xref: /linux/arch/x86/kernel/cpu/mce/amd.c (revision c39f2d9db0fd81ea20bb5cce9b3f082ca63753e2)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  (c) 2005-2016 Advanced Micro Devices, Inc.
4  *
5  *  Written by Jacob Shin - AMD, Inc.
6  *  Maintained by: Borislav Petkov <bp@alien8.de>
7  *
8  *  All MC4_MISCi registers are shared between cores on a node.
9  */
10 #include <linux/interrupt.h>
11 #include <linux/notifier.h>
12 #include <linux/kobject.h>
13 #include <linux/percpu.h>
14 #include <linux/errno.h>
15 #include <linux/sched.h>
16 #include <linux/sysfs.h>
17 #include <linux/slab.h>
18 #include <linux/init.h>
19 #include <linux/cpu.h>
20 #include <linux/smp.h>
21 #include <linux/string.h>
22 
23 #include <asm/amd_nb.h>
24 #include <asm/traps.h>
25 #include <asm/apic.h>
26 #include <asm/mce.h>
27 #include <asm/msr.h>
28 #include <asm/trace/irq_vectors.h>
29 
30 #include "internal.h"
31 
32 #define NR_BLOCKS         5
33 #define THRESHOLD_MAX     0xFFF
34 #define INT_TYPE_APIC     0x00020000
35 #define MASK_VALID_HI     0x80000000
36 #define MASK_CNTP_HI      0x40000000
37 #define MASK_LOCKED_HI    0x20000000
38 #define MASK_LVTOFF_HI    0x00F00000
39 #define MASK_COUNT_EN_HI  0x00080000
40 #define MASK_INT_TYPE_HI  0x00060000
41 #define MASK_OVERFLOW_HI  0x00010000
42 #define MASK_ERR_COUNT_HI 0x00000FFF
43 #define MASK_BLKPTR_LO    0xFF000000
44 #define MCG_XBLK_ADDR     0xC0000400
45 
46 /* Deferred error settings */
47 #define MSR_CU_DEF_ERR		0xC0000410
48 #define MASK_DEF_LVTOFF		0x000000F0
49 #define MASK_DEF_INT_TYPE	0x00000006
50 #define DEF_LVT_OFF		0x2
51 #define DEF_INT_TYPE_APIC	0x2
52 
53 /* Scalable MCA: */
54 
55 /* Threshold LVT offset is at MSR0xC0000410[15:12] */
56 #define SMCA_THR_LVT_OFF	0xF000
57 
58 static bool thresholding_irq_en;
59 
60 static const char * const th_names[] = {
61 	"load_store",
62 	"insn_fetch",
63 	"combined_unit",
64 	"decode_unit",
65 	"northbridge",
66 	"execution_unit",
67 };
68 
69 static const char * const smca_umc_block_names[] = {
70 	"dram_ecc",
71 	"misc_umc"
72 };
73 
74 struct smca_bank_name {
75 	const char *name;	/* Short name for sysfs */
76 	const char *long_name;	/* Long name for pretty-printing */
77 };
78 
79 static struct smca_bank_name smca_names[] = {
80 	[SMCA_LS]	= { "load_store",	"Load Store Unit" },
81 	[SMCA_IF]	= { "insn_fetch",	"Instruction Fetch Unit" },
82 	[SMCA_L2_CACHE]	= { "l2_cache",		"L2 Cache" },
83 	[SMCA_DE]	= { "decode_unit",	"Decode Unit" },
84 	[SMCA_RESERVED]	= { "reserved",		"Reserved" },
85 	[SMCA_EX]	= { "execution_unit",	"Execution Unit" },
86 	[SMCA_FP]	= { "floating_point",	"Floating Point Unit" },
87 	[SMCA_L3_CACHE]	= { "l3_cache",		"L3 Cache" },
88 	[SMCA_CS]	= { "coherent_slave",	"Coherent Slave" },
89 	[SMCA_CS_V2]	= { "coherent_slave",	"Coherent Slave" },
90 	[SMCA_PIE]	= { "pie",		"Power, Interrupts, etc." },
91 	[SMCA_UMC]	= { "umc",		"Unified Memory Controller" },
92 	[SMCA_PB]	= { "param_block",	"Parameter Block" },
93 	[SMCA_PSP]	= { "psp",		"Platform Security Processor" },
94 	[SMCA_PSP_V2]	= { "psp",		"Platform Security Processor" },
95 	[SMCA_SMU]	= { "smu",		"System Management Unit" },
96 	[SMCA_SMU_V2]	= { "smu",		"System Management Unit" },
97 	[SMCA_MP5]	= { "mp5",		"Microprocessor 5 Unit" },
98 	[SMCA_NBIO]	= { "nbio",		"Northbridge IO Unit" },
99 	[SMCA_PCIE]	= { "pcie",		"PCI Express Unit" },
100 };
101 
102 static u32 smca_bank_addrs[MAX_NR_BANKS][NR_BLOCKS] __ro_after_init =
103 {
104 	[0 ... MAX_NR_BANKS - 1] = { [0 ... NR_BLOCKS - 1] = -1 }
105 };
106 
107 static const char *smca_get_name(enum smca_bank_types t)
108 {
109 	if (t >= N_SMCA_BANK_TYPES)
110 		return NULL;
111 
112 	return smca_names[t].name;
113 }
114 
115 const char *smca_get_long_name(enum smca_bank_types t)
116 {
117 	if (t >= N_SMCA_BANK_TYPES)
118 		return NULL;
119 
120 	return smca_names[t].long_name;
121 }
122 EXPORT_SYMBOL_GPL(smca_get_long_name);
123 
124 static enum smca_bank_types smca_get_bank_type(unsigned int bank)
125 {
126 	struct smca_bank *b;
127 
128 	if (bank >= MAX_NR_BANKS)
129 		return N_SMCA_BANK_TYPES;
130 
131 	b = &smca_banks[bank];
132 	if (!b->hwid)
133 		return N_SMCA_BANK_TYPES;
134 
135 	return b->hwid->bank_type;
136 }
137 
138 static struct smca_hwid smca_hwid_mcatypes[] = {
139 	/* { bank_type, hwid_mcatype, xec_bitmap } */
140 
141 	/* Reserved type */
142 	{ SMCA_RESERVED, HWID_MCATYPE(0x00, 0x0), 0x0 },
143 
144 	/* ZN Core (HWID=0xB0) MCA types */
145 	{ SMCA_LS,	 HWID_MCATYPE(0xB0, 0x0), 0x1FFFFF },
146 	{ SMCA_IF,	 HWID_MCATYPE(0xB0, 0x1), 0x3FFF },
147 	{ SMCA_L2_CACHE, HWID_MCATYPE(0xB0, 0x2), 0xF },
148 	{ SMCA_DE,	 HWID_MCATYPE(0xB0, 0x3), 0x1FF },
149 	/* HWID 0xB0 MCATYPE 0x4 is Reserved */
150 	{ SMCA_EX,	 HWID_MCATYPE(0xB0, 0x5), 0xFFF },
151 	{ SMCA_FP,	 HWID_MCATYPE(0xB0, 0x6), 0x7F },
152 	{ SMCA_L3_CACHE, HWID_MCATYPE(0xB0, 0x7), 0xFF },
153 
154 	/* Data Fabric MCA types */
155 	{ SMCA_CS,	 HWID_MCATYPE(0x2E, 0x0), 0x1FF },
156 	{ SMCA_PIE,	 HWID_MCATYPE(0x2E, 0x1), 0x1F },
157 	{ SMCA_CS_V2,	 HWID_MCATYPE(0x2E, 0x2), 0x3FFF },
158 
159 	/* Unified Memory Controller MCA type */
160 	{ SMCA_UMC,	 HWID_MCATYPE(0x96, 0x0), 0xFF },
161 
162 	/* Parameter Block MCA type */
163 	{ SMCA_PB,	 HWID_MCATYPE(0x05, 0x0), 0x1 },
164 
165 	/* Platform Security Processor MCA type */
166 	{ SMCA_PSP,	 HWID_MCATYPE(0xFF, 0x0), 0x1 },
167 	{ SMCA_PSP_V2,	 HWID_MCATYPE(0xFF, 0x1), 0x3FFFF },
168 
169 	/* System Management Unit MCA type */
170 	{ SMCA_SMU,	 HWID_MCATYPE(0x01, 0x0), 0x1 },
171 	{ SMCA_SMU_V2,	 HWID_MCATYPE(0x01, 0x1), 0x7FF },
172 
173 	/* Microprocessor 5 Unit MCA type */
174 	{ SMCA_MP5,	 HWID_MCATYPE(0x01, 0x2), 0x3FF },
175 
176 	/* Northbridge IO Unit MCA type */
177 	{ SMCA_NBIO,	 HWID_MCATYPE(0x18, 0x0), 0x1F },
178 
179 	/* PCI Express Unit MCA type */
180 	{ SMCA_PCIE,	 HWID_MCATYPE(0x46, 0x0), 0x1F },
181 };
182 
183 struct smca_bank smca_banks[MAX_NR_BANKS];
184 EXPORT_SYMBOL_GPL(smca_banks);
185 
186 /*
187  * In SMCA enabled processors, we can have multiple banks for a given IP type.
188  * So to define a unique name for each bank, we use a temp c-string to append
189  * the MCA_IPID[InstanceId] to type's name in get_name().
190  *
191  * InstanceId is 32 bits which is 8 characters. Make sure MAX_MCATYPE_NAME_LEN
192  * is greater than 8 plus 1 (for underscore) plus length of longest type name.
193  */
194 #define MAX_MCATYPE_NAME_LEN	30
195 static char buf_mcatype[MAX_MCATYPE_NAME_LEN];
196 
197 static DEFINE_PER_CPU(struct threshold_bank **, threshold_banks);
198 static DEFINE_PER_CPU(unsigned int, bank_map);	/* see which banks are on */
199 
200 static void amd_threshold_interrupt(void);
201 static void amd_deferred_error_interrupt(void);
202 
203 static void default_deferred_error_interrupt(void)
204 {
205 	pr_err("Unexpected deferred interrupt at vector %x\n", DEFERRED_ERROR_VECTOR);
206 }
207 void (*deferred_error_int_vector)(void) = default_deferred_error_interrupt;
208 
209 static void smca_configure(unsigned int bank, unsigned int cpu)
210 {
211 	unsigned int i, hwid_mcatype;
212 	struct smca_hwid *s_hwid;
213 	u32 high, low;
214 	u32 smca_config = MSR_AMD64_SMCA_MCx_CONFIG(bank);
215 
216 	/* Set appropriate bits in MCA_CONFIG */
217 	if (!rdmsr_safe(smca_config, &low, &high)) {
218 		/*
219 		 * OS is required to set the MCAX bit to acknowledge that it is
220 		 * now using the new MSR ranges and new registers under each
221 		 * bank. It also means that the OS will configure deferred
222 		 * errors in the new MCx_CONFIG register. If the bit is not set,
223 		 * uncorrectable errors will cause a system panic.
224 		 *
225 		 * MCA_CONFIG[MCAX] is bit 32 (0 in the high portion of the MSR.)
226 		 */
227 		high |= BIT(0);
228 
229 		/*
230 		 * SMCA sets the Deferred Error Interrupt type per bank.
231 		 *
232 		 * MCA_CONFIG[DeferredIntTypeSupported] is bit 5, and tells us
233 		 * if the DeferredIntType bit field is available.
234 		 *
235 		 * MCA_CONFIG[DeferredIntType] is bits [38:37] ([6:5] in the
236 		 * high portion of the MSR). OS should set this to 0x1 to enable
237 		 * APIC based interrupt. First, check that no interrupt has been
238 		 * set.
239 		 */
240 		if ((low & BIT(5)) && !((high >> 5) & 0x3))
241 			high |= BIT(5);
242 
243 		wrmsr(smca_config, low, high);
244 	}
245 
246 	/* Return early if this bank was already initialized. */
247 	if (smca_banks[bank].hwid)
248 		return;
249 
250 	if (rdmsr_safe_on_cpu(cpu, MSR_AMD64_SMCA_MCx_IPID(bank), &low, &high)) {
251 		pr_warn("Failed to read MCA_IPID for bank %d\n", bank);
252 		return;
253 	}
254 
255 	hwid_mcatype = HWID_MCATYPE(high & MCI_IPID_HWID,
256 				    (high & MCI_IPID_MCATYPE) >> 16);
257 
258 	for (i = 0; i < ARRAY_SIZE(smca_hwid_mcatypes); i++) {
259 		s_hwid = &smca_hwid_mcatypes[i];
260 		if (hwid_mcatype == s_hwid->hwid_mcatype) {
261 			smca_banks[bank].hwid = s_hwid;
262 			smca_banks[bank].id = low;
263 			smca_banks[bank].sysfs_id = s_hwid->count++;
264 			break;
265 		}
266 	}
267 }
268 
269 struct thresh_restart {
270 	struct threshold_block	*b;
271 	int			reset;
272 	int			set_lvt_off;
273 	int			lvt_off;
274 	u16			old_limit;
275 };
276 
277 static inline bool is_shared_bank(int bank)
278 {
279 	/*
280 	 * Scalable MCA provides for only one core to have access to the MSRs of
281 	 * a shared bank.
282 	 */
283 	if (mce_flags.smca)
284 		return false;
285 
286 	/* Bank 4 is for northbridge reporting and is thus shared */
287 	return (bank == 4);
288 }
289 
290 static const char *bank4_names(const struct threshold_block *b)
291 {
292 	switch (b->address) {
293 	/* MSR4_MISC0 */
294 	case 0x00000413:
295 		return "dram";
296 
297 	case 0xc0000408:
298 		return "ht_links";
299 
300 	case 0xc0000409:
301 		return "l3_cache";
302 
303 	default:
304 		WARN(1, "Funny MSR: 0x%08x\n", b->address);
305 		return "";
306 	}
307 };
308 
309 
310 static bool lvt_interrupt_supported(unsigned int bank, u32 msr_high_bits)
311 {
312 	/*
313 	 * bank 4 supports APIC LVT interrupts implicitly since forever.
314 	 */
315 	if (bank == 4)
316 		return true;
317 
318 	/*
319 	 * IntP: interrupt present; if this bit is set, the thresholding
320 	 * bank can generate APIC LVT interrupts
321 	 */
322 	return msr_high_bits & BIT(28);
323 }
324 
325 static int lvt_off_valid(struct threshold_block *b, int apic, u32 lo, u32 hi)
326 {
327 	int msr = (hi & MASK_LVTOFF_HI) >> 20;
328 
329 	if (apic < 0) {
330 		pr_err(FW_BUG "cpu %d, failed to setup threshold interrupt "
331 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n", b->cpu,
332 		       b->bank, b->block, b->address, hi, lo);
333 		return 0;
334 	}
335 
336 	if (apic != msr) {
337 		/*
338 		 * On SMCA CPUs, LVT offset is programmed at a different MSR, and
339 		 * the BIOS provides the value. The original field where LVT offset
340 		 * was set is reserved. Return early here:
341 		 */
342 		if (mce_flags.smca)
343 			return 0;
344 
345 		pr_err(FW_BUG "cpu %d, invalid threshold interrupt offset %d "
346 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n",
347 		       b->cpu, apic, b->bank, b->block, b->address, hi, lo);
348 		return 0;
349 	}
350 
351 	return 1;
352 };
353 
354 /* Reprogram MCx_MISC MSR behind this threshold bank. */
355 static void threshold_restart_bank(void *_tr)
356 {
357 	struct thresh_restart *tr = _tr;
358 	u32 hi, lo;
359 
360 	rdmsr(tr->b->address, lo, hi);
361 
362 	if (tr->b->threshold_limit < (hi & THRESHOLD_MAX))
363 		tr->reset = 1;	/* limit cannot be lower than err count */
364 
365 	if (tr->reset) {		/* reset err count and overflow bit */
366 		hi =
367 		    (hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
368 		    (THRESHOLD_MAX - tr->b->threshold_limit);
369 	} else if (tr->old_limit) {	/* change limit w/o reset */
370 		int new_count = (hi & THRESHOLD_MAX) +
371 		    (tr->old_limit - tr->b->threshold_limit);
372 
373 		hi = (hi & ~MASK_ERR_COUNT_HI) |
374 		    (new_count & THRESHOLD_MAX);
375 	}
376 
377 	/* clear IntType */
378 	hi &= ~MASK_INT_TYPE_HI;
379 
380 	if (!tr->b->interrupt_capable)
381 		goto done;
382 
383 	if (tr->set_lvt_off) {
384 		if (lvt_off_valid(tr->b, tr->lvt_off, lo, hi)) {
385 			/* set new lvt offset */
386 			hi &= ~MASK_LVTOFF_HI;
387 			hi |= tr->lvt_off << 20;
388 		}
389 	}
390 
391 	if (tr->b->interrupt_enable)
392 		hi |= INT_TYPE_APIC;
393 
394  done:
395 
396 	hi |= MASK_COUNT_EN_HI;
397 	wrmsr(tr->b->address, lo, hi);
398 }
399 
400 static void mce_threshold_block_init(struct threshold_block *b, int offset)
401 {
402 	struct thresh_restart tr = {
403 		.b			= b,
404 		.set_lvt_off		= 1,
405 		.lvt_off		= offset,
406 	};
407 
408 	b->threshold_limit		= THRESHOLD_MAX;
409 	threshold_restart_bank(&tr);
410 };
411 
412 static int setup_APIC_mce_threshold(int reserved, int new)
413 {
414 	if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
415 					      APIC_EILVT_MSG_FIX, 0))
416 		return new;
417 
418 	return reserved;
419 }
420 
421 static int setup_APIC_deferred_error(int reserved, int new)
422 {
423 	if (reserved < 0 && !setup_APIC_eilvt(new, DEFERRED_ERROR_VECTOR,
424 					      APIC_EILVT_MSG_FIX, 0))
425 		return new;
426 
427 	return reserved;
428 }
429 
430 static void deferred_error_interrupt_enable(struct cpuinfo_x86 *c)
431 {
432 	u32 low = 0, high = 0;
433 	int def_offset = -1, def_new;
434 
435 	if (rdmsr_safe(MSR_CU_DEF_ERR, &low, &high))
436 		return;
437 
438 	def_new = (low & MASK_DEF_LVTOFF) >> 4;
439 	if (!(low & MASK_DEF_LVTOFF)) {
440 		pr_err(FW_BUG "Your BIOS is not setting up LVT offset 0x2 for deferred error IRQs correctly.\n");
441 		def_new = DEF_LVT_OFF;
442 		low = (low & ~MASK_DEF_LVTOFF) | (DEF_LVT_OFF << 4);
443 	}
444 
445 	def_offset = setup_APIC_deferred_error(def_offset, def_new);
446 	if ((def_offset == def_new) &&
447 	    (deferred_error_int_vector != amd_deferred_error_interrupt))
448 		deferred_error_int_vector = amd_deferred_error_interrupt;
449 
450 	if (!mce_flags.smca)
451 		low = (low & ~MASK_DEF_INT_TYPE) | DEF_INT_TYPE_APIC;
452 
453 	wrmsr(MSR_CU_DEF_ERR, low, high);
454 }
455 
456 static u32 smca_get_block_address(unsigned int bank, unsigned int block)
457 {
458 	u32 low, high;
459 	u32 addr = 0;
460 
461 	if (smca_get_bank_type(bank) == SMCA_RESERVED)
462 		return addr;
463 
464 	if (!block)
465 		return MSR_AMD64_SMCA_MCx_MISC(bank);
466 
467 	/* Check our cache first: */
468 	if (smca_bank_addrs[bank][block] != -1)
469 		return smca_bank_addrs[bank][block];
470 
471 	/*
472 	 * For SMCA enabled processors, BLKPTR field of the first MISC register
473 	 * (MCx_MISC0) indicates presence of additional MISC regs set (MISC1-4).
474 	 */
475 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_CONFIG(bank), &low, &high))
476 		goto out;
477 
478 	if (!(low & MCI_CONFIG_MCAX))
479 		goto out;
480 
481 	if (!rdmsr_safe(MSR_AMD64_SMCA_MCx_MISC(bank), &low, &high) &&
482 	    (low & MASK_BLKPTR_LO))
483 		addr = MSR_AMD64_SMCA_MCx_MISCy(bank, block - 1);
484 
485 out:
486 	smca_bank_addrs[bank][block] = addr;
487 	return addr;
488 }
489 
490 static u32 get_block_address(u32 current_addr, u32 low, u32 high,
491 			     unsigned int bank, unsigned int block)
492 {
493 	u32 addr = 0, offset = 0;
494 
495 	if ((bank >= mca_cfg.banks) || (block >= NR_BLOCKS))
496 		return addr;
497 
498 	if (mce_flags.smca)
499 		return smca_get_block_address(bank, block);
500 
501 	/* Fall back to method we used for older processors: */
502 	switch (block) {
503 	case 0:
504 		addr = msr_ops.misc(bank);
505 		break;
506 	case 1:
507 		offset = ((low & MASK_BLKPTR_LO) >> 21);
508 		if (offset)
509 			addr = MCG_XBLK_ADDR + offset;
510 		break;
511 	default:
512 		addr = ++current_addr;
513 	}
514 	return addr;
515 }
516 
517 static int
518 prepare_threshold_block(unsigned int bank, unsigned int block, u32 addr,
519 			int offset, u32 misc_high)
520 {
521 	unsigned int cpu = smp_processor_id();
522 	u32 smca_low, smca_high;
523 	struct threshold_block b;
524 	int new;
525 
526 	if (!block)
527 		per_cpu(bank_map, cpu) |= (1 << bank);
528 
529 	memset(&b, 0, sizeof(b));
530 	b.cpu			= cpu;
531 	b.bank			= bank;
532 	b.block			= block;
533 	b.address		= addr;
534 	b.interrupt_capable	= lvt_interrupt_supported(bank, misc_high);
535 
536 	if (!b.interrupt_capable)
537 		goto done;
538 
539 	b.interrupt_enable = 1;
540 
541 	if (!mce_flags.smca) {
542 		new = (misc_high & MASK_LVTOFF_HI) >> 20;
543 		goto set_offset;
544 	}
545 
546 	/* Gather LVT offset for thresholding: */
547 	if (rdmsr_safe(MSR_CU_DEF_ERR, &smca_low, &smca_high))
548 		goto out;
549 
550 	new = (smca_low & SMCA_THR_LVT_OFF) >> 12;
551 
552 set_offset:
553 	offset = setup_APIC_mce_threshold(offset, new);
554 	if (offset == new)
555 		thresholding_irq_en = true;
556 
557 done:
558 	mce_threshold_block_init(&b, offset);
559 
560 out:
561 	return offset;
562 }
563 
564 bool amd_filter_mce(struct mce *m)
565 {
566 	enum smca_bank_types bank_type = smca_get_bank_type(m->bank);
567 	struct cpuinfo_x86 *c = &boot_cpu_data;
568 	u8 xec = (m->status >> 16) & 0x3F;
569 
570 	/* See Family 17h Models 10h-2Fh Erratum #1114. */
571 	if (c->x86 == 0x17 &&
572 	    c->x86_model >= 0x10 && c->x86_model <= 0x2F &&
573 	    bank_type == SMCA_IF && xec == 10)
574 		return true;
575 
576 	return false;
577 }
578 
579 /*
580  * Turn off thresholding banks for the following conditions:
581  * - MC4_MISC thresholding is not supported on Family 0x15.
582  * - Prevent possible spurious interrupts from the IF bank on Family 0x17
583  *   Models 0x10-0x2F due to Erratum #1114.
584  */
585 void disable_err_thresholding(struct cpuinfo_x86 *c, unsigned int bank)
586 {
587 	int i, num_msrs;
588 	u64 hwcr;
589 	bool need_toggle;
590 	u32 msrs[NR_BLOCKS];
591 
592 	if (c->x86 == 0x15 && bank == 4) {
593 		msrs[0] = 0x00000413; /* MC4_MISC0 */
594 		msrs[1] = 0xc0000408; /* MC4_MISC1 */
595 		num_msrs = 2;
596 	} else if (c->x86 == 0x17 &&
597 		   (c->x86_model >= 0x10 && c->x86_model <= 0x2F)) {
598 
599 		if (smca_get_bank_type(bank) != SMCA_IF)
600 			return;
601 
602 		msrs[0] = MSR_AMD64_SMCA_MCx_MISC(bank);
603 		num_msrs = 1;
604 	} else {
605 		return;
606 	}
607 
608 	rdmsrl(MSR_K7_HWCR, hwcr);
609 
610 	/* McStatusWrEn has to be set */
611 	need_toggle = !(hwcr & BIT(18));
612 	if (need_toggle)
613 		wrmsrl(MSR_K7_HWCR, hwcr | BIT(18));
614 
615 	/* Clear CntP bit safely */
616 	for (i = 0; i < num_msrs; i++)
617 		msr_clear_bit(msrs[i], 62);
618 
619 	/* restore old settings */
620 	if (need_toggle)
621 		wrmsrl(MSR_K7_HWCR, hwcr);
622 }
623 
624 /* cpu init entry point, called from mce.c with preempt off */
625 void mce_amd_feature_init(struct cpuinfo_x86 *c)
626 {
627 	u32 low = 0, high = 0, address = 0;
628 	unsigned int bank, block, cpu = smp_processor_id();
629 	int offset = -1;
630 
631 	for (bank = 0; bank < mca_cfg.banks; ++bank) {
632 		if (mce_flags.smca)
633 			smca_configure(bank, cpu);
634 
635 		disable_err_thresholding(c, bank);
636 
637 		for (block = 0; block < NR_BLOCKS; ++block) {
638 			address = get_block_address(address, low, high, bank, block);
639 			if (!address)
640 				break;
641 
642 			if (rdmsr_safe(address, &low, &high))
643 				break;
644 
645 			if (!(high & MASK_VALID_HI))
646 				continue;
647 
648 			if (!(high & MASK_CNTP_HI)  ||
649 			     (high & MASK_LOCKED_HI))
650 				continue;
651 
652 			offset = prepare_threshold_block(bank, block, address, offset, high);
653 		}
654 	}
655 
656 	if (mce_flags.succor)
657 		deferred_error_interrupt_enable(c);
658 }
659 
660 int umc_normaddr_to_sysaddr(u64 norm_addr, u16 nid, u8 umc, u64 *sys_addr)
661 {
662 	u64 dram_base_addr, dram_limit_addr, dram_hole_base;
663 	/* We start from the normalized address */
664 	u64 ret_addr = norm_addr;
665 
666 	u32 tmp;
667 
668 	u8 die_id_shift, die_id_mask, socket_id_shift, socket_id_mask;
669 	u8 intlv_num_dies, intlv_num_chan, intlv_num_sockets;
670 	u8 intlv_addr_sel, intlv_addr_bit;
671 	u8 num_intlv_bits, hashed_bit;
672 	u8 lgcy_mmio_hole_en, base = 0;
673 	u8 cs_mask, cs_id = 0;
674 	bool hash_enabled = false;
675 
676 	/* Read D18F0x1B4 (DramOffset), check if base 1 is used. */
677 	if (amd_df_indirect_read(nid, 0, 0x1B4, umc, &tmp))
678 		goto out_err;
679 
680 	/* Remove HiAddrOffset from normalized address, if enabled: */
681 	if (tmp & BIT(0)) {
682 		u64 hi_addr_offset = (tmp & GENMASK_ULL(31, 20)) << 8;
683 
684 		if (norm_addr >= hi_addr_offset) {
685 			ret_addr -= hi_addr_offset;
686 			base = 1;
687 		}
688 	}
689 
690 	/* Read D18F0x110 (DramBaseAddress). */
691 	if (amd_df_indirect_read(nid, 0, 0x110 + (8 * base), umc, &tmp))
692 		goto out_err;
693 
694 	/* Check if address range is valid. */
695 	if (!(tmp & BIT(0))) {
696 		pr_err("%s: Invalid DramBaseAddress range: 0x%x.\n",
697 			__func__, tmp);
698 		goto out_err;
699 	}
700 
701 	lgcy_mmio_hole_en = tmp & BIT(1);
702 	intlv_num_chan	  = (tmp >> 4) & 0xF;
703 	intlv_addr_sel	  = (tmp >> 8) & 0x7;
704 	dram_base_addr	  = (tmp & GENMASK_ULL(31, 12)) << 16;
705 
706 	/* {0, 1, 2, 3} map to address bits {8, 9, 10, 11} respectively */
707 	if (intlv_addr_sel > 3) {
708 		pr_err("%s: Invalid interleave address select %d.\n",
709 			__func__, intlv_addr_sel);
710 		goto out_err;
711 	}
712 
713 	/* Read D18F0x114 (DramLimitAddress). */
714 	if (amd_df_indirect_read(nid, 0, 0x114 + (8 * base), umc, &tmp))
715 		goto out_err;
716 
717 	intlv_num_sockets = (tmp >> 8) & 0x1;
718 	intlv_num_dies	  = (tmp >> 10) & 0x3;
719 	dram_limit_addr	  = ((tmp & GENMASK_ULL(31, 12)) << 16) | GENMASK_ULL(27, 0);
720 
721 	intlv_addr_bit = intlv_addr_sel + 8;
722 
723 	/* Re-use intlv_num_chan by setting it equal to log2(#channels) */
724 	switch (intlv_num_chan) {
725 	case 0:	intlv_num_chan = 0; break;
726 	case 1: intlv_num_chan = 1; break;
727 	case 3: intlv_num_chan = 2; break;
728 	case 5:	intlv_num_chan = 3; break;
729 	case 7:	intlv_num_chan = 4; break;
730 
731 	case 8: intlv_num_chan = 1;
732 		hash_enabled = true;
733 		break;
734 	default:
735 		pr_err("%s: Invalid number of interleaved channels %d.\n",
736 			__func__, intlv_num_chan);
737 		goto out_err;
738 	}
739 
740 	num_intlv_bits = intlv_num_chan;
741 
742 	if (intlv_num_dies > 2) {
743 		pr_err("%s: Invalid number of interleaved nodes/dies %d.\n",
744 			__func__, intlv_num_dies);
745 		goto out_err;
746 	}
747 
748 	num_intlv_bits += intlv_num_dies;
749 
750 	/* Add a bit if sockets are interleaved. */
751 	num_intlv_bits += intlv_num_sockets;
752 
753 	/* Assert num_intlv_bits <= 4 */
754 	if (num_intlv_bits > 4) {
755 		pr_err("%s: Invalid interleave bits %d.\n",
756 			__func__, num_intlv_bits);
757 		goto out_err;
758 	}
759 
760 	if (num_intlv_bits > 0) {
761 		u64 temp_addr_x, temp_addr_i, temp_addr_y;
762 		u8 die_id_bit, sock_id_bit, cs_fabric_id;
763 
764 		/*
765 		 * Read FabricBlockInstanceInformation3_CS[BlockFabricID].
766 		 * This is the fabric id for this coherent slave. Use
767 		 * umc/channel# as instance id of the coherent slave
768 		 * for FICAA.
769 		 */
770 		if (amd_df_indirect_read(nid, 0, 0x50, umc, &tmp))
771 			goto out_err;
772 
773 		cs_fabric_id = (tmp >> 8) & 0xFF;
774 		die_id_bit   = 0;
775 
776 		/* If interleaved over more than 1 channel: */
777 		if (intlv_num_chan) {
778 			die_id_bit = intlv_num_chan;
779 			cs_mask	   = (1 << die_id_bit) - 1;
780 			cs_id	   = cs_fabric_id & cs_mask;
781 		}
782 
783 		sock_id_bit = die_id_bit;
784 
785 		/* Read D18F1x208 (SystemFabricIdMask). */
786 		if (intlv_num_dies || intlv_num_sockets)
787 			if (amd_df_indirect_read(nid, 1, 0x208, umc, &tmp))
788 				goto out_err;
789 
790 		/* If interleaved over more than 1 die. */
791 		if (intlv_num_dies) {
792 			sock_id_bit  = die_id_bit + intlv_num_dies;
793 			die_id_shift = (tmp >> 24) & 0xF;
794 			die_id_mask  = (tmp >> 8) & 0xFF;
795 
796 			cs_id |= ((cs_fabric_id & die_id_mask) >> die_id_shift) << die_id_bit;
797 		}
798 
799 		/* If interleaved over more than 1 socket. */
800 		if (intlv_num_sockets) {
801 			socket_id_shift	= (tmp >> 28) & 0xF;
802 			socket_id_mask	= (tmp >> 16) & 0xFF;
803 
804 			cs_id |= ((cs_fabric_id & socket_id_mask) >> socket_id_shift) << sock_id_bit;
805 		}
806 
807 		/*
808 		 * The pre-interleaved address consists of XXXXXXIIIYYYYY
809 		 * where III is the ID for this CS, and XXXXXXYYYYY are the
810 		 * address bits from the post-interleaved address.
811 		 * "num_intlv_bits" has been calculated to tell us how many "I"
812 		 * bits there are. "intlv_addr_bit" tells us how many "Y" bits
813 		 * there are (where "I" starts).
814 		 */
815 		temp_addr_y = ret_addr & GENMASK_ULL(intlv_addr_bit-1, 0);
816 		temp_addr_i = (cs_id << intlv_addr_bit);
817 		temp_addr_x = (ret_addr & GENMASK_ULL(63, intlv_addr_bit)) << num_intlv_bits;
818 		ret_addr    = temp_addr_x | temp_addr_i | temp_addr_y;
819 	}
820 
821 	/* Add dram base address */
822 	ret_addr += dram_base_addr;
823 
824 	/* If legacy MMIO hole enabled */
825 	if (lgcy_mmio_hole_en) {
826 		if (amd_df_indirect_read(nid, 0, 0x104, umc, &tmp))
827 			goto out_err;
828 
829 		dram_hole_base = tmp & GENMASK(31, 24);
830 		if (ret_addr >= dram_hole_base)
831 			ret_addr += (BIT_ULL(32) - dram_hole_base);
832 	}
833 
834 	if (hash_enabled) {
835 		/* Save some parentheses and grab ls-bit at the end. */
836 		hashed_bit =	(ret_addr >> 12) ^
837 				(ret_addr >> 18) ^
838 				(ret_addr >> 21) ^
839 				(ret_addr >> 30) ^
840 				cs_id;
841 
842 		hashed_bit &= BIT(0);
843 
844 		if (hashed_bit != ((ret_addr >> intlv_addr_bit) & BIT(0)))
845 			ret_addr ^= BIT(intlv_addr_bit);
846 	}
847 
848 	/* Is calculated system address is above DRAM limit address? */
849 	if (ret_addr > dram_limit_addr)
850 		goto out_err;
851 
852 	*sys_addr = ret_addr;
853 	return 0;
854 
855 out_err:
856 	return -EINVAL;
857 }
858 EXPORT_SYMBOL_GPL(umc_normaddr_to_sysaddr);
859 
860 bool amd_mce_is_memory_error(struct mce *m)
861 {
862 	/* ErrCodeExt[20:16] */
863 	u8 xec = (m->status >> 16) & 0x1f;
864 
865 	if (mce_flags.smca)
866 		return smca_get_bank_type(m->bank) == SMCA_UMC && xec == 0x0;
867 
868 	return m->bank == 4 && xec == 0x8;
869 }
870 
871 static void __log_error(unsigned int bank, u64 status, u64 addr, u64 misc)
872 {
873 	struct mce m;
874 
875 	mce_setup(&m);
876 
877 	m.status = status;
878 	m.misc   = misc;
879 	m.bank   = bank;
880 	m.tsc	 = rdtsc();
881 
882 	if (m.status & MCI_STATUS_ADDRV) {
883 		m.addr = addr;
884 
885 		/*
886 		 * Extract [55:<lsb>] where lsb is the least significant
887 		 * *valid* bit of the address bits.
888 		 */
889 		if (mce_flags.smca) {
890 			u8 lsb = (m.addr >> 56) & 0x3f;
891 
892 			m.addr &= GENMASK_ULL(55, lsb);
893 		}
894 	}
895 
896 	if (mce_flags.smca) {
897 		rdmsrl(MSR_AMD64_SMCA_MCx_IPID(bank), m.ipid);
898 
899 		if (m.status & MCI_STATUS_SYNDV)
900 			rdmsrl(MSR_AMD64_SMCA_MCx_SYND(bank), m.synd);
901 	}
902 
903 	mce_log(&m);
904 }
905 
906 asmlinkage __visible void __irq_entry smp_deferred_error_interrupt(struct pt_regs *regs)
907 {
908 	entering_irq();
909 	trace_deferred_error_apic_entry(DEFERRED_ERROR_VECTOR);
910 	inc_irq_stat(irq_deferred_error_count);
911 	deferred_error_int_vector();
912 	trace_deferred_error_apic_exit(DEFERRED_ERROR_VECTOR);
913 	exiting_ack_irq();
914 }
915 
916 /*
917  * Returns true if the logged error is deferred. False, otherwise.
918  */
919 static inline bool
920 _log_error_bank(unsigned int bank, u32 msr_stat, u32 msr_addr, u64 misc)
921 {
922 	u64 status, addr = 0;
923 
924 	rdmsrl(msr_stat, status);
925 	if (!(status & MCI_STATUS_VAL))
926 		return false;
927 
928 	if (status & MCI_STATUS_ADDRV)
929 		rdmsrl(msr_addr, addr);
930 
931 	__log_error(bank, status, addr, misc);
932 
933 	wrmsrl(msr_stat, 0);
934 
935 	return status & MCI_STATUS_DEFERRED;
936 }
937 
938 /*
939  * We have three scenarios for checking for Deferred errors:
940  *
941  * 1) Non-SMCA systems check MCA_STATUS and log error if found.
942  * 2) SMCA systems check MCA_STATUS. If error is found then log it and also
943  *    clear MCA_DESTAT.
944  * 3) SMCA systems check MCA_DESTAT, if error was not found in MCA_STATUS, and
945  *    log it.
946  */
947 static void log_error_deferred(unsigned int bank)
948 {
949 	bool defrd;
950 
951 	defrd = _log_error_bank(bank, msr_ops.status(bank),
952 					msr_ops.addr(bank), 0);
953 
954 	if (!mce_flags.smca)
955 		return;
956 
957 	/* Clear MCA_DESTAT if we logged the deferred error from MCA_STATUS. */
958 	if (defrd) {
959 		wrmsrl(MSR_AMD64_SMCA_MCx_DESTAT(bank), 0);
960 		return;
961 	}
962 
963 	/*
964 	 * Only deferred errors are logged in MCA_DE{STAT,ADDR} so just check
965 	 * for a valid error.
966 	 */
967 	_log_error_bank(bank, MSR_AMD64_SMCA_MCx_DESTAT(bank),
968 			      MSR_AMD64_SMCA_MCx_DEADDR(bank), 0);
969 }
970 
971 /* APIC interrupt handler for deferred errors */
972 static void amd_deferred_error_interrupt(void)
973 {
974 	unsigned int bank;
975 
976 	for (bank = 0; bank < mca_cfg.banks; ++bank)
977 		log_error_deferred(bank);
978 }
979 
980 static void log_error_thresholding(unsigned int bank, u64 misc)
981 {
982 	_log_error_bank(bank, msr_ops.status(bank), msr_ops.addr(bank), misc);
983 }
984 
985 static void log_and_reset_block(struct threshold_block *block)
986 {
987 	struct thresh_restart tr;
988 	u32 low = 0, high = 0;
989 
990 	if (!block)
991 		return;
992 
993 	if (rdmsr_safe(block->address, &low, &high))
994 		return;
995 
996 	if (!(high & MASK_OVERFLOW_HI))
997 		return;
998 
999 	/* Log the MCE which caused the threshold event. */
1000 	log_error_thresholding(block->bank, ((u64)high << 32) | low);
1001 
1002 	/* Reset threshold block after logging error. */
1003 	memset(&tr, 0, sizeof(tr));
1004 	tr.b = block;
1005 	threshold_restart_bank(&tr);
1006 }
1007 
1008 /*
1009  * Threshold interrupt handler will service THRESHOLD_APIC_VECTOR. The interrupt
1010  * goes off when error_count reaches threshold_limit.
1011  */
1012 static void amd_threshold_interrupt(void)
1013 {
1014 	struct threshold_block *first_block = NULL, *block = NULL, *tmp = NULL;
1015 	unsigned int bank, cpu = smp_processor_id();
1016 
1017 	for (bank = 0; bank < mca_cfg.banks; ++bank) {
1018 		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
1019 			continue;
1020 
1021 		first_block = per_cpu(threshold_banks, cpu)[bank]->blocks;
1022 		if (!first_block)
1023 			continue;
1024 
1025 		/*
1026 		 * The first block is also the head of the list. Check it first
1027 		 * before iterating over the rest.
1028 		 */
1029 		log_and_reset_block(first_block);
1030 		list_for_each_entry_safe(block, tmp, &first_block->miscj, miscj)
1031 			log_and_reset_block(block);
1032 	}
1033 }
1034 
1035 /*
1036  * Sysfs Interface
1037  */
1038 
1039 struct threshold_attr {
1040 	struct attribute attr;
1041 	ssize_t (*show) (struct threshold_block *, char *);
1042 	ssize_t (*store) (struct threshold_block *, const char *, size_t count);
1043 };
1044 
1045 #define SHOW_FIELDS(name)						\
1046 static ssize_t show_ ## name(struct threshold_block *b, char *buf)	\
1047 {									\
1048 	return sprintf(buf, "%lu\n", (unsigned long) b->name);		\
1049 }
1050 SHOW_FIELDS(interrupt_enable)
1051 SHOW_FIELDS(threshold_limit)
1052 
1053 static ssize_t
1054 store_interrupt_enable(struct threshold_block *b, const char *buf, size_t size)
1055 {
1056 	struct thresh_restart tr;
1057 	unsigned long new;
1058 
1059 	if (!b->interrupt_capable)
1060 		return -EINVAL;
1061 
1062 	if (kstrtoul(buf, 0, &new) < 0)
1063 		return -EINVAL;
1064 
1065 	b->interrupt_enable = !!new;
1066 
1067 	memset(&tr, 0, sizeof(tr));
1068 	tr.b		= b;
1069 
1070 	smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
1071 
1072 	return size;
1073 }
1074 
1075 static ssize_t
1076 store_threshold_limit(struct threshold_block *b, const char *buf, size_t size)
1077 {
1078 	struct thresh_restart tr;
1079 	unsigned long new;
1080 
1081 	if (kstrtoul(buf, 0, &new) < 0)
1082 		return -EINVAL;
1083 
1084 	if (new > THRESHOLD_MAX)
1085 		new = THRESHOLD_MAX;
1086 	if (new < 1)
1087 		new = 1;
1088 
1089 	memset(&tr, 0, sizeof(tr));
1090 	tr.old_limit = b->threshold_limit;
1091 	b->threshold_limit = new;
1092 	tr.b = b;
1093 
1094 	smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
1095 
1096 	return size;
1097 }
1098 
1099 static ssize_t show_error_count(struct threshold_block *b, char *buf)
1100 {
1101 	u32 lo, hi;
1102 
1103 	rdmsr_on_cpu(b->cpu, b->address, &lo, &hi);
1104 
1105 	return sprintf(buf, "%u\n", ((hi & THRESHOLD_MAX) -
1106 				     (THRESHOLD_MAX - b->threshold_limit)));
1107 }
1108 
1109 static struct threshold_attr error_count = {
1110 	.attr = {.name = __stringify(error_count), .mode = 0444 },
1111 	.show = show_error_count,
1112 };
1113 
1114 #define RW_ATTR(val)							\
1115 static struct threshold_attr val = {					\
1116 	.attr	= {.name = __stringify(val), .mode = 0644 },		\
1117 	.show	= show_## val,						\
1118 	.store	= store_## val,						\
1119 };
1120 
1121 RW_ATTR(interrupt_enable);
1122 RW_ATTR(threshold_limit);
1123 
1124 static struct attribute *default_attrs[] = {
1125 	&threshold_limit.attr,
1126 	&error_count.attr,
1127 	NULL,	/* possibly interrupt_enable if supported, see below */
1128 	NULL,
1129 };
1130 
1131 #define to_block(k)	container_of(k, struct threshold_block, kobj)
1132 #define to_attr(a)	container_of(a, struct threshold_attr, attr)
1133 
1134 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
1135 {
1136 	struct threshold_block *b = to_block(kobj);
1137 	struct threshold_attr *a = to_attr(attr);
1138 	ssize_t ret;
1139 
1140 	ret = a->show ? a->show(b, buf) : -EIO;
1141 
1142 	return ret;
1143 }
1144 
1145 static ssize_t store(struct kobject *kobj, struct attribute *attr,
1146 		     const char *buf, size_t count)
1147 {
1148 	struct threshold_block *b = to_block(kobj);
1149 	struct threshold_attr *a = to_attr(attr);
1150 	ssize_t ret;
1151 
1152 	ret = a->store ? a->store(b, buf, count) : -EIO;
1153 
1154 	return ret;
1155 }
1156 
1157 static const struct sysfs_ops threshold_ops = {
1158 	.show			= show,
1159 	.store			= store,
1160 };
1161 
1162 static struct kobj_type threshold_ktype = {
1163 	.sysfs_ops		= &threshold_ops,
1164 	.default_attrs		= default_attrs,
1165 };
1166 
1167 static const char *get_name(unsigned int bank, struct threshold_block *b)
1168 {
1169 	enum smca_bank_types bank_type;
1170 
1171 	if (!mce_flags.smca) {
1172 		if (b && bank == 4)
1173 			return bank4_names(b);
1174 
1175 		return th_names[bank];
1176 	}
1177 
1178 	bank_type = smca_get_bank_type(bank);
1179 	if (bank_type >= N_SMCA_BANK_TYPES)
1180 		return NULL;
1181 
1182 	if (b && bank_type == SMCA_UMC) {
1183 		if (b->block < ARRAY_SIZE(smca_umc_block_names))
1184 			return smca_umc_block_names[b->block];
1185 		return NULL;
1186 	}
1187 
1188 	if (smca_banks[bank].hwid->count == 1)
1189 		return smca_get_name(bank_type);
1190 
1191 	snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN,
1192 		 "%s_%x", smca_get_name(bank_type),
1193 			  smca_banks[bank].sysfs_id);
1194 	return buf_mcatype;
1195 }
1196 
1197 static int allocate_threshold_blocks(unsigned int cpu, unsigned int bank,
1198 				     unsigned int block, u32 address)
1199 {
1200 	struct threshold_block *b = NULL;
1201 	u32 low, high;
1202 	int err;
1203 
1204 	if ((bank >= mca_cfg.banks) || (block >= NR_BLOCKS))
1205 		return 0;
1206 
1207 	if (rdmsr_safe_on_cpu(cpu, address, &low, &high))
1208 		return 0;
1209 
1210 	if (!(high & MASK_VALID_HI)) {
1211 		if (block)
1212 			goto recurse;
1213 		else
1214 			return 0;
1215 	}
1216 
1217 	if (!(high & MASK_CNTP_HI)  ||
1218 	     (high & MASK_LOCKED_HI))
1219 		goto recurse;
1220 
1221 	b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
1222 	if (!b)
1223 		return -ENOMEM;
1224 
1225 	b->block		= block;
1226 	b->bank			= bank;
1227 	b->cpu			= cpu;
1228 	b->address		= address;
1229 	b->interrupt_enable	= 0;
1230 	b->interrupt_capable	= lvt_interrupt_supported(bank, high);
1231 	b->threshold_limit	= THRESHOLD_MAX;
1232 
1233 	if (b->interrupt_capable) {
1234 		threshold_ktype.default_attrs[2] = &interrupt_enable.attr;
1235 		b->interrupt_enable = 1;
1236 	} else {
1237 		threshold_ktype.default_attrs[2] = NULL;
1238 	}
1239 
1240 	INIT_LIST_HEAD(&b->miscj);
1241 
1242 	if (per_cpu(threshold_banks, cpu)[bank]->blocks) {
1243 		list_add(&b->miscj,
1244 			 &per_cpu(threshold_banks, cpu)[bank]->blocks->miscj);
1245 	} else {
1246 		per_cpu(threshold_banks, cpu)[bank]->blocks = b;
1247 	}
1248 
1249 	err = kobject_init_and_add(&b->kobj, &threshold_ktype,
1250 				   per_cpu(threshold_banks, cpu)[bank]->kobj,
1251 				   get_name(bank, b));
1252 	if (err)
1253 		goto out_free;
1254 recurse:
1255 	address = get_block_address(address, low, high, bank, ++block);
1256 	if (!address)
1257 		return 0;
1258 
1259 	err = allocate_threshold_blocks(cpu, bank, block, address);
1260 	if (err)
1261 		goto out_free;
1262 
1263 	if (b)
1264 		kobject_uevent(&b->kobj, KOBJ_ADD);
1265 
1266 	return err;
1267 
1268 out_free:
1269 	if (b) {
1270 		kobject_put(&b->kobj);
1271 		list_del(&b->miscj);
1272 		kfree(b);
1273 	}
1274 	return err;
1275 }
1276 
1277 static int __threshold_add_blocks(struct threshold_bank *b)
1278 {
1279 	struct list_head *head = &b->blocks->miscj;
1280 	struct threshold_block *pos = NULL;
1281 	struct threshold_block *tmp = NULL;
1282 	int err = 0;
1283 
1284 	err = kobject_add(&b->blocks->kobj, b->kobj, b->blocks->kobj.name);
1285 	if (err)
1286 		return err;
1287 
1288 	list_for_each_entry_safe(pos, tmp, head, miscj) {
1289 
1290 		err = kobject_add(&pos->kobj, b->kobj, pos->kobj.name);
1291 		if (err) {
1292 			list_for_each_entry_safe_reverse(pos, tmp, head, miscj)
1293 				kobject_del(&pos->kobj);
1294 
1295 			return err;
1296 		}
1297 	}
1298 	return err;
1299 }
1300 
1301 static int threshold_create_bank(unsigned int cpu, unsigned int bank)
1302 {
1303 	struct device *dev = per_cpu(mce_device, cpu);
1304 	struct amd_northbridge *nb = NULL;
1305 	struct threshold_bank *b = NULL;
1306 	const char *name = get_name(bank, NULL);
1307 	int err = 0;
1308 
1309 	if (!dev)
1310 		return -ENODEV;
1311 
1312 	if (is_shared_bank(bank)) {
1313 		nb = node_to_amd_nb(amd_get_nb_id(cpu));
1314 
1315 		/* threshold descriptor already initialized on this node? */
1316 		if (nb && nb->bank4) {
1317 			/* yes, use it */
1318 			b = nb->bank4;
1319 			err = kobject_add(b->kobj, &dev->kobj, name);
1320 			if (err)
1321 				goto out;
1322 
1323 			per_cpu(threshold_banks, cpu)[bank] = b;
1324 			refcount_inc(&b->cpus);
1325 
1326 			err = __threshold_add_blocks(b);
1327 
1328 			goto out;
1329 		}
1330 	}
1331 
1332 	b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
1333 	if (!b) {
1334 		err = -ENOMEM;
1335 		goto out;
1336 	}
1337 
1338 	b->kobj = kobject_create_and_add(name, &dev->kobj);
1339 	if (!b->kobj) {
1340 		err = -EINVAL;
1341 		goto out_free;
1342 	}
1343 
1344 	per_cpu(threshold_banks, cpu)[bank] = b;
1345 
1346 	if (is_shared_bank(bank)) {
1347 		refcount_set(&b->cpus, 1);
1348 
1349 		/* nb is already initialized, see above */
1350 		if (nb) {
1351 			WARN_ON(nb->bank4);
1352 			nb->bank4 = b;
1353 		}
1354 	}
1355 
1356 	err = allocate_threshold_blocks(cpu, bank, 0, msr_ops.misc(bank));
1357 	if (!err)
1358 		goto out;
1359 
1360  out_free:
1361 	kfree(b);
1362 
1363  out:
1364 	return err;
1365 }
1366 
1367 static void deallocate_threshold_block(unsigned int cpu,
1368 						 unsigned int bank)
1369 {
1370 	struct threshold_block *pos = NULL;
1371 	struct threshold_block *tmp = NULL;
1372 	struct threshold_bank *head = per_cpu(threshold_banks, cpu)[bank];
1373 
1374 	if (!head)
1375 		return;
1376 
1377 	list_for_each_entry_safe(pos, tmp, &head->blocks->miscj, miscj) {
1378 		kobject_put(&pos->kobj);
1379 		list_del(&pos->miscj);
1380 		kfree(pos);
1381 	}
1382 
1383 	kfree(per_cpu(threshold_banks, cpu)[bank]->blocks);
1384 	per_cpu(threshold_banks, cpu)[bank]->blocks = NULL;
1385 }
1386 
1387 static void __threshold_remove_blocks(struct threshold_bank *b)
1388 {
1389 	struct threshold_block *pos = NULL;
1390 	struct threshold_block *tmp = NULL;
1391 
1392 	kobject_del(b->kobj);
1393 
1394 	list_for_each_entry_safe(pos, tmp, &b->blocks->miscj, miscj)
1395 		kobject_del(&pos->kobj);
1396 }
1397 
1398 static void threshold_remove_bank(unsigned int cpu, int bank)
1399 {
1400 	struct amd_northbridge *nb;
1401 	struct threshold_bank *b;
1402 
1403 	b = per_cpu(threshold_banks, cpu)[bank];
1404 	if (!b)
1405 		return;
1406 
1407 	if (!b->blocks)
1408 		goto free_out;
1409 
1410 	if (is_shared_bank(bank)) {
1411 		if (!refcount_dec_and_test(&b->cpus)) {
1412 			__threshold_remove_blocks(b);
1413 			per_cpu(threshold_banks, cpu)[bank] = NULL;
1414 			return;
1415 		} else {
1416 			/*
1417 			 * the last CPU on this node using the shared bank is
1418 			 * going away, remove that bank now.
1419 			 */
1420 			nb = node_to_amd_nb(amd_get_nb_id(cpu));
1421 			nb->bank4 = NULL;
1422 		}
1423 	}
1424 
1425 	deallocate_threshold_block(cpu, bank);
1426 
1427 free_out:
1428 	kobject_del(b->kobj);
1429 	kobject_put(b->kobj);
1430 	kfree(b);
1431 	per_cpu(threshold_banks, cpu)[bank] = NULL;
1432 }
1433 
1434 int mce_threshold_remove_device(unsigned int cpu)
1435 {
1436 	unsigned int bank;
1437 
1438 	for (bank = 0; bank < mca_cfg.banks; ++bank) {
1439 		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
1440 			continue;
1441 		threshold_remove_bank(cpu, bank);
1442 	}
1443 	kfree(per_cpu(threshold_banks, cpu));
1444 	per_cpu(threshold_banks, cpu) = NULL;
1445 	return 0;
1446 }
1447 
1448 /* create dir/files for all valid threshold banks */
1449 int mce_threshold_create_device(unsigned int cpu)
1450 {
1451 	unsigned int bank;
1452 	struct threshold_bank **bp;
1453 	int err = 0;
1454 
1455 	bp = per_cpu(threshold_banks, cpu);
1456 	if (bp)
1457 		return 0;
1458 
1459 	bp = kcalloc(mca_cfg.banks, sizeof(struct threshold_bank *),
1460 		     GFP_KERNEL);
1461 	if (!bp)
1462 		return -ENOMEM;
1463 
1464 	per_cpu(threshold_banks, cpu) = bp;
1465 
1466 	for (bank = 0; bank < mca_cfg.banks; ++bank) {
1467 		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
1468 			continue;
1469 		err = threshold_create_bank(cpu, bank);
1470 		if (err)
1471 			goto err;
1472 	}
1473 	return err;
1474 err:
1475 	mce_threshold_remove_device(cpu);
1476 	return err;
1477 }
1478 
1479 static __init int threshold_init_device(void)
1480 {
1481 	unsigned lcpu = 0;
1482 
1483 	/* to hit CPUs online before the notifier is up */
1484 	for_each_online_cpu(lcpu) {
1485 		int err = mce_threshold_create_device(lcpu);
1486 
1487 		if (err)
1488 			return err;
1489 	}
1490 
1491 	if (thresholding_irq_en)
1492 		mce_threshold_vector = amd_threshold_interrupt;
1493 
1494 	return 0;
1495 }
1496 /*
1497  * there are 3 funcs which need to be _initcalled in a logic sequence:
1498  * 1. xen_late_init_mcelog
1499  * 2. mcheck_init_device
1500  * 3. threshold_init_device
1501  *
1502  * xen_late_init_mcelog must register xen_mce_chrdev_device before
1503  * native mce_chrdev_device registration if running under xen platform;
1504  *
1505  * mcheck_init_device should be inited before threshold_init_device to
1506  * initialize mce_device, otherwise a NULL ptr dereference will cause panic.
1507  *
1508  * so we use following _initcalls
1509  * 1. device_initcall(xen_late_init_mcelog);
1510  * 2. device_initcall_sync(mcheck_init_device);
1511  * 3. late_initcall(threshold_init_device);
1512  *
1513  * when running under xen, the initcall order is 1,2,3;
1514  * on baremetal, we skip 1 and we do only 2 and 3.
1515  */
1516 late_initcall(threshold_init_device);
1517