xref: /rk3399_ARM-atf/plat/nvidia/tegra/soc/t194/plat_ras.c (revision 0aa9f3c0f2f2ff675c3c12ae5ac6ceb475d6a16f)
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("MPIDR 0x%lx: exception reason=%u syndrome=0x%llx\n",
46 		read_mpidr(), ea_reason, syndrome);
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 /*
64  * Function to enable all supported RAS error report.
65  *
66  * Uncorrected errors are set to report as External abort (SError)
67  * Corrected errors are set to report as interrupt.
68  */
69 void tegra194_ras_enable(void)
70 {
71 	VERBOSE("%s\n", __func__);
72 
73 	/* skip RAS enablement if not a silicon platform. */
74 	if (!tegra_platform_is_silicon()) {
75 		return;
76 	}
77 
78 	/*
79 	 * Iterate for each group(num_idx ERRSELRs starting from idx_start)
80 	 * use normal for loop instead of for_each_err_record_info to get rid
81 	 * of MISRA noise..
82 	 */
83 	for (uint32_t i = 0U; i < err_record_mappings.num_err_records; i++) {
84 
85 		const struct err_record_info *info = &err_record_mappings.err_records[i];
86 
87 		uint32_t idx_start = info->sysreg.idx_start;
88 		uint32_t num_idx = info->sysreg.num_idx;
89 		const struct ras_aux_data *aux_data = (const struct ras_aux_data *)info->aux_data;
90 
91 		assert(aux_data != NULL);
92 
93 		for (uint32_t j = 0; j < num_idx; j++) {
94 
95 			/* ERR<n>CTLR register value. */
96 			uint64_t err_ctrl = 0ULL;
97 			/* all supported errors for this node. */
98 			uint64_t err_fr;
99 			/* uncorrectable errors */
100 			uint64_t uncorr_errs;
101 			/* correctable errors */
102 			uint64_t corr_errs;
103 
104 			/*
105 			 * Catch error if something wrong with the RAS aux data
106 			 * record table.
107 			 */
108 			assert(aux_data[j].err_ctrl != NULL);
109 
110 			/*
111 			 * Write to ERRSELR_EL1 to select the RAS error node.
112 			 * Always program this at first to select corresponding
113 			 * RAS node before any other RAS register r/w.
114 			 */
115 			ser_sys_select_record(idx_start + j);
116 
117 			err_fr = read_erxfr_el1() & ERR_FR_EN_BITS_MASK;
118 			uncorr_errs = aux_data[j].err_ctrl();
119 			corr_errs = ~uncorr_errs & err_fr;
120 
121 			/* enable error reporting */
122 			ERR_CTLR_ENABLE_FIELD(err_ctrl, ED);
123 
124 			/* enable SError reporting for uncorrectable errors */
125 			if ((uncorr_errs & err_fr) != 0ULL) {
126 				ERR_CTLR_ENABLE_FIELD(err_ctrl, UE);
127 			}
128 
129 			/* generate interrupt for corrected errors. */
130 			if (corr_errs != 0ULL) {
131 				ERR_CTLR_ENABLE_FIELD(err_ctrl, CFI);
132 			}
133 
134 			/* enable the supported errors */
135 			err_ctrl |= err_fr;
136 
137 			VERBOSE("errselr_el1:0x%x, erxfr:0x%llx, err_ctrl:0x%llx\n",
138 				idx_start + j, err_fr, err_ctrl);
139 
140 			/* enable specified errors, or set to 0 if no supported error */
141 			write_erxctlr_el1(err_ctrl);
142 
143 			/*
144 			 * Check if all the bit settings have been enabled to detect
145 			 * uncorrected/corrected errors, if not assert.
146 			 */
147 			assert(read_erxctlr_el1() == err_ctrl);
148 		}
149 	}
150 }
151 
152 /*
153  * Function to clear RAS ERR<n>STATUS for corrected RAS error.
154  * This function ignores any new RAS error signaled during clearing; it is not
155  * multi-core safe(no ras_lock is taken to reduce overhead).
156  */
157 void tegra194_ras_corrected_err_clear(void)
158 {
159 	uint64_t clear_ce_status = 0ULL;
160 
161 	ERR_STATUS_SET_FIELD(clear_ce_status, AV, 0x1UL);
162 	ERR_STATUS_SET_FIELD(clear_ce_status, V, 0x1UL);
163 	ERR_STATUS_SET_FIELD(clear_ce_status, OF, 0x1UL);
164 	ERR_STATUS_SET_FIELD(clear_ce_status, MV, 0x1UL);
165 	ERR_STATUS_SET_FIELD(clear_ce_status, CE, 0x3UL);
166 
167 	for (uint32_t i = 0U; i < err_record_mappings.num_err_records; i++) {
168 
169 		const struct err_record_info *info = &err_record_mappings.err_records[i];
170 		uint32_t idx_start = info->sysreg.idx_start;
171 		uint32_t num_idx = info->sysreg.num_idx;
172 
173 		for (uint32_t j = 0U; j < num_idx; j++) {
174 
175 			uint64_t status;
176 			uint32_t err_idx = idx_start + j;
177 
178 			write_errselr_el1(err_idx);
179 			status = read_erxstatus_el1();
180 
181 			if (ERR_STATUS_GET_FIELD(status, CE) != 0U) {
182 				write_erxstatus_el1(clear_ce_status);
183 			}
184 		}
185 	}
186 }
187 
188 /* Function to probe an error from error record group. */
189 static int32_t tegra194_ras_record_probe(const struct err_record_info *info,
190 		int *probe_data)
191 {
192 	/* Skip probing if not a silicon platform */
193 	if (!tegra_platform_is_silicon()) {
194 		return 0;
195 	}
196 
197 	return ser_probe_sysreg(info->sysreg.idx_start, info->sysreg.num_idx, probe_data);
198 }
199 
200 /* Function to handle error from one given node */
201 static int32_t tegra194_ras_node_handler(uint32_t errselr, const char *name,
202 		const struct ras_error *errors, uint64_t status)
203 {
204 	bool found = false;
205 	uint32_t ierr = (uint32_t)ERR_STATUS_GET_FIELD(status, IERR);
206 	uint32_t serr = (uint32_t)ERR_STATUS_GET_FIELD(status, SERR);
207 	uint64_t val = 0;
208 
209 	/* not a valid error. */
210 	if (ERR_STATUS_GET_FIELD(status, V) == 0U) {
211 		return 0;
212 	}
213 
214 	ERR_STATUS_SET_FIELD(val, V, 1);
215 
216 	/* keep the log print same as linux arm64_ras driver. */
217 	ERROR("**************************************\n");
218 	ERROR("RAS Error in %s, ERRSELR_EL1=0x%x:\n", name, errselr);
219 	ERROR("\tStatus = 0x%llx\n", status);
220 
221 	/* Print uncorrectable errror information. */
222 	if (ERR_STATUS_GET_FIELD(status, UE) != 0U) {
223 
224 		ERR_STATUS_SET_FIELD(val, UE, 1);
225 		ERR_STATUS_SET_FIELD(val, UET, 1);
226 
227 		/* IERR to error message */
228 		for (uint32_t i = 0; errors[i].error_msg != NULL; i++) {
229 			if (ierr == errors[i].error_code) {
230 				ERROR("\tIERR = %s: 0x%x\n",
231 					errors[i].error_msg, ierr);
232 
233 				found = true;
234 				break;
235 			}
236 		}
237 
238 		if (!found) {
239 			ERROR("\tUnknown IERR: 0x%x\n", ierr);
240 		}
241 
242 		ERROR("SERR = %s: 0x%x\n", ras_serr_to_str(serr), serr);
243 
244 		/* Overflow, multiple errors have been detected. */
245 		if (ERR_STATUS_GET_FIELD(status, OF) != 0U) {
246 			ERROR("\tOverflow (there may be more errors) - "
247 				"Uncorrectable\n");
248 			ERR_STATUS_SET_FIELD(val, OF, 1);
249 		}
250 
251 		ERROR("\tUncorrectable (this is fatal)\n");
252 
253 		/* Miscellaneous Register Valid. */
254 		if (ERR_STATUS_GET_FIELD(status, MV) != 0U) {
255 			ERROR("\tMISC0 = 0x%lx\n", read_erxmisc0_el1());
256 			ERROR("\tMISC1 = 0x%lx\n", read_erxmisc1_el1());
257 			ERR_STATUS_SET_FIELD(val, MV, 1);
258 		}
259 
260 		/* Address Valid. */
261 		if (ERR_STATUS_GET_FIELD(status, AV) != 0U) {
262 			ERROR("\tADDR = 0x%lx\n", read_erxaddr_el1());
263 			ERR_STATUS_SET_FIELD(val, AV, 1);
264 		}
265 
266 		/* Deferred error */
267 		if (ERR_STATUS_GET_FIELD(status, DE) != 0U) {
268 			ERROR("\tDeferred error\n");
269 			ERR_STATUS_SET_FIELD(val, DE, 1);
270 		}
271 
272 	} else {
273 		/* For corrected error, simply clear it. */
274 		VERBOSE("corrected RAS error is cleared: ERRSELR_EL1:0x%x, "
275 			"IERR:0x%x, SERR:0x%x\n", errselr, ierr, serr);
276 		ERR_STATUS_SET_FIELD(val, CE, 1);
277 	}
278 
279 	ERROR("**************************************\n");
280 
281 	/* Write to clear reported errors. */
282 	write_erxstatus_el1(val);
283 
284 	/* error handled */
285 	return 0;
286 }
287 
288 /* Function to handle one error node from an error record group. */
289 static int32_t tegra194_ras_record_handler(const struct err_record_info *info,
290 		int probe_data, const struct err_handler_data *const data __unused)
291 {
292 	uint32_t num_idx = info->sysreg.num_idx;
293 	uint32_t idx_start = info->sysreg.idx_start;
294 	const struct ras_aux_data *aux_data = info->aux_data;
295 	const struct ras_error *errors;
296 	uint32_t offset;
297 	const char *node_name;
298 
299 	uint64_t status = 0ULL;
300 
301 	VERBOSE("%s\n", __func__);
302 
303 	assert(probe_data >= 0);
304 	assert((uint32_t)probe_data < num_idx);
305 
306 	offset = (uint32_t)probe_data;
307 	errors = aux_data[offset].error_records;
308 	node_name = aux_data[offset].name;
309 
310 	assert(errors != NULL);
311 
312 	/* Write to ERRSELR_EL1 to select the error record */
313 	ser_sys_select_record(idx_start + offset);
314 
315 	/* Retrieve status register from the error record */
316 	status = read_erxstatus_el1();
317 
318 	return tegra194_ras_node_handler(idx_start + offset, node_name,
319 			errors, status);
320 }
321 
322 
323 /* Instantiate RAS nodes */
324 PER_CORE_RAS_NODE_LIST(DEFINE_ONE_RAS_NODE)
325 PER_CLUSTER_RAS_NODE_LIST(DEFINE_ONE_RAS_NODE)
326 SCF_L3_BANK_RAS_NODE_LIST(DEFINE_ONE_RAS_NODE)
327 CCPLEX_RAS_NODE_LIST(DEFINE_ONE_RAS_NODE)
328 
329 /* Instantiate RAS node groups */
330 static struct ras_aux_data per_core_ras_group[] = {
331 	PER_CORE_RAS_GROUP_NODES
332 };
333 
334 static struct ras_aux_data per_cluster_ras_group[] = {
335 	PER_CLUSTER_RAS_GROUP_NODES
336 };
337 
338 static struct ras_aux_data scf_l3_ras_group[] = {
339 	SCF_L3_BANK_RAS_GROUP_NODES
340 };
341 
342 static struct ras_aux_data ccplex_ras_group[] = {
343     CCPLEX_RAS_GROUP_NODES
344 };
345 
346 /*
347  * We have same probe and handler for each error record group, use a macro to
348  * simply the record definition.
349  */
350 #define ADD_ONE_ERR_GROUP(errselr_start, group) \
351 	ERR_RECORD_SYSREG_V1((errselr_start), (uint32_t)ARRAY_SIZE((group)), \
352 			&tegra194_ras_record_probe, \
353 			&tegra194_ras_record_handler, (group))
354 
355 /* RAS error record group information */
356 static struct err_record_info carmel_ras_records[] = {
357 	/*
358 	 * Per core ras error records
359 	 * ERRSELR starts from 0*256 + Logical_CPU_ID*16 + 0 to
360 	 * 0*256 + Logical_CPU_ID*16 + 5 for each group.
361 	 * 8 cores/groups, 6 * 8 nodes in total.
362 	 */
363 	ADD_ONE_ERR_GROUP(0x000, per_core_ras_group),
364 	ADD_ONE_ERR_GROUP(0x010, per_core_ras_group),
365 	ADD_ONE_ERR_GROUP(0x020, per_core_ras_group),
366 	ADD_ONE_ERR_GROUP(0x030, per_core_ras_group),
367 	ADD_ONE_ERR_GROUP(0x040, per_core_ras_group),
368 	ADD_ONE_ERR_GROUP(0x050, per_core_ras_group),
369 	ADD_ONE_ERR_GROUP(0x060, per_core_ras_group),
370 	ADD_ONE_ERR_GROUP(0x070, per_core_ras_group),
371 
372 	/*
373 	 * Per cluster ras error records
374 	 * ERRSELR starts from 2*256 + Logical_Cluster_ID*16 + 0 to
375 	 * 2*256 + Logical_Cluster_ID*16 + 3.
376 	 * 4 clusters/groups, 3 * 4 nodes in total.
377 	 */
378 	ADD_ONE_ERR_GROUP(0x200, per_cluster_ras_group),
379 	ADD_ONE_ERR_GROUP(0x210, per_cluster_ras_group),
380 	ADD_ONE_ERR_GROUP(0x220, per_cluster_ras_group),
381 	ADD_ONE_ERR_GROUP(0x230, per_cluster_ras_group),
382 
383 	/*
384 	 * SCF L3_Bank ras error records
385 	 * ERRSELR: 3*256 + L3_Bank_ID, L3_Bank_ID: 0-3
386 	 * 1 groups, 4 nodes in total.
387 	 */
388 	ADD_ONE_ERR_GROUP(0x300, scf_l3_ras_group),
389 
390 	/*
391 	 * CCPLEX ras error records
392 	 * ERRSELR: 4*256 + Unit_ID, Unit_ID: 0 - 4
393 	 * 1 groups, 5 nodes in total.
394 	 */
395 	ADD_ONE_ERR_GROUP(0x400, ccplex_ras_group),
396 };
397 
398 REGISTER_ERR_RECORD_INFO(carmel_ras_records);
399 
400 /* dummy RAS interrupt */
401 static struct ras_interrupt carmel_ras_interrupts[] = {};
402 REGISTER_RAS_INTERRUPTS(carmel_ras_interrupts);
403 
404 /*******************************************************************************
405  * RAS handler for the platform
406  ******************************************************************************/
407 void plat_ea_handler(unsigned int ea_reason, uint64_t syndrome, void *cookie,
408 		void *handle, uint64_t flags)
409 {
410 #if RAS_EXTENSION
411 	tegra194_ea_handler(ea_reason, syndrome, cookie, handle, flags);
412 #else
413 	ERROR("Unhandled External Abort received on 0x%llx at EL3!\n",
414 			read_mpidr_el1());
415 	ERROR(" exception reason=%u syndrome=0x%lx\n", ea_reason, syndrome);
416 	panic();
417 #endif
418 }
419