xref: /openbmc/linux/arch/x86/kernel/cpu/mce/amd.c (revision 55b7acbd15b15e75c6df468c72177a6b32e648cf)
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 struct smca_bank_name {
91 	const char *name;	/* Short name for sysfs */
92 	const char *long_name;	/* Long name for pretty-printing */
93 };
94 
95 static struct smca_bank_name smca_names[] = {
96 	[SMCA_LS ... SMCA_LS_V2]	= { "load_store",	"Load Store Unit" },
97 	[SMCA_IF]			= { "insn_fetch",	"Instruction Fetch Unit" },
98 	[SMCA_L2_CACHE]			= { "l2_cache",		"L2 Cache" },
99 	[SMCA_DE]			= { "decode_unit",	"Decode Unit" },
100 	[SMCA_RESERVED]			= { "reserved",		"Reserved" },
101 	[SMCA_EX]			= { "execution_unit",	"Execution Unit" },
102 	[SMCA_FP]			= { "floating_point",	"Floating Point Unit" },
103 	[SMCA_L3_CACHE]			= { "l3_cache",		"L3 Cache" },
104 	[SMCA_CS ... SMCA_CS_V2]	= { "coherent_slave",	"Coherent Slave" },
105 	[SMCA_PIE]			= { "pie",		"Power, Interrupts, etc." },
106 
107 	/* UMC v2 is separate because both of them can exist in a single system. */
108 	[SMCA_UMC]			= { "umc",		"Unified Memory Controller" },
109 	[SMCA_UMC_V2]			= { "umc_v2",		"Unified Memory Controller v2" },
110 	[SMCA_PB]			= { "param_block",	"Parameter Block" },
111 	[SMCA_PSP ... SMCA_PSP_V2]	= { "psp",		"Platform Security Processor" },
112 	[SMCA_SMU ... SMCA_SMU_V2]	= { "smu",		"System Management Unit" },
113 	[SMCA_MP5]			= { "mp5",		"Microprocessor 5 Unit" },
114 	[SMCA_MPDMA]			= { "mpdma",		"MPDMA Unit" },
115 	[SMCA_NBIO]			= { "nbio",		"Northbridge IO Unit" },
116 	[SMCA_PCIE ... SMCA_PCIE_V2]	= { "pcie",		"PCI Express Unit" },
117 	[SMCA_XGMI_PCS]			= { "xgmi_pcs",		"Ext Global Memory Interconnect PCS Unit" },
118 	[SMCA_NBIF]			= { "nbif",		"NBIF Unit" },
119 	[SMCA_SHUB]			= { "shub",		"System Hub Unit" },
120 	[SMCA_SATA]			= { "sata",		"SATA Unit" },
121 	[SMCA_USB]			= { "usb",		"USB Unit" },
122 	[SMCA_GMI_PCS]			= { "gmi_pcs",		"Global Memory Interconnect PCS Unit" },
123 	[SMCA_XGMI_PHY]			= { "xgmi_phy",		"Ext Global Memory Interconnect PHY Unit" },
124 	[SMCA_WAFL_PHY]			= { "wafl_phy",		"WAFL PHY Unit" },
125 	[SMCA_GMI_PHY]			= { "gmi_phy",		"Global Memory Interconnect PHY Unit" },
126 };
127 
smca_get_name(enum smca_bank_types t)128 static const char *smca_get_name(enum smca_bank_types t)
129 {
130 	if (t >= N_SMCA_BANK_TYPES)
131 		return NULL;
132 
133 	return smca_names[t].name;
134 }
135 
smca_get_long_name(enum smca_bank_types t)136 const char *smca_get_long_name(enum smca_bank_types t)
137 {
138 	if (t >= N_SMCA_BANK_TYPES)
139 		return NULL;
140 
141 	return smca_names[t].long_name;
142 }
143 EXPORT_SYMBOL_GPL(smca_get_long_name);
144 
smca_get_bank_type(unsigned int cpu,unsigned int bank)145 enum smca_bank_types smca_get_bank_type(unsigned int cpu, unsigned int bank)
146 {
147 	struct smca_bank *b;
148 
149 	if (bank >= MAX_NR_BANKS)
150 		return N_SMCA_BANK_TYPES;
151 
152 	b = &per_cpu(smca_banks, cpu)[bank];
153 	if (!b->hwid)
154 		return N_SMCA_BANK_TYPES;
155 
156 	return b->hwid->bank_type;
157 }
158 EXPORT_SYMBOL_GPL(smca_get_bank_type);
159 
160 static const struct smca_hwid smca_hwid_mcatypes[] = {
161 	/* { bank_type, hwid_mcatype } */
162 
163 	/* Reserved type */
164 	{ SMCA_RESERVED, HWID_MCATYPE(0x00, 0x0)	},
165 
166 	/* ZN Core (HWID=0xB0) MCA types */
167 	{ SMCA_LS,	 HWID_MCATYPE(0xB0, 0x0)	},
168 	{ SMCA_LS_V2,	 HWID_MCATYPE(0xB0, 0x10)	},
169 	{ SMCA_IF,	 HWID_MCATYPE(0xB0, 0x1)	},
170 	{ SMCA_L2_CACHE, HWID_MCATYPE(0xB0, 0x2)	},
171 	{ SMCA_DE,	 HWID_MCATYPE(0xB0, 0x3)	},
172 	/* HWID 0xB0 MCATYPE 0x4 is Reserved */
173 	{ SMCA_EX,	 HWID_MCATYPE(0xB0, 0x5)	},
174 	{ SMCA_FP,	 HWID_MCATYPE(0xB0, 0x6)	},
175 	{ SMCA_L3_CACHE, HWID_MCATYPE(0xB0, 0x7)	},
176 
177 	/* Data Fabric MCA types */
178 	{ SMCA_CS,	 HWID_MCATYPE(0x2E, 0x0)	},
179 	{ SMCA_PIE,	 HWID_MCATYPE(0x2E, 0x1)	},
180 	{ SMCA_CS_V2,	 HWID_MCATYPE(0x2E, 0x2)	},
181 
182 	/* Unified Memory Controller MCA type */
183 	{ SMCA_UMC,	 HWID_MCATYPE(0x96, 0x0)	},
184 	{ SMCA_UMC_V2,	 HWID_MCATYPE(0x96, 0x1)	},
185 
186 	/* Parameter Block MCA type */
187 	{ SMCA_PB,	 HWID_MCATYPE(0x05, 0x0)	},
188 
189 	/* Platform Security Processor MCA type */
190 	{ SMCA_PSP,	 HWID_MCATYPE(0xFF, 0x0)	},
191 	{ SMCA_PSP_V2,	 HWID_MCATYPE(0xFF, 0x1)	},
192 
193 	/* System Management Unit MCA type */
194 	{ SMCA_SMU,	 HWID_MCATYPE(0x01, 0x0)	},
195 	{ SMCA_SMU_V2,	 HWID_MCATYPE(0x01, 0x1)	},
196 
197 	/* Microprocessor 5 Unit MCA type */
198 	{ SMCA_MP5,	 HWID_MCATYPE(0x01, 0x2)	},
199 
200 	/* MPDMA MCA type */
201 	{ SMCA_MPDMA,	 HWID_MCATYPE(0x01, 0x3)	},
202 
203 	/* Northbridge IO Unit MCA type */
204 	{ SMCA_NBIO,	 HWID_MCATYPE(0x18, 0x0)	},
205 
206 	/* PCI Express Unit MCA type */
207 	{ SMCA_PCIE,	 HWID_MCATYPE(0x46, 0x0)	},
208 	{ SMCA_PCIE_V2,	 HWID_MCATYPE(0x46, 0x1)	},
209 
210 	{ SMCA_XGMI_PCS, HWID_MCATYPE(0x50, 0x0)	},
211 	{ SMCA_NBIF,	 HWID_MCATYPE(0x6C, 0x0)	},
212 	{ SMCA_SHUB,	 HWID_MCATYPE(0x80, 0x0)	},
213 	{ SMCA_SATA,	 HWID_MCATYPE(0xA8, 0x0)	},
214 	{ SMCA_USB,	 HWID_MCATYPE(0xAA, 0x0)	},
215 	{ SMCA_GMI_PCS,  HWID_MCATYPE(0x241, 0x0)	},
216 	{ SMCA_XGMI_PHY, HWID_MCATYPE(0x259, 0x0)	},
217 	{ SMCA_WAFL_PHY, HWID_MCATYPE(0x267, 0x0)	},
218 	{ SMCA_GMI_PHY,	 HWID_MCATYPE(0x269, 0x0)	},
219 };
220 
221 /*
222  * In SMCA enabled processors, we can have multiple banks for a given IP type.
223  * So to define a unique name for each bank, we use a temp c-string to append
224  * the MCA_IPID[InstanceId] to type's name in get_name().
225  *
226  * InstanceId is 32 bits which is 8 characters. Make sure MAX_MCATYPE_NAME_LEN
227  * is greater than 8 plus 1 (for underscore) plus length of longest type name.
228  */
229 #define MAX_MCATYPE_NAME_LEN	30
230 static char buf_mcatype[MAX_MCATYPE_NAME_LEN];
231 
232 static DEFINE_PER_CPU(struct threshold_bank **, threshold_banks);
233 
234 /*
235  * A list of the banks enabled on each logical CPU. Controls which respective
236  * descriptors to initialize later in mce_threshold_create_device().
237  */
238 static DEFINE_PER_CPU(u64, bank_map);
239 
240 /* Map of banks that have more than MCA_MISC0 available. */
241 static DEFINE_PER_CPU(u64, smca_misc_banks_map);
242 
243 static void amd_threshold_interrupt(void);
244 static void amd_deferred_error_interrupt(void);
245 
default_deferred_error_interrupt(void)246 static void default_deferred_error_interrupt(void)
247 {
248 	pr_err("Unexpected deferred interrupt at vector %x\n", DEFERRED_ERROR_VECTOR);
249 }
250 void (*deferred_error_int_vector)(void) = default_deferred_error_interrupt;
251 
smca_set_misc_banks_map(unsigned int bank,unsigned int cpu)252 static void smca_set_misc_banks_map(unsigned int bank, unsigned int cpu)
253 {
254 	u32 low, high;
255 
256 	/*
257 	 * For SMCA enabled processors, BLKPTR field of the first MISC register
258 	 * (MCx_MISC0) indicates presence of additional MISC regs set (MISC1-4).
259 	 */
260 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_CONFIG(bank), &low, &high))
261 		return;
262 
263 	if (!(low & MCI_CONFIG_MCAX))
264 		return;
265 
266 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_MISC(bank), &low, &high))
267 		return;
268 
269 	if (low & MASK_BLKPTR_LO)
270 		per_cpu(smca_misc_banks_map, cpu) |= BIT_ULL(bank);
271 
272 }
273 
smca_configure(unsigned int bank,unsigned int cpu)274 static void smca_configure(unsigned int bank, unsigned int cpu)
275 {
276 	u8 *bank_counts = this_cpu_ptr(smca_bank_counts);
277 	const struct smca_hwid *s_hwid;
278 	unsigned int i, hwid_mcatype;
279 	u32 high, low;
280 	u32 smca_config = MSR_AMD64_SMCA_MCx_CONFIG(bank);
281 
282 	/* Set appropriate bits in MCA_CONFIG */
283 	if (!rdmsr_safe(smca_config, &low, &high)) {
284 		/*
285 		 * OS is required to set the MCAX bit to acknowledge that it is
286 		 * now using the new MSR ranges and new registers under each
287 		 * bank. It also means that the OS will configure deferred
288 		 * errors in the new MCx_CONFIG register. If the bit is not set,
289 		 * uncorrectable errors will cause a system panic.
290 		 *
291 		 * MCA_CONFIG[MCAX] is bit 32 (0 in the high portion of the MSR.)
292 		 */
293 		high |= BIT(0);
294 
295 		/*
296 		 * SMCA sets the Deferred Error Interrupt type per bank.
297 		 *
298 		 * MCA_CONFIG[DeferredIntTypeSupported] is bit 5, and tells us
299 		 * if the DeferredIntType bit field is available.
300 		 *
301 		 * MCA_CONFIG[DeferredIntType] is bits [38:37] ([6:5] in the
302 		 * high portion of the MSR). OS should set this to 0x1 to enable
303 		 * APIC based interrupt. First, check that no interrupt has been
304 		 * set.
305 		 */
306 		if ((low & BIT(5)) && !((high >> 5) & 0x3))
307 			high |= BIT(5);
308 
309 		this_cpu_ptr(mce_banks_array)[bank].lsb_in_status = !!(low & BIT(8));
310 
311 		wrmsr(smca_config, low, high);
312 	}
313 
314 	smca_set_misc_banks_map(bank, cpu);
315 
316 	if (rdmsr_safe(MSR_AMD64_SMCA_MCx_IPID(bank), &low, &high)) {
317 		pr_warn("Failed to read MCA_IPID for bank %d\n", bank);
318 		return;
319 	}
320 
321 	hwid_mcatype = HWID_MCATYPE(high & MCI_IPID_HWID,
322 				    (high & MCI_IPID_MCATYPE) >> 16);
323 
324 	for (i = 0; i < ARRAY_SIZE(smca_hwid_mcatypes); i++) {
325 		s_hwid = &smca_hwid_mcatypes[i];
326 
327 		if (hwid_mcatype == s_hwid->hwid_mcatype) {
328 			this_cpu_ptr(smca_banks)[bank].hwid = s_hwid;
329 			this_cpu_ptr(smca_banks)[bank].id = low;
330 			this_cpu_ptr(smca_banks)[bank].sysfs_id = bank_counts[s_hwid->bank_type]++;
331 			break;
332 		}
333 	}
334 }
335 
336 struct thresh_restart {
337 	struct threshold_block	*b;
338 	int			set_lvt_off;
339 	int			lvt_off;
340 	u16			old_limit;
341 };
342 
is_shared_bank(int bank)343 static inline bool is_shared_bank(int bank)
344 {
345 	/*
346 	 * Scalable MCA provides for only one core to have access to the MSRs of
347 	 * a shared bank.
348 	 */
349 	if (mce_flags.smca)
350 		return false;
351 
352 	/* Bank 4 is for northbridge reporting and is thus shared */
353 	return (bank == 4);
354 }
355 
bank4_names(const struct threshold_block * b)356 static const char *bank4_names(const struct threshold_block *b)
357 {
358 	switch (b->address) {
359 	/* MSR4_MISC0 */
360 	case 0x00000413:
361 		return "dram";
362 
363 	case 0xc0000408:
364 		return "ht_links";
365 
366 	case 0xc0000409:
367 		return "l3_cache";
368 
369 	default:
370 		WARN(1, "Funny MSR: 0x%08x\n", b->address);
371 		return "";
372 	}
373 };
374 
375 
lvt_interrupt_supported(unsigned int bank,u32 msr_high_bits)376 static bool lvt_interrupt_supported(unsigned int bank, u32 msr_high_bits)
377 {
378 	/*
379 	 * bank 4 supports APIC LVT interrupts implicitly since forever.
380 	 */
381 	if (bank == 4)
382 		return true;
383 
384 	/*
385 	 * IntP: interrupt present; if this bit is set, the thresholding
386 	 * bank can generate APIC LVT interrupts
387 	 */
388 	return msr_high_bits & BIT(28);
389 }
390 
lvt_off_valid(struct threshold_block * b,int apic,u32 lo,u32 hi)391 static int lvt_off_valid(struct threshold_block *b, int apic, u32 lo, u32 hi)
392 {
393 	int msr = (hi & MASK_LVTOFF_HI) >> 20;
394 
395 	if (apic < 0) {
396 		pr_err(FW_BUG "cpu %d, failed to setup threshold interrupt "
397 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n", b->cpu,
398 		       b->bank, b->block, b->address, hi, lo);
399 		return 0;
400 	}
401 
402 	if (apic != msr) {
403 		/*
404 		 * On SMCA CPUs, LVT offset is programmed at a different MSR, and
405 		 * the BIOS provides the value. The original field where LVT offset
406 		 * was set is reserved. Return early here:
407 		 */
408 		if (mce_flags.smca)
409 			return 0;
410 
411 		pr_err(FW_BUG "cpu %d, invalid threshold interrupt offset %d "
412 		       "for bank %d, block %d (MSR%08X=0x%x%08x)\n",
413 		       b->cpu, apic, b->bank, b->block, b->address, hi, lo);
414 		return 0;
415 	}
416 
417 	return 1;
418 };
419 
420 /* Reprogram MCx_MISC MSR behind this threshold bank. */
threshold_restart_bank(void * _tr)421 static void threshold_restart_bank(void *_tr)
422 {
423 	struct thresh_restart *tr = _tr;
424 	u32 hi, lo;
425 
426 	/* sysfs write might race against an offline operation */
427 	if (!this_cpu_read(threshold_banks) && !tr->set_lvt_off)
428 		return;
429 
430 	rdmsr(tr->b->address, lo, hi);
431 
432 	/*
433 	 * Reset error count and overflow bit.
434 	 * This is done during init or after handling an interrupt.
435 	 */
436 	if (hi & MASK_OVERFLOW_HI || tr->set_lvt_off) {
437 		hi &= ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI);
438 		hi |= THRESHOLD_MAX - tr->b->threshold_limit;
439 	} else if (tr->old_limit) {	/* change limit w/o reset */
440 		int new_count = (hi & THRESHOLD_MAX) +
441 		    (tr->old_limit - tr->b->threshold_limit);
442 
443 		hi = (hi & ~MASK_ERR_COUNT_HI) |
444 		    (new_count & THRESHOLD_MAX);
445 	}
446 
447 	/* clear IntType */
448 	hi &= ~MASK_INT_TYPE_HI;
449 
450 	if (!tr->b->interrupt_capable)
451 		goto done;
452 
453 	if (tr->set_lvt_off) {
454 		if (lvt_off_valid(tr->b, tr->lvt_off, lo, hi)) {
455 			/* set new lvt offset */
456 			hi &= ~MASK_LVTOFF_HI;
457 			hi |= tr->lvt_off << 20;
458 		}
459 	}
460 
461 	if (tr->b->interrupt_enable)
462 		hi |= INT_TYPE_APIC;
463 
464  done:
465 
466 	hi |= MASK_COUNT_EN_HI;
467 	wrmsr(tr->b->address, lo, hi);
468 }
469 
mce_threshold_block_init(struct threshold_block * b,int offset)470 static void mce_threshold_block_init(struct threshold_block *b, int offset)
471 {
472 	struct thresh_restart tr = {
473 		.b			= b,
474 		.set_lvt_off		= 1,
475 		.lvt_off		= offset,
476 	};
477 
478 	b->threshold_limit		= THRESHOLD_MAX;
479 	threshold_restart_bank(&tr);
480 };
481 
setup_APIC_mce_threshold(int reserved,int new)482 static int setup_APIC_mce_threshold(int reserved, int new)
483 {
484 	if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
485 					      APIC_EILVT_MSG_FIX, 0))
486 		return new;
487 
488 	return reserved;
489 }
490 
setup_APIC_deferred_error(int reserved,int new)491 static int setup_APIC_deferred_error(int reserved, int new)
492 {
493 	if (reserved < 0 && !setup_APIC_eilvt(new, DEFERRED_ERROR_VECTOR,
494 					      APIC_EILVT_MSG_FIX, 0))
495 		return new;
496 
497 	return reserved;
498 }
499 
deferred_error_interrupt_enable(struct cpuinfo_x86 * c)500 static void deferred_error_interrupt_enable(struct cpuinfo_x86 *c)
501 {
502 	u32 low = 0, high = 0;
503 	int def_offset = -1, def_new;
504 
505 	if (rdmsr_safe(MSR_CU_DEF_ERR, &low, &high))
506 		return;
507 
508 	def_new = (low & MASK_DEF_LVTOFF) >> 4;
509 	if (!(low & MASK_DEF_LVTOFF)) {
510 		pr_err(FW_BUG "Your BIOS is not setting up LVT offset 0x2 for deferred error IRQs correctly.\n");
511 		def_new = DEF_LVT_OFF;
512 		low = (low & ~MASK_DEF_LVTOFF) | (DEF_LVT_OFF << 4);
513 	}
514 
515 	def_offset = setup_APIC_deferred_error(def_offset, def_new);
516 	if ((def_offset == def_new) &&
517 	    (deferred_error_int_vector != amd_deferred_error_interrupt))
518 		deferred_error_int_vector = amd_deferred_error_interrupt;
519 
520 	if (!mce_flags.smca)
521 		low = (low & ~MASK_DEF_INT_TYPE) | DEF_INT_TYPE_APIC;
522 
523 	wrmsr(MSR_CU_DEF_ERR, low, high);
524 }
525 
smca_get_block_address(unsigned int bank,unsigned int block,unsigned int cpu)526 static u32 smca_get_block_address(unsigned int bank, unsigned int block,
527 				  unsigned int cpu)
528 {
529 	if (!block)
530 		return MSR_AMD64_SMCA_MCx_MISC(bank);
531 
532 	if (!(per_cpu(smca_misc_banks_map, cpu) & BIT_ULL(bank)))
533 		return 0;
534 
535 	return MSR_AMD64_SMCA_MCx_MISCy(bank, block - 1);
536 }
537 
get_block_address(u32 current_addr,u32 low,u32 high,unsigned int bank,unsigned int block,unsigned int cpu)538 static u32 get_block_address(u32 current_addr, u32 low, u32 high,
539 			     unsigned int bank, unsigned int block,
540 			     unsigned int cpu)
541 {
542 	u32 addr = 0, offset = 0;
543 
544 	if ((bank >= per_cpu(mce_num_banks, cpu)) || (block >= NR_BLOCKS))
545 		return addr;
546 
547 	if (mce_flags.smca)
548 		return smca_get_block_address(bank, block, cpu);
549 
550 	/* Fall back to method we used for older processors: */
551 	switch (block) {
552 	case 0:
553 		addr = mca_msr_reg(bank, MCA_MISC);
554 		break;
555 	case 1:
556 		offset = ((low & MASK_BLKPTR_LO) >> 21);
557 		if (offset)
558 			addr = MCG_XBLK_ADDR + offset;
559 		break;
560 	default:
561 		addr = ++current_addr;
562 	}
563 	return addr;
564 }
565 
566 static int
prepare_threshold_block(unsigned int bank,unsigned int block,u32 addr,int offset,u32 misc_high)567 prepare_threshold_block(unsigned int bank, unsigned int block, u32 addr,
568 			int offset, u32 misc_high)
569 {
570 	unsigned int cpu = smp_processor_id();
571 	u32 smca_low, smca_high;
572 	struct threshold_block b;
573 	int new;
574 
575 	if (!block)
576 		per_cpu(bank_map, cpu) |= BIT_ULL(bank);
577 
578 	memset(&b, 0, sizeof(b));
579 	b.cpu			= cpu;
580 	b.bank			= bank;
581 	b.block			= block;
582 	b.address		= addr;
583 	b.interrupt_capable	= lvt_interrupt_supported(bank, misc_high);
584 
585 	if (!b.interrupt_capable)
586 		goto done;
587 
588 	b.interrupt_enable = 1;
589 
590 	if (!mce_flags.smca) {
591 		new = (misc_high & MASK_LVTOFF_HI) >> 20;
592 		goto set_offset;
593 	}
594 
595 	/* Gather LVT offset for thresholding: */
596 	if (rdmsr_safe(MSR_CU_DEF_ERR, &smca_low, &smca_high))
597 		goto out;
598 
599 	new = (smca_low & SMCA_THR_LVT_OFF) >> 12;
600 
601 set_offset:
602 	offset = setup_APIC_mce_threshold(offset, new);
603 	if (offset == new)
604 		thresholding_irq_en = true;
605 
606 done:
607 	mce_threshold_block_init(&b, offset);
608 
609 out:
610 	return offset;
611 }
612 
amd_filter_mce(struct mce * m)613 bool amd_filter_mce(struct mce *m)
614 {
615 	enum smca_bank_types bank_type = smca_get_bank_type(m->extcpu, m->bank);
616 	struct cpuinfo_x86 *c = &boot_cpu_data;
617 
618 	/* See Family 17h Models 10h-2Fh Erratum #1114. */
619 	if (c->x86 == 0x17 &&
620 	    c->x86_model >= 0x10 && c->x86_model <= 0x2F &&
621 	    bank_type == SMCA_IF && XEC(m->status, 0x3f) == 10)
622 		return true;
623 
624 	/* NB GART TLB error reporting is disabled by default. */
625 	if (c->x86 < 0x17) {
626 		if (m->bank == 4 && XEC(m->status, 0x1f) == 0x5)
627 			return true;
628 	}
629 
630 	return false;
631 }
632 
633 /*
634  * Turn off thresholding banks for the following conditions:
635  * - MC4_MISC thresholding is not supported on Family 0x15.
636  * - Prevent possible spurious interrupts from the IF bank on Family 0x17
637  *   Models 0x10-0x2F due to Erratum #1114.
638  */
disable_err_thresholding(struct cpuinfo_x86 * c,unsigned int bank)639 static void disable_err_thresholding(struct cpuinfo_x86 *c, unsigned int bank)
640 {
641 	int i, num_msrs;
642 	u64 hwcr;
643 	bool need_toggle;
644 	u32 msrs[NR_BLOCKS];
645 
646 	if (c->x86 == 0x15 && bank == 4) {
647 		msrs[0] = 0x00000413; /* MC4_MISC0 */
648 		msrs[1] = 0xc0000408; /* MC4_MISC1 */
649 		num_msrs = 2;
650 	} else if (c->x86 == 0x17 &&
651 		   (c->x86_model >= 0x10 && c->x86_model <= 0x2F)) {
652 
653 		if (smca_get_bank_type(smp_processor_id(), bank) != SMCA_IF)
654 			return;
655 
656 		msrs[0] = MSR_AMD64_SMCA_MCx_MISC(bank);
657 		num_msrs = 1;
658 	} else {
659 		return;
660 	}
661 
662 	rdmsrl(MSR_K7_HWCR, hwcr);
663 
664 	/* McStatusWrEn has to be set */
665 	need_toggle = !(hwcr & BIT(18));
666 	if (need_toggle)
667 		wrmsrl(MSR_K7_HWCR, hwcr | BIT(18));
668 
669 	/* Clear CntP bit safely */
670 	for (i = 0; i < num_msrs; i++)
671 		msr_clear_bit(msrs[i], 62);
672 
673 	/* restore old settings */
674 	if (need_toggle)
675 		wrmsrl(MSR_K7_HWCR, hwcr);
676 }
677 
678 /* cpu init entry point, called from mce.c with preempt off */
mce_amd_feature_init(struct cpuinfo_x86 * c)679 void mce_amd_feature_init(struct cpuinfo_x86 *c)
680 {
681 	unsigned int bank, block, cpu = smp_processor_id();
682 	u32 low = 0, high = 0, address = 0;
683 	int offset = -1;
684 
685 
686 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
687 		if (mce_flags.smca)
688 			smca_configure(bank, cpu);
689 
690 		disable_err_thresholding(c, bank);
691 
692 		for (block = 0; block < NR_BLOCKS; ++block) {
693 			address = get_block_address(address, low, high, bank, block, cpu);
694 			if (!address)
695 				break;
696 
697 			if (rdmsr_safe(address, &low, &high))
698 				break;
699 
700 			if (!(high & MASK_VALID_HI))
701 				continue;
702 
703 			if (!(high & MASK_CNTP_HI)  ||
704 			     (high & MASK_LOCKED_HI))
705 				continue;
706 
707 			offset = prepare_threshold_block(bank, block, address, offset, high);
708 		}
709 	}
710 
711 	if (mce_flags.succor)
712 		deferred_error_interrupt_enable(c);
713 }
714 
amd_mce_is_memory_error(struct mce * m)715 bool amd_mce_is_memory_error(struct mce *m)
716 {
717 	enum smca_bank_types bank_type;
718 	/* ErrCodeExt[20:16] */
719 	u8 xec = (m->status >> 16) & 0x1f;
720 
721 	bank_type = smca_get_bank_type(m->extcpu, m->bank);
722 	if (mce_flags.smca)
723 		return (bank_type == SMCA_UMC || bank_type == SMCA_UMC_V2) && xec == 0x0;
724 
725 	return m->bank == 4 && xec == 0x8;
726 }
727 
__log_error(unsigned int bank,u64 status,u64 addr,u64 misc)728 static void __log_error(unsigned int bank, u64 status, u64 addr, u64 misc)
729 {
730 	struct mce m;
731 
732 	mce_setup(&m);
733 
734 	m.status = status;
735 	m.misc   = misc;
736 	m.bank   = bank;
737 	m.tsc	 = rdtsc();
738 
739 	if (m.status & MCI_STATUS_ADDRV) {
740 		m.addr = addr;
741 
742 		smca_extract_err_addr(&m);
743 	}
744 
745 	if (mce_flags.smca) {
746 		rdmsrl(MSR_AMD64_SMCA_MCx_IPID(bank), m.ipid);
747 
748 		if (m.status & MCI_STATUS_SYNDV)
749 			rdmsrl(MSR_AMD64_SMCA_MCx_SYND(bank), m.synd);
750 	}
751 
752 	mce_log(&m);
753 }
754 
DEFINE_IDTENTRY_SYSVEC(sysvec_deferred_error)755 DEFINE_IDTENTRY_SYSVEC(sysvec_deferred_error)
756 {
757 	trace_deferred_error_apic_entry(DEFERRED_ERROR_VECTOR);
758 	inc_irq_stat(irq_deferred_error_count);
759 	deferred_error_int_vector();
760 	trace_deferred_error_apic_exit(DEFERRED_ERROR_VECTOR);
761 	apic_eoi();
762 }
763 
764 /*
765  * Returns true if the logged error is deferred. False, otherwise.
766  */
767 static inline bool
_log_error_bank(unsigned int bank,u32 msr_stat,u32 msr_addr,u64 misc)768 _log_error_bank(unsigned int bank, u32 msr_stat, u32 msr_addr, u64 misc)
769 {
770 	u64 status, addr = 0;
771 
772 	rdmsrl(msr_stat, status);
773 	if (!(status & MCI_STATUS_VAL))
774 		return false;
775 
776 	if (status & MCI_STATUS_ADDRV)
777 		rdmsrl(msr_addr, addr);
778 
779 	__log_error(bank, status, addr, misc);
780 
781 	wrmsrl(msr_stat, 0);
782 
783 	return status & MCI_STATUS_DEFERRED;
784 }
785 
_log_error_deferred(unsigned int bank,u32 misc)786 static bool _log_error_deferred(unsigned int bank, u32 misc)
787 {
788 	if (!_log_error_bank(bank, mca_msr_reg(bank, MCA_STATUS),
789 			     mca_msr_reg(bank, MCA_ADDR), misc))
790 		return false;
791 
792 	/*
793 	 * Non-SMCA systems don't have MCA_DESTAT/MCA_DEADDR registers.
794 	 * Return true here to avoid accessing these registers.
795 	 */
796 	if (!mce_flags.smca)
797 		return true;
798 
799 	/* Clear MCA_DESTAT if the deferred error was logged from MCA_STATUS. */
800 	wrmsrl(MSR_AMD64_SMCA_MCx_DESTAT(bank), 0);
801 	return true;
802 }
803 
804 /*
805  * We have three scenarios for checking for Deferred errors:
806  *
807  * 1) Non-SMCA systems check MCA_STATUS and log error if found.
808  * 2) SMCA systems check MCA_STATUS. If error is found then log it and also
809  *    clear MCA_DESTAT.
810  * 3) SMCA systems check MCA_DESTAT, if error was not found in MCA_STATUS, and
811  *    log it.
812  */
log_error_deferred(unsigned int bank)813 static void log_error_deferred(unsigned int bank)
814 {
815 	if (_log_error_deferred(bank, 0))
816 		return;
817 
818 	/*
819 	 * Only deferred errors are logged in MCA_DE{STAT,ADDR} so just check
820 	 * for a valid error.
821 	 */
822 	_log_error_bank(bank, MSR_AMD64_SMCA_MCx_DESTAT(bank),
823 			      MSR_AMD64_SMCA_MCx_DEADDR(bank), 0);
824 }
825 
826 /* APIC interrupt handler for deferred errors */
amd_deferred_error_interrupt(void)827 static void amd_deferred_error_interrupt(void)
828 {
829 	unsigned int bank;
830 
831 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank)
832 		log_error_deferred(bank);
833 }
834 
log_error_thresholding(unsigned int bank,u64 misc)835 static void log_error_thresholding(unsigned int bank, u64 misc)
836 {
837 	_log_error_deferred(bank, misc);
838 }
839 
log_and_reset_block(struct threshold_block * block)840 static void log_and_reset_block(struct threshold_block *block)
841 {
842 	struct thresh_restart tr;
843 	u32 low = 0, high = 0;
844 
845 	if (!block)
846 		return;
847 
848 	if (rdmsr_safe(block->address, &low, &high))
849 		return;
850 
851 	if (!(high & MASK_OVERFLOW_HI))
852 		return;
853 
854 	/* Log the MCE which caused the threshold event. */
855 	log_error_thresholding(block->bank, ((u64)high << 32) | low);
856 
857 	/* Reset threshold block after logging error. */
858 	memset(&tr, 0, sizeof(tr));
859 	tr.b = block;
860 	threshold_restart_bank(&tr);
861 }
862 
863 /*
864  * Threshold interrupt handler will service THRESHOLD_APIC_VECTOR. The interrupt
865  * goes off when error_count reaches threshold_limit.
866  */
amd_threshold_interrupt(void)867 static void amd_threshold_interrupt(void)
868 {
869 	struct threshold_block *first_block = NULL, *block = NULL, *tmp = NULL;
870 	struct threshold_bank **bp = this_cpu_read(threshold_banks);
871 	unsigned int bank, cpu = smp_processor_id();
872 
873 	/*
874 	 * Validate that the threshold bank has been initialized already. The
875 	 * handler is installed at boot time, but on a hotplug event the
876 	 * interrupt might fire before the data has been initialized.
877 	 */
878 	if (!bp)
879 		return;
880 
881 	for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
882 		if (!(per_cpu(bank_map, cpu) & BIT_ULL(bank)))
883 			continue;
884 
885 		first_block = bp[bank]->blocks;
886 		if (!first_block)
887 			continue;
888 
889 		/*
890 		 * The first block is also the head of the list. Check it first
891 		 * before iterating over the rest.
892 		 */
893 		log_and_reset_block(first_block);
894 		list_for_each_entry_safe(block, tmp, &first_block->miscj, miscj)
895 			log_and_reset_block(block);
896 	}
897 }
898 
899 /*
900  * Sysfs Interface
901  */
902 
903 struct threshold_attr {
904 	struct attribute attr;
905 	ssize_t (*show) (struct threshold_block *, char *);
906 	ssize_t (*store) (struct threshold_block *, const char *, size_t count);
907 };
908 
909 #define SHOW_FIELDS(name)						\
910 static ssize_t show_ ## name(struct threshold_block *b, char *buf)	\
911 {									\
912 	return sprintf(buf, "%lu\n", (unsigned long) b->name);		\
913 }
914 SHOW_FIELDS(interrupt_enable)
SHOW_FIELDS(threshold_limit)915 SHOW_FIELDS(threshold_limit)
916 
917 static ssize_t
918 store_interrupt_enable(struct threshold_block *b, const char *buf, size_t size)
919 {
920 	struct thresh_restart tr;
921 	unsigned long new;
922 
923 	if (!b->interrupt_capable)
924 		return -EINVAL;
925 
926 	if (kstrtoul(buf, 0, &new) < 0)
927 		return -EINVAL;
928 
929 	b->interrupt_enable = !!new;
930 
931 	memset(&tr, 0, sizeof(tr));
932 	tr.b		= b;
933 
934 	if (smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1))
935 		return -ENODEV;
936 
937 	return size;
938 }
939 
940 static ssize_t
store_threshold_limit(struct threshold_block * b,const char * buf,size_t size)941 store_threshold_limit(struct threshold_block *b, const char *buf, size_t size)
942 {
943 	struct thresh_restart tr;
944 	unsigned long new;
945 
946 	if (kstrtoul(buf, 0, &new) < 0)
947 		return -EINVAL;
948 
949 	if (new > THRESHOLD_MAX)
950 		new = THRESHOLD_MAX;
951 	if (new < 1)
952 		new = 1;
953 
954 	memset(&tr, 0, sizeof(tr));
955 	tr.old_limit = b->threshold_limit;
956 	b->threshold_limit = new;
957 	tr.b = b;
958 
959 	if (smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1))
960 		return -ENODEV;
961 
962 	return size;
963 }
964 
show_error_count(struct threshold_block * b,char * buf)965 static ssize_t show_error_count(struct threshold_block *b, char *buf)
966 {
967 	u32 lo, hi;
968 
969 	/* CPU might be offline by now */
970 	if (rdmsr_on_cpu(b->cpu, b->address, &lo, &hi))
971 		return -ENODEV;
972 
973 	return sprintf(buf, "%u\n", ((hi & THRESHOLD_MAX) -
974 				     (THRESHOLD_MAX - b->threshold_limit)));
975 }
976 
977 static struct threshold_attr error_count = {
978 	.attr = {.name = __stringify(error_count), .mode = 0444 },
979 	.show = show_error_count,
980 };
981 
982 #define RW_ATTR(val)							\
983 static struct threshold_attr val = {					\
984 	.attr	= {.name = __stringify(val), .mode = 0644 },		\
985 	.show	= show_## val,						\
986 	.store	= store_## val,						\
987 };
988 
989 RW_ATTR(interrupt_enable);
990 RW_ATTR(threshold_limit);
991 
992 static struct attribute *default_attrs[] = {
993 	&threshold_limit.attr,
994 	&error_count.attr,
995 	NULL,	/* possibly interrupt_enable if supported, see below */
996 	NULL,
997 };
998 ATTRIBUTE_GROUPS(default);
999 
1000 #define to_block(k)	container_of(k, struct threshold_block, kobj)
1001 #define to_attr(a)	container_of(a, struct threshold_attr, attr)
1002 
show(struct kobject * kobj,struct attribute * attr,char * buf)1003 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
1004 {
1005 	struct threshold_block *b = to_block(kobj);
1006 	struct threshold_attr *a = to_attr(attr);
1007 	ssize_t ret;
1008 
1009 	ret = a->show ? a->show(b, buf) : -EIO;
1010 
1011 	return ret;
1012 }
1013 
store(struct kobject * kobj,struct attribute * attr,const char * buf,size_t count)1014 static ssize_t store(struct kobject *kobj, struct attribute *attr,
1015 		     const char *buf, size_t count)
1016 {
1017 	struct threshold_block *b = to_block(kobj);
1018 	struct threshold_attr *a = to_attr(attr);
1019 	ssize_t ret;
1020 
1021 	ret = a->store ? a->store(b, buf, count) : -EIO;
1022 
1023 	return ret;
1024 }
1025 
1026 static const struct sysfs_ops threshold_ops = {
1027 	.show			= show,
1028 	.store			= store,
1029 };
1030 
1031 static void threshold_block_release(struct kobject *kobj);
1032 
1033 static const struct kobj_type threshold_ktype = {
1034 	.sysfs_ops		= &threshold_ops,
1035 	.default_groups		= default_groups,
1036 	.release		= threshold_block_release,
1037 };
1038 
get_name(unsigned int cpu,unsigned int bank,struct threshold_block * b)1039 static const char *get_name(unsigned int cpu, unsigned int bank, struct threshold_block *b)
1040 {
1041 	enum smca_bank_types bank_type;
1042 
1043 	if (!mce_flags.smca) {
1044 		if (b && bank == 4)
1045 			return bank4_names(b);
1046 
1047 		return th_names[bank];
1048 	}
1049 
1050 	bank_type = smca_get_bank_type(cpu, bank);
1051 
1052 	if (b && (bank_type == SMCA_UMC || bank_type == SMCA_UMC_V2)) {
1053 		if (b->block < ARRAY_SIZE(smca_umc_block_names))
1054 			return smca_umc_block_names[b->block];
1055 	}
1056 
1057 	if (b && b->block) {
1058 		snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN, "th_block_%u", b->block);
1059 		return buf_mcatype;
1060 	}
1061 
1062 	if (bank_type >= N_SMCA_BANK_TYPES) {
1063 		snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN, "th_bank_%u", bank);
1064 		return buf_mcatype;
1065 	}
1066 
1067 	if (per_cpu(smca_bank_counts, cpu)[bank_type] == 1)
1068 		return smca_get_name(bank_type);
1069 
1070 	snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN,
1071 		 "%s_%u", smca_get_name(bank_type),
1072 			  per_cpu(smca_banks, cpu)[bank].sysfs_id);
1073 	return buf_mcatype;
1074 }
1075 
allocate_threshold_blocks(unsigned int cpu,struct threshold_bank * tb,unsigned int bank,unsigned int block,u32 address)1076 static int allocate_threshold_blocks(unsigned int cpu, struct threshold_bank *tb,
1077 				     unsigned int bank, unsigned int block,
1078 				     u32 address)
1079 {
1080 	struct threshold_block *b = NULL;
1081 	u32 low, high;
1082 	int err;
1083 
1084 	if ((bank >= this_cpu_read(mce_num_banks)) || (block >= NR_BLOCKS))
1085 		return 0;
1086 
1087 	if (rdmsr_safe(address, &low, &high))
1088 		return 0;
1089 
1090 	if (!(high & MASK_VALID_HI)) {
1091 		if (block)
1092 			goto recurse;
1093 		else
1094 			return 0;
1095 	}
1096 
1097 	if (!(high & MASK_CNTP_HI)  ||
1098 	     (high & MASK_LOCKED_HI))
1099 		goto recurse;
1100 
1101 	b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
1102 	if (!b)
1103 		return -ENOMEM;
1104 
1105 	b->block		= block;
1106 	b->bank			= bank;
1107 	b->cpu			= cpu;
1108 	b->address		= address;
1109 	b->interrupt_enable	= 0;
1110 	b->interrupt_capable	= lvt_interrupt_supported(bank, high);
1111 	b->threshold_limit	= THRESHOLD_MAX;
1112 
1113 	if (b->interrupt_capable) {
1114 		default_attrs[2] = &interrupt_enable.attr;
1115 		b->interrupt_enable = 1;
1116 	} else {
1117 		default_attrs[2] = NULL;
1118 	}
1119 
1120 	INIT_LIST_HEAD(&b->miscj);
1121 
1122 	/* This is safe as @tb is not visible yet */
1123 	if (tb->blocks)
1124 		list_add(&b->miscj, &tb->blocks->miscj);
1125 	else
1126 		tb->blocks = b;
1127 
1128 	err = kobject_init_and_add(&b->kobj, &threshold_ktype, tb->kobj, get_name(cpu, bank, b));
1129 	if (err)
1130 		goto out_free;
1131 recurse:
1132 	address = get_block_address(address, low, high, bank, ++block, cpu);
1133 	if (!address)
1134 		return 0;
1135 
1136 	err = allocate_threshold_blocks(cpu, tb, bank, block, address);
1137 	if (err)
1138 		goto out_free;
1139 
1140 	if (b)
1141 		kobject_uevent(&b->kobj, KOBJ_ADD);
1142 
1143 	return 0;
1144 
1145 out_free:
1146 	if (b) {
1147 		list_del(&b->miscj);
1148 		kobject_put(&b->kobj);
1149 	}
1150 	return err;
1151 }
1152 
__threshold_add_blocks(struct threshold_bank * b)1153 static int __threshold_add_blocks(struct threshold_bank *b)
1154 {
1155 	struct list_head *head = &b->blocks->miscj;
1156 	struct threshold_block *pos = NULL;
1157 	struct threshold_block *tmp = NULL;
1158 	int err = 0;
1159 
1160 	err = kobject_add(&b->blocks->kobj, b->kobj, b->blocks->kobj.name);
1161 	if (err)
1162 		return err;
1163 
1164 	list_for_each_entry_safe(pos, tmp, head, miscj) {
1165 
1166 		err = kobject_add(&pos->kobj, b->kobj, pos->kobj.name);
1167 		if (err) {
1168 			list_for_each_entry_safe_reverse(pos, tmp, head, miscj)
1169 				kobject_del(&pos->kobj);
1170 
1171 			return err;
1172 		}
1173 	}
1174 	return err;
1175 }
1176 
threshold_create_bank(struct threshold_bank ** bp,unsigned int cpu,unsigned int bank)1177 static int threshold_create_bank(struct threshold_bank **bp, unsigned int cpu,
1178 				 unsigned int bank)
1179 {
1180 	struct device *dev = this_cpu_read(mce_device);
1181 	struct amd_northbridge *nb = NULL;
1182 	struct threshold_bank *b = NULL;
1183 	const char *name = get_name(cpu, bank, NULL);
1184 	int err = 0;
1185 
1186 	if (!dev)
1187 		return -ENODEV;
1188 
1189 	if (is_shared_bank(bank)) {
1190 		nb = node_to_amd_nb(topology_die_id(cpu));
1191 
1192 		/* threshold descriptor already initialized on this node? */
1193 		if (nb && nb->bank4) {
1194 			/* yes, use it */
1195 			b = nb->bank4;
1196 			err = kobject_add(b->kobj, &dev->kobj, name);
1197 			if (err)
1198 				goto out;
1199 
1200 			bp[bank] = b;
1201 			refcount_inc(&b->cpus);
1202 
1203 			err = __threshold_add_blocks(b);
1204 
1205 			goto out;
1206 		}
1207 	}
1208 
1209 	b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
1210 	if (!b) {
1211 		err = -ENOMEM;
1212 		goto out;
1213 	}
1214 
1215 	/* Associate the bank with the per-CPU MCE device */
1216 	b->kobj = kobject_create_and_add(name, &dev->kobj);
1217 	if (!b->kobj) {
1218 		err = -EINVAL;
1219 		goto out_free;
1220 	}
1221 
1222 	if (is_shared_bank(bank)) {
1223 		b->shared = 1;
1224 		refcount_set(&b->cpus, 1);
1225 
1226 		/* nb is already initialized, see above */
1227 		if (nb) {
1228 			WARN_ON(nb->bank4);
1229 			nb->bank4 = b;
1230 		}
1231 	}
1232 
1233 	err = allocate_threshold_blocks(cpu, b, bank, 0, mca_msr_reg(bank, MCA_MISC));
1234 	if (err)
1235 		goto out_kobj;
1236 
1237 	bp[bank] = b;
1238 	return 0;
1239 
1240 out_kobj:
1241 	kobject_put(b->kobj);
1242 out_free:
1243 	kfree(b);
1244 out:
1245 	return err;
1246 }
1247 
threshold_block_release(struct kobject * kobj)1248 static void threshold_block_release(struct kobject *kobj)
1249 {
1250 	kfree(to_block(kobj));
1251 }
1252 
deallocate_threshold_blocks(struct threshold_bank * bank)1253 static void deallocate_threshold_blocks(struct threshold_bank *bank)
1254 {
1255 	struct threshold_block *pos, *tmp;
1256 
1257 	list_for_each_entry_safe(pos, tmp, &bank->blocks->miscj, miscj) {
1258 		list_del(&pos->miscj);
1259 		kobject_put(&pos->kobj);
1260 	}
1261 
1262 	kobject_put(&bank->blocks->kobj);
1263 }
1264 
__threshold_remove_blocks(struct threshold_bank * b)1265 static void __threshold_remove_blocks(struct threshold_bank *b)
1266 {
1267 	struct threshold_block *pos = NULL;
1268 	struct threshold_block *tmp = NULL;
1269 
1270 	kobject_put(b->kobj);
1271 
1272 	list_for_each_entry_safe(pos, tmp, &b->blocks->miscj, miscj)
1273 		kobject_put(b->kobj);
1274 }
1275 
threshold_remove_bank(struct threshold_bank * bank)1276 static void threshold_remove_bank(struct threshold_bank *bank)
1277 {
1278 	struct amd_northbridge *nb;
1279 
1280 	if (!bank->blocks)
1281 		goto out_free;
1282 
1283 	if (!bank->shared)
1284 		goto out_dealloc;
1285 
1286 	if (!refcount_dec_and_test(&bank->cpus)) {
1287 		__threshold_remove_blocks(bank);
1288 		return;
1289 	} else {
1290 		/*
1291 		 * The last CPU on this node using the shared bank is going
1292 		 * away, remove that bank now.
1293 		 */
1294 		nb = node_to_amd_nb(topology_die_id(smp_processor_id()));
1295 		nb->bank4 = NULL;
1296 	}
1297 
1298 out_dealloc:
1299 	deallocate_threshold_blocks(bank);
1300 
1301 out_free:
1302 	kobject_put(bank->kobj);
1303 	kfree(bank);
1304 }
1305 
__threshold_remove_device(struct threshold_bank ** bp)1306 static void __threshold_remove_device(struct threshold_bank **bp)
1307 {
1308 	unsigned int bank, numbanks = this_cpu_read(mce_num_banks);
1309 
1310 	for (bank = 0; bank < numbanks; bank++) {
1311 		if (!bp[bank])
1312 			continue;
1313 
1314 		threshold_remove_bank(bp[bank]);
1315 		bp[bank] = NULL;
1316 	}
1317 	kfree(bp);
1318 }
1319 
mce_threshold_remove_device(unsigned int cpu)1320 int mce_threshold_remove_device(unsigned int cpu)
1321 {
1322 	struct threshold_bank **bp = this_cpu_read(threshold_banks);
1323 
1324 	if (!bp)
1325 		return 0;
1326 
1327 	/*
1328 	 * Clear the pointer before cleaning up, so that the interrupt won't
1329 	 * touch anything of this.
1330 	 */
1331 	this_cpu_write(threshold_banks, NULL);
1332 
1333 	__threshold_remove_device(bp);
1334 	return 0;
1335 }
1336 
1337 /**
1338  * mce_threshold_create_device - Create the per-CPU MCE threshold device
1339  * @cpu:	The plugged in CPU
1340  *
1341  * Create directories and files for all valid threshold banks.
1342  *
1343  * This is invoked from the CPU hotplug callback which was installed in
1344  * mcheck_init_device(). The invocation happens in context of the hotplug
1345  * thread running on @cpu.  The callback is invoked on all CPUs which are
1346  * online when the callback is installed or during a real hotplug event.
1347  */
mce_threshold_create_device(unsigned int cpu)1348 int mce_threshold_create_device(unsigned int cpu)
1349 {
1350 	unsigned int numbanks, bank;
1351 	struct threshold_bank **bp;
1352 	int err;
1353 
1354 	if (!mce_flags.amd_threshold)
1355 		return 0;
1356 
1357 	bp = this_cpu_read(threshold_banks);
1358 	if (bp)
1359 		return 0;
1360 
1361 	numbanks = this_cpu_read(mce_num_banks);
1362 	bp = kcalloc(numbanks, sizeof(*bp), GFP_KERNEL);
1363 	if (!bp)
1364 		return -ENOMEM;
1365 
1366 	for (bank = 0; bank < numbanks; ++bank) {
1367 		if (!(this_cpu_read(bank_map) & BIT_ULL(bank)))
1368 			continue;
1369 		err = threshold_create_bank(bp, cpu, bank);
1370 		if (err) {
1371 			__threshold_remove_device(bp);
1372 			return err;
1373 		}
1374 	}
1375 	this_cpu_write(threshold_banks, bp);
1376 
1377 	if (thresholding_irq_en)
1378 		mce_threshold_vector = amd_threshold_interrupt;
1379 	return 0;
1380 }
1381