1
2
3
4
5
6
7
8
9
10
11
12
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/pci.h>
16#include <linux/slab.h>
17#include <linux/irq_work.h>
18#include <linux/llist.h>
19#include <linux/genalloc.h>
20#include <linux/edac.h>
21#include <linux/bits.h>
22#include <linux/io.h>
23#include <asm/mach_traps.h>
24#include <asm/nmi.h>
25#include <asm/mce.h>
26
27#include "edac_mc.h"
28#include "edac_module.h"
29
30#define IGEN6_REVISION "v2.5"
31
32#define EDAC_MOD_STR "igen6_edac"
33#define IGEN6_NMI_NAME "igen6_ibecc"
34
35
36#define igen6_printk(level, fmt, arg...) \
37 edac_printk(level, "igen6", fmt, ##arg)
38
39#define igen6_mc_printk(mci, level, fmt, arg...) \
40 edac_mc_chipset_printk(mci, level, "igen6", fmt, ##arg)
41
42#define GET_BITFIELD(v, lo, hi) (((v) & GENMASK_ULL(hi, lo)) >> (lo))
43
44#define NUM_IMC 2
45#define NUM_CHANNELS 2
46#define NUM_DIMMS 2
47
48#define _4GB BIT_ULL(32)
49
50
51#define TOM_OFFSET 0xa0
52
53#define TOLUD_OFFSET 0xbc
54
55#define CAPID_C_OFFSET 0xec
56#define CAPID_C_IBECC BIT(15)
57
58
59#define CAPID_E_OFFSET 0xf0
60#define CAPID_E_IBECC BIT(12)
61
62
63#define ERRSTS_OFFSET 0xc8
64#define ERRSTS_CE BIT_ULL(6)
65#define ERRSTS_UE BIT_ULL(7)
66
67
68#define ERRCMD_OFFSET 0xca
69#define ERRCMD_CE BIT_ULL(6)
70#define ERRCMD_UE BIT_ULL(7)
71
72
73#define IBECC_BASE (res_cfg->ibecc_base)
74#define IBECC_ACTIVATE_OFFSET IBECC_BASE
75#define IBECC_ACTIVATE_EN BIT(0)
76
77
78#define ECC_ERROR_LOG_OFFSET (IBECC_BASE + res_cfg->ibecc_error_log_offset)
79#define ECC_ERROR_LOG_CE BIT_ULL(62)
80#define ECC_ERROR_LOG_UE BIT_ULL(63)
81#define ECC_ERROR_LOG_ADDR_SHIFT 5
82#define ECC_ERROR_LOG_ADDR(v) GET_BITFIELD(v, 5, 38)
83#define ECC_ERROR_LOG_SYND(v) GET_BITFIELD(v, 46, 61)
84
85
86#define MCHBAR_OFFSET 0x48
87#define MCHBAR_EN BIT_ULL(0)
88#define MCHBAR_BASE(v) (GET_BITFIELD(v, 16, 38) << 16)
89#define MCHBAR_SIZE 0x10000
90
91
92#define IMC_BASE (res_cfg->imc_base)
93#define MAD_INTER_CHANNEL_OFFSET IMC_BASE
94#define MAD_INTER_CHANNEL_DDR_TYPE(v) GET_BITFIELD(v, 0, 2)
95#define MAD_INTER_CHANNEL_ECHM(v) GET_BITFIELD(v, 3, 3)
96#define MAD_INTER_CHANNEL_CH_L_MAP(v) GET_BITFIELD(v, 4, 4)
97#define MAD_INTER_CHANNEL_CH_S_SIZE(v) ((u64)GET_BITFIELD(v, 12, 19) << 29)
98
99
100#define MAD_INTRA_CH0_OFFSET (IMC_BASE + 4)
101#define MAD_INTRA_CH_DIMM_L_MAP(v) GET_BITFIELD(v, 0, 0)
102
103
104#define MAD_DIMM_CH0_OFFSET (IMC_BASE + 0xc)
105#define MAD_DIMM_CH_DIMM_L_SIZE(v) ((u64)GET_BITFIELD(v, 0, 6) << 29)
106#define MAD_DIMM_CH_DLW(v) GET_BITFIELD(v, 7, 8)
107#define MAD_DIMM_CH_DIMM_S_SIZE(v) ((u64)GET_BITFIELD(v, 16, 22) << 29)
108#define MAD_DIMM_CH_DSW(v) GET_BITFIELD(v, 24, 25)
109
110
111#define MAD_MC_HASH_OFFSET (IMC_BASE + 0x1b8)
112#define MAC_MC_HASH_LSB(v) GET_BITFIELD(v, 1, 3)
113
114
115#define CHANNEL_HASH_OFFSET (IMC_BASE + 0x24)
116
117#define CHANNEL_EHASH_OFFSET (IMC_BASE + 0x28)
118#define CHANNEL_HASH_MASK(v) (GET_BITFIELD(v, 6, 19) << 6)
119#define CHANNEL_HASH_LSB_MASK_BIT(v) GET_BITFIELD(v, 24, 26)
120#define CHANNEL_HASH_MODE(v) GET_BITFIELD(v, 28, 28)
121
122
123#define MEM_SLICE_HASH_MASK(v) (GET_BITFIELD(v, 6, 19) << 6)
124#define MEM_SLICE_HASH_LSB_MASK_BIT(v) GET_BITFIELD(v, 24, 26)
125
126static struct res_config {
127 bool machine_check;
128 int num_imc;
129 u32 imc_base;
130 u32 cmf_base;
131 u32 cmf_size;
132 u32 ms_hash_offset;
133 u32 ibecc_base;
134 u32 ibecc_error_log_offset;
135 bool (*ibecc_available)(struct pci_dev *pdev);
136
137 u64 (*err_addr_to_sys_addr)(u64 eaddr, int mc);
138
139 u64 (*err_addr_to_imc_addr)(u64 eaddr, int mc);
140} *res_cfg;
141
142struct igen6_imc {
143 int mc;
144 struct mem_ctl_info *mci;
145 struct pci_dev *pdev;
146 struct device dev;
147 void __iomem *window;
148 u64 size;
149 u64 ch_s_size;
150 int ch_l_map;
151 u64 dimm_s_size[NUM_CHANNELS];
152 u64 dimm_l_size[NUM_CHANNELS];
153 int dimm_l_map[NUM_CHANNELS];
154};
155
156static struct igen6_pvt {
157 struct igen6_imc imc[NUM_IMC];
158 u64 ms_hash;
159 u64 ms_s_size;
160 int ms_l_map;
161} *igen6_pvt;
162
163
164static u32 igen6_tolud;
165
166static u64 igen6_tom;
167
168struct decoded_addr {
169 int mc;
170 u64 imc_addr;
171 u64 sys_addr;
172 int channel_idx;
173 u64 channel_addr;
174 int sub_channel_idx;
175 u64 sub_channel_addr;
176};
177
178struct ecclog_node {
179 struct llist_node llnode;
180 int mc;
181 u64 ecclog;
182};
183
184
185
186
187
188
189
190#define ECCLOG_POOL_SIZE PAGE_SIZE
191static LLIST_HEAD(ecclog_llist);
192static struct gen_pool *ecclog_pool;
193static char ecclog_buf[ECCLOG_POOL_SIZE];
194static struct irq_work ecclog_irq_work;
195static struct work_struct ecclog_work;
196
197
198#define DID_EHL_SKU5 0x4514
199#define DID_EHL_SKU6 0x4528
200#define DID_EHL_SKU7 0x452a
201#define DID_EHL_SKU8 0x4516
202#define DID_EHL_SKU9 0x452c
203#define DID_EHL_SKU10 0x452e
204#define DID_EHL_SKU11 0x4532
205#define DID_EHL_SKU12 0x4518
206#define DID_EHL_SKU13 0x451a
207#define DID_EHL_SKU14 0x4534
208#define DID_EHL_SKU15 0x4536
209
210
211#define DID_ICL_SKU8 0x4581
212#define DID_ICL_SKU10 0x4585
213#define DID_ICL_SKU11 0x4589
214#define DID_ICL_SKU12 0x458d
215
216
217#define DID_TGL_SKU 0x9a14
218
219
220#define DID_ADL_SKU1 0x4601
221#define DID_ADL_SKU2 0x4602
222#define DID_ADL_SKU3 0x4621
223#define DID_ADL_SKU4 0x4641
224
225static bool ehl_ibecc_available(struct pci_dev *pdev)
226{
227 u32 v;
228
229 if (pci_read_config_dword(pdev, CAPID_C_OFFSET, &v))
230 return false;
231
232 return !!(CAPID_C_IBECC & v);
233}
234
235static u64 ehl_err_addr_to_sys_addr(u64 eaddr, int mc)
236{
237 return eaddr;
238}
239
240static u64 ehl_err_addr_to_imc_addr(u64 eaddr, int mc)
241{
242 if (eaddr < igen6_tolud)
243 return eaddr;
244
245 if (igen6_tom <= _4GB)
246 return eaddr + igen6_tolud - _4GB;
247
248 if (eaddr < _4GB)
249 return eaddr + igen6_tolud - igen6_tom;
250
251 return eaddr;
252}
253
254static bool icl_ibecc_available(struct pci_dev *pdev)
255{
256 u32 v;
257
258 if (pci_read_config_dword(pdev, CAPID_C_OFFSET, &v))
259 return false;
260
261 return !(CAPID_C_IBECC & v) &&
262 (boot_cpu_data.x86_stepping >= 1);
263}
264
265static bool tgl_ibecc_available(struct pci_dev *pdev)
266{
267 u32 v;
268
269 if (pci_read_config_dword(pdev, CAPID_E_OFFSET, &v))
270 return false;
271
272 return !(CAPID_E_IBECC & v);
273}
274
275static u64 mem_addr_to_sys_addr(u64 maddr)
276{
277 if (maddr < igen6_tolud)
278 return maddr;
279
280 if (igen6_tom <= _4GB)
281 return maddr - igen6_tolud + _4GB;
282
283 if (maddr < _4GB)
284 return maddr - igen6_tolud + igen6_tom;
285
286 return maddr;
287}
288
289static u64 mem_slice_hash(u64 addr, u64 mask, u64 hash_init, int intlv_bit)
290{
291 u64 hash_addr = addr & mask, hash = hash_init;
292 u64 intlv = (addr >> intlv_bit) & 1;
293 int i;
294
295 for (i = 6; i < 20; i++)
296 hash ^= (hash_addr >> i) & 1;
297
298 return hash ^ intlv;
299}
300
301static u64 tgl_err_addr_to_mem_addr(u64 eaddr, int mc)
302{
303 u64 maddr, hash, mask, ms_s_size;
304 int intlv_bit;
305 u32 ms_hash;
306
307 ms_s_size = igen6_pvt->ms_s_size;
308 if (eaddr >= ms_s_size)
309 return eaddr + ms_s_size;
310
311 ms_hash = igen6_pvt->ms_hash;
312
313 mask = MEM_SLICE_HASH_MASK(ms_hash);
314 intlv_bit = MEM_SLICE_HASH_LSB_MASK_BIT(ms_hash) + 6;
315
316 maddr = GET_BITFIELD(eaddr, intlv_bit, 63) << (intlv_bit + 1) |
317 GET_BITFIELD(eaddr, 0, intlv_bit - 1);
318
319 hash = mem_slice_hash(maddr, mask, mc, intlv_bit);
320
321 return maddr | (hash << intlv_bit);
322}
323
324static u64 tgl_err_addr_to_sys_addr(u64 eaddr, int mc)
325{
326 u64 maddr = tgl_err_addr_to_mem_addr(eaddr, mc);
327
328 return mem_addr_to_sys_addr(maddr);
329}
330
331static u64 tgl_err_addr_to_imc_addr(u64 eaddr, int mc)
332{
333 return eaddr;
334}
335
336static u64 adl_err_addr_to_sys_addr(u64 eaddr, int mc)
337{
338 return mem_addr_to_sys_addr(eaddr);
339}
340
341static u64 adl_err_addr_to_imc_addr(u64 eaddr, int mc)
342{
343 u64 imc_addr, ms_s_size = igen6_pvt->ms_s_size;
344 struct igen6_imc *imc = &igen6_pvt->imc[mc];
345 int intlv_bit;
346 u32 mc_hash;
347
348 if (eaddr >= 2 * ms_s_size)
349 return eaddr - ms_s_size;
350
351 mc_hash = readl(imc->window + MAD_MC_HASH_OFFSET);
352
353 intlv_bit = MAC_MC_HASH_LSB(mc_hash) + 6;
354
355 imc_addr = GET_BITFIELD(eaddr, intlv_bit + 1, 63) << intlv_bit |
356 GET_BITFIELD(eaddr, 0, intlv_bit - 1);
357
358 return imc_addr;
359}
360
361static struct res_config ehl_cfg = {
362 .num_imc = 1,
363 .imc_base = 0x5000,
364 .ibecc_base = 0xdc00,
365 .ibecc_available = ehl_ibecc_available,
366 .ibecc_error_log_offset = 0x170,
367 .err_addr_to_sys_addr = ehl_err_addr_to_sys_addr,
368 .err_addr_to_imc_addr = ehl_err_addr_to_imc_addr,
369};
370
371static struct res_config icl_cfg = {
372 .num_imc = 1,
373 .imc_base = 0x5000,
374 .ibecc_base = 0xd800,
375 .ibecc_error_log_offset = 0x170,
376 .ibecc_available = icl_ibecc_available,
377 .err_addr_to_sys_addr = ehl_err_addr_to_sys_addr,
378 .err_addr_to_imc_addr = ehl_err_addr_to_imc_addr,
379};
380
381static struct res_config tgl_cfg = {
382 .machine_check = true,
383 .num_imc = 2,
384 .imc_base = 0x5000,
385 .cmf_base = 0x11000,
386 .cmf_size = 0x800,
387 .ms_hash_offset = 0xac,
388 .ibecc_base = 0xd400,
389 .ibecc_error_log_offset = 0x170,
390 .ibecc_available = tgl_ibecc_available,
391 .err_addr_to_sys_addr = tgl_err_addr_to_sys_addr,
392 .err_addr_to_imc_addr = tgl_err_addr_to_imc_addr,
393};
394
395static struct res_config adl_cfg = {
396 .machine_check = true,
397 .num_imc = 2,
398 .imc_base = 0xd800,
399 .ibecc_base = 0xd400,
400 .ibecc_error_log_offset = 0x68,
401 .ibecc_available = tgl_ibecc_available,
402 .err_addr_to_sys_addr = adl_err_addr_to_sys_addr,
403 .err_addr_to_imc_addr = adl_err_addr_to_imc_addr,
404};
405
406static const struct pci_device_id igen6_pci_tbl[] = {
407 { PCI_VDEVICE(INTEL, DID_EHL_SKU5), (kernel_ulong_t)&ehl_cfg },
408 { PCI_VDEVICE(INTEL, DID_EHL_SKU6), (kernel_ulong_t)&ehl_cfg },
409 { PCI_VDEVICE(INTEL, DID_EHL_SKU7), (kernel_ulong_t)&ehl_cfg },
410 { PCI_VDEVICE(INTEL, DID_EHL_SKU8), (kernel_ulong_t)&ehl_cfg },
411 { PCI_VDEVICE(INTEL, DID_EHL_SKU9), (kernel_ulong_t)&ehl_cfg },
412 { PCI_VDEVICE(INTEL, DID_EHL_SKU10), (kernel_ulong_t)&ehl_cfg },
413 { PCI_VDEVICE(INTEL, DID_EHL_SKU11), (kernel_ulong_t)&ehl_cfg },
414 { PCI_VDEVICE(INTEL, DID_EHL_SKU12), (kernel_ulong_t)&ehl_cfg },
415 { PCI_VDEVICE(INTEL, DID_EHL_SKU13), (kernel_ulong_t)&ehl_cfg },
416 { PCI_VDEVICE(INTEL, DID_EHL_SKU14), (kernel_ulong_t)&ehl_cfg },
417 { PCI_VDEVICE(INTEL, DID_EHL_SKU15), (kernel_ulong_t)&ehl_cfg },
418 { PCI_VDEVICE(INTEL, DID_ICL_SKU8), (kernel_ulong_t)&icl_cfg },
419 { PCI_VDEVICE(INTEL, DID_ICL_SKU10), (kernel_ulong_t)&icl_cfg },
420 { PCI_VDEVICE(INTEL, DID_ICL_SKU11), (kernel_ulong_t)&icl_cfg },
421 { PCI_VDEVICE(INTEL, DID_ICL_SKU12), (kernel_ulong_t)&icl_cfg },
422 { PCI_VDEVICE(INTEL, DID_TGL_SKU), (kernel_ulong_t)&tgl_cfg },
423 { PCI_VDEVICE(INTEL, DID_ADL_SKU1), (kernel_ulong_t)&adl_cfg },
424 { PCI_VDEVICE(INTEL, DID_ADL_SKU2), (kernel_ulong_t)&adl_cfg },
425 { PCI_VDEVICE(INTEL, DID_ADL_SKU3), (kernel_ulong_t)&adl_cfg },
426 { PCI_VDEVICE(INTEL, DID_ADL_SKU4), (kernel_ulong_t)&adl_cfg },
427 { },
428};
429MODULE_DEVICE_TABLE(pci, igen6_pci_tbl);
430
431static enum dev_type get_width(int dimm_l, u32 mad_dimm)
432{
433 u32 w = dimm_l ? MAD_DIMM_CH_DLW(mad_dimm) :
434 MAD_DIMM_CH_DSW(mad_dimm);
435
436 switch (w) {
437 case 0:
438 return DEV_X8;
439 case 1:
440 return DEV_X16;
441 case 2:
442 return DEV_X32;
443 default:
444 return DEV_UNKNOWN;
445 }
446}
447
448static enum mem_type get_memory_type(u32 mad_inter)
449{
450 u32 t = MAD_INTER_CHANNEL_DDR_TYPE(mad_inter);
451
452 switch (t) {
453 case 0:
454 return MEM_DDR4;
455 case 1:
456 return MEM_DDR3;
457 case 2:
458 return MEM_LPDDR3;
459 case 3:
460 return MEM_LPDDR4;
461 case 4:
462 return MEM_WIO2;
463 default:
464 return MEM_UNKNOWN;
465 }
466}
467
468static int decode_chan_idx(u64 addr, u64 mask, int intlv_bit)
469{
470 u64 hash_addr = addr & mask, hash = 0;
471 u64 intlv = (addr >> intlv_bit) & 1;
472 int i;
473
474 for (i = 6; i < 20; i++)
475 hash ^= (hash_addr >> i) & 1;
476
477 return (int)hash ^ intlv;
478}
479
480static u64 decode_channel_addr(u64 addr, int intlv_bit)
481{
482 u64 channel_addr;
483
484
485 channel_addr = GET_BITFIELD(addr, intlv_bit + 1, 63) << intlv_bit;
486 channel_addr |= GET_BITFIELD(addr, 0, intlv_bit - 1);
487
488 return channel_addr;
489}
490
491static void decode_addr(u64 addr, u32 hash, u64 s_size, int l_map,
492 int *idx, u64 *sub_addr)
493{
494 int intlv_bit = CHANNEL_HASH_LSB_MASK_BIT(hash) + 6;
495
496 if (addr > 2 * s_size) {
497 *sub_addr = addr - s_size;
498 *idx = l_map;
499 return;
500 }
501
502 if (CHANNEL_HASH_MODE(hash)) {
503 *sub_addr = decode_channel_addr(addr, intlv_bit);
504 *idx = decode_chan_idx(addr, CHANNEL_HASH_MASK(hash), intlv_bit);
505 } else {
506 *sub_addr = decode_channel_addr(addr, 6);
507 *idx = GET_BITFIELD(addr, 6, 6);
508 }
509}
510
511static int igen6_decode(struct decoded_addr *res)
512{
513 struct igen6_imc *imc = &igen6_pvt->imc[res->mc];
514 u64 addr = res->imc_addr, sub_addr, s_size;
515 int idx, l_map;
516 u32 hash;
517
518 if (addr >= igen6_tom) {
519 edac_dbg(0, "Address 0x%llx out of range\n", addr);
520 return -EINVAL;
521 }
522
523
524 hash = readl(imc->window + CHANNEL_HASH_OFFSET);
525 s_size = imc->ch_s_size;
526 l_map = imc->ch_l_map;
527 decode_addr(addr, hash, s_size, l_map, &idx, &sub_addr);
528 res->channel_idx = idx;
529 res->channel_addr = sub_addr;
530
531
532 hash = readl(imc->window + CHANNEL_EHASH_OFFSET);
533 s_size = imc->dimm_s_size[idx];
534 l_map = imc->dimm_l_map[idx];
535 decode_addr(res->channel_addr, hash, s_size, l_map, &idx, &sub_addr);
536 res->sub_channel_idx = idx;
537 res->sub_channel_addr = sub_addr;
538
539 return 0;
540}
541
542static void igen6_output_error(struct decoded_addr *res,
543 struct mem_ctl_info *mci, u64 ecclog)
544{
545 enum hw_event_mc_err_type type = ecclog & ECC_ERROR_LOG_UE ?
546 HW_EVENT_ERR_UNCORRECTED :
547 HW_EVENT_ERR_CORRECTED;
548
549 edac_mc_handle_error(type, mci, 1,
550 res->sys_addr >> PAGE_SHIFT,
551 res->sys_addr & ~PAGE_MASK,
552 ECC_ERROR_LOG_SYND(ecclog),
553 res->channel_idx, res->sub_channel_idx,
554 -1, "", "");
555}
556
557static struct gen_pool *ecclog_gen_pool_create(void)
558{
559 struct gen_pool *pool;
560
561 pool = gen_pool_create(ilog2(sizeof(struct ecclog_node)), -1);
562 if (!pool)
563 return NULL;
564
565 if (gen_pool_add(pool, (unsigned long)ecclog_buf, ECCLOG_POOL_SIZE, -1)) {
566 gen_pool_destroy(pool);
567 return NULL;
568 }
569
570 return pool;
571}
572
573static int ecclog_gen_pool_add(int mc, u64 ecclog)
574{
575 struct ecclog_node *node;
576
577 node = (void *)gen_pool_alloc(ecclog_pool, sizeof(*node));
578 if (!node)
579 return -ENOMEM;
580
581 node->mc = mc;
582 node->ecclog = ecclog;
583 llist_add(&node->llnode, &ecclog_llist);
584
585 return 0;
586}
587
588
589
590
591
592
593
594
595static u64 ecclog_read_and_clear(struct igen6_imc *imc)
596{
597 u64 ecclog = readq(imc->window + ECC_ERROR_LOG_OFFSET);
598
599 if (ecclog & (ECC_ERROR_LOG_CE | ECC_ERROR_LOG_UE)) {
600
601 writeq(ecclog, imc->window + ECC_ERROR_LOG_OFFSET);
602 return ecclog;
603 }
604
605 return 0;
606}
607
608static void errsts_clear(struct igen6_imc *imc)
609{
610 u16 errsts;
611
612 if (pci_read_config_word(imc->pdev, ERRSTS_OFFSET, &errsts)) {
613 igen6_printk(KERN_ERR, "Failed to read ERRSTS\n");
614 return;
615 }
616
617
618 if (errsts & (ERRSTS_CE | ERRSTS_UE))
619 pci_write_config_word(imc->pdev, ERRSTS_OFFSET, errsts);
620}
621
622static int errcmd_enable_error_reporting(bool enable)
623{
624 struct igen6_imc *imc = &igen6_pvt->imc[0];
625 u16 errcmd;
626 int rc;
627
628 rc = pci_read_config_word(imc->pdev, ERRCMD_OFFSET, &errcmd);
629 if (rc)
630 return rc;
631
632 if (enable)
633 errcmd |= ERRCMD_CE | ERRSTS_UE;
634 else
635 errcmd &= ~(ERRCMD_CE | ERRSTS_UE);
636
637 rc = pci_write_config_word(imc->pdev, ERRCMD_OFFSET, errcmd);
638 if (rc)
639 return rc;
640
641 return 0;
642}
643
644static int ecclog_handler(void)
645{
646 struct igen6_imc *imc;
647 int i, n = 0;
648 u64 ecclog;
649
650 for (i = 0; i < res_cfg->num_imc; i++) {
651 imc = &igen6_pvt->imc[i];
652
653
654
655 ecclog = ecclog_read_and_clear(imc);
656 if (!ecclog)
657 continue;
658
659 if (!ecclog_gen_pool_add(i, ecclog))
660 irq_work_queue(&ecclog_irq_work);
661
662 n++;
663 }
664
665 return n;
666}
667
668static void ecclog_work_cb(struct work_struct *work)
669{
670 struct ecclog_node *node, *tmp;
671 struct mem_ctl_info *mci;
672 struct llist_node *head;
673 struct decoded_addr res;
674 u64 eaddr;
675
676 head = llist_del_all(&ecclog_llist);
677 if (!head)
678 return;
679
680 llist_for_each_entry_safe(node, tmp, head, llnode) {
681 memset(&res, 0, sizeof(res));
682 eaddr = ECC_ERROR_LOG_ADDR(node->ecclog) <<
683 ECC_ERROR_LOG_ADDR_SHIFT;
684 res.mc = node->mc;
685 res.sys_addr = res_cfg->err_addr_to_sys_addr(eaddr, res.mc);
686 res.imc_addr = res_cfg->err_addr_to_imc_addr(eaddr, res.mc);
687
688 mci = igen6_pvt->imc[res.mc].mci;
689
690 edac_dbg(2, "MC %d, ecclog = 0x%llx\n", node->mc, node->ecclog);
691 igen6_mc_printk(mci, KERN_DEBUG, "HANDLING IBECC MEMORY ERROR\n");
692 igen6_mc_printk(mci, KERN_DEBUG, "ADDR 0x%llx ", res.sys_addr);
693
694 if (!igen6_decode(&res))
695 igen6_output_error(&res, mci, node->ecclog);
696
697 gen_pool_free(ecclog_pool, (unsigned long)node, sizeof(*node));
698 }
699}
700
701static void ecclog_irq_work_cb(struct irq_work *irq_work)
702{
703 int i;
704
705 for (i = 0; i < res_cfg->num_imc; i++)
706 errsts_clear(&igen6_pvt->imc[i]);
707
708 if (!llist_empty(&ecclog_llist))
709 schedule_work(&ecclog_work);
710}
711
712static int ecclog_nmi_handler(unsigned int cmd, struct pt_regs *regs)
713{
714 unsigned char reason;
715
716 if (!ecclog_handler())
717 return NMI_DONE;
718
719
720
721
722
723
724
725
726 reason = x86_platform.get_nmi_reason() & NMI_REASON_CLEAR_MASK;
727 reason |= NMI_REASON_CLEAR_SERR;
728 outb(reason, NMI_REASON_PORT);
729 reason &= ~NMI_REASON_CLEAR_SERR;
730 outb(reason, NMI_REASON_PORT);
731
732 return NMI_HANDLED;
733}
734
735static int ecclog_mce_handler(struct notifier_block *nb, unsigned long val,
736 void *data)
737{
738 struct mce *mce = (struct mce *)data;
739 char *type;
740
741 if (mce->kflags & MCE_HANDLED_CEC)
742 return NOTIFY_DONE;
743
744
745
746
747
748
749 if ((mce->status & 0xefff) >> 7 != 1)
750 return NOTIFY_DONE;
751
752 if (mce->mcgstatus & MCG_STATUS_MCIP)
753 type = "Exception";
754 else
755 type = "Event";
756
757 edac_dbg(0, "CPU %d: Machine Check %s: 0x%llx Bank %d: 0x%llx\n",
758 mce->extcpu, type, mce->mcgstatus,
759 mce->bank, mce->status);
760 edac_dbg(0, "TSC 0x%llx\n", mce->tsc);
761 edac_dbg(0, "ADDR 0x%llx\n", mce->addr);
762 edac_dbg(0, "MISC 0x%llx\n", mce->misc);
763 edac_dbg(0, "PROCESSOR %u:0x%x TIME %llu SOCKET %u APIC 0x%x\n",
764 mce->cpuvendor, mce->cpuid, mce->time,
765 mce->socketid, mce->apicid);
766
767
768
769
770
771
772
773 if (!ecclog_handler())
774 return NOTIFY_DONE;
775
776 mce->kflags |= MCE_HANDLED_EDAC;
777
778 return NOTIFY_DONE;
779}
780
781static struct notifier_block ecclog_mce_dec = {
782 .notifier_call = ecclog_mce_handler,
783 .priority = MCE_PRIO_EDAC,
784};
785
786static bool igen6_check_ecc(struct igen6_imc *imc)
787{
788 u32 activate = readl(imc->window + IBECC_ACTIVATE_OFFSET);
789
790 return !!(activate & IBECC_ACTIVATE_EN);
791}
792
793static int igen6_get_dimm_config(struct mem_ctl_info *mci)
794{
795 struct igen6_imc *imc = mci->pvt_info;
796 u32 mad_inter, mad_intra, mad_dimm;
797 int i, j, ndimms, mc = imc->mc;
798 struct dimm_info *dimm;
799 enum mem_type mtype;
800 enum dev_type dtype;
801 u64 dsize;
802 bool ecc;
803
804 edac_dbg(2, "\n");
805
806 mad_inter = readl(imc->window + MAD_INTER_CHANNEL_OFFSET);
807 mtype = get_memory_type(mad_inter);
808 ecc = igen6_check_ecc(imc);
809 imc->ch_s_size = MAD_INTER_CHANNEL_CH_S_SIZE(mad_inter);
810 imc->ch_l_map = MAD_INTER_CHANNEL_CH_L_MAP(mad_inter);
811
812 for (i = 0; i < NUM_CHANNELS; i++) {
813 mad_intra = readl(imc->window + MAD_INTRA_CH0_OFFSET + i * 4);
814 mad_dimm = readl(imc->window + MAD_DIMM_CH0_OFFSET + i * 4);
815
816 imc->dimm_l_size[i] = MAD_DIMM_CH_DIMM_L_SIZE(mad_dimm);
817 imc->dimm_s_size[i] = MAD_DIMM_CH_DIMM_S_SIZE(mad_dimm);
818 imc->dimm_l_map[i] = MAD_INTRA_CH_DIMM_L_MAP(mad_intra);
819 imc->size += imc->dimm_s_size[i];
820 imc->size += imc->dimm_l_size[i];
821 ndimms = 0;
822
823 for (j = 0; j < NUM_DIMMS; j++) {
824 dimm = edac_get_dimm(mci, i, j, 0);
825
826 if (j ^ imc->dimm_l_map[i]) {
827 dtype = get_width(0, mad_dimm);
828 dsize = imc->dimm_s_size[i];
829 } else {
830 dtype = get_width(1, mad_dimm);
831 dsize = imc->dimm_l_size[i];
832 }
833
834 if (!dsize)
835 continue;
836
837 dimm->grain = 64;
838 dimm->mtype = mtype;
839 dimm->dtype = dtype;
840 dimm->nr_pages = MiB_TO_PAGES(dsize >> 20);
841 dimm->edac_mode = EDAC_SECDED;
842 snprintf(dimm->label, sizeof(dimm->label),
843 "MC#%d_Chan#%d_DIMM#%d", mc, i, j);
844 edac_dbg(0, "MC %d, Channel %d, DIMM %d, Size %llu MiB (%u pages)\n",
845 mc, i, j, dsize >> 20, dimm->nr_pages);
846
847 ndimms++;
848 }
849
850 if (ndimms && !ecc) {
851 igen6_printk(KERN_ERR, "MC%d In-Band ECC is disabled\n", mc);
852 return -ENODEV;
853 }
854 }
855
856 edac_dbg(0, "MC %d, total size %llu MiB\n", mc, imc->size >> 20);
857
858 return 0;
859}
860
861#ifdef CONFIG_EDAC_DEBUG
862
863static u64 igen6_touud;
864#define TOUUD_OFFSET 0xa8
865
866static void igen6_reg_dump(struct igen6_imc *imc)
867{
868 int i;
869
870 edac_dbg(2, "CHANNEL_HASH : 0x%x\n",
871 readl(imc->window + CHANNEL_HASH_OFFSET));
872 edac_dbg(2, "CHANNEL_EHASH : 0x%x\n",
873 readl(imc->window + CHANNEL_EHASH_OFFSET));
874 edac_dbg(2, "MAD_INTER_CHANNEL: 0x%x\n",
875 readl(imc->window + MAD_INTER_CHANNEL_OFFSET));
876 edac_dbg(2, "ECC_ERROR_LOG : 0x%llx\n",
877 readq(imc->window + ECC_ERROR_LOG_OFFSET));
878
879 for (i = 0; i < NUM_CHANNELS; i++) {
880 edac_dbg(2, "MAD_INTRA_CH%d : 0x%x\n", i,
881 readl(imc->window + MAD_INTRA_CH0_OFFSET + i * 4));
882 edac_dbg(2, "MAD_DIMM_CH%d : 0x%x\n", i,
883 readl(imc->window + MAD_DIMM_CH0_OFFSET + i * 4));
884 }
885 edac_dbg(2, "TOLUD : 0x%x", igen6_tolud);
886 edac_dbg(2, "TOUUD : 0x%llx", igen6_touud);
887 edac_dbg(2, "TOM : 0x%llx", igen6_tom);
888}
889
890static struct dentry *igen6_test;
891
892static int debugfs_u64_set(void *data, u64 val)
893{
894 u64 ecclog;
895
896 if ((val >= igen6_tolud && val < _4GB) || val >= igen6_touud) {
897 edac_dbg(0, "Address 0x%llx out of range\n", val);
898 return 0;
899 }
900
901 pr_warn_once("Fake error to 0x%llx injected via debugfs\n", val);
902
903 val >>= ECC_ERROR_LOG_ADDR_SHIFT;
904 ecclog = (val << ECC_ERROR_LOG_ADDR_SHIFT) | ECC_ERROR_LOG_CE;
905
906 if (!ecclog_gen_pool_add(0, ecclog))
907 irq_work_queue(&ecclog_irq_work);
908
909 return 0;
910}
911DEFINE_SIMPLE_ATTRIBUTE(fops_u64_wo, NULL, debugfs_u64_set, "%llu\n");
912
913static void igen6_debug_setup(void)
914{
915 igen6_test = edac_debugfs_create_dir("igen6_test");
916 if (!igen6_test)
917 return;
918
919 if (!edac_debugfs_create_file("addr", 0200, igen6_test,
920 NULL, &fops_u64_wo)) {
921 debugfs_remove(igen6_test);
922 igen6_test = NULL;
923 }
924}
925
926static void igen6_debug_teardown(void)
927{
928 debugfs_remove_recursive(igen6_test);
929}
930#else
931static void igen6_reg_dump(struct igen6_imc *imc) {}
932static void igen6_debug_setup(void) {}
933static void igen6_debug_teardown(void) {}
934#endif
935
936static int igen6_pci_setup(struct pci_dev *pdev, u64 *mchbar)
937{
938 union {
939 u64 v;
940 struct {
941 u32 v_lo;
942 u32 v_hi;
943 };
944 } u;
945
946 edac_dbg(2, "\n");
947
948 if (!res_cfg->ibecc_available(pdev)) {
949 edac_dbg(2, "No In-Band ECC IP\n");
950 goto fail;
951 }
952
953 if (pci_read_config_dword(pdev, TOLUD_OFFSET, &igen6_tolud)) {
954 igen6_printk(KERN_ERR, "Failed to read TOLUD\n");
955 goto fail;
956 }
957
958 igen6_tolud &= GENMASK(31, 20);
959
960 if (pci_read_config_dword(pdev, TOM_OFFSET, &u.v_lo)) {
961 igen6_printk(KERN_ERR, "Failed to read lower TOM\n");
962 goto fail;
963 }
964
965 if (pci_read_config_dword(pdev, TOM_OFFSET + 4, &u.v_hi)) {
966 igen6_printk(KERN_ERR, "Failed to read upper TOM\n");
967 goto fail;
968 }
969
970 igen6_tom = u.v & GENMASK_ULL(38, 20);
971
972 if (pci_read_config_dword(pdev, MCHBAR_OFFSET, &u.v_lo)) {
973 igen6_printk(KERN_ERR, "Failed to read lower MCHBAR\n");
974 goto fail;
975 }
976
977 if (pci_read_config_dword(pdev, MCHBAR_OFFSET + 4, &u.v_hi)) {
978 igen6_printk(KERN_ERR, "Failed to read upper MCHBAR\n");
979 goto fail;
980 }
981
982 if (!(u.v & MCHBAR_EN)) {
983 igen6_printk(KERN_ERR, "MCHBAR is disabled\n");
984 goto fail;
985 }
986
987 *mchbar = MCHBAR_BASE(u.v);
988
989#ifdef CONFIG_EDAC_DEBUG
990 if (pci_read_config_dword(pdev, TOUUD_OFFSET, &u.v_lo))
991 edac_dbg(2, "Failed to read lower TOUUD\n");
992 else if (pci_read_config_dword(pdev, TOUUD_OFFSET + 4, &u.v_hi))
993 edac_dbg(2, "Failed to read upper TOUUD\n");
994 else
995 igen6_touud = u.v & GENMASK_ULL(38, 20);
996#endif
997
998 return 0;
999fail:
1000 return -ENODEV;
1001}
1002
1003static int igen6_register_mci(int mc, u64 mchbar, struct pci_dev *pdev)
1004{
1005 struct edac_mc_layer layers[2];
1006 struct mem_ctl_info *mci;
1007 struct igen6_imc *imc;
1008 void __iomem *window;
1009 int rc;
1010
1011 edac_dbg(2, "\n");
1012
1013 mchbar += mc * MCHBAR_SIZE;
1014 window = ioremap(mchbar, MCHBAR_SIZE);
1015 if (!window) {
1016 igen6_printk(KERN_ERR, "Failed to ioremap 0x%llx\n", mchbar);
1017 return -ENODEV;
1018 }
1019
1020 layers[0].type = EDAC_MC_LAYER_CHANNEL;
1021 layers[0].size = NUM_CHANNELS;
1022 layers[0].is_virt_csrow = false;
1023 layers[1].type = EDAC_MC_LAYER_SLOT;
1024 layers[1].size = NUM_DIMMS;
1025 layers[1].is_virt_csrow = true;
1026
1027 mci = edac_mc_alloc(mc, ARRAY_SIZE(layers), layers, 0);
1028 if (!mci) {
1029 rc = -ENOMEM;
1030 goto fail;
1031 }
1032
1033 mci->ctl_name = kasprintf(GFP_KERNEL, "Intel_client_SoC MC#%d", mc);
1034 if (!mci->ctl_name) {
1035 rc = -ENOMEM;
1036 goto fail2;
1037 }
1038
1039 mci->mtype_cap = MEM_FLAG_LPDDR4 | MEM_FLAG_DDR4;
1040 mci->edac_ctl_cap = EDAC_FLAG_SECDED;
1041 mci->edac_cap = EDAC_FLAG_SECDED;
1042 mci->mod_name = EDAC_MOD_STR;
1043 mci->dev_name = pci_name(pdev);
1044 mci->pvt_info = &igen6_pvt->imc[mc];
1045
1046 imc = mci->pvt_info;
1047 device_initialize(&imc->dev);
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058 mci->pdev = mc ? &imc->dev : &pdev->dev;
1059 imc->mc = mc;
1060 imc->pdev = pdev;
1061 imc->window = window;
1062
1063 igen6_reg_dump(imc);
1064
1065 rc = igen6_get_dimm_config(mci);
1066 if (rc)
1067 goto fail3;
1068
1069 rc = edac_mc_add_mc(mci);
1070 if (rc) {
1071 igen6_printk(KERN_ERR, "Failed to register mci#%d\n", mc);
1072 goto fail3;
1073 }
1074
1075 imc->mci = mci;
1076 return 0;
1077fail3:
1078 kfree(mci->ctl_name);
1079fail2:
1080 edac_mc_free(mci);
1081fail:
1082 iounmap(window);
1083 return rc;
1084}
1085
1086static void igen6_unregister_mcis(void)
1087{
1088 struct mem_ctl_info *mci;
1089 struct igen6_imc *imc;
1090 int i;
1091
1092 edac_dbg(2, "\n");
1093
1094 for (i = 0; i < res_cfg->num_imc; i++) {
1095 imc = &igen6_pvt->imc[i];
1096 mci = imc->mci;
1097 if (!mci)
1098 continue;
1099
1100 edac_mc_del_mc(mci->pdev);
1101 kfree(mci->ctl_name);
1102 edac_mc_free(mci);
1103 iounmap(imc->window);
1104 }
1105}
1106
1107static int igen6_mem_slice_setup(u64 mchbar)
1108{
1109 struct igen6_imc *imc = &igen6_pvt->imc[0];
1110 u64 base = mchbar + res_cfg->cmf_base;
1111 u32 offset = res_cfg->ms_hash_offset;
1112 u32 size = res_cfg->cmf_size;
1113 u64 ms_s_size, ms_hash;
1114 void __iomem *cmf;
1115 int ms_l_map;
1116
1117 edac_dbg(2, "\n");
1118
1119 if (imc[0].size < imc[1].size) {
1120 ms_s_size = imc[0].size;
1121 ms_l_map = 1;
1122 } else {
1123 ms_s_size = imc[1].size;
1124 ms_l_map = 0;
1125 }
1126
1127 igen6_pvt->ms_s_size = ms_s_size;
1128 igen6_pvt->ms_l_map = ms_l_map;
1129
1130 edac_dbg(0, "ms_s_size: %llu MiB, ms_l_map %d\n",
1131 ms_s_size >> 20, ms_l_map);
1132
1133 if (!size)
1134 return 0;
1135
1136 cmf = ioremap(base, size);
1137 if (!cmf) {
1138 igen6_printk(KERN_ERR, "Failed to ioremap cmf 0x%llx\n", base);
1139 return -ENODEV;
1140 }
1141
1142 ms_hash = readq(cmf + offset);
1143 igen6_pvt->ms_hash = ms_hash;
1144
1145 edac_dbg(0, "MEM_SLICE_HASH: 0x%llx\n", ms_hash);
1146
1147 iounmap(cmf);
1148
1149 return 0;
1150}
1151
1152static int register_err_handler(void)
1153{
1154 int rc;
1155
1156 if (res_cfg->machine_check) {
1157 mce_register_decode_chain(&ecclog_mce_dec);
1158 return 0;
1159 }
1160
1161 rc = register_nmi_handler(NMI_SERR, ecclog_nmi_handler,
1162 0, IGEN6_NMI_NAME);
1163 if (rc) {
1164 igen6_printk(KERN_ERR, "Failed to register NMI handler\n");
1165 return rc;
1166 }
1167
1168 return 0;
1169}
1170
1171static void unregister_err_handler(void)
1172{
1173 if (res_cfg->machine_check) {
1174 mce_unregister_decode_chain(&ecclog_mce_dec);
1175 return;
1176 }
1177
1178 unregister_nmi_handler(NMI_SERR, IGEN6_NMI_NAME);
1179}
1180
1181static int igen6_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1182{
1183 u64 mchbar;
1184 int i, rc;
1185
1186 edac_dbg(2, "\n");
1187
1188 igen6_pvt = kzalloc(sizeof(*igen6_pvt), GFP_KERNEL);
1189 if (!igen6_pvt)
1190 return -ENOMEM;
1191
1192 res_cfg = (struct res_config *)ent->driver_data;
1193
1194 rc = igen6_pci_setup(pdev, &mchbar);
1195 if (rc)
1196 goto fail;
1197
1198 for (i = 0; i < res_cfg->num_imc; i++) {
1199 rc = igen6_register_mci(i, mchbar, pdev);
1200 if (rc)
1201 goto fail2;
1202 }
1203
1204 if (res_cfg->num_imc > 1) {
1205 rc = igen6_mem_slice_setup(mchbar);
1206 if (rc)
1207 goto fail2;
1208 }
1209
1210 ecclog_pool = ecclog_gen_pool_create();
1211 if (!ecclog_pool) {
1212 rc = -ENOMEM;
1213 goto fail2;
1214 }
1215
1216 INIT_WORK(&ecclog_work, ecclog_work_cb);
1217 init_irq_work(&ecclog_irq_work, ecclog_irq_work_cb);
1218
1219
1220 ecclog_handler();
1221
1222 rc = register_err_handler();
1223 if (rc)
1224 goto fail3;
1225
1226
1227 rc = errcmd_enable_error_reporting(true);
1228 if (rc) {
1229 igen6_printk(KERN_ERR, "Failed to enable error reporting\n");
1230 goto fail4;
1231 }
1232
1233 igen6_debug_setup();
1234 return 0;
1235fail4:
1236 unregister_nmi_handler(NMI_SERR, IGEN6_NMI_NAME);
1237fail3:
1238 gen_pool_destroy(ecclog_pool);
1239fail2:
1240 igen6_unregister_mcis();
1241fail:
1242 kfree(igen6_pvt);
1243 return rc;
1244}
1245
1246static void igen6_remove(struct pci_dev *pdev)
1247{
1248 edac_dbg(2, "\n");
1249
1250 igen6_debug_teardown();
1251 errcmd_enable_error_reporting(false);
1252 unregister_err_handler();
1253 irq_work_sync(&ecclog_irq_work);
1254 flush_work(&ecclog_work);
1255 gen_pool_destroy(ecclog_pool);
1256 igen6_unregister_mcis();
1257 kfree(igen6_pvt);
1258}
1259
1260static struct pci_driver igen6_driver = {
1261 .name = EDAC_MOD_STR,
1262 .probe = igen6_probe,
1263 .remove = igen6_remove,
1264 .id_table = igen6_pci_tbl,
1265};
1266
1267static int __init igen6_init(void)
1268{
1269 const char *owner;
1270 int rc;
1271
1272 edac_dbg(2, "\n");
1273
1274 owner = edac_get_owner();
1275 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR)))
1276 return -ENODEV;
1277
1278 edac_op_state = EDAC_OPSTATE_NMI;
1279
1280 rc = pci_register_driver(&igen6_driver);
1281 if (rc)
1282 return rc;
1283
1284 igen6_printk(KERN_INFO, "%s\n", IGEN6_REVISION);
1285
1286 return 0;
1287}
1288
1289static void __exit igen6_exit(void)
1290{
1291 edac_dbg(2, "\n");
1292
1293 pci_unregister_driver(&igen6_driver);
1294}
1295
1296module_init(igen6_init);
1297module_exit(igen6_exit);
1298
1299MODULE_LICENSE("GPL v2");
1300MODULE_AUTHOR("Qiuxu Zhuo");
1301MODULE_DESCRIPTION("MC Driver for Intel client SoC using In-Band ECC");
1302