procfs: make proc_get_link to use dentry instead of inode
[pohmelfs.git] / drivers / edac / i7core_edac.c
blob8568d9b618750e211c79b2d25e30827455e0d040
1 /* Intel i7 core/Nehalem Memory Controller kernel module
3 * This driver supports the memory controllers found on the Intel
4 * processor families i7core, i7core 7xx/8xx, i5core, Xeon 35xx,
5 * Xeon 55xx and Xeon 56xx also known as Nehalem, Nehalem-EP, Lynnfield
6 * and Westmere-EP.
8 * This file may be distributed under the terms of the
9 * GNU General Public License version 2 only.
11 * Copyright (c) 2009-2010 by:
12 * Mauro Carvalho Chehab <mchehab@redhat.com>
14 * Red Hat Inc. http://www.redhat.com
16 * Forked and adapted from the i5400_edac driver
18 * Based on the following public Intel datasheets:
19 * Intel Core i7 Processor Extreme Edition and Intel Core i7 Processor
20 * Datasheet, Volume 2:
21 * http://download.intel.com/design/processor/datashts/320835.pdf
22 * Intel Xeon Processor 5500 Series Datasheet Volume 2
23 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
24 * also available at:
25 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
28 #include <linux/module.h>
29 #include <linux/init.h>
30 #include <linux/pci.h>
31 #include <linux/pci_ids.h>
32 #include <linux/slab.h>
33 #include <linux/delay.h>
34 #include <linux/dmi.h>
35 #include <linux/edac.h>
36 #include <linux/mmzone.h>
37 #include <linux/smp.h>
38 #include <asm/mce.h>
39 #include <asm/processor.h>
40 #include <asm/div64.h>
42 #include "edac_core.h"
44 /* Static vars */
45 static LIST_HEAD(i7core_edac_list);
46 static DEFINE_MUTEX(i7core_edac_lock);
47 static int probed;
49 static int use_pci_fixup;
50 module_param(use_pci_fixup, int, 0444);
51 MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices");
53 * This is used for Nehalem-EP and Nehalem-EX devices, where the non-core
54 * registers start at bus 255, and are not reported by BIOS.
55 * We currently find devices with only 2 sockets. In order to support more QPI
56 * Quick Path Interconnect, just increment this number.
58 #define MAX_SOCKET_BUSES 2
62 * Alter this version for the module when modifications are made
64 #define I7CORE_REVISION " Ver: 1.0.0"
65 #define EDAC_MOD_STR "i7core_edac"
68 * Debug macros
70 #define i7core_printk(level, fmt, arg...) \
71 edac_printk(level, "i7core", fmt, ##arg)
73 #define i7core_mc_printk(mci, level, fmt, arg...) \
74 edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
77 * i7core Memory Controller Registers
80 /* OFFSETS for Device 0 Function 0 */
82 #define MC_CFG_CONTROL 0x90
83 #define MC_CFG_UNLOCK 0x02
84 #define MC_CFG_LOCK 0x00
86 /* OFFSETS for Device 3 Function 0 */
88 #define MC_CONTROL 0x48
89 #define MC_STATUS 0x4c
90 #define MC_MAX_DOD 0x64
93 * OFFSETS for Device 3 Function 4, as inicated on Xeon 5500 datasheet:
94 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
97 #define MC_TEST_ERR_RCV1 0x60
98 #define DIMM2_COR_ERR(r) ((r) & 0x7fff)
100 #define MC_TEST_ERR_RCV0 0x64
101 #define DIMM1_COR_ERR(r) (((r) >> 16) & 0x7fff)
102 #define DIMM0_COR_ERR(r) ((r) & 0x7fff)
104 /* OFFSETS for Device 3 Function 2, as inicated on Xeon 5500 datasheet */
105 #define MC_SSRCONTROL 0x48
106 #define SSR_MODE_DISABLE 0x00
107 #define SSR_MODE_ENABLE 0x01
108 #define SSR_MODE_MASK 0x03
110 #define MC_SCRUB_CONTROL 0x4c
111 #define STARTSCRUB (1 << 24)
112 #define SCRUBINTERVAL_MASK 0xffffff
114 #define MC_COR_ECC_CNT_0 0x80
115 #define MC_COR_ECC_CNT_1 0x84
116 #define MC_COR_ECC_CNT_2 0x88
117 #define MC_COR_ECC_CNT_3 0x8c
118 #define MC_COR_ECC_CNT_4 0x90
119 #define MC_COR_ECC_CNT_5 0x94
121 #define DIMM_TOP_COR_ERR(r) (((r) >> 16) & 0x7fff)
122 #define DIMM_BOT_COR_ERR(r) ((r) & 0x7fff)
125 /* OFFSETS for Devices 4,5 and 6 Function 0 */
127 #define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
128 #define THREE_DIMMS_PRESENT (1 << 24)
129 #define SINGLE_QUAD_RANK_PRESENT (1 << 23)
130 #define QUAD_RANK_PRESENT (1 << 22)
131 #define REGISTERED_DIMM (1 << 15)
133 #define MC_CHANNEL_MAPPER 0x60
134 #define RDLCH(r, ch) ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
135 #define WRLCH(r, ch) ((((r) >> (ch * 6)) & 0x07) - 1)
137 #define MC_CHANNEL_RANK_PRESENT 0x7c
138 #define RANK_PRESENT_MASK 0xffff
140 #define MC_CHANNEL_ADDR_MATCH 0xf0
141 #define MC_CHANNEL_ERROR_MASK 0xf8
142 #define MC_CHANNEL_ERROR_INJECT 0xfc
143 #define INJECT_ADDR_PARITY 0x10
144 #define INJECT_ECC 0x08
145 #define MASK_CACHELINE 0x06
146 #define MASK_FULL_CACHELINE 0x06
147 #define MASK_MSB32_CACHELINE 0x04
148 #define MASK_LSB32_CACHELINE 0x02
149 #define NO_MASK_CACHELINE 0x00
150 #define REPEAT_EN 0x01
152 /* OFFSETS for Devices 4,5 and 6 Function 1 */
154 #define MC_DOD_CH_DIMM0 0x48
155 #define MC_DOD_CH_DIMM1 0x4c
156 #define MC_DOD_CH_DIMM2 0x50
157 #define RANKOFFSET_MASK ((1 << 12) | (1 << 11) | (1 << 10))
158 #define RANKOFFSET(x) ((x & RANKOFFSET_MASK) >> 10)
159 #define DIMM_PRESENT_MASK (1 << 9)
160 #define DIMM_PRESENT(x) (((x) & DIMM_PRESENT_MASK) >> 9)
161 #define MC_DOD_NUMBANK_MASK ((1 << 8) | (1 << 7))
162 #define MC_DOD_NUMBANK(x) (((x) & MC_DOD_NUMBANK_MASK) >> 7)
163 #define MC_DOD_NUMRANK_MASK ((1 << 6) | (1 << 5))
164 #define MC_DOD_NUMRANK(x) (((x) & MC_DOD_NUMRANK_MASK) >> 5)
165 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3) | (1 << 2))
166 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2)
167 #define MC_DOD_NUMCOL_MASK 3
168 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK)
170 #define MC_RANK_PRESENT 0x7c
172 #define MC_SAG_CH_0 0x80
173 #define MC_SAG_CH_1 0x84
174 #define MC_SAG_CH_2 0x88
175 #define MC_SAG_CH_3 0x8c
176 #define MC_SAG_CH_4 0x90
177 #define MC_SAG_CH_5 0x94
178 #define MC_SAG_CH_6 0x98
179 #define MC_SAG_CH_7 0x9c
181 #define MC_RIR_LIMIT_CH_0 0x40
182 #define MC_RIR_LIMIT_CH_1 0x44
183 #define MC_RIR_LIMIT_CH_2 0x48
184 #define MC_RIR_LIMIT_CH_3 0x4C
185 #define MC_RIR_LIMIT_CH_4 0x50
186 #define MC_RIR_LIMIT_CH_5 0x54
187 #define MC_RIR_LIMIT_CH_6 0x58
188 #define MC_RIR_LIMIT_CH_7 0x5C
189 #define MC_RIR_LIMIT_MASK ((1 << 10) - 1)
191 #define MC_RIR_WAY_CH 0x80
192 #define MC_RIR_WAY_OFFSET_MASK (((1 << 14) - 1) & ~0x7)
193 #define MC_RIR_WAY_RANK_MASK 0x7
196 * i7core structs
199 #define NUM_CHANS 3
200 #define MAX_DIMMS 3 /* Max DIMMS per channel */
201 #define MAX_MCR_FUNC 4
202 #define MAX_CHAN_FUNC 3
204 struct i7core_info {
205 u32 mc_control;
206 u32 mc_status;
207 u32 max_dod;
208 u32 ch_map;
212 struct i7core_inject {
213 int enable;
215 u32 section;
216 u32 type;
217 u32 eccmask;
219 /* Error address mask */
220 int channel, dimm, rank, bank, page, col;
223 struct i7core_channel {
224 u32 ranks;
225 u32 dimms;
228 struct pci_id_descr {
229 int dev;
230 int func;
231 int dev_id;
232 int optional;
235 struct pci_id_table {
236 const struct pci_id_descr *descr;
237 int n_devs;
240 struct i7core_dev {
241 struct list_head list;
242 u8 socket;
243 struct pci_dev **pdev;
244 int n_devs;
245 struct mem_ctl_info *mci;
248 struct i7core_pvt {
249 struct pci_dev *pci_noncore;
250 struct pci_dev *pci_mcr[MAX_MCR_FUNC + 1];
251 struct pci_dev *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
253 struct i7core_dev *i7core_dev;
255 struct i7core_info info;
256 struct i7core_inject inject;
257 struct i7core_channel channel[NUM_CHANS];
259 int ce_count_available;
260 int csrow_map[NUM_CHANS][MAX_DIMMS];
262 /* ECC corrected errors counts per udimm */
263 unsigned long udimm_ce_count[MAX_DIMMS];
264 int udimm_last_ce_count[MAX_DIMMS];
265 /* ECC corrected errors counts per rdimm */
266 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
267 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
269 bool is_registered, enable_scrub;
271 /* Fifo double buffers */
272 struct mce mce_entry[MCE_LOG_LEN];
273 struct mce mce_outentry[MCE_LOG_LEN];
275 /* Fifo in/out counters */
276 unsigned mce_in, mce_out;
278 /* Count indicator to show errors not got */
279 unsigned mce_overrun;
281 /* DCLK Frequency used for computing scrub rate */
282 int dclk_freq;
284 /* Struct to control EDAC polling */
285 struct edac_pci_ctl_info *i7core_pci;
288 #define PCI_DESCR(device, function, device_id) \
289 .dev = (device), \
290 .func = (function), \
291 .dev_id = (device_id)
293 static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
294 /* Memory controller */
295 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
296 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
297 /* Exists only for RDIMM */
298 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 },
299 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
301 /* Channel 0 */
302 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
303 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
304 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
305 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) },
307 /* Channel 1 */
308 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
309 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
310 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
311 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) },
313 /* Channel 2 */
314 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
315 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
316 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
317 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
319 /* Generic Non-core registers */
321 * This is the PCI device on i7core and on Xeon 35xx (8086:2c41)
322 * On Xeon 55xx, however, it has a different id (8086:2c40). So,
323 * the probing code needs to test for the other address in case of
324 * failure of this one
326 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) },
330 static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
331 { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR) },
332 { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD) },
333 { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST) },
335 { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) },
336 { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) },
337 { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) },
338 { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC) },
340 { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) },
341 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
342 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
343 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) },
346 * This is the PCI device has an alternate address on some
347 * processors like Core i7 860
349 { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) },
352 static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
353 /* Memory controller */
354 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2) },
355 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2) },
356 /* Exists only for RDIMM */
357 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1 },
358 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) },
360 /* Channel 0 */
361 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) },
362 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) },
363 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) },
364 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2) },
366 /* Channel 1 */
367 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) },
368 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) },
369 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) },
370 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2) },
372 /* Channel 2 */
373 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) },
374 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
375 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
376 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) },
378 /* Generic Non-core registers */
379 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) },
383 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
384 static const struct pci_id_table pci_dev_table[] = {
385 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem),
386 PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield),
387 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere),
388 {0,} /* 0 terminated list. */
392 * pci_device_id table for which devices we are looking for
394 static const struct pci_device_id i7core_pci_tbl[] __devinitdata = {
395 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
396 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
397 {0,} /* 0 terminated list. */
400 /****************************************************************************
401 Anciliary status routines
402 ****************************************************************************/
404 /* MC_CONTROL bits */
405 #define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
406 #define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
408 /* MC_STATUS bits */
409 #define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 4))
410 #define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
412 /* MC_MAX_DOD read functions */
413 static inline int numdimms(u32 dimms)
415 return (dimms & 0x3) + 1;
418 static inline int numrank(u32 rank)
420 static int ranks[4] = { 1, 2, 4, -EINVAL };
422 return ranks[rank & 0x3];
425 static inline int numbank(u32 bank)
427 static int banks[4] = { 4, 8, 16, -EINVAL };
429 return banks[bank & 0x3];
432 static inline int numrow(u32 row)
434 static int rows[8] = {
435 1 << 12, 1 << 13, 1 << 14, 1 << 15,
436 1 << 16, -EINVAL, -EINVAL, -EINVAL,
439 return rows[row & 0x7];
442 static inline int numcol(u32 col)
444 static int cols[8] = {
445 1 << 10, 1 << 11, 1 << 12, -EINVAL,
447 return cols[col & 0x3];
450 static struct i7core_dev *get_i7core_dev(u8 socket)
452 struct i7core_dev *i7core_dev;
454 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
455 if (i7core_dev->socket == socket)
456 return i7core_dev;
459 return NULL;
462 static struct i7core_dev *alloc_i7core_dev(u8 socket,
463 const struct pci_id_table *table)
465 struct i7core_dev *i7core_dev;
467 i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
468 if (!i7core_dev)
469 return NULL;
471 i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * table->n_devs,
472 GFP_KERNEL);
473 if (!i7core_dev->pdev) {
474 kfree(i7core_dev);
475 return NULL;
478 i7core_dev->socket = socket;
479 i7core_dev->n_devs = table->n_devs;
480 list_add_tail(&i7core_dev->list, &i7core_edac_list);
482 return i7core_dev;
485 static void free_i7core_dev(struct i7core_dev *i7core_dev)
487 list_del(&i7core_dev->list);
488 kfree(i7core_dev->pdev);
489 kfree(i7core_dev);
492 /****************************************************************************
493 Memory check routines
494 ****************************************************************************/
495 static struct pci_dev *get_pdev_slot_func(u8 socket, unsigned slot,
496 unsigned func)
498 struct i7core_dev *i7core_dev = get_i7core_dev(socket);
499 int i;
501 if (!i7core_dev)
502 return NULL;
504 for (i = 0; i < i7core_dev->n_devs; i++) {
505 if (!i7core_dev->pdev[i])
506 continue;
508 if (PCI_SLOT(i7core_dev->pdev[i]->devfn) == slot &&
509 PCI_FUNC(i7core_dev->pdev[i]->devfn) == func) {
510 return i7core_dev->pdev[i];
514 return NULL;
518 * i7core_get_active_channels() - gets the number of channels and csrows
519 * @socket: Quick Path Interconnect socket
520 * @channels: Number of channels that will be returned
521 * @csrows: Number of csrows found
523 * Since EDAC core needs to know in advance the number of available channels
524 * and csrows, in order to allocate memory for csrows/channels, it is needed
525 * to run two similar steps. At the first step, implemented on this function,
526 * it checks the number of csrows/channels present at one socket.
527 * this is used in order to properly allocate the size of mci components.
529 * It should be noticed that none of the current available datasheets explain
530 * or even mention how csrows are seen by the memory controller. So, we need
531 * to add a fake description for csrows.
532 * So, this driver is attributing one DIMM memory for one csrow.
534 static int i7core_get_active_channels(const u8 socket, unsigned *channels,
535 unsigned *csrows)
537 struct pci_dev *pdev = NULL;
538 int i, j;
539 u32 status, control;
541 *channels = 0;
542 *csrows = 0;
544 pdev = get_pdev_slot_func(socket, 3, 0);
545 if (!pdev) {
546 i7core_printk(KERN_ERR, "Couldn't find socket %d fn 3.0!!!\n",
547 socket);
548 return -ENODEV;
551 /* Device 3 function 0 reads */
552 pci_read_config_dword(pdev, MC_STATUS, &status);
553 pci_read_config_dword(pdev, MC_CONTROL, &control);
555 for (i = 0; i < NUM_CHANS; i++) {
556 u32 dimm_dod[3];
557 /* Check if the channel is active */
558 if (!(control & (1 << (8 + i))))
559 continue;
561 /* Check if the channel is disabled */
562 if (status & (1 << i))
563 continue;
565 pdev = get_pdev_slot_func(socket, i + 4, 1);
566 if (!pdev) {
567 i7core_printk(KERN_ERR, "Couldn't find socket %d "
568 "fn %d.%d!!!\n",
569 socket, i + 4, 1);
570 return -ENODEV;
572 /* Devices 4-6 function 1 */
573 pci_read_config_dword(pdev,
574 MC_DOD_CH_DIMM0, &dimm_dod[0]);
575 pci_read_config_dword(pdev,
576 MC_DOD_CH_DIMM1, &dimm_dod[1]);
577 pci_read_config_dword(pdev,
578 MC_DOD_CH_DIMM2, &dimm_dod[2]);
580 (*channels)++;
582 for (j = 0; j < 3; j++) {
583 if (!DIMM_PRESENT(dimm_dod[j]))
584 continue;
585 (*csrows)++;
589 debugf0("Number of active channels on socket %d: %d\n",
590 socket, *channels);
592 return 0;
595 static int get_dimm_config(const struct mem_ctl_info *mci)
597 struct i7core_pvt *pvt = mci->pvt_info;
598 struct csrow_info *csr;
599 struct pci_dev *pdev;
600 int i, j;
601 int csrow = 0;
602 unsigned long last_page = 0;
603 enum edac_type mode;
604 enum mem_type mtype;
606 /* Get data from the MC register, function 0 */
607 pdev = pvt->pci_mcr[0];
608 if (!pdev)
609 return -ENODEV;
611 /* Device 3 function 0 reads */
612 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
613 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
614 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
615 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
617 debugf0("QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
618 pvt->i7core_dev->socket, pvt->info.mc_control, pvt->info.mc_status,
619 pvt->info.max_dod, pvt->info.ch_map);
621 if (ECC_ENABLED(pvt)) {
622 debugf0("ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
623 if (ECCx8(pvt))
624 mode = EDAC_S8ECD8ED;
625 else
626 mode = EDAC_S4ECD4ED;
627 } else {
628 debugf0("ECC disabled\n");
629 mode = EDAC_NONE;
632 /* FIXME: need to handle the error codes */
633 debugf0("DOD Max limits: DIMMS: %d, %d-ranked, %d-banked "
634 "x%x x 0x%x\n",
635 numdimms(pvt->info.max_dod),
636 numrank(pvt->info.max_dod >> 2),
637 numbank(pvt->info.max_dod >> 4),
638 numrow(pvt->info.max_dod >> 6),
639 numcol(pvt->info.max_dod >> 9));
641 for (i = 0; i < NUM_CHANS; i++) {
642 u32 data, dimm_dod[3], value[8];
644 if (!pvt->pci_ch[i][0])
645 continue;
647 if (!CH_ACTIVE(pvt, i)) {
648 debugf0("Channel %i is not active\n", i);
649 continue;
651 if (CH_DISABLED(pvt, i)) {
652 debugf0("Channel %i is disabled\n", i);
653 continue;
656 /* Devices 4-6 function 0 */
657 pci_read_config_dword(pvt->pci_ch[i][0],
658 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
660 pvt->channel[i].ranks = (data & QUAD_RANK_PRESENT) ?
661 4 : 2;
663 if (data & REGISTERED_DIMM)
664 mtype = MEM_RDDR3;
665 else
666 mtype = MEM_DDR3;
667 #if 0
668 if (data & THREE_DIMMS_PRESENT)
669 pvt->channel[i].dimms = 3;
670 else if (data & SINGLE_QUAD_RANK_PRESENT)
671 pvt->channel[i].dimms = 1;
672 else
673 pvt->channel[i].dimms = 2;
674 #endif
676 /* Devices 4-6 function 1 */
677 pci_read_config_dword(pvt->pci_ch[i][1],
678 MC_DOD_CH_DIMM0, &dimm_dod[0]);
679 pci_read_config_dword(pvt->pci_ch[i][1],
680 MC_DOD_CH_DIMM1, &dimm_dod[1]);
681 pci_read_config_dword(pvt->pci_ch[i][1],
682 MC_DOD_CH_DIMM2, &dimm_dod[2]);
684 debugf0("Ch%d phy rd%d, wr%d (0x%08x): "
685 "%d ranks, %cDIMMs\n",
687 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
688 data,
689 pvt->channel[i].ranks,
690 (data & REGISTERED_DIMM) ? 'R' : 'U');
692 for (j = 0; j < 3; j++) {
693 u32 banks, ranks, rows, cols;
694 u32 size, npages;
696 if (!DIMM_PRESENT(dimm_dod[j]))
697 continue;
699 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
700 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
701 rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
702 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
704 /* DDR3 has 8 I/O banks */
705 size = (rows * cols * banks * ranks) >> (20 - 3);
707 pvt->channel[i].dimms++;
709 debugf0("\tdimm %d %d Mb offset: %x, "
710 "bank: %d, rank: %d, row: %#x, col: %#x\n",
711 j, size,
712 RANKOFFSET(dimm_dod[j]),
713 banks, ranks, rows, cols);
715 npages = MiB_TO_PAGES(size);
717 csr = &mci->csrows[csrow];
718 csr->first_page = last_page + 1;
719 last_page += npages;
720 csr->last_page = last_page;
721 csr->nr_pages = npages;
723 csr->page_mask = 0;
724 csr->grain = 8;
725 csr->csrow_idx = csrow;
726 csr->nr_channels = 1;
728 csr->channels[0].chan_idx = i;
729 csr->channels[0].ce_count = 0;
731 pvt->csrow_map[i][j] = csrow;
733 switch (banks) {
734 case 4:
735 csr->dtype = DEV_X4;
736 break;
737 case 8:
738 csr->dtype = DEV_X8;
739 break;
740 case 16:
741 csr->dtype = DEV_X16;
742 break;
743 default:
744 csr->dtype = DEV_UNKNOWN;
747 csr->edac_mode = mode;
748 csr->mtype = mtype;
749 snprintf(csr->channels[0].label,
750 sizeof(csr->channels[0].label),
751 "CPU#%uChannel#%u_DIMM#%u",
752 pvt->i7core_dev->socket, i, j);
754 csrow++;
757 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
758 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
759 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
760 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
761 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
762 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
763 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
764 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
765 debugf1("\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
766 for (j = 0; j < 8; j++)
767 debugf1("\t\t%#x\t%#x\t%#x\n",
768 (value[j] >> 27) & 0x1,
769 (value[j] >> 24) & 0x7,
770 (value[j] & ((1 << 24) - 1)));
773 return 0;
776 /****************************************************************************
777 Error insertion routines
778 ****************************************************************************/
780 /* The i7core has independent error injection features per channel.
781 However, to have a simpler code, we don't allow enabling error injection
782 on more than one channel.
783 Also, since a change at an inject parameter will be applied only at enable,
784 we're disabling error injection on all write calls to the sysfs nodes that
785 controls the error code injection.
787 static int disable_inject(const struct mem_ctl_info *mci)
789 struct i7core_pvt *pvt = mci->pvt_info;
791 pvt->inject.enable = 0;
793 if (!pvt->pci_ch[pvt->inject.channel][0])
794 return -ENODEV;
796 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
797 MC_CHANNEL_ERROR_INJECT, 0);
799 return 0;
803 * i7core inject inject.section
805 * accept and store error injection inject.section value
806 * bit 0 - refers to the lower 32-byte half cacheline
807 * bit 1 - refers to the upper 32-byte half cacheline
809 static ssize_t i7core_inject_section_store(struct mem_ctl_info *mci,
810 const char *data, size_t count)
812 struct i7core_pvt *pvt = mci->pvt_info;
813 unsigned long value;
814 int rc;
816 if (pvt->inject.enable)
817 disable_inject(mci);
819 rc = strict_strtoul(data, 10, &value);
820 if ((rc < 0) || (value > 3))
821 return -EIO;
823 pvt->inject.section = (u32) value;
824 return count;
827 static ssize_t i7core_inject_section_show(struct mem_ctl_info *mci,
828 char *data)
830 struct i7core_pvt *pvt = mci->pvt_info;
831 return sprintf(data, "0x%08x\n", pvt->inject.section);
835 * i7core inject.type
837 * accept and store error injection inject.section value
838 * bit 0 - repeat enable - Enable error repetition
839 * bit 1 - inject ECC error
840 * bit 2 - inject parity error
842 static ssize_t i7core_inject_type_store(struct mem_ctl_info *mci,
843 const char *data, size_t count)
845 struct i7core_pvt *pvt = mci->pvt_info;
846 unsigned long value;
847 int rc;
849 if (pvt->inject.enable)
850 disable_inject(mci);
852 rc = strict_strtoul(data, 10, &value);
853 if ((rc < 0) || (value > 7))
854 return -EIO;
856 pvt->inject.type = (u32) value;
857 return count;
860 static ssize_t i7core_inject_type_show(struct mem_ctl_info *mci,
861 char *data)
863 struct i7core_pvt *pvt = mci->pvt_info;
864 return sprintf(data, "0x%08x\n", pvt->inject.type);
868 * i7core_inject_inject.eccmask_store
870 * The type of error (UE/CE) will depend on the inject.eccmask value:
871 * Any bits set to a 1 will flip the corresponding ECC bit
872 * Correctable errors can be injected by flipping 1 bit or the bits within
873 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
874 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
875 * uncorrectable error to be injected.
877 static ssize_t i7core_inject_eccmask_store(struct mem_ctl_info *mci,
878 const char *data, size_t count)
880 struct i7core_pvt *pvt = mci->pvt_info;
881 unsigned long value;
882 int rc;
884 if (pvt->inject.enable)
885 disable_inject(mci);
887 rc = strict_strtoul(data, 10, &value);
888 if (rc < 0)
889 return -EIO;
891 pvt->inject.eccmask = (u32) value;
892 return count;
895 static ssize_t i7core_inject_eccmask_show(struct mem_ctl_info *mci,
896 char *data)
898 struct i7core_pvt *pvt = mci->pvt_info;
899 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
903 * i7core_addrmatch
905 * The type of error (UE/CE) will depend on the inject.eccmask value:
906 * Any bits set to a 1 will flip the corresponding ECC bit
907 * Correctable errors can be injected by flipping 1 bit or the bits within
908 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
909 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
910 * uncorrectable error to be injected.
913 #define DECLARE_ADDR_MATCH(param, limit) \
914 static ssize_t i7core_inject_store_##param( \
915 struct mem_ctl_info *mci, \
916 const char *data, size_t count) \
918 struct i7core_pvt *pvt; \
919 long value; \
920 int rc; \
922 debugf1("%s()\n", __func__); \
923 pvt = mci->pvt_info; \
925 if (pvt->inject.enable) \
926 disable_inject(mci); \
928 if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
929 value = -1; \
930 else { \
931 rc = strict_strtoul(data, 10, &value); \
932 if ((rc < 0) || (value >= limit)) \
933 return -EIO; \
936 pvt->inject.param = value; \
938 return count; \
941 static ssize_t i7core_inject_show_##param( \
942 struct mem_ctl_info *mci, \
943 char *data) \
945 struct i7core_pvt *pvt; \
947 pvt = mci->pvt_info; \
948 debugf1("%s() pvt=%p\n", __func__, pvt); \
949 if (pvt->inject.param < 0) \
950 return sprintf(data, "any\n"); \
951 else \
952 return sprintf(data, "%d\n", pvt->inject.param);\
955 #define ATTR_ADDR_MATCH(param) \
957 .attr = { \
958 .name = #param, \
959 .mode = (S_IRUGO | S_IWUSR) \
960 }, \
961 .show = i7core_inject_show_##param, \
962 .store = i7core_inject_store_##param, \
965 DECLARE_ADDR_MATCH(channel, 3);
966 DECLARE_ADDR_MATCH(dimm, 3);
967 DECLARE_ADDR_MATCH(rank, 4);
968 DECLARE_ADDR_MATCH(bank, 32);
969 DECLARE_ADDR_MATCH(page, 0x10000);
970 DECLARE_ADDR_MATCH(col, 0x4000);
972 static int write_and_test(struct pci_dev *dev, const int where, const u32 val)
974 u32 read;
975 int count;
977 debugf0("setting pci %02x:%02x.%x reg=%02x value=%08x\n",
978 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
979 where, val);
981 for (count = 0; count < 10; count++) {
982 if (count)
983 msleep(100);
984 pci_write_config_dword(dev, where, val);
985 pci_read_config_dword(dev, where, &read);
987 if (read == val)
988 return 0;
991 i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x "
992 "write=%08x. Read=%08x\n",
993 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
994 where, val, read);
996 return -EINVAL;
1000 * This routine prepares the Memory Controller for error injection.
1001 * The error will be injected when some process tries to write to the
1002 * memory that matches the given criteria.
1003 * The criteria can be set in terms of a mask where dimm, rank, bank, page
1004 * and col can be specified.
1005 * A -1 value for any of the mask items will make the MCU to ignore
1006 * that matching criteria for error injection.
1008 * It should be noticed that the error will only happen after a write operation
1009 * on a memory that matches the condition. if REPEAT_EN is not enabled at
1010 * inject mask, then it will produce just one error. Otherwise, it will repeat
1011 * until the injectmask would be cleaned.
1013 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
1014 * is reliable enough to check if the MC is using the
1015 * three channels. However, this is not clear at the datasheet.
1017 static ssize_t i7core_inject_enable_store(struct mem_ctl_info *mci,
1018 const char *data, size_t count)
1020 struct i7core_pvt *pvt = mci->pvt_info;
1021 u32 injectmask;
1022 u64 mask = 0;
1023 int rc;
1024 long enable;
1026 if (!pvt->pci_ch[pvt->inject.channel][0])
1027 return 0;
1029 rc = strict_strtoul(data, 10, &enable);
1030 if ((rc < 0))
1031 return 0;
1033 if (enable) {
1034 pvt->inject.enable = 1;
1035 } else {
1036 disable_inject(mci);
1037 return count;
1040 /* Sets pvt->inject.dimm mask */
1041 if (pvt->inject.dimm < 0)
1042 mask |= 1LL << 41;
1043 else {
1044 if (pvt->channel[pvt->inject.channel].dimms > 2)
1045 mask |= (pvt->inject.dimm & 0x3LL) << 35;
1046 else
1047 mask |= (pvt->inject.dimm & 0x1LL) << 36;
1050 /* Sets pvt->inject.rank mask */
1051 if (pvt->inject.rank < 0)
1052 mask |= 1LL << 40;
1053 else {
1054 if (pvt->channel[pvt->inject.channel].dimms > 2)
1055 mask |= (pvt->inject.rank & 0x1LL) << 34;
1056 else
1057 mask |= (pvt->inject.rank & 0x3LL) << 34;
1060 /* Sets pvt->inject.bank mask */
1061 if (pvt->inject.bank < 0)
1062 mask |= 1LL << 39;
1063 else
1064 mask |= (pvt->inject.bank & 0x15LL) << 30;
1066 /* Sets pvt->inject.page mask */
1067 if (pvt->inject.page < 0)
1068 mask |= 1LL << 38;
1069 else
1070 mask |= (pvt->inject.page & 0xffff) << 14;
1072 /* Sets pvt->inject.column mask */
1073 if (pvt->inject.col < 0)
1074 mask |= 1LL << 37;
1075 else
1076 mask |= (pvt->inject.col & 0x3fff);
1079 * bit 0: REPEAT_EN
1080 * bits 1-2: MASK_HALF_CACHELINE
1081 * bit 3: INJECT_ECC
1082 * bit 4: INJECT_ADDR_PARITY
1085 injectmask = (pvt->inject.type & 1) |
1086 (pvt->inject.section & 0x3) << 1 |
1087 (pvt->inject.type & 0x6) << (3 - 1);
1089 /* Unlock writes to registers - this register is write only */
1090 pci_write_config_dword(pvt->pci_noncore,
1091 MC_CFG_CONTROL, 0x2);
1093 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1094 MC_CHANNEL_ADDR_MATCH, mask);
1095 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1096 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
1098 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1099 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
1101 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1102 MC_CHANNEL_ERROR_INJECT, injectmask);
1105 * This is something undocumented, based on my tests
1106 * Without writing 8 to this register, errors aren't injected. Not sure
1107 * why.
1109 pci_write_config_dword(pvt->pci_noncore,
1110 MC_CFG_CONTROL, 8);
1112 debugf0("Error inject addr match 0x%016llx, ecc 0x%08x,"
1113 " inject 0x%08x\n",
1114 mask, pvt->inject.eccmask, injectmask);
1117 return count;
1120 static ssize_t i7core_inject_enable_show(struct mem_ctl_info *mci,
1121 char *data)
1123 struct i7core_pvt *pvt = mci->pvt_info;
1124 u32 injectmask;
1126 if (!pvt->pci_ch[pvt->inject.channel][0])
1127 return 0;
1129 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
1130 MC_CHANNEL_ERROR_INJECT, &injectmask);
1132 debugf0("Inject error read: 0x%018x\n", injectmask);
1134 if (injectmask & 0x0c)
1135 pvt->inject.enable = 1;
1137 return sprintf(data, "%d\n", pvt->inject.enable);
1140 #define DECLARE_COUNTER(param) \
1141 static ssize_t i7core_show_counter_##param( \
1142 struct mem_ctl_info *mci, \
1143 char *data) \
1145 struct i7core_pvt *pvt = mci->pvt_info; \
1147 debugf1("%s() \n", __func__); \
1148 if (!pvt->ce_count_available || (pvt->is_registered)) \
1149 return sprintf(data, "data unavailable\n"); \
1150 return sprintf(data, "%lu\n", \
1151 pvt->udimm_ce_count[param]); \
1154 #define ATTR_COUNTER(param) \
1156 .attr = { \
1157 .name = __stringify(udimm##param), \
1158 .mode = (S_IRUGO | S_IWUSR) \
1159 }, \
1160 .show = i7core_show_counter_##param \
1163 DECLARE_COUNTER(0);
1164 DECLARE_COUNTER(1);
1165 DECLARE_COUNTER(2);
1168 * Sysfs struct
1171 static const struct mcidev_sysfs_attribute i7core_addrmatch_attrs[] = {
1172 ATTR_ADDR_MATCH(channel),
1173 ATTR_ADDR_MATCH(dimm),
1174 ATTR_ADDR_MATCH(rank),
1175 ATTR_ADDR_MATCH(bank),
1176 ATTR_ADDR_MATCH(page),
1177 ATTR_ADDR_MATCH(col),
1178 { } /* End of list */
1181 static const struct mcidev_sysfs_group i7core_inject_addrmatch = {
1182 .name = "inject_addrmatch",
1183 .mcidev_attr = i7core_addrmatch_attrs,
1186 static const struct mcidev_sysfs_attribute i7core_udimm_counters_attrs[] = {
1187 ATTR_COUNTER(0),
1188 ATTR_COUNTER(1),
1189 ATTR_COUNTER(2),
1190 { .attr = { .name = NULL } }
1193 static const struct mcidev_sysfs_group i7core_udimm_counters = {
1194 .name = "all_channel_counts",
1195 .mcidev_attr = i7core_udimm_counters_attrs,
1198 static const struct mcidev_sysfs_attribute i7core_sysfs_rdimm_attrs[] = {
1200 .attr = {
1201 .name = "inject_section",
1202 .mode = (S_IRUGO | S_IWUSR)
1204 .show = i7core_inject_section_show,
1205 .store = i7core_inject_section_store,
1206 }, {
1207 .attr = {
1208 .name = "inject_type",
1209 .mode = (S_IRUGO | S_IWUSR)
1211 .show = i7core_inject_type_show,
1212 .store = i7core_inject_type_store,
1213 }, {
1214 .attr = {
1215 .name = "inject_eccmask",
1216 .mode = (S_IRUGO | S_IWUSR)
1218 .show = i7core_inject_eccmask_show,
1219 .store = i7core_inject_eccmask_store,
1220 }, {
1221 .grp = &i7core_inject_addrmatch,
1222 }, {
1223 .attr = {
1224 .name = "inject_enable",
1225 .mode = (S_IRUGO | S_IWUSR)
1227 .show = i7core_inject_enable_show,
1228 .store = i7core_inject_enable_store,
1230 { } /* End of list */
1233 static const struct mcidev_sysfs_attribute i7core_sysfs_udimm_attrs[] = {
1235 .attr = {
1236 .name = "inject_section",
1237 .mode = (S_IRUGO | S_IWUSR)
1239 .show = i7core_inject_section_show,
1240 .store = i7core_inject_section_store,
1241 }, {
1242 .attr = {
1243 .name = "inject_type",
1244 .mode = (S_IRUGO | S_IWUSR)
1246 .show = i7core_inject_type_show,
1247 .store = i7core_inject_type_store,
1248 }, {
1249 .attr = {
1250 .name = "inject_eccmask",
1251 .mode = (S_IRUGO | S_IWUSR)
1253 .show = i7core_inject_eccmask_show,
1254 .store = i7core_inject_eccmask_store,
1255 }, {
1256 .grp = &i7core_inject_addrmatch,
1257 }, {
1258 .attr = {
1259 .name = "inject_enable",
1260 .mode = (S_IRUGO | S_IWUSR)
1262 .show = i7core_inject_enable_show,
1263 .store = i7core_inject_enable_store,
1264 }, {
1265 .grp = &i7core_udimm_counters,
1267 { } /* End of list */
1270 /****************************************************************************
1271 Device initialization routines: put/get, init/exit
1272 ****************************************************************************/
1275 * i7core_put_all_devices 'put' all the devices that we have
1276 * reserved via 'get'
1278 static void i7core_put_devices(struct i7core_dev *i7core_dev)
1280 int i;
1282 debugf0(__FILE__ ": %s()\n", __func__);
1283 for (i = 0; i < i7core_dev->n_devs; i++) {
1284 struct pci_dev *pdev = i7core_dev->pdev[i];
1285 if (!pdev)
1286 continue;
1287 debugf0("Removing dev %02x:%02x.%d\n",
1288 pdev->bus->number,
1289 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1290 pci_dev_put(pdev);
1294 static void i7core_put_all_devices(void)
1296 struct i7core_dev *i7core_dev, *tmp;
1298 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
1299 i7core_put_devices(i7core_dev);
1300 free_i7core_dev(i7core_dev);
1304 static void __init i7core_xeon_pci_fixup(const struct pci_id_table *table)
1306 struct pci_dev *pdev = NULL;
1307 int i;
1310 * On Xeon 55xx, the Intel Quick Path Arch Generic Non-core pci buses
1311 * aren't announced by acpi. So, we need to use a legacy scan probing
1312 * to detect them
1314 while (table && table->descr) {
1315 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1316 if (unlikely(!pdev)) {
1317 for (i = 0; i < MAX_SOCKET_BUSES; i++)
1318 pcibios_scan_specific_bus(255-i);
1320 pci_dev_put(pdev);
1321 table++;
1325 static unsigned i7core_pci_lastbus(void)
1327 int last_bus = 0, bus;
1328 struct pci_bus *b = NULL;
1330 while ((b = pci_find_next_bus(b)) != NULL) {
1331 bus = b->number;
1332 debugf0("Found bus %d\n", bus);
1333 if (bus > last_bus)
1334 last_bus = bus;
1337 debugf0("Last bus %d\n", last_bus);
1339 return last_bus;
1343 * i7core_get_all_devices Find and perform 'get' operation on the MCH's
1344 * device/functions we want to reference for this driver
1346 * Need to 'get' device 16 func 1 and func 2
1348 static int i7core_get_onedevice(struct pci_dev **prev,
1349 const struct pci_id_table *table,
1350 const unsigned devno,
1351 const unsigned last_bus)
1353 struct i7core_dev *i7core_dev;
1354 const struct pci_id_descr *dev_descr = &table->descr[devno];
1356 struct pci_dev *pdev = NULL;
1357 u8 bus = 0;
1358 u8 socket = 0;
1360 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1361 dev_descr->dev_id, *prev);
1364 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core regs
1365 * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1366 * to probe for the alternate address in case of failure
1368 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev)
1369 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1370 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
1372 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE && !pdev)
1373 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1374 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1375 *prev);
1377 if (!pdev) {
1378 if (*prev) {
1379 *prev = pdev;
1380 return 0;
1383 if (dev_descr->optional)
1384 return 0;
1386 if (devno == 0)
1387 return -ENODEV;
1389 i7core_printk(KERN_INFO,
1390 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1391 dev_descr->dev, dev_descr->func,
1392 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1394 /* End of list, leave */
1395 return -ENODEV;
1397 bus = pdev->bus->number;
1399 socket = last_bus - bus;
1401 i7core_dev = get_i7core_dev(socket);
1402 if (!i7core_dev) {
1403 i7core_dev = alloc_i7core_dev(socket, table);
1404 if (!i7core_dev) {
1405 pci_dev_put(pdev);
1406 return -ENOMEM;
1410 if (i7core_dev->pdev[devno]) {
1411 i7core_printk(KERN_ERR,
1412 "Duplicated device for "
1413 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1414 bus, dev_descr->dev, dev_descr->func,
1415 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1416 pci_dev_put(pdev);
1417 return -ENODEV;
1420 i7core_dev->pdev[devno] = pdev;
1422 /* Sanity check */
1423 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1424 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1425 i7core_printk(KERN_ERR,
1426 "Device PCI ID %04x:%04x "
1427 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
1428 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1429 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1430 bus, dev_descr->dev, dev_descr->func);
1431 return -ENODEV;
1434 /* Be sure that the device is enabled */
1435 if (unlikely(pci_enable_device(pdev) < 0)) {
1436 i7core_printk(KERN_ERR,
1437 "Couldn't enable "
1438 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1439 bus, dev_descr->dev, dev_descr->func,
1440 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1441 return -ENODEV;
1444 debugf0("Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
1445 socket, bus, dev_descr->dev,
1446 dev_descr->func,
1447 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1450 * As stated on drivers/pci/search.c, the reference count for
1451 * @from is always decremented if it is not %NULL. So, as we need
1452 * to get all devices up to null, we need to do a get for the device
1454 pci_dev_get(pdev);
1456 *prev = pdev;
1458 return 0;
1461 static int i7core_get_all_devices(void)
1463 int i, rc, last_bus;
1464 struct pci_dev *pdev = NULL;
1465 const struct pci_id_table *table = pci_dev_table;
1467 last_bus = i7core_pci_lastbus();
1469 while (table && table->descr) {
1470 for (i = 0; i < table->n_devs; i++) {
1471 pdev = NULL;
1472 do {
1473 rc = i7core_get_onedevice(&pdev, table, i,
1474 last_bus);
1475 if (rc < 0) {
1476 if (i == 0) {
1477 i = table->n_devs;
1478 break;
1480 i7core_put_all_devices();
1481 return -ENODEV;
1483 } while (pdev);
1485 table++;
1488 return 0;
1491 static int mci_bind_devs(struct mem_ctl_info *mci,
1492 struct i7core_dev *i7core_dev)
1494 struct i7core_pvt *pvt = mci->pvt_info;
1495 struct pci_dev *pdev;
1496 int i, func, slot;
1497 char *family;
1499 pvt->is_registered = false;
1500 pvt->enable_scrub = false;
1501 for (i = 0; i < i7core_dev->n_devs; i++) {
1502 pdev = i7core_dev->pdev[i];
1503 if (!pdev)
1504 continue;
1506 func = PCI_FUNC(pdev->devfn);
1507 slot = PCI_SLOT(pdev->devfn);
1508 if (slot == 3) {
1509 if (unlikely(func > MAX_MCR_FUNC))
1510 goto error;
1511 pvt->pci_mcr[func] = pdev;
1512 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1513 if (unlikely(func > MAX_CHAN_FUNC))
1514 goto error;
1515 pvt->pci_ch[slot - 4][func] = pdev;
1516 } else if (!slot && !func) {
1517 pvt->pci_noncore = pdev;
1519 /* Detect the processor family */
1520 switch (pdev->device) {
1521 case PCI_DEVICE_ID_INTEL_I7_NONCORE:
1522 family = "Xeon 35xx/ i7core";
1523 pvt->enable_scrub = false;
1524 break;
1525 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT:
1526 family = "i7-800/i5-700";
1527 pvt->enable_scrub = false;
1528 break;
1529 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE:
1530 family = "Xeon 34xx";
1531 pvt->enable_scrub = false;
1532 break;
1533 case PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT:
1534 family = "Xeon 55xx";
1535 pvt->enable_scrub = true;
1536 break;
1537 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2:
1538 family = "Xeon 56xx / i7-900";
1539 pvt->enable_scrub = true;
1540 break;
1541 default:
1542 family = "unknown";
1543 pvt->enable_scrub = false;
1545 debugf0("Detected a processor type %s\n", family);
1546 } else
1547 goto error;
1549 debugf0("Associated fn %d.%d, dev = %p, socket %d\n",
1550 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1551 pdev, i7core_dev->socket);
1553 if (PCI_SLOT(pdev->devfn) == 3 &&
1554 PCI_FUNC(pdev->devfn) == 2)
1555 pvt->is_registered = true;
1558 return 0;
1560 error:
1561 i7core_printk(KERN_ERR, "Device %d, function %d "
1562 "is out of the expected range\n",
1563 slot, func);
1564 return -EINVAL;
1567 /****************************************************************************
1568 Error check routines
1569 ****************************************************************************/
1570 static void i7core_rdimm_update_csrow(struct mem_ctl_info *mci,
1571 const int chan,
1572 const int dimm,
1573 const int add)
1575 char *msg;
1576 struct i7core_pvt *pvt = mci->pvt_info;
1577 int row = pvt->csrow_map[chan][dimm], i;
1579 for (i = 0; i < add; i++) {
1580 msg = kasprintf(GFP_KERNEL, "Corrected error "
1581 "(Socket=%d channel=%d dimm=%d)",
1582 pvt->i7core_dev->socket, chan, dimm);
1584 edac_mc_handle_fbd_ce(mci, row, 0, msg);
1585 kfree (msg);
1589 static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
1590 const int chan,
1591 const int new0,
1592 const int new1,
1593 const int new2)
1595 struct i7core_pvt *pvt = mci->pvt_info;
1596 int add0 = 0, add1 = 0, add2 = 0;
1597 /* Updates CE counters if it is not the first time here */
1598 if (pvt->ce_count_available) {
1599 /* Updates CE counters */
1601 add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1602 add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1603 add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
1605 if (add2 < 0)
1606 add2 += 0x7fff;
1607 pvt->rdimm_ce_count[chan][2] += add2;
1609 if (add1 < 0)
1610 add1 += 0x7fff;
1611 pvt->rdimm_ce_count[chan][1] += add1;
1613 if (add0 < 0)
1614 add0 += 0x7fff;
1615 pvt->rdimm_ce_count[chan][0] += add0;
1616 } else
1617 pvt->ce_count_available = 1;
1619 /* Store the new values */
1620 pvt->rdimm_last_ce_count[chan][2] = new2;
1621 pvt->rdimm_last_ce_count[chan][1] = new1;
1622 pvt->rdimm_last_ce_count[chan][0] = new0;
1624 /*updated the edac core */
1625 if (add0 != 0)
1626 i7core_rdimm_update_csrow(mci, chan, 0, add0);
1627 if (add1 != 0)
1628 i7core_rdimm_update_csrow(mci, chan, 1, add1);
1629 if (add2 != 0)
1630 i7core_rdimm_update_csrow(mci, chan, 2, add2);
1634 static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1636 struct i7core_pvt *pvt = mci->pvt_info;
1637 u32 rcv[3][2];
1638 int i, new0, new1, new2;
1640 /*Read DEV 3: FUN 2: MC_COR_ECC_CNT regs directly*/
1641 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
1642 &rcv[0][0]);
1643 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
1644 &rcv[0][1]);
1645 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
1646 &rcv[1][0]);
1647 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
1648 &rcv[1][1]);
1649 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
1650 &rcv[2][0]);
1651 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
1652 &rcv[2][1]);
1653 for (i = 0 ; i < 3; i++) {
1654 debugf3("MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1655 (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1656 /*if the channel has 3 dimms*/
1657 if (pvt->channel[i].dimms > 2) {
1658 new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1659 new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1660 new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1661 } else {
1662 new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1663 DIMM_BOT_COR_ERR(rcv[i][0]);
1664 new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1665 DIMM_BOT_COR_ERR(rcv[i][1]);
1666 new2 = 0;
1669 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
1673 /* This function is based on the device 3 function 4 registers as described on:
1674 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1675 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1676 * also available at:
1677 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1679 static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1681 struct i7core_pvt *pvt = mci->pvt_info;
1682 u32 rcv1, rcv0;
1683 int new0, new1, new2;
1685 if (!pvt->pci_mcr[4]) {
1686 debugf0("%s MCR registers not found\n", __func__);
1687 return;
1690 /* Corrected test errors */
1691 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1692 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
1694 /* Store the new values */
1695 new2 = DIMM2_COR_ERR(rcv1);
1696 new1 = DIMM1_COR_ERR(rcv0);
1697 new0 = DIMM0_COR_ERR(rcv0);
1699 /* Updates CE counters if it is not the first time here */
1700 if (pvt->ce_count_available) {
1701 /* Updates CE counters */
1702 int add0, add1, add2;
1704 add2 = new2 - pvt->udimm_last_ce_count[2];
1705 add1 = new1 - pvt->udimm_last_ce_count[1];
1706 add0 = new0 - pvt->udimm_last_ce_count[0];
1708 if (add2 < 0)
1709 add2 += 0x7fff;
1710 pvt->udimm_ce_count[2] += add2;
1712 if (add1 < 0)
1713 add1 += 0x7fff;
1714 pvt->udimm_ce_count[1] += add1;
1716 if (add0 < 0)
1717 add0 += 0x7fff;
1718 pvt->udimm_ce_count[0] += add0;
1720 if (add0 | add1 | add2)
1721 i7core_printk(KERN_ERR, "New Corrected error(s): "
1722 "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1723 add0, add1, add2);
1724 } else
1725 pvt->ce_count_available = 1;
1727 /* Store the new values */
1728 pvt->udimm_last_ce_count[2] = new2;
1729 pvt->udimm_last_ce_count[1] = new1;
1730 pvt->udimm_last_ce_count[0] = new0;
1734 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1735 * Architectures Software Developer’s Manual Volume 3B.
1736 * Nehalem are defined as family 0x06, model 0x1a
1738 * The MCA registers used here are the following ones:
1739 * struct mce field MCA Register
1740 * m->status MSR_IA32_MC8_STATUS
1741 * m->addr MSR_IA32_MC8_ADDR
1742 * m->misc MSR_IA32_MC8_MISC
1743 * In the case of Nehalem, the error information is masked at .status and .misc
1744 * fields
1746 static void i7core_mce_output_error(struct mem_ctl_info *mci,
1747 const struct mce *m)
1749 struct i7core_pvt *pvt = mci->pvt_info;
1750 char *type, *optype, *err, *msg;
1751 unsigned long error = m->status & 0x1ff0000l;
1752 u32 optypenum = (m->status >> 4) & 0x07;
1753 u32 core_err_cnt = (m->status >> 38) & 0x7fff;
1754 u32 dimm = (m->misc >> 16) & 0x3;
1755 u32 channel = (m->misc >> 18) & 0x3;
1756 u32 syndrome = m->misc >> 32;
1757 u32 errnum = find_first_bit(&error, 32);
1758 int csrow;
1760 if (m->mcgstatus & 1)
1761 type = "FATAL";
1762 else
1763 type = "NON_FATAL";
1765 switch (optypenum) {
1766 case 0:
1767 optype = "generic undef request";
1768 break;
1769 case 1:
1770 optype = "read error";
1771 break;
1772 case 2:
1773 optype = "write error";
1774 break;
1775 case 3:
1776 optype = "addr/cmd error";
1777 break;
1778 case 4:
1779 optype = "scrubbing error";
1780 break;
1781 default:
1782 optype = "reserved";
1783 break;
1786 switch (errnum) {
1787 case 16:
1788 err = "read ECC error";
1789 break;
1790 case 17:
1791 err = "RAS ECC error";
1792 break;
1793 case 18:
1794 err = "write parity error";
1795 break;
1796 case 19:
1797 err = "redundacy loss";
1798 break;
1799 case 20:
1800 err = "reserved";
1801 break;
1802 case 21:
1803 err = "memory range error";
1804 break;
1805 case 22:
1806 err = "RTID out of range";
1807 break;
1808 case 23:
1809 err = "address parity error";
1810 break;
1811 case 24:
1812 err = "byte enable parity error";
1813 break;
1814 default:
1815 err = "unknown";
1818 /* FIXME: should convert addr into bank and rank information */
1819 msg = kasprintf(GFP_ATOMIC,
1820 "%s (addr = 0x%08llx, cpu=%d, Dimm=%d, Channel=%d, "
1821 "syndrome=0x%08x, count=%d, Err=%08llx:%08llx (%s: %s))\n",
1822 type, (long long) m->addr, m->cpu, dimm, channel,
1823 syndrome, core_err_cnt, (long long)m->status,
1824 (long long)m->misc, optype, err);
1826 debugf0("%s", msg);
1828 csrow = pvt->csrow_map[channel][dimm];
1830 /* Call the helper to output message */
1831 if (m->mcgstatus & 1)
1832 edac_mc_handle_fbd_ue(mci, csrow, 0,
1833 0 /* FIXME: should be channel here */, msg);
1834 else if (!pvt->is_registered)
1835 edac_mc_handle_fbd_ce(mci, csrow,
1836 0 /* FIXME: should be channel here */, msg);
1838 kfree(msg);
1842 * i7core_check_error Retrieve and process errors reported by the
1843 * hardware. Called by the Core module.
1845 static void i7core_check_error(struct mem_ctl_info *mci)
1847 struct i7core_pvt *pvt = mci->pvt_info;
1848 int i;
1849 unsigned count = 0;
1850 struct mce *m;
1853 * MCE first step: Copy all mce errors into a temporary buffer
1854 * We use a double buffering here, to reduce the risk of
1855 * losing an error.
1857 smp_rmb();
1858 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1859 % MCE_LOG_LEN;
1860 if (!count)
1861 goto check_ce_error;
1863 m = pvt->mce_outentry;
1864 if (pvt->mce_in + count > MCE_LOG_LEN) {
1865 unsigned l = MCE_LOG_LEN - pvt->mce_in;
1867 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1868 smp_wmb();
1869 pvt->mce_in = 0;
1870 count -= l;
1871 m += l;
1873 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1874 smp_wmb();
1875 pvt->mce_in += count;
1877 smp_rmb();
1878 if (pvt->mce_overrun) {
1879 i7core_printk(KERN_ERR, "Lost %d memory errors\n",
1880 pvt->mce_overrun);
1881 smp_wmb();
1882 pvt->mce_overrun = 0;
1886 * MCE second step: parse errors and display
1888 for (i = 0; i < count; i++)
1889 i7core_mce_output_error(mci, &pvt->mce_outentry[i]);
1892 * Now, let's increment CE error counts
1894 check_ce_error:
1895 if (!pvt->is_registered)
1896 i7core_udimm_check_mc_ecc_err(mci);
1897 else
1898 i7core_rdimm_check_mc_ecc_err(mci);
1902 * i7core_mce_check_error Replicates mcelog routine to get errors
1903 * This routine simply queues mcelog errors, and
1904 * return. The error itself should be handled later
1905 * by i7core_check_error.
1906 * WARNING: As this routine should be called at NMI time, extra care should
1907 * be taken to avoid deadlocks, and to be as fast as possible.
1909 static int i7core_mce_check_error(struct notifier_block *nb, unsigned long val,
1910 void *data)
1912 struct mce *mce = (struct mce *)data;
1913 struct i7core_dev *i7_dev;
1914 struct mem_ctl_info *mci;
1915 struct i7core_pvt *pvt;
1917 i7_dev = get_i7core_dev(mce->socketid);
1918 if (!i7_dev)
1919 return NOTIFY_BAD;
1921 mci = i7_dev->mci;
1922 pvt = mci->pvt_info;
1925 * Just let mcelog handle it if the error is
1926 * outside the memory controller
1928 if (((mce->status & 0xffff) >> 7) != 1)
1929 return NOTIFY_DONE;
1931 /* Bank 8 registers are the only ones that we know how to handle */
1932 if (mce->bank != 8)
1933 return NOTIFY_DONE;
1935 #ifdef CONFIG_SMP
1936 /* Only handle if it is the right mc controller */
1937 if (mce->socketid != pvt->i7core_dev->socket)
1938 return NOTIFY_DONE;
1939 #endif
1941 smp_rmb();
1942 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1943 smp_wmb();
1944 pvt->mce_overrun++;
1945 return NOTIFY_DONE;
1948 /* Copy memory error at the ringbuffer */
1949 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1950 smp_wmb();
1951 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1953 /* Handle fatal errors immediately */
1954 if (mce->mcgstatus & 1)
1955 i7core_check_error(mci);
1957 /* Advise mcelog that the errors were handled */
1958 return NOTIFY_STOP;
1961 static struct notifier_block i7_mce_dec = {
1962 .notifier_call = i7core_mce_check_error,
1965 struct memdev_dmi_entry {
1966 u8 type;
1967 u8 length;
1968 u16 handle;
1969 u16 phys_mem_array_handle;
1970 u16 mem_err_info_handle;
1971 u16 total_width;
1972 u16 data_width;
1973 u16 size;
1974 u8 form;
1975 u8 device_set;
1976 u8 device_locator;
1977 u8 bank_locator;
1978 u8 memory_type;
1979 u16 type_detail;
1980 u16 speed;
1981 u8 manufacturer;
1982 u8 serial_number;
1983 u8 asset_tag;
1984 u8 part_number;
1985 u8 attributes;
1986 u32 extended_size;
1987 u16 conf_mem_clk_speed;
1988 } __attribute__((__packed__));
1992 * Decode the DRAM Clock Frequency, be paranoid, make sure that all
1993 * memory devices show the same speed, and if they don't then consider
1994 * all speeds to be invalid.
1996 static void decode_dclk(const struct dmi_header *dh, void *_dclk_freq)
1998 int *dclk_freq = _dclk_freq;
1999 u16 dmi_mem_clk_speed;
2001 if (*dclk_freq == -1)
2002 return;
2004 if (dh->type == DMI_ENTRY_MEM_DEVICE) {
2005 struct memdev_dmi_entry *memdev_dmi_entry =
2006 (struct memdev_dmi_entry *)dh;
2007 unsigned long conf_mem_clk_speed_offset =
2008 (unsigned long)&memdev_dmi_entry->conf_mem_clk_speed -
2009 (unsigned long)&memdev_dmi_entry->type;
2010 unsigned long speed_offset =
2011 (unsigned long)&memdev_dmi_entry->speed -
2012 (unsigned long)&memdev_dmi_entry->type;
2014 /* Check that a DIMM is present */
2015 if (memdev_dmi_entry->size == 0)
2016 return;
2019 * Pick the configured speed if it's available, otherwise
2020 * pick the DIMM speed, or we don't have a speed.
2022 if (memdev_dmi_entry->length > conf_mem_clk_speed_offset) {
2023 dmi_mem_clk_speed =
2024 memdev_dmi_entry->conf_mem_clk_speed;
2025 } else if (memdev_dmi_entry->length > speed_offset) {
2026 dmi_mem_clk_speed = memdev_dmi_entry->speed;
2027 } else {
2028 *dclk_freq = -1;
2029 return;
2032 if (*dclk_freq == 0) {
2033 /* First pass, speed was 0 */
2034 if (dmi_mem_clk_speed > 0) {
2035 /* Set speed if a valid speed is read */
2036 *dclk_freq = dmi_mem_clk_speed;
2037 } else {
2038 /* Otherwise we don't have a valid speed */
2039 *dclk_freq = -1;
2041 } else if (*dclk_freq > 0 &&
2042 *dclk_freq != dmi_mem_clk_speed) {
2044 * If we have a speed, check that all DIMMS are the same
2045 * speed, otherwise set the speed as invalid.
2047 *dclk_freq = -1;
2053 * The default DCLK frequency is used as a fallback if we
2054 * fail to find anything reliable in the DMI. The value
2055 * is taken straight from the datasheet.
2057 #define DEFAULT_DCLK_FREQ 800
2059 static int get_dclk_freq(void)
2061 int dclk_freq = 0;
2063 dmi_walk(decode_dclk, (void *)&dclk_freq);
2065 if (dclk_freq < 1)
2066 return DEFAULT_DCLK_FREQ;
2068 return dclk_freq;
2072 * set_sdram_scrub_rate This routine sets byte/sec bandwidth scrub rate
2073 * to hardware according to SCRUBINTERVAL formula
2074 * found in datasheet.
2076 static int set_sdram_scrub_rate(struct mem_ctl_info *mci, u32 new_bw)
2078 struct i7core_pvt *pvt = mci->pvt_info;
2079 struct pci_dev *pdev;
2080 u32 dw_scrub;
2081 u32 dw_ssr;
2083 /* Get data from the MC register, function 2 */
2084 pdev = pvt->pci_mcr[2];
2085 if (!pdev)
2086 return -ENODEV;
2088 pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &dw_scrub);
2090 if (new_bw == 0) {
2091 /* Prepare to disable petrol scrub */
2092 dw_scrub &= ~STARTSCRUB;
2093 /* Stop the patrol scrub engine */
2094 write_and_test(pdev, MC_SCRUB_CONTROL,
2095 dw_scrub & ~SCRUBINTERVAL_MASK);
2097 /* Get current status of scrub rate and set bit to disable */
2098 pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr);
2099 dw_ssr &= ~SSR_MODE_MASK;
2100 dw_ssr |= SSR_MODE_DISABLE;
2101 } else {
2102 const int cache_line_size = 64;
2103 const u32 freq_dclk_mhz = pvt->dclk_freq;
2104 unsigned long long scrub_interval;
2106 * Translate the desired scrub rate to a register value and
2107 * program the corresponding register value.
2109 scrub_interval = (unsigned long long)freq_dclk_mhz *
2110 cache_line_size * 1000000;
2111 do_div(scrub_interval, new_bw);
2113 if (!scrub_interval || scrub_interval > SCRUBINTERVAL_MASK)
2114 return -EINVAL;
2116 dw_scrub = SCRUBINTERVAL_MASK & scrub_interval;
2118 /* Start the patrol scrub engine */
2119 pci_write_config_dword(pdev, MC_SCRUB_CONTROL,
2120 STARTSCRUB | dw_scrub);
2122 /* Get current status of scrub rate and set bit to enable */
2123 pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr);
2124 dw_ssr &= ~SSR_MODE_MASK;
2125 dw_ssr |= SSR_MODE_ENABLE;
2127 /* Disable or enable scrubbing */
2128 pci_write_config_dword(pdev, MC_SSRCONTROL, dw_ssr);
2130 return new_bw;
2134 * get_sdram_scrub_rate This routine convert current scrub rate value
2135 * into byte/sec bandwidth accourding to
2136 * SCRUBINTERVAL formula found in datasheet.
2138 static int get_sdram_scrub_rate(struct mem_ctl_info *mci)
2140 struct i7core_pvt *pvt = mci->pvt_info;
2141 struct pci_dev *pdev;
2142 const u32 cache_line_size = 64;
2143 const u32 freq_dclk_mhz = pvt->dclk_freq;
2144 unsigned long long scrub_rate;
2145 u32 scrubval;
2147 /* Get data from the MC register, function 2 */
2148 pdev = pvt->pci_mcr[2];
2149 if (!pdev)
2150 return -ENODEV;
2152 /* Get current scrub control data */
2153 pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &scrubval);
2155 /* Mask highest 8-bits to 0 */
2156 scrubval &= SCRUBINTERVAL_MASK;
2157 if (!scrubval)
2158 return 0;
2160 /* Calculate scrub rate value into byte/sec bandwidth */
2161 scrub_rate = (unsigned long long)freq_dclk_mhz *
2162 1000000 * cache_line_size;
2163 do_div(scrub_rate, scrubval);
2164 return (int)scrub_rate;
2167 static void enable_sdram_scrub_setting(struct mem_ctl_info *mci)
2169 struct i7core_pvt *pvt = mci->pvt_info;
2170 u32 pci_lock;
2172 /* Unlock writes to pci registers */
2173 pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock);
2174 pci_lock &= ~0x3;
2175 pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL,
2176 pci_lock | MC_CFG_UNLOCK);
2178 mci->set_sdram_scrub_rate = set_sdram_scrub_rate;
2179 mci->get_sdram_scrub_rate = get_sdram_scrub_rate;
2182 static void disable_sdram_scrub_setting(struct mem_ctl_info *mci)
2184 struct i7core_pvt *pvt = mci->pvt_info;
2185 u32 pci_lock;
2187 /* Lock writes to pci registers */
2188 pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock);
2189 pci_lock &= ~0x3;
2190 pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL,
2191 pci_lock | MC_CFG_LOCK);
2194 static void i7core_pci_ctl_create(struct i7core_pvt *pvt)
2196 pvt->i7core_pci = edac_pci_create_generic_ctl(
2197 &pvt->i7core_dev->pdev[0]->dev,
2198 EDAC_MOD_STR);
2199 if (unlikely(!pvt->i7core_pci))
2200 i7core_printk(KERN_WARNING,
2201 "Unable to setup PCI error report via EDAC\n");
2204 static void i7core_pci_ctl_release(struct i7core_pvt *pvt)
2206 if (likely(pvt->i7core_pci))
2207 edac_pci_release_generic_ctl(pvt->i7core_pci);
2208 else
2209 i7core_printk(KERN_ERR,
2210 "Couldn't find mem_ctl_info for socket %d\n",
2211 pvt->i7core_dev->socket);
2212 pvt->i7core_pci = NULL;
2215 static void i7core_unregister_mci(struct i7core_dev *i7core_dev)
2217 struct mem_ctl_info *mci = i7core_dev->mci;
2218 struct i7core_pvt *pvt;
2220 if (unlikely(!mci || !mci->pvt_info)) {
2221 debugf0("MC: " __FILE__ ": %s(): dev = %p\n",
2222 __func__, &i7core_dev->pdev[0]->dev);
2224 i7core_printk(KERN_ERR, "Couldn't find mci handler\n");
2225 return;
2228 pvt = mci->pvt_info;
2230 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
2231 __func__, mci, &i7core_dev->pdev[0]->dev);
2233 /* Disable scrubrate setting */
2234 if (pvt->enable_scrub)
2235 disable_sdram_scrub_setting(mci);
2237 mce_unregister_decode_chain(&i7_mce_dec);
2239 /* Disable EDAC polling */
2240 i7core_pci_ctl_release(pvt);
2242 /* Remove MC sysfs nodes */
2243 edac_mc_del_mc(mci->dev);
2245 debugf1("%s: free mci struct\n", mci->ctl_name);
2246 kfree(mci->ctl_name);
2247 edac_mc_free(mci);
2248 i7core_dev->mci = NULL;
2251 static int i7core_register_mci(struct i7core_dev *i7core_dev)
2253 struct mem_ctl_info *mci;
2254 struct i7core_pvt *pvt;
2255 int rc, channels, csrows;
2257 /* Check the number of active and not disabled channels */
2258 rc = i7core_get_active_channels(i7core_dev->socket, &channels, &csrows);
2259 if (unlikely(rc < 0))
2260 return rc;
2262 /* allocate a new MC control structure */
2263 mci = edac_mc_alloc(sizeof(*pvt), csrows, channels, i7core_dev->socket);
2264 if (unlikely(!mci))
2265 return -ENOMEM;
2267 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
2268 __func__, mci, &i7core_dev->pdev[0]->dev);
2270 pvt = mci->pvt_info;
2271 memset(pvt, 0, sizeof(*pvt));
2273 /* Associates i7core_dev and mci for future usage */
2274 pvt->i7core_dev = i7core_dev;
2275 i7core_dev->mci = mci;
2278 * FIXME: how to handle RDDR3 at MCI level? It is possible to have
2279 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
2280 * memory channels
2282 mci->mtype_cap = MEM_FLAG_DDR3;
2283 mci->edac_ctl_cap = EDAC_FLAG_NONE;
2284 mci->edac_cap = EDAC_FLAG_NONE;
2285 mci->mod_name = "i7core_edac.c";
2286 mci->mod_ver = I7CORE_REVISION;
2287 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
2288 i7core_dev->socket);
2289 mci->dev_name = pci_name(i7core_dev->pdev[0]);
2290 mci->ctl_page_to_phys = NULL;
2292 /* Store pci devices at mci for faster access */
2293 rc = mci_bind_devs(mci, i7core_dev);
2294 if (unlikely(rc < 0))
2295 goto fail0;
2297 if (pvt->is_registered)
2298 mci->mc_driver_sysfs_attributes = i7core_sysfs_rdimm_attrs;
2299 else
2300 mci->mc_driver_sysfs_attributes = i7core_sysfs_udimm_attrs;
2302 /* Get dimm basic config */
2303 get_dimm_config(mci);
2304 /* record ptr to the generic device */
2305 mci->dev = &i7core_dev->pdev[0]->dev;
2306 /* Set the function pointer to an actual operation function */
2307 mci->edac_check = i7core_check_error;
2309 /* Enable scrubrate setting */
2310 if (pvt->enable_scrub)
2311 enable_sdram_scrub_setting(mci);
2313 /* add this new MC control structure to EDAC's list of MCs */
2314 if (unlikely(edac_mc_add_mc(mci))) {
2315 debugf0("MC: " __FILE__
2316 ": %s(): failed edac_mc_add_mc()\n", __func__);
2317 /* FIXME: perhaps some code should go here that disables error
2318 * reporting if we just enabled it
2321 rc = -EINVAL;
2322 goto fail0;
2325 /* Default error mask is any memory */
2326 pvt->inject.channel = 0;
2327 pvt->inject.dimm = -1;
2328 pvt->inject.rank = -1;
2329 pvt->inject.bank = -1;
2330 pvt->inject.page = -1;
2331 pvt->inject.col = -1;
2333 /* allocating generic PCI control info */
2334 i7core_pci_ctl_create(pvt);
2336 /* DCLK for scrub rate setting */
2337 pvt->dclk_freq = get_dclk_freq();
2339 mce_register_decode_chain(&i7_mce_dec);
2341 return 0;
2343 fail0:
2344 kfree(mci->ctl_name);
2345 edac_mc_free(mci);
2346 i7core_dev->mci = NULL;
2347 return rc;
2351 * i7core_probe Probe for ONE instance of device to see if it is
2352 * present.
2353 * return:
2354 * 0 for FOUND a device
2355 * < 0 for error code
2358 static int __devinit i7core_probe(struct pci_dev *pdev,
2359 const struct pci_device_id *id)
2361 int rc, count = 0;
2362 struct i7core_dev *i7core_dev;
2364 /* get the pci devices we want to reserve for our use */
2365 mutex_lock(&i7core_edac_lock);
2368 * All memory controllers are allocated at the first pass.
2370 if (unlikely(probed >= 1)) {
2371 mutex_unlock(&i7core_edac_lock);
2372 return -ENODEV;
2374 probed++;
2376 rc = i7core_get_all_devices();
2377 if (unlikely(rc < 0))
2378 goto fail0;
2380 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
2381 count++;
2382 rc = i7core_register_mci(i7core_dev);
2383 if (unlikely(rc < 0))
2384 goto fail1;
2388 * Nehalem-EX uses a different memory controller. However, as the
2389 * memory controller is not visible on some Nehalem/Nehalem-EP, we
2390 * need to indirectly probe via a X58 PCI device. The same devices
2391 * are found on (some) Nehalem-EX. So, on those machines, the
2392 * probe routine needs to return -ENODEV, as the actual Memory
2393 * Controller registers won't be detected.
2395 if (!count) {
2396 rc = -ENODEV;
2397 goto fail1;
2400 i7core_printk(KERN_INFO,
2401 "Driver loaded, %d memory controller(s) found.\n",
2402 count);
2404 mutex_unlock(&i7core_edac_lock);
2405 return 0;
2407 fail1:
2408 list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2409 i7core_unregister_mci(i7core_dev);
2411 i7core_put_all_devices();
2412 fail0:
2413 mutex_unlock(&i7core_edac_lock);
2414 return rc;
2418 * i7core_remove destructor for one instance of device
2421 static void __devexit i7core_remove(struct pci_dev *pdev)
2423 struct i7core_dev *i7core_dev;
2425 debugf0(__FILE__ ": %s()\n", __func__);
2428 * we have a trouble here: pdev value for removal will be wrong, since
2429 * it will point to the X58 register used to detect that the machine
2430 * is a Nehalem or upper design. However, due to the way several PCI
2431 * devices are grouped together to provide MC functionality, we need
2432 * to use a different method for releasing the devices
2435 mutex_lock(&i7core_edac_lock);
2437 if (unlikely(!probed)) {
2438 mutex_unlock(&i7core_edac_lock);
2439 return;
2442 list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2443 i7core_unregister_mci(i7core_dev);
2445 /* Release PCI resources */
2446 i7core_put_all_devices();
2448 probed--;
2450 mutex_unlock(&i7core_edac_lock);
2453 MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2456 * i7core_driver pci_driver structure for this module
2459 static struct pci_driver i7core_driver = {
2460 .name = "i7core_edac",
2461 .probe = i7core_probe,
2462 .remove = __devexit_p(i7core_remove),
2463 .id_table = i7core_pci_tbl,
2467 * i7core_init Module entry function
2468 * Try to initialize this module for its devices
2470 static int __init i7core_init(void)
2472 int pci_rc;
2474 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2476 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2477 opstate_init();
2479 if (use_pci_fixup)
2480 i7core_xeon_pci_fixup(pci_dev_table);
2482 pci_rc = pci_register_driver(&i7core_driver);
2484 if (pci_rc >= 0)
2485 return 0;
2487 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2488 pci_rc);
2490 return pci_rc;
2494 * i7core_exit() Module exit function
2495 * Unregister the driver
2497 static void __exit i7core_exit(void)
2499 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2500 pci_unregister_driver(&i7core_driver);
2503 module_init(i7core_init);
2504 module_exit(i7core_exit);
2506 MODULE_LICENSE("GPL");
2507 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
2508 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2509 MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2510 I7CORE_REVISION);
2512 module_param(edac_op_state, int, 0444);
2513 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");