1 // SPDX-License-Identifier: GPL-2.0-only
3 * (c) 2005-2016 Advanced Micro Devices, Inc.
5 * Written by Jacob Shin - AMD, Inc.
6 * Maintained by: Borislav Petkov <bp@alien8.de>
8 * All MC4_MISCi registers are shared between cores on a node.
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>
23 #include <asm/amd_nb.h>
24 #include <asm/traps.h>
28 #include <asm/trace/irq_vectors.h>
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
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
55 /* Threshold LVT offset is at MSR0xC0000410[15:12] */
56 #define SMCA_THR_LVT_OFF 0xF000
58 static bool thresholding_irq_en;
60 static const char * const th_names[] = {
69 static const char * const smca_umc_block_names[] = {
74 struct smca_bank_name {
75 const char *name; /* Short name for sysfs */
76 const char *long_name; /* Long name for pretty-printing */
79 static struct smca_bank_name smca_names[] = {
80 [SMCA_LS] = { "load_store", "Load Store Unit" },
81 [SMCA_LS_V2] = { "load_store", "Load Store Unit" },
82 [SMCA_IF] = { "insn_fetch", "Instruction Fetch Unit" },
83 [SMCA_L2_CACHE] = { "l2_cache", "L2 Cache" },
84 [SMCA_DE] = { "decode_unit", "Decode Unit" },
85 [SMCA_RESERVED] = { "reserved", "Reserved" },
86 [SMCA_EX] = { "execution_unit", "Execution Unit" },
87 [SMCA_FP] = { "floating_point", "Floating Point Unit" },
88 [SMCA_L3_CACHE] = { "l3_cache", "L3 Cache" },
89 [SMCA_CS] = { "coherent_slave", "Coherent Slave" },
90 [SMCA_CS_V2] = { "coherent_slave", "Coherent Slave" },
91 [SMCA_PIE] = { "pie", "Power, Interrupts, etc." },
92 [SMCA_UMC] = { "umc", "Unified Memory Controller" },
93 [SMCA_PB] = { "param_block", "Parameter Block" },
94 [SMCA_PSP] = { "psp", "Platform Security Processor" },
95 [SMCA_PSP_V2] = { "psp", "Platform Security Processor" },
96 [SMCA_SMU] = { "smu", "System Management Unit" },
97 [SMCA_SMU_V2] = { "smu", "System Management Unit" },
98 [SMCA_MP5] = { "mp5", "Microprocessor 5 Unit" },
99 [SMCA_NBIO] = { "nbio", "Northbridge IO Unit" },
100 [SMCA_PCIE] = { "pcie", "PCI Express Unit" },
103 static const char *smca_get_name(enum smca_bank_types t)
105 if (t >= N_SMCA_BANK_TYPES)
108 return smca_names[t].name;
111 const char *smca_get_long_name(enum smca_bank_types t)
113 if (t >= N_SMCA_BANK_TYPES)
116 return smca_names[t].long_name;
118 EXPORT_SYMBOL_GPL(smca_get_long_name);
120 static enum smca_bank_types smca_get_bank_type(unsigned int bank)
124 if (bank >= MAX_NR_BANKS)
125 return N_SMCA_BANK_TYPES;
127 b = &smca_banks[bank];
129 return N_SMCA_BANK_TYPES;
131 return b->hwid->bank_type;
134 static struct smca_hwid smca_hwid_mcatypes[] = {
135 /* { bank_type, hwid_mcatype, xec_bitmap } */
138 { SMCA_RESERVED, HWID_MCATYPE(0x00, 0x0), 0x0 },
140 /* ZN Core (HWID=0xB0) MCA types */
141 { SMCA_LS, HWID_MCATYPE(0xB0, 0x0), 0x1FFFFF },
142 { SMCA_LS_V2, HWID_MCATYPE(0xB0, 0x10), 0xFFFFFF },
143 { SMCA_IF, HWID_MCATYPE(0xB0, 0x1), 0x3FFF },
144 { SMCA_L2_CACHE, HWID_MCATYPE(0xB0, 0x2), 0xF },
145 { SMCA_DE, HWID_MCATYPE(0xB0, 0x3), 0x1FF },
146 /* HWID 0xB0 MCATYPE 0x4 is Reserved */
147 { SMCA_EX, HWID_MCATYPE(0xB0, 0x5), 0xFFF },
148 { SMCA_FP, HWID_MCATYPE(0xB0, 0x6), 0x7F },
149 { SMCA_L3_CACHE, HWID_MCATYPE(0xB0, 0x7), 0xFF },
151 /* Data Fabric MCA types */
152 { SMCA_CS, HWID_MCATYPE(0x2E, 0x0), 0x1FF },
153 { SMCA_PIE, HWID_MCATYPE(0x2E, 0x1), 0x1F },
154 { SMCA_CS_V2, HWID_MCATYPE(0x2E, 0x2), 0x3FFF },
156 /* Unified Memory Controller MCA type */
157 { SMCA_UMC, HWID_MCATYPE(0x96, 0x0), 0xFF },
159 /* Parameter Block MCA type */
160 { SMCA_PB, HWID_MCATYPE(0x05, 0x0), 0x1 },
162 /* Platform Security Processor MCA type */
163 { SMCA_PSP, HWID_MCATYPE(0xFF, 0x0), 0x1 },
164 { SMCA_PSP_V2, HWID_MCATYPE(0xFF, 0x1), 0x3FFFF },
166 /* System Management Unit MCA type */
167 { SMCA_SMU, HWID_MCATYPE(0x01, 0x0), 0x1 },
168 { SMCA_SMU_V2, HWID_MCATYPE(0x01, 0x1), 0x7FF },
170 /* Microprocessor 5 Unit MCA type */
171 { SMCA_MP5, HWID_MCATYPE(0x01, 0x2), 0x3FF },
173 /* Northbridge IO Unit MCA type */
174 { SMCA_NBIO, HWID_MCATYPE(0x18, 0x0), 0x1F },
176 /* PCI Express Unit MCA type */
177 { SMCA_PCIE, HWID_MCATYPE(0x46, 0x0), 0x1F },
180 struct smca_bank smca_banks[MAX_NR_BANKS];
181 EXPORT_SYMBOL_GPL(smca_banks);
184 * In SMCA enabled processors, we can have multiple banks for a given IP type.
185 * So to define a unique name for each bank, we use a temp c-string to append
186 * the MCA_IPID[InstanceId] to type's name in get_name().
188 * InstanceId is 32 bits which is 8 characters. Make sure MAX_MCATYPE_NAME_LEN
189 * is greater than 8 plus 1 (for underscore) plus length of longest type name.
191 #define MAX_MCATYPE_NAME_LEN 30
192 static char buf_mcatype[MAX_MCATYPE_NAME_LEN];
194 static DEFINE_PER_CPU(struct threshold_bank **, threshold_banks);
195 static DEFINE_PER_CPU(unsigned int, bank_map); /* see which banks are on */
197 /* Map of banks that have more than MCA_MISC0 available. */
198 static DEFINE_PER_CPU(u32, smca_misc_banks_map);
200 static void amd_threshold_interrupt(void);
201 static void amd_deferred_error_interrupt(void);
203 static void default_deferred_error_interrupt(void)
205 pr_err("Unexpected deferred interrupt at vector %x\n", DEFERRED_ERROR_VECTOR);
207 void (*deferred_error_int_vector)(void) = default_deferred_error_interrupt;
209 static void smca_set_misc_banks_map(unsigned int bank, unsigned int cpu)
214 * For SMCA enabled processors, BLKPTR field of the first MISC register
215 * (MCx_MISC0) indicates presence of additional MISC regs set (MISC1-4).
217 if (rdmsr_safe(MSR_AMD64_SMCA_MCx_CONFIG(bank), &low, &high))
220 if (!(low & MCI_CONFIG_MCAX))
223 if (rdmsr_safe(MSR_AMD64_SMCA_MCx_MISC(bank), &low, &high))
226 if (low & MASK_BLKPTR_LO)
227 per_cpu(smca_misc_banks_map, cpu) |= BIT(bank);
231 static void smca_configure(unsigned int bank, unsigned int cpu)
233 unsigned int i, hwid_mcatype;
234 struct smca_hwid *s_hwid;
236 u32 smca_config = MSR_AMD64_SMCA_MCx_CONFIG(bank);
238 /* Set appropriate bits in MCA_CONFIG */
239 if (!rdmsr_safe(smca_config, &low, &high)) {
241 * OS is required to set the MCAX bit to acknowledge that it is
242 * now using the new MSR ranges and new registers under each
243 * bank. It also means that the OS will configure deferred
244 * errors in the new MCx_CONFIG register. If the bit is not set,
245 * uncorrectable errors will cause a system panic.
247 * MCA_CONFIG[MCAX] is bit 32 (0 in the high portion of the MSR.)
252 * SMCA sets the Deferred Error Interrupt type per bank.
254 * MCA_CONFIG[DeferredIntTypeSupported] is bit 5, and tells us
255 * if the DeferredIntType bit field is available.
257 * MCA_CONFIG[DeferredIntType] is bits [38:37] ([6:5] in the
258 * high portion of the MSR). OS should set this to 0x1 to enable
259 * APIC based interrupt. First, check that no interrupt has been
262 if ((low & BIT(5)) && !((high >> 5) & 0x3))
265 wrmsr(smca_config, low, high);
268 smca_set_misc_banks_map(bank, cpu);
270 /* Return early if this bank was already initialized. */
271 if (smca_banks[bank].hwid && smca_banks[bank].hwid->hwid_mcatype != 0)
274 if (rdmsr_safe(MSR_AMD64_SMCA_MCx_IPID(bank), &low, &high)) {
275 pr_warn("Failed to read MCA_IPID for bank %d\n", bank);
279 hwid_mcatype = HWID_MCATYPE(high & MCI_IPID_HWID,
280 (high & MCI_IPID_MCATYPE) >> 16);
282 for (i = 0; i < ARRAY_SIZE(smca_hwid_mcatypes); i++) {
283 s_hwid = &smca_hwid_mcatypes[i];
284 if (hwid_mcatype == s_hwid->hwid_mcatype) {
285 smca_banks[bank].hwid = s_hwid;
286 smca_banks[bank].id = low;
287 smca_banks[bank].sysfs_id = s_hwid->count++;
293 struct thresh_restart {
294 struct threshold_block *b;
301 static inline bool is_shared_bank(int bank)
304 * Scalable MCA provides for only one core to have access to the MSRs of
310 /* Bank 4 is for northbridge reporting and is thus shared */
314 static const char *bank4_names(const struct threshold_block *b)
316 switch (b->address) {
328 WARN(1, "Funny MSR: 0x%08x\n", b->address);
334 static bool lvt_interrupt_supported(unsigned int bank, u32 msr_high_bits)
337 * bank 4 supports APIC LVT interrupts implicitly since forever.
343 * IntP: interrupt present; if this bit is set, the thresholding
344 * bank can generate APIC LVT interrupts
346 return msr_high_bits & BIT(28);
349 static int lvt_off_valid(struct threshold_block *b, int apic, u32 lo, u32 hi)
351 int msr = (hi & MASK_LVTOFF_HI) >> 20;
354 pr_err(FW_BUG "cpu %d, failed to setup threshold interrupt "
355 "for bank %d, block %d (MSR%08X=0x%x%08x)\n", b->cpu,
356 b->bank, b->block, b->address, hi, lo);
362 * On SMCA CPUs, LVT offset is programmed at a different MSR, and
363 * the BIOS provides the value. The original field where LVT offset
364 * was set is reserved. Return early here:
369 pr_err(FW_BUG "cpu %d, invalid threshold interrupt offset %d "
370 "for bank %d, block %d (MSR%08X=0x%x%08x)\n",
371 b->cpu, apic, b->bank, b->block, b->address, hi, lo);
378 /* Reprogram MCx_MISC MSR behind this threshold bank. */
379 static void threshold_restart_bank(void *_tr)
381 struct thresh_restart *tr = _tr;
384 rdmsr(tr->b->address, lo, hi);
386 if (tr->b->threshold_limit < (hi & THRESHOLD_MAX))
387 tr->reset = 1; /* limit cannot be lower than err count */
389 if (tr->reset) { /* reset err count and overflow bit */
391 (hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
392 (THRESHOLD_MAX - tr->b->threshold_limit);
393 } else if (tr->old_limit) { /* change limit w/o reset */
394 int new_count = (hi & THRESHOLD_MAX) +
395 (tr->old_limit - tr->b->threshold_limit);
397 hi = (hi & ~MASK_ERR_COUNT_HI) |
398 (new_count & THRESHOLD_MAX);
402 hi &= ~MASK_INT_TYPE_HI;
404 if (!tr->b->interrupt_capable)
407 if (tr->set_lvt_off) {
408 if (lvt_off_valid(tr->b, tr->lvt_off, lo, hi)) {
409 /* set new lvt offset */
410 hi &= ~MASK_LVTOFF_HI;
411 hi |= tr->lvt_off << 20;
415 if (tr->b->interrupt_enable)
420 hi |= MASK_COUNT_EN_HI;
421 wrmsr(tr->b->address, lo, hi);
424 static void mce_threshold_block_init(struct threshold_block *b, int offset)
426 struct thresh_restart tr = {
432 b->threshold_limit = THRESHOLD_MAX;
433 threshold_restart_bank(&tr);
436 static int setup_APIC_mce_threshold(int reserved, int new)
438 if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
439 APIC_EILVT_MSG_FIX, 0))
445 static int setup_APIC_deferred_error(int reserved, int new)
447 if (reserved < 0 && !setup_APIC_eilvt(new, DEFERRED_ERROR_VECTOR,
448 APIC_EILVT_MSG_FIX, 0))
454 static void deferred_error_interrupt_enable(struct cpuinfo_x86 *c)
456 u32 low = 0, high = 0;
457 int def_offset = -1, def_new;
459 if (rdmsr_safe(MSR_CU_DEF_ERR, &low, &high))
462 def_new = (low & MASK_DEF_LVTOFF) >> 4;
463 if (!(low & MASK_DEF_LVTOFF)) {
464 pr_err(FW_BUG "Your BIOS is not setting up LVT offset 0x2 for deferred error IRQs correctly.\n");
465 def_new = DEF_LVT_OFF;
466 low = (low & ~MASK_DEF_LVTOFF) | (DEF_LVT_OFF << 4);
469 def_offset = setup_APIC_deferred_error(def_offset, def_new);
470 if ((def_offset == def_new) &&
471 (deferred_error_int_vector != amd_deferred_error_interrupt))
472 deferred_error_int_vector = amd_deferred_error_interrupt;
475 low = (low & ~MASK_DEF_INT_TYPE) | DEF_INT_TYPE_APIC;
477 wrmsr(MSR_CU_DEF_ERR, low, high);
480 static u32 smca_get_block_address(unsigned int bank, unsigned int block,
484 return MSR_AMD64_SMCA_MCx_MISC(bank);
486 if (!(per_cpu(smca_misc_banks_map, cpu) & BIT(bank)))
489 return MSR_AMD64_SMCA_MCx_MISCy(bank, block - 1);
492 static u32 get_block_address(u32 current_addr, u32 low, u32 high,
493 unsigned int bank, unsigned int block,
496 u32 addr = 0, offset = 0;
498 if ((bank >= per_cpu(mce_num_banks, cpu)) || (block >= NR_BLOCKS))
502 return smca_get_block_address(bank, block, cpu);
504 /* Fall back to method we used for older processors: */
507 addr = msr_ops.misc(bank);
510 offset = ((low & MASK_BLKPTR_LO) >> 21);
512 addr = MCG_XBLK_ADDR + offset;
515 addr = ++current_addr;
521 prepare_threshold_block(unsigned int bank, unsigned int block, u32 addr,
522 int offset, u32 misc_high)
524 unsigned int cpu = smp_processor_id();
525 u32 smca_low, smca_high;
526 struct threshold_block b;
530 per_cpu(bank_map, cpu) |= (1 << bank);
532 memset(&b, 0, sizeof(b));
537 b.interrupt_capable = lvt_interrupt_supported(bank, misc_high);
539 if (!b.interrupt_capable)
542 b.interrupt_enable = 1;
544 if (!mce_flags.smca) {
545 new = (misc_high & MASK_LVTOFF_HI) >> 20;
549 /* Gather LVT offset for thresholding: */
550 if (rdmsr_safe(MSR_CU_DEF_ERR, &smca_low, &smca_high))
553 new = (smca_low & SMCA_THR_LVT_OFF) >> 12;
556 offset = setup_APIC_mce_threshold(offset, new);
558 thresholding_irq_en = true;
561 mce_threshold_block_init(&b, offset);
567 bool amd_filter_mce(struct mce *m)
569 enum smca_bank_types bank_type = smca_get_bank_type(m->bank);
570 struct cpuinfo_x86 *c = &boot_cpu_data;
571 u8 xec = (m->status >> 16) & 0x3F;
573 /* See Family 17h Models 10h-2Fh Erratum #1114. */
574 if (c->x86 == 0x17 &&
575 c->x86_model >= 0x10 && c->x86_model <= 0x2F &&
576 bank_type == SMCA_IF && xec == 10)
583 * Turn off thresholding banks for the following conditions:
584 * - MC4_MISC thresholding is not supported on Family 0x15.
585 * - Prevent possible spurious interrupts from the IF bank on Family 0x17
586 * Models 0x10-0x2F due to Erratum #1114.
588 static void disable_err_thresholding(struct cpuinfo_x86 *c, unsigned int bank)
595 if (c->x86 == 0x15 && bank == 4) {
596 msrs[0] = 0x00000413; /* MC4_MISC0 */
597 msrs[1] = 0xc0000408; /* MC4_MISC1 */
599 } else if (c->x86 == 0x17 &&
600 (c->x86_model >= 0x10 && c->x86_model <= 0x2F)) {
602 if (smca_get_bank_type(bank) != SMCA_IF)
605 msrs[0] = MSR_AMD64_SMCA_MCx_MISC(bank);
611 rdmsrl(MSR_K7_HWCR, hwcr);
613 /* McStatusWrEn has to be set */
614 need_toggle = !(hwcr & BIT(18));
616 wrmsrl(MSR_K7_HWCR, hwcr | BIT(18));
618 /* Clear CntP bit safely */
619 for (i = 0; i < num_msrs; i++)
620 msr_clear_bit(msrs[i], 62);
622 /* restore old settings */
624 wrmsrl(MSR_K7_HWCR, hwcr);
627 /* cpu init entry point, called from mce.c with preempt off */
628 void mce_amd_feature_init(struct cpuinfo_x86 *c)
630 unsigned int bank, block, cpu = smp_processor_id();
631 u32 low = 0, high = 0, address = 0;
635 for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
637 smca_configure(bank, cpu);
639 disable_err_thresholding(c, bank);
641 for (block = 0; block < NR_BLOCKS; ++block) {
642 address = get_block_address(address, low, high, bank, block, cpu);
646 if (rdmsr_safe(address, &low, &high))
649 if (!(high & MASK_VALID_HI))
652 if (!(high & MASK_CNTP_HI) ||
653 (high & MASK_LOCKED_HI))
656 offset = prepare_threshold_block(bank, block, address, offset, high);
660 if (mce_flags.succor)
661 deferred_error_interrupt_enable(c);
664 int umc_normaddr_to_sysaddr(u64 norm_addr, u16 nid, u8 umc, u64 *sys_addr)
666 u64 dram_base_addr, dram_limit_addr, dram_hole_base;
667 /* We start from the normalized address */
668 u64 ret_addr = norm_addr;
672 u8 die_id_shift, die_id_mask, socket_id_shift, socket_id_mask;
673 u8 intlv_num_dies, intlv_num_chan, intlv_num_sockets;
674 u8 intlv_addr_sel, intlv_addr_bit;
675 u8 num_intlv_bits, hashed_bit;
676 u8 lgcy_mmio_hole_en, base = 0;
677 u8 cs_mask, cs_id = 0;
678 bool hash_enabled = false;
680 /* Read D18F0x1B4 (DramOffset), check if base 1 is used. */
681 if (amd_df_indirect_read(nid, 0, 0x1B4, umc, &tmp))
684 /* Remove HiAddrOffset from normalized address, if enabled: */
686 u64 hi_addr_offset = (tmp & GENMASK_ULL(31, 20)) << 8;
688 if (norm_addr >= hi_addr_offset) {
689 ret_addr -= hi_addr_offset;
694 /* Read D18F0x110 (DramBaseAddress). */
695 if (amd_df_indirect_read(nid, 0, 0x110 + (8 * base), umc, &tmp))
698 /* Check if address range is valid. */
699 if (!(tmp & BIT(0))) {
700 pr_err("%s: Invalid DramBaseAddress range: 0x%x.\n",
705 lgcy_mmio_hole_en = tmp & BIT(1);
706 intlv_num_chan = (tmp >> 4) & 0xF;
707 intlv_addr_sel = (tmp >> 8) & 0x7;
708 dram_base_addr = (tmp & GENMASK_ULL(31, 12)) << 16;
710 /* {0, 1, 2, 3} map to address bits {8, 9, 10, 11} respectively */
711 if (intlv_addr_sel > 3) {
712 pr_err("%s: Invalid interleave address select %d.\n",
713 __func__, intlv_addr_sel);
717 /* Read D18F0x114 (DramLimitAddress). */
718 if (amd_df_indirect_read(nid, 0, 0x114 + (8 * base), umc, &tmp))
721 intlv_num_sockets = (tmp >> 8) & 0x1;
722 intlv_num_dies = (tmp >> 10) & 0x3;
723 dram_limit_addr = ((tmp & GENMASK_ULL(31, 12)) << 16) | GENMASK_ULL(27, 0);
725 intlv_addr_bit = intlv_addr_sel + 8;
727 /* Re-use intlv_num_chan by setting it equal to log2(#channels) */
728 switch (intlv_num_chan) {
729 case 0: intlv_num_chan = 0; break;
730 case 1: intlv_num_chan = 1; break;
731 case 3: intlv_num_chan = 2; break;
732 case 5: intlv_num_chan = 3; break;
733 case 7: intlv_num_chan = 4; break;
735 case 8: intlv_num_chan = 1;
739 pr_err("%s: Invalid number of interleaved channels %d.\n",
740 __func__, intlv_num_chan);
744 num_intlv_bits = intlv_num_chan;
746 if (intlv_num_dies > 2) {
747 pr_err("%s: Invalid number of interleaved nodes/dies %d.\n",
748 __func__, intlv_num_dies);
752 num_intlv_bits += intlv_num_dies;
754 /* Add a bit if sockets are interleaved. */
755 num_intlv_bits += intlv_num_sockets;
757 /* Assert num_intlv_bits <= 4 */
758 if (num_intlv_bits > 4) {
759 pr_err("%s: Invalid interleave bits %d.\n",
760 __func__, num_intlv_bits);
764 if (num_intlv_bits > 0) {
765 u64 temp_addr_x, temp_addr_i, temp_addr_y;
766 u8 die_id_bit, sock_id_bit, cs_fabric_id;
769 * Read FabricBlockInstanceInformation3_CS[BlockFabricID].
770 * This is the fabric id for this coherent slave. Use
771 * umc/channel# as instance id of the coherent slave
774 if (amd_df_indirect_read(nid, 0, 0x50, umc, &tmp))
777 cs_fabric_id = (tmp >> 8) & 0xFF;
780 /* If interleaved over more than 1 channel: */
781 if (intlv_num_chan) {
782 die_id_bit = intlv_num_chan;
783 cs_mask = (1 << die_id_bit) - 1;
784 cs_id = cs_fabric_id & cs_mask;
787 sock_id_bit = die_id_bit;
789 /* Read D18F1x208 (SystemFabricIdMask). */
790 if (intlv_num_dies || intlv_num_sockets)
791 if (amd_df_indirect_read(nid, 1, 0x208, umc, &tmp))
794 /* If interleaved over more than 1 die. */
795 if (intlv_num_dies) {
796 sock_id_bit = die_id_bit + intlv_num_dies;
797 die_id_shift = (tmp >> 24) & 0xF;
798 die_id_mask = (tmp >> 8) & 0xFF;
800 cs_id |= ((cs_fabric_id & die_id_mask) >> die_id_shift) << die_id_bit;
803 /* If interleaved over more than 1 socket. */
804 if (intlv_num_sockets) {
805 socket_id_shift = (tmp >> 28) & 0xF;
806 socket_id_mask = (tmp >> 16) & 0xFF;
808 cs_id |= ((cs_fabric_id & socket_id_mask) >> socket_id_shift) << sock_id_bit;
812 * The pre-interleaved address consists of XXXXXXIIIYYYYY
813 * where III is the ID for this CS, and XXXXXXYYYYY are the
814 * address bits from the post-interleaved address.
815 * "num_intlv_bits" has been calculated to tell us how many "I"
816 * bits there are. "intlv_addr_bit" tells us how many "Y" bits
817 * there are (where "I" starts).
819 temp_addr_y = ret_addr & GENMASK_ULL(intlv_addr_bit-1, 0);
820 temp_addr_i = (cs_id << intlv_addr_bit);
821 temp_addr_x = (ret_addr & GENMASK_ULL(63, intlv_addr_bit)) << num_intlv_bits;
822 ret_addr = temp_addr_x | temp_addr_i | temp_addr_y;
825 /* Add dram base address */
826 ret_addr += dram_base_addr;
828 /* If legacy MMIO hole enabled */
829 if (lgcy_mmio_hole_en) {
830 if (amd_df_indirect_read(nid, 0, 0x104, umc, &tmp))
833 dram_hole_base = tmp & GENMASK(31, 24);
834 if (ret_addr >= dram_hole_base)
835 ret_addr += (BIT_ULL(32) - dram_hole_base);
839 /* Save some parentheses and grab ls-bit at the end. */
840 hashed_bit = (ret_addr >> 12) ^
846 hashed_bit &= BIT(0);
848 if (hashed_bit != ((ret_addr >> intlv_addr_bit) & BIT(0)))
849 ret_addr ^= BIT(intlv_addr_bit);
852 /* Is calculated system address is above DRAM limit address? */
853 if (ret_addr > dram_limit_addr)
856 *sys_addr = ret_addr;
862 EXPORT_SYMBOL_GPL(umc_normaddr_to_sysaddr);
864 bool amd_mce_is_memory_error(struct mce *m)
866 /* ErrCodeExt[20:16] */
867 u8 xec = (m->status >> 16) & 0x1f;
870 return smca_get_bank_type(m->bank) == SMCA_UMC && xec == 0x0;
872 return m->bank == 4 && xec == 0x8;
875 static void __log_error(unsigned int bank, u64 status, u64 addr, u64 misc)
886 if (m.status & MCI_STATUS_ADDRV) {
890 * Extract [55:<lsb>] where lsb is the least significant
891 * *valid* bit of the address bits.
893 if (mce_flags.smca) {
894 u8 lsb = (m.addr >> 56) & 0x3f;
896 m.addr &= GENMASK_ULL(55, lsb);
900 if (mce_flags.smca) {
901 rdmsrl(MSR_AMD64_SMCA_MCx_IPID(bank), m.ipid);
903 if (m.status & MCI_STATUS_SYNDV)
904 rdmsrl(MSR_AMD64_SMCA_MCx_SYND(bank), m.synd);
910 asmlinkage __visible void __irq_entry smp_deferred_error_interrupt(struct pt_regs *regs)
913 trace_deferred_error_apic_entry(DEFERRED_ERROR_VECTOR);
914 inc_irq_stat(irq_deferred_error_count);
915 deferred_error_int_vector();
916 trace_deferred_error_apic_exit(DEFERRED_ERROR_VECTOR);
921 * Returns true if the logged error is deferred. False, otherwise.
924 _log_error_bank(unsigned int bank, u32 msr_stat, u32 msr_addr, u64 misc)
926 u64 status, addr = 0;
928 rdmsrl(msr_stat, status);
929 if (!(status & MCI_STATUS_VAL))
932 if (status & MCI_STATUS_ADDRV)
933 rdmsrl(msr_addr, addr);
935 __log_error(bank, status, addr, misc);
939 return status & MCI_STATUS_DEFERRED;
943 * We have three scenarios for checking for Deferred errors:
945 * 1) Non-SMCA systems check MCA_STATUS and log error if found.
946 * 2) SMCA systems check MCA_STATUS. If error is found then log it and also
948 * 3) SMCA systems check MCA_DESTAT, if error was not found in MCA_STATUS, and
951 static void log_error_deferred(unsigned int bank)
955 defrd = _log_error_bank(bank, msr_ops.status(bank),
956 msr_ops.addr(bank), 0);
961 /* Clear MCA_DESTAT if we logged the deferred error from MCA_STATUS. */
963 wrmsrl(MSR_AMD64_SMCA_MCx_DESTAT(bank), 0);
968 * Only deferred errors are logged in MCA_DE{STAT,ADDR} so just check
971 _log_error_bank(bank, MSR_AMD64_SMCA_MCx_DESTAT(bank),
972 MSR_AMD64_SMCA_MCx_DEADDR(bank), 0);
975 /* APIC interrupt handler for deferred errors */
976 static void amd_deferred_error_interrupt(void)
980 for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank)
981 log_error_deferred(bank);
984 static void log_error_thresholding(unsigned int bank, u64 misc)
986 _log_error_bank(bank, msr_ops.status(bank), msr_ops.addr(bank), misc);
989 static void log_and_reset_block(struct threshold_block *block)
991 struct thresh_restart tr;
992 u32 low = 0, high = 0;
997 if (rdmsr_safe(block->address, &low, &high))
1000 if (!(high & MASK_OVERFLOW_HI))
1003 /* Log the MCE which caused the threshold event. */
1004 log_error_thresholding(block->bank, ((u64)high << 32) | low);
1006 /* Reset threshold block after logging error. */
1007 memset(&tr, 0, sizeof(tr));
1009 threshold_restart_bank(&tr);
1013 * Threshold interrupt handler will service THRESHOLD_APIC_VECTOR. The interrupt
1014 * goes off when error_count reaches threshold_limit.
1016 static void amd_threshold_interrupt(void)
1018 struct threshold_block *first_block = NULL, *block = NULL, *tmp = NULL;
1019 unsigned int bank, cpu = smp_processor_id();
1021 for (bank = 0; bank < this_cpu_read(mce_num_banks); ++bank) {
1022 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
1025 first_block = per_cpu(threshold_banks, cpu)[bank]->blocks;
1030 * The first block is also the head of the list. Check it first
1031 * before iterating over the rest.
1033 log_and_reset_block(first_block);
1034 list_for_each_entry_safe(block, tmp, &first_block->miscj, miscj)
1035 log_and_reset_block(block);
1043 struct threshold_attr {
1044 struct attribute attr;
1045 ssize_t (*show) (struct threshold_block *, char *);
1046 ssize_t (*store) (struct threshold_block *, const char *, size_t count);
1049 #define SHOW_FIELDS(name) \
1050 static ssize_t show_ ## name(struct threshold_block *b, char *buf) \
1052 return sprintf(buf, "%lu\n", (unsigned long) b->name); \
1054 SHOW_FIELDS(interrupt_enable)
1055 SHOW_FIELDS(threshold_limit)
1058 store_interrupt_enable(struct threshold_block *b, const char *buf, size_t size)
1060 struct thresh_restart tr;
1063 if (!b->interrupt_capable)
1066 if (kstrtoul(buf, 0, &new) < 0)
1069 b->interrupt_enable = !!new;
1071 memset(&tr, 0, sizeof(tr));
1074 smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
1080 store_threshold_limit(struct threshold_block *b, const char *buf, size_t size)
1082 struct thresh_restart tr;
1085 if (kstrtoul(buf, 0, &new) < 0)
1088 if (new > THRESHOLD_MAX)
1089 new = THRESHOLD_MAX;
1093 memset(&tr, 0, sizeof(tr));
1094 tr.old_limit = b->threshold_limit;
1095 b->threshold_limit = new;
1098 smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
1103 static ssize_t show_error_count(struct threshold_block *b, char *buf)
1107 rdmsr_on_cpu(b->cpu, b->address, &lo, &hi);
1109 return sprintf(buf, "%u\n", ((hi & THRESHOLD_MAX) -
1110 (THRESHOLD_MAX - b->threshold_limit)));
1113 static struct threshold_attr error_count = {
1114 .attr = {.name = __stringify(error_count), .mode = 0444 },
1115 .show = show_error_count,
1118 #define RW_ATTR(val) \
1119 static struct threshold_attr val = { \
1120 .attr = {.name = __stringify(val), .mode = 0644 }, \
1121 .show = show_## val, \
1122 .store = store_## val, \
1125 RW_ATTR(interrupt_enable);
1126 RW_ATTR(threshold_limit);
1128 static struct attribute *default_attrs[] = {
1129 &threshold_limit.attr,
1131 NULL, /* possibly interrupt_enable if supported, see below */
1135 #define to_block(k) container_of(k, struct threshold_block, kobj)
1136 #define to_attr(a) container_of(a, struct threshold_attr, attr)
1138 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
1140 struct threshold_block *b = to_block(kobj);
1141 struct threshold_attr *a = to_attr(attr);
1144 ret = a->show ? a->show(b, buf) : -EIO;
1149 static ssize_t store(struct kobject *kobj, struct attribute *attr,
1150 const char *buf, size_t count)
1152 struct threshold_block *b = to_block(kobj);
1153 struct threshold_attr *a = to_attr(attr);
1156 ret = a->store ? a->store(b, buf, count) : -EIO;
1161 static const struct sysfs_ops threshold_ops = {
1166 static void threshold_block_release(struct kobject *kobj);
1168 static struct kobj_type threshold_ktype = {
1169 .sysfs_ops = &threshold_ops,
1170 .default_attrs = default_attrs,
1171 .release = threshold_block_release,
1174 static const char *get_name(unsigned int bank, struct threshold_block *b)
1176 enum smca_bank_types bank_type;
1178 if (!mce_flags.smca) {
1180 return bank4_names(b);
1182 return th_names[bank];
1185 bank_type = smca_get_bank_type(bank);
1186 if (bank_type >= N_SMCA_BANK_TYPES)
1189 if (b && bank_type == SMCA_UMC) {
1190 if (b->block < ARRAY_SIZE(smca_umc_block_names))
1191 return smca_umc_block_names[b->block];
1195 if (smca_banks[bank].hwid->count == 1)
1196 return smca_get_name(bank_type);
1198 snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN,
1199 "%s_%x", smca_get_name(bank_type),
1200 smca_banks[bank].sysfs_id);
1204 static int allocate_threshold_blocks(unsigned int cpu, struct threshold_bank *tb,
1205 unsigned int bank, unsigned int block,
1208 struct threshold_block *b = NULL;
1212 if ((bank >= per_cpu(mce_num_banks, cpu)) || (block >= NR_BLOCKS))
1215 if (rdmsr_safe_on_cpu(cpu, address, &low, &high))
1218 if (!(high & MASK_VALID_HI)) {
1225 if (!(high & MASK_CNTP_HI) ||
1226 (high & MASK_LOCKED_HI))
1229 b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
1236 b->address = address;
1237 b->interrupt_enable = 0;
1238 b->interrupt_capable = lvt_interrupt_supported(bank, high);
1239 b->threshold_limit = THRESHOLD_MAX;
1241 if (b->interrupt_capable) {
1242 threshold_ktype.default_attrs[2] = &interrupt_enable.attr;
1243 b->interrupt_enable = 1;
1245 threshold_ktype.default_attrs[2] = NULL;
1248 INIT_LIST_HEAD(&b->miscj);
1251 list_add(&b->miscj, &tb->blocks->miscj);
1255 err = kobject_init_and_add(&b->kobj, &threshold_ktype, tb->kobj, get_name(bank, b));
1259 address = get_block_address(address, low, high, bank, ++block, cpu);
1263 err = allocate_threshold_blocks(cpu, tb, bank, block, address);
1268 kobject_uevent(&b->kobj, KOBJ_ADD);
1274 kobject_put(&b->kobj);
1275 list_del(&b->miscj);
1281 static int __threshold_add_blocks(struct threshold_bank *b)
1283 struct list_head *head = &b->blocks->miscj;
1284 struct threshold_block *pos = NULL;
1285 struct threshold_block *tmp = NULL;
1288 err = kobject_add(&b->blocks->kobj, b->kobj, b->blocks->kobj.name);
1292 list_for_each_entry_safe(pos, tmp, head, miscj) {
1294 err = kobject_add(&pos->kobj, b->kobj, pos->kobj.name);
1296 list_for_each_entry_safe_reverse(pos, tmp, head, miscj)
1297 kobject_del(&pos->kobj);
1305 static int threshold_create_bank(unsigned int cpu, unsigned int bank)
1307 struct device *dev = per_cpu(mce_device, cpu);
1308 struct amd_northbridge *nb = NULL;
1309 struct threshold_bank *b = NULL;
1310 const char *name = get_name(bank, NULL);
1316 if (is_shared_bank(bank)) {
1317 nb = node_to_amd_nb(amd_get_nb_id(cpu));
1319 /* threshold descriptor already initialized on this node? */
1320 if (nb && nb->bank4) {
1323 err = kobject_add(b->kobj, &dev->kobj, name);
1327 per_cpu(threshold_banks, cpu)[bank] = b;
1328 refcount_inc(&b->cpus);
1330 err = __threshold_add_blocks(b);
1336 b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
1342 b->kobj = kobject_create_and_add(name, &dev->kobj);
1348 if (is_shared_bank(bank)) {
1349 refcount_set(&b->cpus, 1);
1351 /* nb is already initialized, see above */
1358 err = allocate_threshold_blocks(cpu, b, bank, 0, msr_ops.misc(bank));
1362 per_cpu(threshold_banks, cpu)[bank] = b;
1373 static void threshold_block_release(struct kobject *kobj)
1375 kfree(to_block(kobj));
1378 static void deallocate_threshold_block(unsigned int cpu, unsigned int bank)
1380 struct threshold_block *pos = NULL;
1381 struct threshold_block *tmp = NULL;
1382 struct threshold_bank *head = per_cpu(threshold_banks, cpu)[bank];
1387 list_for_each_entry_safe(pos, tmp, &head->blocks->miscj, miscj) {
1388 list_del(&pos->miscj);
1389 kobject_put(&pos->kobj);
1392 kobject_put(&head->blocks->kobj);
1395 static void __threshold_remove_blocks(struct threshold_bank *b)
1397 struct threshold_block *pos = NULL;
1398 struct threshold_block *tmp = NULL;
1400 kobject_del(b->kobj);
1402 list_for_each_entry_safe(pos, tmp, &b->blocks->miscj, miscj)
1403 kobject_del(&pos->kobj);
1406 static void threshold_remove_bank(unsigned int cpu, int bank)
1408 struct amd_northbridge *nb;
1409 struct threshold_bank *b;
1411 b = per_cpu(threshold_banks, cpu)[bank];
1418 if (is_shared_bank(bank)) {
1419 if (!refcount_dec_and_test(&b->cpus)) {
1420 __threshold_remove_blocks(b);
1421 per_cpu(threshold_banks, cpu)[bank] = NULL;
1425 * the last CPU on this node using the shared bank is
1426 * going away, remove that bank now.
1428 nb = node_to_amd_nb(amd_get_nb_id(cpu));
1433 deallocate_threshold_block(cpu, bank);
1436 kobject_del(b->kobj);
1437 kobject_put(b->kobj);
1439 per_cpu(threshold_banks, cpu)[bank] = NULL;
1442 int mce_threshold_remove_device(unsigned int cpu)
1446 for (bank = 0; bank < per_cpu(mce_num_banks, cpu); ++bank) {
1447 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
1449 threshold_remove_bank(cpu, bank);
1451 kfree(per_cpu(threshold_banks, cpu));
1452 per_cpu(threshold_banks, cpu) = NULL;
1456 /* create dir/files for all valid threshold banks */
1457 int mce_threshold_create_device(unsigned int cpu)
1460 struct threshold_bank **bp;
1463 bp = per_cpu(threshold_banks, cpu);
1467 bp = kcalloc(per_cpu(mce_num_banks, cpu), sizeof(struct threshold_bank *),
1472 per_cpu(threshold_banks, cpu) = bp;
1474 for (bank = 0; bank < per_cpu(mce_num_banks, cpu); ++bank) {
1475 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
1477 err = threshold_create_bank(cpu, bank);
1483 mce_threshold_remove_device(cpu);
1487 static __init int threshold_init_device(void)
1491 /* to hit CPUs online before the notifier is up */
1492 for_each_online_cpu(lcpu) {
1493 int err = mce_threshold_create_device(lcpu);
1499 if (thresholding_irq_en)
1500 mce_threshold_vector = amd_threshold_interrupt;
1505 * there are 3 funcs which need to be _initcalled in a logic sequence:
1506 * 1. xen_late_init_mcelog
1507 * 2. mcheck_init_device
1508 * 3. threshold_init_device
1510 * xen_late_init_mcelog must register xen_mce_chrdev_device before
1511 * native mce_chrdev_device registration if running under xen platform;
1513 * mcheck_init_device should be inited before threshold_init_device to
1514 * initialize mce_device, otherwise a NULL ptr dereference will cause panic.
1516 * so we use following _initcalls
1517 * 1. device_initcall(xen_late_init_mcelog);
1518 * 2. device_initcall_sync(mcheck_init_device);
1519 * 3. late_initcall(threshold_init_device);
1521 * when running under xen, the initcall order is 1,2,3;
1522 * on baremetal, we skip 1 and we do only 2 and 3.
1524 late_initcall(threshold_init_device);