xref: /openbmc/linux/arch/x86/kernel/cpu/mce/amd.c (revision 8121b8f9)
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 ... SMCA_LS_V2]	= { "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 ... SMCA_CS_V2]	= { "coherent_slave",	"Coherent Slave" },
89 	[SMCA_PIE]			= { "pie",		"Power, Interrupts, etc." },
90 
91 	/* UMC v2 is separate because both of them can exist in a single system. */
92 	[SMCA_UMC]			= { "umc",		"Unified Memory Controller" },
93 	[SMCA_UMC_V2]			= { "umc_v2",		"Unified Memory Controller v2" },
94 	[SMCA_PB]			= { "param_block",	"Parameter Block" },
95 	[SMCA_PSP ... SMCA_PSP_V2]	= { "psp",		"Platform Security Processor" },
96 	[SMCA_SMU ... SMCA_SMU_V2]	= { "smu",		"System Management Unit" },
97 	[SMCA_MP5]			= { "mp5",		"Microprocessor 5 Unit" },
98 	[SMCA_NBIO]			= { "nbio",		"Northbridge IO Unit" },
99 	[SMCA_PCIE ... SMCA_PCIE_V2]	= { "pcie",		"PCI Express Unit" },
100 	[SMCA_XGMI_PCS]			= { "xgmi_pcs",		"Ext Global Memory Interconnect PCS Unit" },
101 	[SMCA_XGMI_PHY]			= { "xgmi_phy",		"Ext Global Memory Interconnect PHY Unit" },
102 	[SMCA_WAFL_PHY]			= { "wafl_phy",		"WAFL PHY Unit" },
103 };
104 
105 static const char *smca_get_name(enum smca_bank_types t)
106 {
107 	if (t >= N_SMCA_BANK_TYPES)
108 		return NULL;
109 
110 	return smca_names[t].name;
111 }
112 
113 const char *smca_get_long_name(enum smca_bank_types t)
114 {
115 	if (t >= N_SMCA_BANK_TYPES)
116 		return NULL;
117 
118 	return smca_names[t].long_name;
119 }
120 EXPORT_SYMBOL_GPL(smca_get_long_name);
121 
122 static enum smca_bank_types smca_get_bank_type(unsigned int bank)
123 {
124 	struct smca_bank *b;
125 
126 	if (bank >= MAX_NR_BANKS)
127 		return N_SMCA_BANK_TYPES;
128 
129 	b = &smca_banks[bank];
130 	if (!b->hwid)
131 		return N_SMCA_BANK_TYPES;
132 
133 	return b->hwid->bank_type;
134 }
135 
136 static struct smca_hwid smca_hwid_mcatypes[] = {
137 	/* { bank_type, hwid_mcatype } */
138 
139 	/* Reserved type */
140 	{ SMCA_RESERVED, HWID_MCATYPE(0x00, 0x0)	},
141 
142 	/* ZN Core (HWID=0xB0) MCA types */
143 	{ SMCA_LS,	 HWID_MCATYPE(0xB0, 0x0)	},
144 	{ SMCA_LS_V2,	 HWID_MCATYPE(0xB0, 0x10)	},
145 	{ SMCA_IF,	 HWID_MCATYPE(0xB0, 0x1)	},
146 	{ SMCA_L2_CACHE, HWID_MCATYPE(0xB0, 0x2)	},
147 	{ SMCA_DE,	 HWID_MCATYPE(0xB0, 0x3)	},
148 	/* HWID 0xB0 MCATYPE 0x4 is Reserved */
149 	{ SMCA_EX,	 HWID_MCATYPE(0xB0, 0x5)	},
150 	{ SMCA_FP,	 HWID_MCATYPE(0xB0, 0x6)	},
151 	{ SMCA_L3_CACHE, HWID_MCATYPE(0xB0, 0x7)	},
152 
153 	/* Data Fabric MCA types */
154 	{ SMCA_CS,	 HWID_MCATYPE(0x2E, 0x0)	},
155 	{ SMCA_PIE,	 HWID_MCATYPE(0x2E, 0x1)	},
156 	{ SMCA_CS_V2,	 HWID_MCATYPE(0x2E, 0x2)	},
157 
158 	/* Unified Memory Controller MCA type */
159 	{ SMCA_UMC,	 HWID_MCATYPE(0x96, 0x0)	},
160 	{ SMCA_UMC_V2,	 HWID_MCATYPE(0x96, 0x1)	},
161 
162 	/* Parameter Block MCA type */
163 	{ SMCA_PB,	 HWID_MCATYPE(0x05, 0x0)	},
164 
165 	/* Platform Security Processor MCA type */
166 	{ SMCA_PSP,	 HWID_MCATYPE(0xFF, 0x0)	},
167 	{ SMCA_PSP_V2,	 HWID_MCATYPE(0xFF, 0x1)	},
168 
169 	/* System Management Unit MCA type */
170 	{ SMCA_SMU,	 HWID_MCATYPE(0x01, 0x0)	},
171 	{ SMCA_SMU_V2,	 HWID_MCATYPE(0x01, 0x1)	},
172 
173 	/* Microprocessor 5 Unit MCA type */
174 	{ SMCA_MP5,	 HWID_MCATYPE(0x01, 0x2)	},
175 
176 	/* Northbridge IO Unit MCA type */
177 	{ SMCA_NBIO,	 HWID_MCATYPE(0x18, 0x0)	},
178 
179 	/* PCI Express Unit MCA type */
180 	{ SMCA_PCIE,	 HWID_MCATYPE(0x46, 0x0)	},
181 	{ SMCA_PCIE_V2,	 HWID_MCATYPE(0x46, 0x1)	},
182 
183 	/* xGMI PCS MCA type */
184 	{ SMCA_XGMI_PCS, HWID_MCATYPE(0x50, 0x0)	},
185 
186 	/* xGMI PHY MCA type */
187 	{ SMCA_XGMI_PHY, HWID_MCATYPE(0x259, 0x0)	},
188 
189 	/* WAFL PHY MCA type */
190 	{ SMCA_WAFL_PHY, HWID_MCATYPE(0x267, 0x0)	},
191 };
192 
193 struct smca_bank smca_banks[MAX_NR_BANKS];
194 EXPORT_SYMBOL_GPL(smca_banks);
195 
196 /*
197  * In SMCA enabled processors, we can have multiple banks for a given IP type.
198  * So to define a unique name for each bank, we use a temp c-string to append
199  * the MCA_IPID[InstanceId] to type's name in get_name().
200  *
201  * InstanceId is 32 bits which is 8 characters. Make sure MAX_MCATYPE_NAME_LEN
202  * is greater than 8 plus 1 (for underscore) plus length of longest type name.
203  */
204 #define MAX_MCATYPE_NAME_LEN	30
205 static char buf_mcatype[MAX_MCATYPE_NAME_LEN];
206 
207 static DEFINE_PER_CPU(struct threshold_bank **, threshold_banks);
208 
209 /*
210  * A list of the banks enabled on each logical CPU. Controls which respective
211  * descriptors to initialize later in mce_threshold_create_device().
212  */
213 static DEFINE_PER_CPU(unsigned int, bank_map);
214 
215 /* Map of banks that have more than MCA_MISC0 available. */
216 static DEFINE_PER_CPU(u32, smca_misc_banks_map);
217 
218 static void amd_threshold_interrupt(void);
219 static void amd_deferred_error_interrupt(void);
220 
221 static void default_deferred_error_interrupt(void)
222 {
223 	pr_err("Unexpected deferred interrupt at vector %x\n", DEFERRED_ERROR_VECTOR);
224 }
225 void (*deferred_error_int_vector)(void) = default_deferred_error_interrupt;
226 
227 static void smca_set_misc_banks_map(unsigned int bank, unsigned int cpu)
228 {
229 	u32 low, high;
230 
231 	/*
232 	 * For SMCA enabled processors, BLKPTR field of the first MISC register
233 	 * (MCx_MISC0) indicates presence of additional MISC regs set (MISC1-4).
234 	 */
235 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_CONFIG(bank), &low, &high))
236 		return;
237 
238 	if (!(low & MCI_CONFIG_MCAX))
239 		return;
240 
241 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_MISC(bank), &low, &high))
242 		return;
243 
244 	if (low & MASK_BLKPTR_LO)
245 		per_cpu(smca_misc_banks_map, cpu) |= BIT(bank);
246 
247 }
248 
249 static void smca_configure(unsigned int bank, unsigned int cpu)
250 {
251 	unsigned int i, hwid_mcatype;
252 	struct smca_hwid *s_hwid;
253 	u32 high, low;
254 	u32 smca_config = MSR_AMD64_SMCA_MCx_CONFIG(bank);
255 
256 	/* Set appropriate bits in MCA_CONFIG */
257 	if (!rdmsr_safe(smca_config, &low, &high)) {
258 		/*
259 		 * OS is required to set the MCAX bit to acknowledge that it is
260 		 * now using the new MSR ranges and new registers under each
261 		 * bank. It also means that the OS will configure deferred
262 		 * errors in the new MCx_CONFIG register. If the bit is not set,
263 		 * uncorrectable errors will cause a system panic.
264 		 *
265 		 * MCA_CONFIG[MCAX] is bit 32 (0 in the high portion of the MSR.)
266 		 */
267 		high |= BIT(0);
268 
269 		/*
270 		 * SMCA sets the Deferred Error Interrupt type per bank.
271 		 *
272 		 * MCA_CONFIG[DeferredIntTypeSupported] is bit 5, and tells us
273 		 * if the DeferredIntType bit field is available.
274 		 *
275 		 * MCA_CONFIG[DeferredIntType] is bits [38:37] ([6:5] in the
276 		 * high portion of the MSR). OS should set this to 0x1 to enable
277 		 * APIC based interrupt. First, check that no interrupt has been
278 		 * set.
279 		 */
280 		if ((low & BIT(5)) && !((high >> 5) & 0x3))
281 			high |= BIT(5);
282 
283 		wrmsr(smca_config, low, high);
284 	}
285 
286 	smca_set_misc_banks_map(bank, cpu);
287 
288 	/* Return early if this bank was already initialized. */
289 	if (smca_banks[bank].hwid && smca_banks[bank].hwid->hwid_mcatype != 0)
290 		return;
291 
292 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_IPID(bank), &low, &high)) {
293 		pr_warn("Failed to read MCA_IPID for bank %d\n", bank);
294 		return;
295 	}
296 
297 	hwid_mcatype = HWID_MCATYPE(high & MCI_IPID_HWID,
298 				    (high & MCI_IPID_MCATYPE) >> 16);
299 
300 	for (i = 0; i < ARRAY_SIZE(smca_hwid_mcatypes); i++) {
301 		s_hwid = &smca_hwid_mcatypes[i];
302 		if (hwid_mcatype == s_hwid->hwid_mcatype) {
303 			smca_banks[bank].hwid = s_hwid;
304 			smca_banks[bank].id = low;
305 			smca_banks[bank].sysfs_id = s_hwid->count++;
306 			break;
307 		}
308 	}
309 }
310 
311 struct thresh_restart {
312 	struct threshold_block	*b;
313 	int			reset;
314 	int			set_lvt_off;
315 	int			lvt_off;
316 	u16			old_limit;
317 };
318 
319 static inline bool is_shared_bank(int bank)
320 {
321 	/*
322 	 * Scalable MCA provides for only one core to have access to the MSRs of
323 	 * a shared bank.
324 	 */
325 	if (mce_flags.smca)
326 		return false;
327 
328 	/* Bank 4 is for northbridge reporting and is thus shared */
329 	return (bank == 4);
330 }
331 
332 static const char *bank4_names(const struct threshold_block *b)
333 {
334 	switch (b->address) {
335 	/* MSR4_MISC0 */
336 	case 0x00000413:
337 		return "dram";
338 
339 	case 0xc0000408:
340 		return "ht_links";
341 
342 	case 0xc0000409:
343 		return "l3_cache";
344 
345 	default:
346 		WARN(1, "Funny MSR: 0x%08x\n", b->address);
347 		return "";
348 	}
349 };
350 
351 
352 static bool lvt_interrupt_supported(unsigned int bank, u32 msr_high_bits)
353 {
354 	/*
355 	 * bank 4 supports APIC LVT interrupts implicitly since forever.
356 	 */
357 	if (bank == 4)
358 		return true;
359 
360 	/*
361 	 * IntP: interrupt present; if this bit is set, the thresholding
362 	 * bank can generate APIC LVT interrupts
363 	 */
364 	return msr_high_bits & BIT(28);
365 }
366 
367 static int lvt_off_valid(struct threshold_block *b, int apic, u32 lo, u32 hi)
368 {
369 	int msr = (hi & MASK_LVTOFF_HI) >> 20;
370 
371 	if (apic < 0) {
372 		pr_err(FW_BUG "cpu %d, failed to setup threshold interrupt "
373 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n", b->cpu,
374 		       b->bank, b->block, b->address, hi, lo);
375 		return 0;
376 	}
377 
378 	if (apic != msr) {
379 		/*
380 		 * On SMCA CPUs, LVT offset is programmed at a different MSR, and
381 		 * the BIOS provides the value. The original field where LVT offset
382 		 * was set is reserved. Return early here:
383 		 */
384 		if (mce_flags.smca)
385 			return 0;
386 
387 		pr_err(FW_BUG "cpu %d, invalid threshold interrupt offset %d "
388 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n",
389 		       b->cpu, apic, b->bank, b->block, b->address, hi, lo);
390 		return 0;
391 	}
392 
393 	return 1;
394 };
395 
396 /* Reprogram MCx_MISC MSR behind this threshold bank. */
397 static void threshold_restart_bank(void *_tr)
398 {
399 	struct thresh_restart *tr = _tr;
400 	u32 hi, lo;
401 
402 	/* sysfs write might race against an offline operation */
403 	if (this_cpu_read(threshold_banks))
404 		return;
405 
406 	rdmsr(tr->b->address, lo, hi);
407 
408 	if (tr->b->threshold_limit < (hi & THRESHOLD_MAX))
409 		tr->reset = 1;	/* limit cannot be lower than err count */
410 
411 	if (tr->reset) {		/* reset err count and overflow bit */
412 		hi =
413 		    (hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
414 		    (THRESHOLD_MAX - tr->b->threshold_limit);
415 	} else if (tr->old_limit) {	/* change limit w/o reset */
416 		int new_count = (hi & THRESHOLD_MAX) +
417 		    (tr->old_limit - tr->b->threshold_limit);
418 
419 		hi = (hi & ~MASK_ERR_COUNT_HI) |
420 		    (new_count & THRESHOLD_MAX);
421 	}
422 
423 	/* clear IntType */
424 	hi &= ~MASK_INT_TYPE_HI;
425 
426 	if (!tr->b->interrupt_capable)
427 		goto done;
428 
429 	if (tr->set_lvt_off) {
430 		if (lvt_off_valid(tr->b, tr->lvt_off, lo, hi)) {
431 			/* set new lvt offset */
432 			hi &= ~MASK_LVTOFF_HI;
433 			hi |= tr->lvt_off << 20;
434 		}
435 	}
436 
437 	if (tr->b->interrupt_enable)
438 		hi |= INT_TYPE_APIC;
439 
440  done:
441 
442 	hi |= MASK_COUNT_EN_HI;
443 	wrmsr(tr->b->address, lo, hi);
444 }
445 
446 static void mce_threshold_block_init(struct threshold_block *b, int offset)
447 {
448 	struct thresh_restart tr = {
449 		.b			= b,
450 		.set_lvt_off		= 1,
451 		.lvt_off		= offset,
452 	};
453 
454 	b->threshold_limit		= THRESHOLD_MAX;
455 	threshold_restart_bank(&tr);
456 };
457 
458 static int setup_APIC_mce_threshold(int reserved, int new)
459 {
460 	if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
461 					      APIC_EILVT_MSG_FIX, 0))
462 		return new;
463 
464 	return reserved;
465 }
466 
467 static int setup_APIC_deferred_error(int reserved, int new)
468 {
469 	if (reserved < 0 && !setup_APIC_eilvt(new, DEFERRED_ERROR_VECTOR,
470 					      APIC_EILVT_MSG_FIX, 0))
471 		return new;
472 
473 	return reserved;
474 }
475 
476 static void deferred_error_interrupt_enable(struct cpuinfo_x86 *c)
477 {
478 	u32 low = 0, high = 0;
479 	int def_offset = -1, def_new;
480 
481 	if (rdmsr_safe(MSR_CU_DEF_ERR, &low, &high))
482 		return;
483 
484 	def_new = (low & MASK_DEF_LVTOFF) >> 4;
485 	if (!(low & MASK_DEF_LVTOFF)) {
486 		pr_err(FW_BUG "Your BIOS is not setting up LVT offset 0x2 for deferred error IRQs correctly.\n");
487 		def_new = DEF_LVT_OFF;
488 		low = (low & ~MASK_DEF_LVTOFF) | (DEF_LVT_OFF << 4);
489 	}
490 
491 	def_offset = setup_APIC_deferred_error(def_offset, def_new);
492 	if ((def_offset == def_new) &&
493 	    (deferred_error_int_vector != amd_deferred_error_interrupt))
494 		deferred_error_int_vector = amd_deferred_error_interrupt;
495 
496 	if (!mce_flags.smca)
497 		low = (low & ~MASK_DEF_INT_TYPE) | DEF_INT_TYPE_APIC;
498 
499 	wrmsr(MSR_CU_DEF_ERR, low, high);
500 }
501 
502 static u32 smca_get_block_address(unsigned int bank, unsigned int block,
503 				  unsigned int cpu)
504 {
505 	if (!block)
506 		return MSR_AMD64_SMCA_MCx_MISC(bank);
507 
508 	if (!(per_cpu(smca_misc_banks_map, cpu) & BIT(bank)))
509 		return 0;
510 
511 	return MSR_AMD64_SMCA_MCx_MISCy(bank, block - 1);
512 }
513 
514 static u32 get_block_address(u32 current_addr, u32 low, u32 high,
515 			     unsigned int bank, unsigned int block,
516 			     unsigned int cpu)
517 {
518 	u32 addr = 0, offset = 0;
519 
520 	if ((bank >= per_cpu(mce_num_banks, cpu)) || (block >= NR_BLOCKS))
521 		return addr;
522 
523 	if (mce_flags.smca)
524 		return smca_get_block_address(bank, block, cpu);
525 
526 	/* Fall back to method we used for older processors: */
527 	switch (block) {
528 	case 0:
529 		addr = mca_msr_reg(bank, MCA_MISC);
530 		break;
531 	case 1:
532 		offset = ((low & MASK_BLKPTR_LO) >> 21);
533 		if (offset)
534 			addr = MCG_XBLK_ADDR + offset;
535 		break;
536 	default:
537 		addr = ++current_addr;
538 	}
539 	return addr;
540 }
541 
542 static int
543 prepare_threshold_block(unsigned int bank, unsigned int block, u32 addr,
544 			int offset, u32 misc_high)
545 {
546 	unsigned int cpu = smp_processor_id();
547 	u32 smca_low, smca_high;
548 	struct threshold_block b;
549 	int new;
550 
551 	if (!block)
552 		per_cpu(bank_map, cpu) |= (1 << bank);
553 
554 	memset(&b, 0, sizeof(b));
555 	b.cpu			= cpu;
556 	b.bank			= bank;
557 	b.block			= block;
558 	b.address		= addr;
559 	b.interrupt_capable	= lvt_interrupt_supported(bank, misc_high);
560 
561 	if (!b.interrupt_capable)
562 		goto done;
563 
564 	b.interrupt_enable = 1;
565 
566 	if (!mce_flags.smca) {
567 		new = (misc_high & MASK_LVTOFF_HI) >> 20;
568 		goto set_offset;
569 	}
570 
571 	/* Gather LVT offset for thresholding: */
572 	if (rdmsr_safe(MSR_CU_DEF_ERR, &smca_low, &smca_high))
573 		goto out;
574 
575 	new = (smca_low & SMCA_THR_LVT_OFF) >> 12;
576 
577 set_offset:
578 	offset = setup_APIC_mce_threshold(offset, new);
579 	if (offset == new)
580 		thresholding_irq_en = true;
581 
582 done:
583 	mce_threshold_block_init(&b, offset);
584 
585 out:
586 	return offset;
587 }
588 
589 bool amd_filter_mce(struct mce *m)
590 {
591 	enum smca_bank_types bank_type = smca_get_bank_type(m->bank);
592 	struct cpuinfo_x86 *c = &boot_cpu_data;
593 
594 	/* See Family 17h Models 10h-2Fh Erratum #1114. */
595 	if (c->x86 == 0x17 &&
596 	    c->x86_model >= 0x10 && c->x86_model <= 0x2F &&
597 	    bank_type == SMCA_IF && XEC(m->status, 0x3f) == 10)
598 		return true;
599 
600 	/* NB GART TLB error reporting is disabled by default. */
601 	if (c->x86 < 0x17) {
602 		if (m->bank == 4 && XEC(m->status, 0x1f) == 0x5)
603 			return true;
604 	}
605 
606 	return false;
607 }
608 
609 /*
610  * Turn off thresholding banks for the following conditions:
611  * - MC4_MISC thresholding is not supported on Family 0x15.
612  * - Prevent possible spurious interrupts from the IF bank on Family 0x17
613  *   Models 0x10-0x2F due to Erratum #1114.
614  */
615 static void disable_err_thresholding(struct cpuinfo_x86 *c, unsigned int bank)
616 {
617 	int i, num_msrs;
618 	u64 hwcr;
619 	bool need_toggle;
620 	u32 msrs[NR_BLOCKS];
621 
622 	if (c->x86 == 0x15 && bank == 4) {
623 		msrs[0] = 0x00000413; /* MC4_MISC0 */
624 		msrs[1] = 0xc0000408; /* MC4_MISC1 */
625 		num_msrs = 2;
626 	} else if (c->x86 == 0x17 &&
627 		   (c->x86_model >= 0x10 && c->x86_model <= 0x2F)) {
628 
629 		if (smca_get_bank_type(bank) != SMCA_IF)
630 			return;
631 
632 		msrs[0] = MSR_AMD64_SMCA_MCx_MISC(bank);
633 		num_msrs = 1;
634 	} else {
635 		return;
636 	}
637 
638 	rdmsrl(MSR_K7_HWCR, hwcr);
639 
640 	/* McStatusWrEn has to be set */
641 	need_toggle = !(hwcr & BIT(18));
642 	if (need_toggle)
643 		wrmsrl(MSR_K7_HWCR, hwcr | BIT(18));
644 
645 	/* Clear CntP bit safely */
646 	for (i = 0; i < num_msrs; i++)
647 		msr_clear_bit(msrs[i], 62);
648 
649 	/* restore old settings */
650 	if (need_toggle)
651 		wrmsrl(MSR_K7_HWCR, hwcr);
652 }
653 
654 /* cpu init entry point, called from mce.c with preempt off */
655 void mce_amd_feature_init(struct cpuinfo_x86 *c)
656 {
657 	unsigned int bank, block, cpu = smp_processor_id();
658 	u32 low = 0, high = 0, address = 0;
659 	int offset = -1;
660 
661 
662 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
663 		if (mce_flags.smca)
664 			smca_configure(bank, cpu);
665 
666 		disable_err_thresholding(c, bank);
667 
668 		for (block = 0; block < NR_BLOCKS; ++block) {
669 			address = get_block_address(address, low, high, bank, block, cpu);
670 			if (!address)
671 				break;
672 
673 			if (rdmsr_safe(address, &low, &high))
674 				break;
675 
676 			if (!(high & MASK_VALID_HI))
677 				continue;
678 
679 			if (!(high & MASK_CNTP_HI)  ||
680 			     (high & MASK_LOCKED_HI))
681 				continue;
682 
683 			offset = prepare_threshold_block(bank, block, address, offset, high);
684 		}
685 	}
686 
687 	if (mce_flags.succor)
688 		deferred_error_interrupt_enable(c);
689 }
690 
691 int umc_normaddr_to_sysaddr(u64 norm_addr, u16 nid, u8 umc, u64 *sys_addr)
692 {
693 	u64 dram_base_addr, dram_limit_addr, dram_hole_base;
694 	/* We start from the normalized address */
695 	u64 ret_addr = norm_addr;
696 
697 	u32 tmp;
698 
699 	u8 die_id_shift, die_id_mask, socket_id_shift, socket_id_mask;
700 	u8 intlv_num_dies, intlv_num_chan, intlv_num_sockets;
701 	u8 intlv_addr_sel, intlv_addr_bit;
702 	u8 num_intlv_bits, hashed_bit;
703 	u8 lgcy_mmio_hole_en, base = 0;
704 	u8 cs_mask, cs_id = 0;
705 	bool hash_enabled = false;
706 
707 	/* Read D18F0x1B4 (DramOffset), check if base 1 is used. */
708 	if (amd_df_indirect_read(nid, 0, 0x1B4, umc, &tmp))
709 		goto out_err;
710 
711 	/* Remove HiAddrOffset from normalized address, if enabled: */
712 	if (tmp & BIT(0)) {
713 		u64 hi_addr_offset = (tmp & GENMASK_ULL(31, 20)) << 8;
714 
715 		if (norm_addr >= hi_addr_offset) {
716 			ret_addr -= hi_addr_offset;
717 			base = 1;
718 		}
719 	}
720 
721 	/* Read D18F0x110 (DramBaseAddress). */
722 	if (amd_df_indirect_read(nid, 0, 0x110 + (8 * base), umc, &tmp))
723 		goto out_err;
724 
725 	/* Check if address range is valid. */
726 	if (!(tmp & BIT(0))) {
727 		pr_err("%s: Invalid DramBaseAddress range: 0x%x.\n",
728 			__func__, tmp);
729 		goto out_err;
730 	}
731 
732 	lgcy_mmio_hole_en = tmp & BIT(1);
733 	intlv_num_chan	  = (tmp >> 4) & 0xF;
734 	intlv_addr_sel	  = (tmp >> 8) & 0x7;
735 	dram_base_addr	  = (tmp & GENMASK_ULL(31, 12)) << 16;
736 
737 	/* {0, 1, 2, 3} map to address bits {8, 9, 10, 11} respectively */
738 	if (intlv_addr_sel > 3) {
739 		pr_err("%s: Invalid interleave address select %d.\n",
740 			__func__, intlv_addr_sel);
741 		goto out_err;
742 	}
743 
744 	/* Read D18F0x114 (DramLimitAddress). */
745 	if (amd_df_indirect_read(nid, 0, 0x114 + (8 * base), umc, &tmp))
746 		goto out_err;
747 
748 	intlv_num_sockets = (tmp >> 8) & 0x1;
749 	intlv_num_dies	  = (tmp >> 10) & 0x3;
750 	dram_limit_addr	  = ((tmp & GENMASK_ULL(31, 12)) << 16) | GENMASK_ULL(27, 0);
751 
752 	intlv_addr_bit = intlv_addr_sel + 8;
753 
754 	/* Re-use intlv_num_chan by setting it equal to log2(#channels) */
755 	switch (intlv_num_chan) {
756 	case 0:	intlv_num_chan = 0; break;
757 	case 1: intlv_num_chan = 1; break;
758 	case 3: intlv_num_chan = 2; break;
759 	case 5:	intlv_num_chan = 3; break;
760 	case 7:	intlv_num_chan = 4; break;
761 
762 	case 8: intlv_num_chan = 1;
763 		hash_enabled = true;
764 		break;
765 	default:
766 		pr_err("%s: Invalid number of interleaved channels %d.\n",
767 			__func__, intlv_num_chan);
768 		goto out_err;
769 	}
770 
771 	num_intlv_bits = intlv_num_chan;
772 
773 	if (intlv_num_dies > 2) {
774 		pr_err("%s: Invalid number of interleaved nodes/dies %d.\n",
775 			__func__, intlv_num_dies);
776 		goto out_err;
777 	}
778 
779 	num_intlv_bits += intlv_num_dies;
780 
781 	/* Add a bit if sockets are interleaved. */
782 	num_intlv_bits += intlv_num_sockets;
783 
784 	/* Assert num_intlv_bits <= 4 */
785 	if (num_intlv_bits > 4) {
786 		pr_err("%s: Invalid interleave bits %d.\n",
787 			__func__, num_intlv_bits);
788 		goto out_err;
789 	}
790 
791 	if (num_intlv_bits > 0) {
792 		u64 temp_addr_x, temp_addr_i, temp_addr_y;
793 		u8 die_id_bit, sock_id_bit, cs_fabric_id;
794 
795 		/*
796 		 * Read FabricBlockInstanceInformation3_CS[BlockFabricID].
797 		 * This is the fabric id for this coherent slave. Use
798 		 * umc/channel# as instance id of the coherent slave
799 		 * for FICAA.
800 		 */
801 		if (amd_df_indirect_read(nid, 0, 0x50, umc, &tmp))
802 			goto out_err;
803 
804 		cs_fabric_id = (tmp >> 8) & 0xFF;
805 		die_id_bit   = 0;
806 
807 		/* If interleaved over more than 1 channel: */
808 		if (intlv_num_chan) {
809 			die_id_bit = intlv_num_chan;
810 			cs_mask	   = (1 << die_id_bit) - 1;
811 			cs_id	   = cs_fabric_id & cs_mask;
812 		}
813 
814 		sock_id_bit = die_id_bit;
815 
816 		/* Read D18F1x208 (SystemFabricIdMask). */
817 		if (intlv_num_dies || intlv_num_sockets)
818 			if (amd_df_indirect_read(nid, 1, 0x208, umc, &tmp))
819 				goto out_err;
820 
821 		/* If interleaved over more than 1 die. */
822 		if (intlv_num_dies) {
823 			sock_id_bit  = die_id_bit + intlv_num_dies;
824 			die_id_shift = (tmp >> 24) & 0xF;
825 			die_id_mask  = (tmp >> 8) & 0xFF;
826 
827 			cs_id |= ((cs_fabric_id & die_id_mask) >> die_id_shift) << die_id_bit;
828 		}
829 
830 		/* If interleaved over more than 1 socket. */
831 		if (intlv_num_sockets) {
832 			socket_id_shift	= (tmp >> 28) & 0xF;
833 			socket_id_mask	= (tmp >> 16) & 0xFF;
834 
835 			cs_id |= ((cs_fabric_id & socket_id_mask) >> socket_id_shift) << sock_id_bit;
836 		}
837 
838 		/*
839 		 * The pre-interleaved address consists of XXXXXXIIIYYYYY
840 		 * where III is the ID for this CS, and XXXXXXYYYYY are the
841 		 * address bits from the post-interleaved address.
842 		 * "num_intlv_bits" has been calculated to tell us how many "I"
843 		 * bits there are. "intlv_addr_bit" tells us how many "Y" bits
844 		 * there are (where "I" starts).
845 		 */
846 		temp_addr_y = ret_addr & GENMASK_ULL(intlv_addr_bit-1, 0);
847 		temp_addr_i = (cs_id << intlv_addr_bit);
848 		temp_addr_x = (ret_addr & GENMASK_ULL(63, intlv_addr_bit)) << num_intlv_bits;
849 		ret_addr    = temp_addr_x | temp_addr_i | temp_addr_y;
850 	}
851 
852 	/* Add dram base address */
853 	ret_addr += dram_base_addr;
854 
855 	/* If legacy MMIO hole enabled */
856 	if (lgcy_mmio_hole_en) {
857 		if (amd_df_indirect_read(nid, 0, 0x104, umc, &tmp))
858 			goto out_err;
859 
860 		dram_hole_base = tmp & GENMASK(31, 24);
861 		if (ret_addr >= dram_hole_base)
862 			ret_addr += (BIT_ULL(32) - dram_hole_base);
863 	}
864 
865 	if (hash_enabled) {
866 		/* Save some parentheses and grab ls-bit at the end. */
867 		hashed_bit =	(ret_addr >> 12) ^
868 				(ret_addr >> 18) ^
869 				(ret_addr >> 21) ^
870 				(ret_addr >> 30) ^
871 				cs_id;
872 
873 		hashed_bit &= BIT(0);
874 
875 		if (hashed_bit != ((ret_addr >> intlv_addr_bit) & BIT(0)))
876 			ret_addr ^= BIT(intlv_addr_bit);
877 	}
878 
879 	/* Is calculated system address is above DRAM limit address? */
880 	if (ret_addr > dram_limit_addr)
881 		goto out_err;
882 
883 	*sys_addr = ret_addr;
884 	return 0;
885 
886 out_err:
887 	return -EINVAL;
888 }
889 EXPORT_SYMBOL_GPL(umc_normaddr_to_sysaddr);
890 
891 bool amd_mce_is_memory_error(struct mce *m)
892 {
893 	/* ErrCodeExt[20:16] */
894 	u8 xec = (m->status >> 16) & 0x1f;
895 
896 	if (mce_flags.smca)
897 		return smca_get_bank_type(m->bank) == SMCA_UMC && xec == 0x0;
898 
899 	return m->bank == 4 && xec == 0x8;
900 }
901 
902 static void __log_error(unsigned int bank, u64 status, u64 addr, u64 misc)
903 {
904 	struct mce m;
905 
906 	mce_setup(&m);
907 
908 	m.status = status;
909 	m.misc   = misc;
910 	m.bank   = bank;
911 	m.tsc	 = rdtsc();
912 
913 	if (m.status & MCI_STATUS_ADDRV) {
914 		m.addr = addr;
915 
916 		/*
917 		 * Extract [55:<lsb>] where lsb is the least significant
918 		 * *valid* bit of the address bits.
919 		 */
920 		if (mce_flags.smca) {
921 			u8 lsb = (m.addr >> 56) & 0x3f;
922 
923 			m.addr &= GENMASK_ULL(55, lsb);
924 		}
925 	}
926 
927 	if (mce_flags.smca) {
928 		rdmsrl(MSR_AMD64_SMCA_MCx_IPID(bank), m.ipid);
929 
930 		if (m.status & MCI_STATUS_SYNDV)
931 			rdmsrl(MSR_AMD64_SMCA_MCx_SYND(bank), m.synd);
932 	}
933 
934 	mce_log(&m);
935 }
936 
937 DEFINE_IDTENTRY_SYSVEC(sysvec_deferred_error)
938 {
939 	trace_deferred_error_apic_entry(DEFERRED_ERROR_VECTOR);
940 	inc_irq_stat(irq_deferred_error_count);
941 	deferred_error_int_vector();
942 	trace_deferred_error_apic_exit(DEFERRED_ERROR_VECTOR);
943 	ack_APIC_irq();
944 }
945 
946 /*
947  * Returns true if the logged error is deferred. False, otherwise.
948  */
949 static inline bool
950 _log_error_bank(unsigned int bank, u32 msr_stat, u32 msr_addr, u64 misc)
951 {
952 	u64 status, addr = 0;
953 
954 	rdmsrl(msr_stat, status);
955 	if (!(status & MCI_STATUS_VAL))
956 		return false;
957 
958 	if (status & MCI_STATUS_ADDRV)
959 		rdmsrl(msr_addr, addr);
960 
961 	__log_error(bank, status, addr, misc);
962 
963 	wrmsrl(msr_stat, 0);
964 
965 	return status & MCI_STATUS_DEFERRED;
966 }
967 
968 /*
969  * We have three scenarios for checking for Deferred errors:
970  *
971  * 1) Non-SMCA systems check MCA_STATUS and log error if found.
972  * 2) SMCA systems check MCA_STATUS. If error is found then log it and also
973  *    clear MCA_DESTAT.
974  * 3) SMCA systems check MCA_DESTAT, if error was not found in MCA_STATUS, and
975  *    log it.
976  */
977 static void log_error_deferred(unsigned int bank)
978 {
979 	bool defrd;
980 
981 	defrd = _log_error_bank(bank, mca_msr_reg(bank, MCA_STATUS),
982 				mca_msr_reg(bank, MCA_ADDR), 0);
983 
984 	if (!mce_flags.smca)
985 		return;
986 
987 	/* Clear MCA_DESTAT if we logged the deferred error from MCA_STATUS. */
988 	if (defrd) {
989 		wrmsrl(MSR_AMD64_SMCA_MCx_DESTAT(bank), 0);
990 		return;
991 	}
992 
993 	/*
994 	 * Only deferred errors are logged in MCA_DE{STAT,ADDR} so just check
995 	 * for a valid error.
996 	 */
997 	_log_error_bank(bank, MSR_AMD64_SMCA_MCx_DESTAT(bank),
998 			      MSR_AMD64_SMCA_MCx_DEADDR(bank), 0);
999 }
1000 
1001 /* APIC interrupt handler for deferred errors */
1002 static void amd_deferred_error_interrupt(void)
1003 {
1004 	unsigned int bank;
1005 
1006 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank)
1007 		log_error_deferred(bank);
1008 }
1009 
1010 static void log_error_thresholding(unsigned int bank, u64 misc)
1011 {
1012 	_log_error_bank(bank, mca_msr_reg(bank, MCA_STATUS), mca_msr_reg(bank, MCA_ADDR), misc);
1013 }
1014 
1015 static void log_and_reset_block(struct threshold_block *block)
1016 {
1017 	struct thresh_restart tr;
1018 	u32 low = 0, high = 0;
1019 
1020 	if (!block)
1021 		return;
1022 
1023 	if (rdmsr_safe(block->address, &low, &high))
1024 		return;
1025 
1026 	if (!(high & MASK_OVERFLOW_HI))
1027 		return;
1028 
1029 	/* Log the MCE which caused the threshold event. */
1030 	log_error_thresholding(block->bank, ((u64)high << 32) | low);
1031 
1032 	/* Reset threshold block after logging error. */
1033 	memset(&tr, 0, sizeof(tr));
1034 	tr.b = block;
1035 	threshold_restart_bank(&tr);
1036 }
1037 
1038 /*
1039  * Threshold interrupt handler will service THRESHOLD_APIC_VECTOR. The interrupt
1040  * goes off when error_count reaches threshold_limit.
1041  */
1042 static void amd_threshold_interrupt(void)
1043 {
1044 	struct threshold_block *first_block = NULL, *block = NULL, *tmp = NULL;
1045 	struct threshold_bank **bp = this_cpu_read(threshold_banks);
1046 	unsigned int bank, cpu = smp_processor_id();
1047 
1048 	/*
1049 	 * Validate that the threshold bank has been initialized already. The
1050 	 * handler is installed at boot time, but on a hotplug event the
1051 	 * interrupt might fire before the data has been initialized.
1052 	 */
1053 	if (!bp)
1054 		return;
1055 
1056 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
1057 		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
1058 			continue;
1059 
1060 		first_block = bp[bank]->blocks;
1061 		if (!first_block)
1062 			continue;
1063 
1064 		/*
1065 		 * The first block is also the head of the list. Check it first
1066 		 * before iterating over the rest.
1067 		 */
1068 		log_and_reset_block(first_block);
1069 		list_for_each_entry_safe(block, tmp, &first_block->miscj, miscj)
1070 			log_and_reset_block(block);
1071 	}
1072 }
1073 
1074 /*
1075  * Sysfs Interface
1076  */
1077 
1078 struct threshold_attr {
1079 	struct attribute attr;
1080 	ssize_t (*show) (struct threshold_block *, char *);
1081 	ssize_t (*store) (struct threshold_block *, const char *, size_t count);
1082 };
1083 
1084 #define SHOW_FIELDS(name)						\
1085 static ssize_t show_ ## name(struct threshold_block *b, char *buf)	\
1086 {									\
1087 	return sprintf(buf, "%lu\n", (unsigned long) b->name);		\
1088 }
1089 SHOW_FIELDS(interrupt_enable)
1090 SHOW_FIELDS(threshold_limit)
1091 
1092 static ssize_t
1093 store_interrupt_enable(struct threshold_block *b, const char *buf, size_t size)
1094 {
1095 	struct thresh_restart tr;
1096 	unsigned long new;
1097 
1098 	if (!b->interrupt_capable)
1099 		return -EINVAL;
1100 
1101 	if (kstrtoul(buf, 0, &new) < 0)
1102 		return -EINVAL;
1103 
1104 	b->interrupt_enable = !!new;
1105 
1106 	memset(&tr, 0, sizeof(tr));
1107 	tr.b		= b;
1108 
1109 	if (smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1))
1110 		return -ENODEV;
1111 
1112 	return size;
1113 }
1114 
1115 static ssize_t
1116 store_threshold_limit(struct threshold_block *b, const char *buf, size_t size)
1117 {
1118 	struct thresh_restart tr;
1119 	unsigned long new;
1120 
1121 	if (kstrtoul(buf, 0, &new) < 0)
1122 		return -EINVAL;
1123 
1124 	if (new > THRESHOLD_MAX)
1125 		new = THRESHOLD_MAX;
1126 	if (new < 1)
1127 		new = 1;
1128 
1129 	memset(&tr, 0, sizeof(tr));
1130 	tr.old_limit = b->threshold_limit;
1131 	b->threshold_limit = new;
1132 	tr.b = b;
1133 
1134 	if (smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1))
1135 		return -ENODEV;
1136 
1137 	return size;
1138 }
1139 
1140 static ssize_t show_error_count(struct threshold_block *b, char *buf)
1141 {
1142 	u32 lo, hi;
1143 
1144 	/* CPU might be offline by now */
1145 	if (rdmsr_on_cpu(b->cpu, b->address, &lo, &hi))
1146 		return -ENODEV;
1147 
1148 	return sprintf(buf, "%u\n", ((hi & THRESHOLD_MAX) -
1149 				     (THRESHOLD_MAX - b->threshold_limit)));
1150 }
1151 
1152 static struct threshold_attr error_count = {
1153 	.attr = {.name = __stringify(error_count), .mode = 0444 },
1154 	.show = show_error_count,
1155 };
1156 
1157 #define RW_ATTR(val)							\
1158 static struct threshold_attr val = {					\
1159 	.attr	= {.name = __stringify(val), .mode = 0644 },		\
1160 	.show	= show_## val,						\
1161 	.store	= store_## val,						\
1162 };
1163 
1164 RW_ATTR(interrupt_enable);
1165 RW_ATTR(threshold_limit);
1166 
1167 static struct attribute *default_attrs[] = {
1168 	&threshold_limit.attr,
1169 	&error_count.attr,
1170 	NULL,	/* possibly interrupt_enable if supported, see below */
1171 	NULL,
1172 };
1173 
1174 #define to_block(k)	container_of(k, struct threshold_block, kobj)
1175 #define to_attr(a)	container_of(a, struct threshold_attr, attr)
1176 
1177 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
1178 {
1179 	struct threshold_block *b = to_block(kobj);
1180 	struct threshold_attr *a = to_attr(attr);
1181 	ssize_t ret;
1182 
1183 	ret = a->show ? a->show(b, buf) : -EIO;
1184 
1185 	return ret;
1186 }
1187 
1188 static ssize_t store(struct kobject *kobj, struct attribute *attr,
1189 		     const char *buf, size_t count)
1190 {
1191 	struct threshold_block *b = to_block(kobj);
1192 	struct threshold_attr *a = to_attr(attr);
1193 	ssize_t ret;
1194 
1195 	ret = a->store ? a->store(b, buf, count) : -EIO;
1196 
1197 	return ret;
1198 }
1199 
1200 static const struct sysfs_ops threshold_ops = {
1201 	.show			= show,
1202 	.store			= store,
1203 };
1204 
1205 static void threshold_block_release(struct kobject *kobj);
1206 
1207 static struct kobj_type threshold_ktype = {
1208 	.sysfs_ops		= &threshold_ops,
1209 	.default_attrs		= default_attrs,
1210 	.release		= threshold_block_release,
1211 };
1212 
1213 static const char *get_name(unsigned int bank, struct threshold_block *b)
1214 {
1215 	enum smca_bank_types bank_type;
1216 
1217 	if (!mce_flags.smca) {
1218 		if (b && bank == 4)
1219 			return bank4_names(b);
1220 
1221 		return th_names[bank];
1222 	}
1223 
1224 	bank_type = smca_get_bank_type(bank);
1225 	if (bank_type >= N_SMCA_BANK_TYPES)
1226 		return NULL;
1227 
1228 	if (b && bank_type == SMCA_UMC) {
1229 		if (b->block < ARRAY_SIZE(smca_umc_block_names))
1230 			return smca_umc_block_names[b->block];
1231 		return NULL;
1232 	}
1233 
1234 	if (smca_banks[bank].hwid->count == 1)
1235 		return smca_get_name(bank_type);
1236 
1237 	snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN,
1238 		 "%s_%x", smca_get_name(bank_type),
1239 			  smca_banks[bank].sysfs_id);
1240 	return buf_mcatype;
1241 }
1242 
1243 static int allocate_threshold_blocks(unsigned int cpu, struct threshold_bank *tb,
1244 				     unsigned int bank, unsigned int block,
1245 				     u32 address)
1246 {
1247 	struct threshold_block *b = NULL;
1248 	u32 low, high;
1249 	int err;
1250 
1251 	if ((bank >= this_cpu_read(mce_num_banks)) || (block >= NR_BLOCKS))
1252 		return 0;
1253 
1254 	if (rdmsr_safe(address, &low, &high))
1255 		return 0;
1256 
1257 	if (!(high & MASK_VALID_HI)) {
1258 		if (block)
1259 			goto recurse;
1260 		else
1261 			return 0;
1262 	}
1263 
1264 	if (!(high & MASK_CNTP_HI)  ||
1265 	     (high & MASK_LOCKED_HI))
1266 		goto recurse;
1267 
1268 	b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
1269 	if (!b)
1270 		return -ENOMEM;
1271 
1272 	b->block		= block;
1273 	b->bank			= bank;
1274 	b->cpu			= cpu;
1275 	b->address		= address;
1276 	b->interrupt_enable	= 0;
1277 	b->interrupt_capable	= lvt_interrupt_supported(bank, high);
1278 	b->threshold_limit	= THRESHOLD_MAX;
1279 
1280 	if (b->interrupt_capable) {
1281 		threshold_ktype.default_attrs[2] = &interrupt_enable.attr;
1282 		b->interrupt_enable = 1;
1283 	} else {
1284 		threshold_ktype.default_attrs[2] = NULL;
1285 	}
1286 
1287 	INIT_LIST_HEAD(&b->miscj);
1288 
1289 	/* This is safe as @tb is not visible yet */
1290 	if (tb->blocks)
1291 		list_add(&b->miscj, &tb->blocks->miscj);
1292 	else
1293 		tb->blocks = b;
1294 
1295 	err = kobject_init_and_add(&b->kobj, &threshold_ktype, tb->kobj, get_name(bank, b));
1296 	if (err)
1297 		goto out_free;
1298 recurse:
1299 	address = get_block_address(address, low, high, bank, ++block, cpu);
1300 	if (!address)
1301 		return 0;
1302 
1303 	err = allocate_threshold_blocks(cpu, tb, bank, block, address);
1304 	if (err)
1305 		goto out_free;
1306 
1307 	if (b)
1308 		kobject_uevent(&b->kobj, KOBJ_ADD);
1309 
1310 	return 0;
1311 
1312 out_free:
1313 	if (b) {
1314 		list_del(&b->miscj);
1315 		kobject_put(&b->kobj);
1316 	}
1317 	return err;
1318 }
1319 
1320 static int __threshold_add_blocks(struct threshold_bank *b)
1321 {
1322 	struct list_head *head = &b->blocks->miscj;
1323 	struct threshold_block *pos = NULL;
1324 	struct threshold_block *tmp = NULL;
1325 	int err = 0;
1326 
1327 	err = kobject_add(&b->blocks->kobj, b->kobj, b->blocks->kobj.name);
1328 	if (err)
1329 		return err;
1330 
1331 	list_for_each_entry_safe(pos, tmp, head, miscj) {
1332 
1333 		err = kobject_add(&pos->kobj, b->kobj, pos->kobj.name);
1334 		if (err) {
1335 			list_for_each_entry_safe_reverse(pos, tmp, head, miscj)
1336 				kobject_del(&pos->kobj);
1337 
1338 			return err;
1339 		}
1340 	}
1341 	return err;
1342 }
1343 
1344 static int threshold_create_bank(struct threshold_bank **bp, unsigned int cpu,
1345 				 unsigned int bank)
1346 {
1347 	struct device *dev = this_cpu_read(mce_device);
1348 	struct amd_northbridge *nb = NULL;
1349 	struct threshold_bank *b = NULL;
1350 	const char *name = get_name(bank, NULL);
1351 	int err = 0;
1352 
1353 	if (!dev)
1354 		return -ENODEV;
1355 
1356 	if (is_shared_bank(bank)) {
1357 		nb = node_to_amd_nb(topology_die_id(cpu));
1358 
1359 		/* threshold descriptor already initialized on this node? */
1360 		if (nb && nb->bank4) {
1361 			/* yes, use it */
1362 			b = nb->bank4;
1363 			err = kobject_add(b->kobj, &dev->kobj, name);
1364 			if (err)
1365 				goto out;
1366 
1367 			bp[bank] = b;
1368 			refcount_inc(&b->cpus);
1369 
1370 			err = __threshold_add_blocks(b);
1371 
1372 			goto out;
1373 		}
1374 	}
1375 
1376 	b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
1377 	if (!b) {
1378 		err = -ENOMEM;
1379 		goto out;
1380 	}
1381 
1382 	/* Associate the bank with the per-CPU MCE device */
1383 	b->kobj = kobject_create_and_add(name, &dev->kobj);
1384 	if (!b->kobj) {
1385 		err = -EINVAL;
1386 		goto out_free;
1387 	}
1388 
1389 	if (is_shared_bank(bank)) {
1390 		b->shared = 1;
1391 		refcount_set(&b->cpus, 1);
1392 
1393 		/* nb is already initialized, see above */
1394 		if (nb) {
1395 			WARN_ON(nb->bank4);
1396 			nb->bank4 = b;
1397 		}
1398 	}
1399 
1400 	err = allocate_threshold_blocks(cpu, b, bank, 0, mca_msr_reg(bank, MCA_MISC));
1401 	if (err)
1402 		goto out_kobj;
1403 
1404 	bp[bank] = b;
1405 	return 0;
1406 
1407 out_kobj:
1408 	kobject_put(b->kobj);
1409 out_free:
1410 	kfree(b);
1411 out:
1412 	return err;
1413 }
1414 
1415 static void threshold_block_release(struct kobject *kobj)
1416 {
1417 	kfree(to_block(kobj));
1418 }
1419 
1420 static void deallocate_threshold_blocks(struct threshold_bank *bank)
1421 {
1422 	struct threshold_block *pos, *tmp;
1423 
1424 	list_for_each_entry_safe(pos, tmp, &bank->blocks->miscj, miscj) {
1425 		list_del(&pos->miscj);
1426 		kobject_put(&pos->kobj);
1427 	}
1428 
1429 	kobject_put(&bank->blocks->kobj);
1430 }
1431 
1432 static void __threshold_remove_blocks(struct threshold_bank *b)
1433 {
1434 	struct threshold_block *pos = NULL;
1435 	struct threshold_block *tmp = NULL;
1436 
1437 	kobject_del(b->kobj);
1438 
1439 	list_for_each_entry_safe(pos, tmp, &b->blocks->miscj, miscj)
1440 		kobject_del(&pos->kobj);
1441 }
1442 
1443 static void threshold_remove_bank(struct threshold_bank *bank)
1444 {
1445 	struct amd_northbridge *nb;
1446 
1447 	if (!bank->blocks)
1448 		goto out_free;
1449 
1450 	if (!bank->shared)
1451 		goto out_dealloc;
1452 
1453 	if (!refcount_dec_and_test(&bank->cpus)) {
1454 		__threshold_remove_blocks(bank);
1455 		return;
1456 	} else {
1457 		/*
1458 		 * The last CPU on this node using the shared bank is going
1459 		 * away, remove that bank now.
1460 		 */
1461 		nb = node_to_amd_nb(topology_die_id(smp_processor_id()));
1462 		nb->bank4 = NULL;
1463 	}
1464 
1465 out_dealloc:
1466 	deallocate_threshold_blocks(bank);
1467 
1468 out_free:
1469 	kobject_put(bank->kobj);
1470 	kfree(bank);
1471 }
1472 
1473 int mce_threshold_remove_device(unsigned int cpu)
1474 {
1475 	struct threshold_bank **bp = this_cpu_read(threshold_banks);
1476 	unsigned int bank, numbanks = this_cpu_read(mce_num_banks);
1477 
1478 	if (!bp)
1479 		return 0;
1480 
1481 	/*
1482 	 * Clear the pointer before cleaning up, so that the interrupt won't
1483 	 * touch anything of this.
1484 	 */
1485 	this_cpu_write(threshold_banks, NULL);
1486 
1487 	for (bank = 0; bank < numbanks; bank++) {
1488 		if (bp[bank]) {
1489 			threshold_remove_bank(bp[bank]);
1490 			bp[bank] = NULL;
1491 		}
1492 	}
1493 	kfree(bp);
1494 	return 0;
1495 }
1496 
1497 /**
1498  * mce_threshold_create_device - Create the per-CPU MCE threshold device
1499  * @cpu:	The plugged in CPU
1500  *
1501  * Create directories and files for all valid threshold banks.
1502  *
1503  * This is invoked from the CPU hotplug callback which was installed in
1504  * mcheck_init_device(). The invocation happens in context of the hotplug
1505  * thread running on @cpu.  The callback is invoked on all CPUs which are
1506  * online when the callback is installed or during a real hotplug event.
1507  */
1508 int mce_threshold_create_device(unsigned int cpu)
1509 {
1510 	unsigned int numbanks, bank;
1511 	struct threshold_bank **bp;
1512 	int err;
1513 
1514 	if (!mce_flags.amd_threshold)
1515 		return 0;
1516 
1517 	bp = this_cpu_read(threshold_banks);
1518 	if (bp)
1519 		return 0;
1520 
1521 	numbanks = this_cpu_read(mce_num_banks);
1522 	bp = kcalloc(numbanks, sizeof(*bp), GFP_KERNEL);
1523 	if (!bp)
1524 		return -ENOMEM;
1525 
1526 	for (bank = 0; bank < numbanks; ++bank) {
1527 		if (!(this_cpu_read(bank_map) & (1 << bank)))
1528 			continue;
1529 		err = threshold_create_bank(bp, cpu, bank);
1530 		if (err)
1531 			goto out_err;
1532 	}
1533 	this_cpu_write(threshold_banks, bp);
1534 
1535 	if (thresholding_irq_en)
1536 		mce_threshold_vector = amd_threshold_interrupt;
1537 	return 0;
1538 out_err:
1539 	mce_threshold_remove_device(cpu);
1540 	return err;
1541 }
1542