBW_RD_RQ_S_ARB, 0x01010101);
1753 	WREG32(mmTPC6_RTR_HBW_RD_RQ_E_ARB, 0x01010601);
1754 	WREG32(mmTPC6_RTR_HBW_WR_RQ_N_ARB, 0x01010101);
1755 	WREG32(mmTPC6_RTR_HBW_WR_RQ_S_ARB, 0x01010101);
1756 	WREG32(mmTPC6_RTR_HBW_WR_RQ_E_ARB, 0x02020702);
1757 	WREG32(mmTPC6_RTR_HBW_RD_RS_N_ARB, 0x01010101);
1758 	WREG32(mmTPC6_RTR_HBW_RD_RS_S_ARB, 0x01010101);
1759 	WREG32(mmTPC6_RTR_HBW_RD_RS_W_ARB, 0x01020702);
1760 	WREG32(mmTPC6_RTR_HBW_WR_RS_N_ARB, 0x01050101);
1761 	WREG32(mmTPC6_RTR_HBW_WR_RS_S_ARB, 0x01010101);
1762 	WREG32(mmTPC6_RTR_HBW_WR_RS_W_ARB, 0x01010501);
1763 
1764 	for (i = 0, offset = 0 ; i < 10 ; i++, offset += 4) {
1765 		WREG32(mmMME1_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1766 		WREG32(mmMME2_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1767 		WREG32(mmMME3_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1768 		WREG32(mmMME4_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1769 		WREG32(mmMME5_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1770 		WREG32(mmMME6_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1771 
1772 		WREG32(mmTPC0_NRTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1773 		WREG32(mmTPC1_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1774 		WREG32(mmTPC2_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1775 		WREG32(mmTPC3_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1776 		WREG32(mmTPC4_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1777 		WREG32(mmTPC5_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1778 		WREG32(mmTPC6_RTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1779 		WREG32(mmTPC7_NRTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1780 
1781 		WREG32(mmPCI_NRTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1782 		WREG32(mmDMA_NRTR_SPLIT_COEF_0 + offset, polynom[i] >> 7);
1783 	}
1784 
1785 	for (i = 0, offset = 0 ; i < 6 ; i++, offset += 0x40000) {
1786 		WREG32(mmMME1_RTR_SCRAMB_EN + offset,
1787 				1 << MME1_RTR_SCRAMB_EN_VAL_SHIFT);
1788 		WREG32(mmMME1_RTR_NON_LIN_SCRAMB + offset,
1789 				1 << MME1_RTR_NON_LIN_SCRAMB_EN_SHIFT);
1790 	}
1791 
1792 	for (i = 0, offset = 0 ; i < 8 ; i++, offset += 0x40000) {
1793 		/*
1794 		 * Workaround for Bug H2 #2441 :
1795 		 * "ST.NOP set trace event illegal opcode"
1796 		 */
1797 		WREG32(mmTPC0_CFG_TPC_INTR_MASK + offset, tpc_intr_mask);
1798 
1799 		WREG32(mmTPC0_NRTR_SCRAMB_EN + offset,
1800 				1 << TPC0_NRTR_SCRAMB_EN_VAL_SHIFT);
1801 		WREG32(mmTPC0_NRTR_NON_LIN_SCRAMB + offset,
1802 				1 << TPC0_NRTR_NON_LIN_SCRAMB_EN_SHIFT);
1803 
1804 		WREG32_FIELD(TPC0_CFG_MSS_CONFIG, offset,
1805 				ICACHE_FETCH_LINE_NUM, 2);
1806 	}
1807 
1808 	WREG32(mmDMA_NRTR_SCRAMB_EN, 1 << DMA_NRTR_SCRAMB_EN_VAL_SHIFT);
1809 	WREG32(mmDMA_NRTR_NON_LIN_SCRAMB,
1810 			1 << DMA_NRTR_NON_LIN_SCRAMB_EN_SHIFT);
1811 
1812 	WREG32(mmPCI_NRTR_SCRAMB_EN, 1 << PCI_NRTR_SCRAMB_EN_VAL_SHIFT);
1813 	WREG32(mmPCI_NRTR_NON_LIN_SCRAMB,
1814 			1 << PCI_NRTR_NON_LIN_SCRAMB_EN_SHIFT);
1815 
1816 	/*
1817 	 * Workaround for H2 #HW-23 bug
1818 	 * Set DMA max outstanding read requests to 240 on DMA CH 1.
1819 	 * This limitation is still large enough to not affect Gen4 bandwidth.
1820 	 * We need to only limit that DMA channel because the user can only read
1821 	 * from Host using DMA CH 1
1822 	 */
1823 	WREG32(mmDMA_CH_1_CFG0, 0x0fff00F0);
1824 
1825 	WREG32(mmTPC_PLL_CLK_RLX_0, 0x200020);
1826 
1827 	goya->hw_cap_initialized |= HW_CAP_GOLDEN;
1828 }
1829 
1830 static void goya_init_mme_qman(struct hl_device *hdev)
1831 {
1832 	u32 mtr_base_lo, mtr_base_hi;
1833 	u32 so_base_lo, so_base_hi;
1834 	u32 gic_base_lo, gic_base_hi;
1835 	u64 qman_base_addr;
1836 
1837 	mtr_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_MON_PAY_ADDRL_0);
1838 	mtr_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_MON_PAY_ADDRL_0);
1839 	so_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1840 	so_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1841 
1842 	gic_base_lo =
1843 		lower_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
1844 	gic_base_hi =
1845 		upper_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
1846 
1847 	qman_base_addr = hdev->asic_prop.sram_base_address +
1848 				MME_QMAN_BASE_OFFSET;
1849 
1850 	WREG32(mmMME_QM_PQ_BASE_LO, lower_32_bits(qman_base_addr));
1851 	WREG32(mmMME_QM_PQ_BASE_HI, upper_32_bits(qman_base_addr));
1852 	WREG32(mmMME_QM_PQ_SIZE, ilog2(MME_QMAN_LENGTH));
1853 	WREG32(mmMME_QM_PQ_PI, 0);
1854 	WREG32(mmMME_QM_PQ_CI, 0);
1855 	WREG32(mmMME_QM_CP_LDMA_SRC_BASE_LO_OFFSET, 0x10C0);
1856 	WREG32(mmMME_QM_CP_LDMA_SRC_BASE_HI_OFFSET, 0x10C4);
1857 	WREG32(mmMME_QM_CP_LDMA_TSIZE_OFFSET, 0x10C8);
1858 	WREG32(mmMME_QM_CP_LDMA_COMMIT_OFFSET, 0x10CC);
1859 
1860 	WREG32(mmMME_QM_CP_MSG_BASE0_ADDR_LO, mtr_base_lo);
1861 	WREG32(mmMME_QM_CP_MSG_BASE0_ADDR_HI, mtr_base_hi);
1862 	WREG32(mmMME_QM_CP_MSG_BASE1_ADDR_LO, so_base_lo);
1863 	WREG32(mmMME_QM_CP_MSG_BASE1_ADDR_HI, so_base_hi);
1864 
1865 	/* QMAN CQ has 8 cache lines */
1866 	WREG32(mmMME_QM_CQ_CFG1, 0x00080008);
1867 
1868 	WREG32(mmMME_QM_GLBL_ERR_ADDR_LO, gic_base_lo);
1869 	WREG32(mmMME_QM_GLBL_ERR_ADDR_HI, gic_base_hi);
1870 
1871 	WREG32(mmMME_QM_GLBL_ERR_WDATA, GOYA_ASYNC_EVENT_ID_MME_QM);
1872 
1873 	WREG32(mmMME_QM_GLBL_ERR_CFG, QMAN_MME_ERR_MSG_EN);
1874 
1875 	WREG32(mmMME_QM_GLBL_PROT, QMAN_MME_ERR_PROT);
1876 
1877 	WREG32(mmMME_QM_GLBL_CFG0, QMAN_MME_ENABLE);
1878 }
1879 
1880 static void goya_init_mme_cmdq(struct hl_device *hdev)
1881 {
1882 	u32 mtr_base_lo, mtr_base_hi;
1883 	u32 so_base_lo, so_base_hi;
1884 	u32 gic_base_lo, gic_base_hi;
1885 
1886 	mtr_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_MON_PAY_ADDRL_0);
1887 	mtr_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_MON_PAY_ADDRL_0);
1888 	so_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1889 	so_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1890 
1891 	gic_base_lo =
1892 		lower_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
1893 	gic_base_hi =
1894 		upper_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
1895 
1896 	WREG32(mmMME_CMDQ_CP_MSG_BASE0_ADDR_LO, mtr_base_lo);
1897 	WREG32(mmMME_CMDQ_CP_MSG_BASE0_ADDR_HI, mtr_base_hi);
1898 	WREG32(mmMME_CMDQ_CP_MSG_BASE1_ADDR_LO,	so_base_lo);
1899 	WREG32(mmMME_CMDQ_CP_MSG_BASE1_ADDR_HI, so_base_hi);
1900 
1901 	/* CMDQ CQ has 20 cache lines */
1902 	WREG32(mmMME_CMDQ_CQ_CFG1, 0x00140014);
1903 
1904 	WREG32(mmMME_CMDQ_GLBL_ERR_ADDR_LO, gic_base_lo);
1905 	WREG32(mmMME_CMDQ_GLBL_ERR_ADDR_HI, gic_base_hi);
1906 
1907 	WREG32(mmMME_CMDQ_GLBL_ERR_WDATA, GOYA_ASYNC_EVENT_ID_MME_CMDQ);
1908 
1909 	WREG32(mmMME_CMDQ_GLBL_ERR_CFG, CMDQ_MME_ERR_MSG_EN);
1910 
1911 	WREG32(mmMME_CMDQ_GLBL_PROT, CMDQ_MME_ERR_PROT);
1912 
1913 	WREG32(mmMME_CMDQ_GLBL_CFG0, CMDQ_MME_ENABLE);
1914 }
1915 
1916 void goya_init_mme_qmans(struct hl_device *hdev)
1917 {
1918 	struct goya_device *goya = hdev->asic_specific;
1919 	u32 so_base_lo, so_base_hi;
1920 
1921 	if (goya->hw_cap_initialized & HW_CAP_MME)
1922 		return;
1923 
1924 	so_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1925 	so_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1926 
1927 	WREG32(mmMME_SM_BASE_ADDRESS_LOW, so_base_lo);
1928 	WREG32(mmMME_SM_BASE_ADDRESS_HIGH, so_base_hi);
1929 
1930 	goya_init_mme_qman(hdev);
1931 	goya_init_mme_cmdq(hdev);
1932 
1933 	goya->hw_cap_initialized |= HW_CAP_MME;
1934 }
1935 
1936 static void goya_init_tpc_qman(struct hl_device *hdev, u32 base_off, int tpc_id)
1937 {
1938 	u32 mtr_base_lo, mtr_base_hi;
1939 	u32 so_base_lo, so_base_hi;
1940 	u32 gic_base_lo, gic_base_hi;
1941 	u64 qman_base_addr;
1942 	u32 reg_off = tpc_id * (mmTPC1_QM_PQ_PI - mmTPC0_QM_PQ_PI);
1943 
1944 	mtr_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_MON_PAY_ADDRL_0);
1945 	mtr_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_MON_PAY_ADDRL_0);
1946 	so_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1947 	so_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1948 
1949 	gic_base_lo =
1950 		lower_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
1951 	gic_base_hi =
1952 		upper_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
1953 
1954 	qman_base_addr = hdev->asic_prop.sram_base_address + base_off;
1955 
1956 	WREG32(mmTPC0_QM_PQ_BASE_LO + reg_off, lower_32_bits(qman_base_addr));
1957 	WREG32(mmTPC0_QM_PQ_BASE_HI + reg_off, upper_32_bits(qman_base_addr));
1958 	WREG32(mmTPC0_QM_PQ_SIZE + reg_off, ilog2(TPC_QMAN_LENGTH));
1959 	WREG32(mmTPC0_QM_PQ_PI + reg_off, 0);
1960 	WREG32(mmTPC0_QM_PQ_CI + reg_off, 0);
1961 	WREG32(mmTPC0_QM_CP_LDMA_SRC_BASE_LO_OFFSET + reg_off, 0x10C0);
1962 	WREG32(mmTPC0_QM_CP_LDMA_SRC_BASE_HI_OFFSET + reg_off, 0x10C4);
1963 	WREG32(mmTPC0_QM_CP_LDMA_TSIZE_OFFSET + reg_off, 0x10C8);
1964 	WREG32(mmTPC0_QM_CP_LDMA_COMMIT_OFFSET + reg_off, 0x10CC);
1965 
1966 	WREG32(mmTPC0_QM_CP_MSG_BASE0_ADDR_LO + reg_off, mtr_base_lo);
1967 	WREG32(mmTPC0_QM_CP_MSG_BASE0_ADDR_HI + reg_off, mtr_base_hi);
1968 	WREG32(mmTPC0_QM_CP_MSG_BASE1_ADDR_LO + reg_off, so_base_lo);
1969 	WREG32(mmTPC0_QM_CP_MSG_BASE1_ADDR_HI + reg_off, so_base_hi);
1970 
1971 	WREG32(mmTPC0_QM_CQ_CFG1 + reg_off, 0x00080008);
1972 
1973 	WREG32(mmTPC0_QM_GLBL_ERR_ADDR_LO + reg_off, gic_base_lo);
1974 	WREG32(mmTPC0_QM_GLBL_ERR_ADDR_HI + reg_off, gic_base_hi);
1975 
1976 	WREG32(mmTPC0_QM_GLBL_ERR_WDATA + reg_off,
1977 			GOYA_ASYNC_EVENT_ID_TPC0_QM + tpc_id);
1978 
1979 	WREG32(mmTPC0_QM_GLBL_ERR_CFG + reg_off, QMAN_TPC_ERR_MSG_EN);
1980 
1981 	WREG32(mmTPC0_QM_GLBL_PROT + reg_off, QMAN_TPC_ERR_PROT);
1982 
1983 	WREG32(mmTPC0_QM_GLBL_CFG0 + reg_off, QMAN_TPC_ENABLE);
1984 }
1985 
1986 static void goya_init_tpc_cmdq(struct hl_device *hdev, int tpc_id)
1987 {
1988 	u32 mtr_base_lo, mtr_base_hi;
1989 	u32 so_base_lo, so_base_hi;
1990 	u32 gic_base_lo, gic_base_hi;
1991 	u32 reg_off = tpc_id * (mmTPC1_CMDQ_CQ_CFG1 - mmTPC0_CMDQ_CQ_CFG1);
1992 
1993 	mtr_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_MON_PAY_ADDRL_0);
1994 	mtr_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_MON_PAY_ADDRL_0);
1995 	so_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1996 	so_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
1997 
1998 	gic_base_lo =
1999 		lower_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
2000 	gic_base_hi =
2001 		upper_32_bits(CFG_BASE + mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR);
2002 
2003 	WREG32(mmTPC0_CMDQ_CP_MSG_BASE0_ADDR_LO + reg_off, mtr_base_lo);
2004 	WREG32(mmTPC0_CMDQ_CP_MSG_BASE0_ADDR_HI + reg_off, mtr_base_hi);
2005 	WREG32(mmTPC0_CMDQ_CP_MSG_BASE1_ADDR_LO + reg_off, so_base_lo);
2006 	WREG32(mmTPC0_CMDQ_CP_MSG_BASE1_ADDR_HI + reg_off, so_base_hi);
2007 
2008 	WREG32(mmTPC0_CMDQ_CQ_CFG1 + reg_off, 0x00140014);
2009 
2010 	WREG32(mmTPC0_CMDQ_GLBL_ERR_ADDR_LO + reg_off, gic_base_lo);
2011 	WREG32(mmTPC0_CMDQ_GLBL_ERR_ADDR_HI + reg_off, gic_base_hi);
2012 
2013 	WREG32(mmTPC0_CMDQ_GLBL_ERR_WDATA + reg_off,
2014 			GOYA_ASYNC_EVENT_ID_TPC0_CMDQ + tpc_id);
2015 
2016 	WREG32(mmTPC0_CMDQ_GLBL_ERR_CFG + reg_off, CMDQ_TPC_ERR_MSG_EN);
2017 
2018 	WREG32(mmTPC0_CMDQ_GLBL_PROT + reg_off, CMDQ_TPC_ERR_PROT);
2019 
2020 	WREG32(mmTPC0_CMDQ_GLBL_CFG0 + reg_off, CMDQ_TPC_ENABLE);
2021 }
2022 
2023 void goya_init_tpc_qmans(struct hl_device *hdev)
2024 {
2025 	struct goya_device *goya = hdev->asic_specific;
2026 	u32 so_base_lo, so_base_hi;
2027 	u32 cfg_off = mmTPC1_CFG_SM_BASE_ADDRESS_LOW -
2028 			mmTPC0_CFG_SM_BASE_ADDRESS_LOW;
2029 	int i;
2030 
2031 	if (goya->hw_cap_initialized & HW_CAP_TPC)
2032 		return;
2033 
2034 	so_base_lo = lower_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
2035 	so_base_hi = upper_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
2036 
2037 	for (i = 0 ; i < TPC_MAX_NUM ; i++) {
2038 		WREG32(mmTPC0_CFG_SM_BASE_ADDRESS_LOW + i * cfg_off,
2039 				so_base_lo);
2040 		WREG32(mmTPC0_CFG_SM_BASE_ADDRESS_HIGH + i * cfg_off,
2041 				so_base_hi);
2042 	}
2043 
2044 	goya_init_tpc_qman(hdev, TPC0_QMAN_BASE_OFFSET, 0);
2045 	goya_init_tpc_qman(hdev, TPC1_QMAN_BASE_OFFSET, 1);
2046 	goya_init_tpc_qman(hdev, TPC2_QMAN_BASE_OFFSET, 2);
2047 	goya_init_tpc_qman(hdev, TPC3_QMAN_BASE_OFFSET, 3);
2048 	goya_init_tpc_qman(hdev, TPC4_QMAN_BASE_OFFSET, 4);
2049 	goya_init_tpc_qman(hdev, TPC5_QMAN_BASE_OFFSET, 5);
2050 	goya_init_tpc_qman(hdev, TPC6_QMAN_BASE_OFFSET, 6);
2051 	goya_init_tpc_qman(hdev, TPC7_QMAN_BASE_OFFSET, 7);
2052 
2053 	for (i = 0 ; i < TPC_MAX_NUM ; i++)
2054 		goya_init_tpc_cmdq(hdev, i);
2055 
2056 	goya->hw_cap_initialized |= HW_CAP_TPC;
2057 }
2058 
2059 /*
2060  * goya_disable_internal_queues - Disable internal queues
2061  *
2062  * @hdev: pointer to hl_device structure
2063  *
2064  */
2065 static void goya_disable_internal_queues(struct hl_device *hdev)
2066 {
2067 	struct goya_device *goya = hdev->asic_specific;
2068 
2069 	if (!(goya->hw_cap_initialized & HW_CAP_MME))
2070 		goto disable_tpc;
2071 
2072 	WREG32(mmMME_QM_GLBL_CFG0, 0);
2073 	WREG32(mmMME_CMDQ_GLBL_CFG0, 0);
2074 
2075 disable_tpc:
2076 	if (!(goya->hw_cap_initialized & HW_CAP_TPC))
2077 		return;
2078 
2079 	WREG32(mmTPC0_QM_GLBL_CFG0, 0);
2080 	WREG32(mmTPC0_CMDQ_GLBL_CFG0, 0);
2081 
2082 	WREG32(mmTPC1_QM_GLBL_CFG0, 0);
2083 	WREG32(mmTPC1_CMDQ_GLBL_CFG0, 0);
2084 
2085 	WREG32(mmTPC2_QM_GLBL_CFG0, 0);
2086 	WREG32(mmTPC2_CMDQ_GLBL_CFG0, 0);
2087 
2088 	WREG32(mmTPC3_QM_GLBL_CFG0, 0);
2089 	WREG32(mmTPC3_CMDQ_GLBL_CFG0, 0);
2090 
2091 	WREG32(mmTPC4_QM_GLBL_CFG0, 0);
2092 	WREG32(mmTPC4_CMDQ_GLBL_CFG0, 0);
2093 
2094 	WREG32(mmTPC5_QM_GLBL_CFG0, 0);
2095 	WREG32(mmTPC5_CMDQ_GLBL_CFG0, 0);
2096 
2097 	WREG32(mmTPC6_QM_GLBL_CFG0, 0);
2098 	WREG32(mmTPC6_CMDQ_GLBL_CFG0, 0);
2099 
2100 	WREG32(mmTPC7_QM_GLBL_CFG0, 0);
2101 	WREG32(mmTPC7_CMDQ_GLBL_CFG0, 0);
2102 }
2103 
2104 /*
2105  * goya_stop_internal_queues - Stop internal queues
2106  *
2107  * @hdev: pointer to hl_device structure
2108  *
2109  * Returns 0 on success
2110  *
2111  */
2112 static int goya_stop_internal_queues(struct hl_device *hdev)
2113 {
2114 	struct goya_device *goya = hdev->asic_specific;
2115 	int rc, retval = 0;
2116 
2117 	if (!(goya->hw_cap_initialized & HW_CAP_MME))
2118 		goto stop_tpc;
2119 
2120 	/*
2121 	 * Each queue (QMAN) is a separate H/W logic. That means that each
2122 	 * QMAN can be stopped independently and failure to stop one does NOT
2123 	 * mandate we should not try to stop other QMANs
2124 	 */
2125 
2126 	rc = goya_stop_queue(hdev,
2127 			mmMME_QM_GLBL_CFG1,
2128 			mmMME_QM_CP_STS,
2129 			mmMME_QM_GLBL_STS0);
2130 
2131 	if (rc) {
2132 		dev_err(hdev->dev, "failed to stop MME QMAN\n");
2133 		retval = -EIO;
2134 	}
2135 
2136 	rc = goya_stop_queue(hdev,
2137 			mmMME_CMDQ_GLBL_CFG1,
2138 			mmMME_CMDQ_CP_STS,
2139 			mmMME_CMDQ_GLBL_STS0);
2140 
2141 	if (rc) {
2142 		dev_err(hdev->dev, "failed to stop MME CMDQ\n");
2143 		retval = -EIO;
2144 	}
2145 
2146 stop_tpc:
2147 	if (!(goya->hw_cap_initialized & HW_CAP_TPC))
2148 		return retval;
2149 
2150 	rc = goya_stop_queue(hdev,
2151 			mmTPC0_QM_GLBL_CFG1,
2152 			mmTPC0_QM_CP_STS,
2153 			mmTPC0_QM_GLBL_STS0);
2154 
2155 	if (rc) {
2156 		dev_err(hdev->dev, "failed to stop TPC 0 QMAN\n");
2157 		retval = -EIO;
2158 	}
2159 
2160 	rc = goya_stop_queue(hdev,
2161 			mmTPC0_CMDQ_GLBL_CFG1,
2162 			mmTPC0_CMDQ_CP_STS,
2163 			mmTPC0_CMDQ_GLBL_STS0);
2164 
2165 	if (rc) {
2166 		dev_err(hdev->dev, "failed to stop TPC 0 CMDQ\n");
2167 		retval = -EIO;
2168 	}
2169 
2170 	rc = goya_stop_queue(hdev,
2171 			mmTPC1_QM_GLBL_CFG1,
2172 			mmTPC1_QM_CP_STS,
2173 			mmTPC1_QM_GLBL_STS0);
2174 
2175 	if (rc) {
2176 		dev_err(hdev->dev, "failed to stop TPC 1 QMAN\n");
2177 		retval = -EIO;
2178 	}
2179 
2180 	rc = goya_stop_queue(hdev,
2181 			mmTPC1_CMDQ_GLBL_CFG1,
2182 			mmTPC1_CMDQ_CP_STS,
2183 			mmTPC1_CMDQ_GLBL_STS0);
2184 
2185 	if (rc) {
2186 		dev_err(hdev->dev, "failed to stop TPC 1 CMDQ\n");
2187 		retval = -EIO;
2188 	}
2189 
2190 	rc = goya_stop_queue(hdev,
2191 			mmTPC2_QM_GLBL_CFG1,
2192 			mmTPC2_QM_CP_STS,
2193 			mmTPC2_QM_GLBL_STS0);
2194 
2195 	if (rc) {
2196 		dev_err(hdev->dev, "failed to stop TPC 2 QMAN\n");
2197 		retval = -EIO;
2198 	}
2199 
2200 	rc = goya_stop_queue(hdev,
2201 			mmTPC2_CMDQ_GLBL_CFG1,
2202 			mmTPC2_CMDQ_CP_STS,
2203 			mmTPC2_CMDQ_GLBL_STS0);
2204 
2205 	if (rc) {
2206 		dev_err(hdev->dev, "failed to stop TPC 2 CMDQ\n");
2207 		retval = -EIO;
2208 	}
2209 
2210 	rc = goya_stop_queue(hdev,
2211 			mmTPC3_QM_GLBL_CFG1,
2212 			mmTPC3_QM_CP_STS,
2213 			mmTPC3_QM_GLBL_STS0);
2214 
2215 	if (rc) {
2216 		dev_err(hdev->dev, "failed to stop TPC 3 QMAN\n");
2217 		retval = -EIO;
2218 	}
2219 
2220 	rc = goya_stop_queue(hdev,
2221 			mmTPC3_CMDQ_GLBL_CFG1,
2222 			mmTPC3_CMDQ_CP_STS,
2223 			mmTPC3_CMDQ_GLBL_STS0);
2224 
2225 	if (rc) {
2226 		dev_err(hdev->dev, "failed to stop TPC 3 CMDQ\n");
2227 		retval = -EIO;
2228 	}
2229 
2230 	rc = goya_stop_queue(hdev,
2231 			mmTPC4_QM_GLBL_CFG1,
2232 			mmTPC4_QM_CP_STS,
2233 			mmTPC4_QM_GLBL_STS0);
2234 
2235 	if (rc) {
2236 		dev_err(hdev->dev, "failed to stop TPC 4 QMAN\n");
2237 		retval = -EIO;
2238 	}
2239 
2240 	rc = goya_stop_queue(hdev,
2241 			mmTPC4_CMDQ_GLBL_CFG1,
2242 			mmTPC4_CMDQ_CP_STS,
2243 			mmTPC4_CMDQ_GLBL_STS0);
2244 
2245 	if (rc) {
2246 		dev_err(hdev->dev, "failed to stop TPC 4 CMDQ\n");
2247 		retval = -EIO;
2248 	}
2249 
2250 	rc = goya_stop_queue(hdev,
2251 			mmTPC5_QM_GLBL_CFG1,
2252 			mmTPC5_QM_CP_STS,
2253 			mmTPC5_QM_GLBL_STS0);
2254 
2255 	if (rc) {
2256 		dev_err(hdev->dev, "failed to stop TPC 5 QMAN\n");
2257 		retval = -EIO;
2258 	}
2259 
2260 	rc = goya_stop_queue(hdev,
2261 			mmTPC5_CMDQ_GLBL_CFG1,
2262 			mmTPC5_CMDQ_CP_STS,
2263 			mmTPC5_CMDQ_GLBL_STS0);
2264 
2265 	if (rc) {
2266 		dev_err(hdev->dev, "failed to stop TPC 5 CMDQ\n");
2267 		retval = -EIO;
2268 	}
2269 
2270 	rc = goya_stop_queue(hdev,
2271 			mmTPC6_QM_GLBL_CFG1,
2272 			mmTPC6_QM_CP_STS,
2273 			mmTPC6_QM_GLBL_STS0);
2274 
2275 	if (rc) {
2276 		dev_err(hdev->dev, "failed to stop TPC 6 QMAN\n");
2277 		retval = -EIO;
2278 	}
2279 
2280 	rc = goya_stop_queue(hdev,
2281 			mmTPC6_CMDQ_GLBL_CFG1,
2282 			mmTPC6_CMDQ_CP_STS,
2283 			mmTPC6_CMDQ_GLBL_STS0);
2284 
2285 	if (rc) {
2286 		dev_err(hdev->dev, "failed to stop TPC 6 CMDQ\n");
2287 		retval = -EIO;
2288 	}
2289 
2290 	rc = goya_stop_queue(hdev,
2291 			mmTPC7_QM_GLBL_CFG1,
2292 			mmTPC7_QM_CP_STS,
2293 			mmTPC7_QM_GLBL_STS0);
2294 
2295 	if (rc) {
2296 		dev_err(hdev->dev, "failed to stop TPC 7 QMAN\n");
2297 		retval = -EIO;
2298 	}
2299 
2300 	rc = goya_stop_queue(hdev,
2301 			mmTPC7_CMDQ_GLBL_CFG1,
2302 			mmTPC7_CMDQ_CP_STS,
2303 			mmTPC7_CMDQ_GLBL_STS0);
2304 
2305 	if (rc) {
2306 		dev_err(hdev->dev, "failed to stop TPC 7 CMDQ\n");
2307 		retval = -EIO;
2308 	}
2309 
2310 	return retval;
2311 }
2312 
2313 static void goya_dma_stall(struct hl_device *hdev)
2314 {
2315 	struct goya_device *goya = hdev->asic_specific;
2316 
2317 	if (!(goya->hw_cap_initialized & HW_CAP_DMA))
2318 		return;
2319 
2320 	WREG32(mmDMA_QM_0_GLBL_CFG1, 1 << DMA_QM_0_GLBL_CFG1_DMA_STOP_SHIFT);
2321 	WREG32(mmDMA_QM_1_GLBL_CFG1, 1 << DMA_QM_1_GLBL_CFG1_DMA_STOP_SHIFT);
2322 	WREG32(mmDMA_QM_2_GLBL_CFG1, 1 << DMA_QM_2_GLBL_CFG1_DMA_STOP_SHIFT);
2323 	WREG32(mmDMA_QM_3_GLBL_CFG1, 1 << DMA_QM_3_GLBL_CFG1_DMA_STOP_SHIFT);
2324 	WREG32(mmDMA_QM_4_GLBL_CFG1, 1 << DMA_QM_4_GLBL_CFG1_DMA_STOP_SHIFT);
2325 }
2326 
2327 static void goya_tpc_stall(struct hl_device *hdev)
2328 {
2329 	struct goya_device *goya = hdev->asic_specific;
2330 
2331 	if (!(goya->hw_cap_initialized & HW_CAP_TPC))
2332 		return;
2333 
2334 	WREG32(mmTPC0_CFG_TPC_STALL, 1 << TPC0_CFG_TPC_STALL_V_SHIFT);
2335 	WREG32(mmTPC1_CFG_TPC_STALL, 1 << TPC1_CFG_TPC_STALL_V_SHIFT);
2336 	WREG32(mmTPC2_CFG_TPC_STALL, 1 << TPC2_CFG_TPC_STALL_V_SHIFT);
2337 	WREG32(mmTPC3_CFG_TPC_STALL, 1 << TPC3_CFG_TPC_STALL_V_SHIFT);
2338 	WREG32(mmTPC4_CFG_TPC_STALL, 1 << TPC4_CFG_TPC_STALL_V_SHIFT);
2339 	WREG32(mmTPC5_CFG_TPC_STALL, 1 << TPC5_CFG_TPC_STALL_V_SHIFT);
2340 	WREG32(mmTPC6_CFG_TPC_STALL, 1 << TPC6_CFG_TPC_STALL_V_SHIFT);
2341 	WREG32(mmTPC7_CFG_TPC_STALL, 1 << TPC7_CFG_TPC_STALL_V_SHIFT);
2342 }
2343 
2344 static void goya_mme_stall(struct hl_device *hdev)
2345 {
2346 	struct goya_device *goya = hdev->asic_specific;
2347 
2348 	if (!(goya->hw_cap_initialized & HW_CAP_MME))
2349 		return;
2350 
2351 	WREG32(mmMME_STALL, 0xFFFFFFFF);
2352 }
2353 
2354 static int goya_enable_msix(struct hl_device *hdev)
2355 {
2356 	struct goya_device *goya = hdev->asic_specific;
2357 	int cq_cnt = hdev->asic_prop.completion_queues_count;
2358 	int rc, i, irq_cnt_init, irq;
2359 
2360 	if (goya->hw_cap_initialized & HW_CAP_MSIX)
2361 		return 0;
2362 
2363 	rc = pci_alloc_irq_vectors(hdev->pdev, GOYA_MSIX_ENTRIES,
2364 				GOYA_MSIX_ENTRIES, PCI_IRQ_MSIX);
2365 	if (rc < 0) {
2366 		dev_err(hdev->dev,
2367 			"MSI-X: Failed to enable support -- %d/%d\n",
2368 			GOYA_MSIX_ENTRIES, rc);
2369 		return rc;
2370 	}
2371 
2372 	for (i = 0, irq_cnt_init = 0 ; i < cq_cnt ; i++, irq_cnt_init++) {
2373 		irq = pci_irq_vector(hdev->pdev, i);
2374 		rc = request_irq(irq, hl_irq_handler_cq, 0, goya_irq_name[i],
2375 				&hdev->completion_queue[i]);
2376 		if (rc) {
2377 			dev_err(hdev->dev, "Failed to request IRQ %d", irq);
2378 			goto free_irqs;
2379 		}
2380 	}
2381 
2382 	irq = pci_irq_vector(hdev->pdev, GOYA_EVENT_QUEUE_MSIX_IDX);
2383 
2384 	rc = request_irq(irq, hl_irq_handler_eq, 0,
2385 			goya_irq_name[GOYA_EVENT_QUEUE_MSIX_IDX],
2386 			&hdev->event_queue);
2387 	if (rc) {
2388 		dev_err(hdev->dev, "Failed to request IRQ %d", irq);
2389 		goto free_irqs;
2390 	}
2391 
2392 	goya->hw_cap_initialized |= HW_CAP_MSIX;
2393 	return 0;
2394 
2395 free_irqs:
2396 	for (i = 0 ; i < irq_cnt_init ; i++)
2397 		free_irq(pci_irq_vector(hdev->pdev, i),
2398 			&hdev->completion_queue[i]);
2399 
2400 	pci_free_irq_vectors(hdev->pdev);
2401 	return rc;
2402 }
2403 
2404 static void goya_sync_irqs(struct hl_device *hdev)
2405 {
2406 	struct goya_device *goya = hdev->asic_specific;
2407 	int i;
2408 
2409 	if (!(goya->hw_cap_initialized & HW_CAP_MSIX))
2410 		return;
2411 
2412 	/* Wait for all pending IRQs to be finished */
2413 	for (i = 0 ; i < hdev->asic_prop.completion_queues_count ; i++)
2414 		synchronize_irq(pci_irq_vector(hdev->pdev, i));
2415 
2416 	synchronize_irq(pci_irq_vector(hdev->pdev, GOYA_EVENT_QUEUE_MSIX_IDX));
2417 }
2418 
2419 static void goya_disable_msix(struct hl_device *hdev)
2420 {
2421 	struct goya_device *goya = hdev->asic_specific;
2422 	int i, irq;
2423 
2424 	if (!(goya->hw_cap_initialized & HW_CAP_MSIX))
2425 		return;
2426 
2427 	goya_sync_irqs(hdev);
2428 
2429 	irq = pci_irq_vector(hdev->pdev, GOYA_EVENT_QUEUE_MSIX_IDX);
2430 	free_irq(irq, &hdev->event_queue);
2431 
2432 	for (i = 0 ; i < hdev->asic_prop.completion_queues_count ; i++) {
2433 		irq = pci_irq_vector(hdev->pdev, i);
2434 		free_irq(irq, &hdev->completion_queue[i]);
2435 	}
2436 
2437 	pci_free_irq_vectors(hdev->pdev);
2438 
2439 	goya->hw_cap_initialized &= ~HW_CAP_MSIX;
2440 }
2441 
2442 static void goya_enable_timestamp(struct hl_device *hdev)
2443 {
2444 	/* Disable the timestamp counter */
2445 	WREG32(mmPSOC_TIMESTAMP_BASE - CFG_BASE, 0);
2446 
2447 	/* Zero the lower/upper parts of the 64-bit counter */
2448 	WREG32(mmPSOC_TIMESTAMP_BASE - CFG_BASE + 0xC, 0);
2449 	WREG32(mmPSOC_TIMESTAMP_BASE - CFG_BASE + 0x8, 0);
2450 
2451 	/* Enable the counter */
2452 	WREG32(mmPSOC_TIMESTAMP_BASE - CFG_BASE, 1);
2453 }
2454 
2455 static void goya_disable_timestamp(struct hl_device *hdev)
2456 {
2457 	/* Disable the timestamp counter */
2458 	WREG32(mmPSOC_TIMESTAMP_BASE - CFG_BASE, 0);
2459 }
2460 
2461 static void goya_halt_engines(struct hl_device *hdev, bool hard_reset, bool fw_reset)
2462 {
2463 	u32 wait_timeout_ms;
2464 
2465 	if (hdev->pldm)
2466 		wait_timeout_ms = GOYA_PLDM_RESET_WAIT_MSEC;
2467 	else
2468 		wait_timeout_ms = GOYA_RESET_WAIT_MSEC;
2469 
2470 	goya_stop_external_queues(hdev);
2471 	goya_stop_internal_queues(hdev);
2472 
2473 	msleep(wait_timeout_ms);
2474 
2475 	goya_dma_stall(hdev);
2476 	goya_tpc_stall(hdev);
2477 	goya_mme_stall(hdev);
2478 
2479 	msleep(wait_timeout_ms);
2480 
2481 	goya_disable_external_queues(hdev);
2482 	goya_disable_internal_queues(hdev);
2483 
2484 	goya_disable_timestamp(hdev);
2485 
2486 	if (hard_reset) {
2487 		goya_disable_msix(hdev);
2488 		goya_mmu_remove_device_cpu_mappings(hdev);
2489 	} else {
2490 		goya_sync_irqs(hdev);
2491 	}
2492 }
2493 
2494 /*
2495  * goya_load_firmware_to_device() - Load LINUX FW code to device.
2496  * @hdev: Pointer to hl_device structure.
2497  *
2498  * Copy LINUX fw code from firmware file to HBM BAR.
2499  *
2500  * Return: 0 on success, non-zero for failure.
2501  */
2502 static int goya_load_firmware_to_device(struct hl_device *hdev)
2503 {
2504 	void __iomem *dst;
2505 
2506 	dst = hdev->pcie_bar[DDR_BAR_ID] + LINUX_FW_OFFSET;
2507 
2508 	return hl_fw_load_fw_to_device(hdev, GOYA_LINUX_FW_FILE, dst, 0, 0);
2509 }
2510 
2511 /*
2512  * goya_load_boot_fit_to_device() - Load boot fit to device.
2513  * @hdev: Pointer to hl_device structure.
2514  *
2515  * Copy boot fit file to SRAM BAR.
2516  *
2517  * Return: 0 on success, non-zero for failure.
2518  */
2519 static int goya_load_boot_fit_to_device(struct hl_device *hdev)
2520 {
2521 	void __iomem *dst;
2522 
2523 	dst = hdev->pcie_bar[SRAM_CFG_BAR_ID] + BOOT_FIT_SRAM_OFFSET;
2524 
2525 	return hl_fw_load_fw_to_device(hdev, GOYA_BOOT_FIT_FILE, dst, 0, 0);
2526 }
2527 
2528 static void goya_init_dynamic_firmware_loader(struct hl_device *hdev)
2529 {
2530 	struct dynamic_fw_load_mgr *dynamic_loader;
2531 	struct cpu_dyn_regs *dyn_regs;
2532 
2533 	dynamic_loader = &hdev->fw_loader.dynamic_loader;
2534 
2535 	/*
2536 	 * here we update initial values for few specific dynamic regs (as
2537 	 * before reading the first descriptor from FW those value has to be
2538 	 * hard-coded) in later stages of the protocol those values will be
2539 	 * updated automatically by reading the FW descriptor so data there
2540 	 * will always be up-to-date
2541 	 */
2542 	dyn_regs = &dynamic_loader->comm_desc.cpu_dyn_regs;
2543 	dyn_regs->kmd_msg_to_cpu =
2544 				cpu_to_le32(mmPSOC_GLOBAL_CONF_KMD_MSG_TO_CPU);
2545 	dyn_regs->cpu_cmd_status_to_host =
2546 				cpu_to_le32(mmCPU_CMD_STATUS_TO_HOST);
2547 
2548 	dynamic_loader->wait_for_bl_timeout = GOYA_WAIT_FOR_BL_TIMEOUT_USEC;
2549 }
2550 
2551 static void goya_init_static_firmware_loader(struct hl_device *hdev)
2552 {
2553 	struct static_fw_load_mgr *static_loader;
2554 
2555 	static_loader = &hdev->fw_loader.static_loader;
2556 
2557 	static_loader->preboot_version_max_off = SRAM_SIZE - VERSION_MAX_LEN;
2558 	static_loader->boot_fit_version_max_off = SRAM_SIZE - VERSION_MAX_LEN;
2559 	static_loader->kmd_msg_to_cpu_reg = mmPSOC_GLOBAL_CONF_KMD_MSG_TO_CPU;
2560 	static_loader->cpu_cmd_status_to_host_reg = mmCPU_CMD_STATUS_TO_HOST;
2561 	static_loader->cpu_boot_status_reg = mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS;
2562 	static_loader->cpu_boot_dev_status0_reg = mmCPU_BOOT_DEV_STS0;
2563 	static_loader->cpu_boot_dev_status1_reg = mmCPU_BOOT_DEV_STS1;
2564 	static_loader->boot_err0_reg = mmCPU_BOOT_ERR0;
2565 	static_loader->boot_err1_reg = mmCPU_BOOT_ERR1;
2566 	static_loader->preboot_version_offset_reg = mmPREBOOT_VER_OFFSET;
2567 	static_loader->boot_fit_version_offset_reg = mmUBOOT_VER_OFFSET;
2568 	static_loader->sram_offset_mask = ~(lower_32_bits(SRAM_BASE_ADDR));
2569 }
2570 
2571 static void goya_init_firmware_preload_params(struct hl_device *hdev)
2572 {
2573 	struct pre_fw_load_props *pre_fw_load = &hdev->fw_loader.pre_fw_load;
2574 
2575 	pre_fw_load->cpu_boot_status_reg = mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS;
2576 	pre_fw_load->sts_boot_dev_sts0_reg = mmCPU_BOOT_DEV_STS0;
2577 	pre_fw_load->sts_boot_dev_sts1_reg = mmCPU_BOOT_DEV_STS1;
2578 	pre_fw_load->boot_err0_reg = mmCPU_BOOT_ERR0;
2579 	pre_fw_load->boot_err1_reg = mmCPU_BOOT_ERR1;
2580 	pre_fw_load->wait_for_preboot_timeout = GOYA_BOOT_FIT_REQ_TIMEOUT_USEC;
2581 }
2582 
2583 static void goya_init_firmware_loader(struct hl_device *hdev)
2584 {
2585 	struct asic_fixed_properties *prop = &hdev->asic_prop;
2586 	struct fw_load_mgr *fw_loader = &hdev->fw_loader;
2587 
2588 	/* fill common fields */
2589 	fw_loader->fw_comp_loaded = FW_TYPE_NONE;
2590 	fw_loader->boot_fit_img.image_name = GOYA_BOOT_FIT_FILE;
2591 	fw_loader->linux_img.image_name = GOYA_LINUX_FW_FILE;
2592 	fw_loader->cpu_timeout = GOYA_CPU_TIMEOUT_USEC;
2593 	fw_loader->boot_fit_timeout = GOYA_BOOT_FIT_REQ_TIMEOUT_USEC;
2594 	fw_loader->skip_bmc = false;
2595 	fw_loader->sram_bar_id = SRAM_CFG_BAR_ID;
2596 	fw_loader->dram_bar_id = DDR_BAR_ID;
2597 
2598 	if (prop->dynamic_fw_load)
2599 		goya_init_dynamic_firmware_loader(hdev);
2600 	else
2601 		goya_init_static_firmware_loader(hdev);
2602 }
2603 
2604 static int goya_init_cpu(struct hl_device *hdev)
2605 {
2606 	struct goya_device *goya = hdev->asic_specific;
2607 	int rc;
2608 
2609 	if (!(hdev->fw_components & FW_TYPE_PREBOOT_CPU))
2610 		return 0;
2611 
2612 	if (goya->hw_cap_initialized & HW_CAP_CPU)
2613 		return 0;
2614 
2615 	/*
2616 	 * Before pushing u-boot/linux to device, need to set the ddr bar to
2617 	 * base address of dram
2618 	 */
2619 	if (goya_set_ddr_bar_base(hdev, DRAM_PHYS_BASE) == U64_MAX) {
2620 		dev_err(hdev->dev,
2621 			"failed to map DDR bar to DRAM base address\n");
2622 		return -EIO;
2623 	}
2624 
2625 	rc = hl_fw_init_cpu(hdev);
2626 
2627 	if (rc)
2628 		return rc;
2629 
2630 	goya->hw_cap_initialized |= HW_CAP_CPU;
2631 
2632 	return 0;
2633 }
2634 
2635 static int goya_mmu_update_asid_hop0_addr(struct hl_device *hdev, u32 asid,
2636 						u64 phys_addr)
2637 {
2638 	u32 status, timeout_usec;
2639 	int rc;
2640 
2641 	if (hdev->pldm)
2642 		timeout_usec = GOYA_PLDM_MMU_TIMEOUT_USEC;
2643 	else
2644 		timeout_usec = MMU_CONFIG_TIMEOUT_USEC;
2645 
2646 	WREG32(MMU_HOP0_PA43_12, phys_addr >> MMU_HOP0_PA43_12_SHIFT);
2647 	WREG32(MMU_HOP0_PA49_44, phys_addr >> MMU_HOP0_PA49_44_SHIFT);
2648 	WREG32(MMU_ASID_BUSY, 0x80000000 | asid);
2649 
2650 	rc = hl_poll_timeout(
2651 		hdev,
2652 		MMU_ASID_BUSY,
2653 		status,
2654 		!(status & 0x80000000),
2655 		1000,
2656 		timeout_usec);
2657 
2658 	if (rc) {
2659 		dev_err(hdev->dev,
2660 			"Timeout during MMU hop0 config of asid %d\n", asid);
2661 		return rc;
2662 	}
2663 
2664 	return 0;
2665 }
2666 
2667 int goya_mmu_init(struct hl_device *hdev)
2668 {
2669 	struct asic_fixed_properties *prop = &hdev->asic_prop;
2670 	struct goya_device *goya = hdev->asic_specific;
2671 	u64 hop0_addr;
2672 	int rc, i;
2673 
2674 	if (goya->hw_cap_initialized & HW_CAP_MMU)
2675 		return 0;
2676 
2677 	hdev->dram_default_page_mapping = true;
2678 
2679 	for (i = 0 ; i < prop->max_asid ; i++) {
2680 		hop0_addr = prop->mmu_pgt_addr +
2681 				(i * prop->mmu_hop_table_size);
2682 
2683 		rc = goya_mmu_update_asid_hop0_addr(hdev, i, hop0_addr);
2684 		if (rc) {
2685 			dev_err(hdev->dev,
2686 				"failed to set hop0 addr for asid %d\n", i);
2687 			goto err;
2688 		}
2689 	}
2690 
2691 	goya->hw_cap_initialized |= HW_CAP_MMU;
2692 
2693 	/* init MMU cache manage page */
2694 	WREG32(mmSTLB_CACHE_INV_BASE_39_8,
2695 				lower_32_bits(MMU_CACHE_MNG_ADDR >> 8));
2696 	WREG32(mmSTLB_CACHE_INV_BASE_49_40, MMU_CACHE_MNG_ADDR >> 40);
2697 
2698 	/* Remove follower feature due to performance bug */
2699 	WREG32_AND(mmSTLB_STLB_FEATURE_EN,
2700 			(~STLB_STLB_FEATURE_EN_FOLLOWER_EN_MASK));
2701 
2702 	hl_mmu_invalidate_cache(hdev, true, MMU_OP_USERPTR | MMU_OP_PHYS_PACK);
2703 
2704 	WREG32(mmMMU_MMU_ENABLE, 1);
2705 	WREG32(mmMMU_SPI_MASK, 0xF);
2706 
2707 	return 0;
2708 
2709 err:
2710 	return rc;
2711 }
2712 
2713 /*
2714  * goya_hw_init - Goya hardware initialization code
2715  *
2716  * @hdev: pointer to hl_device structure
2717  *
2718  * Returns 0 on success
2719  *
2720  */
2721 static int goya_hw_init(struct hl_device *hdev)
2722 {
2723 	struct asic_fixed_properties *prop = &hdev->asic_prop;
2724 	int rc;
2725 
2726 	/* Perform read from the device to make sure device is up */
2727 	RREG32(mmPCIE_DBI_DEVICE_ID_VENDOR_ID_REG);
2728 
2729 	/*
2730 	 * Let's mark in the H/W that we have reached this point. We check
2731 	 * this value in the reset_before_init function to understand whether
2732 	 * we need to reset the chip before doing H/W init. This register is
2733 	 * cleared by the H/W upon H/W reset
2734 	 */
2735 	WREG32(mmHW_STATE, HL_DEVICE_HW_STATE_DIRTY);
2736 
2737 	rc = goya_init_cpu(hdev);
2738 	if (rc) {
2739 		dev_err(hdev->dev, "failed to initialize CPU\n");
2740 		return rc;
2741 	}
2742 
2743 	goya_tpc_mbist_workaround(hdev);
2744 
2745 	goya_init_golden_registers(hdev);
2746 
2747 	/*
2748 	 * After CPU initialization is finished, change DDR bar mapping inside
2749 	 * iATU to point to the start address of the MMU page tables
2750 	 */
2751 	if (goya_set_ddr_bar_base(hdev, (MMU_PAGE_TABLES_ADDR &
2752 			~(prop->dram_pci_bar_size - 0x1ull))) == U64_MAX) {
2753 		dev_err(hdev->dev,
2754 			"failed to map DDR bar to MMU page tables\n");
2755 		return -EIO;
2756 	}
2757 
2758 	rc = goya_mmu_init(hdev);
2759 	if (rc)
2760 		return rc;
2761 
2762 	goya_init_security(hdev);
2763 
2764 	goya_init_dma_qmans(hdev);
2765 
2766 	goya_init_mme_qmans(hdev);
2767 
2768 	goya_init_tpc_qmans(hdev);
2769 
2770 	goya_enable_timestamp(hdev);
2771 
2772 	/* MSI-X must be enabled before CPU queues are initialized */
2773 	rc = goya_enable_msix(hdev);
2774 	if (rc)
2775 		goto disable_queues;
2776 
2777 	/* Perform read from the device to flush all MSI-X configuration */
2778 	RREG32(mmPCIE_DBI_DEVICE_ID_VENDOR_ID_REG);
2779 
2780 	return 0;
2781 
2782 disable_queues:
2783 	goya_disable_internal_queues(hdev);
2784 	goya_disable_external_queues(hdev);
2785 
2786 	return rc;
2787 }
2788 
2789 static int goya_hw_fini(struct hl_device *hdev, bool hard_reset, bool fw_reset)
2790 {
2791 	struct goya_device *goya = hdev->asic_specific;
2792 	u32 reset_timeout_ms, cpu_timeout_ms, status;
2793 
2794 	if (hdev->pldm) {
2795 		reset_timeout_ms = GOYA_PLDM_RESET_TIMEOUT_MSEC;
2796 		cpu_timeout_ms = GOYA_PLDM_RESET_WAIT_MSEC;
2797 	} else {
2798 		reset_timeout_ms = GOYA_RESET_TIMEOUT_MSEC;
2799 		cpu_timeout_ms = GOYA_CPU_RESET_WAIT_MSEC;
2800 	}
2801 
2802 	if (hard_reset) {
2803 		/* I don't know what is the state of the CPU so make sure it is
2804 		 * stopped in any means necessary
2805 		 */
2806 		WREG32(mmPSOC_GLOBAL_CONF_UBOOT_MAGIC, KMD_MSG_GOTO_WFE);
2807 		WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR,
2808 			GOYA_ASYNC_EVENT_ID_HALT_MACHINE);
2809 
2810 		msleep(cpu_timeout_ms);
2811 
2812 		goya_set_ddr_bar_base(hdev, DRAM_PHYS_BASE);
2813 		goya_disable_clk_rlx(hdev);
2814 		goya_set_pll_refclk(hdev);
2815 
2816 		WREG32(mmPSOC_GLOBAL_CONF_SW_ALL_RST_CFG, RESET_ALL);
2817 		dev_dbg(hdev->dev,
2818 			"Issued HARD reset command, going to wait %dms\n",
2819 			reset_timeout_ms);
2820 	} else {
2821 		WREG32(mmPSOC_GLOBAL_CONF_SW_ALL_RST_CFG, DMA_MME_TPC_RESET);
2822 		dev_dbg(hdev->dev,
2823 			"Issued SOFT reset command, going to wait %dms\n",
2824 			reset_timeout_ms);
2825 	}
2826 
2827 	/*
2828 	 * After hard reset, we can't poll the BTM_FSM register because the PSOC
2829 	 * itself is in reset. In either reset we need to wait until the reset
2830 	 * is deasserted
2831 	 */
2832 	msleep(reset_timeout_ms);
2833 
2834 	status = RREG32(mmPSOC_GLOBAL_CONF_BTM_FSM);
2835 	if (status & PSOC_GLOBAL_CONF_BTM_FSM_STATE_MASK) {
2836 		dev_err(hdev->dev, "Timeout while waiting for device to reset 0x%x\n", status);
2837 		return -ETIMEDOUT;
2838 	}
2839 
2840 	if (!hard_reset && goya) {
2841 		goya->hw_cap_initialized &= ~(HW_CAP_DMA | HW_CAP_MME |
2842 						HW_CAP_GOLDEN | HW_CAP_TPC);
2843 		WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR,
2844 				GOYA_ASYNC_EVENT_ID_SOFT_RESET);
2845 		return 0;
2846 	}
2847 
2848 	/* Chicken bit to re-initiate boot sequencer flow */
2849 	WREG32(mmPSOC_GLOBAL_CONF_BOOT_SEQ_RE_START,
2850 		1 << PSOC_GLOBAL_CONF_BOOT_SEQ_RE_START_IND_SHIFT);
2851 	/* Move boot manager FSM to pre boot sequencer init state */
2852 	WREG32(mmPSOC_GLOBAL_CONF_SW_BTM_FSM,
2853 			0xA << PSOC_GLOBAL_CONF_SW_BTM_FSM_CTRL_SHIFT);
2854 
2855 	if (goya) {
2856 		goya->hw_cap_initialized &= ~(HW_CAP_CPU | HW_CAP_CPU_Q |
2857 				HW_CAP_DDR_0 | HW_CAP_DDR_1 |
2858 				HW_CAP_DMA | HW_CAP_MME |
2859 				HW_CAP_MMU | HW_CAP_TPC_MBIST |
2860 				HW_CAP_GOLDEN | HW_CAP_TPC);
2861 
2862 		memset(goya->events_stat, 0, sizeof(goya->events_stat));
2863 	}
2864 	return 0;
2865 }
2866 
2867 int goya_suspend(struct hl_device *hdev)
2868 {
2869 	int rc;
2870 
2871 	rc = hl_fw_send_pci_access_msg(hdev, CPUCP_PACKET_DISABLE_PCI_ACCESS, 0x0);
2872 	if (rc)
2873 		dev_err(hdev->dev, "Failed to disable PCI access from CPU\n");
2874 
2875 	return rc;
2876 }
2877 
2878 int goya_resume(struct hl_device *hdev)
2879 {
2880 	return goya_init_iatu(hdev);
2881 }
2882 
2883 static int goya_mmap(struct hl_device *hdev, struct vm_area_struct *vma,
2884 			void *cpu_addr, dma_addr_t dma_addr, size_t size)
2885 {
2886 	int rc;
2887 
2888 	vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP |
2889 			VM_DONTCOPY | VM_NORESERVE);
2890 
2891 	rc = dma_mmap_coherent(hdev->dev, vma, cpu_addr,
2892 				(dma_addr - HOST_PHYS_BASE), size);
2893 	if (rc)
2894 		dev_err(hdev->dev, "dma_mmap_coherent error %d", rc);
2895 
2896 	return rc;
2897 }
2898 
2899 void goya_ring_doorbell(struct hl_device *hdev, u32 hw_queue_id, u32 pi)
2900 {
2901 	u32 db_reg_offset, db_value;
2902 
2903 	switch (hw_queue_id) {
2904 	case GOYA_QUEUE_ID_DMA_0:
2905 		db_reg_offset = mmDMA_QM_0_PQ_PI;
2906 		break;
2907 
2908 	case GOYA_QUEUE_ID_DMA_1:
2909 		db_reg_offset = mmDMA_QM_1_PQ_PI;
2910 		break;
2911 
2912 	case GOYA_QUEUE_ID_DMA_2:
2913 		db_reg_offset = mmDMA_QM_2_PQ_PI;
2914 		break;
2915 
2916 	case GOYA_QUEUE_ID_DMA_3:
2917 		db_reg_offset = mmDMA_QM_3_PQ_PI;
2918 		break;
2919 
2920 	case GOYA_QUEUE_ID_DMA_4:
2921 		db_reg_offset = mmDMA_QM_4_PQ_PI;
2922 		break;
2923 
2924 	case GOYA_QUEUE_ID_CPU_PQ:
2925 		db_reg_offset = mmCPU_IF_PF_PQ_PI;
2926 		break;
2927 
2928 	case GOYA_QUEUE_ID_MME:
2929 		db_reg_offset = mmMME_QM_PQ_PI;
2930 		break;
2931 
2932 	case GOYA_QUEUE_ID_TPC0:
2933 		db_reg_offset = mmTPC0_QM_PQ_PI;
2934 		break;
2935 
2936 	case GOYA_QUEUE_ID_TPC1:
2937 		db_reg_offset = mmTPC1_QM_PQ_PI;
2938 		break;
2939 
2940 	case GOYA_QUEUE_ID_TPC2:
2941 		db_reg_offset = mmTPC2_QM_PQ_PI;
2942 		break;
2943 
2944 	case GOYA_QUEUE_ID_TPC3:
2945 		db_reg_offset = mmTPC3_QM_PQ_PI;
2946 		break;
2947 
2948 	case GOYA_QUEUE_ID_TPC4:
2949 		db_reg_offset = mmTPC4_QM_PQ_PI;
2950 		break;
2951 
2952 	case GOYA_QUEUE_ID_TPC5:
2953 		db_reg_offset = mmTPC5_QM_PQ_PI;
2954 		break;
2955 
2956 	case GOYA_QUEUE_ID_TPC6:
2957 		db_reg_offset = mmTPC6_QM_PQ_PI;
2958 		break;
2959 
2960 	case GOYA_QUEUE_ID_TPC7:
2961 		db_reg_offset = mmTPC7_QM_PQ_PI;
2962 		break;
2963 
2964 	default:
2965 		/* Should never get here */
2966 		dev_err(hdev->dev, "H/W queue %d is invalid. Can't set pi\n",
2967 			hw_queue_id);
2968 		return;
2969 	}
2970 
2971 	db_value = pi;
2972 
2973 	/* ring the doorbell */
2974 	WREG32(db_reg_offset, db_value);
2975 
2976 	if (hw_queue_id == GOYA_QUEUE_ID_CPU_PQ) {
2977 		/* make sure device CPU will read latest data from host */
2978 		mb();
2979 		WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR,
2980 				GOYA_ASYNC_EVENT_ID_PI_UPDATE);
2981 	}
2982 }
2983 
2984 void goya_pqe_write(struct hl_device *hdev, __le64 *pqe, struct hl_bd *bd)
2985 {
2986 	/* The QMANs are on the SRAM so need to copy to IO space */
2987 	memcpy_toio((void __iomem *) pqe, bd, sizeof(struct hl_bd));
2988 }
2989 
2990 static void *goya_dma_alloc_coherent(struct hl_device *hdev, size_t size,
2991 					dma_addr_t *dma_handle, gfp_t flags)
2992 {
2993 	void *kernel_addr = dma_alloc_coherent(&hdev->pdev->dev, size,
2994 						dma_handle, flags);
2995 
2996 	/* Shift to the device's base physical address of host memory */
2997 	if (kernel_addr)
2998 		*dma_handle += HOST_PHYS_BASE;
2999 
3000 	return kernel_addr;
3001 }
3002 
3003 static void goya_dma_free_coherent(struct hl_device *hdev, size_t size,
3004 					void *cpu_addr, dma_addr_t dma_handle)
3005 {
3006 	/* Cancel the device's base physical address of host memory */
3007 	dma_addr_t fixed_dma_handle = dma_handle - HOST_PHYS_BASE;
3008 
3009 	dma_free_coherent(&hdev->pdev->dev, size, cpu_addr, fixed_dma_handle);
3010 }
3011 
3012 int goya_scrub_device_mem(struct hl_device *hdev)
3013 {
3014 	return 0;
3015 }
3016 
3017 void *goya_get_int_queue_base(struct hl_device *hdev, u32 queue_id,
3018 				dma_addr_t *dma_handle,	u16 *queue_len)
3019 {
3020 	void *base;
3021 	u32 offset;
3022 
3023 	*dma_handle = hdev->asic_prop.sram_base_address;
3024 
3025 	base = (__force void *) hdev->pcie_bar[SRAM_CFG_BAR_ID];
3026 
3027 	switch (queue_id) {
3028 	case GOYA_QUEUE_ID_MME:
3029 		offset = MME_QMAN_BASE_OFFSET;
3030 		*queue_len = MME_QMAN_LENGTH;
3031 		break;
3032 	case GOYA_QUEUE_ID_TPC0:
3033 		offset = TPC0_QMAN_BASE_OFFSET;
3034 		*queue_len = TPC_QMAN_LENGTH;
3035 		break;
3036 	case GOYA_QUEUE_ID_TPC1:
3037 		offset = TPC1_QMAN_BASE_OFFSET;
3038 		*queue_len = TPC_QMAN_LENGTH;
3039 		break;
3040 	case GOYA_QUEUE_ID_TPC2:
3041 		offset = TPC2_QMAN_BASE_OFFSET;
3042 		*queue_len = TPC_QMAN_LENGTH;
3043 		break;
3044 	case GOYA_QUEUE_ID_TPC3:
3045 		offset = TPC3_QMAN_BASE_OFFSET;
3046 		*queue_len = TPC_QMAN_LENGTH;
3047 		break;
3048 	case GOYA_QUEUE_ID_TPC4:
3049 		offset = TPC4_QMAN_BASE_OFFSET;
3050 		*queue_len = TPC_QMAN_LENGTH;
3051 		break;
3052 	case GOYA_QUEUE_ID_TPC5:
3053 		offset = TPC5_QMAN_BASE_OFFSET;
3054 		*queue_len = TPC_QMAN_LENGTH;
3055 		break;
3056 	case GOYA_QUEUE_ID_TPC6:
3057 		offset = TPC6_QMAN_BASE_OFFSET;
3058 		*queue_len = TPC_QMAN_LENGTH;
3059 		break;
3060 	case GOYA_QUEUE_ID_TPC7:
3061 		offset = TPC7_QMAN_BASE_OFFSET;
3062 		*queue_len = TPC_QMAN_LENGTH;
3063 		break;
3064 	default:
3065 		dev_err(hdev->dev, "Got invalid queue id %d\n", queue_id);
3066 		return NULL;
3067 	}
3068 
3069 	base += offset;
3070 	*dma_handle += offset;
3071 
3072 	return base;
3073 }
3074 
3075 static int goya_send_job_on_qman0(struct hl_device *hdev, struct hl_cs_job *job)
3076 {
3077 	struct packet_msg_prot *fence_pkt;
3078 	u32 *fence_ptr;
3079 	dma_addr_t fence_dma_addr;
3080 	struct hl_cb *cb;
3081 	u32 tmp, timeout;
3082 	int rc;
3083 
3084 	if (hdev->pldm)
3085 		timeout = GOYA_PLDM_QMAN0_TIMEOUT_USEC;
3086 	else
3087 		timeout = HL_DEVICE_TIMEOUT_USEC;
3088 
3089 	if (!hdev->asic_funcs->is_device_idle(hdev, NULL, 0, NULL)) {
3090 		dev_err_ratelimited(hdev->dev,
3091 			"Can't send driver job on QMAN0 because the device is not idle\n");
3092 		return -EBUSY;
3093 	}
3094 
3095 	fence_ptr = hl_asic_dma_pool_zalloc(hdev, 4, GFP_KERNEL, &fence_dma_addr);
3096 	if (!fence_ptr) {
3097 		dev_err(hdev->dev,
3098 			"Failed to allocate fence memory for QMAN0\n");
3099 		return -ENOMEM;
3100 	}
3101 
3102 	goya_qman0_set_security(hdev, true);
3103 
3104 	cb = job->patched_cb;
3105 
3106 	fence_pkt = cb->kernel_address +
3107 			job->job_cb_size - sizeof(struct packet_msg_prot);
3108 
3109 	tmp = (PACKET_MSG_PROT << GOYA_PKT_CTL_OPCODE_SHIFT) |
3110 			(1 << GOYA_PKT_CTL_EB_SHIFT) |
3111 			(1 << GOYA_PKT_CTL_MB_SHIFT);
3112 	fence_pkt->ctl = cpu_to_le32(tmp);
3113 	fence_pkt->value = cpu_to_le32(GOYA_QMAN0_FENCE_VAL);
3114 	fence_pkt->addr = cpu_to_le64(fence_dma_addr);
3115 
3116 	rc = hl_hw_queue_send_cb_no_cmpl(hdev, GOYA_QUEUE_ID_DMA_0,
3117 					job->job_cb_size, cb->bus_address);
3118 	if (rc) {
3119 		dev_err(hdev->dev, "Failed to send CB on QMAN0, %d\n", rc);
3120 		goto free_fence_ptr;
3121 	}
3122 
3123 	rc = hl_poll_timeout_memory(hdev, fence_ptr, tmp,
3124 				(tmp == GOYA_QMAN0_FENCE_VAL), 1000,
3125 				timeout, true);
3126 
3127 	hl_hw_queue_inc_ci_kernel(hdev, GOYA_QUEUE_ID_DMA_0);
3128 
3129 	if (rc == -ETIMEDOUT) {
3130 		dev_err(hdev->dev, "QMAN0 Job timeout (0x%x)\n", tmp);
3131 		goto free_fence_ptr;
3132 	}
3133 
3134 free_fence_ptr:
3135 	hl_asic_dma_pool_free(hdev, (void *) fence_ptr, fence_dma_addr);
3136 
3137 	goya_qman0_set_security(hdev, false);
3138 
3139 	return rc;
3140 }
3141 
3142 int goya_send_cpu_message(struct hl_device *hdev, u32 *msg, u16 len,
3143 				u32 timeout, u64 *result)
3144 {
3145 	struct goya_device *goya = hdev->asic_specific;
3146 
3147 	if (!(goya->hw_cap_initialized & HW_CAP_CPU_Q)) {
3148 		if (result)
3149 			*result = 0;
3150 		return 0;
3151 	}
3152 
3153 	if (!timeout)
3154 		timeout = GOYA_MSG_TO_CPU_TIMEOUT_USEC;
3155 
3156 	return hl_fw_send_cpu_message(hdev, GOYA_QUEUE_ID_CPU_PQ, msg, len,
3157 					timeout, result);
3158 }
3159 
3160 int goya_test_queue(struct hl_device *hdev, u32 hw_queue_id)
3161 {
3162 	struct packet_msg_prot *fence_pkt;
3163 	dma_addr_t pkt_dma_addr;
3164 	u32 fence_val, tmp;
3165 	dma_addr_t fence_dma_addr;
3166 	u32 *fence_ptr;
3167 	int rc;
3168 
3169 	fence_val = GOYA_QMAN0_FENCE_VAL;
3170 
3171 	fence_ptr = hl_asic_dma_pool_zalloc(hdev, 4, GFP_KERNEL, &fence_dma_addr);
3172 	if (!fence_ptr) {
3173 		dev_err(hdev->dev,
3174 			"Failed to allocate memory for H/W queue %d testing\n",
3175 			hw_queue_id);
3176 		return -ENOMEM;
3177 	}
3178 
3179 	*fence_ptr = 0;
3180 
3181 	fence_pkt = hl_asic_dma_pool_zalloc(hdev, sizeof(struct packet_msg_prot), GFP_KERNEL,
3182 						&pkt_dma_addr);
3183 	if (!fence_pkt) {
3184 		dev_err(hdev->dev,
3185 			"Failed to allocate packet for H/W queue %d testing\n",
3186 			hw_queue_id);
3187 		rc = -ENOMEM;
3188 		goto free_fence_ptr;
3189 	}
3190 
3191 	tmp = (PACKET_MSG_PROT << GOYA_PKT_CTL_OPCODE_SHIFT) |
3192 			(1 << GOYA_PKT_CTL_EB_SHIFT) |
3193 			(1 << GOYA_PKT_CTL_MB_SHIFT);
3194 	fence_pkt->ctl = cpu_to_le32(tmp);
3195 	fence_pkt->value = cpu_to_le32(fence_val);
3196 	fence_pkt->addr = cpu_to_le64(fence_dma_addr);
3197 
3198 	rc = hl_hw_queue_send_cb_no_cmpl(hdev, hw_queue_id,
3199 					sizeof(struct packet_msg_prot),
3200 					pkt_dma_addr);
3201 	if (rc) {
3202 		dev_err(hdev->dev,
3203 			"Failed to send fence packet to H/W queue %d\n",
3204 			hw_queue_id);
3205 		goto free_pkt;
3206 	}
3207 
3208 	rc = hl_poll_timeout_memory(hdev, fence_ptr, tmp, (tmp == fence_val),
3209 					1000, GOYA_TEST_QUEUE_WAIT_USEC, true);
3210 
3211 	hl_hw_queue_inc_ci_kernel(hdev, hw_queue_id);
3212 
3213 	if (rc == -ETIMEDOUT) {
3214 		dev_err(hdev->dev,
3215 			"H/W queue %d test failed (scratch(0x%08llX) == 0x%08X)\n",
3216 			hw_queue_id, (unsigned long long) fence_dma_addr, tmp);
3217 		rc = -EIO;
3218 	}
3219 
3220 free_pkt:
3221 	hl_asic_dma_pool_free(hdev, (void *) fence_pkt, pkt_dma_addr);
3222 free_fence_ptr:
3223 	hl_asic_dma_pool_free(hdev, (void *) fence_ptr, fence_dma_addr);
3224 	return rc;
3225 }
3226 
3227 int goya_test_cpu_queue(struct hl_device *hdev)
3228 {
3229 	struct goya_device *goya = hdev->asic_specific;
3230 
3231 	/*
3232 	 * check capability here as send_cpu_message() won't update the result
3233 	 * value if no capability
3234 	 */
3235 	if (!(goya->hw_cap_initialized & HW_CAP_CPU_Q))
3236 		return 0;
3237 
3238 	return hl_fw_test_cpu_queue(hdev);
3239 }
3240 
3241 int goya_test_queues(struct hl_device *hdev)
3242 {
3243 	int i, rc, ret_val = 0;
3244 
3245 	for (i = 0 ; i < NUMBER_OF_EXT_HW_QUEUES ; i++) {
3246 		rc = goya_test_queue(hdev, i);
3247 		if (rc)
3248 			ret_val = -EINVAL;
3249 	}
3250 
3251 	return ret_val;
3252 }
3253 
3254 static void *goya_dma_pool_zalloc(struct hl_device *hdev, size_t size,
3255 					gfp_t mem_flags, dma_addr_t *dma_handle)
3256 {
3257 	void *kernel_addr;
3258 
3259 	if (size > GOYA_DMA_POOL_BLK_SIZE)
3260 		return NULL;
3261 
3262 	kernel_addr =  dma_pool_zalloc(hdev->dma_pool, mem_flags, dma_handle);
3263 
3264 	/* Shift to the device's base physical address of host memory */
3265 	if (kernel_addr)
3266 		*dma_handle += HOST_PHYS_BASE;
3267 
3268 	return kernel_addr;
3269 }
3270 
3271 static void goya_dma_pool_free(struct hl_device *hdev, void *vaddr,
3272 				dma_addr_t dma_addr)
3273 {
3274 	/* Cancel the device's base physical address of host memory */
3275 	dma_addr_t fixed_dma_addr = dma_addr - HOST_PHYS_BASE;
3276 
3277 	dma_pool_free(hdev->dma_pool, vaddr, fixed_dma_addr);
3278 }
3279 
3280 void *goya_cpu_accessible_dma_pool_alloc(struct hl_device *hdev, size_t size,
3281 					dma_addr_t *dma_handle)
3282 {
3283 	void *vaddr;
3284 
3285 	vaddr = hl_fw_cpu_accessible_dma_pool_alloc(hdev, size, dma_handle);
3286 	*dma_handle = (*dma_handle) - hdev->cpu_accessible_dma_address +
3287 			VA_CPU_ACCESSIBLE_MEM_ADDR;
3288 
3289 	return vaddr;
3290 }
3291 
3292 void goya_cpu_accessible_dma_pool_free(struct hl_device *hdev, size_t size,
3293 					void *vaddr)
3294 {
3295 	hl_fw_cpu_accessible_dma_pool_free(hdev, size, vaddr);
3296 }
3297 
3298 u32 goya_get_dma_desc_list_size(struct hl_device *hdev, struct sg_table *sgt)
3299 {
3300 	struct scatterlist *sg, *sg_next_iter;
3301 	u32 count, dma_desc_cnt;
3302 	u64 len, len_next;
3303 	dma_addr_t addr, addr_next;
3304 
3305 	dma_desc_cnt = 0;
3306 
3307 	for_each_sgtable_dma_sg(sgt, sg, count) {
3308 		len = sg_dma_len(sg);
3309 		addr = sg_dma_address(sg);
3310 
3311 		if (len == 0)
3312 			break;
3313 
3314 		while ((count + 1) < sgt->nents) {
3315 			sg_next_iter = sg_next(sg);
3316 			len_next = sg_dma_len(sg_next_iter);
3317 			addr_next = sg_dma_address(sg_next_iter);
3318 
3319 			if (len_next == 0)
3320 				break;
3321 
3322 			if ((addr + len == addr_next) &&
3323 				(len + len_next <= DMA_MAX_TRANSFER_SIZE)) {
3324 				len += len_next;
3325 				count++;
3326 				sg = sg_next_iter;
3327 			} else {
3328 				break;
3329 			}
3330 		}
3331 
3332 		dma_desc_cnt++;
3333 	}
3334 
3335 	return dma_desc_cnt * sizeof(struct packet_lin_dma);
3336 }
3337 
3338 static int goya_pin_memory_before_cs(struct hl_device *hdev,
3339 				struct hl_cs_parser *parser,
3340 				struct packet_lin_dma *user_dma_pkt,
3341 				u64 addr, enum dma_data_direction dir)
3342 {
3343 	struct hl_userptr *userptr;
3344 	int rc;
3345 
3346 	if (hl_userptr_is_pinned(hdev, addr, le32_to_cpu(user_dma_pkt->tsize),
3347 			parser->job_userptr_list, &userptr))
3348 		goto already_pinned;
3349 
3350 	userptr = kzalloc(sizeof(*userptr), GFP_KERNEL);
3351 	if (!userptr)
3352 		return -ENOMEM;
3353 
3354 	rc = hl_pin_host_memory(hdev, addr, le32_to_cpu(user_dma_pkt->tsize),
3355 				userptr);
3356 	if (rc)
3357 		goto free_userptr;
3358 
3359 	list_add_tail(&userptr->job_node, parser->job_userptr_list);
3360 
3361 	rc = hl_dma_map_sgtable(hdev, userptr->sgt, dir);
3362 	if (rc) {
3363 		dev_err(hdev->dev, "failed to map sgt with DMA region\n");
3364 		goto unpin_memory;
3365 	}
3366 
3367 	userptr->dma_mapped = true;
3368 	userptr->dir = dir;
3369 
3370 already_pinned:
3371 	parser->patched_cb_size +=
3372 			goya_get_dma_desc_list_size(hdev, userptr->sgt);
3373 
3374 	return 0;
3375 
3376 unpin_memory:
3377 	list_del(&userptr->job_node);
3378 	hl_unpin_host_memory(hdev, userptr);
3379 free_userptr:
3380 	kfree(userptr);
3381 	return rc;
3382 }
3383 
3384 static int goya_validate_dma_pkt_host(struct hl_device *hdev,
3385 				struct hl_cs_parser *parser,
3386 				struct packet_lin_dma *user_dma_pkt)
3387 {
3388 	u64 device_memory_addr, addr;
3389 	enum dma_data_direction dir;
3390 	enum hl_goya_dma_direction user_dir;
3391 	bool sram_addr = true;
3392 	bool skip_host_mem_pin = false;
3393 	bool user_memset;
3394 	u32 ctl;
3395 	int rc = 0;
3396 
3397 	ctl = le32_to_cpu(user_dma_pkt->ctl);
3398 
3399 	user_dir = (ctl & GOYA_PKT_LIN_DMA_CTL_DMA_DIR_MASK) >>
3400 			GOYA_PKT_LIN_DMA_CTL_DMA_DIR_SHIFT;
3401 
3402 	user_memset = (ctl & GOYA_PKT_LIN_DMA_CTL_MEMSET_MASK) >>
3403 			GOYA_PKT_LIN_DMA_CTL_MEMSET_SHIFT;
3404 
3405 	switch (user_dir) {
3406 	case HL_DMA_HOST_TO_DRAM:
3407 		dev_dbg(hdev->dev, "DMA direction is HOST --> DRAM\n");
3408 		dir = DMA_TO_DEVICE;
3409 		sram_addr = false;
3410 		addr = le64_to_cpu(user_dma_pkt->src_addr);
3411 		device_memory_addr = le64_to_cpu(user_dma_pkt->dst_addr);
3412 		if (user_memset)
3413 			skip_host_mem_pin = true;
3414 		break;
3415 
3416 	case HL_DMA_DRAM_TO_HOST:
3417 		dev_dbg(hdev->dev, "DMA direction is DRAM --> HOST\n");
3418 		dir = DMA_FROM_DEVICE;
3419 		sram_addr = false;
3420 		addr = le64_to_cpu(user_dma_pkt->dst_addr);
3421 		device_memory_addr = le64_to_cpu(user_dma_pkt->src_addr);
3422 		break;
3423 
3424 	case HL_DMA_HOST_TO_SRAM:
3425 		dev_dbg(hdev->dev, "DMA direction is HOST --> SRAM\n");
3426 		dir = DMA_TO_DEVICE;
3427 		addr = le64_to_cpu(user_dma_pkt->src_addr);
3428 		device_memory_addr = le64_to_cpu(user_dma_pkt->dst_addr);
3429 		if (user_memset)
3430 			skip_host_mem_pin = true;
3431 		break;
3432 
3433 	case HL_DMA_SRAM_TO_HOST:
3434 		dev_dbg(hdev->dev, "DMA direction is SRAM --> HOST\n");
3435 		dir = DMA_FROM_DEVICE;
3436 		addr = le64_to_cpu(user_dma_pkt->dst_addr);
3437 		device_memory_addr = le64_to_cpu(user_dma_pkt->src_addr);
3438 		break;
3439 	default:
3440 		dev_err(hdev->dev, "DMA direction %d is unsupported/undefined\n", user_dir);
3441 		return -EFAULT;
3442 	}
3443 
3444 	if (sram_addr) {
3445 		if (!hl_mem_area_inside_range(device_memory_addr,
3446 				le32_to_cpu(user_dma_pkt->tsize),
3447 				hdev->asic_prop.sram_user_base_address,
3448 				hdev->asic_prop.sram_end_address)) {
3449 
3450 			dev_err(hdev->dev,
3451 				"SRAM address 0x%llx + 0x%x is invalid\n",
3452 				device_memory_addr,
3453 				user_dma_pkt->tsize);
3454 			return -EFAULT;
3455 		}
3456 	} else {
3457 		if (!hl_mem_area_inside_range(device_memory_addr,
3458 				le32_to_cpu(user_dma_pkt->tsize),
3459 				hdev->asic_prop.dram_user_base_address,
3460 				hdev->asic_prop.dram_end_address)) {
3461 
3462 			dev_err(hdev->dev,
3463 				"DRAM address 0x%llx + 0x%x is invalid\n",
3464 				device_memory_addr,
3465 				user_dma_pkt->tsize);
3466 			return -EFAULT;
3467 		}
3468 	}
3469 
3470 	if (skip_host_mem_pin)
3471 		parser->patched_cb_size += sizeof(*user_dma_pkt);
3472 	else {
3473 		if ((dir == DMA_TO_DEVICE) &&
3474 				(parser->hw_queue_id > GOYA_QUEUE_ID_DMA_1)) {
3475 			dev_err(hdev->dev,
3476 				"Can't DMA from host on queue other then 1\n");
3477 			return -EFAULT;
3478 		}
3479 
3480 		rc = goya_pin_memory_before_cs(hdev, parser, user_dma_pkt,
3481 						addr, dir);
3482 	}
3483 
3484 	return rc;
3485 }
3486 
3487 static int goya_validate_dma_pkt_no_host(struct hl_device *hdev,
3488 				struct hl_cs_parser *parser,
3489 				struct packet_lin_dma *user_dma_pkt)
3490 {
3491 	u64 sram_memory_addr, dram_memory_addr;
3492 	enum hl_goya_dma_direction user_dir;
3493 	u32 ctl;
3494 
3495 	ctl = le32_to_cpu(user_dma_pkt->ctl);
3496 	user_dir = (ctl & GOYA_PKT_LIN_DMA_CTL_DMA_DIR_MASK) >>
3497 			GOYA_PKT_LIN_DMA_CTL_DMA_DIR_SHIFT;
3498 
3499 	if (user_dir == HL_DMA_DRAM_TO_SRAM) {
3500 		dev_dbg(hdev->dev, "DMA direction is DRAM --> SRAM\n");
3501 		dram_memory_addr = le64_to_cpu(user_dma_pkt->src_addr);
3502 		sram_memory_addr = le64_to_cpu(user_dma_pkt->dst_addr);
3503 	} else {
3504 		dev_dbg(hdev->dev, "DMA direction is SRAM --> DRAM\n");
3505 		sram_memory_addr = le64_to_cpu(user_dma_pkt->src_addr);
3506 		dram_memory_addr = le64_to_cpu(user_dma_pkt->dst_addr);
3507 	}
3508 
3509 	if (!hl_mem_area_inside_range(sram_memory_addr,
3510 				le32_to_cpu(user_dma_pkt->tsize),
3511 				hdev->asic_prop.sram_user_base_address,
3512 				hdev->asic_prop.sram_end_address)) {
3513 		dev_err(hdev->dev, "SRAM address 0x%llx + 0x%x is invalid\n",
3514 			sram_memory_addr, user_dma_pkt->tsize);
3515 		return -EFAULT;
3516 	}
3517 
3518 	if (!hl_mem_area_inside_range(dram_memory_addr,
3519 				le32_to_cpu(user_dma_pkt->tsize),
3520 				hdev->asic_prop.dram_user_base_address,
3521 				hdev->asic_prop.dram_end_address)) {
3522 		dev_err(hdev->dev, "DRAM address 0x%llx + 0x%x is invalid\n",
3523 			dram_memory_addr, user_dma_pkt->tsize);
3524 		return -EFAULT;
3525 	}
3526 
3527 	parser->patched_cb_size += sizeof(*user_dma_pkt);
3528 
3529 	return 0;
3530 }
3531 
3532 static int goya_validate_dma_pkt_no_mmu(struct hl_device *hdev,
3533 				struct hl_cs_parser *parser,
3534 				struct packet_lin_dma *user_dma_pkt)
3535 {
3536 	enum hl_goya_dma_direction user_dir;
3537 	u32 ctl;
3538 	int rc;
3539 
3540 	dev_dbg(hdev->dev, "DMA packet details:\n");
3541 	dev_dbg(hdev->dev, "source == 0x%llx\n",
3542 		le64_to_cpu(user_dma_pkt->src_addr));
3543 	dev_dbg(hdev->dev, "destination == 0x%llx\n",
3544 		le64_to_cpu(user_dma_pkt->dst_addr));
3545 	dev_dbg(hdev->dev, "size == %u\n", le32_to_cpu(user_dma_pkt->tsize));
3546 
3547 	ctl = le32_to_cpu(user_dma_pkt->ctl);
3548 	user_dir = (ctl & GOYA_PKT_LIN_DMA_CTL_DMA_DIR_MASK) >>
3549 			GOYA_PKT_LIN_DMA_CTL_DMA_DIR_SHIFT;
3550 
3551 	/*
3552 	 * Special handling for DMA with size 0. The H/W has a bug where
3553 	 * this can cause the QMAN DMA to get stuck, so block it here.
3554 	 */
3555 	if (user_dma_pkt->tsize == 0) {
3556 		dev_err(hdev->dev,
3557 			"Got DMA with size 0, might reset the device\n");
3558 		return -EINVAL;
3559 	}
3560 
3561 	if ((user_dir == HL_DMA_DRAM_TO_SRAM) || (user_dir == HL_DMA_SRAM_TO_DRAM))
3562 		rc = goya_validate_dma_pkt_no_host(hdev, parser, user_dma_pkt);
3563 	else
3564 		rc = goya_validate_dma_pkt_host(hdev, parser, user_dma_pkt);
3565 
3566 	return rc;
3567 }
3568 
3569 static int goya_validate_dma_pkt_mmu(struct hl_device *hdev,
3570 				struct hl_cs_parser *parser,
3571 				struct packet_lin_dma *user_dma_pkt)
3572 {
3573 	dev_dbg(hdev->dev, "DMA packet details:\n");
3574 	dev_dbg(hdev->dev, "source == 0x%llx\n",
3575 		le64_to_cpu(user_dma_pkt->src_addr));
3576 	dev_dbg(hdev->dev, "destination == 0x%llx\n",
3577 		le64_to_cpu(user_dma_pkt->dst_addr));
3578 	dev_dbg(hdev->dev, "size == %u\n", le32_to_cpu(user_dma_pkt->tsize));
3579 
3580 	/*
3581 	 * WA for HW-23.
3582 	 * We can't allow user to read from Host using QMANs other than 1.
3583 	 * PMMU and HPMMU addresses are equal, check only one of them.
3584 	 */
3585 	if (parser->hw_queue_id != GOYA_QUEUE_ID_DMA_1 &&
3586 		hl_mem_area_inside_range(le64_to_cpu(user_dma_pkt->src_addr),
3587 				le32_to_cpu(user_dma_pkt->tsize),
3588 				hdev->asic_prop.pmmu.start_addr,
3589 				hdev->asic_prop.pmmu.end_addr)) {
3590 		dev_err(hdev->dev,
3591 			"Can't DMA from host on queue other then 1\n");
3592 		return -EFAULT;
3593 	}
3594 
3595 	if (user_dma_pkt->tsize == 0) {
3596 		dev_err(hdev->dev,
3597 			"Got DMA with size 0, might reset the device\n");
3598 		return -EINVAL;
3599 	}
3600 
3601 	parser->patched_cb_size += sizeof(*user_dma_pkt);
3602 
3603 	return 0;
3604 }
3605 
3606 static int goya_validate_wreg32(struct hl_device *hdev,
3607 				struct hl_cs_parser *parser,
3608 				struct packet_wreg32 *wreg_pkt)
3609 {
3610 	struct goya_device *goya = hdev->asic_specific;
3611 	u32 sob_start_addr, sob_end_addr;
3612 	u16 reg_offset;
3613 
3614 	reg_offset = le32_to_cpu(wreg_pkt->ctl) &
3615 			GOYA_PKT_WREG32_CTL_REG_OFFSET_MASK;
3616 
3617 	dev_dbg(hdev->dev, "WREG32 packet details:\n");
3618 	dev_dbg(hdev->dev, "reg_offset == 0x%x\n", reg_offset);
3619 	dev_dbg(hdev->dev, "value      == 0x%x\n",
3620 		le32_to_cpu(wreg_pkt->value));
3621 
3622 	if (reg_offset != (mmDMA_CH_0_WR_COMP_ADDR_LO & 0x1FFF)) {
3623 		dev_err(hdev->dev, "WREG32 packet with illegal address 0x%x\n",
3624 			reg_offset);
3625 		return -EPERM;
3626 	}
3627 
3628 	/*
3629 	 * With MMU, DMA channels are not secured, so it doesn't matter where
3630 	 * the WR COMP will be written to because it will go out with
3631 	 * non-secured property
3632 	 */
3633 	if (goya->hw_cap_initialized & HW_CAP_MMU)
3634 		return 0;
3635 
3636 	sob_start_addr = lower_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_0);
3637 	sob_end_addr = lower_32_bits(CFG_BASE + mmSYNC_MNGR_SOB_OBJ_1023);
3638 
3639 	if ((le32_to_cpu(wreg_pkt->value) < sob_start_addr) ||
3640 			(le32_to_cpu(wreg_pkt->value) > sob_end_addr)) {
3641 
3642 		dev_err(hdev->dev, "WREG32 packet with illegal value 0x%x\n",
3643 			wreg_pkt->value);
3644 		return -EPERM;
3645 	}
3646 
3647 	return 0;
3648 }
3649 
3650 static int goya_validate_cb(struct hl_device *hdev,
3651 			struct hl_cs_parser *parser, bool is_mmu)
3652 {
3653 	u32 cb_parsed_length = 0;
3654 	int rc = 0;
3655 
3656 	parser->patched_cb_size = 0;
3657 
3658 	/* cb_user_size is more than 0 so loop will always be executed */
3659 	while (cb_parsed_length < parser->user_cb_size) {
3660 		enum packet_id pkt_id;
3661 		u16 pkt_size;
3662 		struct goya_packet *user_pkt;
3663 
3664 		user_pkt = parser->user_cb->kernel_address + cb_parsed_length;
3665 
3666 		pkt_id = (enum packet_id) (
3667 				(le64_to_cpu(user_pkt->header) &
3668 				PACKET_HEADER_PACKET_ID_MASK) >>
3669 					PACKET_HEADER_PACKET_ID_SHIFT);
3670 
3671 		if (!validate_packet_id(pkt_id)) {
3672 			dev_err(hdev->dev, "Invalid packet id %u\n", pkt_id);
3673 			rc = -EINVAL;
3674 			break;
3675 		}
3676 
3677 		pkt_size = goya_packet_sizes[pkt_id];
3678 		cb_parsed_length += pkt_size;
3679 		if (cb_parsed_length > parser->user_cb_size) {
3680 			dev_err(hdev->dev,
3681 				"packet 0x%x is out of CB boundary\n", pkt_id);
3682 			rc = -EINVAL;
3683 			break;
3684 		}
3685 
3686 		switch (pkt_id) {
3687 		case PACKET_WREG_32:
3688 			/*
3689 			 * Although it is validated after copy in patch_cb(),
3690 			 * need to validate here as well because patch_cb() is
3691 			 * not called in MMU path while this function is called
3692 			 */
3693 			rc = goya_validate_wreg32(hdev,
3694 				parser, (struct packet_wreg32 *) user_pkt);
3695 			parser->patched_cb_size += pkt_size;
3696 			break;
3697 
3698 		case PACKET_WREG_BULK:
3699 			dev_err(hdev->dev,
3700 				"User not allowed to use WREG_BULK\n");
3701 			rc = -EPERM;
3702 			break;
3703 
3704 		case PACKET_MSG_PROT:
3705 			dev_err(hdev->dev,
3706 				"User not allowed to use MSG_PROT\n");
3707 			rc = -EPERM;
3708 			break;
3709 
3710 		case PACKET_CP_DMA:
3711 			dev_err(hdev->dev, "User not allowed to use CP_DMA\n");
3712 			rc = -EPERM;
3713 			break;
3714 
3715 		case PACKET_STOP:
3716 			dev_err(hdev->dev, "User not allowed to use STOP\n");
3717 			rc = -EPERM;
3718 			break;
3719 
3720 		case PACKET_LIN_DMA:
3721 			if (is_mmu)
3722 				rc = goya_validate_dma_pkt_mmu(hdev, parser,
3723 					(struct packet_lin_dma *) user_pkt);
3724 			else
3725 				rc = goya_validate_dma_pkt_no_mmu(hdev, parser,
3726 					(struct packet_lin_dma *) user_pkt);
3727 			break;
3728 
3729 		case PACKET_MSG_LONG:
3730 		case PACKET_MSG_SHORT:
3731 		case PACKET_FENCE:
3732 		case PACKET_NOP:
3733 			parser->patched_cb_size += pkt_size;
3734 			break;
3735 
3736 		default:
3737 			dev_err(hdev->dev, "Invalid packet header 0x%x\n",
3738 				pkt_id);
3739 			rc = -EINVAL;
3740 			break;
3741 		}
3742 
3743 		if (rc)
3744 			break;
3745 	}
3746 
3747 	/*
3748 	 * The new CB should have space at the end for two MSG_PROT packets:
3749 	 * 1. A packet that will act as a completion packet
3750 	 * 2. A packet that will generate MSI-X interrupt
3751 	 */
3752 	parser->patched_cb_size += sizeof(struct packet_msg_prot) * 2;
3753 
3754 	return rc;
3755 }
3756 
3757 static int goya_patch_dma_packet(struct hl_device *hdev,
3758 				struct hl_cs_parser *parser,
3759 				struct packet_lin_dma *user_dma_pkt,
3760 				struct packet_lin_dma *new_dma_pkt,
3761 				u32 *new_dma_pkt_size)
3762 {
3763 	struct hl_userptr *userptr;
3764 	struct scatterlist *sg, *sg_next_iter;
3765 	u32 count, dma_desc_cnt;
3766 	u64 len, len_next;
3767 	dma_addr_t dma_addr, dma_addr_next;
3768 	enum hl_goya_dma_direction user_dir;
3769 	u64 device_memory_addr, addr;
3770 	enum dma_data_direction dir;
3771 	struct sg_table *sgt;
3772 	bool skip_host_mem_pin = false;
3773 	bool user_memset;
3774 	u32 user_rdcomp_mask, user_wrcomp_mask, ctl;
3775 
3776 	ctl = le32_to_cpu(user_dma_pkt->ctl);
3777 
3778 	user_dir = (ctl & GOYA_PKT_LIN_DMA_CTL_DMA_DIR_MASK) >>
3779 			GOYA_PKT_LIN_DMA_CTL_DMA_DIR_SHIFT;
3780 
3781 	user_memset = (ctl & GOYA_PKT_LIN_DMA_CTL_MEMSET_MASK) >>
3782 			GOYA_PKT_LIN_DMA_CTL_MEMSET_SHIFT;
3783 
3784 	if ((user_dir == HL_DMA_DRAM_TO_SRAM) || (user_dir == HL_DMA_SRAM_TO_DRAM) ||
3785 			(user_dma_pkt->tsize == 0)) {
3786 		memcpy(new_dma_pkt, user_dma_pkt, sizeof(*new_dma_pkt));
3787 		*new_dma_pkt_size = sizeof(*new_dma_pkt);
3788 		return 0;
3789 	}
3790 
3791 	if ((user_dir == HL_DMA_HOST_TO_DRAM) || (user_dir == HL_DMA_HOST_TO_SRAM)) {
3792 		addr = le64_to_cpu(user_dma_pkt->src_addr);
3793 		device_memory_addr = le64_to_cpu(user_dma_pkt->dst_addr);
3794 		dir = DMA_TO_DEVICE;
3795 		if (user_memset)
3796 			skip_host_mem_pin = true;
3797 	} else {
3798 		addr = le64_to_cpu(user_dma_pkt->dst_addr);
3799 		device_memory_addr = le64_to_cpu(user_dma_pkt->src_addr);
3800 		dir = DMA_FROM_DEVICE;
3801 	}
3802 
3803 	if ((!skip_host_mem_pin) &&
3804 		(hl_userptr_is_pinned(hdev, addr,
3805 			le32_to_cpu(user_dma_pkt->tsize),
3806 			parser->job_userptr_list, &userptr) == false)) {
3807 		dev_err(hdev->dev, "Userptr 0x%llx + 0x%x NOT mapped\n",
3808 				addr, user_dma_pkt->tsize);
3809 		return -EFAULT;
3810 	}
3811 
3812 	if ((user_memset) && (dir == DMA_TO_DEVICE)) {
3813 		memcpy(new_dma_pkt, user_dma_pkt, sizeof(*user_dma_pkt));
3814 		*new_dma_pkt_size = sizeof(*user_dma_pkt);
3815 		return 0;
3816 	}
3817 
3818 	user_rdcomp_mask = ctl & GOYA_PKT_LIN_DMA_CTL_RDCOMP_MASK;
3819 
3820 	user_wrcomp_mask = ctl & GOYA_PKT_LIN_DMA_CTL_WRCOMP_MASK;
3821 
3822 	sgt = userptr->sgt;
3823 	dma_desc_cnt = 0;
3824 
3825 	for_each_sgtable_dma_sg(sgt, sg, count) {
3826 		len = sg_dma_len(sg);
3827 		dma_addr = sg_dma_address(sg);
3828 
3829 		if (len == 0)
3830 			break;
3831 
3832 		while ((count + 1) < sgt->nents) {
3833 			sg_next_iter = sg_next(sg);
3834 			len_next = sg_dma_len(sg_next_iter);
3835 			dma_addr_next = sg_dma_address(sg_next_iter);
3836 
3837 			if (len_next == 0)
3838 				break;
3839 
3840 			if ((dma_addr + len == dma_addr_next) &&
3841 				(len + len_next <= DMA_MAX_TRANSFER_SIZE)) {
3842 				len += len_next;
3843 				count++;
3844 				sg = sg_next_iter;
3845 			} else {
3846 				break;
3847 			}
3848 		}
3849 
3850 		ctl = le32_to_cpu(user_dma_pkt->ctl);
3851 		if (likely(dma_desc_cnt))
3852 			ctl &= ~GOYA_PKT_CTL_EB_MASK;
3853 		ctl &= ~(GOYA_PKT_LIN_DMA_CTL_RDCOMP_MASK |
3854 				GOYA_PKT_LIN_DMA_CTL_WRCOMP_MASK);
3855 		new_dma_pkt->ctl = cpu_to_le32(ctl);
3856 		new_dma_pkt->tsize = cpu_to_le32((u32) len);
3857 
3858 		if (dir == DMA_TO_DEVICE) {
3859 			new_dma_pkt->src_addr = cpu_to_le64(dma_addr);
3860 			new_dma_pkt->dst_addr = cpu_to_le64(device_memory_addr);
3861 		} else {
3862 			new_dma_pkt->src_addr = cpu_to_le64(device_memory_addr);
3863 			new_dma_pkt->dst_addr = cpu_to_le64(dma_addr);
3864 		}
3865 
3866 		if (!user_memset)
3867 			device_memory_addr += len;
3868 		dma_desc_cnt++;
3869 		new_dma_pkt++;
3870 	}
3871 
3872 	if (!dma_desc_cnt) {
3873 		dev_err(hdev->dev,
3874 			"Error of 0 SG entries when patching DMA packet\n");
3875 		return -EFAULT;
3876 	}
3877 
3878 	/* Fix the last dma packet - rdcomp/wrcomp must be as user set them */
3879 	new_dma_pkt--;
3880 	new_dma_pkt->ctl |= cpu_to_le32(user_rdcomp_mask | user_wrcomp_mask);
3881 
3882 	*new_dma_pkt_size = dma_desc_cnt * sizeof(struct packet_lin_dma);
3883 
3884 	return 0;
3885 }
3886 
3887 static int goya_patch_cb(struct hl_device *hdev,
3888 				struct hl_cs_parser *parser)
3889 {
3890 	u32 cb_parsed_length = 0;
3891 	u32 cb_patched_cur_length = 0;
3892 	int rc = 0;
3893 
3894 	/* cb_user_size is more than 0 so loop will always be executed */
3895 	while (cb_parsed_length < parser->user_cb_size) {
3896 		enum packet_id pkt_id;
3897 		u16 pkt_size;
3898 		u32 new_pkt_size = 0;
3899 		struct goya_packet *user_pkt, *kernel_pkt;
3900 
3901 		user_pkt = parser->user_cb->kernel_address + cb_parsed_length;
3902 		kernel_pkt = parser->patched_cb->kernel_address +
3903 					cb_patched_cur_length;
3904 
3905 		pkt_id = (enum packet_id) (
3906 				(le64_to_cpu(user_pkt->header) &
3907 				PACKET_HEADER_PACKET_ID_MASK) >>
3908 					PACKET_HEADER_PACKET_ID_SHIFT);
3909 
3910 		if (!validate_packet_id(pkt_id)) {
3911 			dev_err(hdev->dev, "Invalid packet id %u\n", pkt_id);
3912 			rc = -EINVAL;
3913 			break;
3914 		}
3915 
3916 		pkt_size = goya_packet_sizes[pkt_id];
3917 		cb_parsed_length += pkt_size;
3918 		if (cb_parsed_length > parser->user_cb_size) {
3919 			dev_err(hdev->dev,
3920 				"packet 0x%x is out of CB boundary\n", pkt_id);
3921 			rc = -EINVAL;
3922 			break;
3923 		}
3924 
3925 		switch (pkt_id) {
3926 		case PACKET_LIN_DMA:
3927 			rc = goya_patch_dma_packet(hdev, parser,
3928 					(struct packet_lin_dma *) user_pkt,
3929 					(struct packet_lin_dma *) kernel_pkt,
3930 					&new_pkt_size);
3931 			cb_patched_cur_length += new_pkt_size;
3932 			break;
3933 
3934 		case PACKET_WREG_32:
3935 			memcpy(kernel_pkt, user_pkt, pkt_size);
3936 			cb_patched_cur_length += pkt_size;
3937 			rc = goya_validate_wreg32(hdev, parser,
3938 					(struct packet_wreg32 *) kernel_pkt);
3939 			break;
3940 
3941 		case PACKET_WREG_BULK:
3942 			dev_err(hdev->dev,
3943 				"User not allowed to use WREG_BULK\n");
3944 			rc = -EPERM;
3945 			break;
3946 
3947 		case PACKET_MSG_PROT:
3948 			dev_err(hdev->dev,
3949 				"User not allowed to use MSG_PROT\n");
3950 			rc = -EPERM;
3951 			break;
3952 
3953 		case PACKET_CP_DMA:
3954 			dev_err(hdev->dev, "User not allowed to use CP_DMA\n");
3955 			rc = -EPERM;
3956 			break;
3957 
3958 		case PACKET_STOP:
3959 			dev_err(hdev->dev, "User not allowed to use STOP\n");
3960 			rc = -EPERM;
3961 			break;
3962 
3963 		case PACKET_MSG_LONG:
3964 		case PACKET_MSG_SHORT:
3965 		case PACKET_FENCE:
3966 		case PACKET_NOP:
3967 			memcpy(kernel_pkt, user_pkt, pkt_size);
3968 			cb_patched_cur_length += pkt_size;
3969 			break;
3970 
3971 		default:
3972 			dev_err(hdev->dev, "Invalid packet header 0x%x\n",
3973 				pkt_id);
3974 			rc = -EINVAL;
3975 			break;
3976 		}
3977 
3978 		if (rc)
3979 			break;
3980 	}
3981 
3982 	return rc;
3983 }
3984 
3985 static int goya_parse_cb_mmu(struct hl_device *hdev,
3986 		struct hl_cs_parser *parser)
3987 {
3988 	u64 handle;
3989 	u32 patched_cb_size;
3990 	struct hl_cb *user_cb;
3991 	int rc;
3992 
3993 	/*
3994 	 * The new CB should have space at the end for two MSG_PROT pkt:
3995 	 * 1. A packet that will act as a completion packet
3996 	 * 2. A packet that will generate MSI-X interrupt
3997 	 */
3998 	parser->patched_cb_size = parser->user_cb_size +
3999 			sizeof(struct packet_msg_prot) * 2;
4000 
4001 	rc = hl_cb_create(hdev, &hdev->kernel_mem_mgr, hdev->kernel_ctx,
4002 				parser->patched_cb_size, false, false,
4003 				&handle);
4004 
4005 	if (rc) {
4006 		dev_err(hdev->dev,
4007 			"Failed to allocate patched CB for DMA CS %d\n",
4008 			rc);
4009 		return rc;
4010 	}
4011 
4012 	parser->patched_cb = hl_cb_get(&hdev->kernel_mem_mgr, handle);
4013 	/* hl_cb_get should never fail here */
4014 	if (!parser->patched_cb) {
4015 		dev_crit(hdev->dev, "DMA CB handle invalid 0x%llx\n", handle);
4016 		rc = -EFAULT;
4017 		goto out;
4018 	}
4019 
4020 	/*
4021 	 * The check that parser->user_cb_size <= parser->user_cb->size was done
4022 	 * in validate_queue_index().
4023 	 */
4024 	memcpy(parser->patched_cb->kernel_address,
4025 		parser->user_cb->kernel_address,
4026 		parser->user_cb_size);
4027 
4028 	patched_cb_size = parser->patched_cb_size;
4029 
4030 	/* validate patched CB instead of user CB */
4031 	user_cb = parser->user_cb;
4032 	parser->user_cb = parser->patched_cb;
4033 	rc = goya_validate_cb(hdev, parser, true);
4034 	parser->user_cb = user_cb;
4035 
4036 	if (rc) {
4037 		hl_cb_put(parser->patched_cb);
4038 		goto out;
4039 	}
4040 
4041 	if (patched_cb_size != parser->patched_cb_size) {
4042 		dev_err(hdev->dev, "user CB size mismatch\n");
4043 		hl_cb_put(parser->patched_cb);
4044 		rc = -EINVAL;
4045 		goto out;
4046 	}
4047 
4048 out:
4049 	/*
4050 	 * Always call cb destroy here because we still have 1 reference
4051 	 * to it by calling cb_get earlier. After the job will be completed,
4052 	 * cb_put will release it, but here we want to remove it from the
4053 	 * idr
4054 	 */
4055 	hl_cb_destroy(&hdev->kernel_mem_mgr, handle);
4056 
4057 	return rc;
4058 }
4059 
4060 static int goya_parse_cb_no_mmu(struct hl_device *hdev,
4061 				struct hl_cs_parser *parser)
4062 {
4063 	u64 handle;
4064 	int rc;
4065 
4066 	rc = goya_validate_cb(hdev, parser, false);
4067 
4068 	if (rc)
4069 		goto free_userptr;
4070 
4071 	rc = hl_cb_create(hdev, &hdev->kernel_mem_mgr, hdev->kernel_ctx,
4072 				parser->patched_cb_size, false, false,
4073 				&handle);
4074 	if (rc) {
4075 		dev_err(hdev->dev,
4076 			"Failed to allocate patched CB for DMA CS %d\n", rc);
4077 		goto free_userptr;
4078 	}
4079 
4080 	parser->patched_cb = hl_cb_get(&hdev->kernel_mem_mgr, handle);
4081 	/* hl_cb_get should never fail here */
4082 	if (!parser->patched_cb) {
4083 		dev_crit(hdev->dev, "DMA CB handle invalid 0x%llx\n", handle);
4084 		rc = -EFAULT;
4085 		goto out;
4086 	}
4087 
4088 	rc = goya_patch_cb(hdev, parser);
4089 
4090 	if (rc)
4091 		hl_cb_put(parser->patched_cb);
4092 
4093 out:
4094 	/*
4095 	 * Always call cb destroy here because we still have 1 reference
4096 	 * to it by calling cb_get earlier. After the job will be completed,
4097 	 * cb_put will release it, but here we want to remove it from the
4098 	 * idr
4099 	 */
4100 	hl_cb_destroy(&hdev->kernel_mem_mgr, handle);
4101 
4102 free_userptr:
4103 	if (rc)
4104 		hl_userptr_delete_list(hdev, parser->job_userptr_list);
4105 	return rc;
4106 }
4107 
4108 static int goya_parse_cb_no_ext_queue(struct hl_device *hdev,
4109 					struct hl_cs_parser *parser)
4110 {
4111 	struct asic_fixed_properties *asic_prop = &hdev->asic_prop;
4112 	struct goya_device *goya = hdev->asic_specific;
4113 
4114 	if (goya->hw_cap_initialized & HW_CAP_MMU)
4115 		return 0;
4116 
4117 	/* For internal queue jobs, just check if CB address is valid */
4118 	if (hl_mem_area_inside_range(
4119 			(u64) (uintptr_t) parser->user_cb,
4120 			parser->user_cb_size,
4121 			asic_prop->sram_user_base_address,
4122 			asic_prop->sram_end_address))
4123 		return 0;
4124 
4125 	if (hl_mem_area_inside_range(
4126 			(u64) (uintptr_t) parser->user_cb,
4127 			parser->user_cb_size,
4128 			asic_prop->dram_user_base_address,
4129 			asic_prop->dram_end_address))
4130 		return 0;
4131 
4132 	dev_err(hdev->dev,
4133 		"Internal CB address 0x%px + 0x%x is not in SRAM nor in DRAM\n",
4134 		parser->user_cb, parser->user_cb_size);
4135 
4136 	return -EFAULT;
4137 }
4138 
4139 int goya_cs_parser(struct hl_device *hdev, struct hl_cs_parser *parser)
4140 {
4141 	struct goya_device *goya = hdev->asic_specific;
4142 
4143 	if (parser->queue_type == QUEUE_TYPE_INT)
4144 		return goya_parse_cb_no_ext_queue(hdev, parser);
4145 
4146 	if (goya->hw_cap_initialized & HW_CAP_MMU)
4147 		return goya_parse_cb_mmu(hdev, parser);
4148 	else
4149 		return goya_parse_cb_no_mmu(hdev, parser);
4150 }
4151 
4152 void goya_add_end_of_cb_packets(struct hl_device *hdev, void *kernel_address,
4153 				u32 len, u32 original_len, u64 cq_addr, u32 cq_val,
4154 				u32 msix_vec, bool eb)
4155 {
4156 	struct packet_msg_prot *cq_pkt;
4157 	u32 tmp;
4158 
4159 	cq_pkt = kernel_address + len - (sizeof(struct packet_msg_prot) * 2);
4160 
4161 	tmp = (PACKET_MSG_PROT << GOYA_PKT_CTL_OPCODE_SHIFT) |
4162 			(1 << GOYA_PKT_CTL_EB_SHIFT) |
4163 			(1 << GOYA_PKT_CTL_MB_SHIFT);
4164 	cq_pkt->ctl = cpu_to_le32(tmp);
4165 	cq_pkt->value = cpu_to_le32(cq_val);
4166 	cq_pkt->addr = cpu_to_le64(cq_addr);
4167 
4168 	cq_pkt++;
4169 
4170 	tmp = (PACKET_MSG_PROT << GOYA_PKT_CTL_OPCODE_SHIFT) |
4171 			(1 << GOYA_PKT_CTL_MB_SHIFT);
4172 	cq_pkt->ctl = cpu_to_le32(tmp);
4173 	cq_pkt->value = cpu_to_le32(msix_vec & 0x7FF);
4174 	cq_pkt->addr = cpu_to_le64(CFG_BASE + mmPCIE_DBI_MSIX_DOORBELL_OFF);
4175 }
4176 
4177 void goya_update_eq_ci(struct hl_device *hdev, u32 val)
4178 {
4179 	WREG32(mmCPU_EQ_CI, val);
4180 }
4181 
4182 void goya_restore_phase_topology(struct hl_device *hdev)
4183 {
4184 
4185 }
4186 
4187 static void goya_clear_sm_regs(struct hl_device *hdev)
4188 {
4189 	int i, num_of_sob_in_longs, num_of_mon_in_longs;
4190 
4191 	num_of_sob_in_longs =
4192 		((mmSYNC_MNGR_SOB_OBJ_1023 - mmSYNC_MNGR_SOB_OBJ_0) + 4);
4193 
4194 	num_of_mon_in_longs =
4195 		((mmSYNC_MNGR_MON_STATUS_255 - mmSYNC_MNGR_MON_STATUS_0) + 4);
4196 
4197 	for (i = 0 ; i < num_of_sob_in_longs ; i += 4)
4198 		WREG32(mmSYNC_MNGR_SOB_OBJ_0 + i, 0);
4199 
4200 	for (i = 0 ; i < num_of_mon_in_longs ; i += 4)
4201 		WREG32(mmSYNC_MNGR_MON_STATUS_0 + i, 0);
4202 
4203 	/* Flush all WREG to prevent race */
4204 	i = RREG32(mmSYNC_MNGR_SOB_OBJ_0);
4205 }
4206 
4207 static int goya_debugfs_read_dma(struct hl_device *hdev, u64 addr, u32 size, void *blob_addr)
4208 {
4209 	dev_err(hdev->dev, "Reading via DMA is unimplemented yet\n");
4210 	return -EPERM;
4211 }
4212 
4213 static u64 goya_read_pte(struct hl_device *hdev, u64 addr)
4214 {
4215 	struct goya_device *goya = hdev->asic_specific;
4216 
4217 	if (hdev->reset_info.hard_reset_pending)
4218 		return U64_MAX;
4219 
4220 	return readq(hdev->pcie_bar[DDR_BAR_ID] +
4221 			(addr - goya->ddr_bar_cur_addr));
4222 }
4223 
4224 static void goya_write_pte(struct hl_device *hdev, u64 addr, u64 val)
4225 {
4226 	struct goya_device *goya = hdev->asic_specific;
4227 
4228 	if (hdev->reset_info.hard_reset_pending)
4229 		return;
4230 
4231 	writeq(val, hdev->pcie_bar[DDR_BAR_ID] +
4232 			(addr - goya->ddr_bar_cur_addr));
4233 }
4234 
4235 static const char *_goya_get_event_desc(u16 event_type)
4236 {
4237 	switch (event_type) {
4238 	case GOYA_ASYNC_EVENT_ID_PCIE_IF:
4239 		return "PCIe_if";
4240 	case GOYA_ASYNC_EVENT_ID_TPC0_ECC:
4241 	case GOYA_ASYNC_EVENT_ID_TPC1_ECC:
4242 	case GOYA_ASYNC_EVENT_ID_TPC2_ECC:
4243 	case GOYA_ASYNC_EVENT_ID_TPC3_ECC:
4244 	case GOYA_ASYNC_EVENT_ID_TPC4_ECC:
4245 	case GOYA_ASYNC_EVENT_ID_TPC5_ECC:
4246 	case GOYA_ASYNC_EVENT_ID_TPC6_ECC:
4247 	case GOYA_ASYNC_EVENT_ID_TPC7_ECC:
4248 		return "TPC%d_ecc";
4249 	case GOYA_ASYNC_EVENT_ID_MME_ECC:
4250 		return "MME_ecc";
4251 	case GOYA_ASYNC_EVENT_ID_MME_ECC_EXT:
4252 		return "MME_ecc_ext";
4253 	case GOYA_ASYNC_EVENT_ID_MMU_ECC:
4254 		return "MMU_ecc";
4255 	case GOYA_ASYNC_EVENT_ID_DMA_MACRO:
4256 		return "DMA_macro";
4257 	case GOYA_ASYNC_EVENT_ID_DMA_ECC:
4258 		return "DMA_ecc";
4259 	case GOYA_ASYNC_EVENT_ID_CPU_IF_ECC:
4260 		return "CPU_if_ecc";
4261 	case GOYA_ASYNC_EVENT_ID_PSOC_MEM:
4262 		return "PSOC_mem";
4263 	case GOYA_ASYNC_EVENT_ID_PSOC_CORESIGHT:
4264 		return "PSOC_coresight";
4265 	case GOYA_ASYNC_EVENT_ID_SRAM0 ... GOYA_ASYNC_EVENT_ID_SRAM29:
4266 		return "SRAM%d";
4267 	case GOYA_ASYNC_EVENT_ID_GIC500:
4268 		return "GIC500";
4269 	case GOYA_ASYNC_EVENT_ID_PLL0 ... GOYA_ASYNC_EVENT_ID_PLL6:
4270 		return "PLL%d";
4271 	case GOYA_ASYNC_EVENT_ID_AXI_ECC:
4272 		return "AXI_ecc";
4273 	case GOYA_ASYNC_EVENT_ID_L2_RAM_ECC:
4274 		return "L2_ram_ecc";
4275 	case GOYA_ASYNC_EVENT_ID_PSOC_GPIO_05_SW_RESET:
4276 		return "PSOC_gpio_05_sw_reset";
4277 	case GOYA_ASYNC_EVENT_ID_PSOC_GPIO_10_VRHOT_ICRIT:
4278 		return "PSOC_gpio_10_vrhot_icrit";
4279 	case GOYA_ASYNC_EVENT_ID_PCIE_DEC:
4280 		return "PCIe_dec";
4281 	case GOYA_ASYNC_EVENT_ID_TPC0_DEC:
4282 	case GOYA_ASYNC_EVENT_ID_TPC1_DEC:
4283 	case GOYA_ASYNC_EVENT_ID_TPC2_DEC:
4284 	case GOYA_ASYNC_EVENT_ID_TPC3_DEC:
4285 	case GOYA_ASYNC_EVENT_ID_TPC4_DEC:
4286 	case GOYA_ASYNC_EVENT_ID_TPC5_DEC:
4287 	case GOYA_ASYNC_EVENT_ID_TPC6_DEC:
4288 	case GOYA_ASYNC_EVENT_ID_TPC7_DEC:
4289 		return "TPC%d_dec";
4290 	case GOYA_ASYNC_EVENT_ID_MME_WACS:
4291 		return "MME_wacs";
4292 	case GOYA_ASYNC_EVENT_ID_MME_WACSD:
4293 		return "MME_wacsd";
4294 	case GOYA_ASYNC_EVENT_ID_CPU_AXI_SPLITTER:
4295 		return "CPU_axi_splitter";
4296 	case GOYA_ASYNC_EVENT_ID_PSOC_AXI_DEC:
4297 		return "PSOC_axi_dec";
4298 	case GOYA_ASYNC_EVENT_ID_PSOC:
4299 		return "PSOC";
4300 	case GOYA_ASYNC_EVENT_ID_TPC0_KRN_ERR:
4301 	case GOYA_ASYNC_EVENT_ID_TPC1_KRN_ERR:
4302 	case GOYA_ASYNC_EVENT_ID_TPC2_KRN_ERR:
4303 	case GOYA_ASYNC_EVENT_ID_TPC3_KRN_ERR:
4304 	case GOYA_ASYNC_EVENT_ID_TPC4_KRN_ERR:
4305 	case GOYA_ASYNC_EVENT_ID_TPC5_KRN_ERR:
4306 	case GOYA_ASYNC_EVENT_ID_TPC6_KRN_ERR:
4307 	case GOYA_ASYNC_EVENT_ID_TPC7_KRN_ERR:
4308 		return "TPC%d_krn_err";
4309 	case GOYA_ASYNC_EVENT_ID_TPC0_CMDQ ... GOYA_ASYNC_EVENT_ID_TPC7_CMDQ:
4310 		return "TPC%d_cq";
4311 	case GOYA_ASYNC_EVENT_ID_TPC0_QM ... GOYA_ASYNC_EVENT_ID_TPC7_QM:
4312 		return "TPC%d_qm";
4313 	case GOYA_ASYNC_EVENT_ID_MME_QM:
4314 		return "MME_qm";
4315 	case GOYA_ASYNC_EVENT_ID_MME_CMDQ:
4316 		return "MME_cq";
4317 	case GOYA_ASYNC_EVENT_ID_DMA0_QM ... GOYA_ASYNC_EVENT_ID_DMA4_QM:
4318 		return "DMA%d_qm";
4319 	case GOYA_ASYNC_EVENT_ID_DMA0_CH ... GOYA_ASYNC_EVENT_ID_DMA4_CH:
4320 		return "DMA%d_ch";
4321 	case GOYA_ASYNC_EVENT_ID_TPC0_BMON_SPMU:
4322 	case GOYA_ASYNC_EVENT_ID_TPC1_BMON_SPMU:
4323 	case GOYA_ASYNC_EVENT_ID_TPC2_BMON_SPMU:
4324 	case GOYA_ASYNC_EVENT_ID_TPC3_BMON_SPMU:
4325 	case GOYA_ASYNC_EVENT_ID_TPC4_BMON_SPMU:
4326 	case GOYA_ASYNC_EVENT_ID_TPC5_BMON_SPMU:
4327 	case GOYA_ASYNC_EVENT_ID_TPC6_BMON_SPMU:
4328 	case GOYA_ASYNC_EVENT_ID_TPC7_BMON_SPMU:
4329 		return "TPC%d_bmon_spmu";
4330 	case GOYA_ASYNC_EVENT_ID_DMA_BM_CH0 ... GOYA_ASYNC_EVENT_ID_DMA_BM_CH4:
4331 		return "DMA_bm_ch%d";
4332 	case GOYA_ASYNC_EVENT_ID_FIX_POWER_ENV_S:
4333 		return "POWER_ENV_S";
4334 	case GOYA_ASYNC_EVENT_ID_FIX_POWER_ENV_E:
4335 		return "POWER_ENV_E";
4336 	case GOYA_ASYNC_EVENT_ID_FIX_THERMAL_ENV_S:
4337 		return "THERMAL_ENV_S";
4338 	case GOYA_ASYNC_EVENT_ID_FIX_THERMAL_ENV_E:
4339 		return "THERMAL_ENV_E";
4340 	case GOYA_ASYNC_EVENT_PKT_QUEUE_OUT_SYNC:
4341 		return "QUEUE_OUT_OF_SYNC";
4342 	default:
4343 		return "N/A";
4344 	}
4345 }
4346 
4347 static void goya_get_event_desc(u16 event_type, char *desc, size_t size)
4348 {
4349 	u8 index;
4350 
4351 	switch (event_type) {
4352 	case GOYA_ASYNC_EVENT_ID_TPC0_ECC:
4353 	case GOYA_ASYNC_EVENT_ID_TPC1_ECC:
4354 	case GOYA_ASYNC_EVENT_ID_TPC2_ECC:
4355 	case GOYA_ASYNC_EVENT_ID_TPC3_ECC:
4356 	case GOYA_ASYNC_EVENT_ID_TPC4_ECC:
4357 	case GOYA_ASYNC_EVENT_ID_TPC5_ECC:
4358 	case GOYA_ASYNC_EVENT_ID_TPC6_ECC:
4359 	case GOYA_ASYNC_EVENT_ID_TPC7_ECC:
4360 		index = (event_type - GOYA_ASYNC_EVENT_ID_TPC0_ECC) / 3;
4361 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4362 		break;
4363 	case GOYA_ASYNC_EVENT_ID_SRAM0 ... GOYA_ASYNC_EVENT_ID_SRAM29:
4364 		index = event_type - GOYA_ASYNC_EVENT_ID_SRAM0;
4365 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4366 		break;
4367 	case GOYA_ASYNC_EVENT_ID_PLL0 ... GOYA_ASYNC_EVENT_ID_PLL6:
4368 		index = event_type - GOYA_ASYNC_EVENT_ID_PLL0;
4369 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4370 		break;
4371 	case GOYA_ASYNC_EVENT_ID_TPC0_DEC:
4372 	case GOYA_ASYNC_EVENT_ID_TPC1_DEC:
4373 	case GOYA_ASYNC_EVENT_ID_TPC2_DEC:
4374 	case GOYA_ASYNC_EVENT_ID_TPC3_DEC:
4375 	case GOYA_ASYNC_EVENT_ID_TPC4_DEC:
4376 	case GOYA_ASYNC_EVENT_ID_TPC5_DEC:
4377 	case GOYA_ASYNC_EVENT_ID_TPC6_DEC:
4378 	case GOYA_ASYNC_EVENT_ID_TPC7_DEC:
4379 		index = (event_type - GOYA_ASYNC_EVENT_ID_TPC0_DEC) / 3;
4380 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4381 		break;
4382 	case GOYA_ASYNC_EVENT_ID_TPC0_KRN_ERR:
4383 	case GOYA_ASYNC_EVENT_ID_TPC1_KRN_ERR:
4384 	case GOYA_ASYNC_EVENT_ID_TPC2_KRN_ERR:
4385 	case GOYA_ASYNC_EVENT_ID_TPC3_KRN_ERR:
4386 	case GOYA_ASYNC_EVENT_ID_TPC4_KRN_ERR:
4387 	case GOYA_ASYNC_EVENT_ID_TPC5_KRN_ERR:
4388 	case GOYA_ASYNC_EVENT_ID_TPC6_KRN_ERR:
4389 	case GOYA_ASYNC_EVENT_ID_TPC7_KRN_ERR:
4390 		index = (event_type - GOYA_ASYNC_EVENT_ID_TPC0_KRN_ERR) / 10;
4391 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4392 		break;
4393 	case GOYA_ASYNC_EVENT_ID_TPC0_CMDQ ... GOYA_ASYNC_EVENT_ID_TPC7_CMDQ:
4394 		index = event_type - GOYA_ASYNC_EVENT_ID_TPC0_CMDQ;
4395 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4396 		break;
4397 	case GOYA_ASYNC_EVENT_ID_TPC0_QM ... GOYA_ASYNC_EVENT_ID_TPC7_QM:
4398 		index = event_type - GOYA_ASYNC_EVENT_ID_TPC0_QM;
4399 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4400 		break;
4401 	case GOYA_ASYNC_EVENT_ID_DMA0_QM ... GOYA_ASYNC_EVENT_ID_DMA4_QM:
4402 		index = event_type - GOYA_ASYNC_EVENT_ID_DMA0_QM;
4403 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4404 		break;
4405 	case GOYA_ASYNC_EVENT_ID_DMA0_CH ... GOYA_ASYNC_EVENT_ID_DMA4_CH:
4406 		index = event_type - GOYA_ASYNC_EVENT_ID_DMA0_CH;
4407 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4408 		break;
4409 	case GOYA_ASYNC_EVENT_ID_TPC0_BMON_SPMU:
4410 	case GOYA_ASYNC_EVENT_ID_TPC1_BMON_SPMU:
4411 	case GOYA_ASYNC_EVENT_ID_TPC2_BMON_SPMU:
4412 	case GOYA_ASYNC_EVENT_ID_TPC3_BMON_SPMU:
4413 	case GOYA_ASYNC_EVENT_ID_TPC4_BMON_SPMU:
4414 	case GOYA_ASYNC_EVENT_ID_TPC5_BMON_SPMU:
4415 	case GOYA_ASYNC_EVENT_ID_TPC6_BMON_SPMU:
4416 	case GOYA_ASYNC_EVENT_ID_TPC7_BMON_SPMU:
4417 		index = (event_type - GOYA_ASYNC_EVENT_ID_TPC0_BMON_SPMU) / 10;
4418 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4419 		break;
4420 	case GOYA_ASYNC_EVENT_ID_DMA_BM_CH0 ... GOYA_ASYNC_EVENT_ID_DMA_BM_CH4:
4421 		index = event_type - GOYA_ASYNC_EVENT_ID_DMA_BM_CH0;
4422 		snprintf(desc, size, _goya_get_event_desc(event_type), index);
4423 		break;
4424 	case GOYA_ASYNC_EVENT_PKT_QUEUE_OUT_SYNC:
4425 		snprintf(desc, size, _goya_get_event_desc(event_type));
4426 		break;
4427 	default:
4428 		snprintf(desc, size, _goya_get_event_desc(event_type));
4429 		break;
4430 	}
4431 }
4432 
4433 static void goya_print_razwi_info(struct hl_device *hdev)
4434 {
4435 	if (RREG32(mmDMA_MACRO_RAZWI_LBW_WT_VLD)) {
4436 		dev_err_ratelimited(hdev->dev, "Illegal write to LBW\n");
4437 		WREG32(mmDMA_MACRO_RAZWI_LBW_WT_VLD, 0);
4438 	}
4439 
4440 	if (RREG32(mmDMA_MACRO_RAZWI_LBW_RD_VLD)) {
4441 		dev_err_ratelimited(hdev->dev, "Illegal read from LBW\n");
4442 		WREG32(mmDMA_MACRO_RAZWI_LBW_RD_VLD, 0);
4443 	}
4444 
4445 	if (RREG32(mmDMA_MACRO_RAZWI_HBW_WT_VLD)) {
4446 		dev_err_ratelimited(hdev->dev, "Illegal write to HBW\n");
4447 		WREG32(mmDMA_MACRO_RAZWI_HBW_WT_VLD, 0);
4448 	}
4449 
4450 	if (RREG32(mmDMA_MACRO_RAZWI_HBW_RD_VLD)) {
4451 		dev_err_ratelimited(hdev->dev, "Illegal read from HBW\n");
4452 		WREG32(mmDMA_MACRO_RAZWI_HBW_RD_VLD, 0);
4453 	}
4454 }
4455 
4456 static void goya_print_mmu_error_info(struct hl_device *hdev)
4457 {
4458 	struct goya_device *goya = hdev->asic_specific;
4459 	u64 addr;
4460 	u32 val;
4461 
4462 	if (!(goya->hw_cap_initialized & HW_CAP_MMU))
4463 		return;
4464 
4465 	val = RREG32(mmMMU_PAGE_ERROR_CAPTURE);
4466 	if (val & MMU_PAGE_ERROR_CAPTURE_ENTRY_VALID_MASK) {
4467 		addr = val & MMU_PAGE_ERROR_CAPTURE_VA_49_32_MASK;
4468 		addr <<= 32;
4469 		addr |= RREG32(mmMMU_PAGE_ERROR_CAPTURE_VA);
4470 
4471 		dev_err_ratelimited(hdev->dev, "MMU page fault on va 0x%llx\n",
4472 					addr);
4473 
4474 		WREG32(mmMMU_PAGE_ERROR_CAPTURE, 0);
4475 	}
4476 }
4477 
4478 static void goya_print_out_of_sync_info(struct hl_device *hdev,
4479 					struct cpucp_pkt_sync_err *sync_err)
4480 {
4481 	struct hl_hw_queue *q = &hdev->kernel_queues[GOYA_QUEUE_ID_CPU_PQ];
4482 
4483 	dev_err(hdev->dev, "Out of sync with FW, FW: pi=%u, ci=%u, LKD: pi=%u, ci=%d\n",
4484 		le32_to_cpu(sync_err->pi), le32_to_cpu(sync_err->ci), q->pi, atomic_read(&q->ci));
4485 }
4486 
4487 static void goya_print_irq_info(struct hl_device *hdev, u16 event_type,
4488 				bool razwi)
4489 {
4490 	char desc[20] = "";
4491 
4492 	goya_get_event_desc(event_type, desc, sizeof(desc));
4493 	dev_err_ratelimited(hdev->dev, "Received H/W interrupt %d [\"%s\"]\n",
4494 		event_type, desc);
4495 
4496 	if (razwi) {
4497 		goya_print_razwi_info(hdev);
4498 		goya_print_mmu_error_info(hdev);
4499 	}
4500 }
4501 
4502 static int goya_unmask_irq_arr(struct hl_device *hdev, u32 *irq_arr,
4503 		size_t irq_arr_size)
4504 {
4505 	struct cpucp_unmask_irq_arr_packet *pkt;
4506 	size_t total_pkt_size;
4507 	u64 result;
4508 	int rc;
4509 	int irq_num_entries, irq_arr_index;
4510 	__le32 *goya_irq_arr;
4511 
4512 	total_pkt_size = sizeof(struct cpucp_unmask_irq_arr_packet) +
4513 			irq_arr_size;
4514 
4515 	/* data should be aligned to 8 bytes in order to CPU-CP to copy it */
4516 	total_pkt_size = (total_pkt_size + 0x7) & ~0x7;
4517 
4518 	/* total_pkt_size is casted to u16 later on */
4519 	if (total_pkt_size > USHRT_MAX) {
4520 		dev_err(hdev->dev, "too many elements in IRQ array\n");
4521 		return -EINVAL;
4522 	}
4523 
4524 	pkt = kzalloc(total_pkt_size, GFP_KERNEL);
4525 	if (!pkt)
4526 		return -ENOMEM;
4527 
4528 	irq_num_entries = irq_arr_size / sizeof(irq_arr[0]);
4529 	pkt->length = cpu_to_le32(irq_num_entries);
4530 
4531 	/* We must perform any necessary endianness conversation on the irq
4532 	 * array being passed to the goya hardware
4533 	 */
4534 	for (irq_arr_index = 0, goya_irq_arr = (__le32 *) &pkt->irqs;
4535 			irq_arr_index < irq_num_entries ; irq_arr_index++)
4536 		goya_irq_arr[irq_arr_index] =
4537 				cpu_to_le32(irq_arr[irq_arr_index]);
4538 
4539 	pkt->cpucp_pkt.ctl = cpu_to_le32(CPUCP_PACKET_UNMASK_RAZWI_IRQ_ARRAY <<
4540 						CPUCP_PKT_CTL_OPCODE_SHIFT);
4541 
4542 	rc = hdev->asic_funcs->send_cpu_message(hdev, (u32 *) pkt,
4543 						total_pkt_size,	0, &result);
4544 
4545 	if (rc)
4546 		dev_err(hdev->dev, "failed to unmask IRQ array\n");
4547 
4548 	kfree(pkt);
4549 
4550 	return rc;
4551 }
4552 
4553 static int goya_compute_reset_late_init(struct hl_device *hdev)
4554 {
4555 	/*
4556 	 * Unmask all IRQs since some could have been received
4557 	 * during the soft reset
4558 	 */
4559 	return goya_unmask_irq_arr(hdev, goya_all_events,
4560 					sizeof(goya_all_events));
4561 }
4562 
4563 static int goya_unmask_irq(struct hl_device *hdev, u16 event_type)
4564 {
4565 	struct cpucp_packet pkt;
4566 	u64 result;
4567 	int rc;
4568 
4569 	memset(&pkt, 0, sizeof(pkt));
4570 
4571 	pkt.ctl = cpu_to_le32(CPUCP_PACKET_UNMASK_RAZWI_IRQ <<
4572 				CPUCP_PKT_CTL_OPCODE_SHIFT);
4573 	pkt.value = cpu_to_le64(event_type);
4574 
4575 	rc = hdev->asic_funcs->send_cpu_message(hdev, (u32 *) &pkt, sizeof(pkt),
4576 						0, &result);
4577 
4578 	if (rc)
4579 		dev_err(hdev->dev, "failed to unmask RAZWI IRQ %d", event_type);
4580 
4581 	return rc;
4582 }
4583 
4584 static void goya_print_clk_change_info(struct hl_device *hdev, u16 event_type)
4585 {
4586 	ktime_t zero_time = ktime_set(0, 0);
4587 
4588 	mutex_lock(&hdev->clk_throttling.lock);
4589 
4590 	switch (event_type) {
4591 	case GOYA_ASYNC_EVENT_ID_FIX_POWER_ENV_S:
4592 		hdev->clk_throttling.current_reason |= HL_CLK_THROTTLE_POWER;
4593 		hdev->clk_throttling.aggregated_reason |= HL_CLK_THROTTLE_POWER;
4594 		hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_POWER].start = ktime_get();
4595 		hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_POWER].end = zero_time;
4596 		dev_info_ratelimited(hdev->dev,
4597 			"Clock throttling due to power consumption\n");
4598 		break;
4599 
4600 	case GOYA_ASYNC_EVENT_ID_FIX_POWER_ENV_E:
4601 		hdev->clk_throttling.current_reason &= ~HL_CLK_THROTTLE_POWER;
4602 		hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_POWER].end = ktime_get();
4603 		dev_info_ratelimited(hdev->dev,
4604 			"Power envelop is safe, back to optimal clock\n");
4605 		break;
4606 
4607 	case GOYA_ASYNC_EVENT_ID_FIX_THERMAL_ENV_S:
4608 		hdev->clk_throttling.current_reason |= HL_CLK_THROTTLE_THERMAL;
4609 		hdev->clk_throttling.aggregated_reason |= HL_CLK_THROTTLE_THERMAL;
4610 		hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_THERMAL].start = ktime_get();
4611 		hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_THERMAL].end = zero_time;
4612 		dev_info_ratelimited(hdev->dev,
4613 			"Clock throttling due to overheating\n");
4614 		break;
4615 
4616 	case GOYA_ASYNC_EVENT_ID_FIX_THERMAL_ENV_E:
4617 		hdev->clk_throttling.current_reason &= ~HL_CLK_THROTTLE_THERMAL;
4618 		hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_THERMAL].end = ktime_get();
4619 		dev_info_ratelimited(hdev->dev,
4620 			"Thermal envelop is safe, back to optimal clock\n");
4621 		break;
4622 
4623 	default:
4624 		dev_err(hdev->dev, "Received invalid clock change event %d\n",
4625 			event_type);
4626 		break;
4627 	}
4628 
4629 	mutex_unlock(&hdev->clk_throttling.lock);
4630 }
4631 
4632 void goya_handle_eqe(struct hl_device *hdev, struct hl_eq_entry *eq_entry)
4633 {
4634 	u32 ctl = le32_to_cpu(eq_entry->hdr.ctl);
4635 	u16 event_type = ((ctl & EQ_CTL_EVENT_TYPE_MASK)
4636 				>> EQ_CTL_EVENT_TYPE_SHIFT);
4637 	struct goya_device *goya = hdev->asic_specific;
4638 
4639 	if (event_type >= GOYA_ASYNC_EVENT_ID_SIZE) {
4640 		dev_err(hdev->dev, "Event type %u exceeds maximum of %u",
4641 				event_type, GOYA_ASYNC_EVENT_ID_SIZE - 1);
4642 		return;
4643 	}
4644 
4645 	goya->events_stat[event_type]++;
4646 	goya->events_stat_aggregate[event_type]++;
4647 
4648 	switch (event_type) {
4649 	case GOYA_ASYNC_EVENT_ID_PCIE_IF:
4650 	case GOYA_ASYNC_EVENT_ID_TPC0_ECC:
4651 	case GOYA_ASYNC_EVENT_ID_TPC1_ECC:
4652 	case GOYA_ASYNC_EVENT_ID_TPC2_ECC:
4653 	case GOYA_ASYNC_EVENT_ID_TPC3_ECC:
4654 	case GOYA_ASYNC_EVENT_ID_TPC4_ECC:
4655 	case GOYA_ASYNC_EVENT_ID_TPC5_ECC:
4656 	case GOYA_ASYNC_EVENT_ID_TPC6_ECC:
4657 	case GOYA_ASYNC_EVENT_ID_TPC7_ECC:
4658 	case GOYA_ASYNC_EVENT_ID_MME_ECC:
4659 	case GOYA_ASYNC_EVENT_ID_MME_ECC_EXT:
4660 	case GOYA_ASYNC_EVENT_ID_MMU_ECC:
4661 	case GOYA_ASYNC_EVENT_ID_DMA_MACRO:
4662 	case GOYA_ASYNC_EVENT_ID_DMA_ECC:
4663 	case GOYA_ASYNC_EVENT_ID_CPU_IF_ECC:
4664 	case GOYA_ASYNC_EVENT_ID_PSOC_MEM:
4665 	case GOYA_ASYNC_EVENT_ID_PSOC_CORESIGHT:
4666 	case GOYA_ASYNC_EVENT_ID_SRAM0 ... GOYA_ASYNC_EVENT_ID_SRAM29:
4667 	case GOYA_ASYNC_EVENT_ID_GIC500:
4668 	case GOYA_ASYNC_EVENT_ID_PLL0 ... GOYA_ASYNC_EVENT_ID_PLL6:
4669 	case GOYA_ASYNC_EVENT_ID_AXI_ECC:
4670 	case GOYA_ASYNC_EVENT_ID_L2_RAM_ECC:
4671 		goya_print_irq_info(hdev, event_type, false);
4672 		if (hdev->hard_reset_on_fw_events)
4673 			hl_device_reset(hdev, (HL_DRV_RESET_HARD |
4674 						HL_DRV_RESET_FW_FATAL_ERR));
4675 		break;
4676 
4677 	case GOYA_ASYNC_EVENT_ID_PSOC_GPIO_05_SW_RESET:
4678 		goya_print_irq_info(hdev, event_type, false);
4679 		if (hdev->hard_reset_on_fw_events)
4680 			hl_device_reset(hdev, HL_DRV_RESET_HARD);
4681 		break;
4682 
4683 	case GOYA_ASYNC_EVENT_ID_PCIE_DEC:
4684 	case GOYA_ASYNC_EVENT_ID_TPC0_DEC:
4685 	case GOYA_ASYNC_EVENT_ID_TPC1_DEC:
4686 	case GOYA_ASYNC_EVENT_ID_TPC2_DEC:
4687 	case GOYA_ASYNC_EVENT_ID_TPC3_DEC:
4688 	case GOYA_ASYNC_EVENT_ID_TPC4_DEC:
4689 	case GOYA_ASYNC_EVENT_ID_TPC5_DEC:
4690 	case GOYA_ASYNC_EVENT_ID_TPC6_DEC:
4691 	case GOYA_ASYNC_EVENT_ID_TPC7_DEC:
4692 	case GOYA_ASYNC_EVENT_ID_MME_WACS:
4693 	case GOYA_ASYNC_EVENT_ID_MME_WACSD:
4694 	case GOYA_ASYNC_EVENT_ID_CPU_AXI_SPLITTER:
4695 	case GOYA_ASYNC_EVENT_ID_PSOC_AXI_DEC:
4696 	case GOYA_ASYNC_EVENT_ID_PSOC:
4697 	case GOYA_ASYNC_EVENT_ID_TPC0_KRN_ERR:
4698 	case GOYA_ASYNC_EVENT_ID_TPC1_KRN_ERR:
4699 	case GOYA_ASYNC_EVENT_ID_TPC2_KRN_ERR:
4700 	case GOYA_ASYNC_EVENT_ID_TPC3_KRN_ERR:
4701 	case GOYA_ASYNC_EVENT_ID_TPC4_KRN_ERR:
4702 	case GOYA_ASYNC_EVENT_ID_TPC5_KRN_ERR:
4703 	case GOYA_ASYNC_EVENT_ID_TPC6_KRN_ERR:
4704 	case GOYA_ASYNC_EVENT_ID_TPC7_KRN_ERR:
4705 	case GOYA_ASYNC_EVENT_ID_TPC0_CMDQ ... GOYA_ASYNC_EVENT_ID_TPC7_QM:
4706 	case GOYA_ASYNC_EVENT_ID_MME_QM:
4707 	case GOYA_ASYNC_EVENT_ID_MME_CMDQ:
4708 	case GOYA_ASYNC_EVENT_ID_DMA0_QM ... GOYA_ASYNC_EVENT_ID_DMA4_QM:
4709 	case GOYA_ASYNC_EVENT_ID_DMA0_CH ... GOYA_ASYNC_EVENT_ID_DMA4_CH:
4710 		goya_print_irq_info(hdev, event_type, true);
4711 		goya_unmask_irq(hdev, event_type);
4712 		break;
4713 
4714 	case GOYA_ASYNC_EVENT_ID_PSOC_GPIO_10_VRHOT_ICRIT:
4715 	case GOYA_ASYNC_EVENT_ID_TPC0_BMON_SPMU:
4716 	case GOYA_ASYNC_EVENT_ID_TPC1_BMON_SPMU:
4717 	case GOYA_ASYNC_EVENT_ID_TPC2_BMON_SPMU:
4718 	case GOYA_ASYNC_EVENT_ID_TPC3_BMON_SPMU:
4719 	case GOYA_ASYNC_EVENT_ID_TPC4_BMON_SPMU:
4720 	case GOYA_ASYNC_EVENT_ID_TPC5_BMON_SPMU:
4721 	case GOYA_ASYNC_EVENT_ID_TPC6_BMON_SPMU:
4722 	case GOYA_ASYNC_EVENT_ID_TPC7_BMON_SPMU:
4723 	case GOYA_ASYNC_EVENT_ID_DMA_BM_CH0 ... GOYA_ASYNC_EVENT_ID_DMA_BM_CH4:
4724 		goya_print_irq_info(hdev, event_type, false);
4725 		goya_unmask_irq(hdev, event_type);
4726 		break;
4727 
4728 	case GOYA_ASYNC_EVENT_ID_FIX_POWER_ENV_S:
4729 	case GOYA_ASYNC_EVENT_ID_FIX_POWER_ENV_E:
4730 	case GOYA_ASYNC_EVENT_ID_FIX_THERMAL_ENV_S:
4731 	case GOYA_ASYNC_EVENT_ID_FIX_THERMAL_ENV_E:
4732 		goya_print_clk_change_info(hdev, event_type);
4733 		goya_unmask_irq(hdev, event_type);
4734 		break;
4735 
4736 	case GOYA_ASYNC_EVENT_PKT_QUEUE_OUT_SYNC:
4737 		goya_print_irq_info(hdev, event_type, false);
4738 		goya_print_out_of_sync_info(hdev, &eq_entry->pkt_sync_err);
4739 		if (hdev->hard_reset_on_fw_events)
4740 			hl_device_reset(hdev, HL_DRV_RESET_HARD);
4741 		else
4742 			hl_fw_unmask_irq(hdev, event_type);
4743 		break;
4744 
4745 	default:
4746 		dev_err(hdev->dev, "Received invalid H/W interrupt %d\n",
4747 				event_type);
4748 		break;
4749 	}
4750 }
4751 
4752 void *goya_get_events_stat(struct hl_device *hdev, bool aggregate, u32 *size)
4753 {
4754 	struct goya_device *goya = hdev->asic_specific;
4755 
4756 	if (aggregate) {
4757 		*size = (u32) sizeof(goya->events_stat_aggregate);
4758 		return goya->events_stat_aggregate;
4759 	}
4760 
4761 	*size = (u32) sizeof(goya->events_stat);
4762 	return goya->events_stat;
4763 }
4764 
4765 static int goya_memset_device_memory(struct hl_device *hdev, u64 addr, u64 size,
4766 				u64 val, bool is_dram)
4767 {
4768 	struct packet_lin_dma *lin_dma_pkt;
4769 	struct hl_cs_job *job;
4770 	u32 cb_size, ctl;
4771 	struct hl_cb *cb;
4772 	int rc, lin_dma_pkts_cnt;
4773 
4774 	lin_dma_pkts_cnt = DIV_ROUND_UP_ULL(size, SZ_2G);
4775 	cb_size = lin_dma_pkts_cnt * sizeof(struct packet_lin_dma) +
4776 						sizeof(struct packet_msg_prot);
4777 	cb = hl_cb_kernel_create(hdev, cb_size, false);
4778 	if (!cb)
4779 		return -ENOMEM;
4780 
4781 	lin_dma_pkt = cb->kernel_address;
4782 
4783 	do {
4784 		memset(lin_dma_pkt, 0, sizeof(*lin_dma_pkt));
4785 
4786 		ctl = ((PACKET_LIN_DMA << GOYA_PKT_CTL_OPCODE_SHIFT) |
4787 				(1 << GOYA_PKT_LIN_DMA_CTL_MEMSET_SHIFT) |
4788 				(1 << GOYA_PKT_LIN_DMA_CTL_WO_SHIFT) |
4789 				(1 << GOYA_PKT_CTL_RB_SHIFT) |
4790 				(1 << GOYA_PKT_CTL_MB_SHIFT));
4791 		ctl |= (is_dram ? HL_DMA_HOST_TO_DRAM : HL_DMA_HOST_TO_SRAM) <<
4792 				GOYA_PKT_LIN_DMA_CTL_DMA_DIR_SHIFT;
4793 		lin_dma_pkt->ctl = cpu_to_le32(ctl);
4794 
4795 		lin_dma_pkt->src_addr = cpu_to_le64(val);
4796 		lin_dma_pkt->dst_addr = cpu_to_le64(addr);
4797 		if (lin_dma_pkts_cnt > 1)
4798 			lin_dma_pkt->tsize = cpu_to_le32(SZ_2G);
4799 		else
4800 			lin_dma_pkt->tsize = cpu_to_le32(size);
4801 
4802 		size -= SZ_2G;
4803 		addr += SZ_2G;
4804 		lin_dma_pkt++;
4805 	} while (--lin_dma_pkts_cnt);
4806 
4807 	job = hl_cs_allocate_job(hdev, QUEUE_TYPE_EXT, true);
4808 	if (!job) {
4809 		dev_err(hdev->dev, "Failed to allocate a new job\n");
4810 		rc = -ENOMEM;
4811 		goto release_cb;
4812 	}
4813 
4814 	job->id = 0;
4815 	job->user_cb = cb;
4816 	atomic_inc(&job->user_cb->cs_cnt);
4817 	job->user_cb_size = cb_size;
4818 	job->hw_queue_id = GOYA_QUEUE_ID_DMA_0;
4819 	job->patched_cb = job->user_cb;
4820 	job->job_cb_size = job->user_cb_size;
4821 
4822 	hl_debugfs_add_job(hdev, job);
4823 
4824 	rc = goya_send_job_on_qman0(hdev, job);
4825 
4826 	hl_debugfs_remove_job(hdev, job);
4827 	kfree(job);
4828 	atomic_dec(&cb->cs_cnt);
4829 
4830 release_cb:
4831 	hl_cb_put(cb);
4832 	hl_cb_destroy(&hdev->kernel_mem_mgr, cb->buf->handle);
4833 
4834 	return rc;
4835 }
4836 
4837 int goya_context_switch(struct hl_device *hdev, u32 asid)
4838 {
4839 	struct asic_fixed_properties *prop = &hdev->asic_prop;
4840 	u64 addr = prop->sram_base_address, sob_addr;
4841 	u32 size = hdev->pldm ? 0x10000 : prop->sram_size;
4842 	u64 val = 0x7777777777777777ull;
4843 	int rc, dma_id;
4844 	u32 channel_off = mmDMA_CH_1_WR_COMP_ADDR_LO -
4845 					mmDMA_CH_0_WR_COMP_ADDR_LO;
4846 
4847 	rc = goya_memset_device_memory(hdev, addr, size, val, false);
4848 	if (rc) {
4849 		dev_err(hdev->dev, "Failed to clear SRAM in context switch\n");
4850 		return rc;
4851 	}
4852 
4853 	/* we need to reset registers that the user is allowed to change */
4854 	sob_addr = CFG_BASE + mmSYNC_MNGR_SOB_OBJ_1007;
4855 	WREG32(mmDMA_CH_0_WR_COMP_ADDR_LO, lower_32_bits(sob_addr));
4856 
4857 	for (dma_id = 1 ; dma_id < NUMBER_OF_EXT_HW_QUEUES ; dma_id++) {
4858 		sob_addr = CFG_BASE + mmSYNC_MNGR_SOB_OBJ_1000 +
4859 							(dma_id - 1) * 4;
4860 		WREG32(mmDMA_CH_0_WR_COMP_ADDR_LO + channel_off * dma_id,
4861 						lower_32_bits(sob_addr));
4862 	}
4863 
4864 	WREG32(mmTPC_PLL_CLK_RLX_0, 0x200020);
4865 
4866 	goya_clear_sm_regs(hdev);
4867 
4868 	return 0;
4869 }
4870 
4871 static int goya_mmu_clear_pgt_range(struct hl_device *hdev)
4872 {
4873 	struct asic_fixed_properties *prop = &hdev->asic_prop;
4874 	struct goya_device *goya = hdev->asic_specific;
4875 	u64 addr = prop->mmu_pgt_addr;
4876 	u32 size = prop->mmu_pgt_size + MMU_DRAM_DEFAULT_PAGE_SIZE +
4877 			MMU_CACHE_MNG_SIZE;
4878 
4879 	if (!(goya->hw_cap_initialized & HW_CAP_MMU))
4880 		return 0;
4881 
4882 	return goya_memset_device_memory(hdev, addr, size, 0, true);
4883 }
4884 
4885 static int goya_mmu_set_dram_default_page(struct hl_device *hdev)
4886 {
4887 	struct goya_device *goya = hdev->asic_specific;
4888 	u64 addr = hdev->asic_prop.mmu_dram_default_page_addr;
4889 	u32 size = MMU_DRAM_DEFAULT_PAGE_SIZE;
4890 	u64 val = 0x9999999999999999ull;
4891 
4892 	if (!(goya->hw_cap_initialized & HW_CAP_MMU))
4893 		return 0;
4894 
4895 	return goya_memset_device_memory(hdev, addr, size, val, true);
4896 }
4897 
4898 static int goya_mmu_add_mappings_for_device_cpu(struct hl_device *hdev)
4899 {
4900 	struct asic_fixed_properties *prop = &hdev->asic_prop;
4901 	struct goya_device *goya = hdev->asic_specific;
4902 	s64 off, cpu_off;
4903 	int rc;
4904 
4905 	if (!(goya->hw_cap_initialized & HW_CAP_MMU))
4906 		return 0;
4907 
4908 	for (off = 0 ; off < CPU_FW_IMAGE_SIZE ; off += PAGE_SIZE_2MB) {
4909 		rc = hl_mmu_map_page(hdev->kernel_ctx,
4910 			prop->dram_base_address + off,
4911 			prop->dram_base_address + off, PAGE_SIZE_2MB,
4912 			(off + PAGE_SIZE_2MB) == CPU_FW_IMAGE_SIZE);
4913 		if (rc) {
4914 			dev_err(hdev->dev, "Map failed for address 0x%llx\n",
4915 				prop->dram_base_address + off);
4916 			goto unmap;
4917 		}
4918 	}
4919 
4920 	if (!(hdev->cpu_accessible_dma_address & (PAGE_SIZE_2MB - 1))) {
4921 		rc = hl_mmu_map_page(hdev->kernel_ctx,
4922 			VA_CPU_ACCESSIBLE_MEM_ADDR,
4923 			hdev->cpu_accessible_dma_address,
4924 			PAGE_SIZE_2MB, true);
4925 
4926 		if (rc) {
4927 			dev_err(hdev->dev,
4928 				"Map failed for CPU accessible memory\n");
4929 			off -= PAGE_SIZE_2MB;
4930 			goto unmap;
4931 		}
4932 	} else {
4933 		for (cpu_off = 0 ; cpu_off < SZ_2M ; cpu_off += PAGE_SIZE_4KB) {
4934 			rc = hl_mmu_map_page(hdev->kernel_ctx,
4935 				VA_CPU_ACCESSIBLE_MEM_ADDR + cpu_off,
4936 				hdev->cpu_accessible_dma_address + cpu_off,
4937 				PAGE_SIZE_4KB, true);
4938 			if (rc) {
4939 				dev_err(hdev->dev,
4940 					"Map failed for CPU accessible memory\n");
4941 				cpu_off -= PAGE_SIZE_4KB;
4942 				goto unmap_cpu;
4943 			}
4944 		}
4945 	}
4946 
4947 	goya_mmu_prepare_reg(hdev, mmCPU_IF_ARUSER_OVR, HL_KERNEL_ASID_ID);
4948 	goya_mmu_prepare_reg(hdev, mmCPU_IF_AWUSER_OVR, HL_KERNEL_ASID_ID);
4949 	WREG32(mmCPU_IF_ARUSER_OVR_EN, 0x7FF);
4950 	WREG32(mmCPU_IF_AWUSER_OVR_EN, 0x7FF);
4951 
4952 	/* Make sure configuration is flushed to device */
4953 	RREG32(mmCPU_IF_AWUSER_OVR_EN);
4954 
4955 	goya->device_cpu_mmu_mappings_done = true;
4956 
4957 	return 0;
4958 
4959 unmap_cpu:
4960 	for (; cpu_off >= 0 ; cpu_off -= PAGE_SIZE_4KB)
4961 		if (hl_mmu_unmap_page(hdev->kernel_ctx,
4962 				VA_CPU_ACCESSIBLE_MEM_ADDR + cpu_off,
4963 				PAGE_SIZE_4KB, true))
4964 			dev_warn_ratelimited(hdev->dev,
4965 				"failed to unmap address 0x%llx\n",
4966 				VA_CPU_ACCESSIBLE_MEM_ADDR + cpu_off);
4967 unmap:
4968 	for (; off >= 0 ; off -= PAGE_SIZE_2MB)
4969 		if (hl_mmu_unmap_page(hdev->kernel_ctx,
4970 				prop->dram_base_address + off, PAGE_SIZE_2MB,
4971 				true))
4972 			dev_warn_ratelimited(hdev->dev,
4973 				"failed to unmap address 0x%llx\n",
4974 				prop->dram_base_address + off);
4975 
4976 	return rc;
4977 }
4978 
4979 void goya_mmu_remove_device_cpu_mappings(struct hl_device *hdev)
4980 {
4981 	struct asic_fixed_properties *prop = &hdev->asic_prop;
4982 	struct goya_device *goya = hdev->asic_specific;
4983 	u32 off, cpu_off;
4984 
4985 	if (!(goya->hw_cap_initialized & HW_CAP_MMU))
4986 		return;
4987 
4988 	if (!goya->device_cpu_mmu_mappings_done)
4989 		return;
4990 
4991 	WREG32(mmCPU_IF_ARUSER_OVR_EN, 0);
4992 	WREG32(mmCPU_IF_AWUSER_OVR_EN, 0);
4993 
4994 	if (!(hdev->cpu_accessible_dma_address & (PAGE_SIZE_2MB - 1))) {
4995 		if (hl_mmu_unmap_page(hdev->kernel_ctx,
4996 				VA_CPU_ACCESSIBLE_MEM_ADDR,
4997 				PAGE_SIZE_2MB, true))
4998 			dev_warn(hdev->dev,
4999 				"Failed to unmap CPU accessible memory\n");
5000 	} else {
5001 		for (cpu_off = 0 ; cpu_off < SZ_2M ; cpu_off += PAGE_SIZE_4KB)
5002 			if (hl_mmu_unmap_page(hdev->kernel_ctx,
5003 					VA_CPU_ACCESSIBLE_MEM_ADDR + cpu_off,
5004 					PAGE_SIZE_4KB,
5005 					(cpu_off + PAGE_SIZE_4KB) >= SZ_2M))
5006 				dev_warn_ratelimited(hdev->dev,
5007 					"failed to unmap address 0x%llx\n",
5008 					VA_CPU_ACCESSIBLE_MEM_ADDR + cpu_off);
5009 	}
5010 
5011 	for (off = 0 ; off < CPU_FW_IMAGE_SIZE ; off += PAGE_SIZE_2MB)
5012 		if (hl_mmu_unmap_page(hdev->kernel_ctx,
5013 				prop->dram_base_address + off, PAGE_SIZE_2MB,
5014 				(off + PAGE_SIZE_2MB) >= CPU_FW_IMAGE_SIZE))
5015 			dev_warn_ratelimited(hdev->dev,
5016 					"Failed to unmap address 0x%llx\n",
5017 					prop->dram_base_address + off);
5018 
5019 	goya->device_cpu_mmu_mappings_done = false;
5020 }
5021 
5022 static void goya_mmu_prepare(struct hl_device *hdev, u32 asid)
5023 {
5024 	struct goya_device *goya = hdev->asic_specific;
5025 	int i;
5026 
5027 	if (!(goya->hw_cap_initialized & HW_CAP_MMU))
5028 		return;
5029 
5030 	if (asid & ~MME_QM_GLBL_SECURE_PROPS_ASID_MASK) {
5031 		dev_crit(hdev->dev, "asid %u is too big\n", asid);
5032 		return;
5033 	}
5034 
5035 	/* zero the MMBP and ASID bits and then set the ASID */
5036 	for (i = 0 ; i < GOYA_MMU_REGS_NUM ; i++)
5037 		goya_mmu_prepare_reg(hdev, goya_mmu_regs[i], asid);
5038 }
5039 
5040 static int goya_mmu_invalidate_cache(struct hl_device *hdev, bool is_hard,
5041 					u32 flags)
5042 {
5043 	struct goya_device *goya = hdev->asic_specific;
5044 	u32 status, timeout_usec;
5045 	int rc;
5046 
5047 	if (!(goya->hw_cap_initialized & HW_CAP_MMU) ||
5048 		hdev->reset_info.hard_reset_pending)
5049 		return 0;
5050 
5051 	/* no need in L1 only invalidation in Goya */
5052 	if (!is_hard)
5053 		return 0;
5054 
5055 	if (hdev->pldm)
5056 		timeout_usec = GOYA_PLDM_MMU_TIMEOUT_USEC;
5057 	else
5058 		timeout_usec = MMU_CONFIG_TIMEOUT_USEC;
5059 
5060 	/* L0 & L1 invalidation */
5061 	WREG32(mmSTLB_INV_ALL_START, 1);
5062 
5063 	rc = hl_poll_timeout(
5064 		hdev,
5065 		mmSTLB_INV_ALL_START,
5066 		status,
5067 		!status,
5068 		1000,
5069 		timeout_usec);
5070 
5071 	return rc;
5072 }
5073 
5074 static int goya_mmu_invalidate_cache_range(struct hl_device *hdev,
5075 						bool is_hard, u32 flags,
5076 						u32 asid, u64 va, u64 size)
5077 {
5078 	/* Treat as invalidate all because there is no range invalidation
5079 	 * in Goya
5080 	 */
5081 	return hl_mmu_invalidate_cache(hdev, is_hard, flags);
5082 }
5083 
5084 int goya_send_heartbeat(struct hl_device *hdev)
5085 {
5086 	struct goya_device *goya = hdev->asic_specific;
5087 
5088 	if (!(goya->hw_cap_initialized & HW_CAP_CPU_Q))
5089 		return 0;
5090 
5091 	return hl_fw_send_heartbeat(hdev);
5092 }
5093 
5094 int goya_cpucp_info_get(struct hl_device *hdev)
5095 {
5096 	struct goya_device *goya = hdev->asic_specific;
5097 	struct asic_fixed_properties *prop = &hdev->asic_prop;
5098 	u64 dram_size;
5099 	int rc;
5100 
5101 	if (!(goya->hw_cap_initialized & HW_CAP_CPU_Q))
5102 		return 0;
5103 
5104 	rc = hl_fw_cpucp_handshake(hdev, mmCPU_BOOT_DEV_STS0,
5105 					mmCPU_BOOT_DEV_STS1, mmCPU_BOOT_ERR0,
5106 					mmCPU_BOOT_ERR1);
5107 	if (rc)
5108 		return rc;
5109 
5110 	dram_size = le64_to_cpu(prop->cpucp_info.dram_size);
5111 	if (dram_size) {
5112 		if ((!is_power_of_2(dram_size)) ||
5113 				(dram_size < DRAM_PHYS_DEFAULT_SIZE)) {
5114 			dev_err(hdev->dev,
5115 				"F/W reported invalid DRAM size %llu. Trying to use default size\n",
5116 				dram_size);
5117 			dram_size = DRAM_PHYS_DEFAULT_SIZE;
5118 		}
5119 
5120 		prop->dram_size = dram_size;
5121 		prop->dram_end_address = prop->dram_base_address + dram_size;
5122 	}
5123 
5124 	if (!strlen(prop->cpucp_info.card_name))
5125 		strscpy_pad(prop->cpucp_info.card_name, GOYA_DEFAULT_CARD_NAME,
5126 				CARD_NAME_MAX_LEN);
5127 
5128 	return 0;
5129 }
5130 
5131 static bool goya_is_device_idle(struct hl_device *hdev, u64 *mask_arr, u8 mask_len,
5132 				struct engines_data *e)
5133 {
5134 	const char *fmt = "%-5d%-9s%#-14x%#-16x%#x\n";
5135 	const char *dma_fmt = "%-5d%-9s%#-14x%#x\n";
5136 	unsigned long *mask = (unsigned long *)mask_arr;
5137 	u32 qm_glbl_sts0, cmdq_glbl_sts0, dma_core_sts0, tpc_cfg_sts,
5138 		mme_arch_sts;
5139 	bool is_idle = true, is_eng_idle;
5140 	u64 offset;
5141 	int i;
5142 
5143 	if (e)
5144 		hl_engine_data_sprintf(e, "\nDMA  is_idle  QM_GLBL_STS0  DMA_CORE_STS0\n"
5145 					"---  -------  ------------  -------------\n");
5146 
5147 	offset = mmDMA_QM_1_GLBL_STS0 - mmDMA_QM_0_GLBL_STS0;
5148 
5149 	for (i = 0 ; i < DMA_MAX_NUM ; i++) {
5150 		qm_glbl_sts0 = RREG32(mmDMA_QM_0_GLBL_STS0 + i * offset);
5151 		dma_core_sts0 = RREG32(mmDMA_CH_0_STS0 + i * offset);
5152 		is_eng_idle = IS_DMA_QM_IDLE(qm_glbl_sts0) &&
5153 				IS_DMA_IDLE(dma_core_sts0);
5154 		is_idle &= is_eng_idle;
5155 
5156 		if (mask && !is_eng_idle)
5157 			set_bit(GOYA_ENGINE_ID_DMA_0 + i, mask);
5158 		if (e)
5159 			hl_engine_data_sprintf(e, dma_fmt, i, is_eng_idle ? "Y" : "N",
5160 					qm_glbl_sts0, dma_core_sts0);
5161 	}
5162 
5163 	if (e)
5164 		hl_engine_data_sprintf(e,
5165 			"\nTPC  is_idle  QM_GLBL_STS0  CMDQ_GLBL_STS0  CFG_STATUS\n"
5166 			"---  -------  ------------  --------------  ----------\n");
5167 
5168 	offset = mmTPC1_QM_GLBL_STS0 - mmTPC0_QM_GLBL_STS0;
5169 
5170 	for (i = 0 ; i < TPC_MAX_NUM ; i++) {
5171 		qm_glbl_sts0 = RREG32(mmTPC0_QM_GLBL_STS0 + i * offset);
5172 		cmdq_glbl_sts0 = RREG32(mmTPC0_CMDQ_GLBL_STS0 + i * offset);
5173 		tpc_cfg_sts = RREG32(mmTPC0_CFG_STATUS + i * offset);
5174 		is_eng_idle = IS_TPC_QM_IDLE(qm_glbl_sts0) &&
5175 				IS_TPC_CMDQ_IDLE(cmdq_glbl_sts0) &&
5176 				IS_TPC_IDLE(tpc_cfg_sts);
5177 		is_idle &= is_eng_idle;
5178 
5179 		if (mask && !is_eng_idle)
5180 			set_bit(GOYA_ENGINE_ID_TPC_0 + i, mask);
5181 		if (e)
5182 			hl_engine_data_sprintf(e, fmt, i, is_eng_idle ? "Y" : "N",
5183 				qm_glbl_sts0, cmdq_glbl_sts0, tpc_cfg_sts);
5184 	}
5185 
5186 	if (e)
5187 		hl_engine_data_sprintf(e,
5188 			"\nMME  is_idle  QM_GLBL_STS0  CMDQ_GLBL_STS0  ARCH_STATUS\n"
5189 			"---  -------  ------------  --------------  -----------\n");
5190 
5191 	qm_glbl_sts0 = RREG32(mmMME_QM_GLBL_STS0);
5192 	cmdq_glbl_sts0 = RREG32(mmMME_CMDQ_GLBL_STS0);
5193 	mme_arch_sts = RREG32(mmMME_ARCH_STATUS);
5194 	is_eng_idle = IS_MME_QM_IDLE(qm_glbl_sts0) &&
5195 			IS_MME_CMDQ_IDLE(cmdq_glbl_sts0) &&
5196 			IS_MME_IDLE(mme_arch_sts);
5197 	is_idle &= is_eng_idle;
5198 
5199 	if (mask && !is_eng_idle)
5200 		set_bit(GOYA_ENGINE_ID_MME_0, mask);
5201 	if (e) {
5202 		hl_engine_data_sprintf(e, fmt, 0, is_eng_idle ? "Y" : "N", qm_glbl_sts0,
5203 				cmdq_glbl_sts0, mme_arch_sts);
5204 		hl_engine_data_sprintf(e, "\n");
5205 	}
5206 
5207 	return is_idle;
5208 }
5209 
5210 static void goya_hw_queues_lock(struct hl_device *hdev)
5211 	__acquires(&goya->hw_queues_lock)
5212 {
5213 	struct goya_device *goya = hdev->asic_specific;
5214 
5215 	spin_lock(&goya->hw_queues_lock);
5216 }
5217 
5218 static void goya_hw_queues_unlock(struct hl_device *hdev)
5219 	__releases(&goya->hw_queues_lock)
5220 {
5221 	struct goya_device *goya = hdev->asic_specific;
5222 
5223 	spin_unlock(&goya->hw_queues_lock);
5224 }
5225 
5226 static u32 goya_get_pci_id(struct hl_device *hdev)
5227 {
5228 	return hdev->pdev->device;
5229 }
5230 
5231 static int goya_get_eeprom_data(struct hl_device *hdev, void *data,
5232 				size_t max_size)
5233 {
5234 	struct goya_device *goya = hdev->asic_specific;
5235 
5236 	if (!(goya->hw_cap_initialized & HW_CAP_CPU_Q))
5237 		return 0;
5238 
5239 	return hl_fw_get_eeprom_data(hdev, data, max_size);
5240 }
5241 
5242 static void goya_cpu_init_scrambler_dram(struct hl_device *hdev)
5243 {
5244 
5245 }
5246 
5247 static int goya_ctx_init(struct hl_ctx *ctx)
5248 {
5249 	if (ctx->asid != HL_KERNEL_ASID_ID)
5250 		goya_mmu_prepare(ctx->hdev, ctx->asid);
5251 
5252 	return 0;
5253 }
5254 
5255 static int goya_pre_schedule_cs(struct hl_cs *cs)
5256 {
5257 	return 0;
5258 }
5259 
5260 u32 goya_get_queue_id_for_cq(struct hl_device *hdev, u32 cq_idx)
5261 {
5262 	return cq_idx;
5263 }
5264 
5265 static u32 goya_get_signal_cb_size(struct hl_device *hdev)
5266 {
5267 	return 0;
5268 }
5269 
5270 static u32 goya_get_wait_cb_size(struct hl_device *hdev)
5271 {
5272 	return 0;
5273 }
5274 
5275 static u32 goya_gen_signal_cb(struct hl_device *hdev, void *data, u16 sob_id,
5276 				u32 size, bool eb)
5277 {
5278 	return 0;
5279 }
5280 
5281 static u32 goya_gen_wait_cb(struct hl_device *hdev,
5282 		struct hl_gen_wait_properties *prop)
5283 {
5284 	return 0;
5285 }
5286 
5287 static void goya_reset_sob(struct hl_device *hdev, void *data)
5288 {
5289 
5290 }
5291 
5292 static void goya_reset_sob_group(struct hl_device *hdev, u16 sob_group)
5293 {
5294 
5295 }
5296 
5297 u64 goya_get_device_time(struct hl_device *hdev)
5298 {
5299 	u64 device_time = ((u64) RREG32(mmPSOC_TIMESTAMP_CNTCVU)) << 32;
5300 
5301 	return device_time | RREG32(mmPSOC_TIMESTAMP_CNTCVL);
5302 }
5303 
5304 static int goya_collective_wait_init_cs(struct hl_cs *cs)
5305 {
5306 	return 0;
5307 }
5308 
5309 static int goya_collective_wait_create_jobs(struct hl_device *hdev,
5310 		struct hl_ctx *ctx, struct hl_cs *cs, u32 wait_queue_id,
5311 		u32 collective_engine_id, u32 encaps_signal_offset)
5312 {
5313 	return -EINVAL;
5314 }
5315 
5316 static void goya_ctx_fini(struct hl_ctx *ctx)
5317 {
5318 
5319 }
5320 
5321 static int goya_get_hw_block_id(struct hl_device *hdev, u64 block_addr,
5322 			u32 *block_size, u32 *block_id)
5323 {
5324 	return -EPERM;
5325 }
5326 
5327 static int goya_block_mmap(struct hl_device *hdev, struct vm_area_struct *vma,
5328 				u32 block_id, u32 block_size)
5329 {
5330 	return -EPERM;
5331 }
5332 
5333 static void goya_enable_events_from_fw(struct hl_device *hdev)
5334 {
5335 	WREG32(mmGIC_DISTRIBUTOR__5_GICD_SETSPI_NSR,
5336 			GOYA_ASYNC_EVENT_ID_INTS_REGISTER);
5337 }
5338 
5339 static int goya_ack_mmu_page_fault_or_access_error(struct hl_device *hdev, u64 mmu_cap_mask)
5340 {
5341 	return -EINVAL;
5342 }
5343 
5344 static int goya_map_pll_idx_to_fw_idx(u32 pll_idx)
5345 {
5346 	switch (pll_idx) {
5347 	case HL_GOYA_CPU_PLL: return CPU_PLL;
5348 	case HL_GOYA_PCI_PLL: return PCI_PLL;
5349 	case HL_GOYA_MME_PLL: return MME_PLL;
5350 	case HL_GOYA_TPC_PLL: return TPC_PLL;
5351 	case HL_GOYA_IC_PLL: return IC_PLL;
5352 	case HL_GOYA_MC_PLL: return MC_PLL;
5353 	case HL_GOYA_EMMC_PLL: return EMMC_PLL;
5354 	default: return -EINVAL;
5355 	}
5356 }
5357 
5358 static int goya_gen_sync_to_engine_map(struct hl_device *hdev,
5359 				struct hl_sync_to_engine_map *map)
5360 {
5361 	/* Not implemented */
5362 	return 0;
5363 }
5364 
5365 static int goya_monitor_valid(struct hl_mon_state_dump *mon)
5366 {
5367 	/* Not implemented */
5368 	return 0;
5369 }
5370 
5371 static int goya_print_single_monitor(char **buf, size_t *size, size_t *offset,
5372 				struct hl_device *hdev,
5373 				struct hl_mon_state_dump *mon)
5374 {
5375 	/* Not implemented */
5376 	return 0;
5377 }
5378 
5379 
5380 static int goya_print_fences_single_engine(
5381 	struct hl_device *hdev, u64 base_offset, u64 status_base_offset,
5382 	enum hl_sync_engine_type engine_type, u32 engine_id, char **buf,
5383 	size_t *size, size_t *offset)
5384 {
5385 	/* Not implemented */
5386 	return 0;
5387 }
5388 
5389 
5390 static struct hl_state_dump_specs_funcs goya_state_dump_funcs = {
5391 	.monitor_valid = goya_monitor_valid,
5392 	.print_single_monitor = goya_print_single_monitor,
5393 	.gen_sync_to_engine_map = goya_gen_sync_to_engine_map,
5394 	.print_fences_single_engine = goya_print_fences_single_engine,
5395 };
5396 
5397 static void goya_state_dump_init(struct hl_device *hdev)
5398 {
5399 	/* Not implemented */
5400 	hdev->state_dump_specs.props = goya_state_dump_specs_props;
5401 	hdev->state_dump_specs.funcs = goya_state_dump_funcs;
5402 }
5403 
5404 static u32 goya_get_sob_addr(struct hl_device *hdev, u32 sob_id)
5405 {
5406 	return 0;
5407 }
5408 
5409 static u32 *goya_get_stream_master_qid_arr(void)
5410 {
5411 	return NULL;
5412 }
5413 
5414 static int goya_get_monitor_dump(struct hl_device *hdev, void *data)
5415 {
5416 	return -EOPNOTSUPP;
5417 }
5418 
5419 static void goya_check_if_razwi_happened(struct hl_device *hdev)
5420 {
5421 }
5422 
5423 static int goya_scrub_device_dram(struct hl_device *hdev, u64 val)
5424 {
5425 	return -EOPNOTSUPP;
5426 }
5427 
5428 static int goya_set_dram_properties(struct hl_device *hdev)
5429 {
5430 	return 0;
5431 }
5432 
5433 static int goya_set_binning_masks(struct hl_device *hdev)
5434 {
5435 	return 0;
5436 }
5437 
5438 static int goya_send_device_activity(struct hl_device *hdev, bool open)
5439 {
5440 	return 0;
5441 }
5442 
5443 static const struct hl_asic_funcs goya_funcs = {
5444 	.early_init = goya_early_init,
5445 	.early_fini = goya_early_fini,
5446 	.late_init = goya_late_init,
5447 	.late_fini = goya_late_fini,
5448 	.sw_init = goya_sw_init,
5449 	.sw_fini = goya_sw_fini,
5450 	.hw_init = goya_hw_init,
5451 	.hw_fini = goya_hw_fini,
5452 	.halt_engines = goya_halt_engines,
5453 	.suspend = goya_suspend,
5454 	.resume = goya_resume,
5455 	.mmap = goya_mmap,
5456 	.ring_doorbell = goya_ring_doorbell,
5457 	.pqe_write = goya_pqe_write,
5458 	.asic_dma_alloc_coherent = goya_dma_alloc_coherent,
5459 	.asic_dma_free_coherent = goya_dma_free_coherent,
5460 	.scrub_device_mem = goya_scrub_device_mem,
5461 	.scrub_device_dram = goya_scrub_device_dram,
5462 	.get_int_queue_base = goya_get_int_queue_base,
5463 	.test_queues = goya_test_queues,
5464 	.asic_dma_pool_zalloc = goya_dma_pool_zalloc,
5465 	.asic_dma_pool_free = goya_dma_pool_free,
5466 	.cpu_accessible_dma_pool_alloc = goya_cpu_accessible_dma_pool_alloc,
5467 	.cpu_accessible_dma_pool_free = goya_cpu_accessible_dma_pool_free,
5468 	.dma_unmap_sgtable = hl_asic_dma_unmap_sgtable,
5469 	.cs_parser = goya_cs_parser,
5470 	.dma_map_sgtable = hl_asic_dma_map_sgtable,
5471 	.add_end_of_cb_packets = goya_add_end_of_cb_packets,
5472 	.update_eq_ci = goya_update_eq_ci,
5473 	.context_switch = goya_context_switch,
5474 	.restore_phase_topology = goya_restore_phase_topology,
5475 	.debugfs_read_dma = goya_debugfs_read_dma,
5476 	.add_device_attr = goya_add_device_attr,
5477 	.handle_eqe = goya_handle_eqe,
5478 	.get_events_stat = goya_get_events_stat,
5479 	.read_pte = goya_read_pte,
5480 	.write_pte = goya_write_pte,
5481 	.mmu_invalidate_cache = goya_mmu_invalidate_cache,
5482 	.mmu_invalidate_cache_range = goya_mmu_invalidate_cache_range,
5483 	.mmu_prefetch_cache_range = NULL,
5484 	.send_heartbeat = goya_send_heartbeat,
5485 	.debug_coresight = goya_debug_coresight,
5486 	.is_device_idle = goya_is_device_idle,
5487 	.compute_reset_late_init = goya_compute_reset_late_init,
5488 	.hw_queues_lock = goya_hw_queues_lock,
5489 	.hw_queues_unlock = goya_hw_queues_unlock,
5490 	.get_pci_id = goya_get_pci_id,
5491 	.get_eeprom_data = goya_get_eeprom_data,
5492 	.get_monitor_dump = goya_get_monitor_dump,
5493 	.send_cpu_message = goya_send_cpu_message,
5494 	.pci_bars_map = goya_pci_bars_map,
5495 	.init_iatu = goya_init_iatu,
5496 	.rreg = hl_rreg,
5497 	.wreg = hl_wreg,
5498 	.halt_coresight = goya_halt_coresight,
5499 	.ctx_init = goya_ctx_init,
5500 	.ctx_fini = goya_ctx_fini,
5501 	.pre_schedule_cs = goya_pre_schedule_cs,
5502 	.get_queue_id_for_cq = goya_get_queue_id_for_cq,
5503 	.load_firmware_to_device = goya_load_firmware_to_device,
5504 	.load_boot_fit_to_device = goya_load_boot_fit_to_device,
5505 	.get_signal_cb_size = goya_get_signal_cb_size,
5506 	.get_wait_cb_size = goya_get_wait_cb_size,
5507 	.gen_signal_cb = goya_gen_signal_cb,
5508 	.gen_wait_cb = goya_gen_wait_cb,
5509 	.reset_sob = goya_reset_sob,
5510 	.reset_sob_group = goya_reset_sob_group,
5511 	.get_device_time = goya_get_device_time,
5512 	.pb_print_security_errors = NULL,
5513 	.collective_wait_init_cs = goya_collective_wait_init_cs,
5514 	.collective_wait_create_jobs = goya_collective_wait_create_jobs,
5515 	.get_dec_base_addr = NULL,
5516 	.scramble_addr = hl_mmu_scramble_addr,
5517 	.descramble_addr = hl_mmu_descramble_addr,
5518 	.ack_protection_bits_errors = goya_ack_protection_bits_errors,
5519 	.get_hw_block_id = goya_get_hw_block_id,
5520 	.hw_block_mmap = goya_block_mmap,
5521 	.enable_events_from_fw = goya_enable_events_from_fw,
5522 	.ack_mmu_errors = goya_ack_mmu_page_fault_or_access_error,
5523 	.map_pll_idx_to_fw_idx = goya_map_pll_idx_to_fw_idx,
5524 	.init_firmware_preload_params = goya_init_firmware_preload_params,
5525 	.init_firmware_loader = goya_init_firmware_loader,
5526 	.init_cpu_scrambler_dram = goya_cpu_init_scrambler_dram,
5527 	.state_dump_init = goya_state_dump_init,
5528 	.get_sob_addr = &goya_get_sob_addr,
5529 	.set_pci_memory_regions = goya_set_pci_memory_regions,
5530 	.get_stream_master_qid_arr = goya_get_stream_master_qid_arr,
5531 	.check_if_razwi_happened = goya_check_if_razwi_happened,
5532 	.mmu_get_real_page_size = hl_mmu_get_real_page_size,
5533 	.access_dev_mem = hl_access_dev_mem,
5534 	.set_dram_bar_base = goya_set_ddr_bar_base,
5535 	.send_device_activity = goya_send_device_activity,
5536 	.set_dram_properties = goya_set_dram_properties,
5537 	.set_binning_masks = goya_set_binning_masks,
5538 };
5539 
5540 /*
5541  * goya_set_asic_funcs - set Goya function pointers
5542  *
5543  * @*hdev: pointer to hl_device structure
5544  *
5545  */
5546 void goya_set_asic_funcs(struct hl_device *hdev)
5547 {
5548 	hdev->asic_funcs = &goya_funcs;
5549 }
5550