2 * Intel e7xxx Memory Controller kernel module
3 * (C) 2003 Linux Networx (http://lnxi.com)
4 * This file may be distributed under the terms of the
5 * GNU General Public License.
7 * See "enum e7xxx_chips" below for supported chipsets
9 * Written by Thayne Harbaugh
10 * Based on work by Dan Hollis <goemon at anime dot net> and others.
11 * http://www.anime.net/~goemon/linux-ecc/
14 * http://www.intel.com/content/www/us/en/chipsets/e7501-chipset-memory-controller-hub-datasheet.html
17 * Eric Biederman (Linux Networx)
18 * Tom Zimmerman (Linux Networx)
19 * Jim Garlick (Lawrence Livermore National Labs)
20 * Dave Peterson (Lawrence Livermore National Labs)
21 * That One Guy (Some other place)
22 * Wang Zhenyu (intel.com)
24 * $Id: edac_e7xxx.c,v 1.5.2.9 2005/10/05 00:43:44 dsp_llnl Exp $
28 #include <linux/module.h>
29 #include <linux/init.h>
30 #include <linux/pci.h>
31 #include <linux/pci_ids.h>
32 #include <linux/edac.h>
33 #include "edac_module.h"
35 #define EDAC_MOD_STR "e7xxx_edac"
37 #define e7xxx_printk(level, fmt, arg...) \
38 edac_printk(level, "e7xxx", fmt, ##arg)
40 #define e7xxx_mc_printk(mci, level, fmt, arg...) \
41 edac_mc_chipset_printk(mci, level, "e7xxx", fmt, ##arg)
43 #ifndef PCI_DEVICE_ID_INTEL_7205_0
44 #define PCI_DEVICE_ID_INTEL_7205_0 0x255d
45 #endif /* PCI_DEVICE_ID_INTEL_7205_0 */
47 #ifndef PCI_DEVICE_ID_INTEL_7205_1_ERR
48 #define PCI_DEVICE_ID_INTEL_7205_1_ERR 0x2551
49 #endif /* PCI_DEVICE_ID_INTEL_7205_1_ERR */
51 #ifndef PCI_DEVICE_ID_INTEL_7500_0
52 #define PCI_DEVICE_ID_INTEL_7500_0 0x2540
53 #endif /* PCI_DEVICE_ID_INTEL_7500_0 */
55 #ifndef PCI_DEVICE_ID_INTEL_7500_1_ERR
56 #define PCI_DEVICE_ID_INTEL_7500_1_ERR 0x2541
57 #endif /* PCI_DEVICE_ID_INTEL_7500_1_ERR */
59 #ifndef PCI_DEVICE_ID_INTEL_7501_0
60 #define PCI_DEVICE_ID_INTEL_7501_0 0x254c
61 #endif /* PCI_DEVICE_ID_INTEL_7501_0 */
63 #ifndef PCI_DEVICE_ID_INTEL_7501_1_ERR
64 #define PCI_DEVICE_ID_INTEL_7501_1_ERR 0x2541
65 #endif /* PCI_DEVICE_ID_INTEL_7501_1_ERR */
67 #ifndef PCI_DEVICE_ID_INTEL_7505_0
68 #define PCI_DEVICE_ID_INTEL_7505_0 0x2550
69 #endif /* PCI_DEVICE_ID_INTEL_7505_0 */
71 #ifndef PCI_DEVICE_ID_INTEL_7505_1_ERR
72 #define PCI_DEVICE_ID_INTEL_7505_1_ERR 0x2551
73 #endif /* PCI_DEVICE_ID_INTEL_7505_1_ERR */
75 #define E7XXX_NR_CSROWS 8 /* number of csrows */
76 #define E7XXX_NR_DIMMS 8 /* 2 channels, 4 dimms/channel */
78 /* E7XXX register addresses - device 0 function 0 */
79 #define E7XXX_DRB 0x60 /* DRAM row boundary register (8b) */
80 #define E7XXX_DRA 0x70 /* DRAM row attribute register (8b) */
82 * 31 Device width row 7 0=x8 1=x4
83 * 27 Device width row 6
84 * 23 Device width row 5
85 * 19 Device width row 4
86 * 15 Device width row 3
87 * 11 Device width row 2
88 * 7 Device width row 1
89 * 3 Device width row 0
91 #define E7XXX_DRC 0x7C /* DRAM controller mode reg (32b) */
93 * 22 Number channels 0=1,1=2
94 * 19:18 DRB Granularity 32/64MB
96 #define E7XXX_TOLM 0xC4 /* DRAM top of low memory reg (16b) */
97 #define E7XXX_REMAPBASE 0xC6 /* DRAM remap base address reg (16b) */
98 #define E7XXX_REMAPLIMIT 0xC8 /* DRAM remap limit address reg (16b) */
100 /* E7XXX register addresses - device 0 function 1 */
101 #define E7XXX_DRAM_FERR 0x80 /* DRAM first error register (8b) */
102 #define E7XXX_DRAM_NERR 0x82 /* DRAM next error register (8b) */
103 #define E7XXX_DRAM_CELOG_ADD 0xA0 /* DRAM first correctable memory */
104 /* error address register (32b) */
107 * 27:6 CE address (4k block 33:12)
110 #define E7XXX_DRAM_UELOG_ADD 0xB0 /* DRAM first uncorrectable memory */
111 /* error address register (32b) */
114 * 27:6 CE address (4k block 33:12)
117 #define E7XXX_DRAM_CELOG_SYNDROME 0xD0 /* DRAM first correctable memory */
118 /* error syndrome register (16b) */
128 struct pci_dev
*bridge_ck
;
132 const struct e7xxx_dev_info
*dev_info
;
135 struct e7xxx_dev_info
{
137 const char *ctl_name
;
140 struct e7xxx_error_info
{
144 u16 dram_celog_syndrome
;
148 static struct edac_pci_ctl_info
*e7xxx_pci
;
150 static const struct e7xxx_dev_info e7xxx_devs
[] = {
152 .err_dev
= PCI_DEVICE_ID_INTEL_7500_1_ERR
,
153 .ctl_name
= "E7500"},
155 .err_dev
= PCI_DEVICE_ID_INTEL_7501_1_ERR
,
156 .ctl_name
= "E7501"},
158 .err_dev
= PCI_DEVICE_ID_INTEL_7505_1_ERR
,
159 .ctl_name
= "E7505"},
161 .err_dev
= PCI_DEVICE_ID_INTEL_7205_1_ERR
,
162 .ctl_name
= "E7205"},
165 /* FIXME - is this valid for both SECDED and S4ECD4ED? */
166 static inline int e7xxx_find_channel(u16 syndrome
)
170 if ((syndrome
& 0xff00) == 0)
173 if ((syndrome
& 0x00ff) == 0)
176 if ((syndrome
& 0xf000) == 0 || (syndrome
& 0x0f00) == 0)
182 static unsigned long ctl_page_to_phys(struct mem_ctl_info
*mci
,
186 struct e7xxx_pvt
*pvt
= (struct e7xxx_pvt
*)mci
->pvt_info
;
190 if ((page
< pvt
->tolm
) ||
191 ((page
>= 0x100000) && (page
< pvt
->remapbase
)))
194 remap
= (page
- pvt
->tolm
) + pvt
->remapbase
;
196 if (remap
< pvt
->remaplimit
)
199 e7xxx_printk(KERN_ERR
, "Invalid page %lx - out of range\n", page
);
200 return pvt
->tolm
- 1;
203 static void process_ce(struct mem_ctl_info
*mci
, struct e7xxx_error_info
*info
)
211 /* read the error address */
212 error_1b
= info
->dram_celog_add
;
213 /* FIXME - should use PAGE_SHIFT */
214 page
= error_1b
>> 6; /* convert the address to 4k page */
215 /* read the syndrome */
216 syndrome
= info
->dram_celog_syndrome
;
217 /* FIXME - check for -1 */
218 row
= edac_mc_find_csrow_by_page(mci
, page
);
219 /* convert syndrome to channel */
220 channel
= e7xxx_find_channel(syndrome
);
221 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED
, mci
, 1, page
, 0, syndrome
,
222 row
, channel
, -1, "e7xxx CE", "");
225 static void process_ce_no_info(struct mem_ctl_info
*mci
)
228 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED
, mci
, 1, 0, 0, 0, -1, -1, -1,
229 "e7xxx CE log register overflow", "");
232 static void process_ue(struct mem_ctl_info
*mci
, struct e7xxx_error_info
*info
)
234 u32 error_2b
, block_page
;
238 /* read the error address */
239 error_2b
= info
->dram_uelog_add
;
240 /* FIXME - should use PAGE_SHIFT */
241 block_page
= error_2b
>> 6; /* convert to 4k address */
242 row
= edac_mc_find_csrow_by_page(mci
, block_page
);
244 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1, block_page
, 0, 0,
245 row
, -1, -1, "e7xxx UE", "");
248 static void process_ue_no_info(struct mem_ctl_info
*mci
)
252 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1, 0, 0, 0, -1, -1, -1,
253 "e7xxx UE log register overflow", "");
256 static void e7xxx_get_error_info(struct mem_ctl_info
*mci
,
257 struct e7xxx_error_info
*info
)
259 struct e7xxx_pvt
*pvt
;
261 pvt
= (struct e7xxx_pvt
*)mci
->pvt_info
;
262 pci_read_config_byte(pvt
->bridge_ck
, E7XXX_DRAM_FERR
, &info
->dram_ferr
);
263 pci_read_config_byte(pvt
->bridge_ck
, E7XXX_DRAM_NERR
, &info
->dram_nerr
);
265 if ((info
->dram_ferr
& 1) || (info
->dram_nerr
& 1)) {
266 pci_read_config_dword(pvt
->bridge_ck
, E7XXX_DRAM_CELOG_ADD
,
267 &info
->dram_celog_add
);
268 pci_read_config_word(pvt
->bridge_ck
,
269 E7XXX_DRAM_CELOG_SYNDROME
,
270 &info
->dram_celog_syndrome
);
273 if ((info
->dram_ferr
& 2) || (info
->dram_nerr
& 2))
274 pci_read_config_dword(pvt
->bridge_ck
, E7XXX_DRAM_UELOG_ADD
,
275 &info
->dram_uelog_add
);
277 if (info
->dram_ferr
& 3)
278 pci_write_bits8(pvt
->bridge_ck
, E7XXX_DRAM_FERR
, 0x03, 0x03);
280 if (info
->dram_nerr
& 3)
281 pci_write_bits8(pvt
->bridge_ck
, E7XXX_DRAM_NERR
, 0x03, 0x03);
284 static int e7xxx_process_error_info(struct mem_ctl_info
*mci
,
285 struct e7xxx_error_info
*info
,
292 /* decode and report errors */
293 if (info
->dram_ferr
& 1) { /* check first error correctable */
297 process_ce(mci
, info
);
300 if (info
->dram_ferr
& 2) { /* check first error uncorrectable */
304 process_ue(mci
, info
);
307 if (info
->dram_nerr
& 1) { /* check next error correctable */
311 if (info
->dram_ferr
& 1)
312 process_ce_no_info(mci
);
314 process_ce(mci
, info
);
318 if (info
->dram_nerr
& 2) { /* check next error uncorrectable */
322 if (info
->dram_ferr
& 2)
323 process_ue_no_info(mci
);
325 process_ue(mci
, info
);
332 static void e7xxx_check(struct mem_ctl_info
*mci
)
334 struct e7xxx_error_info info
;
337 e7xxx_get_error_info(mci
, &info
);
338 e7xxx_process_error_info(mci
, &info
, 1);
341 /* Return 1 if dual channel mode is active. Else return 0. */
342 static inline int dual_channel_active(u32 drc
, int dev_idx
)
344 return (dev_idx
== E7501
) ? ((drc
>> 22) & 0x1) : 1;
347 /* Return DRB granularity (0=32mb, 1=64mb). */
348 static inline int drb_granularity(u32 drc
, int dev_idx
)
350 /* only e7501 can be single channel */
351 return (dev_idx
== E7501
) ? ((drc
>> 18) & 0x3) : 1;
354 static void e7xxx_init_csrows(struct mem_ctl_info
*mci
, struct pci_dev
*pdev
,
355 int dev_idx
, u32 drc
)
357 unsigned long last_cumul_size
;
360 u32 dra
, cumul_size
, nr_pages
;
361 int drc_chan
, drc_drbg
, drc_ddim
, mem_dev
;
362 struct csrow_info
*csrow
;
363 struct dimm_info
*dimm
;
364 enum edac_type edac_mode
;
366 pci_read_config_dword(pdev
, E7XXX_DRA
, &dra
);
367 drc_chan
= dual_channel_active(drc
, dev_idx
);
368 drc_drbg
= drb_granularity(drc
, dev_idx
);
369 drc_ddim
= (drc
>> 20) & 0x3;
372 /* The dram row boundary (DRB) reg values are boundary address
373 * for each DRAM row with a granularity of 32 or 64MB (single/dual
374 * channel operation). DRB regs are cumulative; therefore DRB7 will
375 * contain the total memory contained in all eight rows.
377 for (index
= 0; index
< mci
->nr_csrows
; index
++) {
378 /* mem_dev 0=x8, 1=x4 */
379 mem_dev
= (dra
>> (index
* 4 + 3)) & 0x1;
380 csrow
= mci
->csrows
[index
];
382 pci_read_config_byte(pdev
, E7XXX_DRB
+ index
, &value
);
383 /* convert a 64 or 32 MiB DRB to a page size. */
384 cumul_size
= value
<< (25 + drc_drbg
- PAGE_SHIFT
);
385 edac_dbg(3, "(%d) cumul_size 0x%x\n", index
, cumul_size
);
386 if (cumul_size
== last_cumul_size
)
387 continue; /* not populated */
389 csrow
->first_page
= last_cumul_size
;
390 csrow
->last_page
= cumul_size
- 1;
391 nr_pages
= cumul_size
- last_cumul_size
;
392 last_cumul_size
= cumul_size
;
395 * if single channel or x8 devices then SECDED
396 * if dual channel and x4 then S4ECD4ED
399 if (drc_chan
&& mem_dev
) {
400 edac_mode
= EDAC_S4ECD4ED
;
401 mci
->edac_cap
|= EDAC_FLAG_S4ECD4ED
;
403 edac_mode
= EDAC_SECDED
;
404 mci
->edac_cap
|= EDAC_FLAG_SECDED
;
407 edac_mode
= EDAC_NONE
;
409 for (j
= 0; j
< drc_chan
+ 1; j
++) {
410 dimm
= csrow
->channels
[j
]->dimm
;
412 dimm
->nr_pages
= nr_pages
/ (drc_chan
+ 1);
413 dimm
->grain
= 1 << 12; /* 4KiB - resolution of CELOG */
414 dimm
->mtype
= MEM_RDDR
; /* only one type supported */
415 dimm
->dtype
= mem_dev
? DEV_X4
: DEV_X8
;
416 dimm
->edac_mode
= edac_mode
;
421 static int e7xxx_probe1(struct pci_dev
*pdev
, int dev_idx
)
424 struct mem_ctl_info
*mci
= NULL
;
425 struct edac_mc_layer layers
[2];
426 struct e7xxx_pvt
*pvt
= NULL
;
429 struct e7xxx_error_info discard
;
431 edac_dbg(0, "mci\n");
433 pci_read_config_dword(pdev
, E7XXX_DRC
, &drc
);
435 drc_chan
= dual_channel_active(drc
, dev_idx
);
437 * According with the datasheet, this device has a maximum of
438 * 4 DIMMS per channel, either single-rank or dual-rank. So, the
439 * total amount of dimms is 8 (E7XXX_NR_DIMMS).
440 * That means that the DIMM is mapped as CSROWs, and the channel
441 * will map the rank. So, an error to either channel should be
442 * attributed to the same dimm.
444 layers
[0].type
= EDAC_MC_LAYER_CHIP_SELECT
;
445 layers
[0].size
= E7XXX_NR_CSROWS
;
446 layers
[0].is_virt_csrow
= true;
447 layers
[1].type
= EDAC_MC_LAYER_CHANNEL
;
448 layers
[1].size
= drc_chan
+ 1;
449 layers
[1].is_virt_csrow
= false;
450 mci
= edac_mc_alloc(0, ARRAY_SIZE(layers
), layers
, sizeof(*pvt
));
454 edac_dbg(3, "init mci\n");
455 mci
->mtype_cap
= MEM_FLAG_RDDR
;
456 mci
->edac_ctl_cap
= EDAC_FLAG_NONE
| EDAC_FLAG_SECDED
|
458 /* FIXME - what if different memory types are in different csrows? */
459 mci
->mod_name
= EDAC_MOD_STR
;
460 mci
->pdev
= &pdev
->dev
;
461 edac_dbg(3, "init pvt\n");
462 pvt
= (struct e7xxx_pvt
*)mci
->pvt_info
;
463 pvt
->dev_info
= &e7xxx_devs
[dev_idx
];
464 pvt
->bridge_ck
= pci_get_device(PCI_VENDOR_ID_INTEL
,
465 pvt
->dev_info
->err_dev
, pvt
->bridge_ck
);
467 if (!pvt
->bridge_ck
) {
468 e7xxx_printk(KERN_ERR
, "error reporting device not found:"
469 "vendor %x device 0x%x (broken BIOS?)\n",
470 PCI_VENDOR_ID_INTEL
, e7xxx_devs
[dev_idx
].err_dev
);
474 edac_dbg(3, "more mci init\n");
475 mci
->ctl_name
= pvt
->dev_info
->ctl_name
;
476 mci
->dev_name
= pci_name(pdev
);
477 mci
->edac_check
= e7xxx_check
;
478 mci
->ctl_page_to_phys
= ctl_page_to_phys
;
479 e7xxx_init_csrows(mci
, pdev
, dev_idx
, drc
);
480 mci
->edac_cap
|= EDAC_FLAG_NONE
;
481 edac_dbg(3, "tolm, remapbase, remaplimit\n");
482 /* load the top of low memory, remap base, and remap limit vars */
483 pci_read_config_word(pdev
, E7XXX_TOLM
, &pci_data
);
484 pvt
->tolm
= ((u32
) pci_data
) << 4;
485 pci_read_config_word(pdev
, E7XXX_REMAPBASE
, &pci_data
);
486 pvt
->remapbase
= ((u32
) pci_data
) << 14;
487 pci_read_config_word(pdev
, E7XXX_REMAPLIMIT
, &pci_data
);
488 pvt
->remaplimit
= ((u32
) pci_data
) << 14;
489 e7xxx_printk(KERN_INFO
,
490 "tolm = %x, remapbase = %x, remaplimit = %x\n", pvt
->tolm
,
491 pvt
->remapbase
, pvt
->remaplimit
);
493 /* clear any pending errors, or initial state bits */
494 e7xxx_get_error_info(mci
, &discard
);
496 /* Here we assume that we will never see multiple instances of this
497 * type of memory controller. The ID is therefore hardcoded to 0.
499 if (edac_mc_add_mc(mci
)) {
500 edac_dbg(3, "failed edac_mc_add_mc()\n");
504 /* allocating generic PCI control info */
505 e7xxx_pci
= edac_pci_create_generic_ctl(&pdev
->dev
, EDAC_MOD_STR
);
508 "%s(): Unable to create PCI control\n",
511 "%s(): PCI error report via EDAC not setup\n",
515 /* get this far and it's successful */
516 edac_dbg(3, "success\n");
520 pci_dev_put(pvt
->bridge_ck
);
528 /* returns count (>= 0), or negative on error */
529 static int e7xxx_init_one(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
533 /* wake up and enable device */
534 return pci_enable_device(pdev
) ?
535 -EIO
: e7xxx_probe1(pdev
, ent
->driver_data
);
538 static void e7xxx_remove_one(struct pci_dev
*pdev
)
540 struct mem_ctl_info
*mci
;
541 struct e7xxx_pvt
*pvt
;
546 edac_pci_release_generic_ctl(e7xxx_pci
);
548 if ((mci
= edac_mc_del_mc(&pdev
->dev
)) == NULL
)
551 pvt
= (struct e7xxx_pvt
*)mci
->pvt_info
;
552 pci_dev_put(pvt
->bridge_ck
);
556 static const struct pci_device_id e7xxx_pci_tbl
[] = {
558 PCI_VEND_DEV(INTEL
, 7205_0
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
561 PCI_VEND_DEV(INTEL
, 7500_0
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
564 PCI_VEND_DEV(INTEL
, 7501_0
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
567 PCI_VEND_DEV(INTEL
, 7505_0
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
571 } /* 0 terminated list. */
574 MODULE_DEVICE_TABLE(pci
, e7xxx_pci_tbl
);
576 static struct pci_driver e7xxx_driver
= {
577 .name
= EDAC_MOD_STR
,
578 .probe
= e7xxx_init_one
,
579 .remove
= e7xxx_remove_one
,
580 .id_table
= e7xxx_pci_tbl
,
583 static int __init
e7xxx_init(void)
585 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
588 return pci_register_driver(&e7xxx_driver
);
591 static void __exit
e7xxx_exit(void)
593 pci_unregister_driver(&e7xxx_driver
);
596 module_init(e7xxx_init
);
597 module_exit(e7xxx_exit
);
599 MODULE_LICENSE("GPL");
600 MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh et al\n"
601 "Based on.work by Dan Hollis et al");
602 MODULE_DESCRIPTION("MC support for Intel e7xxx memory controllers");
603 module_param(edac_op_state
, int, 0444);
604 MODULE_PARM_DESC(edac_op_state
, "EDAC Error Reporting state: 0=Poll,1=NMI");