xref: /linux/arch/x86/kernel/cpu/mce/amd.c (revision 7fc2cd2e4b398c57c9cf961cfea05eadbf34c05c)
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 #include <linux/interrupt.h>
9 #include <linux/notifier.h>
10 #include <linux/kobject.h>
11 #include <linux/percpu.h>
12 #include <linux/errno.h>
13 #include <linux/sched.h>
14 #include <linux/sysfs.h>
15 #include <linux/slab.h>
16 #include <linux/init.h>
17 #include <linux/cpu.h>
18 #include <linux/smp.h>
19 #include <linux/string.h>
20 
21 #include <asm/traps.h>
22 #include <asm/apic.h>
23 #include <asm/mce.h>
24 #include <asm/msr.h>
25 #include <asm/trace/irq_vectors.h>
26 
27 #include "internal.h"
28 
29 #define NR_BLOCKS         5
30 #define THRESHOLD_MAX     0xFFF
31 #define INT_TYPE_APIC     0x00020000
32 #define MASK_VALID_HI     0x80000000
33 #define MASK_CNTP_HI      0x40000000
34 #define MASK_LOCKED_HI    0x20000000
35 #define MASK_LVTOFF_HI    0x00F00000
36 #define MASK_COUNT_EN_HI  0x00080000
37 #define MASK_INT_TYPE_HI  0x00060000
38 #define MASK_OVERFLOW_HI  0x00010000
39 #define MASK_ERR_COUNT_HI 0x00000FFF
40 #define MASK_BLKPTR_LO    0xFF000000
41 #define MCG_XBLK_ADDR     0xC0000400
42 
43 /* Deferred error settings */
44 #define MSR_CU_DEF_ERR		0xC0000410
45 #define MASK_DEF_LVTOFF		0x000000F0
46 
47 /* Scalable MCA: */
48 
49 /* Threshold LVT offset is at MSR0xC0000410[15:12] */
50 #define SMCA_THR_LVT_OFF	0xF000
51 
52 static bool thresholding_irq_en;
53 
54 struct mce_amd_cpu_data {
55 	mce_banks_t     thr_intr_banks;
56 	mce_banks_t     dfr_intr_banks;
57 
58 	u32		thr_intr_en: 1,
59 			dfr_intr_en: 1,
60 			__resv: 30;
61 };
62 
63 static DEFINE_PER_CPU_READ_MOSTLY(struct mce_amd_cpu_data, mce_amd_data);
64 
65 static const char * const th_names[] = {
66 	"load_store",
67 	"insn_fetch",
68 	"combined_unit",
69 	"decode_unit",
70 	"northbridge",
71 	"execution_unit",
72 };
73 
74 static const char * const smca_umc_block_names[] = {
75 	"dram_ecc",
76 	"misc_umc"
77 };
78 
79 #define HWID_MCATYPE(hwid, mcatype) (((hwid) << 16) | (mcatype))
80 
81 struct smca_hwid {
82 	unsigned int bank_type;	/* Use with smca_bank_types for easy indexing. */
83 	u32 hwid_mcatype;	/* (hwid,mcatype) tuple */
84 };
85 
86 struct smca_bank {
87 	const struct smca_hwid *hwid;
88 	u32 id;			/* Value of MCA_IPID[InstanceId]. */
89 	u8 sysfs_id;		/* Value used for sysfs name. */
90 	u64 paddrv	:1,	/* Physical Address Valid bit in MCA_CONFIG */
91 	    __reserved	:63;
92 };
93 
94 static DEFINE_PER_CPU_READ_MOSTLY(struct smca_bank[MAX_NR_BANKS], smca_banks);
95 static DEFINE_PER_CPU_READ_MOSTLY(u8[N_SMCA_BANK_TYPES], smca_bank_counts);
96 
97 static const char * const smca_names[] = {
98 	[SMCA_LS ... SMCA_LS_V2]	= "load_store",
99 	[SMCA_IF]			= "insn_fetch",
100 	[SMCA_L2_CACHE]			= "l2_cache",
101 	[SMCA_DE]			= "decode_unit",
102 	[SMCA_RESERVED]			= "reserved",
103 	[SMCA_EX]			= "execution_unit",
104 	[SMCA_FP]			= "floating_point",
105 	[SMCA_L3_CACHE]			= "l3_cache",
106 	[SMCA_CS ... SMCA_CS_V2]	= "coherent_slave",
107 	[SMCA_PIE]			= "pie",
108 
109 	/* UMC v2 is separate because both of them can exist in a single system. */
110 	[SMCA_UMC]			= "umc",
111 	[SMCA_UMC_V2]			= "umc_v2",
112 	[SMCA_MA_LLC]			= "ma_llc",
113 	[SMCA_PB]			= "param_block",
114 	[SMCA_PSP ... SMCA_PSP_V2]	= "psp",
115 	[SMCA_SMU ... SMCA_SMU_V2]	= "smu",
116 	[SMCA_MP5]			= "mp5",
117 	[SMCA_MPDMA]			= "mpdma",
118 	[SMCA_NBIO]			= "nbio",
119 	[SMCA_PCIE ... SMCA_PCIE_V2]	= "pcie",
120 	[SMCA_XGMI_PCS]			= "xgmi_pcs",
121 	[SMCA_NBIF]			= "nbif",
122 	[SMCA_SHUB]			= "shub",
123 	[SMCA_SATA]			= "sata",
124 	[SMCA_USB]			= "usb",
125 	[SMCA_USR_DP]			= "usr_dp",
126 	[SMCA_USR_CP]			= "usr_cp",
127 	[SMCA_GMI_PCS]			= "gmi_pcs",
128 	[SMCA_XGMI_PHY]			= "xgmi_phy",
129 	[SMCA_WAFL_PHY]			= "wafl_phy",
130 	[SMCA_GMI_PHY]			= "gmi_phy",
131 };
132 
133 static const char *smca_get_name(enum smca_bank_types t)
134 {
135 	if (t >= N_SMCA_BANK_TYPES)
136 		return NULL;
137 
138 	return smca_names[t];
139 }
140 
141 enum smca_bank_types smca_get_bank_type(unsigned int cpu, unsigned int bank)
142 {
143 	struct smca_bank *b;
144 
145 	if (bank >= MAX_NR_BANKS)
146 		return N_SMCA_BANK_TYPES;
147 
148 	b = &per_cpu(smca_banks, cpu)[bank];
149 	if (!b->hwid)
150 		return N_SMCA_BANK_TYPES;
151 
152 	return b->hwid->bank_type;
153 }
154 EXPORT_SYMBOL_GPL(smca_get_bank_type);
155 
156 static const struct smca_hwid smca_hwid_mcatypes[] = {
157 	/* { bank_type, hwid_mcatype } */
158 
159 	/* Reserved type */
160 	{ SMCA_RESERVED, HWID_MCATYPE(0x00, 0x0)	},
161 
162 	/* ZN Core (HWID=0xB0) MCA types */
163 	{ SMCA_LS,	 HWID_MCATYPE(0xB0, 0x0)	},
164 	{ SMCA_LS_V2,	 HWID_MCATYPE(0xB0, 0x10)	},
165 	{ SMCA_IF,	 HWID_MCATYPE(0xB0, 0x1)	},
166 	{ SMCA_L2_CACHE, HWID_MCATYPE(0xB0, 0x2)	},
167 	{ SMCA_DE,	 HWID_MCATYPE(0xB0, 0x3)	},
168 	/* HWID 0xB0 MCATYPE 0x4 is Reserved */
169 	{ SMCA_EX,	 HWID_MCATYPE(0xB0, 0x5)	},
170 	{ SMCA_FP,	 HWID_MCATYPE(0xB0, 0x6)	},
171 	{ SMCA_L3_CACHE, HWID_MCATYPE(0xB0, 0x7)	},
172 
173 	/* Data Fabric MCA types */
174 	{ SMCA_CS,	 HWID_MCATYPE(0x2E, 0x0)	},
175 	{ SMCA_PIE,	 HWID_MCATYPE(0x2E, 0x1)	},
176 	{ SMCA_CS_V2,	 HWID_MCATYPE(0x2E, 0x2)	},
177 	{ SMCA_MA_LLC,	 HWID_MCATYPE(0x2E, 0x4)	},
178 
179 	/* Unified Memory Controller MCA type */
180 	{ SMCA_UMC,	 HWID_MCATYPE(0x96, 0x0)	},
181 	{ SMCA_UMC_V2,	 HWID_MCATYPE(0x96, 0x1)	},
182 
183 	/* Parameter Block MCA type */
184 	{ SMCA_PB,	 HWID_MCATYPE(0x05, 0x0)	},
185 
186 	/* Platform Security Processor MCA type */
187 	{ SMCA_PSP,	 HWID_MCATYPE(0xFF, 0x0)	},
188 	{ SMCA_PSP_V2,	 HWID_MCATYPE(0xFF, 0x1)	},
189 
190 	/* System Management Unit MCA type */
191 	{ SMCA_SMU,	 HWID_MCATYPE(0x01, 0x0)	},
192 	{ SMCA_SMU_V2,	 HWID_MCATYPE(0x01, 0x1)	},
193 
194 	/* Microprocessor 5 Unit MCA type */
195 	{ SMCA_MP5,	 HWID_MCATYPE(0x01, 0x2)	},
196 
197 	/* MPDMA MCA type */
198 	{ SMCA_MPDMA,	 HWID_MCATYPE(0x01, 0x3)	},
199 
200 	/* Northbridge IO Unit MCA type */
201 	{ SMCA_NBIO,	 HWID_MCATYPE(0x18, 0x0)	},
202 
203 	/* PCI Express Unit MCA type */
204 	{ SMCA_PCIE,	 HWID_MCATYPE(0x46, 0x0)	},
205 	{ SMCA_PCIE_V2,	 HWID_MCATYPE(0x46, 0x1)	},
206 
207 	{ SMCA_XGMI_PCS, HWID_MCATYPE(0x50, 0x0)	},
208 	{ SMCA_NBIF,	 HWID_MCATYPE(0x6C, 0x0)	},
209 	{ SMCA_SHUB,	 HWID_MCATYPE(0x80, 0x0)	},
210 	{ SMCA_SATA,	 HWID_MCATYPE(0xA8, 0x0)	},
211 	{ SMCA_USB,	 HWID_MCATYPE(0xAA, 0x0)	},
212 	{ SMCA_USR_DP,	 HWID_MCATYPE(0x170, 0x0)	},
213 	{ SMCA_USR_CP,	 HWID_MCATYPE(0x180, 0x0)	},
214 	{ SMCA_GMI_PCS,  HWID_MCATYPE(0x241, 0x0)	},
215 	{ SMCA_XGMI_PHY, HWID_MCATYPE(0x259, 0x0)	},
216 	{ SMCA_WAFL_PHY, HWID_MCATYPE(0x267, 0x0)	},
217 	{ SMCA_GMI_PHY,	 HWID_MCATYPE(0x269, 0x0)	},
218 };
219 
220 /*
221  * In SMCA enabled processors, we can have multiple banks for a given IP type.
222  * So to define a unique name for each bank, we use a temp c-string to append
223  * the MCA_IPID[InstanceId] to type's name in get_name().
224  *
225  * InstanceId is 32 bits which is 8 characters. Make sure MAX_MCATYPE_NAME_LEN
226  * is greater than 8 plus 1 (for underscore) plus length of longest type name.
227  */
228 #define MAX_MCATYPE_NAME_LEN	30
229 static char buf_mcatype[MAX_MCATYPE_NAME_LEN];
230 
231 struct threshold_block {
232 	/* This block's number within its bank. */
233 	unsigned int		block;
234 	/* MCA bank number that contains this block. */
235 	unsigned int		bank;
236 	/* CPU which controls this block's MCA bank. */
237 	unsigned int		cpu;
238 	/* MCA_MISC MSR address for this block. */
239 	u32			address;
240 	/* Enable/Disable APIC interrupt. */
241 	bool			interrupt_enable;
242 	/* Bank can generate an interrupt. */
243 	bool			interrupt_capable;
244 	/* Value upon which threshold interrupt is generated. */
245 	u16			threshold_limit;
246 	/* sysfs object */
247 	struct kobject		kobj;
248 	/* List of threshold blocks within this block's MCA bank. */
249 	struct list_head	miscj;
250 };
251 
252 struct threshold_bank {
253 	struct kobject		*kobj;
254 	/* List of threshold blocks within this MCA bank. */
255 	struct list_head	miscj;
256 };
257 
258 static DEFINE_PER_CPU(struct threshold_bank **, threshold_banks);
259 
260 /*
261  * A list of the banks enabled on each logical CPU. Controls which respective
262  * descriptors to initialize later in mce_threshold_create_device().
263  */
264 static DEFINE_PER_CPU(u64, bank_map);
265 
266 static void amd_threshold_interrupt(void);
267 static void amd_deferred_error_interrupt(void);
268 
269 static void default_deferred_error_interrupt(void)
270 {
271 	pr_err("Unexpected deferred interrupt at vector %x\n", DEFERRED_ERROR_VECTOR);
272 }
273 void (*deferred_error_int_vector)(void) = default_deferred_error_interrupt;
274 
275 static void smca_configure(unsigned int bank, unsigned int cpu)
276 {
277 	struct mce_amd_cpu_data *data = this_cpu_ptr(&mce_amd_data);
278 	u8 *bank_counts = this_cpu_ptr(smca_bank_counts);
279 	const struct smca_hwid *s_hwid;
280 	unsigned int i, hwid_mcatype;
281 	u32 high, low;
282 	u32 smca_config = MSR_AMD64_SMCA_MCx_CONFIG(bank);
283 
284 	/* Set appropriate bits in MCA_CONFIG */
285 	if (!rdmsr_safe(smca_config, &low, &high)) {
286 		/*
287 		 * OS is required to set the MCAX bit to acknowledge that it is
288 		 * now using the new MSR ranges and new registers under each
289 		 * bank. It also means that the OS will configure deferred
290 		 * errors in the new MCx_CONFIG register. If the bit is not set,
291 		 * uncorrectable errors will cause a system panic.
292 		 *
293 		 * MCA_CONFIG[MCAX] is bit 32 (0 in the high portion of the MSR.)
294 		 */
295 		high |= BIT(0);
296 
297 		/*
298 		 * SMCA sets the Deferred Error Interrupt type per bank.
299 		 *
300 		 * MCA_CONFIG[DeferredIntTypeSupported] is bit 5, and tells us
301 		 * if the DeferredIntType bit field is available.
302 		 *
303 		 * MCA_CONFIG[DeferredIntType] is bits [38:37] ([6:5] in the
304 		 * high portion of the MSR). OS should set this to 0x1 to enable
305 		 * APIC based interrupt. First, check that no interrupt has been
306 		 * set.
307 		 */
308 		if ((low & BIT(5)) && !((high >> 5) & 0x3) && data->dfr_intr_en) {
309 			__set_bit(bank, data->dfr_intr_banks);
310 			high |= BIT(5);
311 		}
312 
313 		/*
314 		 * SMCA Corrected Error Interrupt
315 		 *
316 		 * MCA_CONFIG[IntPresent] is bit 10, and tells us if the bank can
317 		 * send an MCA Thresholding interrupt without the OS initializing
318 		 * this feature. This can be used if the threshold limit is managed
319 		 * by the platform.
320 		 *
321 		 * MCA_CONFIG[IntEn] is bit 40 (8 in the high portion of the MSR).
322 		 * The OS should set this to inform the platform that the OS is ready
323 		 * to handle the MCA Thresholding interrupt.
324 		 */
325 		if ((low & BIT(10)) && data->thr_intr_en) {
326 			__set_bit(bank, data->thr_intr_banks);
327 			high |= BIT(8);
328 		}
329 
330 		this_cpu_ptr(mce_banks_array)[bank].lsb_in_status = !!(low & BIT(8));
331 
332 		if (low & MCI_CONFIG_PADDRV)
333 			this_cpu_ptr(smca_banks)[bank].paddrv = 1;
334 
335 		wrmsr(smca_config, low, high);
336 	}
337 
338 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_IPID(bank), &low, &high)) {
339 		pr_warn("Failed to read MCA_IPID for bank %d\n", bank);
340 		return;
341 	}
342 
343 	hwid_mcatype = HWID_MCATYPE(high & MCI_IPID_HWID,
344 				    (high & MCI_IPID_MCATYPE) >> 16);
345 
346 	for (i = 0; i < ARRAY_SIZE(smca_hwid_mcatypes); i++) {
347 		s_hwid = &smca_hwid_mcatypes[i];
348 
349 		if (hwid_mcatype == s_hwid->hwid_mcatype) {
350 			this_cpu_ptr(smca_banks)[bank].hwid = s_hwid;
351 			this_cpu_ptr(smca_banks)[bank].id = low;
352 			this_cpu_ptr(smca_banks)[bank].sysfs_id = bank_counts[s_hwid->bank_type]++;
353 			break;
354 		}
355 	}
356 }
357 
358 struct thresh_restart {
359 	struct threshold_block	*b;
360 	int			set_lvt_off;
361 	int			lvt_off;
362 	u16			old_limit;
363 };
364 
365 static const char *bank4_names(const struct threshold_block *b)
366 {
367 	switch (b->address) {
368 	/* MSR4_MISC0 */
369 	case 0x00000413:
370 		return "dram";
371 
372 	case 0xc0000408:
373 		return "ht_links";
374 
375 	case 0xc0000409:
376 		return "l3_cache";
377 
378 	default:
379 		WARN(1, "Funny MSR: 0x%08x\n", b->address);
380 		return "";
381 	}
382 };
383 
384 
385 static bool lvt_interrupt_supported(unsigned int bank, u32 msr_high_bits)
386 {
387 	/*
388 	 * bank 4 supports APIC LVT interrupts implicitly since forever.
389 	 */
390 	if (bank == 4)
391 		return true;
392 
393 	/*
394 	 * IntP: interrupt present; if this bit is set, the thresholding
395 	 * bank can generate APIC LVT interrupts
396 	 */
397 	return msr_high_bits & BIT(28);
398 }
399 
400 static bool lvt_off_valid(struct threshold_block *b, int apic, u32 lo, u32 hi)
401 {
402 	int msr = (hi & MASK_LVTOFF_HI) >> 20;
403 
404 	/*
405 	 * On SMCA CPUs, LVT offset is programmed at a different MSR, and
406 	 * the BIOS provides the value. The original field where LVT offset
407 	 * was set is reserved. Return early here:
408 	 */
409 	if (mce_flags.smca)
410 		return false;
411 
412 	if (apic < 0) {
413 		pr_err(FW_BUG "cpu %d, failed to setup threshold interrupt "
414 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n", b->cpu,
415 		       b->bank, b->block, b->address, hi, lo);
416 		return false;
417 	}
418 
419 	if (apic != msr) {
420 		pr_err(FW_BUG "cpu %d, invalid threshold interrupt offset %d "
421 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n",
422 		       b->cpu, apic, b->bank, b->block, b->address, hi, lo);
423 		return false;
424 	}
425 
426 	return true;
427 };
428 
429 /* Reprogram MCx_MISC MSR behind this threshold block. */
430 static void threshold_restart_block(void *_tr)
431 {
432 	struct thresh_restart *tr = _tr;
433 	u32 hi, lo;
434 
435 	/* sysfs write might race against an offline operation */
436 	if (!this_cpu_read(threshold_banks) && !tr->set_lvt_off)
437 		return;
438 
439 	rdmsr(tr->b->address, lo, hi);
440 
441 	/*
442 	 * Reset error count and overflow bit.
443 	 * This is done during init or after handling an interrupt.
444 	 */
445 	if (hi & MASK_OVERFLOW_HI || tr->set_lvt_off) {
446 		hi &= ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI);
447 		hi |= THRESHOLD_MAX - tr->b->threshold_limit;
448 	} else if (tr->old_limit) {	/* change limit w/o reset */
449 		int new_count = (hi & THRESHOLD_MAX) +
450 		    (tr->old_limit - tr->b->threshold_limit);
451 
452 		hi = (hi & ~MASK_ERR_COUNT_HI) |
453 		    (new_count & THRESHOLD_MAX);
454 	}
455 
456 	/* clear IntType */
457 	hi &= ~MASK_INT_TYPE_HI;
458 
459 	if (!tr->b->interrupt_capable)
460 		goto done;
461 
462 	if (tr->set_lvt_off) {
463 		if (lvt_off_valid(tr->b, tr->lvt_off, lo, hi)) {
464 			/* set new lvt offset */
465 			hi &= ~MASK_LVTOFF_HI;
466 			hi |= tr->lvt_off << 20;
467 		}
468 	}
469 
470 	if (tr->b->interrupt_enable)
471 		hi |= INT_TYPE_APIC;
472 
473  done:
474 
475 	hi |= MASK_COUNT_EN_HI;
476 	wrmsr(tr->b->address, lo, hi);
477 }
478 
479 static void threshold_restart_bank(unsigned int bank, bool intr_en)
480 {
481 	struct threshold_bank **thr_banks = this_cpu_read(threshold_banks);
482 	struct threshold_block *block, *tmp;
483 	struct thresh_restart tr;
484 
485 	if (!thr_banks || !thr_banks[bank])
486 		return;
487 
488 	memset(&tr, 0, sizeof(tr));
489 
490 	list_for_each_entry_safe(block, tmp, &thr_banks[bank]->miscj, miscj) {
491 		tr.b = block;
492 		tr.b->interrupt_enable = intr_en;
493 		threshold_restart_block(&tr);
494 	}
495 }
496 
497 /* Try to use the threshold limit reported through APEI. */
498 static u16 get_thr_limit(void)
499 {
500 	u32 thr_limit = mce_get_apei_thr_limit();
501 
502 	/* Fallback to old default if APEI limit is not available. */
503 	if (!thr_limit)
504 		return THRESHOLD_MAX;
505 
506 	return min(thr_limit, THRESHOLD_MAX);
507 }
508 
509 static void mce_threshold_block_init(struct threshold_block *b, int offset)
510 {
511 	struct thresh_restart tr = {
512 		.b			= b,
513 		.set_lvt_off		= 1,
514 		.lvt_off		= offset,
515 	};
516 
517 	b->threshold_limit		= get_thr_limit();
518 	threshold_restart_block(&tr);
519 };
520 
521 static int setup_APIC_mce_threshold(int reserved, int new)
522 {
523 	if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
524 					      APIC_EILVT_MSG_FIX, 0))
525 		return new;
526 
527 	return reserved;
528 }
529 
530 static u32 get_block_address(u32 current_addr, u32 low, u32 high,
531 			     unsigned int bank, unsigned int block,
532 			     unsigned int cpu)
533 {
534 	u32 addr = 0, offset = 0;
535 
536 	if ((bank >= per_cpu(mce_num_banks, cpu)) || (block >= NR_BLOCKS))
537 		return addr;
538 
539 	if (mce_flags.smca) {
540 		if (!block)
541 			return MSR_AMD64_SMCA_MCx_MISC(bank);
542 
543 		if (!(low & MASK_BLKPTR_LO))
544 			return 0;
545 
546 		return MSR_AMD64_SMCA_MCx_MISCy(bank, block - 1);
547 	}
548 
549 	/* Fall back to method we used for older processors: */
550 	switch (block) {
551 	case 0:
552 		addr = mca_msr_reg(bank, MCA_MISC);
553 		break;
554 	case 1:
555 		offset = ((low & MASK_BLKPTR_LO) >> 21);
556 		if (offset)
557 			addr = MCG_XBLK_ADDR + offset;
558 		break;
559 	default:
560 		addr = ++current_addr;
561 	}
562 	return addr;
563 }
564 
565 static int prepare_threshold_block(unsigned int bank, unsigned int block, u32 addr,
566 				   int offset, u32 misc_high)
567 {
568 	unsigned int cpu = smp_processor_id();
569 	struct threshold_block b;
570 	int new;
571 
572 	if (!block)
573 		per_cpu(bank_map, cpu) |= BIT_ULL(bank);
574 
575 	memset(&b, 0, sizeof(b));
576 	b.cpu			= cpu;
577 	b.bank			= bank;
578 	b.block			= block;
579 	b.address		= addr;
580 	b.interrupt_capable	= lvt_interrupt_supported(bank, misc_high);
581 
582 	if (!b.interrupt_capable)
583 		goto done;
584 
585 	__set_bit(bank, this_cpu_ptr(&mce_amd_data)->thr_intr_banks);
586 	b.interrupt_enable = 1;
587 
588 	if (mce_flags.smca)
589 		goto done;
590 
591 	new = (misc_high & MASK_LVTOFF_HI) >> 20;
592 	offset = setup_APIC_mce_threshold(offset, new);
593 	if (offset == new)
594 		thresholding_irq_en = true;
595 
596 done:
597 	mce_threshold_block_init(&b, offset);
598 
599 	return offset;
600 }
601 
602 bool amd_filter_mce(struct mce *m)
603 {
604 	enum smca_bank_types bank_type = smca_get_bank_type(m->extcpu, m->bank);
605 	struct cpuinfo_x86 *c = &boot_cpu_data;
606 
607 	/* See Family 17h Models 10h-2Fh Erratum #1114. */
608 	if (c->x86 == 0x17 &&
609 	    c->x86_model >= 0x10 && c->x86_model <= 0x2F &&
610 	    bank_type == SMCA_IF && XEC(m->status, 0x3f) == 10)
611 		return true;
612 
613 	/* NB GART TLB error reporting is disabled by default. */
614 	if (c->x86 < 0x17) {
615 		if (m->bank == 4 && XEC(m->status, 0x1f) == 0x5)
616 			return true;
617 	}
618 
619 	return false;
620 }
621 
622 /*
623  * Turn off thresholding banks for the following conditions:
624  * - MC4_MISC thresholding is not supported on Family 0x15.
625  * - Prevent possible spurious interrupts from the IF bank on Family 0x17
626  *   Models 0x10-0x2F due to Erratum #1114.
627  */
628 static void disable_err_thresholding(struct cpuinfo_x86 *c, unsigned int bank)
629 {
630 	int i, num_msrs;
631 	u64 hwcr;
632 	bool need_toggle;
633 	u32 msrs[NR_BLOCKS];
634 
635 	if (c->x86 == 0x15 && bank == 4) {
636 		msrs[0] = 0x00000413; /* MC4_MISC0 */
637 		msrs[1] = 0xc0000408; /* MC4_MISC1 */
638 		num_msrs = 2;
639 	} else if (c->x86 == 0x17 &&
640 		   (c->x86_model >= 0x10 && c->x86_model <= 0x2F)) {
641 
642 		if (smca_get_bank_type(smp_processor_id(), bank) != SMCA_IF)
643 			return;
644 
645 		msrs[0] = MSR_AMD64_SMCA_MCx_MISC(bank);
646 		num_msrs = 1;
647 	} else {
648 		return;
649 	}
650 
651 	rdmsrq(MSR_K7_HWCR, hwcr);
652 
653 	/* McStatusWrEn has to be set */
654 	need_toggle = !(hwcr & BIT(18));
655 	if (need_toggle)
656 		wrmsrq(MSR_K7_HWCR, hwcr | BIT(18));
657 
658 	/* Clear CntP bit safely */
659 	for (i = 0; i < num_msrs; i++)
660 		msr_clear_bit(msrs[i], 62);
661 
662 	/* restore old settings */
663 	if (need_toggle)
664 		wrmsrq(MSR_K7_HWCR, hwcr);
665 }
666 
667 static void amd_apply_cpu_quirks(struct cpuinfo_x86 *c)
668 {
669 	struct mce_bank *mce_banks = this_cpu_ptr(mce_banks_array);
670 
671 	/* This should be disabled by the BIOS, but isn't always */
672 	if (c->x86 == 15 && this_cpu_read(mce_num_banks) > 4) {
673 		/*
674 		 * disable GART TBL walk error reporting, which
675 		 * trips off incorrectly with the IOMMU & 3ware
676 		 * & Cerberus:
677 		 */
678 		clear_bit(10, (unsigned long *)&mce_banks[4].ctl);
679 	}
680 
681 	/*
682 	 * Various K7s with broken bank 0 around. Always disable
683 	 * by default.
684 	 */
685 	if (c->x86 == 6 && this_cpu_read(mce_num_banks))
686 		mce_banks[0].ctl = 0;
687 }
688 
689 /*
690  * Enable the APIC LVT interrupt vectors once per-CPU. This should be done before hardware is
691  * ready to send interrupts.
692  *
693  * Individual error sources are enabled later during per-bank init.
694  */
695 static void smca_enable_interrupt_vectors(void)
696 {
697 	struct mce_amd_cpu_data *data = this_cpu_ptr(&mce_amd_data);
698 	u64 mca_intr_cfg, offset;
699 
700 	if (!mce_flags.smca || !mce_flags.succor)
701 		return;
702 
703 	if (rdmsrq_safe(MSR_CU_DEF_ERR, &mca_intr_cfg))
704 		return;
705 
706 	offset = (mca_intr_cfg & SMCA_THR_LVT_OFF) >> 12;
707 	if (!setup_APIC_eilvt(offset, THRESHOLD_APIC_VECTOR, APIC_EILVT_MSG_FIX, 0))
708 		data->thr_intr_en = 1;
709 
710 	offset = (mca_intr_cfg & MASK_DEF_LVTOFF) >> 4;
711 	if (!setup_APIC_eilvt(offset, DEFERRED_ERROR_VECTOR, APIC_EILVT_MSG_FIX, 0))
712 		data->dfr_intr_en = 1;
713 }
714 
715 /* cpu init entry point, called from mce.c with preempt off */
716 void mce_amd_feature_init(struct cpuinfo_x86 *c)
717 {
718 	unsigned int bank, block, cpu = smp_processor_id();
719 	u32 low = 0, high = 0, address = 0;
720 	int offset = -1;
721 
722 	amd_apply_cpu_quirks(c);
723 
724 	mce_flags.amd_threshold	 = 1;
725 
726 	smca_enable_interrupt_vectors();
727 
728 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
729 		if (mce_flags.smca) {
730 			smca_configure(bank, cpu);
731 
732 			if (!this_cpu_ptr(&mce_amd_data)->thr_intr_en)
733 				continue;
734 		}
735 
736 		disable_err_thresholding(c, bank);
737 
738 		for (block = 0; block < NR_BLOCKS; ++block) {
739 			address = get_block_address(address, low, high, bank, block, cpu);
740 			if (!address)
741 				break;
742 
743 			if (rdmsr_safe(address, &low, &high))
744 				break;
745 
746 			if (!(high & MASK_VALID_HI))
747 				continue;
748 
749 			if (!(high & MASK_CNTP_HI)  ||
750 			     (high & MASK_LOCKED_HI))
751 				continue;
752 
753 			offset = prepare_threshold_block(bank, block, address, offset, high);
754 		}
755 	}
756 }
757 
758 void smca_bsp_init(void)
759 {
760 	mce_threshold_vector	  = amd_threshold_interrupt;
761 	deferred_error_int_vector = amd_deferred_error_interrupt;
762 }
763 
764 /*
765  * DRAM ECC errors are reported in the Northbridge (bank 4) with
766  * Extended Error Code 8.
767  */
768 static bool legacy_mce_is_memory_error(struct mce *m)
769 {
770 	return m->bank == 4 && XEC(m->status, 0x1f) == 8;
771 }
772 
773 /*
774  * DRAM ECC errors are reported in Unified Memory Controllers with
775  * Extended Error Code 0.
776  */
777 static bool smca_mce_is_memory_error(struct mce *m)
778 {
779 	enum smca_bank_types bank_type;
780 
781 	if (XEC(m->status, 0x3f))
782 		return false;
783 
784 	bank_type = smca_get_bank_type(m->extcpu, m->bank);
785 
786 	return bank_type == SMCA_UMC || bank_type == SMCA_UMC_V2;
787 }
788 
789 bool amd_mce_is_memory_error(struct mce *m)
790 {
791 	if (mce_flags.smca)
792 		return smca_mce_is_memory_error(m);
793 	else
794 		return legacy_mce_is_memory_error(m);
795 }
796 
797 /*
798  * Some AMD systems have an explicit indicator that the value in MCA_ADDR is a
799  * system physical address. Individual cases though, need to be detected for
800  * other systems. Future cases will be added as needed.
801  *
802  * 1) General case
803  *	a) Assume address is not usable.
804  * 2) Poison errors
805  *	a) Indicated by MCA_STATUS[43]: poison. Defined for all banks except legacy
806  *	   northbridge (bank 4).
807  *	b) Refers to poison consumption in the core. Does not include "no action",
808  *	   "action optional", or "deferred" error severities.
809  *	c) Will include a usable address so that immediate action can be taken.
810  * 3) Northbridge DRAM ECC errors
811  *	a) Reported in legacy bank 4 with extended error code (XEC) 8.
812  *	b) MCA_STATUS[43] is *not* defined as poison in legacy bank 4. Therefore,
813  *	   this bit should not be checked.
814  * 4) MCI_STATUS_PADDRVAL is set
815  *	a) Will provide a valid system physical address.
816  *
817  * NOTE: SMCA UMC memory errors fall into case #1.
818  */
819 bool amd_mce_usable_address(struct mce *m)
820 {
821 	/* Check special northbridge case 3) first. */
822 	if (!mce_flags.smca) {
823 		if (legacy_mce_is_memory_error(m))
824 			return true;
825 		else if (m->bank == 4)
826 			return false;
827 	}
828 
829 	if (this_cpu_ptr(smca_banks)[m->bank].paddrv)
830 		return m->status & MCI_STATUS_PADDRV;
831 
832 	/* Check poison bit for all other bank types. */
833 	if (m->status & MCI_STATUS_POISON)
834 		return true;
835 
836 	/* Assume address is not usable for all others. */
837 	return false;
838 }
839 
840 DEFINE_IDTENTRY_SYSVEC(sysvec_deferred_error)
841 {
842 	trace_deferred_error_apic_entry(DEFERRED_ERROR_VECTOR);
843 	inc_irq_stat(irq_deferred_error_count);
844 	deferred_error_int_vector();
845 	trace_deferred_error_apic_exit(DEFERRED_ERROR_VECTOR);
846 	apic_eoi();
847 }
848 
849 /* APIC interrupt handler for deferred errors */
850 static void amd_deferred_error_interrupt(void)
851 {
852 	machine_check_poll(MCP_TIMESTAMP, &this_cpu_ptr(&mce_amd_data)->dfr_intr_banks);
853 }
854 
855 void mce_amd_handle_storm(unsigned int bank, bool on)
856 {
857 	threshold_restart_bank(bank, on);
858 }
859 
860 static void amd_reset_thr_limit(unsigned int bank)
861 {
862 	threshold_restart_bank(bank, true);
863 }
864 
865 /*
866  * Threshold interrupt handler will service THRESHOLD_APIC_VECTOR. The interrupt
867  * goes off when error_count reaches threshold_limit.
868  */
869 static void amd_threshold_interrupt(void)
870 {
871 	machine_check_poll(MCP_TIMESTAMP, &this_cpu_ptr(&mce_amd_data)->thr_intr_banks);
872 }
873 
874 void amd_clear_bank(struct mce *m)
875 {
876 	amd_reset_thr_limit(m->bank);
877 
878 	/* Clear MCA_DESTAT for all deferred errors even those logged in MCA_STATUS. */
879 	if (m->status & MCI_STATUS_DEFERRED)
880 		mce_wrmsrq(MSR_AMD64_SMCA_MCx_DESTAT(m->bank), 0);
881 
882 	/* Don't clear MCA_STATUS if MCA_DESTAT was used exclusively. */
883 	if (m->kflags & MCE_CHECK_DFR_REGS)
884 		return;
885 
886 	mce_wrmsrq(mca_msr_reg(m->bank, MCA_STATUS), 0);
887 }
888 
889 /*
890  * Sysfs Interface
891  */
892 
893 struct threshold_attr {
894 	struct attribute attr;
895 	ssize_t (*show) (struct threshold_block *, char *);
896 	ssize_t (*store) (struct threshold_block *, const char *, size_t count);
897 };
898 
899 #define SHOW_FIELDS(name)						\
900 static ssize_t show_ ## name(struct threshold_block *b, char *buf)	\
901 {									\
902 	return sprintf(buf, "%lu\n", (unsigned long) b->name);		\
903 }
904 SHOW_FIELDS(interrupt_enable)
905 SHOW_FIELDS(threshold_limit)
906 
907 static ssize_t
908 store_interrupt_enable(struct threshold_block *b, const char *buf, size_t size)
909 {
910 	struct thresh_restart tr;
911 	unsigned long new;
912 
913 	if (!b->interrupt_capable)
914 		return -EINVAL;
915 
916 	if (kstrtoul(buf, 0, &new) < 0)
917 		return -EINVAL;
918 
919 	b->interrupt_enable = !!new;
920 
921 	memset(&tr, 0, sizeof(tr));
922 	tr.b		= b;
923 
924 	if (smp_call_function_single(b->cpu, threshold_restart_block, &tr, 1))
925 		return -ENODEV;
926 
927 	return size;
928 }
929 
930 static ssize_t
931 store_threshold_limit(struct threshold_block *b, const char *buf, size_t size)
932 {
933 	struct thresh_restart tr;
934 	unsigned long new;
935 
936 	if (kstrtoul(buf, 0, &new) < 0)
937 		return -EINVAL;
938 
939 	if (new > THRESHOLD_MAX)
940 		new = THRESHOLD_MAX;
941 	if (new < 1)
942 		new = 1;
943 
944 	memset(&tr, 0, sizeof(tr));
945 	tr.old_limit = b->threshold_limit;
946 	b->threshold_limit = new;
947 	tr.b = b;
948 
949 	if (smp_call_function_single(b->cpu, threshold_restart_block, &tr, 1))
950 		return -ENODEV;
951 
952 	return size;
953 }
954 
955 static ssize_t show_error_count(struct threshold_block *b, char *buf)
956 {
957 	u32 lo, hi;
958 
959 	/* CPU might be offline by now */
960 	if (rdmsr_on_cpu(b->cpu, b->address, &lo, &hi))
961 		return -ENODEV;
962 
963 	return sprintf(buf, "%u\n", ((hi & THRESHOLD_MAX) -
964 				     (THRESHOLD_MAX - b->threshold_limit)));
965 }
966 
967 static struct threshold_attr error_count = {
968 	.attr = {.name = __stringify(error_count), .mode = 0444 },
969 	.show = show_error_count,
970 };
971 
972 #define RW_ATTR(val)							\
973 static struct threshold_attr val = {					\
974 	.attr	= {.name = __stringify(val), .mode = 0644 },		\
975 	.show	= show_## val,						\
976 	.store	= store_## val,						\
977 };
978 
979 RW_ATTR(interrupt_enable);
980 RW_ATTR(threshold_limit);
981 
982 static struct attribute *default_attrs[] = {
983 	&threshold_limit.attr,
984 	&error_count.attr,
985 	NULL,	/* possibly interrupt_enable if supported, see below */
986 	NULL,
987 };
988 ATTRIBUTE_GROUPS(default);
989 
990 #define to_block(k)	container_of(k, struct threshold_block, kobj)
991 #define to_attr(a)	container_of(a, struct threshold_attr, attr)
992 
993 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
994 {
995 	struct threshold_block *b = to_block(kobj);
996 	struct threshold_attr *a = to_attr(attr);
997 	ssize_t ret;
998 
999 	ret = a->show ? a->show(b, buf) : -EIO;
1000 
1001 	return ret;
1002 }
1003 
1004 static ssize_t store(struct kobject *kobj, struct attribute *attr,
1005 		     const char *buf, size_t count)
1006 {
1007 	struct threshold_block *b = to_block(kobj);
1008 	struct threshold_attr *a = to_attr(attr);
1009 	ssize_t ret;
1010 
1011 	ret = a->store ? a->store(b, buf, count) : -EIO;
1012 
1013 	return ret;
1014 }
1015 
1016 static const struct sysfs_ops threshold_ops = {
1017 	.show			= show,
1018 	.store			= store,
1019 };
1020 
1021 static void threshold_block_release(struct kobject *kobj);
1022 
1023 static const struct kobj_type threshold_ktype = {
1024 	.sysfs_ops		= &threshold_ops,
1025 	.default_groups		= default_groups,
1026 	.release		= threshold_block_release,
1027 };
1028 
1029 static const char *get_name(unsigned int cpu, unsigned int bank, struct threshold_block *b)
1030 {
1031 	enum smca_bank_types bank_type;
1032 
1033 	if (!mce_flags.smca) {
1034 		if (b && bank == 4)
1035 			return bank4_names(b);
1036 
1037 		return th_names[bank];
1038 	}
1039 
1040 	bank_type = smca_get_bank_type(cpu, bank);
1041 
1042 	if (b && (bank_type == SMCA_UMC || bank_type == SMCA_UMC_V2)) {
1043 		if (b->block < ARRAY_SIZE(smca_umc_block_names))
1044 			return smca_umc_block_names[b->block];
1045 	}
1046 
1047 	if (b && b->block) {
1048 		snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN, "th_block_%u", b->block);
1049 		return buf_mcatype;
1050 	}
1051 
1052 	if (bank_type >= N_SMCA_BANK_TYPES) {
1053 		snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN, "th_bank_%u", bank);
1054 		return buf_mcatype;
1055 	}
1056 
1057 	if (per_cpu(smca_bank_counts, cpu)[bank_type] == 1)
1058 		return smca_get_name(bank_type);
1059 
1060 	snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN,
1061 		 "%s_%u", smca_get_name(bank_type),
1062 			  per_cpu(smca_banks, cpu)[bank].sysfs_id);
1063 	return buf_mcatype;
1064 }
1065 
1066 static int allocate_threshold_blocks(unsigned int cpu, struct threshold_bank *tb,
1067 				     unsigned int bank, unsigned int block,
1068 				     u32 address)
1069 {
1070 	struct threshold_block *b = NULL;
1071 	u32 low, high;
1072 	int err;
1073 
1074 	if ((bank >= this_cpu_read(mce_num_banks)) || (block >= NR_BLOCKS))
1075 		return 0;
1076 
1077 	if (rdmsr_safe(address, &low, &high))
1078 		return 0;
1079 
1080 	if (!(high & MASK_VALID_HI)) {
1081 		if (block)
1082 			goto recurse;
1083 		else
1084 			return 0;
1085 	}
1086 
1087 	if (!(high & MASK_CNTP_HI)  ||
1088 	     (high & MASK_LOCKED_HI))
1089 		goto recurse;
1090 
1091 	b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
1092 	if (!b)
1093 		return -ENOMEM;
1094 
1095 	b->block		= block;
1096 	b->bank			= bank;
1097 	b->cpu			= cpu;
1098 	b->address		= address;
1099 	b->interrupt_enable	= 0;
1100 	b->interrupt_capable	= lvt_interrupt_supported(bank, high);
1101 	b->threshold_limit	= get_thr_limit();
1102 
1103 	if (b->interrupt_capable) {
1104 		default_attrs[2] = &interrupt_enable.attr;
1105 		b->interrupt_enable = 1;
1106 	} else {
1107 		default_attrs[2] = NULL;
1108 	}
1109 
1110 	list_add(&b->miscj, &tb->miscj);
1111 
1112 	mce_threshold_block_init(b, (high & MASK_LVTOFF_HI) >> 20);
1113 
1114 	err = kobject_init_and_add(&b->kobj, &threshold_ktype, tb->kobj, get_name(cpu, bank, b));
1115 	if (err)
1116 		goto out_free;
1117 recurse:
1118 	address = get_block_address(address, low, high, bank, ++block, cpu);
1119 	if (!address)
1120 		return 0;
1121 
1122 	err = allocate_threshold_blocks(cpu, tb, bank, block, address);
1123 	if (err)
1124 		goto out_free;
1125 
1126 	if (b)
1127 		kobject_uevent(&b->kobj, KOBJ_ADD);
1128 
1129 	return 0;
1130 
1131 out_free:
1132 	if (b) {
1133 		list_del(&b->miscj);
1134 		kobject_put(&b->kobj);
1135 	}
1136 	return err;
1137 }
1138 
1139 static int threshold_create_bank(struct threshold_bank **bp, unsigned int cpu,
1140 				 unsigned int bank)
1141 {
1142 	struct device *dev = this_cpu_read(mce_device);
1143 	struct threshold_bank *b = NULL;
1144 	const char *name = get_name(cpu, bank, NULL);
1145 	int err = 0;
1146 
1147 	if (!dev)
1148 		return -ENODEV;
1149 
1150 	b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
1151 	if (!b) {
1152 		err = -ENOMEM;
1153 		goto out;
1154 	}
1155 
1156 	/* Associate the bank with the per-CPU MCE device */
1157 	b->kobj = kobject_create_and_add(name, &dev->kobj);
1158 	if (!b->kobj) {
1159 		err = -EINVAL;
1160 		goto out_free;
1161 	}
1162 
1163 	INIT_LIST_HEAD(&b->miscj);
1164 
1165 	err = allocate_threshold_blocks(cpu, b, bank, 0, mca_msr_reg(bank, MCA_MISC));
1166 	if (err)
1167 		goto out_kobj;
1168 
1169 	bp[bank] = b;
1170 	return 0;
1171 
1172 out_kobj:
1173 	kobject_put(b->kobj);
1174 out_free:
1175 	kfree(b);
1176 out:
1177 	return err;
1178 }
1179 
1180 static void threshold_block_release(struct kobject *kobj)
1181 {
1182 	kfree(to_block(kobj));
1183 }
1184 
1185 static void threshold_remove_bank(struct threshold_bank *bank)
1186 {
1187 	struct threshold_block *pos, *tmp;
1188 
1189 	list_for_each_entry_safe(pos, tmp, &bank->miscj, miscj) {
1190 		list_del(&pos->miscj);
1191 		kobject_put(&pos->kobj);
1192 	}
1193 
1194 	kobject_put(bank->kobj);
1195 	kfree(bank);
1196 }
1197 
1198 static void __threshold_remove_device(struct threshold_bank **bp)
1199 {
1200 	unsigned int bank, numbanks = this_cpu_read(mce_num_banks);
1201 
1202 	for (bank = 0; bank < numbanks; bank++) {
1203 		if (!bp[bank])
1204 			continue;
1205 
1206 		threshold_remove_bank(bp[bank]);
1207 		bp[bank] = NULL;
1208 	}
1209 	kfree(bp);
1210 }
1211 
1212 void mce_threshold_remove_device(unsigned int cpu)
1213 {
1214 	struct threshold_bank **bp = this_cpu_read(threshold_banks);
1215 
1216 	if (!bp)
1217 		return;
1218 
1219 	/*
1220 	 * Clear the pointer before cleaning up, so that the interrupt won't
1221 	 * touch anything of this.
1222 	 */
1223 	this_cpu_write(threshold_banks, NULL);
1224 
1225 	__threshold_remove_device(bp);
1226 	return;
1227 }
1228 
1229 /**
1230  * mce_threshold_create_device - Create the per-CPU MCE threshold device
1231  * @cpu:	The plugged in CPU
1232  *
1233  * Create directories and files for all valid threshold banks.
1234  *
1235  * This is invoked from the CPU hotplug callback which was installed in
1236  * mcheck_init_device(). The invocation happens in context of the hotplug
1237  * thread running on @cpu.  The callback is invoked on all CPUs which are
1238  * online when the callback is installed or during a real hotplug event.
1239  */
1240 void mce_threshold_create_device(unsigned int cpu)
1241 {
1242 	unsigned int numbanks, bank;
1243 	struct threshold_bank **bp;
1244 
1245 	if (!mce_flags.amd_threshold)
1246 		return;
1247 
1248 	bp = this_cpu_read(threshold_banks);
1249 	if (bp)
1250 		return;
1251 
1252 	numbanks = this_cpu_read(mce_num_banks);
1253 	bp = kcalloc(numbanks, sizeof(*bp), GFP_KERNEL);
1254 	if (!bp)
1255 		return;
1256 
1257 	for (bank = 0; bank < numbanks; ++bank) {
1258 		if (!(this_cpu_read(bank_map) & BIT_ULL(bank)))
1259 			continue;
1260 		if (threshold_create_bank(bp, cpu, bank)) {
1261 			__threshold_remove_device(bp);
1262 			return;
1263 		}
1264 	}
1265 	this_cpu_write(threshold_banks, bp);
1266 
1267 	if (thresholding_irq_en)
1268 		mce_threshold_vector = amd_threshold_interrupt;
1269 	return;
1270 }
1271