xref: /linux/arch/x86/kernel/cpu/mce/amd.c (revision 55d0969c451159cff86949b38c39171cab962069)
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 #define HWID_MCATYPE(hwid, mcatype) (((hwid) << 16) | (mcatype))
75 
76 struct smca_hwid {
77 	unsigned int bank_type;	/* Use with smca_bank_types for easy indexing. */
78 	u32 hwid_mcatype;	/* (hwid,mcatype) tuple */
79 };
80 
81 struct smca_bank {
82 	const struct smca_hwid *hwid;
83 	u32 id;			/* Value of MCA_IPID[InstanceId]. */
84 	u8 sysfs_id;		/* Value used for sysfs name. */
85 };
86 
87 static DEFINE_PER_CPU_READ_MOSTLY(struct smca_bank[MAX_NR_BANKS], smca_banks);
88 static DEFINE_PER_CPU_READ_MOSTLY(u8[N_SMCA_BANK_TYPES], smca_bank_counts);
89 
90 static const char * const smca_names[] = {
91 	[SMCA_LS ... SMCA_LS_V2]	= "load_store",
92 	[SMCA_IF]			= "insn_fetch",
93 	[SMCA_L2_CACHE]			= "l2_cache",
94 	[SMCA_DE]			= "decode_unit",
95 	[SMCA_RESERVED]			= "reserved",
96 	[SMCA_EX]			= "execution_unit",
97 	[SMCA_FP]			= "floating_point",
98 	[SMCA_L3_CACHE]			= "l3_cache",
99 	[SMCA_CS ... SMCA_CS_V2]	= "coherent_slave",
100 	[SMCA_PIE]			= "pie",
101 
102 	/* UMC v2 is separate because both of them can exist in a single system. */
103 	[SMCA_UMC]			= "umc",
104 	[SMCA_UMC_V2]			= "umc_v2",
105 	[SMCA_MA_LLC]			= "ma_llc",
106 	[SMCA_PB]			= "param_block",
107 	[SMCA_PSP ... SMCA_PSP_V2]	= "psp",
108 	[SMCA_SMU ... SMCA_SMU_V2]	= "smu",
109 	[SMCA_MP5]			= "mp5",
110 	[SMCA_MPDMA]			= "mpdma",
111 	[SMCA_NBIO]			= "nbio",
112 	[SMCA_PCIE ... SMCA_PCIE_V2]	= "pcie",
113 	[SMCA_XGMI_PCS]			= "xgmi_pcs",
114 	[SMCA_NBIF]			= "nbif",
115 	[SMCA_SHUB]			= "shub",
116 	[SMCA_SATA]			= "sata",
117 	[SMCA_USB]			= "usb",
118 	[SMCA_USR_DP]			= "usr_dp",
119 	[SMCA_USR_CP]			= "usr_cp",
120 	[SMCA_GMI_PCS]			= "gmi_pcs",
121 	[SMCA_XGMI_PHY]			= "xgmi_phy",
122 	[SMCA_WAFL_PHY]			= "wafl_phy",
123 	[SMCA_GMI_PHY]			= "gmi_phy",
124 };
125 
126 static const char *smca_get_name(enum smca_bank_types t)
127 {
128 	if (t >= N_SMCA_BANK_TYPES)
129 		return NULL;
130 
131 	return smca_names[t];
132 }
133 
134 enum smca_bank_types smca_get_bank_type(unsigned int cpu, unsigned int bank)
135 {
136 	struct smca_bank *b;
137 
138 	if (bank >= MAX_NR_BANKS)
139 		return N_SMCA_BANK_TYPES;
140 
141 	b = &per_cpu(smca_banks, cpu)[bank];
142 	if (!b->hwid)
143 		return N_SMCA_BANK_TYPES;
144 
145 	return b->hwid->bank_type;
146 }
147 EXPORT_SYMBOL_GPL(smca_get_bank_type);
148 
149 static const struct smca_hwid smca_hwid_mcatypes[] = {
150 	/* { bank_type, hwid_mcatype } */
151 
152 	/* Reserved type */
153 	{ SMCA_RESERVED, HWID_MCATYPE(0x00, 0x0)	},
154 
155 	/* ZN Core (HWID=0xB0) MCA types */
156 	{ SMCA_LS,	 HWID_MCATYPE(0xB0, 0x0)	},
157 	{ SMCA_LS_V2,	 HWID_MCATYPE(0xB0, 0x10)	},
158 	{ SMCA_IF,	 HWID_MCATYPE(0xB0, 0x1)	},
159 	{ SMCA_L2_CACHE, HWID_MCATYPE(0xB0, 0x2)	},
160 	{ SMCA_DE,	 HWID_MCATYPE(0xB0, 0x3)	},
161 	/* HWID 0xB0 MCATYPE 0x4 is Reserved */
162 	{ SMCA_EX,	 HWID_MCATYPE(0xB0, 0x5)	},
163 	{ SMCA_FP,	 HWID_MCATYPE(0xB0, 0x6)	},
164 	{ SMCA_L3_CACHE, HWID_MCATYPE(0xB0, 0x7)	},
165 
166 	/* Data Fabric MCA types */
167 	{ SMCA_CS,	 HWID_MCATYPE(0x2E, 0x0)	},
168 	{ SMCA_PIE,	 HWID_MCATYPE(0x2E, 0x1)	},
169 	{ SMCA_CS_V2,	 HWID_MCATYPE(0x2E, 0x2)	},
170 	{ SMCA_MA_LLC,	 HWID_MCATYPE(0x2E, 0x4)	},
171 
172 	/* Unified Memory Controller MCA type */
173 	{ SMCA_UMC,	 HWID_MCATYPE(0x96, 0x0)	},
174 	{ SMCA_UMC_V2,	 HWID_MCATYPE(0x96, 0x1)	},
175 
176 	/* Parameter Block MCA type */
177 	{ SMCA_PB,	 HWID_MCATYPE(0x05, 0x0)	},
178 
179 	/* Platform Security Processor MCA type */
180 	{ SMCA_PSP,	 HWID_MCATYPE(0xFF, 0x0)	},
181 	{ SMCA_PSP_V2,	 HWID_MCATYPE(0xFF, 0x1)	},
182 
183 	/* System Management Unit MCA type */
184 	{ SMCA_SMU,	 HWID_MCATYPE(0x01, 0x0)	},
185 	{ SMCA_SMU_V2,	 HWID_MCATYPE(0x01, 0x1)	},
186 
187 	/* Microprocessor 5 Unit MCA type */
188 	{ SMCA_MP5,	 HWID_MCATYPE(0x01, 0x2)	},
189 
190 	/* MPDMA MCA type */
191 	{ SMCA_MPDMA,	 HWID_MCATYPE(0x01, 0x3)	},
192 
193 	/* Northbridge IO Unit MCA type */
194 	{ SMCA_NBIO,	 HWID_MCATYPE(0x18, 0x0)	},
195 
196 	/* PCI Express Unit MCA type */
197 	{ SMCA_PCIE,	 HWID_MCATYPE(0x46, 0x0)	},
198 	{ SMCA_PCIE_V2,	 HWID_MCATYPE(0x46, 0x1)	},
199 
200 	{ SMCA_XGMI_PCS, HWID_MCATYPE(0x50, 0x0)	},
201 	{ SMCA_NBIF,	 HWID_MCATYPE(0x6C, 0x0)	},
202 	{ SMCA_SHUB,	 HWID_MCATYPE(0x80, 0x0)	},
203 	{ SMCA_SATA,	 HWID_MCATYPE(0xA8, 0x0)	},
204 	{ SMCA_USB,	 HWID_MCATYPE(0xAA, 0x0)	},
205 	{ SMCA_USR_DP,	 HWID_MCATYPE(0x170, 0x0)	},
206 	{ SMCA_USR_CP,	 HWID_MCATYPE(0x180, 0x0)	},
207 	{ SMCA_GMI_PCS,  HWID_MCATYPE(0x241, 0x0)	},
208 	{ SMCA_XGMI_PHY, HWID_MCATYPE(0x259, 0x0)	},
209 	{ SMCA_WAFL_PHY, HWID_MCATYPE(0x267, 0x0)	},
210 	{ SMCA_GMI_PHY,	 HWID_MCATYPE(0x269, 0x0)	},
211 };
212 
213 /*
214  * In SMCA enabled processors, we can have multiple banks for a given IP type.
215  * So to define a unique name for each bank, we use a temp c-string to append
216  * the MCA_IPID[InstanceId] to type's name in get_name().
217  *
218  * InstanceId is 32 bits which is 8 characters. Make sure MAX_MCATYPE_NAME_LEN
219  * is greater than 8 plus 1 (for underscore) plus length of longest type name.
220  */
221 #define MAX_MCATYPE_NAME_LEN	30
222 static char buf_mcatype[MAX_MCATYPE_NAME_LEN];
223 
224 static DEFINE_PER_CPU(struct threshold_bank **, threshold_banks);
225 
226 /*
227  * A list of the banks enabled on each logical CPU. Controls which respective
228  * descriptors to initialize later in mce_threshold_create_device().
229  */
230 static DEFINE_PER_CPU(u64, bank_map);
231 
232 /* Map of banks that have more than MCA_MISC0 available. */
233 static DEFINE_PER_CPU(u64, smca_misc_banks_map);
234 
235 static void amd_threshold_interrupt(void);
236 static void amd_deferred_error_interrupt(void);
237 
238 static void default_deferred_error_interrupt(void)
239 {
240 	pr_err("Unexpected deferred interrupt at vector %x\n", DEFERRED_ERROR_VECTOR);
241 }
242 void (*deferred_error_int_vector)(void) = default_deferred_error_interrupt;
243 
244 static void smca_set_misc_banks_map(unsigned int bank, unsigned int cpu)
245 {
246 	u32 low, high;
247 
248 	/*
249 	 * For SMCA enabled processors, BLKPTR field of the first MISC register
250 	 * (MCx_MISC0) indicates presence of additional MISC regs set (MISC1-4).
251 	 */
252 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_CONFIG(bank), &low, &high))
253 		return;
254 
255 	if (!(low & MCI_CONFIG_MCAX))
256 		return;
257 
258 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_MISC(bank), &low, &high))
259 		return;
260 
261 	if (low & MASK_BLKPTR_LO)
262 		per_cpu(smca_misc_banks_map, cpu) |= BIT_ULL(bank);
263 
264 }
265 
266 static void smca_configure(unsigned int bank, unsigned int cpu)
267 {
268 	u8 *bank_counts = this_cpu_ptr(smca_bank_counts);
269 	const struct smca_hwid *s_hwid;
270 	unsigned int i, hwid_mcatype;
271 	u32 high, low;
272 	u32 smca_config = MSR_AMD64_SMCA_MCx_CONFIG(bank);
273 
274 	/* Set appropriate bits in MCA_CONFIG */
275 	if (!rdmsr_safe(smca_config, &low, &high)) {
276 		/*
277 		 * OS is required to set the MCAX bit to acknowledge that it is
278 		 * now using the new MSR ranges and new registers under each
279 		 * bank. It also means that the OS will configure deferred
280 		 * errors in the new MCx_CONFIG register. If the bit is not set,
281 		 * uncorrectable errors will cause a system panic.
282 		 *
283 		 * MCA_CONFIG[MCAX] is bit 32 (0 in the high portion of the MSR.)
284 		 */
285 		high |= BIT(0);
286 
287 		/*
288 		 * SMCA sets the Deferred Error Interrupt type per bank.
289 		 *
290 		 * MCA_CONFIG[DeferredIntTypeSupported] is bit 5, and tells us
291 		 * if the DeferredIntType bit field is available.
292 		 *
293 		 * MCA_CONFIG[DeferredIntType] is bits [38:37] ([6:5] in the
294 		 * high portion of the MSR). OS should set this to 0x1 to enable
295 		 * APIC based interrupt. First, check that no interrupt has been
296 		 * set.
297 		 */
298 		if ((low & BIT(5)) && !((high >> 5) & 0x3))
299 			high |= BIT(5);
300 
301 		this_cpu_ptr(mce_banks_array)[bank].lsb_in_status = !!(low & BIT(8));
302 
303 		wrmsr(smca_config, low, high);
304 	}
305 
306 	smca_set_misc_banks_map(bank, cpu);
307 
308 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_IPID(bank), &low, &high)) {
309 		pr_warn("Failed to read MCA_IPID for bank %d\n", bank);
310 		return;
311 	}
312 
313 	hwid_mcatype = HWID_MCATYPE(high & MCI_IPID_HWID,
314 				    (high & MCI_IPID_MCATYPE) >> 16);
315 
316 	for (i = 0; i < ARRAY_SIZE(smca_hwid_mcatypes); i++) {
317 		s_hwid = &smca_hwid_mcatypes[i];
318 
319 		if (hwid_mcatype == s_hwid->hwid_mcatype) {
320 			this_cpu_ptr(smca_banks)[bank].hwid = s_hwid;
321 			this_cpu_ptr(smca_banks)[bank].id = low;
322 			this_cpu_ptr(smca_banks)[bank].sysfs_id = bank_counts[s_hwid->bank_type]++;
323 			break;
324 		}
325 	}
326 }
327 
328 struct thresh_restart {
329 	struct threshold_block	*b;
330 	int			reset;
331 	int			set_lvt_off;
332 	int			lvt_off;
333 	u16			old_limit;
334 };
335 
336 static inline bool is_shared_bank(int bank)
337 {
338 	/*
339 	 * Scalable MCA provides for only one core to have access to the MSRs of
340 	 * a shared bank.
341 	 */
342 	if (mce_flags.smca)
343 		return false;
344 
345 	/* Bank 4 is for northbridge reporting and is thus shared */
346 	return (bank == 4);
347 }
348 
349 static const char *bank4_names(const struct threshold_block *b)
350 {
351 	switch (b->address) {
352 	/* MSR4_MISC0 */
353 	case 0x00000413:
354 		return "dram";
355 
356 	case 0xc0000408:
357 		return "ht_links";
358 
359 	case 0xc0000409:
360 		return "l3_cache";
361 
362 	default:
363 		WARN(1, "Funny MSR: 0x%08x\n", b->address);
364 		return "";
365 	}
366 };
367 
368 
369 static bool lvt_interrupt_supported(unsigned int bank, u32 msr_high_bits)
370 {
371 	/*
372 	 * bank 4 supports APIC LVT interrupts implicitly since forever.
373 	 */
374 	if (bank == 4)
375 		return true;
376 
377 	/*
378 	 * IntP: interrupt present; if this bit is set, the thresholding
379 	 * bank can generate APIC LVT interrupts
380 	 */
381 	return msr_high_bits & BIT(28);
382 }
383 
384 static int lvt_off_valid(struct threshold_block *b, int apic, u32 lo, u32 hi)
385 {
386 	int msr = (hi & MASK_LVTOFF_HI) >> 20;
387 
388 	if (apic < 0) {
389 		pr_err(FW_BUG "cpu %d, failed to setup threshold interrupt "
390 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n", b->cpu,
391 		       b->bank, b->block, b->address, hi, lo);
392 		return 0;
393 	}
394 
395 	if (apic != msr) {
396 		/*
397 		 * On SMCA CPUs, LVT offset is programmed at a different MSR, and
398 		 * the BIOS provides the value. The original field where LVT offset
399 		 * was set is reserved. Return early here:
400 		 */
401 		if (mce_flags.smca)
402 			return 0;
403 
404 		pr_err(FW_BUG "cpu %d, invalid threshold interrupt offset %d "
405 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n",
406 		       b->cpu, apic, b->bank, b->block, b->address, hi, lo);
407 		return 0;
408 	}
409 
410 	return 1;
411 };
412 
413 /* Reprogram MCx_MISC MSR behind this threshold bank. */
414 static void threshold_restart_bank(void *_tr)
415 {
416 	struct thresh_restart *tr = _tr;
417 	u32 hi, lo;
418 
419 	/* sysfs write might race against an offline operation */
420 	if (!this_cpu_read(threshold_banks) && !tr->set_lvt_off)
421 		return;
422 
423 	rdmsr(tr->b->address, lo, hi);
424 
425 	if (tr->b->threshold_limit < (hi & THRESHOLD_MAX))
426 		tr->reset = 1;	/* limit cannot be lower than err count */
427 
428 	if (tr->reset) {		/* reset err count and overflow bit */
429 		hi =
430 		    (hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
431 		    (THRESHOLD_MAX - tr->b->threshold_limit);
432 	} else if (tr->old_limit) {	/* change limit w/o reset */
433 		int new_count = (hi & THRESHOLD_MAX) +
434 		    (tr->old_limit - tr->b->threshold_limit);
435 
436 		hi = (hi & ~MASK_ERR_COUNT_HI) |
437 		    (new_count & THRESHOLD_MAX);
438 	}
439 
440 	/* clear IntType */
441 	hi &= ~MASK_INT_TYPE_HI;
442 
443 	if (!tr->b->interrupt_capable)
444 		goto done;
445 
446 	if (tr->set_lvt_off) {
447 		if (lvt_off_valid(tr->b, tr->lvt_off, lo, hi)) {
448 			/* set new lvt offset */
449 			hi &= ~MASK_LVTOFF_HI;
450 			hi |= tr->lvt_off << 20;
451 		}
452 	}
453 
454 	if (tr->b->interrupt_enable)
455 		hi |= INT_TYPE_APIC;
456 
457  done:
458 
459 	hi |= MASK_COUNT_EN_HI;
460 	wrmsr(tr->b->address, lo, hi);
461 }
462 
463 static void mce_threshold_block_init(struct threshold_block *b, int offset)
464 {
465 	struct thresh_restart tr = {
466 		.b			= b,
467 		.set_lvt_off		= 1,
468 		.lvt_off		= offset,
469 	};
470 
471 	b->threshold_limit		= THRESHOLD_MAX;
472 	threshold_restart_bank(&tr);
473 };
474 
475 static int setup_APIC_mce_threshold(int reserved, int new)
476 {
477 	if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
478 					      APIC_EILVT_MSG_FIX, 0))
479 		return new;
480 
481 	return reserved;
482 }
483 
484 static int setup_APIC_deferred_error(int reserved, int new)
485 {
486 	if (reserved < 0 && !setup_APIC_eilvt(new, DEFERRED_ERROR_VECTOR,
487 					      APIC_EILVT_MSG_FIX, 0))
488 		return new;
489 
490 	return reserved;
491 }
492 
493 static void deferred_error_interrupt_enable(struct cpuinfo_x86 *c)
494 {
495 	u32 low = 0, high = 0;
496 	int def_offset = -1, def_new;
497 
498 	if (rdmsr_safe(MSR_CU_DEF_ERR, &low, &high))
499 		return;
500 
501 	def_new = (low & MASK_DEF_LVTOFF) >> 4;
502 	if (!(low & MASK_DEF_LVTOFF)) {
503 		pr_err(FW_BUG "Your BIOS is not setting up LVT offset 0x2 for deferred error IRQs correctly.\n");
504 		def_new = DEF_LVT_OFF;
505 		low = (low & ~MASK_DEF_LVTOFF) | (DEF_LVT_OFF << 4);
506 	}
507 
508 	def_offset = setup_APIC_deferred_error(def_offset, def_new);
509 	if ((def_offset == def_new) &&
510 	    (deferred_error_int_vector != amd_deferred_error_interrupt))
511 		deferred_error_int_vector = amd_deferred_error_interrupt;
512 
513 	if (!mce_flags.smca)
514 		low = (low & ~MASK_DEF_INT_TYPE) | DEF_INT_TYPE_APIC;
515 
516 	wrmsr(MSR_CU_DEF_ERR, low, high);
517 }
518 
519 static u32 smca_get_block_address(unsigned int bank, unsigned int block,
520 				  unsigned int cpu)
521 {
522 	if (!block)
523 		return MSR_AMD64_SMCA_MCx_MISC(bank);
524 
525 	if (!(per_cpu(smca_misc_banks_map, cpu) & BIT_ULL(bank)))
526 		return 0;
527 
528 	return MSR_AMD64_SMCA_MCx_MISCy(bank, block - 1);
529 }
530 
531 static u32 get_block_address(u32 current_addr, u32 low, u32 high,
532 			     unsigned int bank, unsigned int block,
533 			     unsigned int cpu)
534 {
535 	u32 addr = 0, offset = 0;
536 
537 	if ((bank >= per_cpu(mce_num_banks, cpu)) || (block >= NR_BLOCKS))
538 		return addr;
539 
540 	if (mce_flags.smca)
541 		return smca_get_block_address(bank, block, cpu);
542 
543 	/* Fall back to method we used for older processors: */
544 	switch (block) {
545 	case 0:
546 		addr = mca_msr_reg(bank, MCA_MISC);
547 		break;
548 	case 1:
549 		offset = ((low & MASK_BLKPTR_LO) >> 21);
550 		if (offset)
551 			addr = MCG_XBLK_ADDR + offset;
552 		break;
553 	default:
554 		addr = ++current_addr;
555 	}
556 	return addr;
557 }
558 
559 static int
560 prepare_threshold_block(unsigned int bank, unsigned int block, u32 addr,
561 			int offset, u32 misc_high)
562 {
563 	unsigned int cpu = smp_processor_id();
564 	u32 smca_low, smca_high;
565 	struct threshold_block b;
566 	int new;
567 
568 	if (!block)
569 		per_cpu(bank_map, cpu) |= BIT_ULL(bank);
570 
571 	memset(&b, 0, sizeof(b));
572 	b.cpu			= cpu;
573 	b.bank			= bank;
574 	b.block			= block;
575 	b.address		= addr;
576 	b.interrupt_capable	= lvt_interrupt_supported(bank, misc_high);
577 
578 	if (!b.interrupt_capable)
579 		goto done;
580 
581 	b.interrupt_enable = 1;
582 
583 	if (!mce_flags.smca) {
584 		new = (misc_high & MASK_LVTOFF_HI) >> 20;
585 		goto set_offset;
586 	}
587 
588 	/* Gather LVT offset for thresholding: */
589 	if (rdmsr_safe(MSR_CU_DEF_ERR, &smca_low, &smca_high))
590 		goto out;
591 
592 	new = (smca_low & SMCA_THR_LVT_OFF) >> 12;
593 
594 set_offset:
595 	offset = setup_APIC_mce_threshold(offset, new);
596 	if (offset == new)
597 		thresholding_irq_en = true;
598 
599 done:
600 	mce_threshold_block_init(&b, offset);
601 
602 out:
603 	return offset;
604 }
605 
606 bool amd_filter_mce(struct mce *m)
607 {
608 	enum smca_bank_types bank_type = smca_get_bank_type(m->extcpu, m->bank);
609 	struct cpuinfo_x86 *c = &boot_cpu_data;
610 
611 	/* See Family 17h Models 10h-2Fh Erratum #1114. */
612 	if (c->x86 == 0x17 &&
613 	    c->x86_model >= 0x10 && c->x86_model <= 0x2F &&
614 	    bank_type == SMCA_IF && XEC(m->status, 0x3f) == 10)
615 		return true;
616 
617 	/* NB GART TLB error reporting is disabled by default. */
618 	if (c->x86 < 0x17) {
619 		if (m->bank == 4 && XEC(m->status, 0x1f) == 0x5)
620 			return true;
621 	}
622 
623 	return false;
624 }
625 
626 /*
627  * Turn off thresholding banks for the following conditions:
628  * - MC4_MISC thresholding is not supported on Family 0x15.
629  * - Prevent possible spurious interrupts from the IF bank on Family 0x17
630  *   Models 0x10-0x2F due to Erratum #1114.
631  */
632 static void disable_err_thresholding(struct cpuinfo_x86 *c, unsigned int bank)
633 {
634 	int i, num_msrs;
635 	u64 hwcr;
636 	bool need_toggle;
637 	u32 msrs[NR_BLOCKS];
638 
639 	if (c->x86 == 0x15 && bank == 4) {
640 		msrs[0] = 0x00000413; /* MC4_MISC0 */
641 		msrs[1] = 0xc0000408; /* MC4_MISC1 */
642 		num_msrs = 2;
643 	} else if (c->x86 == 0x17 &&
644 		   (c->x86_model >= 0x10 && c->x86_model <= 0x2F)) {
645 
646 		if (smca_get_bank_type(smp_processor_id(), bank) != SMCA_IF)
647 			return;
648 
649 		msrs[0] = MSR_AMD64_SMCA_MCx_MISC(bank);
650 		num_msrs = 1;
651 	} else {
652 		return;
653 	}
654 
655 	rdmsrl(MSR_K7_HWCR, hwcr);
656 
657 	/* McStatusWrEn has to be set */
658 	need_toggle = !(hwcr & BIT(18));
659 	if (need_toggle)
660 		wrmsrl(MSR_K7_HWCR, hwcr | BIT(18));
661 
662 	/* Clear CntP bit safely */
663 	for (i = 0; i < num_msrs; i++)
664 		msr_clear_bit(msrs[i], 62);
665 
666 	/* restore old settings */
667 	if (need_toggle)
668 		wrmsrl(MSR_K7_HWCR, hwcr);
669 }
670 
671 /* cpu init entry point, called from mce.c with preempt off */
672 void mce_amd_feature_init(struct cpuinfo_x86 *c)
673 {
674 	unsigned int bank, block, cpu = smp_processor_id();
675 	u32 low = 0, high = 0, address = 0;
676 	int offset = -1;
677 
678 
679 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
680 		if (mce_flags.smca)
681 			smca_configure(bank, cpu);
682 
683 		disable_err_thresholding(c, bank);
684 
685 		for (block = 0; block < NR_BLOCKS; ++block) {
686 			address = get_block_address(address, low, high, bank, block, cpu);
687 			if (!address)
688 				break;
689 
690 			if (rdmsr_safe(address, &low, &high))
691 				break;
692 
693 			if (!(high & MASK_VALID_HI))
694 				continue;
695 
696 			if (!(high & MASK_CNTP_HI)  ||
697 			     (high & MASK_LOCKED_HI))
698 				continue;
699 
700 			offset = prepare_threshold_block(bank, block, address, offset, high);
701 		}
702 	}
703 
704 	if (mce_flags.succor)
705 		deferred_error_interrupt_enable(c);
706 }
707 
708 /*
709  * DRAM ECC errors are reported in the Northbridge (bank 4) with
710  * Extended Error Code 8.
711  */
712 static bool legacy_mce_is_memory_error(struct mce *m)
713 {
714 	return m->bank == 4 && XEC(m->status, 0x1f) == 8;
715 }
716 
717 /*
718  * DRAM ECC errors are reported in Unified Memory Controllers with
719  * Extended Error Code 0.
720  */
721 static bool smca_mce_is_memory_error(struct mce *m)
722 {
723 	enum smca_bank_types bank_type;
724 
725 	if (XEC(m->status, 0x3f))
726 		return false;
727 
728 	bank_type = smca_get_bank_type(m->extcpu, m->bank);
729 
730 	return bank_type == SMCA_UMC || bank_type == SMCA_UMC_V2;
731 }
732 
733 bool amd_mce_is_memory_error(struct mce *m)
734 {
735 	if (mce_flags.smca)
736 		return smca_mce_is_memory_error(m);
737 	else
738 		return legacy_mce_is_memory_error(m);
739 }
740 
741 /*
742  * AMD systems do not have an explicit indicator that the value in MCA_ADDR is
743  * a system physical address. Therefore, individual cases need to be detected.
744  * Future cases and checks will be added as needed.
745  *
746  * 1) General case
747  *	a) Assume address is not usable.
748  * 2) Poison errors
749  *	a) Indicated by MCA_STATUS[43]: poison. Defined for all banks except legacy
750  *	   northbridge (bank 4).
751  *	b) Refers to poison consumption in the core. Does not include "no action",
752  *	   "action optional", or "deferred" error severities.
753  *	c) Will include a usable address so that immediate action can be taken.
754  * 3) Northbridge DRAM ECC errors
755  *	a) Reported in legacy bank 4 with extended error code (XEC) 8.
756  *	b) MCA_STATUS[43] is *not* defined as poison in legacy bank 4. Therefore,
757  *	   this bit should not be checked.
758  *
759  * NOTE: SMCA UMC memory errors fall into case #1.
760  */
761 bool amd_mce_usable_address(struct mce *m)
762 {
763 	/* Check special northbridge case 3) first. */
764 	if (!mce_flags.smca) {
765 		if (legacy_mce_is_memory_error(m))
766 			return true;
767 		else if (m->bank == 4)
768 			return false;
769 	}
770 
771 	/* Check poison bit for all other bank types. */
772 	if (m->status & MCI_STATUS_POISON)
773 		return true;
774 
775 	/* Assume address is not usable for all others. */
776 	return false;
777 }
778 
779 static void __log_error(unsigned int bank, u64 status, u64 addr, u64 misc)
780 {
781 	struct mce m;
782 
783 	mce_prep_record(&m);
784 
785 	m.status = status;
786 	m.misc   = misc;
787 	m.bank   = bank;
788 	m.tsc	 = rdtsc();
789 
790 	if (m.status & MCI_STATUS_ADDRV) {
791 		m.addr = addr;
792 
793 		smca_extract_err_addr(&m);
794 	}
795 
796 	if (mce_flags.smca) {
797 		rdmsrl(MSR_AMD64_SMCA_MCx_IPID(bank), m.ipid);
798 
799 		if (m.status & MCI_STATUS_SYNDV)
800 			rdmsrl(MSR_AMD64_SMCA_MCx_SYND(bank), m.synd);
801 	}
802 
803 	mce_log(&m);
804 }
805 
806 DEFINE_IDTENTRY_SYSVEC(sysvec_deferred_error)
807 {
808 	trace_deferred_error_apic_entry(DEFERRED_ERROR_VECTOR);
809 	inc_irq_stat(irq_deferred_error_count);
810 	deferred_error_int_vector();
811 	trace_deferred_error_apic_exit(DEFERRED_ERROR_VECTOR);
812 	apic_eoi();
813 }
814 
815 /*
816  * Returns true if the logged error is deferred. False, otherwise.
817  */
818 static inline bool
819 _log_error_bank(unsigned int bank, u32 msr_stat, u32 msr_addr, u64 misc)
820 {
821 	u64 status, addr = 0;
822 
823 	rdmsrl(msr_stat, status);
824 	if (!(status & MCI_STATUS_VAL))
825 		return false;
826 
827 	if (status & MCI_STATUS_ADDRV)
828 		rdmsrl(msr_addr, addr);
829 
830 	__log_error(bank, status, addr, misc);
831 
832 	wrmsrl(msr_stat, 0);
833 
834 	return status & MCI_STATUS_DEFERRED;
835 }
836 
837 static bool _log_error_deferred(unsigned int bank, u32 misc)
838 {
839 	if (!_log_error_bank(bank, mca_msr_reg(bank, MCA_STATUS),
840 			     mca_msr_reg(bank, MCA_ADDR), misc))
841 		return false;
842 
843 	/*
844 	 * Non-SMCA systems don't have MCA_DESTAT/MCA_DEADDR registers.
845 	 * Return true here to avoid accessing these registers.
846 	 */
847 	if (!mce_flags.smca)
848 		return true;
849 
850 	/* Clear MCA_DESTAT if the deferred error was logged from MCA_STATUS. */
851 	wrmsrl(MSR_AMD64_SMCA_MCx_DESTAT(bank), 0);
852 	return true;
853 }
854 
855 /*
856  * We have three scenarios for checking for Deferred errors:
857  *
858  * 1) Non-SMCA systems check MCA_STATUS and log error if found.
859  * 2) SMCA systems check MCA_STATUS. If error is found then log it and also
860  *    clear MCA_DESTAT.
861  * 3) SMCA systems check MCA_DESTAT, if error was not found in MCA_STATUS, and
862  *    log it.
863  */
864 static void log_error_deferred(unsigned int bank)
865 {
866 	if (_log_error_deferred(bank, 0))
867 		return;
868 
869 	/*
870 	 * Only deferred errors are logged in MCA_DE{STAT,ADDR} so just check
871 	 * for a valid error.
872 	 */
873 	_log_error_bank(bank, MSR_AMD64_SMCA_MCx_DESTAT(bank),
874 			      MSR_AMD64_SMCA_MCx_DEADDR(bank), 0);
875 }
876 
877 /* APIC interrupt handler for deferred errors */
878 static void amd_deferred_error_interrupt(void)
879 {
880 	unsigned int bank;
881 
882 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank)
883 		log_error_deferred(bank);
884 }
885 
886 static void log_error_thresholding(unsigned int bank, u64 misc)
887 {
888 	_log_error_deferred(bank, misc);
889 }
890 
891 static void log_and_reset_block(struct threshold_block *block)
892 {
893 	struct thresh_restart tr;
894 	u32 low = 0, high = 0;
895 
896 	if (!block)
897 		return;
898 
899 	if (rdmsr_safe(block->address, &low, &high))
900 		return;
901 
902 	if (!(high & MASK_OVERFLOW_HI))
903 		return;
904 
905 	/* Log the MCE which caused the threshold event. */
906 	log_error_thresholding(block->bank, ((u64)high << 32) | low);
907 
908 	/* Reset threshold block after logging error. */
909 	memset(&tr, 0, sizeof(tr));
910 	tr.b = block;
911 	threshold_restart_bank(&tr);
912 }
913 
914 /*
915  * Threshold interrupt handler will service THRESHOLD_APIC_VECTOR. The interrupt
916  * goes off when error_count reaches threshold_limit.
917  */
918 static void amd_threshold_interrupt(void)
919 {
920 	struct threshold_block *first_block = NULL, *block = NULL, *tmp = NULL;
921 	struct threshold_bank **bp = this_cpu_read(threshold_banks);
922 	unsigned int bank, cpu = smp_processor_id();
923 
924 	/*
925 	 * Validate that the threshold bank has been initialized already. The
926 	 * handler is installed at boot time, but on a hotplug event the
927 	 * interrupt might fire before the data has been initialized.
928 	 */
929 	if (!bp)
930 		return;
931 
932 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
933 		if (!(per_cpu(bank_map, cpu) & BIT_ULL(bank)))
934 			continue;
935 
936 		first_block = bp[bank]->blocks;
937 		if (!first_block)
938 			continue;
939 
940 		/*
941 		 * The first block is also the head of the list. Check it first
942 		 * before iterating over the rest.
943 		 */
944 		log_and_reset_block(first_block);
945 		list_for_each_entry_safe(block, tmp, &first_block->miscj, miscj)
946 			log_and_reset_block(block);
947 	}
948 }
949 
950 /*
951  * Sysfs Interface
952  */
953 
954 struct threshold_attr {
955 	struct attribute attr;
956 	ssize_t (*show) (struct threshold_block *, char *);
957 	ssize_t (*store) (struct threshold_block *, const char *, size_t count);
958 };
959 
960 #define SHOW_FIELDS(name)						\
961 static ssize_t show_ ## name(struct threshold_block *b, char *buf)	\
962 {									\
963 	return sprintf(buf, "%lu\n", (unsigned long) b->name);		\
964 }
965 SHOW_FIELDS(interrupt_enable)
966 SHOW_FIELDS(threshold_limit)
967 
968 static ssize_t
969 store_interrupt_enable(struct threshold_block *b, const char *buf, size_t size)
970 {
971 	struct thresh_restart tr;
972 	unsigned long new;
973 
974 	if (!b->interrupt_capable)
975 		return -EINVAL;
976 
977 	if (kstrtoul(buf, 0, &new) < 0)
978 		return -EINVAL;
979 
980 	b->interrupt_enable = !!new;
981 
982 	memset(&tr, 0, sizeof(tr));
983 	tr.b		= b;
984 
985 	if (smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1))
986 		return -ENODEV;
987 
988 	return size;
989 }
990 
991 static ssize_t
992 store_threshold_limit(struct threshold_block *b, const char *buf, size_t size)
993 {
994 	struct thresh_restart tr;
995 	unsigned long new;
996 
997 	if (kstrtoul(buf, 0, &new) < 0)
998 		return -EINVAL;
999 
1000 	if (new > THRESHOLD_MAX)
1001 		new = THRESHOLD_MAX;
1002 	if (new < 1)
1003 		new = 1;
1004 
1005 	memset(&tr, 0, sizeof(tr));
1006 	tr.old_limit = b->threshold_limit;
1007 	b->threshold_limit = new;
1008 	tr.b = b;
1009 
1010 	if (smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1))
1011 		return -ENODEV;
1012 
1013 	return size;
1014 }
1015 
1016 static ssize_t show_error_count(struct threshold_block *b, char *buf)
1017 {
1018 	u32 lo, hi;
1019 
1020 	/* CPU might be offline by now */
1021 	if (rdmsr_on_cpu(b->cpu, b->address, &lo, &hi))
1022 		return -ENODEV;
1023 
1024 	return sprintf(buf, "%u\n", ((hi & THRESHOLD_MAX) -
1025 				     (THRESHOLD_MAX - b->threshold_limit)));
1026 }
1027 
1028 static struct threshold_attr error_count = {
1029 	.attr = {.name = __stringify(error_count), .mode = 0444 },
1030 	.show = show_error_count,
1031 };
1032 
1033 #define RW_ATTR(val)							\
1034 static struct threshold_attr val = {					\
1035 	.attr	= {.name = __stringify(val), .mode = 0644 },		\
1036 	.show	= show_## val,						\
1037 	.store	= store_## val,						\
1038 };
1039 
1040 RW_ATTR(interrupt_enable);
1041 RW_ATTR(threshold_limit);
1042 
1043 static struct attribute *default_attrs[] = {
1044 	&threshold_limit.attr,
1045 	&error_count.attr,
1046 	NULL,	/* possibly interrupt_enable if supported, see below */
1047 	NULL,
1048 };
1049 ATTRIBUTE_GROUPS(default);
1050 
1051 #define to_block(k)	container_of(k, struct threshold_block, kobj)
1052 #define to_attr(a)	container_of(a, struct threshold_attr, attr)
1053 
1054 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
1055 {
1056 	struct threshold_block *b = to_block(kobj);
1057 	struct threshold_attr *a = to_attr(attr);
1058 	ssize_t ret;
1059 
1060 	ret = a->show ? a->show(b, buf) : -EIO;
1061 
1062 	return ret;
1063 }
1064 
1065 static ssize_t store(struct kobject *kobj, struct attribute *attr,
1066 		     const char *buf, size_t count)
1067 {
1068 	struct threshold_block *b = to_block(kobj);
1069 	struct threshold_attr *a = to_attr(attr);
1070 	ssize_t ret;
1071 
1072 	ret = a->store ? a->store(b, buf, count) : -EIO;
1073 
1074 	return ret;
1075 }
1076 
1077 static const struct sysfs_ops threshold_ops = {
1078 	.show			= show,
1079 	.store			= store,
1080 };
1081 
1082 static void threshold_block_release(struct kobject *kobj);
1083 
1084 static const struct kobj_type threshold_ktype = {
1085 	.sysfs_ops		= &threshold_ops,
1086 	.default_groups		= default_groups,
1087 	.release		= threshold_block_release,
1088 };
1089 
1090 static const char *get_name(unsigned int cpu, unsigned int bank, struct threshold_block *b)
1091 {
1092 	enum smca_bank_types bank_type;
1093 
1094 	if (!mce_flags.smca) {
1095 		if (b && bank == 4)
1096 			return bank4_names(b);
1097 
1098 		return th_names[bank];
1099 	}
1100 
1101 	bank_type = smca_get_bank_type(cpu, bank);
1102 	if (bank_type >= N_SMCA_BANK_TYPES)
1103 		return NULL;
1104 
1105 	if (b && (bank_type == SMCA_UMC || bank_type == SMCA_UMC_V2)) {
1106 		if (b->block < ARRAY_SIZE(smca_umc_block_names))
1107 			return smca_umc_block_names[b->block];
1108 		return NULL;
1109 	}
1110 
1111 	if (per_cpu(smca_bank_counts, cpu)[bank_type] == 1)
1112 		return smca_get_name(bank_type);
1113 
1114 	snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN,
1115 		 "%s_%u", smca_get_name(bank_type),
1116 			  per_cpu(smca_banks, cpu)[bank].sysfs_id);
1117 	return buf_mcatype;
1118 }
1119 
1120 static int allocate_threshold_blocks(unsigned int cpu, struct threshold_bank *tb,
1121 				     unsigned int bank, unsigned int block,
1122 				     u32 address)
1123 {
1124 	struct threshold_block *b = NULL;
1125 	u32 low, high;
1126 	int err;
1127 
1128 	if ((bank >= this_cpu_read(mce_num_banks)) || (block >= NR_BLOCKS))
1129 		return 0;
1130 
1131 	if (rdmsr_safe(address, &low, &high))
1132 		return 0;
1133 
1134 	if (!(high & MASK_VALID_HI)) {
1135 		if (block)
1136 			goto recurse;
1137 		else
1138 			return 0;
1139 	}
1140 
1141 	if (!(high & MASK_CNTP_HI)  ||
1142 	     (high & MASK_LOCKED_HI))
1143 		goto recurse;
1144 
1145 	b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
1146 	if (!b)
1147 		return -ENOMEM;
1148 
1149 	b->block		= block;
1150 	b->bank			= bank;
1151 	b->cpu			= cpu;
1152 	b->address		= address;
1153 	b->interrupt_enable	= 0;
1154 	b->interrupt_capable	= lvt_interrupt_supported(bank, high);
1155 	b->threshold_limit	= THRESHOLD_MAX;
1156 
1157 	if (b->interrupt_capable) {
1158 		default_attrs[2] = &interrupt_enable.attr;
1159 		b->interrupt_enable = 1;
1160 	} else {
1161 		default_attrs[2] = NULL;
1162 	}
1163 
1164 	INIT_LIST_HEAD(&b->miscj);
1165 
1166 	/* This is safe as @tb is not visible yet */
1167 	if (tb->blocks)
1168 		list_add(&b->miscj, &tb->blocks->miscj);
1169 	else
1170 		tb->blocks = b;
1171 
1172 	err = kobject_init_and_add(&b->kobj, &threshold_ktype, tb->kobj, get_name(cpu, bank, b));
1173 	if (err)
1174 		goto out_free;
1175 recurse:
1176 	address = get_block_address(address, low, high, bank, ++block, cpu);
1177 	if (!address)
1178 		return 0;
1179 
1180 	err = allocate_threshold_blocks(cpu, tb, bank, block, address);
1181 	if (err)
1182 		goto out_free;
1183 
1184 	if (b)
1185 		kobject_uevent(&b->kobj, KOBJ_ADD);
1186 
1187 	return 0;
1188 
1189 out_free:
1190 	if (b) {
1191 		list_del(&b->miscj);
1192 		kobject_put(&b->kobj);
1193 	}
1194 	return err;
1195 }
1196 
1197 static int __threshold_add_blocks(struct threshold_bank *b)
1198 {
1199 	struct list_head *head = &b->blocks->miscj;
1200 	struct threshold_block *pos = NULL;
1201 	struct threshold_block *tmp = NULL;
1202 	int err = 0;
1203 
1204 	err = kobject_add(&b->blocks->kobj, b->kobj, b->blocks->kobj.name);
1205 	if (err)
1206 		return err;
1207 
1208 	list_for_each_entry_safe(pos, tmp, head, miscj) {
1209 
1210 		err = kobject_add(&pos->kobj, b->kobj, pos->kobj.name);
1211 		if (err) {
1212 			list_for_each_entry_safe_reverse(pos, tmp, head, miscj)
1213 				kobject_del(&pos->kobj);
1214 
1215 			return err;
1216 		}
1217 	}
1218 	return err;
1219 }
1220 
1221 static int threshold_create_bank(struct threshold_bank **bp, unsigned int cpu,
1222 				 unsigned int bank)
1223 {
1224 	struct device *dev = this_cpu_read(mce_device);
1225 	struct amd_northbridge *nb = NULL;
1226 	struct threshold_bank *b = NULL;
1227 	const char *name = get_name(cpu, bank, NULL);
1228 	int err = 0;
1229 
1230 	if (!dev)
1231 		return -ENODEV;
1232 
1233 	if (is_shared_bank(bank)) {
1234 		nb = node_to_amd_nb(topology_amd_node_id(cpu));
1235 
1236 		/* threshold descriptor already initialized on this node? */
1237 		if (nb && nb->bank4) {
1238 			/* yes, use it */
1239 			b = nb->bank4;
1240 			err = kobject_add(b->kobj, &dev->kobj, name);
1241 			if (err)
1242 				goto out;
1243 
1244 			bp[bank] = b;
1245 			refcount_inc(&b->cpus);
1246 
1247 			err = __threshold_add_blocks(b);
1248 
1249 			goto out;
1250 		}
1251 	}
1252 
1253 	b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
1254 	if (!b) {
1255 		err = -ENOMEM;
1256 		goto out;
1257 	}
1258 
1259 	/* Associate the bank with the per-CPU MCE device */
1260 	b->kobj = kobject_create_and_add(name, &dev->kobj);
1261 	if (!b->kobj) {
1262 		err = -EINVAL;
1263 		goto out_free;
1264 	}
1265 
1266 	if (is_shared_bank(bank)) {
1267 		b->shared = 1;
1268 		refcount_set(&b->cpus, 1);
1269 
1270 		/* nb is already initialized, see above */
1271 		if (nb) {
1272 			WARN_ON(nb->bank4);
1273 			nb->bank4 = b;
1274 		}
1275 	}
1276 
1277 	err = allocate_threshold_blocks(cpu, b, bank, 0, mca_msr_reg(bank, MCA_MISC));
1278 	if (err)
1279 		goto out_kobj;
1280 
1281 	bp[bank] = b;
1282 	return 0;
1283 
1284 out_kobj:
1285 	kobject_put(b->kobj);
1286 out_free:
1287 	kfree(b);
1288 out:
1289 	return err;
1290 }
1291 
1292 static void threshold_block_release(struct kobject *kobj)
1293 {
1294 	kfree(to_block(kobj));
1295 }
1296 
1297 static void deallocate_threshold_blocks(struct threshold_bank *bank)
1298 {
1299 	struct threshold_block *pos, *tmp;
1300 
1301 	list_for_each_entry_safe(pos, tmp, &bank->blocks->miscj, miscj) {
1302 		list_del(&pos->miscj);
1303 		kobject_put(&pos->kobj);
1304 	}
1305 
1306 	kobject_put(&bank->blocks->kobj);
1307 }
1308 
1309 static void __threshold_remove_blocks(struct threshold_bank *b)
1310 {
1311 	struct threshold_block *pos = NULL;
1312 	struct threshold_block *tmp = NULL;
1313 
1314 	kobject_put(b->kobj);
1315 
1316 	list_for_each_entry_safe(pos, tmp, &b->blocks->miscj, miscj)
1317 		kobject_put(b->kobj);
1318 }
1319 
1320 static void threshold_remove_bank(struct threshold_bank *bank)
1321 {
1322 	struct amd_northbridge *nb;
1323 
1324 	if (!bank->blocks)
1325 		goto out_free;
1326 
1327 	if (!bank->shared)
1328 		goto out_dealloc;
1329 
1330 	if (!refcount_dec_and_test(&bank->cpus)) {
1331 		__threshold_remove_blocks(bank);
1332 		return;
1333 	} else {
1334 		/*
1335 		 * The last CPU on this node using the shared bank is going
1336 		 * away, remove that bank now.
1337 		 */
1338 		nb = node_to_amd_nb(topology_amd_node_id(smp_processor_id()));
1339 		nb->bank4 = NULL;
1340 	}
1341 
1342 out_dealloc:
1343 	deallocate_threshold_blocks(bank);
1344 
1345 out_free:
1346 	kobject_put(bank->kobj);
1347 	kfree(bank);
1348 }
1349 
1350 static void __threshold_remove_device(struct threshold_bank **bp)
1351 {
1352 	unsigned int bank, numbanks = this_cpu_read(mce_num_banks);
1353 
1354 	for (bank = 0; bank < numbanks; bank++) {
1355 		if (!bp[bank])
1356 			continue;
1357 
1358 		threshold_remove_bank(bp[bank]);
1359 		bp[bank] = NULL;
1360 	}
1361 	kfree(bp);
1362 }
1363 
1364 int mce_threshold_remove_device(unsigned int cpu)
1365 {
1366 	struct threshold_bank **bp = this_cpu_read(threshold_banks);
1367 
1368 	if (!bp)
1369 		return 0;
1370 
1371 	/*
1372 	 * Clear the pointer before cleaning up, so that the interrupt won't
1373 	 * touch anything of this.
1374 	 */
1375 	this_cpu_write(threshold_banks, NULL);
1376 
1377 	__threshold_remove_device(bp);
1378 	return 0;
1379 }
1380 
1381 /**
1382  * mce_threshold_create_device - Create the per-CPU MCE threshold device
1383  * @cpu:	The plugged in CPU
1384  *
1385  * Create directories and files for all valid threshold banks.
1386  *
1387  * This is invoked from the CPU hotplug callback which was installed in
1388  * mcheck_init_device(). The invocation happens in context of the hotplug
1389  * thread running on @cpu.  The callback is invoked on all CPUs which are
1390  * online when the callback is installed or during a real hotplug event.
1391  */
1392 int mce_threshold_create_device(unsigned int cpu)
1393 {
1394 	unsigned int numbanks, bank;
1395 	struct threshold_bank **bp;
1396 	int err;
1397 
1398 	if (!mce_flags.amd_threshold)
1399 		return 0;
1400 
1401 	bp = this_cpu_read(threshold_banks);
1402 	if (bp)
1403 		return 0;
1404 
1405 	numbanks = this_cpu_read(mce_num_banks);
1406 	bp = kcalloc(numbanks, sizeof(*bp), GFP_KERNEL);
1407 	if (!bp)
1408 		return -ENOMEM;
1409 
1410 	for (bank = 0; bank < numbanks; ++bank) {
1411 		if (!(this_cpu_read(bank_map) & BIT_ULL(bank)))
1412 			continue;
1413 		err = threshold_create_bank(bp, cpu, bank);
1414 		if (err) {
1415 			__threshold_remove_device(bp);
1416 			return err;
1417 		}
1418 	}
1419 	this_cpu_write(threshold_banks, bp);
1420 
1421 	if (thresholding_irq_en)
1422 		mce_threshold_vector = amd_threshold_interrupt;
1423 	return 0;
1424 }
1425