2 * (c) 2005-2016 Advanced Micro Devices, Inc.
3 * Your use of this code is subject to the terms and conditions of the
4 * GNU general public license version 2. See "COPYING" or
5 * http://www.gnu.org/licenses/gpl.html
7 * Written by Jacob Shin - AMD, Inc.
8 * Maintained by: Borislav Petkov <bp@alien8.de>
10 * All MC4_MISCi registers are shared between cores on a node.
12 #include <linux/interrupt.h>
13 #include <linux/notifier.h>
14 #include <linux/kobject.h>
15 #include <linux/percpu.h>
16 #include <linux/errno.h>
17 #include <linux/sched.h>
18 #include <linux/sysfs.h>
19 #include <linux/slab.h>
20 #include <linux/init.h>
21 #include <linux/cpu.h>
22 #include <linux/smp.h>
23 #include <linux/string.h>
25 #include <asm/amd_nb.h>
26 #include <asm/traps.h>
30 #include <asm/trace/irq_vectors.h>
32 #include "mce-internal.h"
35 #define THRESHOLD_MAX 0xFFF
36 #define INT_TYPE_APIC 0x00020000
37 #define MASK_VALID_HI 0x80000000
38 #define MASK_CNTP_HI 0x40000000
39 #define MASK_LOCKED_HI 0x20000000
40 #define MASK_LVTOFF_HI 0x00F00000
41 #define MASK_COUNT_EN_HI 0x00080000
42 #define MASK_INT_TYPE_HI 0x00060000
43 #define MASK_OVERFLOW_HI 0x00010000
44 #define MASK_ERR_COUNT_HI 0x00000FFF
45 #define MASK_BLKPTR_LO 0xFF000000
46 #define MCG_XBLK_ADDR 0xC0000400
48 /* Deferred error settings */
49 #define MSR_CU_DEF_ERR 0xC0000410
50 #define MASK_DEF_LVTOFF 0x000000F0
51 #define MASK_DEF_INT_TYPE 0x00000006
52 #define DEF_LVT_OFF 0x2
53 #define DEF_INT_TYPE_APIC 0x2
57 /* Threshold LVT offset is at MSR0xC0000410[15:12] */
58 #define SMCA_THR_LVT_OFF 0xF000
60 static bool thresholding_irq_en;
62 static const char * const th_names[] = {
71 static const char * const smca_umc_block_names[] = {
76 struct smca_bank_name {
77 const char *name; /* Short name for sysfs */
78 const char *long_name; /* Long name for pretty-printing */
81 static struct smca_bank_name smca_names[] = {
82 [SMCA_LS] = { "load_store", "Load Store Unit" },
83 [SMCA_IF] = { "insn_fetch", "Instruction Fetch Unit" },
84 [SMCA_L2_CACHE] = { "l2_cache", "L2 Cache" },
85 [SMCA_DE] = { "decode_unit", "Decode Unit" },
86 [SMCA_RESERVED] = { "reserved", "Reserved" },
87 [SMCA_EX] = { "execution_unit", "Execution Unit" },
88 [SMCA_FP] = { "floating_point", "Floating Point Unit" },
89 [SMCA_L3_CACHE] = { "l3_cache", "L3 Cache" },
90 [SMCA_CS] = { "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_SMU] = { "smu", "System Management Unit" },
98 static u32 smca_bank_addrs[MAX_NR_BANKS][NR_BLOCKS] __ro_after_init =
100 [0 ... MAX_NR_BANKS - 1] = { [0 ... NR_BLOCKS - 1] = -1 }
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), 0x1FFFEF },
142 { SMCA_IF, HWID_MCATYPE(0xB0, 0x1), 0x3FFF },
143 { SMCA_L2_CACHE, HWID_MCATYPE(0xB0, 0x2), 0xF },
144 { SMCA_DE, HWID_MCATYPE(0xB0, 0x3), 0x1FF },
145 /* HWID 0xB0 MCATYPE 0x4 is Reserved */
146 { SMCA_EX, HWID_MCATYPE(0xB0, 0x5), 0x7FF },
147 { SMCA_FP, HWID_MCATYPE(0xB0, 0x6), 0x7F },
148 { SMCA_L3_CACHE, HWID_MCATYPE(0xB0, 0x7), 0xFF },
150 /* Data Fabric MCA types */
151 { SMCA_CS, HWID_MCATYPE(0x2E, 0x0), 0x1FF },
152 { SMCA_PIE, HWID_MCATYPE(0x2E, 0x1), 0xF },
154 /* Unified Memory Controller MCA type */
155 { SMCA_UMC, HWID_MCATYPE(0x96, 0x0), 0x3F },
157 /* Parameter Block MCA type */
158 { SMCA_PB, HWID_MCATYPE(0x05, 0x0), 0x1 },
160 /* Platform Security Processor MCA type */
161 { SMCA_PSP, HWID_MCATYPE(0xFF, 0x0), 0x1 },
163 /* System Management Unit MCA type */
164 { SMCA_SMU, HWID_MCATYPE(0x01, 0x0), 0x1 },
167 struct smca_bank smca_banks[MAX_NR_BANKS];
168 EXPORT_SYMBOL_GPL(smca_banks);
171 * In SMCA enabled processors, we can have multiple banks for a given IP type.
172 * So to define a unique name for each bank, we use a temp c-string to append
173 * the MCA_IPID[InstanceId] to type's name in get_name().
175 * InstanceId is 32 bits which is 8 characters. Make sure MAX_MCATYPE_NAME_LEN
176 * is greater than 8 plus 1 (for underscore) plus length of longest type name.
178 #define MAX_MCATYPE_NAME_LEN 30
179 static char buf_mcatype[MAX_MCATYPE_NAME_LEN];
181 static DEFINE_PER_CPU(struct threshold_bank **, threshold_banks);
182 static DEFINE_PER_CPU(unsigned int, bank_map); /* see which banks are on */
184 static void amd_threshold_interrupt(void);
185 static void amd_deferred_error_interrupt(void);
187 static void default_deferred_error_interrupt(void)
189 pr_err("Unexpected deferred interrupt at vector %x\n", DEFERRED_ERROR_VECTOR);
191 void (*deferred_error_int_vector)(void) = default_deferred_error_interrupt;
193 static void smca_configure(unsigned int bank, unsigned int cpu)
195 unsigned int i, hwid_mcatype;
196 struct smca_hwid *s_hwid;
198 u32 smca_config = MSR_AMD64_SMCA_MCx_CONFIG(bank);
200 /* Set appropriate bits in MCA_CONFIG */
201 if (!rdmsr_safe(smca_config, &low, &high)) {
203 * OS is required to set the MCAX bit to acknowledge that it is
204 * now using the new MSR ranges and new registers under each
205 * bank. It also means that the OS will configure deferred
206 * errors in the new MCx_CONFIG register. If the bit is not set,
207 * uncorrectable errors will cause a system panic.
209 * MCA_CONFIG[MCAX] is bit 32 (0 in the high portion of the MSR.)
214 * SMCA sets the Deferred Error Interrupt type per bank.
216 * MCA_CONFIG[DeferredIntTypeSupported] is bit 5, and tells us
217 * if the DeferredIntType bit field is available.
219 * MCA_CONFIG[DeferredIntType] is bits [38:37] ([6:5] in the
220 * high portion of the MSR). OS should set this to 0x1 to enable
221 * APIC based interrupt. First, check that no interrupt has been
224 if ((low & BIT(5)) && !((high >> 5) & 0x3))
227 wrmsr(smca_config, low, high);
230 /* Return early if this bank was already initialized. */
231 if (smca_banks[bank].hwid && smca_banks[bank].hwid->hwid_mcatype != 0)
234 if (rdmsr_safe(MSR_AMD64_SMCA_MCx_IPID(bank), &low, &high)) {
235 pr_warn("Failed to read MCA_IPID for bank %d\n", bank);
239 hwid_mcatype = HWID_MCATYPE(high & MCI_IPID_HWID,
240 (high & MCI_IPID_MCATYPE) >> 16);
242 for (i = 0; i < ARRAY_SIZE(smca_hwid_mcatypes); i++) {
243 s_hwid = &smca_hwid_mcatypes[i];
244 if (hwid_mcatype == s_hwid->hwid_mcatype) {
245 smca_banks[bank].hwid = s_hwid;
246 smca_banks[bank].id = low;
247 smca_banks[bank].sysfs_id = s_hwid->count++;
253 struct thresh_restart {
254 struct threshold_block *b;
261 static inline bool is_shared_bank(int bank)
264 * Scalable MCA provides for only one core to have access to the MSRs of
270 /* Bank 4 is for northbridge reporting and is thus shared */
274 static const char *bank4_names(const struct threshold_block *b)
276 switch (b->address) {
288 WARN(1, "Funny MSR: 0x%08x\n", b->address);
294 static bool lvt_interrupt_supported(unsigned int bank, u32 msr_high_bits)
297 * bank 4 supports APIC LVT interrupts implicitly since forever.
303 * IntP: interrupt present; if this bit is set, the thresholding
304 * bank can generate APIC LVT interrupts
306 return msr_high_bits & BIT(28);
309 static int lvt_off_valid(struct threshold_block *b, int apic, u32 lo, u32 hi)
311 int msr = (hi & MASK_LVTOFF_HI) >> 20;
314 pr_err(FW_BUG "cpu %d, failed to setup threshold interrupt "
315 "for bank %d, block %d (MSR%08X=0x%x%08x)\n", b->cpu,
316 b->bank, b->block, b->address, hi, lo);
322 * On SMCA CPUs, LVT offset is programmed at a different MSR, and
323 * the BIOS provides the value. The original field where LVT offset
324 * was set is reserved. Return early here:
329 pr_err(FW_BUG "cpu %d, invalid threshold interrupt offset %d "
330 "for bank %d, block %d (MSR%08X=0x%x%08x)\n",
331 b->cpu, apic, b->bank, b->block, b->address, hi, lo);
338 /* Reprogram MCx_MISC MSR behind this threshold bank. */
339 static void threshold_restart_bank(void *_tr)
341 struct thresh_restart *tr = _tr;
344 rdmsr(tr->b->address, lo, hi);
346 if (tr->b->threshold_limit < (hi & THRESHOLD_MAX))
347 tr->reset = 1; /* limit cannot be lower than err count */
349 if (tr->reset) { /* reset err count and overflow bit */
351 (hi & ~(MASK_ERR_COUNT_HI | MASK_OVERFLOW_HI)) |
352 (THRESHOLD_MAX - tr->b->threshold_limit);
353 } else if (tr->old_limit) { /* change limit w/o reset */
354 int new_count = (hi & THRESHOLD_MAX) +
355 (tr->old_limit - tr->b->threshold_limit);
357 hi = (hi & ~MASK_ERR_COUNT_HI) |
358 (new_count & THRESHOLD_MAX);
362 hi &= ~MASK_INT_TYPE_HI;
364 if (!tr->b->interrupt_capable)
367 if (tr->set_lvt_off) {
368 if (lvt_off_valid(tr->b, tr->lvt_off, lo, hi)) {
369 /* set new lvt offset */
370 hi &= ~MASK_LVTOFF_HI;
371 hi |= tr->lvt_off << 20;
375 if (tr->b->interrupt_enable)
380 hi |= MASK_COUNT_EN_HI;
381 wrmsr(tr->b->address, lo, hi);
384 static void mce_threshold_block_init(struct threshold_block *b, int offset)
386 struct thresh_restart tr = {
392 b->threshold_limit = THRESHOLD_MAX;
393 threshold_restart_bank(&tr);
396 static int setup_APIC_mce_threshold(int reserved, int new)
398 if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
399 APIC_EILVT_MSG_FIX, 0))
405 static int setup_APIC_deferred_error(int reserved, int new)
407 if (reserved < 0 && !setup_APIC_eilvt(new, DEFERRED_ERROR_VECTOR,
408 APIC_EILVT_MSG_FIX, 0))
414 static void deferred_error_interrupt_enable(struct cpuinfo_x86 *c)
416 u32 low = 0, high = 0;
417 int def_offset = -1, def_new;
419 if (rdmsr_safe(MSR_CU_DEF_ERR, &low, &high))
422 def_new = (low & MASK_DEF_LVTOFF) >> 4;
423 if (!(low & MASK_DEF_LVTOFF)) {
424 pr_err(FW_BUG "Your BIOS is not setting up LVT offset 0x2 for deferred error IRQs correctly.\n");
425 def_new = DEF_LVT_OFF;
426 low = (low & ~MASK_DEF_LVTOFF) | (DEF_LVT_OFF << 4);
429 def_offset = setup_APIC_deferred_error(def_offset, def_new);
430 if ((def_offset == def_new) &&
431 (deferred_error_int_vector != amd_deferred_error_interrupt))
432 deferred_error_int_vector = amd_deferred_error_interrupt;
435 low = (low & ~MASK_DEF_INT_TYPE) | DEF_INT_TYPE_APIC;
437 wrmsr(MSR_CU_DEF_ERR, low, high);
440 static u32 smca_get_block_address(unsigned int bank, unsigned int block)
445 if (smca_get_bank_type(bank) == SMCA_RESERVED)
449 return MSR_AMD64_SMCA_MCx_MISC(bank);
451 /* Check our cache first: */
452 if (smca_bank_addrs[bank][block] != -1)
453 return smca_bank_addrs[bank][block];
456 * For SMCA enabled processors, BLKPTR field of the first MISC register
457 * (MCx_MISC0) indicates presence of additional MISC regs set (MISC1-4).
459 if (rdmsr_safe(MSR_AMD64_SMCA_MCx_CONFIG(bank), &low, &high))
462 if (!(low & MCI_CONFIG_MCAX))
465 if (!rdmsr_safe(MSR_AMD64_SMCA_MCx_MISC(bank), &low, &high) &&
466 (low & MASK_BLKPTR_LO))
467 addr = MSR_AMD64_SMCA_MCx_MISCy(bank, block - 1);
470 smca_bank_addrs[bank][block] = addr;
474 static u32 get_block_address(u32 current_addr, u32 low, u32 high,
475 unsigned int bank, unsigned int block)
477 u32 addr = 0, offset = 0;
479 if ((bank >= mca_cfg.banks) || (block >= NR_BLOCKS))
483 return smca_get_block_address(bank, block);
485 /* Fall back to method we used for older processors: */
488 addr = msr_ops.misc(bank);
491 offset = ((low & MASK_BLKPTR_LO) >> 21);
493 addr = MCG_XBLK_ADDR + offset;
496 addr = ++current_addr;
502 prepare_threshold_block(unsigned int bank, unsigned int block, u32 addr,
503 int offset, u32 misc_high)
505 unsigned int cpu = smp_processor_id();
506 u32 smca_low, smca_high;
507 struct threshold_block b;
511 per_cpu(bank_map, cpu) |= (1 << bank);
513 memset(&b, 0, sizeof(b));
518 b.interrupt_capable = lvt_interrupt_supported(bank, misc_high);
520 if (!b.interrupt_capable)
523 b.interrupt_enable = 1;
525 if (!mce_flags.smca) {
526 new = (misc_high & MASK_LVTOFF_HI) >> 20;
530 /* Gather LVT offset for thresholding: */
531 if (rdmsr_safe(MSR_CU_DEF_ERR, &smca_low, &smca_high))
534 new = (smca_low & SMCA_THR_LVT_OFF) >> 12;
537 offset = setup_APIC_mce_threshold(offset, new);
539 thresholding_irq_en = true;
542 mce_threshold_block_init(&b, offset);
549 * Turn off MC4_MISC thresholding banks on all family 0x15 models since
550 * they're not supported there.
552 void disable_err_thresholding(struct cpuinfo_x86 *c)
558 0x00000413, /* MC4_MISC0 */
559 0xc0000408, /* MC4_MISC1 */
565 rdmsrl(MSR_K7_HWCR, hwcr);
567 /* McStatusWrEn has to be set */
568 need_toggle = !(hwcr & BIT(18));
571 wrmsrl(MSR_K7_HWCR, hwcr | BIT(18));
573 /* Clear CntP bit safely */
574 for (i = 0; i < ARRAY_SIZE(msrs); i++)
575 msr_clear_bit(msrs[i], 62);
577 /* restore old settings */
579 wrmsrl(MSR_K7_HWCR, hwcr);
582 /* cpu init entry point, called from mce.c with preempt off */
583 void mce_amd_feature_init(struct cpuinfo_x86 *c)
585 u32 low = 0, high = 0, address = 0;
586 unsigned int bank, block, cpu = smp_processor_id();
589 disable_err_thresholding(c);
591 for (bank = 0; bank < mca_cfg.banks; ++bank) {
593 smca_configure(bank, cpu);
595 for (block = 0; block < NR_BLOCKS; ++block) {
596 address = get_block_address(address, low, high, bank, block);
600 if (rdmsr_safe(address, &low, &high))
603 if (!(high & MASK_VALID_HI))
606 if (!(high & MASK_CNTP_HI) ||
607 (high & MASK_LOCKED_HI))
610 offset = prepare_threshold_block(bank, block, address, offset, high);
614 if (mce_flags.succor)
615 deferred_error_interrupt_enable(c);
618 int umc_normaddr_to_sysaddr(u64 norm_addr, u16 nid, u8 umc, u64 *sys_addr)
620 u64 dram_base_addr, dram_limit_addr, dram_hole_base;
621 /* We start from the normalized address */
622 u64 ret_addr = norm_addr;
626 u8 die_id_shift, die_id_mask, socket_id_shift, socket_id_mask;
627 u8 intlv_num_dies, intlv_num_chan, intlv_num_sockets;
628 u8 intlv_addr_sel, intlv_addr_bit;
629 u8 num_intlv_bits, hashed_bit;
630 u8 lgcy_mmio_hole_en, base = 0;
631 u8 cs_mask, cs_id = 0;
632 bool hash_enabled = false;
634 /* Read D18F0x1B4 (DramOffset), check if base 1 is used. */
635 if (amd_df_indirect_read(nid, 0, 0x1B4, umc, &tmp))
638 /* Remove HiAddrOffset from normalized address, if enabled: */
640 u64 hi_addr_offset = (tmp & GENMASK_ULL(31, 20)) << 8;
642 if (norm_addr >= hi_addr_offset) {
643 ret_addr -= hi_addr_offset;
648 /* Read D18F0x110 (DramBaseAddress). */
649 if (amd_df_indirect_read(nid, 0, 0x110 + (8 * base), umc, &tmp))
652 /* Check if address range is valid. */
653 if (!(tmp & BIT(0))) {
654 pr_err("%s: Invalid DramBaseAddress range: 0x%x.\n",
659 lgcy_mmio_hole_en = tmp & BIT(1);
660 intlv_num_chan = (tmp >> 4) & 0xF;
661 intlv_addr_sel = (tmp >> 8) & 0x7;
662 dram_base_addr = (tmp & GENMASK_ULL(31, 12)) << 16;
664 /* {0, 1, 2, 3} map to address bits {8, 9, 10, 11} respectively */
665 if (intlv_addr_sel > 3) {
666 pr_err("%s: Invalid interleave address select %d.\n",
667 __func__, intlv_addr_sel);
671 /* Read D18F0x114 (DramLimitAddress). */
672 if (amd_df_indirect_read(nid, 0, 0x114 + (8 * base), umc, &tmp))
675 intlv_num_sockets = (tmp >> 8) & 0x1;
676 intlv_num_dies = (tmp >> 10) & 0x3;
677 dram_limit_addr = ((tmp & GENMASK_ULL(31, 12)) << 16) | GENMASK_ULL(27, 0);
679 intlv_addr_bit = intlv_addr_sel + 8;
681 /* Re-use intlv_num_chan by setting it equal to log2(#channels) */
682 switch (intlv_num_chan) {
683 case 0: intlv_num_chan = 0; break;
684 case 1: intlv_num_chan = 1; break;
685 case 3: intlv_num_chan = 2; break;
686 case 5: intlv_num_chan = 3; break;
687 case 7: intlv_num_chan = 4; break;
689 case 8: intlv_num_chan = 1;
693 pr_err("%s: Invalid number of interleaved channels %d.\n",
694 __func__, intlv_num_chan);
698 num_intlv_bits = intlv_num_chan;
700 if (intlv_num_dies > 2) {
701 pr_err("%s: Invalid number of interleaved nodes/dies %d.\n",
702 __func__, intlv_num_dies);
706 num_intlv_bits += intlv_num_dies;
708 /* Add a bit if sockets are interleaved. */
709 num_intlv_bits += intlv_num_sockets;
711 /* Assert num_intlv_bits <= 4 */
712 if (num_intlv_bits > 4) {
713 pr_err("%s: Invalid interleave bits %d.\n",
714 __func__, num_intlv_bits);
718 if (num_intlv_bits > 0) {
719 u64 temp_addr_x, temp_addr_i, temp_addr_y;
720 u8 die_id_bit, sock_id_bit, cs_fabric_id;
723 * Read FabricBlockInstanceInformation3_CS[BlockFabricID].
724 * This is the fabric id for this coherent slave. Use
725 * umc/channel# as instance id of the coherent slave
728 if (amd_df_indirect_read(nid, 0, 0x50, umc, &tmp))
731 cs_fabric_id = (tmp >> 8) & 0xFF;
734 /* If interleaved over more than 1 channel: */
735 if (intlv_num_chan) {
736 die_id_bit = intlv_num_chan;
737 cs_mask = (1 << die_id_bit) - 1;
738 cs_id = cs_fabric_id & cs_mask;
741 sock_id_bit = die_id_bit;
743 /* Read D18F1x208 (SystemFabricIdMask). */
744 if (intlv_num_dies || intlv_num_sockets)
745 if (amd_df_indirect_read(nid, 1, 0x208, umc, &tmp))
748 /* If interleaved over more than 1 die. */
749 if (intlv_num_dies) {
750 sock_id_bit = die_id_bit + intlv_num_dies;
751 die_id_shift = (tmp >> 24) & 0xF;
752 die_id_mask = (tmp >> 8) & 0xFF;
754 cs_id |= ((cs_fabric_id & die_id_mask) >> die_id_shift) << die_id_bit;
757 /* If interleaved over more than 1 socket. */
758 if (intlv_num_sockets) {
759 socket_id_shift = (tmp >> 28) & 0xF;
760 socket_id_mask = (tmp >> 16) & 0xFF;
762 cs_id |= ((cs_fabric_id & socket_id_mask) >> socket_id_shift) << sock_id_bit;
766 * The pre-interleaved address consists of XXXXXXIIIYYYYY
767 * where III is the ID for this CS, and XXXXXXYYYYY are the
768 * address bits from the post-interleaved address.
769 * "num_intlv_bits" has been calculated to tell us how many "I"
770 * bits there are. "intlv_addr_bit" tells us how many "Y" bits
771 * there are (where "I" starts).
773 temp_addr_y = ret_addr & GENMASK_ULL(intlv_addr_bit-1, 0);
774 temp_addr_i = (cs_id << intlv_addr_bit);
775 temp_addr_x = (ret_addr & GENMASK_ULL(63, intlv_addr_bit)) << num_intlv_bits;
776 ret_addr = temp_addr_x | temp_addr_i | temp_addr_y;
779 /* Add dram base address */
780 ret_addr += dram_base_addr;
782 /* If legacy MMIO hole enabled */
783 if (lgcy_mmio_hole_en) {
784 if (amd_df_indirect_read(nid, 0, 0x104, umc, &tmp))
787 dram_hole_base = tmp & GENMASK(31, 24);
788 if (ret_addr >= dram_hole_base)
789 ret_addr += (BIT_ULL(32) - dram_hole_base);
793 /* Save some parentheses and grab ls-bit at the end. */
794 hashed_bit = (ret_addr >> 12) ^
800 hashed_bit &= BIT(0);
802 if (hashed_bit != ((ret_addr >> intlv_addr_bit) & BIT(0)))
803 ret_addr ^= BIT(intlv_addr_bit);
806 /* Is calculated system address is above DRAM limit address? */
807 if (ret_addr > dram_limit_addr)
810 *sys_addr = ret_addr;
816 EXPORT_SYMBOL_GPL(umc_normaddr_to_sysaddr);
818 bool amd_mce_is_memory_error(struct mce *m)
820 /* ErrCodeExt[20:16] */
821 u8 xec = (m->status >> 16) & 0x1f;
824 return smca_get_bank_type(m->bank) == SMCA_UMC && xec == 0x0;
826 return m->bank == 4 && xec == 0x8;
829 static void __log_error(unsigned int bank, u64 status, u64 addr, u64 misc)
840 if (m.status & MCI_STATUS_ADDRV) {
844 * Extract [55:<lsb>] where lsb is the least significant
845 * *valid* bit of the address bits.
847 if (mce_flags.smca) {
848 u8 lsb = (m.addr >> 56) & 0x3f;
850 m.addr &= GENMASK_ULL(55, lsb);
854 if (mce_flags.smca) {
855 rdmsrl(MSR_AMD64_SMCA_MCx_IPID(bank), m.ipid);
857 if (m.status & MCI_STATUS_SYNDV)
858 rdmsrl(MSR_AMD64_SMCA_MCx_SYND(bank), m.synd);
864 asmlinkage __visible void __irq_entry smp_deferred_error_interrupt(struct pt_regs *regs)
867 trace_deferred_error_apic_entry(DEFERRED_ERROR_VECTOR);
868 inc_irq_stat(irq_deferred_error_count);
869 deferred_error_int_vector();
870 trace_deferred_error_apic_exit(DEFERRED_ERROR_VECTOR);
875 * Returns true if the logged error is deferred. False, otherwise.
878 _log_error_bank(unsigned int bank, u32 msr_stat, u32 msr_addr, u64 misc)
880 u64 status, addr = 0;
882 rdmsrl(msr_stat, status);
883 if (!(status & MCI_STATUS_VAL))
886 if (status & MCI_STATUS_ADDRV)
887 rdmsrl(msr_addr, addr);
889 __log_error(bank, status, addr, misc);
893 return status & MCI_STATUS_DEFERRED;
897 * We have three scenarios for checking for Deferred errors:
899 * 1) Non-SMCA systems check MCA_STATUS and log error if found.
900 * 2) SMCA systems check MCA_STATUS. If error is found then log it and also
902 * 3) SMCA systems check MCA_DESTAT, if error was not found in MCA_STATUS, and
905 static void log_error_deferred(unsigned int bank)
909 defrd = _log_error_bank(bank, msr_ops.status(bank),
910 msr_ops.addr(bank), 0);
915 /* Clear MCA_DESTAT if we logged the deferred error from MCA_STATUS. */
917 wrmsrl(MSR_AMD64_SMCA_MCx_DESTAT(bank), 0);
922 * Only deferred errors are logged in MCA_DE{STAT,ADDR} so just check
925 _log_error_bank(bank, MSR_AMD64_SMCA_MCx_DESTAT(bank),
926 MSR_AMD64_SMCA_MCx_DEADDR(bank), 0);
929 /* APIC interrupt handler for deferred errors */
930 static void amd_deferred_error_interrupt(void)
934 for (bank = 0; bank < mca_cfg.banks; ++bank)
935 log_error_deferred(bank);
938 static void log_error_thresholding(unsigned int bank, u64 misc)
940 _log_error_bank(bank, msr_ops.status(bank), msr_ops.addr(bank), misc);
943 static void log_and_reset_block(struct threshold_block *block)
945 struct thresh_restart tr;
946 u32 low = 0, high = 0;
951 if (rdmsr_safe(block->address, &low, &high))
954 if (!(high & MASK_OVERFLOW_HI))
957 /* Log the MCE which caused the threshold event. */
958 log_error_thresholding(block->bank, ((u64)high << 32) | low);
960 /* Reset threshold block after logging error. */
961 memset(&tr, 0, sizeof(tr));
963 threshold_restart_bank(&tr);
967 * Threshold interrupt handler will service THRESHOLD_APIC_VECTOR. The interrupt
968 * goes off when error_count reaches threshold_limit.
970 static void amd_threshold_interrupt(void)
972 struct threshold_block *first_block = NULL, *block = NULL, *tmp = NULL;
973 unsigned int bank, cpu = smp_processor_id();
975 for (bank = 0; bank < mca_cfg.banks; ++bank) {
976 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
979 first_block = per_cpu(threshold_banks, cpu)[bank]->blocks;
984 * The first block is also the head of the list. Check it first
985 * before iterating over the rest.
987 log_and_reset_block(first_block);
988 list_for_each_entry_safe(block, tmp, &first_block->miscj, miscj)
989 log_and_reset_block(block);
997 struct threshold_attr {
998 struct attribute attr;
999 ssize_t (*show) (struct threshold_block *, char *);
1000 ssize_t (*store) (struct threshold_block *, const char *, size_t count);
1003 #define SHOW_FIELDS(name) \
1004 static ssize_t show_ ## name(struct threshold_block *b, char *buf) \
1006 return sprintf(buf, "%lu\n", (unsigned long) b->name); \
1008 SHOW_FIELDS(interrupt_enable)
1009 SHOW_FIELDS(threshold_limit)
1012 store_interrupt_enable(struct threshold_block *b, const char *buf, size_t size)
1014 struct thresh_restart tr;
1017 if (!b->interrupt_capable)
1020 if (kstrtoul(buf, 0, &new) < 0)
1023 b->interrupt_enable = !!new;
1025 memset(&tr, 0, sizeof(tr));
1028 smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
1034 store_threshold_limit(struct threshold_block *b, const char *buf, size_t size)
1036 struct thresh_restart tr;
1039 if (kstrtoul(buf, 0, &new) < 0)
1042 if (new > THRESHOLD_MAX)
1043 new = THRESHOLD_MAX;
1047 memset(&tr, 0, sizeof(tr));
1048 tr.old_limit = b->threshold_limit;
1049 b->threshold_limit = new;
1052 smp_call_function_single(b->cpu, threshold_restart_bank, &tr, 1);
1057 static ssize_t show_error_count(struct threshold_block *b, char *buf)
1061 rdmsr_on_cpu(b->cpu, b->address, &lo, &hi);
1063 return sprintf(buf, "%u\n", ((hi & THRESHOLD_MAX) -
1064 (THRESHOLD_MAX - b->threshold_limit)));
1067 static struct threshold_attr error_count = {
1068 .attr = {.name = __stringify(error_count), .mode = 0444 },
1069 .show = show_error_count,
1072 #define RW_ATTR(val) \
1073 static struct threshold_attr val = { \
1074 .attr = {.name = __stringify(val), .mode = 0644 }, \
1075 .show = show_## val, \
1076 .store = store_## val, \
1079 RW_ATTR(interrupt_enable);
1080 RW_ATTR(threshold_limit);
1082 static struct attribute *default_attrs[] = {
1083 &threshold_limit.attr,
1085 NULL, /* possibly interrupt_enable if supported, see below */
1089 #define to_block(k) container_of(k, struct threshold_block, kobj)
1090 #define to_attr(a) container_of(a, struct threshold_attr, attr)
1092 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
1094 struct threshold_block *b = to_block(kobj);
1095 struct threshold_attr *a = to_attr(attr);
1098 ret = a->show ? a->show(b, buf) : -EIO;
1103 static ssize_t store(struct kobject *kobj, struct attribute *attr,
1104 const char *buf, size_t count)
1106 struct threshold_block *b = to_block(kobj);
1107 struct threshold_attr *a = to_attr(attr);
1110 ret = a->store ? a->store(b, buf, count) : -EIO;
1115 static const struct sysfs_ops threshold_ops = {
1120 static void threshold_block_release(struct kobject *kobj);
1122 static struct kobj_type threshold_ktype = {
1123 .sysfs_ops = &threshold_ops,
1124 .default_attrs = default_attrs,
1125 .release = threshold_block_release,
1128 static const char *get_name(unsigned int bank, struct threshold_block *b)
1130 enum smca_bank_types bank_type;
1132 if (!mce_flags.smca) {
1134 return bank4_names(b);
1136 return th_names[bank];
1139 bank_type = smca_get_bank_type(bank);
1140 if (bank_type >= N_SMCA_BANK_TYPES)
1143 if (b && bank_type == SMCA_UMC) {
1144 if (b->block < ARRAY_SIZE(smca_umc_block_names))
1145 return smca_umc_block_names[b->block];
1149 if (smca_banks[bank].hwid->count == 1)
1150 return smca_get_name(bank_type);
1152 snprintf(buf_mcatype, MAX_MCATYPE_NAME_LEN,
1153 "%s_%x", smca_get_name(bank_type),
1154 smca_banks[bank].sysfs_id);
1158 static int allocate_threshold_blocks(unsigned int cpu, struct threshold_bank *tb,
1159 unsigned int bank, unsigned int block,
1162 struct threshold_block *b = NULL;
1166 if ((bank >= mca_cfg.banks) || (block >= NR_BLOCKS))
1169 if (rdmsr_safe_on_cpu(cpu, address, &low, &high))
1172 if (!(high & MASK_VALID_HI)) {
1179 if (!(high & MASK_CNTP_HI) ||
1180 (high & MASK_LOCKED_HI))
1183 b = kzalloc(sizeof(struct threshold_block), GFP_KERNEL);
1190 b->address = address;
1191 b->interrupt_enable = 0;
1192 b->interrupt_capable = lvt_interrupt_supported(bank, high);
1193 b->threshold_limit = THRESHOLD_MAX;
1195 if (b->interrupt_capable) {
1196 threshold_ktype.default_attrs[2] = &interrupt_enable.attr;
1197 b->interrupt_enable = 1;
1199 threshold_ktype.default_attrs[2] = NULL;
1202 INIT_LIST_HEAD(&b->miscj);
1205 list_add(&b->miscj, &tb->blocks->miscj);
1209 err = kobject_init_and_add(&b->kobj, &threshold_ktype, tb->kobj, get_name(bank, b));
1213 address = get_block_address(address, low, high, bank, ++block);
1217 err = allocate_threshold_blocks(cpu, tb, bank, block, address);
1222 kobject_uevent(&b->kobj, KOBJ_ADD);
1228 kobject_put(&b->kobj);
1229 list_del(&b->miscj);
1235 static int __threshold_add_blocks(struct threshold_bank *b)
1237 struct list_head *head = &b->blocks->miscj;
1238 struct threshold_block *pos = NULL;
1239 struct threshold_block *tmp = NULL;
1242 err = kobject_add(&b->blocks->kobj, b->kobj, b->blocks->kobj.name);
1246 list_for_each_entry_safe(pos, tmp, head, miscj) {
1248 err = kobject_add(&pos->kobj, b->kobj, pos->kobj.name);
1250 list_for_each_entry_safe_reverse(pos, tmp, head, miscj)
1251 kobject_del(&pos->kobj);
1259 static int threshold_create_bank(unsigned int cpu, unsigned int bank)
1261 struct device *dev = per_cpu(mce_device, cpu);
1262 struct amd_northbridge *nb = NULL;
1263 struct threshold_bank *b = NULL;
1264 const char *name = get_name(bank, NULL);
1270 if (is_shared_bank(bank)) {
1271 nb = node_to_amd_nb(amd_get_nb_id(cpu));
1273 /* threshold descriptor already initialized on this node? */
1274 if (nb && nb->bank4) {
1277 err = kobject_add(b->kobj, &dev->kobj, name);
1281 per_cpu(threshold_banks, cpu)[bank] = b;
1282 refcount_inc(&b->cpus);
1284 err = __threshold_add_blocks(b);
1290 b = kzalloc(sizeof(struct threshold_bank), GFP_KERNEL);
1296 b->kobj = kobject_create_and_add(name, &dev->kobj);
1302 if (is_shared_bank(bank)) {
1303 refcount_set(&b->cpus, 1);
1305 /* nb is already initialized, see above */
1312 err = allocate_threshold_blocks(cpu, b, bank, 0, msr_ops.misc(bank));
1316 per_cpu(threshold_banks, cpu)[bank] = b;
1327 static void threshold_block_release(struct kobject *kobj)
1329 kfree(to_block(kobj));
1332 static void deallocate_threshold_block(unsigned int cpu, unsigned int bank)
1334 struct threshold_block *pos = NULL;
1335 struct threshold_block *tmp = NULL;
1336 struct threshold_bank *head = per_cpu(threshold_banks, cpu)[bank];
1341 list_for_each_entry_safe(pos, tmp, &head->blocks->miscj, miscj) {
1342 list_del(&pos->miscj);
1343 kobject_put(&pos->kobj);
1346 kobject_put(&head->blocks->kobj);
1349 static void __threshold_remove_blocks(struct threshold_bank *b)
1351 struct threshold_block *pos = NULL;
1352 struct threshold_block *tmp = NULL;
1354 kobject_del(b->kobj);
1356 list_for_each_entry_safe(pos, tmp, &b->blocks->miscj, miscj)
1357 kobject_del(&pos->kobj);
1360 static void threshold_remove_bank(unsigned int cpu, int bank)
1362 struct amd_northbridge *nb;
1363 struct threshold_bank *b;
1365 b = per_cpu(threshold_banks, cpu)[bank];
1372 if (is_shared_bank(bank)) {
1373 if (!refcount_dec_and_test(&b->cpus)) {
1374 __threshold_remove_blocks(b);
1375 per_cpu(threshold_banks, cpu)[bank] = NULL;
1379 * the last CPU on this node using the shared bank is
1380 * going away, remove that bank now.
1382 nb = node_to_amd_nb(amd_get_nb_id(cpu));
1387 deallocate_threshold_block(cpu, bank);
1390 kobject_del(b->kobj);
1391 kobject_put(b->kobj);
1393 per_cpu(threshold_banks, cpu)[bank] = NULL;
1396 int mce_threshold_remove_device(unsigned int cpu)
1400 for (bank = 0; bank < mca_cfg.banks; ++bank) {
1401 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
1403 threshold_remove_bank(cpu, bank);
1405 kfree(per_cpu(threshold_banks, cpu));
1406 per_cpu(threshold_banks, cpu) = NULL;
1410 /* create dir/files for all valid threshold banks */
1411 int mce_threshold_create_device(unsigned int cpu)
1414 struct threshold_bank **bp;
1417 bp = per_cpu(threshold_banks, cpu);
1421 bp = kcalloc(mca_cfg.banks, sizeof(struct threshold_bank *),
1426 per_cpu(threshold_banks, cpu) = bp;
1428 for (bank = 0; bank < mca_cfg.banks; ++bank) {
1429 if (!(per_cpu(bank_map, cpu) & (1 << bank)))
1431 err = threshold_create_bank(cpu, bank);
1437 mce_threshold_remove_device(cpu);
1441 static __init int threshold_init_device(void)
1445 /* to hit CPUs online before the notifier is up */
1446 for_each_online_cpu(lcpu) {
1447 int err = mce_threshold_create_device(lcpu);
1453 if (thresholding_irq_en)
1454 mce_threshold_vector = amd_threshold_interrupt;
1459 * there are 3 funcs which need to be _initcalled in a logic sequence:
1460 * 1. xen_late_init_mcelog
1461 * 2. mcheck_init_device
1462 * 3. threshold_init_device
1464 * xen_late_init_mcelog must register xen_mce_chrdev_device before
1465 * native mce_chrdev_device registration if running under xen platform;
1467 * mcheck_init_device should be inited before threshold_init_device to
1468 * initialize mce_device, otherwise a NULL ptr dereference will cause panic.
1470 * so we use following _initcalls
1471 * 1. device_initcall(xen_late_init_mcelog);
1472 * 2. device_initcall_sync(mcheck_init_device);
1473 * 3. late_initcall(threshold_init_device);
1475 * when running under xen, the initcall order is 1,2,3;
1476 * on baremetal, we skip 1 and we do only 2 and 3.
1478 late_initcall(threshold_init_device);