xref: /rk3399_ARM-atf/plat/nvidia/tegra/soc/t194/plat_ras.c (revision 8ca61538a0fe3aed6764a012317cbf61f09ebb61)
1 /*
2  * Copyright (c) 2020, NVIDIA Corporation. All rights reserved.
3  *
4  * SPDX-License-Identifier: BSD-3-Clause
5  */
6 
7 #include <stdbool.h>
8 #include <stdint.h>
9 
10 #include <common/debug.h>
11 #include <lib/bakery_lock.h>
12 #include <lib/extensions/ras.h>
13 #include <lib/utils_def.h>
14 #include <services/sdei.h>
15 
16 #include <plat/common/platform.h>
17 #include <platform_def.h>
18 #include <tegra194_ras_private.h>
19 #include <tegra_def.h>
20 #include <tegra_platform.h>
21 #include <tegra_private.h>
22 
23 /*
24  * ERR<n>FR bits[63:32], it indicates supported RAS errors which can be enabled
25  * by setting corresponding bits in ERR<n>CTLR
26  */
27 #define ERR_FR_EN_BITS_MASK	0xFFFFFFFF00000000ULL
28 
29 /* bakery lock for platform RAS handler. */
30 static DEFINE_BAKERY_LOCK(ras_handler_lock);
31 #define ras_lock()		bakery_lock_get(&ras_handler_lock)
32 #define ras_unlock()		bakery_lock_release(&ras_handler_lock)
33 
34 /*
35  * Function to handle an External Abort received at EL3.
36  * This function is invoked by RAS framework.
37  */
38 static void tegra194_ea_handler(unsigned int ea_reason, uint64_t syndrome,
39 		void *cookie, void *handle, uint64_t flags)
40 {
41 	int32_t ret;
42 
43 	ras_lock();
44 
45 	ERROR("exception reason=%u syndrome=0x%llx on 0x%lx at EL3.\n",
46 		ea_reason, syndrome, read_mpidr_el1());
47 
48 	/* Call RAS EA handler */
49 	ret = ras_ea_handler(ea_reason, syndrome, cookie, handle, flags);
50 	if (ret != 0) {
51 		ERROR("RAS error handled!\n");
52 		ret = sdei_dispatch_event(TEGRA_SDEI_EP_EVENT_0 +
53 				plat_my_core_pos());
54 		if (ret != 0)
55 			ERROR("sdei_dispatch_event returned %d\n", ret);
56 	} else {
57 		ERROR("Not a RAS error!\n");
58 	}
59 
60 	ras_unlock();
61 }
62 
63 /* Function to enable uncorrectable errors as External abort (SError) */
64 void tegra194_ras_enable(void)
65 {
66 	VERBOSE("%s\n", __func__);
67 
68 	/* skip RAS enablement if not a silicon platform. */
69 	if (!tegra_platform_is_silicon()) {
70 		return;
71 	}
72 
73 	/*
74 	 * Iterate for each group(num_idx ERRSELRs starting from idx_start)
75 	 * use normal for loop instead of for_each_err_record_info to get rid
76 	 * of MISRA noise..
77 	 */
78 	for (uint32_t i = 0U; i < err_record_mappings.num_err_records; i++) {
79 
80 		const struct err_record_info *info = &err_record_mappings.err_records[i];
81 
82 		uint32_t idx_start = info->sysreg.idx_start;
83 		uint32_t num_idx = info->sysreg.num_idx;
84 		const struct ras_aux_data *aux_data = (const struct ras_aux_data *)info->aux_data;
85 
86 		assert(aux_data != NULL);
87 
88 		for (uint32_t j = 0; j < num_idx; j++) {
89 			uint64_t err_ctrl = 0ULL;
90 
91 			/* enable SError reporting for uncorrectable error */
92 			ERR_CTLR_ENABLE_FIELD(err_ctrl, UE);
93 			ERR_CTLR_ENABLE_FIELD(err_ctrl, ED);
94 
95 			/*
96 			 * Catch error if something wrong with the RAS aux data
97 			 * record table.
98 			 */
99 			assert(aux_data[j].err_ctrl != NULL);
100 
101 			/* enable the specified errors */
102 			err_ctrl |= aux_data[j].err_ctrl();
103 
104 			/* Write to ERRSELR_EL1 to select the error record */
105 			ser_sys_select_record(idx_start + j);
106 
107 			/* enable specified errors */
108 			write_erxctlr_el1(err_ctrl);
109 
110 			/*
111 			 * Check if all the bit settings have been enabled to detect
112 			 * uncorrected/corrected errors, if not assert.
113 			 */
114 			assert(read_erxctlr_el1() == err_ctrl);
115 		}
116 	}
117 }
118 
119 /* Function to probe an error from error record group. */
120 static int32_t tegra194_ras_record_probe(const struct err_record_info *info,
121 		int *probe_data)
122 {
123 	/* Skip probing if not a silicon platform */
124 	if (!tegra_platform_is_silicon()) {
125 		return 0;
126 	}
127 
128 	return ser_probe_sysreg(info->sysreg.idx_start, info->sysreg.num_idx, probe_data);
129 }
130 
131 /* Function to handle error from one given node */
132 static int32_t tegra194_ras_node_handler(const struct ras_error *errors, uint64_t status)
133 {
134 	bool found = false;
135 	uint32_t ierr = (uint32_t)ERR_STATUS_GET_FIELD(status, IERR);
136 	uint32_t serr = (uint32_t)ERR_STATUS_GET_FIELD(status, SERR);
137 
138 	/* IERR to error message */
139 	for (uint32_t i = 0; errors[i].error_msg != NULL; i++) {
140 		if (ierr == errors[i].error_code) {
141 			ERROR("IERR = %s(0x%x)\n",
142 				errors[i].error_msg, errors[i].error_code);
143 			found = true;
144 			break;
145 		}
146 	}
147 	if (!found) {
148 		ERROR("unknown IERR: 0x%x\n", ierr);
149 	}
150 
151 	ERROR("SERR = %s(0x%x)\n", ras_serr_to_str(serr), serr);
152 
153 	/* Write to clear reported errors. */
154 	write_erxstatus_el1(status);
155 
156 	return 0;
157 }
158 
159 /* Function to handle one error node from an error record group. */
160 static int32_t tegra194_ras_record_handler(const struct err_record_info *info,
161 		int probe_data, const struct err_handler_data *const data)
162 {
163 	uint32_t num_idx = info->sysreg.num_idx;
164 	uint32_t idx_start = info->sysreg.idx_start;
165 	const struct ras_aux_data *aux_data = info->aux_data;
166 
167 	uint64_t status = 0ULL;
168 
169 	VERBOSE("%s\n", __func__);
170 
171 	assert(probe_data >= 0);
172 	assert((uint32_t)probe_data < num_idx);
173 
174 	uint32_t offset = (uint32_t)probe_data;
175 	const struct  ras_error *errors = aux_data[offset].error_records;
176 
177 	assert(errors != NULL);
178 
179 	/* Write to ERRSELR_EL1 to select the error record */
180 	ser_sys_select_record(idx_start + offset);
181 
182 	/* Retrieve status register from the error record */
183 	status = read_erxstatus_el1();
184 
185 	assert(ERR_STATUS_GET_FIELD(status, V) != 0U);
186 	assert(ERR_STATUS_GET_FIELD(status, UE) != 0U);
187 
188 	return tegra194_ras_node_handler(errors, status);
189 }
190 
191 
192 /* Instantiate RAS nodes */
193 PER_CORE_RAS_NODE_LIST(DEFINE_ONE_RAS_NODE)
194 PER_CLUSTER_RAS_NODE_LIST(DEFINE_ONE_RAS_NODE)
195 SCF_L3_BANK_RAS_NODE_LIST(DEFINE_ONE_RAS_NODE)
196 CCPLEX_RAS_NODE_LIST(DEFINE_ONE_RAS_NODE)
197 
198 /* Instantiate RAS node groups */
199 static struct ras_aux_data per_core_ras_group[] = {
200 	PER_CORE_RAS_GROUP_NODES
201 };
202 
203 static struct ras_aux_data per_cluster_ras_group[] = {
204 	PER_CLUSTER_RAS_GROUP_NODES
205 };
206 
207 static struct ras_aux_data scf_l3_ras_group[] = {
208 	SCF_L3_BANK_RAS_GROUP_NODES
209 };
210 
211 static struct ras_aux_data ccplex_ras_group[] = {
212     CCPLEX_RAS_GROUP_NODES
213 };
214 
215 /*
216  * We have same probe and handler for each error record group, use a macro to
217  * simply the record definition.
218  */
219 #define ADD_ONE_ERR_GROUP(errselr_start, group) \
220 	ERR_RECORD_SYSREG_V1((errselr_start), (uint32_t)ARRAY_SIZE((group)), \
221 			&tegra194_ras_record_probe, \
222 			&tegra194_ras_record_handler, (group))
223 
224 /* RAS error record group information */
225 static struct err_record_info carmel_ras_records[] = {
226 	/*
227 	 * Per core ras error records
228 	 * ERRSELR starts from 0*256 + Logical_CPU_ID*16 + 0 to
229 	 * 0*256 + Logical_CPU_ID*16 + 5 for each group.
230 	 * 8 cores/groups, 6 * 8 nodes in total.
231 	 */
232 	ADD_ONE_ERR_GROUP(0x000, per_core_ras_group),
233 	ADD_ONE_ERR_GROUP(0x010, per_core_ras_group),
234 	ADD_ONE_ERR_GROUP(0x020, per_core_ras_group),
235 	ADD_ONE_ERR_GROUP(0x030, per_core_ras_group),
236 	ADD_ONE_ERR_GROUP(0x040, per_core_ras_group),
237 	ADD_ONE_ERR_GROUP(0x050, per_core_ras_group),
238 	ADD_ONE_ERR_GROUP(0x060, per_core_ras_group),
239 	ADD_ONE_ERR_GROUP(0x070, per_core_ras_group),
240 
241 	/*
242 	 * Per cluster ras error records
243 	 * ERRSELR starts from 2*256 + Logical_Cluster_ID*16 + 0 to
244 	 * 2*256 + Logical_Cluster_ID*16 + 3.
245 	 * 4 clusters/groups, 3 * 4 nodes in total.
246 	 */
247 	ADD_ONE_ERR_GROUP(0x200, per_cluster_ras_group),
248 	ADD_ONE_ERR_GROUP(0x210, per_cluster_ras_group),
249 	ADD_ONE_ERR_GROUP(0x220, per_cluster_ras_group),
250 	ADD_ONE_ERR_GROUP(0x230, per_cluster_ras_group),
251 
252 	/*
253 	 * SCF L3_Bank ras error records
254 	 * ERRSELR: 3*256 + L3_Bank_ID, L3_Bank_ID: 0-3
255 	 * 1 groups, 4 nodes in total.
256 	 */
257 	ADD_ONE_ERR_GROUP(0x300, scf_l3_ras_group),
258 
259 	/*
260 	 * CCPLEX ras error records
261 	 * ERRSELR: 4*256 + Unit_ID, Unit_ID: 0 - 4
262 	 * 1 groups, 5 nodes in total.
263 	 */
264 	ADD_ONE_ERR_GROUP(0x400, ccplex_ras_group),
265 };
266 
267 REGISTER_ERR_RECORD_INFO(carmel_ras_records);
268 
269 /* dummy RAS interrupt */
270 static struct ras_interrupt carmel_ras_interrupts[] = {};
271 REGISTER_RAS_INTERRUPTS(carmel_ras_interrupts);
272 
273 /*******************************************************************************
274  * RAS handler for the platform
275  ******************************************************************************/
276 void plat_ea_handler(unsigned int ea_reason, uint64_t syndrome, void *cookie,
277 		void *handle, uint64_t flags)
278 {
279 #if RAS_EXTENSION
280 	tegra194_ea_handler(ea_reason, syndrome, cookie, handle, flags);
281 #else
282 	ERROR("Unhandled External Abort received on 0x%llx at EL3!\n",
283 			read_mpidr_el1());
284 	ERROR(" exception reason=%u syndrome=0x%lx\n", ea_reason, syndrome);
285 	panic();
286 #endif
287 }
288