3 * Copyright (C) 2014 Jason Baron <jbaron@akamai.com>
5 * Support for the E3-1200 processor family. Heavily based on previous
8 * Since the DRAM controller is on the cpu chip, we can use its PCI device
9 * id to identify these processors.
11 * PCI DRAM controller device ids (Taken from The PCI ID Repository - http://pci-ids.ucw.cz/)
13 * 0108: Xeon E3-1200 Processor Family DRAM Controller
14 * 010c: Xeon E3-1200/2nd Generation Core Processor Family DRAM Controller
15 * 0150: Xeon E3-1200 v2/3rd Gen Core processor DRAM Controller
16 * 0158: Xeon E3-1200 v2/Ivy Bridge DRAM Controller
17 * 015c: Xeon E3-1200 v2/3rd Gen Core processor DRAM Controller
18 * 0c04: Xeon E3-1200 v3/4th Gen Core Processor DRAM Controller
19 * 0c08: Xeon E3-1200 v3 Processor DRAM Controller
20 * 1918: Xeon E3-1200 v5 Skylake Host Bridge/DRAM Registers
22 * Based on Intel specification:
23 * http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xeon-e3-1200v3-vol-2-datasheet.pdf
24 * http://www.intel.com/content/www/us/en/processors/xeon/xeon-e3-1200-family-vol-2-datasheet.html
26 * According to the above datasheet (p.16):
28 * 6. Software must not access B0/D0/F0 32-bit memory-mapped registers with
29 * requests that cross a DW boundary.
32 * Thus, we make use of the explicit: lo_hi_readq(), which breaks the readq into
33 * 2 readl() calls. This restriction may be lifted in subsequent chip releases,
34 * but lo_hi_readq() ensures that we are safe across all e3-1200 processors.
37 #include <linux/module.h>
38 #include <linux/init.h>
39 #include <linux/pci.h>
40 #include <linux/pci_ids.h>
41 #include <linux/edac.h>
43 #include <linux/io-64-nonatomic-lo-hi.h>
44 #include "edac_core.h"
46 #define IE31200_REVISION "1.0"
47 #define EDAC_MOD_STR "ie31200_edac"
49 #define ie31200_printk(level, fmt, arg...) \
50 edac_printk(level, "ie31200", fmt, ##arg)
52 #define PCI_DEVICE_ID_INTEL_IE31200_HB_1 0x0108
53 #define PCI_DEVICE_ID_INTEL_IE31200_HB_2 0x010c
54 #define PCI_DEVICE_ID_INTEL_IE31200_HB_3 0x0150
55 #define PCI_DEVICE_ID_INTEL_IE31200_HB_4 0x0158
56 #define PCI_DEVICE_ID_INTEL_IE31200_HB_5 0x015c
57 #define PCI_DEVICE_ID_INTEL_IE31200_HB_6 0x0c04
58 #define PCI_DEVICE_ID_INTEL_IE31200_HB_7 0x0c08
59 #define PCI_DEVICE_ID_INTEL_IE31200_HB_8 0x1918
61 #define IE31200_DIMMS 4
62 #define IE31200_RANKS 8
63 #define IE31200_RANKS_PER_CHANNEL 4
64 #define IE31200_DIMMS_PER_CHANNEL 2
65 #define IE31200_CHANNELS 2
67 /* Intel IE31200 register addresses - device 0 function 0 - DRAM Controller */
68 #define IE31200_MCHBAR_LOW 0x48
69 #define IE31200_MCHBAR_HIGH 0x4c
70 #define IE31200_MCHBAR_MASK GENMASK_ULL(38, 15)
71 #define IE31200_MMR_WINDOW_SIZE BIT(15)
74 * Error Status Register (16b)
77 * 14 Isochronous TBWRR Run Behind FIFO Full
79 * 13 Isochronous TBWRR Run Behind FIFO Put
82 * 11 MCH Thermal Sensor Event
83 * for SMI/SCI/SERR (GTSE)
85 * 9 LOCK to non-DRAM Memory Flag (LCKF)
87 * 7 DRAM Throttle Flag (DTF)
89 * 1 Multi-bit DRAM ECC Error Flag (DMERR)
90 * 0 Single-bit DRAM ECC Error Flag (DSERR)
92 #define IE31200_ERRSTS 0xc8
93 #define IE31200_ERRSTS_UE BIT(1)
94 #define IE31200_ERRSTS_CE BIT(0)
95 #define IE31200_ERRSTS_BITS (IE31200_ERRSTS_UE | IE31200_ERRSTS_CE)
98 * Channel 0 ECC Error Log (64b)
100 * 63:48 Error Column Address (ERRCOL)
101 * 47:32 Error Row Address (ERRROW)
102 * 31:29 Error Bank Address (ERRBANK)
103 * 28:27 Error Rank Address (ERRRANK)
105 * 23:16 Error Syndrome (ERRSYND)
107 * 1 Multiple Bit Error Status (MERRSTS)
108 * 0 Correctable Error Status (CERRSTS)
111 #define IE31200_C0ECCERRLOG 0x40c8
112 #define IE31200_C1ECCERRLOG 0x44c8
113 #define IE31200_C0ECCERRLOG_SKL 0x4048
114 #define IE31200_C1ECCERRLOG_SKL 0x4448
115 #define IE31200_ECCERRLOG_CE BIT(0)
116 #define IE31200_ECCERRLOG_UE BIT(1)
117 #define IE31200_ECCERRLOG_RANK_BITS GENMASK_ULL(28, 27)
118 #define IE31200_ECCERRLOG_RANK_SHIFT 27
119 #define IE31200_ECCERRLOG_SYNDROME_BITS GENMASK_ULL(23, 16)
120 #define IE31200_ECCERRLOG_SYNDROME_SHIFT 16
122 #define IE31200_ECCERRLOG_SYNDROME(log) \
123 ((log & IE31200_ECCERRLOG_SYNDROME_BITS) >> \
124 IE31200_ECCERRLOG_SYNDROME_SHIFT)
126 #define IE31200_CAPID0 0xe4
127 #define IE31200_CAPID0_PDCD BIT(4)
128 #define IE31200_CAPID0_DDPCD BIT(6)
129 #define IE31200_CAPID0_ECC BIT(1)
131 #define IE31200_MAD_DIMM_0_OFFSET 0x5004
132 #define IE31200_MAD_DIMM_0_OFFSET_SKL 0x500C
133 #define IE31200_MAD_DIMM_SIZE GENMASK_ULL(7, 0)
134 #define IE31200_MAD_DIMM_A_RANK BIT(17)
135 #define IE31200_MAD_DIMM_A_RANK_SHIFT 17
136 #define IE31200_MAD_DIMM_A_RANK_SKL BIT(10)
137 #define IE31200_MAD_DIMM_A_RANK_SKL_SHIFT 10
138 #define IE31200_MAD_DIMM_A_WIDTH BIT(19)
139 #define IE31200_MAD_DIMM_A_WIDTH_SHIFT 19
140 #define IE31200_MAD_DIMM_A_WIDTH_SKL GENMASK_ULL(9, 8)
141 #define IE31200_MAD_DIMM_A_WIDTH_SKL_SHIFT 8
143 /* Skylake reports 1GB increments, everything else is 256MB */
144 #define IE31200_PAGES(n, skl) \
145 (n << (28 + (2 * skl) - PAGE_SHIFT))
147 static int nr_channels
;
149 struct ie31200_priv
{
150 void __iomem
*window
;
151 void __iomem
*c0errlog
;
152 void __iomem
*c1errlog
;
159 struct ie31200_dev_info
{
160 const char *ctl_name
;
163 struct ie31200_error_info
{
166 u64 eccerrlog
[IE31200_CHANNELS
];
169 static const struct ie31200_dev_info ie31200_devs
[] = {
171 .ctl_name
= "IE31200"
176 u8 size
; /* in multiples of 256MB, except Skylake is 1GB */
178 x16_width
: 2; /* 0 means x8 width */
181 static int how_many_channels(struct pci_dev
*pdev
)
184 unsigned char capid0_2b
; /* 2nd byte of CAPID0 */
186 pci_read_config_byte(pdev
, IE31200_CAPID0
+ 1, &capid0_2b
);
188 /* check PDCD: Dual Channel Disable */
189 if (capid0_2b
& IE31200_CAPID0_PDCD
) {
190 edac_dbg(0, "In single channel mode\n");
193 edac_dbg(0, "In dual channel mode\n");
197 /* check DDPCD - check if both channels are filled */
198 if (capid0_2b
& IE31200_CAPID0_DDPCD
)
199 edac_dbg(0, "2 DIMMS per channel disabled\n");
201 edac_dbg(0, "2 DIMMS per channel enabled\n");
206 static bool ecc_capable(struct pci_dev
*pdev
)
208 unsigned char capid0_4b
; /* 4th byte of CAPID0 */
210 pci_read_config_byte(pdev
, IE31200_CAPID0
+ 3, &capid0_4b
);
211 if (capid0_4b
& IE31200_CAPID0_ECC
)
216 static int eccerrlog_row(u64 log
)
218 return ((log
& IE31200_ECCERRLOG_RANK_BITS
) >>
219 IE31200_ECCERRLOG_RANK_SHIFT
);
222 static void ie31200_clear_error_info(struct mem_ctl_info
*mci
)
225 * Clear any error bits.
226 * (Yes, we really clear bits by writing 1 to them.)
228 pci_write_bits16(to_pci_dev(mci
->pdev
), IE31200_ERRSTS
,
229 IE31200_ERRSTS_BITS
, IE31200_ERRSTS_BITS
);
232 static void ie31200_get_and_clear_error_info(struct mem_ctl_info
*mci
,
233 struct ie31200_error_info
*info
)
235 struct pci_dev
*pdev
;
236 struct ie31200_priv
*priv
= mci
->pvt_info
;
238 pdev
= to_pci_dev(mci
->pdev
);
241 * This is a mess because there is no atomic way to read all the
242 * registers at once and the registers can transition from CE being
245 pci_read_config_word(pdev
, IE31200_ERRSTS
, &info
->errsts
);
246 if (!(info
->errsts
& IE31200_ERRSTS_BITS
))
249 info
->eccerrlog
[0] = lo_hi_readq(priv
->c0errlog
);
250 if (nr_channels
== 2)
251 info
->eccerrlog
[1] = lo_hi_readq(priv
->c1errlog
);
253 pci_read_config_word(pdev
, IE31200_ERRSTS
, &info
->errsts2
);
256 * If the error is the same for both reads then the first set
257 * of reads is valid. If there is a change then there is a CE
258 * with no info and the second set of reads is valid and
261 if ((info
->errsts
^ info
->errsts2
) & IE31200_ERRSTS_BITS
) {
262 info
->eccerrlog
[0] = lo_hi_readq(priv
->c0errlog
);
263 if (nr_channels
== 2)
265 lo_hi_readq(priv
->c1errlog
);
268 ie31200_clear_error_info(mci
);
271 static void ie31200_process_error_info(struct mem_ctl_info
*mci
,
272 struct ie31200_error_info
*info
)
277 if (!(info
->errsts
& IE31200_ERRSTS_BITS
))
280 if ((info
->errsts
^ info
->errsts2
) & IE31200_ERRSTS_BITS
) {
281 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1, 0, 0, 0,
282 -1, -1, -1, "UE overwrote CE", "");
283 info
->errsts
= info
->errsts2
;
286 for (channel
= 0; channel
< nr_channels
; channel
++) {
287 log
= info
->eccerrlog
[channel
];
288 if (log
& IE31200_ECCERRLOG_UE
) {
289 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1,
294 } else if (log
& IE31200_ECCERRLOG_CE
) {
295 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED
, mci
, 1,
297 IE31200_ECCERRLOG_SYNDROME(log
),
305 static void ie31200_check(struct mem_ctl_info
*mci
)
307 struct ie31200_error_info info
;
309 edac_dbg(1, "MC%d\n", mci
->mc_idx
);
310 ie31200_get_and_clear_error_info(mci
, &info
);
311 ie31200_process_error_info(mci
, &info
);
314 static void __iomem
*ie31200_map_mchbar(struct pci_dev
*pdev
)
323 void __iomem
*window
;
325 pci_read_config_dword(pdev
, IE31200_MCHBAR_LOW
, &u
.mchbar_low
);
326 pci_read_config_dword(pdev
, IE31200_MCHBAR_HIGH
, &u
.mchbar_high
);
327 u
.mchbar
&= IE31200_MCHBAR_MASK
;
329 if (u
.mchbar
!= (resource_size_t
)u
.mchbar
) {
330 ie31200_printk(KERN_ERR
, "mmio space beyond accessible range (0x%llx)\n",
331 (unsigned long long)u
.mchbar
);
335 window
= ioremap_nocache(u
.mchbar
, IE31200_MMR_WINDOW_SIZE
);
337 ie31200_printk(KERN_ERR
, "Cannot map mmio space at 0x%llx\n",
338 (unsigned long long)u
.mchbar
);
343 static void __skl_populate_dimm_info(struct dimm_data
*dd
, u32 addr_decode
,
346 dd
->size
= (addr_decode
>> (chan
<< 4)) & IE31200_MAD_DIMM_SIZE
;
347 dd
->dual_rank
= (addr_decode
& (IE31200_MAD_DIMM_A_RANK_SKL
<< (chan
<< 4))) ? 1 : 0;
348 dd
->x16_width
= ((addr_decode
& (IE31200_MAD_DIMM_A_WIDTH_SKL
<< (chan
<< 4))) >>
349 (IE31200_MAD_DIMM_A_WIDTH_SKL_SHIFT
+ (chan
<< 4)));
352 static void __populate_dimm_info(struct dimm_data
*dd
, u32 addr_decode
,
355 dd
->size
= (addr_decode
>> (chan
<< 3)) & IE31200_MAD_DIMM_SIZE
;
356 dd
->dual_rank
= (addr_decode
& (IE31200_MAD_DIMM_A_RANK
<< chan
)) ? 1 : 0;
357 dd
->x16_width
= (addr_decode
& (IE31200_MAD_DIMM_A_WIDTH
<< chan
)) ? 1 : 0;
360 static void populate_dimm_info(struct dimm_data
*dd
, u32 addr_decode
, int chan
,
364 __skl_populate_dimm_info(dd
, addr_decode
, chan
);
366 __populate_dimm_info(dd
, addr_decode
, chan
);
370 static int ie31200_probe1(struct pci_dev
*pdev
, int dev_idx
)
373 struct mem_ctl_info
*mci
= NULL
;
374 struct edac_mc_layer layers
[2];
375 struct dimm_data dimm_info
[IE31200_CHANNELS
][IE31200_DIMMS_PER_CHANNEL
];
376 void __iomem
*window
;
377 struct ie31200_priv
*priv
;
378 u32 addr_decode
, mad_offset
;
379 bool skl
= (pdev
->device
== PCI_DEVICE_ID_INTEL_IE31200_HB_8
);
381 edac_dbg(0, "MC:\n");
383 if (!ecc_capable(pdev
)) {
384 ie31200_printk(KERN_INFO
, "No ECC support\n");
388 nr_channels
= how_many_channels(pdev
);
389 layers
[0].type
= EDAC_MC_LAYER_CHIP_SELECT
;
390 layers
[0].size
= IE31200_DIMMS
;
391 layers
[0].is_virt_csrow
= true;
392 layers
[1].type
= EDAC_MC_LAYER_CHANNEL
;
393 layers
[1].size
= nr_channels
;
394 layers
[1].is_virt_csrow
= false;
395 mci
= edac_mc_alloc(0, ARRAY_SIZE(layers
), layers
,
396 sizeof(struct ie31200_priv
));
400 window
= ie31200_map_mchbar(pdev
);
406 edac_dbg(3, "MC: init mci\n");
407 mci
->pdev
= &pdev
->dev
;
409 mci
->mtype_cap
= MEM_FLAG_DDR4
;
411 mci
->mtype_cap
= MEM_FLAG_DDR3
;
412 mci
->edac_ctl_cap
= EDAC_FLAG_SECDED
;
413 mci
->edac_cap
= EDAC_FLAG_SECDED
;
414 mci
->mod_name
= EDAC_MOD_STR
;
415 mci
->mod_ver
= IE31200_REVISION
;
416 mci
->ctl_name
= ie31200_devs
[dev_idx
].ctl_name
;
417 mci
->dev_name
= pci_name(pdev
);
418 mci
->edac_check
= ie31200_check
;
419 mci
->ctl_page_to_phys
= NULL
;
420 priv
= mci
->pvt_info
;
421 priv
->window
= window
;
423 priv
->c0errlog
= window
+ IE31200_C0ECCERRLOG_SKL
;
424 priv
->c1errlog
= window
+ IE31200_C1ECCERRLOG_SKL
;
425 mad_offset
= IE31200_MAD_DIMM_0_OFFSET_SKL
;
427 priv
->c0errlog
= window
+ IE31200_C0ECCERRLOG
;
428 priv
->c1errlog
= window
+ IE31200_C1ECCERRLOG
;
429 mad_offset
= IE31200_MAD_DIMM_0_OFFSET
;
432 /* populate DIMM info */
433 for (i
= 0; i
< IE31200_CHANNELS
; i
++) {
434 addr_decode
= readl(window
+ mad_offset
+
436 edac_dbg(0, "addr_decode: 0x%x\n", addr_decode
);
437 for (j
= 0; j
< IE31200_DIMMS_PER_CHANNEL
; j
++) {
438 populate_dimm_info(&dimm_info
[i
][j
], addr_decode
, j
,
440 edac_dbg(0, "size: 0x%x, rank: %d, width: %d\n",
441 dimm_info
[i
][j
].size
,
442 dimm_info
[i
][j
].dual_rank
,
443 dimm_info
[i
][j
].x16_width
);
448 * The dram rank boundary (DRB) reg values are boundary addresses
449 * for each DRAM rank with a granularity of 64MB. DRB regs are
450 * cumulative; the last one will contain the total memory
451 * contained in all ranks.
453 for (i
= 0; i
< IE31200_DIMMS_PER_CHANNEL
; i
++) {
454 for (j
= 0; j
< IE31200_CHANNELS
; j
++) {
455 struct dimm_info
*dimm
;
456 unsigned long nr_pages
;
458 nr_pages
= IE31200_PAGES(dimm_info
[j
][i
].size
, skl
);
462 if (dimm_info
[j
][i
].dual_rank
) {
463 nr_pages
= nr_pages
/ 2;
464 dimm
= EDAC_DIMM_PTR(mci
->layers
, mci
->dimms
,
465 mci
->n_layers
, (i
* 2) + 1,
467 dimm
->nr_pages
= nr_pages
;
468 edac_dbg(0, "set nr pages: 0x%lx\n", nr_pages
);
469 dimm
->grain
= 8; /* just a guess */
471 dimm
->mtype
= MEM_DDR4
;
473 dimm
->mtype
= MEM_DDR3
;
474 dimm
->dtype
= DEV_UNKNOWN
;
475 dimm
->edac_mode
= EDAC_UNKNOWN
;
477 dimm
= EDAC_DIMM_PTR(mci
->layers
, mci
->dimms
,
478 mci
->n_layers
, i
* 2, j
, 0);
479 dimm
->nr_pages
= nr_pages
;
480 edac_dbg(0, "set nr pages: 0x%lx\n", nr_pages
);
481 dimm
->grain
= 8; /* same guess */
483 dimm
->mtype
= MEM_DDR4
;
485 dimm
->mtype
= MEM_DDR3
;
486 dimm
->dtype
= DEV_UNKNOWN
;
487 dimm
->edac_mode
= EDAC_UNKNOWN
;
491 ie31200_clear_error_info(mci
);
493 if (edac_mc_add_mc(mci
)) {
494 edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
499 /* get this far and it's successful */
500 edac_dbg(3, "MC: success\n");
512 static int ie31200_init_one(struct pci_dev
*pdev
,
513 const struct pci_device_id
*ent
)
515 edac_dbg(0, "MC:\n");
517 if (pci_enable_device(pdev
) < 0)
520 return ie31200_probe1(pdev
, ent
->driver_data
);
523 static void ie31200_remove_one(struct pci_dev
*pdev
)
525 struct mem_ctl_info
*mci
;
526 struct ie31200_priv
*priv
;
529 mci
= edac_mc_del_mc(&pdev
->dev
);
532 priv
= mci
->pvt_info
;
533 iounmap(priv
->window
);
537 static const struct pci_device_id ie31200_pci_tbl
[] = {
539 PCI_VEND_DEV(INTEL
, IE31200_HB_1
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
542 PCI_VEND_DEV(INTEL
, IE31200_HB_2
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
545 PCI_VEND_DEV(INTEL
, IE31200_HB_3
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
548 PCI_VEND_DEV(INTEL
, IE31200_HB_4
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
551 PCI_VEND_DEV(INTEL
, IE31200_HB_5
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
554 PCI_VEND_DEV(INTEL
, IE31200_HB_6
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
557 PCI_VEND_DEV(INTEL
, IE31200_HB_7
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
560 PCI_VEND_DEV(INTEL
, IE31200_HB_8
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
564 } /* 0 terminated list. */
566 MODULE_DEVICE_TABLE(pci
, ie31200_pci_tbl
);
568 static struct pci_driver ie31200_driver
= {
569 .name
= EDAC_MOD_STR
,
570 .probe
= ie31200_init_one
,
571 .remove
= ie31200_remove_one
,
572 .id_table
= ie31200_pci_tbl
,
575 static int __init
ie31200_init(void)
577 edac_dbg(3, "MC:\n");
578 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
581 return pci_register_driver(&ie31200_driver
);
584 static void __exit
ie31200_exit(void)
586 edac_dbg(3, "MC:\n");
587 pci_unregister_driver(&ie31200_driver
);
590 module_init(ie31200_init
);
591 module_exit(ie31200_exit
);
593 MODULE_LICENSE("GPL");
594 MODULE_AUTHOR("Jason Baron <jbaron@akamai.com>");
595 MODULE_DESCRIPTION("MC support for Intel Processor E31200 memory hub controllers");