Root/drivers/edac/e7xxx_edac.c

1/*
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.
6 *
7 * See "enum e7xxx_chips" below for supported chipsets
8 *
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/
12 *
13 * Datasheet:
14 * http://www.intel.com/content/www/us/en/chipsets/e7501-chipset-memory-controller-hub-datasheet.html
15 *
16 * Contributors:
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)
23 *
24 * $Id: edac_e7xxx.c,v 1.5.2.9 2005/10/05 00:43:44 dsp_llnl Exp $
25 *
26 */
27
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_core.h"
34
35#define E7XXX_REVISION " Ver: 2.0.2"
36#define EDAC_MOD_STR "e7xxx_edac"
37
38#define e7xxx_printk(level, fmt, arg...) \
39    edac_printk(level, "e7xxx", fmt, ##arg)
40
41#define e7xxx_mc_printk(mci, level, fmt, arg...) \
42    edac_mc_chipset_printk(mci, level, "e7xxx", fmt, ##arg)
43
44#ifndef PCI_DEVICE_ID_INTEL_7205_0
45#define PCI_DEVICE_ID_INTEL_7205_0 0x255d
46#endif /* PCI_DEVICE_ID_INTEL_7205_0 */
47
48#ifndef PCI_DEVICE_ID_INTEL_7205_1_ERR
49#define PCI_DEVICE_ID_INTEL_7205_1_ERR 0x2551
50#endif /* PCI_DEVICE_ID_INTEL_7205_1_ERR */
51
52#ifndef PCI_DEVICE_ID_INTEL_7500_0
53#define PCI_DEVICE_ID_INTEL_7500_0 0x2540
54#endif /* PCI_DEVICE_ID_INTEL_7500_0 */
55
56#ifndef PCI_DEVICE_ID_INTEL_7500_1_ERR
57#define PCI_DEVICE_ID_INTEL_7500_1_ERR 0x2541
58#endif /* PCI_DEVICE_ID_INTEL_7500_1_ERR */
59
60#ifndef PCI_DEVICE_ID_INTEL_7501_0
61#define PCI_DEVICE_ID_INTEL_7501_0 0x254c
62#endif /* PCI_DEVICE_ID_INTEL_7501_0 */
63
64#ifndef PCI_DEVICE_ID_INTEL_7501_1_ERR
65#define PCI_DEVICE_ID_INTEL_7501_1_ERR 0x2541
66#endif /* PCI_DEVICE_ID_INTEL_7501_1_ERR */
67
68#ifndef PCI_DEVICE_ID_INTEL_7505_0
69#define PCI_DEVICE_ID_INTEL_7505_0 0x2550
70#endif /* PCI_DEVICE_ID_INTEL_7505_0 */
71
72#ifndef PCI_DEVICE_ID_INTEL_7505_1_ERR
73#define PCI_DEVICE_ID_INTEL_7505_1_ERR 0x2551
74#endif /* PCI_DEVICE_ID_INTEL_7505_1_ERR */
75
76#define E7XXX_NR_CSROWS 8 /* number of csrows */
77#define E7XXX_NR_DIMMS 8 /* 2 channels, 4 dimms/channel */
78
79/* E7XXX register addresses - device 0 function 0 */
80#define E7XXX_DRB 0x60 /* DRAM row boundary register (8b) */
81#define E7XXX_DRA 0x70 /* DRAM row attribute register (8b) */
82                    /*
83                     * 31 Device width row 7 0=x8 1=x4
84                     * 27 Device width row 6
85                     * 23 Device width row 5
86                     * 19 Device width row 4
87                     * 15 Device width row 3
88                     * 11 Device width row 2
89                     * 7 Device width row 1
90                     * 3 Device width row 0
91                     */
92#define E7XXX_DRC 0x7C /* DRAM controller mode reg (32b) */
93                    /*
94                     * 22 Number channels 0=1,1=2
95                     * 19:18 DRB Granularity 32/64MB
96                     */
97#define E7XXX_TOLM 0xC4 /* DRAM top of low memory reg (16b) */
98#define E7XXX_REMAPBASE 0xC6 /* DRAM remap base address reg (16b) */
99#define E7XXX_REMAPLIMIT 0xC8 /* DRAM remap limit address reg (16b) */
100
101/* E7XXX register addresses - device 0 function 1 */
102#define E7XXX_DRAM_FERR 0x80 /* DRAM first error register (8b) */
103#define E7XXX_DRAM_NERR 0x82 /* DRAM next error register (8b) */
104#define E7XXX_DRAM_CELOG_ADD 0xA0 /* DRAM first correctable memory */
105                    /* error address register (32b) */
106                    /*
107                     * 31:28 Reserved
108                     * 27:6 CE address (4k block 33:12)
109                     * 5:0 Reserved
110                     */
111#define E7XXX_DRAM_UELOG_ADD 0xB0 /* DRAM first uncorrectable memory */
112                    /* error address register (32b) */
113                    /*
114                     * 31:28 Reserved
115                     * 27:6 CE address (4k block 33:12)
116                     * 5:0 Reserved
117                     */
118#define E7XXX_DRAM_CELOG_SYNDROME 0xD0 /* DRAM first correctable memory */
119                    /* error syndrome register (16b) */
120
121enum e7xxx_chips {
122    E7500 = 0,
123    E7501,
124    E7505,
125    E7205,
126};
127
128struct e7xxx_pvt {
129    struct pci_dev *bridge_ck;
130    u32 tolm;
131    u32 remapbase;
132    u32 remaplimit;
133    const struct e7xxx_dev_info *dev_info;
134};
135
136struct e7xxx_dev_info {
137    u16 err_dev;
138    const char *ctl_name;
139};
140
141struct e7xxx_error_info {
142    u8 dram_ferr;
143    u8 dram_nerr;
144    u32 dram_celog_add;
145    u16 dram_celog_syndrome;
146    u32 dram_uelog_add;
147};
148
149static struct edac_pci_ctl_info *e7xxx_pci;
150
151static const struct e7xxx_dev_info e7xxx_devs[] = {
152    [E7500] = {
153        .err_dev = PCI_DEVICE_ID_INTEL_7500_1_ERR,
154        .ctl_name = "E7500"},
155    [E7501] = {
156        .err_dev = PCI_DEVICE_ID_INTEL_7501_1_ERR,
157        .ctl_name = "E7501"},
158    [E7505] = {
159        .err_dev = PCI_DEVICE_ID_INTEL_7505_1_ERR,
160        .ctl_name = "E7505"},
161    [E7205] = {
162        .err_dev = PCI_DEVICE_ID_INTEL_7205_1_ERR,
163        .ctl_name = "E7205"},
164};
165
166/* FIXME - is this valid for both SECDED and S4ECD4ED? */
167static inline int e7xxx_find_channel(u16 syndrome)
168{
169    edac_dbg(3, "\n");
170
171    if ((syndrome & 0xff00) == 0)
172        return 0;
173
174    if ((syndrome & 0x00ff) == 0)
175        return 1;
176
177    if ((syndrome & 0xf000) == 0 || (syndrome & 0x0f00) == 0)
178        return 0;
179
180    return 1;
181}
182
183static unsigned long ctl_page_to_phys(struct mem_ctl_info *mci,
184                unsigned long page)
185{
186    u32 remap;
187    struct e7xxx_pvt *pvt = (struct e7xxx_pvt *)mci->pvt_info;
188
189    edac_dbg(3, "\n");
190
191    if ((page < pvt->tolm) ||
192        ((page >= 0x100000) && (page < pvt->remapbase)))
193        return page;
194
195    remap = (page - pvt->tolm) + pvt->remapbase;
196
197    if (remap < pvt->remaplimit)
198        return remap;
199
200    e7xxx_printk(KERN_ERR, "Invalid page %lx - out of range\n", page);
201    return pvt->tolm - 1;
202}
203
204static void process_ce(struct mem_ctl_info *mci, struct e7xxx_error_info *info)
205{
206    u32 error_1b, page;
207    u16 syndrome;
208    int row;
209    int channel;
210
211    edac_dbg(3, "\n");
212    /* read the error address */
213    error_1b = info->dram_celog_add;
214    /* FIXME - should use PAGE_SHIFT */
215    page = error_1b >> 6; /* convert the address to 4k page */
216    /* read the syndrome */
217    syndrome = info->dram_celog_syndrome;
218    /* FIXME - check for -1 */
219    row = edac_mc_find_csrow_by_page(mci, page);
220    /* convert syndrome to channel */
221    channel = e7xxx_find_channel(syndrome);
222    edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1, page, 0, syndrome,
223                 row, channel, -1, "e7xxx CE", "");
224}
225
226static void process_ce_no_info(struct mem_ctl_info *mci)
227{
228    edac_dbg(3, "\n");
229    edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0, -1, -1, -1,
230                 "e7xxx CE log register overflow", "");
231}
232
233static void process_ue(struct mem_ctl_info *mci, struct e7xxx_error_info *info)
234{
235    u32 error_2b, block_page;
236    int row;
237
238    edac_dbg(3, "\n");
239    /* read the error address */
240    error_2b = info->dram_uelog_add;
241    /* FIXME - should use PAGE_SHIFT */
242    block_page = error_2b >> 6; /* convert to 4k address */
243    row = edac_mc_find_csrow_by_page(mci, block_page);
244
245    edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, block_page, 0, 0,
246                 row, -1, -1, "e7xxx UE", "");
247}
248
249static void process_ue_no_info(struct mem_ctl_info *mci)
250{
251    edac_dbg(3, "\n");
252
253    edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0, -1, -1, -1,
254                 "e7xxx UE log register overflow", "");
255}
256
257static void e7xxx_get_error_info(struct mem_ctl_info *mci,
258                 struct e7xxx_error_info *info)
259{
260    struct e7xxx_pvt *pvt;
261
262    pvt = (struct e7xxx_pvt *)mci->pvt_info;
263    pci_read_config_byte(pvt->bridge_ck, E7XXX_DRAM_FERR, &info->dram_ferr);
264    pci_read_config_byte(pvt->bridge_ck, E7XXX_DRAM_NERR, &info->dram_nerr);
265
266    if ((info->dram_ferr & 1) || (info->dram_nerr & 1)) {
267        pci_read_config_dword(pvt->bridge_ck, E7XXX_DRAM_CELOG_ADD,
268                &info->dram_celog_add);
269        pci_read_config_word(pvt->bridge_ck,
270                E7XXX_DRAM_CELOG_SYNDROME,
271                &info->dram_celog_syndrome);
272    }
273
274    if ((info->dram_ferr & 2) || (info->dram_nerr & 2))
275        pci_read_config_dword(pvt->bridge_ck, E7XXX_DRAM_UELOG_ADD,
276                &info->dram_uelog_add);
277
278    if (info->dram_ferr & 3)
279        pci_write_bits8(pvt->bridge_ck, E7XXX_DRAM_FERR, 0x03, 0x03);
280
281    if (info->dram_nerr & 3)
282        pci_write_bits8(pvt->bridge_ck, E7XXX_DRAM_NERR, 0x03, 0x03);
283}
284
285static int e7xxx_process_error_info(struct mem_ctl_info *mci,
286                struct e7xxx_error_info *info,
287                int handle_errors)
288{
289    int error_found;
290
291    error_found = 0;
292
293    /* decode and report errors */
294    if (info->dram_ferr & 1) { /* check first error correctable */
295        error_found = 1;
296
297        if (handle_errors)
298            process_ce(mci, info);
299    }
300
301    if (info->dram_ferr & 2) { /* check first error uncorrectable */
302        error_found = 1;
303
304        if (handle_errors)
305            process_ue(mci, info);
306    }
307
308    if (info->dram_nerr & 1) { /* check next error correctable */
309        error_found = 1;
310
311        if (handle_errors) {
312            if (info->dram_ferr & 1)
313                process_ce_no_info(mci);
314            else
315                process_ce(mci, info);
316        }
317    }
318
319    if (info->dram_nerr & 2) { /* check next error uncorrectable */
320        error_found = 1;
321
322        if (handle_errors) {
323            if (info->dram_ferr & 2)
324                process_ue_no_info(mci);
325            else
326                process_ue(mci, info);
327        }
328    }
329
330    return error_found;
331}
332
333static void e7xxx_check(struct mem_ctl_info *mci)
334{
335    struct e7xxx_error_info info;
336
337    edac_dbg(3, "\n");
338    e7xxx_get_error_info(mci, &info);
339    e7xxx_process_error_info(mci, &info, 1);
340}
341
342/* Return 1 if dual channel mode is active. Else return 0. */
343static inline int dual_channel_active(u32 drc, int dev_idx)
344{
345    return (dev_idx == E7501) ? ((drc >> 22) & 0x1) : 1;
346}
347
348/* Return DRB granularity (0=32mb, 1=64mb). */
349static inline int drb_granularity(u32 drc, int dev_idx)
350{
351    /* only e7501 can be single channel */
352    return (dev_idx == E7501) ? ((drc >> 18) & 0x3) : 1;
353}
354
355static void e7xxx_init_csrows(struct mem_ctl_info *mci, struct pci_dev *pdev,
356            int dev_idx, u32 drc)
357{
358    unsigned long last_cumul_size;
359    int index, j;
360    u8 value;
361    u32 dra, cumul_size, nr_pages;
362    int drc_chan, drc_drbg, drc_ddim, mem_dev;
363    struct csrow_info *csrow;
364    struct dimm_info *dimm;
365    enum edac_type edac_mode;
366
367    pci_read_config_dword(pdev, E7XXX_DRA, &dra);
368    drc_chan = dual_channel_active(drc, dev_idx);
369    drc_drbg = drb_granularity(drc, dev_idx);
370    drc_ddim = (drc >> 20) & 0x3;
371    last_cumul_size = 0;
372
373    /* The dram row boundary (DRB) reg values are boundary address
374     * for each DRAM row with a granularity of 32 or 64MB (single/dual
375     * channel operation). DRB regs are cumulative; therefore DRB7 will
376     * contain the total memory contained in all eight rows.
377     */
378    for (index = 0; index < mci->nr_csrows; index++) {
379        /* mem_dev 0=x8, 1=x4 */
380        mem_dev = (dra >> (index * 4 + 3)) & 0x1;
381        csrow = mci->csrows[index];
382
383        pci_read_config_byte(pdev, E7XXX_DRB + index, &value);
384        /* convert a 64 or 32 MiB DRB to a page size. */
385        cumul_size = value << (25 + drc_drbg - PAGE_SHIFT);
386        edac_dbg(3, "(%d) cumul_size 0x%x\n", index, cumul_size);
387        if (cumul_size == last_cumul_size)
388            continue; /* not populated */
389
390        csrow->first_page = last_cumul_size;
391        csrow->last_page = cumul_size - 1;
392        nr_pages = cumul_size - last_cumul_size;
393        last_cumul_size = cumul_size;
394
395        /*
396        * if single channel or x8 devices then SECDED
397        * if dual channel and x4 then S4ECD4ED
398        */
399        if (drc_ddim) {
400            if (drc_chan && mem_dev) {
401                edac_mode = EDAC_S4ECD4ED;
402                mci->edac_cap |= EDAC_FLAG_S4ECD4ED;
403            } else {
404                edac_mode = EDAC_SECDED;
405                mci->edac_cap |= EDAC_FLAG_SECDED;
406            }
407        } else
408            edac_mode = EDAC_NONE;
409
410        for (j = 0; j < drc_chan + 1; j++) {
411            dimm = csrow->channels[j]->dimm;
412
413            dimm->nr_pages = nr_pages / (drc_chan + 1);
414            dimm->grain = 1 << 12; /* 4KiB - resolution of CELOG */
415            dimm->mtype = MEM_RDDR; /* only one type supported */
416            dimm->dtype = mem_dev ? DEV_X4 : DEV_X8;
417            dimm->edac_mode = edac_mode;
418        }
419    }
420}
421
422static int e7xxx_probe1(struct pci_dev *pdev, int dev_idx)
423{
424    u16 pci_data;
425    struct mem_ctl_info *mci = NULL;
426    struct edac_mc_layer layers[2];
427    struct e7xxx_pvt *pvt = NULL;
428    u32 drc;
429    int drc_chan;
430    struct e7xxx_error_info discard;
431
432    edac_dbg(0, "mci\n");
433
434    pci_read_config_dword(pdev, E7XXX_DRC, &drc);
435
436    drc_chan = dual_channel_active(drc, dev_idx);
437    /*
438     * According with the datasheet, this device has a maximum of
439     * 4 DIMMS per channel, either single-rank or dual-rank. So, the
440     * total amount of dimms is 8 (E7XXX_NR_DIMMS).
441     * That means that the DIMM is mapped as CSROWs, and the channel
442     * will map the rank. So, an error to either channel should be
443     * attributed to the same dimm.
444     */
445    layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
446    layers[0].size = E7XXX_NR_CSROWS;
447    layers[0].is_virt_csrow = true;
448    layers[1].type = EDAC_MC_LAYER_CHANNEL;
449    layers[1].size = drc_chan + 1;
450    layers[1].is_virt_csrow = false;
451    mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, sizeof(*pvt));
452    if (mci == NULL)
453        return -ENOMEM;
454
455    edac_dbg(3, "init mci\n");
456    mci->mtype_cap = MEM_FLAG_RDDR;
457    mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED |
458        EDAC_FLAG_S4ECD4ED;
459    /* FIXME - what if different memory types are in different csrows? */
460    mci->mod_name = EDAC_MOD_STR;
461    mci->mod_ver = E7XXX_REVISION;
462    mci->pdev = &pdev->dev;
463    edac_dbg(3, "init pvt\n");
464    pvt = (struct e7xxx_pvt *)mci->pvt_info;
465    pvt->dev_info = &e7xxx_devs[dev_idx];
466    pvt->bridge_ck = pci_get_device(PCI_VENDOR_ID_INTEL,
467                    pvt->dev_info->err_dev, pvt->bridge_ck);
468
469    if (!pvt->bridge_ck) {
470        e7xxx_printk(KERN_ERR, "error reporting device not found:"
471            "vendor %x device 0x%x (broken BIOS?)\n",
472            PCI_VENDOR_ID_INTEL, e7xxx_devs[dev_idx].err_dev);
473        goto fail0;
474    }
475
476    edac_dbg(3, "more mci init\n");
477    mci->ctl_name = pvt->dev_info->ctl_name;
478    mci->dev_name = pci_name(pdev);
479    mci->edac_check = e7xxx_check;
480    mci->ctl_page_to_phys = ctl_page_to_phys;
481    e7xxx_init_csrows(mci, pdev, dev_idx, drc);
482    mci->edac_cap |= EDAC_FLAG_NONE;
483    edac_dbg(3, "tolm, remapbase, remaplimit\n");
484    /* load the top of low memory, remap base, and remap limit vars */
485    pci_read_config_word(pdev, E7XXX_TOLM, &pci_data);
486    pvt->tolm = ((u32) pci_data) << 4;
487    pci_read_config_word(pdev, E7XXX_REMAPBASE, &pci_data);
488    pvt->remapbase = ((u32) pci_data) << 14;
489    pci_read_config_word(pdev, E7XXX_REMAPLIMIT, &pci_data);
490    pvt->remaplimit = ((u32) pci_data) << 14;
491    e7xxx_printk(KERN_INFO,
492        "tolm = %x, remapbase = %x, remaplimit = %x\n", pvt->tolm,
493        pvt->remapbase, pvt->remaplimit);
494
495    /* clear any pending errors, or initial state bits */
496    e7xxx_get_error_info(mci, &discard);
497
498    /* Here we assume that we will never see multiple instances of this
499     * type of memory controller. The ID is therefore hardcoded to 0.
500     */
501    if (edac_mc_add_mc(mci)) {
502        edac_dbg(3, "failed edac_mc_add_mc()\n");
503        goto fail1;
504    }
505
506    /* allocating generic PCI control info */
507    e7xxx_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
508    if (!e7xxx_pci) {
509        printk(KERN_WARNING
510            "%s(): Unable to create PCI control\n",
511            __func__);
512        printk(KERN_WARNING
513            "%s(): PCI error report via EDAC not setup\n",
514            __func__);
515    }
516
517    /* get this far and it's successful */
518    edac_dbg(3, "success\n");
519    return 0;
520
521fail1:
522    pci_dev_put(pvt->bridge_ck);
523
524fail0:
525    edac_mc_free(mci);
526
527    return -ENODEV;
528}
529
530/* returns count (>= 0), or negative on error */
531static int __devinit e7xxx_init_one(struct pci_dev *pdev,
532                const struct pci_device_id *ent)
533{
534    edac_dbg(0, "\n");
535
536    /* wake up and enable device */
537    return pci_enable_device(pdev) ?
538        -EIO : e7xxx_probe1(pdev, ent->driver_data);
539}
540
541static void __devexit e7xxx_remove_one(struct pci_dev *pdev)
542{
543    struct mem_ctl_info *mci;
544    struct e7xxx_pvt *pvt;
545
546    edac_dbg(0, "\n");
547
548    if (e7xxx_pci)
549        edac_pci_release_generic_ctl(e7xxx_pci);
550
551    if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
552        return;
553
554    pvt = (struct e7xxx_pvt *)mci->pvt_info;
555    pci_dev_put(pvt->bridge_ck);
556    edac_mc_free(mci);
557}
558
559static DEFINE_PCI_DEVICE_TABLE(e7xxx_pci_tbl) = {
560    {
561     PCI_VEND_DEV(INTEL, 7205_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
562     E7205},
563    {
564     PCI_VEND_DEV(INTEL, 7500_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
565     E7500},
566    {
567     PCI_VEND_DEV(INTEL, 7501_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
568     E7501},
569    {
570     PCI_VEND_DEV(INTEL, 7505_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
571     E7505},
572    {
573     0,
574     } /* 0 terminated list. */
575};
576
577MODULE_DEVICE_TABLE(pci, e7xxx_pci_tbl);
578
579static struct pci_driver e7xxx_driver = {
580    .name = EDAC_MOD_STR,
581    .probe = e7xxx_init_one,
582    .remove = __devexit_p(e7xxx_remove_one),
583    .id_table = e7xxx_pci_tbl,
584};
585
586static int __init e7xxx_init(void)
587{
588       /* Ensure that the OPSTATE is set correctly for POLL or NMI */
589       opstate_init();
590
591    return pci_register_driver(&e7xxx_driver);
592}
593
594static void __exit e7xxx_exit(void)
595{
596    pci_unregister_driver(&e7xxx_driver);
597}
598
599module_init(e7xxx_init);
600module_exit(e7xxx_exit);
601
602MODULE_LICENSE("GPL");
603MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh et al\n"
604        "Based on.work by Dan Hollis et al");
605MODULE_DESCRIPTION("MC support for Intel e7xxx memory controllers");
606module_param(edac_op_state, int, 0444);
607MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
608

Archive Download this file



interactive