xref: /openbmc/linux/arch/x86/kernel/cpu/mce/amd.c (revision a037f3ca)
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_LS_V2]	= { "load_store",	"Load Store Unit" },
82 	[SMCA_IF]	= { "insn_fetch",	"Instruction Fetch Unit" },
83 	[SMCA_L2_CACHE]	= { "l2_cache",		"L2 Cache" },
84 	[SMCA_DE]	= { "decode_unit",	"Decode Unit" },
85 	[SMCA_RESERVED]	= { "reserved",		"Reserved" },
86 	[SMCA_EX]	= { "execution_unit",	"Execution Unit" },
87 	[SMCA_FP]	= { "floating_point",	"Floating Point Unit" },
88 	[SMCA_L3_CACHE]	= { "l3_cache",		"L3 Cache" },
89 	[SMCA_CS]	= { "coherent_slave",	"Coherent Slave" },
90 	[SMCA_CS_V2]	= { "coherent_slave",	"Coherent Slave" },
91 	[SMCA_PIE]	= { "pie",		"Power, Interrupts, etc." },
92 	[SMCA_UMC]	= { "umc",		"Unified Memory Controller" },
93 	[SMCA_PB]	= { "param_block",	"Parameter Block" },
94 	[SMCA_PSP]	= { "psp",		"Platform Security Processor" },
95 	[SMCA_PSP_V2]	= { "psp",		"Platform Security Processor" },
96 	[SMCA_SMU]	= { "smu",		"System Management Unit" },
97 	[SMCA_SMU_V2]	= { "smu",		"System Management Unit" },
98 	[SMCA_MP5]	= { "mp5",		"Microprocessor 5 Unit" },
99 	[SMCA_NBIO]	= { "nbio",		"Northbridge IO Unit" },
100 	[SMCA_PCIE]	= { "pcie",		"PCI Express Unit" },
101 };
102 
103 static const char *smca_get_name(enum smca_bank_types t)
104 {
105 	if (t >= N_SMCA_BANK_TYPES)
106 		return NULL;
107 
108 	return smca_names[t].name;
109 }
110 
111 const char *smca_get_long_name(enum smca_bank_types t)
112 {
113 	if (t >= N_SMCA_BANK_TYPES)
114 		return NULL;
115 
116 	return smca_names[t].long_name;
117 }
118 EXPORT_SYMBOL_GPL(smca_get_long_name);
119 
120 static enum smca_bank_types smca_get_bank_type(unsigned int bank)
121 {
122 	struct smca_bank *b;
123 
124 	if (bank >= MAX_NR_BANKS)
125 		return N_SMCA_BANK_TYPES;
126 
127 	b = &smca_banks[bank];
128 	if (!b->hwid)
129 		return N_SMCA_BANK_TYPES;
130 
131 	return b->hwid->bank_type;
132 }
133 
134 static struct smca_hwid smca_hwid_mcatypes[] = {
135 	/* { bank_type, hwid_mcatype, xec_bitmap } */
136 
137 	/* Reserved type */
138 	{ SMCA_RESERVED, HWID_MCATYPE(0x00, 0x0), 0x0 },
139 
140 	/* ZN Core (HWID=0xB0) MCA types */
141 	{ SMCA_LS,	 HWID_MCATYPE(0xB0, 0x0), 0x1FFFFF },
142 	{ SMCA_LS_V2,	 HWID_MCATYPE(0xB0, 0x10), 0xFFFFFF },
143 	{ SMCA_IF,	 HWID_MCATYPE(0xB0, 0x1), 0x3FFF },
144 	{ SMCA_L2_CACHE, HWID_MCATYPE(0xB0, 0x2), 0xF },
145 	{ SMCA_DE,	 HWID_MCATYPE(0xB0, 0x3), 0x1FF },
146 	/* HWID 0xB0 MCATYPE 0x4 is Reserved */
147 	{ SMCA_EX,	 HWID_MCATYPE(0xB0, 0x5), 0xFFF },
148 	{ SMCA_FP,	 HWID_MCATYPE(0xB0, 0x6), 0x7F },
149 	{ SMCA_L3_CACHE, HWID_MCATYPE(0xB0, 0x7), 0xFF },
150 
151 	/* Data Fabric MCA types */
152 	{ SMCA_CS,	 HWID_MCATYPE(0x2E, 0x0), 0x1FF },
153 	{ SMCA_PIE,	 HWID_MCATYPE(0x2E, 0x1), 0x1F },
154 	{ SMCA_CS_V2,	 HWID_MCATYPE(0x2E, 0x2), 0x3FFF },
155 
156 	/* Unified Memory Controller MCA type */
157 	{ SMCA_UMC,	 HWID_MCATYPE(0x96, 0x0), 0xFF },
158 
159 	/* Parameter Block MCA type */
160 	{ SMCA_PB,	 HWID_MCATYPE(0x05, 0x0), 0x1 },
161 
162 	/* Platform Security Processor MCA type */
163 	{ SMCA_PSP,	 HWID_MCATYPE(0xFF, 0x0), 0x1 },
164 	{ SMCA_PSP_V2,	 HWID_MCATYPE(0xFF, 0x1), 0x3FFFF },
165 
166 	/* System Management Unit MCA type */
167 	{ SMCA_SMU,	 HWID_MCATYPE(0x01, 0x0), 0x1 },
168 	{ SMCA_SMU_V2,	 HWID_MCATYPE(0x01, 0x1), 0x7FF },
169 
170 	/* Microprocessor 5 Unit MCA type */
171 	{ SMCA_MP5,	 HWID_MCATYPE(0x01, 0x2), 0x3FF },
172 
173 	/* Northbridge IO Unit MCA type */
174 	{ SMCA_NBIO,	 HWID_MCATYPE(0x18, 0x0), 0x1F },
175 
176 	/* PCI Express Unit MCA type */
177 	{ SMCA_PCIE,	 HWID_MCATYPE(0x46, 0x0), 0x1F },
178 };
179 
180 struct smca_bank smca_banks[MAX_NR_BANKS];
181 EXPORT_SYMBOL_GPL(smca_banks);
182 
183 /*
184  * In SMCA enabled processors, we can have multiple banks for a given IP type.
185  * So to define a unique name for each bank, we use a temp c-string to append
186  * the MCA_IPID[InstanceId] to type's name in get_name().
187  *
188  * InstanceId is 32 bits which is 8 characters. Make sure MAX_MCATYPE_NAME_LEN
189  * is greater than 8 plus 1 (for underscore) plus length of longest type name.
190  */
191 #define MAX_MCATYPE_NAME_LEN	30
192 static char buf_mcatype[MAX_MCATYPE_NAME_LEN];
193 
194 static DEFINE_PER_CPU(struct threshold_bank **, threshold_banks);
195 
196 /*
197  * A list of the banks enabled on each logical CPU. Controls which respective
198  * descriptors to initialize later in mce_threshold_create_device().
199  */
200 static DEFINE_PER_CPU(unsigned int, bank_map);
201 
202 /* Map of banks that have more than MCA_MISC0 available. */
203 static DEFINE_PER_CPU(u32, smca_misc_banks_map);
204 
205 static void amd_threshold_interrupt(void);
206 static void amd_deferred_error_interrupt(void);
207 
208 static void default_deferred_error_interrupt(void)
209 {
210 	pr_err("Unexpected deferred interrupt at vector %x\n", DEFERRED_ERROR_VECTOR);
211 }
212 void (*deferred_error_int_vector)(void) = default_deferred_error_interrupt;
213 
214 static void smca_set_misc_banks_map(unsigned int bank, unsigned int cpu)
215 {
216 	u32 low, high;
217 
218 	/*
219 	 * For SMCA enabled processors, BLKPTR field of the first MISC register
220 	 * (MCx_MISC0) indicates presence of additional MISC regs set (MISC1-4).
221 	 */
222 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_CONFIG(bank), &low, &high))
223 		return;
224 
225 	if (!(low & MCI_CONFIG_MCAX))
226 		return;
227 
228 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_MISC(bank), &low, &high))
229 		return;
230 
231 	if (low & MASK_BLKPTR_LO)
232 		per_cpu(smca_misc_banks_map, cpu) |= BIT(bank);
233 
234 }
235 
236 static void smca_configure(unsigned int bank, unsigned int cpu)
237 {
238 	unsigned int i, hwid_mcatype;
239 	struct smca_hwid *s_hwid;
240 	u32 high, low;
241 	u32 smca_config = MSR_AMD64_SMCA_MCx_CONFIG(bank);
242 
243 	/* Set appropriate bits in MCA_CONFIG */
244 	if (!rdmsr_safe(smca_config, &low, &high)) {
245 		/*
246 		 * OS is required to set the MCAX bit to acknowledge that it is
247 		 * now using the new MSR ranges and new registers under each
248 		 * bank. It also means that the OS will configure deferred
249 		 * errors in the new MCx_CONFIG register. If the bit is not set,
250 		 * uncorrectable errors will cause a system panic.
251 		 *
252 		 * MCA_CONFIG[MCAX] is bit 32 (0 in the high portion of the MSR.)
253 		 */
254 		high |= BIT(0);
255 
256 		/*
257 		 * SMCA sets the Deferred Error Interrupt type per bank.
258 		 *
259 		 * MCA_CONFIG[DeferredIntTypeSupported] is bit 5, and tells us
260 		 * if the DeferredIntType bit field is available.
261 		 *
262 		 * MCA_CONFIG[DeferredIntType] is bits [38:37] ([6:5] in the
263 		 * high portion of the MSR). OS should set this to 0x1 to enable
264 		 * APIC based interrupt. First, check that no interrupt has been
265 		 * set.
266 		 */
267 		if ((low & BIT(5)) && !((high >> 5) & 0x3))
268 			high |= BIT(5);
269 
270 		wrmsr(smca_config, low, high);
271 	}
272 
273 	smca_set_misc_banks_map(bank, cpu);
274 
275 	/* Return early if this bank was already initialized. */
276 	if (smca_banks[bank].hwid && smca_banks[bank].hwid->hwid_mcatype != 0)
277 		return;
278 
279 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_IPID(bank), &low, &high)) {
280 		pr_warn("Failed to read MCA_IPID for bank %d\n", bank);
281 		return;
282 	}
283 
284 	hwid_mcatype = HWID_MCATYPE(high & MCI_IPID_HWID,
285 				    (high & MCI_IPID_MCATYPE) >> 16);
286 
287 	for (i = 0; i < ARRAY_SIZE(smca_hwid_mcatypes); i++) {
288 		s_hwid = &smca_hwid_mcatypes[i];
289 		if (hwid_mcatype == s_hwid->hwid_mcatype) {
290 			smca_banks[bank].hwid = s_hwid;
291 			smca_banks[bank].id = low;
292 			smca_banks[bank].sysfs_id = s_hwid->count++;
293 			break;
294 		}
295 	}
296 }
297 
298 struct thresh_restart {
299 	struct threshold_block	*b;
300 	int			reset;
301 	int			set_lvt_off;
302 	int			lvt_off;
303 	u16			old_limit;
304 };
305 
306 static inline bool is_shared_bank(int bank)
307 {
308 	/*
309 	 * Scalable MCA provides for only one core to have access to the MSRs of
310 	 * a shared bank.
311 	 */
312 	if (mce_flags.smca)
313 		return false;
314 
315 	/* Bank 4 is for northbridge reporting and is thus shared */
316 	return (bank == 4);
317 }
318 
319 static const char *bank4_names(const struct threshold_block *b)
320 {
321 	switch (b->address) {
322 	/* MSR4_MISC0 */
323 	case 0x00000413:
324 		return "dram";
325 
326 	case 0xc0000408:
327 		return "ht_links";
328 
329 	case 0xc0000409:
330 		return "l3_cache";
331 
332 	default:
333 		WARN(1, "Funny MSR: 0x%08x\n", b->address);
334 		return "";
335 	}
336 };
337 
338 
339 static bool lvt_interrupt_supported(unsigned int bank, u32 msr_high_bits)
340 {
341 	/*
342 	 * bank 4 supports APIC LVT interrupts implicitly since forever.
343 	 */
344 	if (bank == 4)
345 		return true;
346 
347 	/*
348 	 * IntP: interrupt present; if this bit is set, the thresholding
349 	 * bank can generate APIC LVT interrupts
350 	 */
351 	return msr_high_bits & BIT(28);
352 }
353 
354 static int lvt_off_valid(struct threshold_block *b, int apic, u32 lo, u32 hi)
355 {
356 	int msr = (hi & MASK_LVTOFF_HI) >> 20;
357 
358 	if (apic < 0) {
359 		pr_err(FW_BUG "cpu %d, failed to setup threshold interrupt "
360 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n", b->cpu,
361 		       b->bank, b->block, b->address, hi, lo);
362 		return 0;
363 	}
364 
365 	if (apic != msr) {
366 		/*
367 		 * On SMCA CPUs, LVT offset is programmed at a different MSR, and
368 		 * the BIOS provides the value. The original field where LVT offset
369 		 * was set is reserved. Return early here:
370 		 */
371 		if (mce_flags.smca)
372 			return 0;
373 
374 		pr_err(FW_BUG "cpu %d, invalid threshold interrupt offset %d "
375 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n",
376 		       b->cpu, apic, b->bank, b->block, b->address, hi, lo);
377 		return 0;
378 	}
379 
380 	return 1;
381 };
382 
383 /* Reprogram MCx_MISC MSR behind this threshold bank. */
384 static void threshold_restart_bank(void *_tr)
385 {
386 	struct thresh_restart *tr = _tr;
387 	u32 hi, lo;
388 
389 	/* sysfs write might race against an offline operation */
390 	if (this_cpu_read(threshold_banks))
391 		return;
392 
393 	rdmsr(tr->b->address, lo, hi);
394 
395 	if (tr->b->threshold_limit < (hi & THRESHOLD_MAX))
396 		tr->reset = 1;	/* limit cannot be lower than err count */
397 
398 	if (tr->reset) {		/* reset err count and overflow bit */
399 		hi =
400 		    (hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
401 		    (THRESHOLD_MAX - tr->b->threshold_limit);
402 	} else if (tr->old_limit) {	/* change limit w/o reset */
403 		int new_count = (hi & THRESHOLD_MAX) +
404 		    (tr->old_limit - tr->b->threshold_limit);
405 
406 		hi = (hi & ~MASK_ERR_COUNT_HI) |
407 		    (new_count & THRESHOLD_MAX);
408 	}
409 
410 	/* clear IntType */
411 	hi &= ~MASK_INT_TYPE_HI;
412 
413 	if (!tr->b->interrupt_capable)
414 		goto done;
415 
416 	if (tr->set_lvt_off) {
417 		if (lvt_off_valid(tr->b, tr->lvt_off, lo, hi)) {
418 			/* set new lvt offset */
419 			hi &= ~MASK_LVTOFF_HI;
420 			hi |= tr->lvt_off << 20;
421 		}
422 	}
423 
424 	if (tr->b->interrupt_enable)
425 		hi |= INT_TYPE_APIC;
426 
427  done:
428 
429 	hi |= MASK_COUNT_EN_HI;
430 	wrmsr(tr->b->address, lo, hi);
431 }
432 
433 static void mce_threshold_block_init(struct threshold_block *b, int offset)
434 {
435 	struct thresh_restart tr = {
436 		.b			= b,
437 		.set_lvt_off		= 1,
438 		.lvt_off		= offset,
439 	};
440 
441 	b->threshold_limit		= THRESHOLD_MAX;
442 	threshold_restart_bank(&tr);
443 };
444 
445 static int setup_APIC_mce_threshold(int reserved, int new)
446 {
447 	if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
448 					      APIC_EILVT_MSG_FIX, 0))
449 		return new;
450 
451 	return reserved;
452 }
453 
454 static int setup_APIC_deferred_error(int reserved, int new)
455 {
456 	if (reserved < 0 && !setup_APIC_eilvt(new, DEFERRED_ERROR_VECTOR,
457 					      APIC_EILVT_MSG_FIX, 0))
458 		return new;
459 
460 	return reserved;
461 }
462 
463 static void deferred_error_interrupt_enable(struct cpuinfo_x86 *c)
464 {
465 	u32 low = 0, high = 0;
466 	int def_offset = -1, def_new;
467 
468 	if (rdmsr_safe(MSR_CU_DEF_ERR, &low, &high))
469 		return;
470 
471 	def_new = (low & MASK_DEF_LVTOFF) >> 4;
472 	if (!(low & MASK_DEF_LVTOFF)) {
473 		pr_err(FW_BUG "Your BIOS is not setting up LVT offset 0x2 for deferred error IRQs correctly.\n");
474 		def_new = DEF_LVT_OFF;
475 		low = (low & ~MASK_DEF_LVTOFF) | (DEF_LVT_OFF << 4);
476 	}
477 
478 	def_offset = setup_APIC_deferred_error(def_offset, def_new);
479 	if ((def_offset == def_new) &&
480 	    (deferred_error_int_vector != amd_deferred_error_interrupt))
481 		deferred_error_int_vector = amd_deferred_error_interrupt;
482 
483 	if (!mce_flags.smca)
484 		low = (low & ~MASK_DEF_INT_TYPE) | DEF_INT_TYPE_APIC;
485 
486 	wrmsr(MSR_CU_DEF_ERR, low, high);
487 }
488 
489 static u32 smca_get_block_address(unsigned int bank, unsigned int block,
490 				  unsigned int cpu)
491 {
492 	if (!block)
493 		return MSR_AMD64_SMCA_MCx_MISC(bank);
494 
495 	if (!(per_cpu(smca_misc_banks_map, cpu) & BIT(bank)))
496 		return 0;
497 
498 	return MSR_AMD64_SMCA_MCx_MISCy(bank, block - 1);
499 }
500 
501 static u32 get_block_address(u32 current_addr, u32 low, u32 high,
502 			     unsigned int bank, unsigned int block,
503 			     unsigned int cpu)
504 {
505 	u32 addr = 0, offset = 0;
506 
507 	if ((bank >= per_cpu(mce_num_banks, cpu)) || (block >= NR_BLOCKS))
508 		return addr;
509 
510 	if (mce_flags.smca)
511 		return smca_get_block_address(bank, block, cpu);
512 
513 	/* Fall back to method we used for older processors: */
514 	switch (block) {
515 	case 0:
516 		addr = msr_ops.misc(bank);
517 		break;
518 	case 1:
519 		offset = ((low & MASK_BLKPTR_LO) >> 21);
520 		if (offset)
521 			addr = MCG_XBLK_ADDR + offset;
522 		break;
523 	default:
524 		addr = ++current_addr;
525 	}
526 	return addr;
527 }
528 
529 static int
530 prepare_threshold_block(unsigned int bank, unsigned int block, u32 addr,
531 			int offset, u32 misc_high)
532 {
533 	unsigned int cpu = smp_processor_id();
534 	u32 smca_low, smca_high;
535 	struct threshold_block b;
536 	int new;
537 
538 	if (!block)
539 		per_cpu(bank_map, cpu) |= (1 << bank);
540 
541 	memset(&b, 0, sizeof(b));
542 	b.cpu			= cpu;
543 	b.bank			= bank;
544 	b.block			= block;
545 	b.address		= addr;
546 	b.interrupt_capable	= lvt_interrupt_supported(bank, misc_high);
547 
548 	if (!b.interrupt_capable)
549 		goto done;
550 
551 	b.interrupt_enable = 1;
552 
553 	if (!mce_flags.smca) {
554 		new = (misc_high & MASK_LVTOFF_HI) >> 20;
555 		goto set_offset;
556 	}
557 
558 	/* Gather LVT offset for thresholding: */
559 	if (rdmsr_safe(MSR_CU_DEF_ERR, &smca_low, &smca_high))
560 		goto out;
561 
562 	new = (smca_low & SMCA_THR_LVT_OFF) >> 12;
563 
564 set_offset:
565 	offset = setup_APIC_mce_threshold(offset, new);
566 	if (offset == new)
567 		thresholding_irq_en = true;
568 
569 done:
570 	mce_threshold_block_init(&b, offset);
571 
572 out:
573 	return offset;
574 }
575 
576 bool amd_filter_mce(struct mce *m)
577 {
578 	enum smca_bank_types bank_type = smca_get_bank_type(m->bank);
579 	struct cpuinfo_x86 *c = &boot_cpu_data;
580 	u8 xec = (m->status >> 16) & 0x3F;
581 
582 	/* See Family 17h Models 10h-2Fh Erratum #1114. */
583 	if (c->x86 == 0x17 &&
584 	    c->x86_model >= 0x10 && c->x86_model <= 0x2F &&
585 	    bank_type == SMCA_IF && xec == 10)
586 		return true;
587 
588 	return false;
589 }
590 
591 /*
592  * Turn off thresholding banks for the following conditions:
593  * - MC4_MISC thresholding is not supported on Family 0x15.
594  * - Prevent possible spurious interrupts from the IF bank on Family 0x17
595  *   Models 0x10-0x2F due to Erratum #1114.
596  */
597 static void disable_err_thresholding(struct cpuinfo_x86 *c, unsigned int bank)
598 {
599 	int i, num_msrs;
600 	u64 hwcr;
601 	bool need_toggle;
602 	u32 msrs[NR_BLOCKS];
603 
604 	if (c->x86 == 0x15 && bank == 4) {
605 		msrs[0] = 0x00000413; /* MC4_MISC0 */
606 		msrs[1] = 0xc0000408; /* MC4_MISC1 */
607 		num_msrs = 2;
608 	} else if (c->x86 == 0x17 &&
609 		   (c->x86_model >= 0x10 && c->x86_model <= 0x2F)) {
610 
611 		if (smca_get_bank_type(bank) != SMCA_IF)
612 			return;
613 
614 		msrs[0] = MSR_AMD64_SMCA_MCx_MISC(bank);
615 		num_msrs = 1;
616 	} else {
617 		return;
618 	}
619 
620 	rdmsrl(MSR_K7_HWCR, hwcr);
621 
622 	/* McStatusWrEn has to be set */
623 	need_toggle = !(hwcr & BIT(18));
624 	if (need_toggle)
625 		wrmsrl(MSR_K7_HWCR, hwcr | BIT(18));
626 
627 	/* Clear CntP bit safely */
628 	for (i = 0; i < num_msrs; i++)
629 		msr_clear_bit(msrs[i], 62);
630 
631 	/* restore old settings */
632 	if (need_toggle)
633 		wrmsrl(MSR_K7_HWCR, hwcr);
634 }
635 
636 /* cpu init entry point, called from mce.c with preempt off */
637 void mce_amd_feature_init(struct cpuinfo_x86 *c)
638 {
639 	unsigned int bank, block, cpu = smp_processor_id();
640 	u32 low = 0, high = 0, address = 0;
641 	int offset = -1;
642 
643 
644 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
645 		if (mce_flags.smca)
646 			smca_configure(bank, cpu);
647 
648 		disable_err_thresholding(c, bank);
649 
650 		for (block = 0; block < NR_BLOCKS; ++block) {
651 			address = get_block_address(address, low, high, bank, block, cpu);
652 			if (!address)
653 				break;
654 
655 			if (rdmsr_safe(address, &low, &high))
656 				break;
657 
658 			if (!(high & MASK_VALID_HI))
659 				continue;
660 
661 			if (!(high & MASK_CNTP_HI)  ||
662 			     (high & MASK_LOCKED_HI))
663 				continue;
664 
665 			offset = prepare_threshold_block(bank, block, address, offset, high);
666 		}
667 	}
668 
669 	if (mce_flags.succor)
670 		deferred_error_interrupt_enable(c);
671 }
672 
673 int umc_normaddr_to_sysaddr(u64 norm_addr, u16 nid, u8 umc, u64 *sys_addr)
674 {
675 	u64 dram_base_addr, dram_limit_addr, dram_hole_base;
676 	/* We start from the normalized address */
677 	u64 ret_addr = norm_addr;
678 
679 	u32 tmp;
680 
681 	u8 die_id_shift, die_id_mask, socket_id_shift, socket_id_mask;
682 	u8 intlv_num_dies, intlv_num_chan, intlv_num_sockets;
683 	u8 intlv_addr_sel, intlv_addr_bit;
684 	u8 num_intlv_bits, hashed_bit;
685 	u8 lgcy_mmio_hole_en, base = 0;
686 	u8 cs_mask, cs_id = 0;
687 	bool hash_enabled = false;
688 
689 	/* Read D18F0x1B4 (DramOffset), check if base 1 is used. */
690 	if (amd_df_indirect_read(nid, 0, 0x1B4, umc, &tmp))
691 		goto out_err;
692 
693 	/* Remove HiAddrOffset from normalized address, if enabled: */
694 	if (tmp & BIT(0)) {
695 		u64 hi_addr_offset = (tmp & GENMASK_ULL(31, 20)) << 8;
696 
697 		if (norm_addr >= hi_addr_offset) {
698 			ret_addr -= hi_addr_offset;
699 			base = 1;
700 		}
701 	}
702 
703 	/* Read D18F0x110 (DramBaseAddress). */
704 	if (amd_df_indirect_read(nid, 0, 0x110 + (8 * base), umc, &tmp))
705 		goto out_err;
706 
707 	/* Check if address range is valid. */
708 	if (!(tmp & BIT(0))) {
709 		pr_err("%s: Invalid DramBaseAddress range: 0x%x.\n",
710 			__func__, tmp);
711 		goto out_err;
712 	}
713 
714 	lgcy_mmio_hole_en = tmp & BIT(1);
715 	intlv_num_chan	  = (tmp >> 4) & 0xF;
716 	intlv_addr_sel	  = (tmp >> 8) & 0x7;
717 	dram_base_addr	  = (tmp & GENMASK_ULL(31, 12)) << 16;
718 
719 	/* {0, 1, 2, 3} map to address bits {8, 9, 10, 11} respectively */
720 	if (intlv_addr_sel > 3) {
721 		pr_err("%s: Invalid interleave address select %d.\n",
722 			__func__, intlv_addr_sel);
723 		goto out_err;
724 	}
725 
726 	/* Read D18F0x114 (DramLimitAddress). */
727 	if (amd_df_indirect_read(nid, 0, 0x114 + (8 * base), umc, &tmp))
728 		goto out_err;
729 
730 	intlv_num_sockets = (tmp >> 8) & 0x1;
731 	intlv_num_dies	  = (tmp >> 10) & 0x3;
732 	dram_limit_addr	  = ((tmp & GENMASK_ULL(31, 12)) << 16) | GENMASK_ULL(27, 0);
733 
734 	intlv_addr_bit = intlv_addr_sel + 8;
735 
736 	/* Re-use intlv_num_chan by setting it equal to log2(#channels) */
737 	switch (intlv_num_chan) {
738 	case 0:	intlv_num_chan = 0; break;
739 	case 1: intlv_num_chan = 1; break;
740 	case 3: intlv_num_chan = 2; break;
741 	case 5:	intlv_num_chan = 3; break;
742 	case 7:	intlv_num_chan = 4; break;
743 
744 	case 8: intlv_num_chan = 1;
745 		hash_enabled = true;
746 		break;
747 	default:
748 		pr_err("%s: Invalid number of interleaved channels %d.\n",
749 			__func__, intlv_num_chan);
750 		goto out_err;
751 	}
752 
753 	num_intlv_bits = intlv_num_chan;
754 
755 	if (intlv_num_dies > 2) {
756 		pr_err("%s: Invalid number of interleaved nodes/dies %d.\n",
757 			__func__, intlv_num_dies);
758 		goto out_err;
759 	}
760 
761 	num_intlv_bits += intlv_num_dies;
762 
763 	/* Add a bit if sockets are interleaved. */
764 	num_intlv_bits += intlv_num_sockets;
765 
766 	/* Assert num_intlv_bits <= 4 */
767 	if (num_intlv_bits > 4) {
768 		pr_err("%s: Invalid interleave bits %d.\n",
769 			__func__, num_intlv_bits);
770 		goto out_err;
771 	}
772 
773 	if (num_intlv_bits > 0) {
774 		u64 temp_addr_x, temp_addr_i, temp_addr_y;
775 		u8 die_id_bit, sock_id_bit, cs_fabric_id;
776 
777 		/*
778 		 * Read FabricBlockInstanceInformation3_CS[BlockFabricID].
779 		 * This is the fabric id for this coherent slave. Use
780 		 * umc/channel# as instance id of the coherent slave
781 		 * for FICAA.
782 		 */
783 		if (amd_df_indirect_read(nid, 0, 0x50, umc, &tmp))
784 			goto out_err;
785 
786 		cs_fabric_id = (tmp >> 8) & 0xFF;
787 		die_id_bit   = 0;
788 
789 		/* If interleaved over more than 1 channel: */
790 		if (intlv_num_chan) {
791 			die_id_bit = intlv_num_chan;
792 			cs_mask	   = (1 << die_id_bit) - 1;
793 			cs_id	   = cs_fabric_id & cs_mask;
794 		}
795 
796 		sock_id_bit = die_id_bit;
797 
798 		/* Read D18F1x208 (SystemFabricIdMask). */
799 		if (intlv_num_dies || intlv_num_sockets)
800 			if (amd_df_indirect_read(nid, 1, 0x208, umc, &tmp))
801 				goto out_err;
802 
803 		/* If interleaved over more than 1 die. */
804 		if (intlv_num_dies) {
805 			sock_id_bit  = die_id_bit + intlv_num_dies;
806 			die_id_shift = (tmp >> 24) & 0xF;
807 			die_id_mask  = (tmp >> 8) & 0xFF;
808 
809 			cs_id |= ((cs_fabric_id & die_id_mask) >> die_id_shift) << die_id_bit;
810 		}
811 
812 		/* If interleaved over more than 1 socket. */
813 		if (intlv_num_sockets) {
814 			socket_id_shift	= (tmp >> 28) & 0xF;
815 			socket_id_mask	= (tmp >> 16) & 0xFF;
816 
817 			cs_id |= ((cs_fabric_id & socket_id_mask) >> socket_id_shift) << sock_id_bit;
818 		}
819 
820 		/*
821 		 * The pre-interleaved address consists of XXXXXXIIIYYYYY
822 		 * where III is the ID for this CS, and XXXXXXYYYYY are the
823 		 * address bits from the post-interleaved address.
824 		 * "num_intlv_bits" has been calculated to tell us how many "I"
825 		 * bits there are. "intlv_addr_bit" tells us how many "Y" bits
826 		 * there are (where "I" starts).
827 		 */
828 		temp_addr_y = ret_addr & GENMASK_ULL(intlv_addr_bit-1, 0);
829 		temp_addr_i = (cs_id << intlv_addr_bit);
830 		temp_addr_x = (ret_addr & GENMASK_ULL(63, intlv_addr_bit)) << num_intlv_bits;
831 		ret_addr    = temp_addr_x | temp_addr_i | temp_addr_y;
832 	}
833 
834 	/* Add dram base address */
835 	ret_addr += dram_base_addr;
836 
837 	/* If legacy MMIO hole enabled */
838 	if (lgcy_mmio_hole_en) {
839 		if (amd_df_indirect_read(nid, 0, 0x104, umc, &tmp))
840 			goto out_err;
841 
842 		dram_hole_base = tmp & GENMASK(31, 24);
843 		if (ret_addr >= dram_hole_base)
844 			ret_addr += (BIT_ULL(32) - dram_hole_base);
845 	}
846 
847 	if (hash_enabled) {
848 		/* Save some parentheses and grab ls-bit at the end. */
849 		hashed_bit =	(ret_addr >> 12) ^
850 				(ret_addr >> 18) ^
851 				(ret_addr >> 21) ^
852 				(ret_addr >> 30) ^
853 				cs_id;
854 
855 		hashed_bit &= BIT(0);
856 
857 		if (hashed_bit != ((ret_addr >> intlv_addr_bit) & BIT(0)))
858 			ret_addr ^= BIT(intlv_addr_bit);
859 	}
860 
861 	/* Is calculated system address is above DRAM limit address? */
862 	if (ret_addr > dram_limit_addr)
863 		goto out_err;
864 
865 	*sys_addr = ret_addr;
866 	return 0;
867 
868 out_err:
869 	return -EINVAL;
870 }
871 EXPORT_SYMBOL_GPL(umc_normaddr_to_sysaddr);
872 
873 bool amd_mce_is_memory_error(struct mce *m)
874 {
875 	/* ErrCodeExt[20:16] */
876 	u8 xec = (m->status >> 16) & 0x1f;
877 
878 	if (mce_flags.smca)
879 		return smca_get_bank_type(m->bank) == SMCA_UMC && xec == 0x0;
880 
881 	return m->bank == 4 && xec == 0x8;
882 }
883 
884 static void __log_error(unsigned int bank, u64 status, u64 addr, u64 misc)
885 {
886 	struct mce m;
887 
888 	mce_setup(&m);
889 
890 	m.status = status;
891 	m.misc   = misc;
892 	m.bank   = bank;
893 	m.tsc	 = rdtsc();
894 
895 	if (m.status & MCI_STATUS_ADDRV) {
896 		m.addr = addr;
897 
898 		/*
899 		 * Extract [55:<lsb>] where lsb is the least significant
900 		 * *valid* bit of the address bits.
901 		 */
902 		if (mce_flags.smca) {
903 			u8 lsb = (m.addr >> 56) & 0x3f;
904 
905 			m.addr &= GENMASK_ULL(55, lsb);
906 		}
907 	}
908 
909 	if (mce_flags.smca) {
910 		rdmsrl(MSR_AMD64_SMCA_MCx_IPID(bank), m.ipid);
911 
912 		if (m.status & MCI_STATUS_SYNDV)
913 			rdmsrl(MSR_AMD64_SMCA_MCx_SYND(bank), m.synd);
914 	}
915 
916 	mce_log(&m);
917 }
918 
919 asmlinkage __visible void __irq_entry smp_deferred_error_interrupt(struct pt_regs *regs)
920 {
921 	entering_irq();
922 	trace_deferred_error_apic_entry(DEFERRED_ERROR_VECTOR);
923 	inc_irq_stat(irq_deferred_error_count);
924 	deferred_error_int_vector();
925 	trace_deferred_error_apic_exit(DEFERRED_ERROR_VECTOR);
926 	exiting_ack_irq();
927 }
928 
929 /*
930  * Returns true if the logged error is deferred. False, otherwise.
931  */
932 static inline bool
933 _log_error_bank(unsigned int bank, u32 msr_stat, u32 msr_addr, u64 misc)
934 {
935 	u64 status, addr = 0;
936 
937 	rdmsrl(msr_stat, status);
938 	if (!(status & MCI_STATUS_VAL))
939 		return false;
940 
941 	if (status & MCI_STATUS_ADDRV)
942 		rdmsrl(msr_addr, addr);
943 
944 	__log_error(bank, status, addr, misc);
945 
946 	wrmsrl(msr_stat, 0);
947 
948 	return status & MCI_STATUS_DEFERRED;
949 }
950 
951 /*
952  * We have three scenarios for checking for Deferred errors:
953  *
954  * 1) Non-SMCA systems check MCA_STATUS and log error if found.
955  * 2) SMCA systems check MCA_STATUS. If error is found then log it and also
956  *    clear MCA_DESTAT.
957  * 3) SMCA systems check MCA_DESTAT, if error was not found in MCA_STATUS, and
958  *    log it.
959  */
960 static void log_error_deferred(unsigned int bank)
961 {
962 	bool defrd;
963 
964 	defrd = _log_error_bank(bank, msr_ops.status(bank),
965 					msr_ops.addr(bank), 0);
966 
967 	if (!mce_flags.smca)
968 		return;
969 
970 	/* Clear MCA_DESTAT if we logged the deferred error from MCA_STATUS. */
971 	if (defrd) {
972 		wrmsrl(MSR_AMD64_SMCA_MCx_DESTAT(bank), 0);
973 		return;
974 	}
975 
976 	/*
977 	 * Only deferred errors are logged in MCA_DE{STAT,ADDR} so just check
978 	 * for a valid error.
979 	 */
980 	_log_error_bank(bank, MSR_AMD64_SMCA_MCx_DESTAT(bank),
981 			      MSR_AMD64_SMCA_MCx_DEADDR(bank), 0);
982 }
983 
984 /* APIC interrupt handler for deferred errors */
985 static void amd_deferred_error_interrupt(void)
986 {
987 	unsigned int bank;
988 
989 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank)
990 		log_error_deferred(bank);
991 }
992 
993 static void log_error_thresholding(unsigned int bank, u64 misc)
994 {
995 	_log_error_bank(bank, msr_ops.status(bank), msr_ops.addr(bank), misc);
996 }
997 
998 static void log_and_reset_block(struct threshold_block *block)
999 {
1000 	struct thresh_restart tr;
1001 	u32 low = 0, high = 0;
1002 
1003 	if (!block)
1004 		return;
1005 
1006 	if (rdmsr_safe(block->address, &low, &high))
1007 		return;
1008 
1009 	if (!(high & MASK_OVERFLOW_HI))
1010 		return;
1011 
1012 	/* Log the MCE which caused the threshold event. */
1013 	log_error_thresholding(block->bank, ((u64)high << 32) | low);
1014 
1015 	/* Reset threshold block after logging error. */
1016 	memset(&tr, 0, sizeof(tr));
1017 	tr.b = block;
1018 	threshold_restart_bank(&tr);
1019 }
1020 
1021 /*
1022  * Threshold interrupt handler will service THRESHOLD_APIC_VECTOR. The interrupt
1023  * goes off when error_count reaches threshold_limit.
1024  */
1025 static void amd_threshold_interrupt(void)
1026 {
1027 	struct threshold_block *first_block = NULL, *block = NULL, *tmp = NULL;
1028 	struct threshold_bank **bp = this_cpu_read(threshold_banks);
1029 	unsigned int bank, cpu = smp_processor_id();
1030 
1031 	/*
1032 	 * Validate that the threshold bank has been initialized already. The
1033 	 * handler is installed at boot time, but on a hotplug event the
1034 	 * interrupt might fire before the data has been initialized.
1035 	 */
1036 	if (!bp)
1037 		return;
1038 
1039 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
1040 		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
1041 			continue;
1042 
1043 		first_block = bp[bank]->blocks;
1044 		if (!first_block)
1045 			continue;
1046 
1047 		/*
1048 		 * The first block is also the head of the list. Check it first
1049 		 * before iterating over the rest.
1050 		 */
1051 		log_and_reset_block(first_block);
1052 		list_for_each_entry_safe(block, tmp, &first_block->miscj, miscj)
1053 			log_and_reset_block(block);
1054 	}
1055 }
1056 
1057 /*
1058  * Sysfs Interface
1059  */
1060 
1061 struct threshold_attr {
1062 	struct attribute attr;
1063 	ssize_t (*show) (struct threshold_block *, char *);
1064 	ssize_t (*store) (struct threshold_block *, const char *, size_t count);
1065 };
1066 
1067 #define SHOW_FIELDS(name)						\
1068 static ssize_t show_ ## name(struct threshold_block *b, char *buf)	\
1069 {									\
1070 	return sprintf(buf, "%lu\n", (unsigned long) b->name);		\
1071 }
1072 SHOW_FIELDS(interrupt_enable)
1073 SHOW_FIELDS(threshold_limit)
1074 
1075 static ssize_t
1076 store_interrupt_enable(struct threshold_block *b, const char *buf, size_t size)
1077 {
1078 	struct thresh_restart tr;
1079 	unsigned long new;
1080 
1081 	if (!b->interrupt_capable)
1082 		return -EINVAL;
1083 
1084 	if (kstrtoul(buf, 0, &new) < 0)
1085 		return -EINVAL;
1086 
1087 	b->interrupt_enable = !!new;
1088 
1089 	memset(&tr, 0, sizeof(tr));
1090 	tr.b		= b;
1091 
1092 	if (smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1))
1093 		return -ENODEV;
1094 
1095 	return size;
1096 }
1097 
1098 static ssize_t
1099 store_threshold_limit(struct threshold_block *b, const char *buf, size_t size)
1100 {
1101 	struct thresh_restart tr;
1102 	unsigned long new;
1103 
1104 	if (kstrtoul(buf, 0, &new) < 0)
1105 		return -EINVAL;
1106 
1107 	if (new > THRESHOLD_MAX)
1108 		new = THRESHOLD_MAX;
1109 	if (new < 1)
1110 		new = 1;
1111 
1112 	memset(&tr, 0, sizeof(tr));
1113 	tr.old_limit = b->threshold_limit;
1114 	b->threshold_limit = new;
1115 	tr.b = b;
1116 
1117 	if (smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1))
1118 		return -ENODEV;
1119 
1120 	return size;
1121 }
1122 
1123 static ssize_t show_error_count(struct threshold_block *b, char *buf)
1124 {
1125 	u32 lo, hi;
1126 
1127 	/* CPU might be offline by now */
1128 	if (rdmsr_on_cpu(b->cpu, b->address, &lo, &hi))
1129 		return -ENODEV;
1130 
1131 	return sprintf(buf, "%u\n", ((hi & THRESHOLD_MAX) -
1132 				     (THRESHOLD_MAX - b->threshold_limit)));
1133 }
1134 
1135 static struct threshold_attr error_count = {
1136 	.attr = {.name = __stringify(error_count), .mode = 0444 },
1137 	.show = show_error_count,
1138 };
1139 
1140 #define RW_ATTR(val)							\
1141 static struct threshold_attr val = {					\
1142 	.attr	= {.name = __stringify(val), .mode = 0644 },		\
1143 	.show	= show_## val,						\
1144 	.store	= store_## val,						\
1145 };
1146 
1147 RW_ATTR(interrupt_enable);
1148 RW_ATTR(threshold_limit);
1149 
1150 static struct attribute *default_attrs[] = {
1151 	&threshold_limit.attr,
1152 	&error_count.attr,
1153 	NULL,	/* possibly interrupt_enable if supported, see below */
1154 	NULL,
1155 };
1156 
1157 #define to_block(k)	container_of(k, struct threshold_block, kobj)
1158 #define to_attr(a)	container_of(a, struct threshold_attr, attr)
1159 
1160 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
1161 {
1162 	struct threshold_block *b = to_block(kobj);
1163 	struct threshold_attr *a = to_attr(attr);
1164 	ssize_t ret;
1165 
1166 	ret = a->show ? a->show(b, buf) : -EIO;
1167 
1168 	return ret;
1169 }
1170 
1171 static ssize_t store(struct kobject *kobj, struct attribute *attr,
1172 		     const char *buf, size_t count)
1173 {
1174 	struct threshold_block *b = to_block(kobj);
1175 	struct threshold_attr *a = to_attr(attr);
1176 	ssize_t ret;
1177 
1178 	ret = a->store ? a->store(b, buf, count) : -EIO;
1179 
1180 	return ret;
1181 }
1182 
1183 static const struct sysfs_ops threshold_ops = {
1184 	.show			= show,
1185 	.store			= store,
1186 };
1187 
1188 static void threshold_block_release(struct kobject *kobj);
1189 
1190 static struct kobj_type threshold_ktype = {
1191 	.sysfs_ops		= &threshold_ops,
1192 	.default_attrs		= default_attrs,
1193 	.release		= threshold_block_release,
1194 };
1195 
1196 static const char *get_name(unsigned int bank, struct threshold_block *b)
1197 {
1198 	enum smca_bank_types bank_type;
1199 
1200 	if (!mce_flags.smca) {
1201 		if (b && bank == 4)
1202 			return bank4_names(b);
1203 
1204 		return th_names[bank];
1205 	}
1206 
1207 	bank_type = smca_get_bank_type(bank);
1208 	if (bank_type >= N_SMCA_BANK_TYPES)
1209 		return NULL;
1210 
1211 	if (b && bank_type == SMCA_UMC) {
1212 		if (b->block < ARRAY_SIZE(smca_umc_block_names))
1213 			return smca_umc_block_names[b->block];
1214 		return NULL;
1215 	}
1216 
1217 	if (smca_banks[bank].hwid->count == 1)
1218 		return smca_get_name(bank_type);
1219 
1220 	snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN,
1221 		 "%s_%x", smca_get_name(bank_type),
1222 			  smca_banks[bank].sysfs_id);
1223 	return buf_mcatype;
1224 }
1225 
1226 static int allocate_threshold_blocks(unsigned int cpu, struct threshold_bank *tb,
1227 				     unsigned int bank, unsigned int block,
1228 				     u32 address)
1229 {
1230 	struct threshold_block *b = NULL;
1231 	u32 low, high;
1232 	int err;
1233 
1234 	if ((bank >= this_cpu_read(mce_num_banks)) || (block >= NR_BLOCKS))
1235 		return 0;
1236 
1237 	if (rdmsr_safe(address, &low, &high))
1238 		return 0;
1239 
1240 	if (!(high & MASK_VALID_HI)) {
1241 		if (block)
1242 			goto recurse;
1243 		else
1244 			return 0;
1245 	}
1246 
1247 	if (!(high & MASK_CNTP_HI)  ||
1248 	     (high & MASK_LOCKED_HI))
1249 		goto recurse;
1250 
1251 	b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
1252 	if (!b)
1253 		return -ENOMEM;
1254 
1255 	b->block		= block;
1256 	b->bank			= bank;
1257 	b->cpu			= cpu;
1258 	b->address		= address;
1259 	b->interrupt_enable	= 0;
1260 	b->interrupt_capable	= lvt_interrupt_supported(bank, high);
1261 	b->threshold_limit	= THRESHOLD_MAX;
1262 
1263 	if (b->interrupt_capable) {
1264 		threshold_ktype.default_attrs[2] = &interrupt_enable.attr;
1265 		b->interrupt_enable = 1;
1266 	} else {
1267 		threshold_ktype.default_attrs[2] = NULL;
1268 	}
1269 
1270 	INIT_LIST_HEAD(&b->miscj);
1271 
1272 	/* This is safe as @tb is not visible yet */
1273 	if (tb->blocks)
1274 		list_add(&b->miscj, &tb->blocks->miscj);
1275 	else
1276 		tb->blocks = b;
1277 
1278 	err = kobject_init_and_add(&b->kobj, &threshold_ktype, tb->kobj, get_name(bank, b));
1279 	if (err)
1280 		goto out_free;
1281 recurse:
1282 	address = get_block_address(address, low, high, bank, ++block, cpu);
1283 	if (!address)
1284 		return 0;
1285 
1286 	err = allocate_threshold_blocks(cpu, tb, bank, block, address);
1287 	if (err)
1288 		goto out_free;
1289 
1290 	if (b)
1291 		kobject_uevent(&b->kobj, KOBJ_ADD);
1292 
1293 	return 0;
1294 
1295 out_free:
1296 	if (b) {
1297 		list_del(&b->miscj);
1298 		kobject_put(&b->kobj);
1299 	}
1300 	return err;
1301 }
1302 
1303 static int __threshold_add_blocks(struct threshold_bank *b)
1304 {
1305 	struct list_head *head = &b->blocks->miscj;
1306 	struct threshold_block *pos = NULL;
1307 	struct threshold_block *tmp = NULL;
1308 	int err = 0;
1309 
1310 	err = kobject_add(&b->blocks->kobj, b->kobj, b->blocks->kobj.name);
1311 	if (err)
1312 		return err;
1313 
1314 	list_for_each_entry_safe(pos, tmp, head, miscj) {
1315 
1316 		err = kobject_add(&pos->kobj, b->kobj, pos->kobj.name);
1317 		if (err) {
1318 			list_for_each_entry_safe_reverse(pos, tmp, head, miscj)
1319 				kobject_del(&pos->kobj);
1320 
1321 			return err;
1322 		}
1323 	}
1324 	return err;
1325 }
1326 
1327 static int threshold_create_bank(struct threshold_bank **bp, unsigned int cpu,
1328 				 unsigned int bank)
1329 {
1330 	struct device *dev = this_cpu_read(mce_device);
1331 	struct amd_northbridge *nb = NULL;
1332 	struct threshold_bank *b = NULL;
1333 	const char *name = get_name(bank, NULL);
1334 	int err = 0;
1335 
1336 	if (!dev)
1337 		return -ENODEV;
1338 
1339 	if (is_shared_bank(bank)) {
1340 		nb = node_to_amd_nb(amd_get_nb_id(cpu));
1341 
1342 		/* threshold descriptor already initialized on this node? */
1343 		if (nb && nb->bank4) {
1344 			/* yes, use it */
1345 			b = nb->bank4;
1346 			err = kobject_add(b->kobj, &dev->kobj, name);
1347 			if (err)
1348 				goto out;
1349 
1350 			bp[bank] = b;
1351 			refcount_inc(&b->cpus);
1352 
1353 			err = __threshold_add_blocks(b);
1354 
1355 			goto out;
1356 		}
1357 	}
1358 
1359 	b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
1360 	if (!b) {
1361 		err = -ENOMEM;
1362 		goto out;
1363 	}
1364 
1365 	/* Associate the bank with the per-CPU MCE device */
1366 	b->kobj = kobject_create_and_add(name, &dev->kobj);
1367 	if (!b->kobj) {
1368 		err = -EINVAL;
1369 		goto out_free;
1370 	}
1371 
1372 	if (is_shared_bank(bank)) {
1373 		b->shared = 1;
1374 		refcount_set(&b->cpus, 1);
1375 
1376 		/* nb is already initialized, see above */
1377 		if (nb) {
1378 			WARN_ON(nb->bank4);
1379 			nb->bank4 = b;
1380 		}
1381 	}
1382 
1383 	err = allocate_threshold_blocks(cpu, b, bank, 0, msr_ops.misc(bank));
1384 	if (err)
1385 		goto out_kobj;
1386 
1387 	bp[bank] = b;
1388 	return 0;
1389 
1390 out_kobj:
1391 	kobject_put(b->kobj);
1392 out_free:
1393 	kfree(b);
1394 out:
1395 	return err;
1396 }
1397 
1398 static void threshold_block_release(struct kobject *kobj)
1399 {
1400 	kfree(to_block(kobj));
1401 }
1402 
1403 static void deallocate_threshold_blocks(struct threshold_bank *bank)
1404 {
1405 	struct threshold_block *pos, *tmp;
1406 
1407 	list_for_each_entry_safe(pos, tmp, &bank->blocks->miscj, miscj) {
1408 		list_del(&pos->miscj);
1409 		kobject_put(&pos->kobj);
1410 	}
1411 
1412 	kobject_put(&bank->blocks->kobj);
1413 }
1414 
1415 static void __threshold_remove_blocks(struct threshold_bank *b)
1416 {
1417 	struct threshold_block *pos = NULL;
1418 	struct threshold_block *tmp = NULL;
1419 
1420 	kobject_del(b->kobj);
1421 
1422 	list_for_each_entry_safe(pos, tmp, &b->blocks->miscj, miscj)
1423 		kobject_del(&pos->kobj);
1424 }
1425 
1426 static void threshold_remove_bank(struct threshold_bank *bank)
1427 {
1428 	struct amd_northbridge *nb;
1429 
1430 	if (!bank->blocks)
1431 		goto out_free;
1432 
1433 	if (!bank->shared)
1434 		goto out_dealloc;
1435 
1436 	if (!refcount_dec_and_test(&bank->cpus)) {
1437 		__threshold_remove_blocks(bank);
1438 		return;
1439 	} else {
1440 		/*
1441 		 * The last CPU on this node using the shared bank is going
1442 		 * away, remove that bank now.
1443 		 */
1444 		nb = node_to_amd_nb(amd_get_nb_id(smp_processor_id()));
1445 		nb->bank4 = NULL;
1446 	}
1447 
1448 out_dealloc:
1449 	deallocate_threshold_blocks(bank);
1450 
1451 out_free:
1452 	kobject_put(bank->kobj);
1453 	kfree(bank);
1454 }
1455 
1456 int mce_threshold_remove_device(unsigned int cpu)
1457 {
1458 	struct threshold_bank **bp = this_cpu_read(threshold_banks);
1459 	unsigned int bank, numbanks = this_cpu_read(mce_num_banks);
1460 
1461 	if (!bp)
1462 		return 0;
1463 
1464 	/*
1465 	 * Clear the pointer before cleaning up, so that the interrupt won't
1466 	 * touch anything of this.
1467 	 */
1468 	this_cpu_write(threshold_banks, NULL);
1469 
1470 	for (bank = 0; bank < numbanks; bank++) {
1471 		if (bp[bank]) {
1472 			threshold_remove_bank(bp[bank]);
1473 			bp[bank] = NULL;
1474 		}
1475 	}
1476 	kfree(bp);
1477 	return 0;
1478 }
1479 
1480 /**
1481  * mce_threshold_create_device - Create the per-CPU MCE threshold device
1482  * @cpu:	The plugged in CPU
1483  *
1484  * Create directories and files for all valid threshold banks.
1485  *
1486  * This is invoked from the CPU hotplug callback which was installed in
1487  * mcheck_init_device(). The invocation happens in context of the hotplug
1488  * thread running on @cpu.  The callback is invoked on all CPUs which are
1489  * online when the callback is installed or during a real hotplug event.
1490  */
1491 int mce_threshold_create_device(unsigned int cpu)
1492 {
1493 	unsigned int numbanks, bank;
1494 	struct threshold_bank **bp;
1495 	int err;
1496 
1497 	if (!mce_flags.amd_threshold)
1498 		return 0;
1499 
1500 	bp = this_cpu_read(threshold_banks);
1501 	if (bp)
1502 		return 0;
1503 
1504 	numbanks = this_cpu_read(mce_num_banks);
1505 	bp = kcalloc(numbanks, sizeof(*bp), GFP_KERNEL);
1506 	if (!bp)
1507 		return -ENOMEM;
1508 
1509 	for (bank = 0; bank < numbanks; ++bank) {
1510 		if (!(this_cpu_read(bank_map) & (1 << bank)))
1511 			continue;
1512 		err = threshold_create_bank(bp, cpu, bank);
1513 		if (err)
1514 			goto out_err;
1515 	}
1516 	this_cpu_write(threshold_banks, bp);
1517 
1518 	if (thresholding_irq_en)
1519 		mce_threshold_vector = amd_threshold_interrupt;
1520 	return 0;
1521 out_err:
1522 	mce_threshold_remove_device(cpu);
1523 	return err;
1524 }
1525