aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/edac
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2011-11-02 19:55:15 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2011-11-02 19:55:15 -0400
commit6681ba7ec480bc839584fd0817991d248b4b9e44 (patch)
tree994fb1de40d58ce8dac821cf1fec727e2f902f47 /drivers/edac
parent06ef93e1b8405acac6ec900564e3ad1a8e3a72b2 (diff)
parent4d096ca7e65584dd5845e64c6400f920e694f672 (diff)
Merge branch 'linux_next' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-edac
* 'linux_next' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-edac: (21 commits) MAINTAINERS: add an entry for Edac Sandy Bridge driver edac: tag sb_edac as EXPERIMENTAL, as it requires more testing EDAC: Fix incorrect edac mode reporting in sb_edac edac: sb_edac: Add it to the building system edac: Add an experimental new driver to support Sandy Bridge CPU's i7300_edac: Fix error cleanup logic i7core_edac: Initialize memory name with cpu, channel, bank i7core_edac: Fix compilation on 32 bits arch i7core_edac: scrubbing fixups EDAC: Correct Kconfig dependencies i7core_edac: return -ENODEV if no MC is found i7core_edac: use edac's own way to print errors MAINTAINERS: remove dropped edac_mce.* from the file i7core_edac: Drop the edac_mce facility x86, MCE: Use notifier chain only for MCE decoding EDAC i7core: Use mce socketid for better compatibility i7core_edac: Don't enable memory scrubbing for Xeon 35xx i7core_edac: Add scrubbing support edac: Move edac main structs to include/linux/edac.h i7core_edac: Fix oops when trying to inject errors ...
Diffstat (limited to 'drivers/edac')
-rw-r--r--drivers/edac/Kconfig16
-rw-r--r--drivers/edac/Makefile2
-rw-r--r--drivers/edac/edac_core.h350
-rw-r--r--drivers/edac/edac_mce.c61
-rw-r--r--drivers/edac/i7300_edac.c51
-rw-r--r--drivers/edac/i7core_edac.c415
-rw-r--r--drivers/edac/sb_edac.c1893
7 files changed, 2308 insertions, 480 deletions
diff --git a/drivers/edac/Kconfig b/drivers/edac/Kconfig
index af1a17d42bd7..203361e4ef94 100644
--- a/drivers/edac/Kconfig
+++ b/drivers/edac/Kconfig
@@ -41,7 +41,7 @@ config EDAC_DEBUG
41 41
42config EDAC_DECODE_MCE 42config EDAC_DECODE_MCE
43 tristate "Decode MCEs in human-readable form (only on AMD for now)" 43 tristate "Decode MCEs in human-readable form (only on AMD for now)"
44 depends on CPU_SUP_AMD && X86_MCE 44 depends on CPU_SUP_AMD && X86_MCE_AMD
45 default y 45 default y
46 ---help--- 46 ---help---
47 Enable this option if you want to decode Machine Check Exceptions 47 Enable this option if you want to decode Machine Check Exceptions
@@ -71,9 +71,6 @@ config EDAC_MM_EDAC
71 occurred so that a particular failing memory module can be 71 occurred so that a particular failing memory module can be
72 replaced. If unsure, select 'Y'. 72 replaced. If unsure, select 'Y'.
73 73
74config EDAC_MCE
75 bool
76
77config EDAC_AMD64 74config EDAC_AMD64
78 tristate "AMD64 (Opteron, Athlon64) K8, F10h" 75 tristate "AMD64 (Opteron, Athlon64) K8, F10h"
79 depends on EDAC_MM_EDAC && AMD_NB && X86_64 && EDAC_DECODE_MCE 76 depends on EDAC_MM_EDAC && AMD_NB && X86_64 && EDAC_DECODE_MCE
@@ -173,8 +170,7 @@ config EDAC_I5400
173 170
174config EDAC_I7CORE 171config EDAC_I7CORE
175 tristate "Intel i7 Core (Nehalem) processors" 172 tristate "Intel i7 Core (Nehalem) processors"
176 depends on EDAC_MM_EDAC && PCI && X86 173 depends on EDAC_MM_EDAC && PCI && X86 && X86_MCE_INTEL
177 select EDAC_MCE
178 help 174 help
179 Support for error detection and correction the Intel 175 Support for error detection and correction the Intel
180 i7 Core (Nehalem) Integrated Memory Controller that exists on 176 i7 Core (Nehalem) Integrated Memory Controller that exists on
@@ -216,6 +212,14 @@ config EDAC_I7300
216 Support for error detection and correction the Intel 212 Support for error detection and correction the Intel
217 Clarksboro MCH (Intel 7300 chipset). 213 Clarksboro MCH (Intel 7300 chipset).
218 214
215config EDAC_SBRIDGE
216 tristate "Intel Sandy-Bridge Integrated MC"
217 depends on EDAC_MM_EDAC && PCI && X86 && X86_MCE_INTEL
218 depends on EXPERIMENTAL
219 help
220 Support for error detection and correction the Intel
221 Sandy Bridge Integrated Memory Controller.
222
219config EDAC_MPC85XX 223config EDAC_MPC85XX
220 tristate "Freescale MPC83xx / MPC85xx" 224 tristate "Freescale MPC83xx / MPC85xx"
221 depends on EDAC_MM_EDAC && FSL_SOC && (PPC_83xx || PPC_85xx) 225 depends on EDAC_MM_EDAC && FSL_SOC && (PPC_83xx || PPC_85xx)
diff --git a/drivers/edac/Makefile b/drivers/edac/Makefile
index 3e239133e29e..196a63dd37c5 100644
--- a/drivers/edac/Makefile
+++ b/drivers/edac/Makefile
@@ -8,7 +8,6 @@
8 8
9obj-$(CONFIG_EDAC) := edac_stub.o 9obj-$(CONFIG_EDAC) := edac_stub.o
10obj-$(CONFIG_EDAC_MM_EDAC) += edac_core.o 10obj-$(CONFIG_EDAC_MM_EDAC) += edac_core.o
11obj-$(CONFIG_EDAC_MCE) += edac_mce.o
12 11
13edac_core-y := edac_mc.o edac_device.o edac_mc_sysfs.o edac_pci_sysfs.o 12edac_core-y := edac_mc.o edac_device.o edac_mc_sysfs.o edac_pci_sysfs.o
14edac_core-y += edac_module.o edac_device_sysfs.o 13edac_core-y += edac_module.o edac_device_sysfs.o
@@ -29,6 +28,7 @@ obj-$(CONFIG_EDAC_I5100) += i5100_edac.o
29obj-$(CONFIG_EDAC_I5400) += i5400_edac.o 28obj-$(CONFIG_EDAC_I5400) += i5400_edac.o
30obj-$(CONFIG_EDAC_I7300) += i7300_edac.o 29obj-$(CONFIG_EDAC_I7300) += i7300_edac.o
31obj-$(CONFIG_EDAC_I7CORE) += i7core_edac.o 30obj-$(CONFIG_EDAC_I7CORE) += i7core_edac.o
31obj-$(CONFIG_EDAC_SBRIDGE) += sb_edac.o
32obj-$(CONFIG_EDAC_E7XXX) += e7xxx_edac.o 32obj-$(CONFIG_EDAC_E7XXX) += e7xxx_edac.o
33obj-$(CONFIG_EDAC_E752X) += e752x_edac.o 33obj-$(CONFIG_EDAC_E752X) += e752x_edac.o
34obj-$(CONFIG_EDAC_I82443BXGX) += i82443bxgx_edac.o 34obj-$(CONFIG_EDAC_I82443BXGX) += i82443bxgx_edac.o
diff --git a/drivers/edac/edac_core.h b/drivers/edac/edac_core.h
index 55b8278bb172..fe90cd4a7ebc 100644
--- a/drivers/edac/edac_core.h
+++ b/drivers/edac/edac_core.h
@@ -34,11 +34,10 @@
34#include <linux/platform_device.h> 34#include <linux/platform_device.h>
35#include <linux/sysdev.h> 35#include <linux/sysdev.h>
36#include <linux/workqueue.h> 36#include <linux/workqueue.h>
37#include <linux/edac.h>
37 38
38#define EDAC_MC_LABEL_LEN 31
39#define EDAC_DEVICE_NAME_LEN 31 39#define EDAC_DEVICE_NAME_LEN 31
40#define EDAC_ATTRIB_VALUE_LEN 15 40#define EDAC_ATTRIB_VALUE_LEN 15
41#define MC_PROC_NAME_MAX_LEN 7
42 41
43#if PAGE_SHIFT < 20 42#if PAGE_SHIFT < 20
44#define PAGES_TO_MiB(pages) ((pages) >> (20 - PAGE_SHIFT)) 43#define PAGES_TO_MiB(pages) ((pages) >> (20 - PAGE_SHIFT))
@@ -101,353 +100,6 @@ extern int edac_debug_level;
101 100
102#define edac_dev_name(dev) (dev)->dev_name 101#define edac_dev_name(dev) (dev)->dev_name
103 102
104/* memory devices */
105enum dev_type {
106 DEV_UNKNOWN = 0,
107 DEV_X1,
108 DEV_X2,
109 DEV_X4,
110 DEV_X8,
111 DEV_X16,
112 DEV_X32, /* Do these parts exist? */
113 DEV_X64 /* Do these parts exist? */
114};
115
116#define DEV_FLAG_UNKNOWN BIT(DEV_UNKNOWN)
117#define DEV_FLAG_X1 BIT(DEV_X1)
118#define DEV_FLAG_X2 BIT(DEV_X2)
119#define DEV_FLAG_X4 BIT(DEV_X4)
120#define DEV_FLAG_X8 BIT(DEV_X8)
121#define DEV_FLAG_X16 BIT(DEV_X16)
122#define DEV_FLAG_X32 BIT(DEV_X32)
123#define DEV_FLAG_X64 BIT(DEV_X64)
124
125/* memory types */
126enum mem_type {
127 MEM_EMPTY = 0, /* Empty csrow */
128 MEM_RESERVED, /* Reserved csrow type */
129 MEM_UNKNOWN, /* Unknown csrow type */
130 MEM_FPM, /* Fast page mode */
131 MEM_EDO, /* Extended data out */
132 MEM_BEDO, /* Burst Extended data out */
133 MEM_SDR, /* Single data rate SDRAM */
134 MEM_RDR, /* Registered single data rate SDRAM */
135 MEM_DDR, /* Double data rate SDRAM */
136 MEM_RDDR, /* Registered Double data rate SDRAM */
137 MEM_RMBS, /* Rambus DRAM */
138 MEM_DDR2, /* DDR2 RAM */
139 MEM_FB_DDR2, /* fully buffered DDR2 */
140 MEM_RDDR2, /* Registered DDR2 RAM */
141 MEM_XDR, /* Rambus XDR */
142 MEM_DDR3, /* DDR3 RAM */
143 MEM_RDDR3, /* Registered DDR3 RAM */
144};
145
146#define MEM_FLAG_EMPTY BIT(MEM_EMPTY)
147#define MEM_FLAG_RESERVED BIT(MEM_RESERVED)
148#define MEM_FLAG_UNKNOWN BIT(MEM_UNKNOWN)
149#define MEM_FLAG_FPM BIT(MEM_FPM)
150#define MEM_FLAG_EDO BIT(MEM_EDO)
151#define MEM_FLAG_BEDO BIT(MEM_BEDO)
152#define MEM_FLAG_SDR BIT(MEM_SDR)
153#define MEM_FLAG_RDR BIT(MEM_RDR)
154#define MEM_FLAG_DDR BIT(MEM_DDR)
155#define MEM_FLAG_RDDR BIT(MEM_RDDR)
156#define MEM_FLAG_RMBS BIT(MEM_RMBS)
157#define MEM_FLAG_DDR2 BIT(MEM_DDR2)
158#define MEM_FLAG_FB_DDR2 BIT(MEM_FB_DDR2)
159#define MEM_FLAG_RDDR2 BIT(MEM_RDDR2)
160#define MEM_FLAG_XDR BIT(MEM_XDR)
161#define MEM_FLAG_DDR3 BIT(MEM_DDR3)
162#define MEM_FLAG_RDDR3 BIT(MEM_RDDR3)
163
164/* chipset Error Detection and Correction capabilities and mode */
165enum edac_type {
166 EDAC_UNKNOWN = 0, /* Unknown if ECC is available */
167 EDAC_NONE, /* Doesn't support ECC */
168 EDAC_RESERVED, /* Reserved ECC type */
169 EDAC_PARITY, /* Detects parity errors */
170 EDAC_EC, /* Error Checking - no correction */
171 EDAC_SECDED, /* Single bit error correction, Double detection */
172 EDAC_S2ECD2ED, /* Chipkill x2 devices - do these exist? */
173 EDAC_S4ECD4ED, /* Chipkill x4 devices */
174 EDAC_S8ECD8ED, /* Chipkill x8 devices */
175 EDAC_S16ECD16ED, /* Chipkill x16 devices */
176};
177
178#define EDAC_FLAG_UNKNOWN BIT(EDAC_UNKNOWN)
179#define EDAC_FLAG_NONE BIT(EDAC_NONE)
180#define EDAC_FLAG_PARITY BIT(EDAC_PARITY)
181#define EDAC_FLAG_EC BIT(EDAC_EC)
182#define EDAC_FLAG_SECDED BIT(EDAC_SECDED)
183#define EDAC_FLAG_S2ECD2ED BIT(EDAC_S2ECD2ED)
184#define EDAC_FLAG_S4ECD4ED BIT(EDAC_S4ECD4ED)
185#define EDAC_FLAG_S8ECD8ED BIT(EDAC_S8ECD8ED)
186#define EDAC_FLAG_S16ECD16ED BIT(EDAC_S16ECD16ED)
187
188/* scrubbing capabilities */
189enum scrub_type {
190 SCRUB_UNKNOWN = 0, /* Unknown if scrubber is available */
191 SCRUB_NONE, /* No scrubber */
192 SCRUB_SW_PROG, /* SW progressive (sequential) scrubbing */
193 SCRUB_SW_SRC, /* Software scrub only errors */
194 SCRUB_SW_PROG_SRC, /* Progressive software scrub from an error */
195 SCRUB_SW_TUNABLE, /* Software scrub frequency is tunable */
196 SCRUB_HW_PROG, /* HW progressive (sequential) scrubbing */
197 SCRUB_HW_SRC, /* Hardware scrub only errors */
198 SCRUB_HW_PROG_SRC, /* Progressive hardware scrub from an error */
199 SCRUB_HW_TUNABLE /* Hardware scrub frequency is tunable */
200};
201
202#define SCRUB_FLAG_SW_PROG BIT(SCRUB_SW_PROG)
203#define SCRUB_FLAG_SW_SRC BIT(SCRUB_SW_SRC)
204#define SCRUB_FLAG_SW_PROG_SRC BIT(SCRUB_SW_PROG_SRC)
205#define SCRUB_FLAG_SW_TUN BIT(SCRUB_SW_SCRUB_TUNABLE)
206#define SCRUB_FLAG_HW_PROG BIT(SCRUB_HW_PROG)
207#define SCRUB_FLAG_HW_SRC BIT(SCRUB_HW_SRC)
208#define SCRUB_FLAG_HW_PROG_SRC BIT(SCRUB_HW_PROG_SRC)
209#define SCRUB_FLAG_HW_TUN BIT(SCRUB_HW_TUNABLE)
210
211/* FIXME - should have notify capabilities: NMI, LOG, PROC, etc */
212
213/* EDAC internal operation states */
214#define OP_ALLOC 0x100
215#define OP_RUNNING_POLL 0x201
216#define OP_RUNNING_INTERRUPT 0x202
217#define OP_RUNNING_POLL_INTR 0x203
218#define OP_OFFLINE 0x300
219
220/*
221 * There are several things to be aware of that aren't at all obvious:
222 *
223 *
224 * SOCKETS, SOCKET SETS, BANKS, ROWS, CHIP-SELECT ROWS, CHANNELS, etc..
225 *
226 * These are some of the many terms that are thrown about that don't always
227 * mean what people think they mean (Inconceivable!). In the interest of
228 * creating a common ground for discussion, terms and their definitions
229 * will be established.
230 *
231 * Memory devices: The individual chip on a memory stick. These devices
232 * commonly output 4 and 8 bits each. Grouping several
233 * of these in parallel provides 64 bits which is common
234 * for a memory stick.
235 *
236 * Memory Stick: A printed circuit board that aggregates multiple
237 * memory devices in parallel. This is the atomic
238 * memory component that is purchaseable by Joe consumer
239 * and loaded into a memory socket.
240 *
241 * Socket: A physical connector on the motherboard that accepts
242 * a single memory stick.
243 *
244 * Channel: Set of memory devices on a memory stick that must be
245 * grouped in parallel with one or more additional
246 * channels from other memory sticks. This parallel
247 * grouping of the output from multiple channels are
248 * necessary for the smallest granularity of memory access.
249 * Some memory controllers are capable of single channel -
250 * which means that memory sticks can be loaded
251 * individually. Other memory controllers are only
252 * capable of dual channel - which means that memory
253 * sticks must be loaded as pairs (see "socket set").
254 *
255 * Chip-select row: All of the memory devices that are selected together.
256 * for a single, minimum grain of memory access.
257 * This selects all of the parallel memory devices across
258 * all of the parallel channels. Common chip-select rows
259 * for single channel are 64 bits, for dual channel 128
260 * bits.
261 *
262 * Single-Ranked stick: A Single-ranked stick has 1 chip-select row of memory.
263 * Motherboards commonly drive two chip-select pins to
264 * a memory stick. A single-ranked stick, will occupy
265 * only one of those rows. The other will be unused.
266 *
267 * Double-Ranked stick: A double-ranked stick has two chip-select rows which
268 * access different sets of memory devices. The two
269 * rows cannot be accessed concurrently.
270 *
271 * Double-sided stick: DEPRECATED TERM, see Double-Ranked stick.
272 * A double-sided stick has two chip-select rows which
273 * access different sets of memory devices. The two
274 * rows cannot be accessed concurrently. "Double-sided"
275 * is irrespective of the memory devices being mounted
276 * on both sides of the memory stick.
277 *
278 * Socket set: All of the memory sticks that are required for
279 * a single memory access or all of the memory sticks
280 * spanned by a chip-select row. A single socket set
281 * has two chip-select rows and if double-sided sticks
282 * are used these will occupy those chip-select rows.
283 *
284 * Bank: This term is avoided because it is unclear when
285 * needing to distinguish between chip-select rows and
286 * socket sets.
287 *
288 * Controller pages:
289 *
290 * Physical pages:
291 *
292 * Virtual pages:
293 *
294 *
295 * STRUCTURE ORGANIZATION AND CHOICES
296 *
297 *
298 *
299 * PS - I enjoyed writing all that about as much as you enjoyed reading it.
300 */
301
302struct channel_info {
303 int chan_idx; /* channel index */
304 u32 ce_count; /* Correctable Errors for this CHANNEL */
305 char label[EDAC_MC_LABEL_LEN + 1]; /* DIMM label on motherboard */
306 struct csrow_info *csrow; /* the parent */
307};
308
309struct csrow_info {
310 unsigned long first_page; /* first page number in dimm */
311 unsigned long last_page; /* last page number in dimm */
312 unsigned long page_mask; /* used for interleaving -
313 * 0UL for non intlv
314 */
315 u32 nr_pages; /* number of pages in csrow */
316 u32 grain; /* granularity of reported error in bytes */
317 int csrow_idx; /* the chip-select row */
318 enum dev_type dtype; /* memory device type */
319 u32 ue_count; /* Uncorrectable Errors for this csrow */
320 u32 ce_count; /* Correctable Errors for this csrow */
321 enum mem_type mtype; /* memory csrow type */
322 enum edac_type edac_mode; /* EDAC mode for this csrow */
323 struct mem_ctl_info *mci; /* the parent */
324
325 struct kobject kobj; /* sysfs kobject for this csrow */
326
327 /* channel information for this csrow */
328 u32 nr_channels;
329 struct channel_info *channels;
330};
331
332struct mcidev_sysfs_group {
333 const char *name; /* group name */
334 const struct mcidev_sysfs_attribute *mcidev_attr; /* group attributes */
335};
336
337struct mcidev_sysfs_group_kobj {
338 struct list_head list; /* list for all instances within a mc */
339
340 struct kobject kobj; /* kobj for the group */
341
342 const struct mcidev_sysfs_group *grp; /* group description table */
343 struct mem_ctl_info *mci; /* the parent */
344};
345
346/* mcidev_sysfs_attribute structure
347 * used for driver sysfs attributes and in mem_ctl_info
348 * sysfs top level entries
349 */
350struct mcidev_sysfs_attribute {
351 /* It should use either attr or grp */
352 struct attribute attr;
353 const struct mcidev_sysfs_group *grp; /* Points to a group of attributes */
354
355 /* Ops for show/store values at the attribute - not used on group */
356 ssize_t (*show)(struct mem_ctl_info *,char *);
357 ssize_t (*store)(struct mem_ctl_info *, const char *,size_t);
358};
359
360/* MEMORY controller information structure
361 */
362struct mem_ctl_info {
363 struct list_head link; /* for global list of mem_ctl_info structs */
364
365 struct module *owner; /* Module owner of this control struct */
366
367 unsigned long mtype_cap; /* memory types supported by mc */
368 unsigned long edac_ctl_cap; /* Mem controller EDAC capabilities */
369 unsigned long edac_cap; /* configuration capabilities - this is
370 * closely related to edac_ctl_cap. The
371 * difference is that the controller may be
372 * capable of s4ecd4ed which would be listed
373 * in edac_ctl_cap, but if channels aren't
374 * capable of s4ecd4ed then the edac_cap would
375 * not have that capability.
376 */
377 unsigned long scrub_cap; /* chipset scrub capabilities */
378 enum scrub_type scrub_mode; /* current scrub mode */
379
380 /* Translates sdram memory scrub rate given in bytes/sec to the
381 internal representation and configures whatever else needs
382 to be configured.
383 */
384 int (*set_sdram_scrub_rate) (struct mem_ctl_info * mci, u32 bw);
385
386 /* Get the current sdram memory scrub rate from the internal
387 representation and converts it to the closest matching
388 bandwidth in bytes/sec.
389 */
390 int (*get_sdram_scrub_rate) (struct mem_ctl_info * mci);
391
392
393 /* pointer to edac checking routine */
394 void (*edac_check) (struct mem_ctl_info * mci);
395
396 /*
397 * Remaps memory pages: controller pages to physical pages.
398 * For most MC's, this will be NULL.
399 */
400 /* FIXME - why not send the phys page to begin with? */
401 unsigned long (*ctl_page_to_phys) (struct mem_ctl_info * mci,
402 unsigned long page);
403 int mc_idx;
404 int nr_csrows;
405 struct csrow_info *csrows;
406 /*
407 * FIXME - what about controllers on other busses? - IDs must be
408 * unique. dev pointer should be sufficiently unique, but
409 * BUS:SLOT.FUNC numbers may not be unique.
410 */
411 struct device *dev;
412 const char *mod_name;
413 const char *mod_ver;
414 const char *ctl_name;
415 const char *dev_name;
416 char proc_name[MC_PROC_NAME_MAX_LEN + 1];
417 void *pvt_info;
418 u32 ue_noinfo_count; /* Uncorrectable Errors w/o info */
419 u32 ce_noinfo_count; /* Correctable Errors w/o info */
420 u32 ue_count; /* Total Uncorrectable Errors for this MC */
421 u32 ce_count; /* Total Correctable Errors for this MC */
422 unsigned long start_time; /* mci load start time (in jiffies) */
423
424 struct completion complete;
425
426 /* edac sysfs device control */
427 struct kobject edac_mci_kobj;
428
429 /* list for all grp instances within a mc */
430 struct list_head grp_kobj_list;
431
432 /* Additional top controller level attributes, but specified
433 * by the low level driver.
434 *
435 * Set by the low level driver to provide attributes at the
436 * controller level, same level as 'ue_count' and 'ce_count' above.
437 * An array of structures, NULL terminated
438 *
439 * If attributes are desired, then set to array of attributes
440 * If no attributes are desired, leave NULL
441 */
442 const struct mcidev_sysfs_attribute *mc_driver_sysfs_attributes;
443
444 /* work struct for this MC */
445 struct delayed_work work;
446
447 /* the internal state of this controller instance */
448 int op_state;
449};
450
451/* 103/*
452 * The following are the structures to provide for a generic 104 * The following are the structures to provide for a generic
453 * or abstract 'edac_device'. This set of structures and the 105 * or abstract 'edac_device'. This set of structures and the
diff --git a/drivers/edac/edac_mce.c b/drivers/edac/edac_mce.c
deleted file mode 100644
index 9ccdc5b140e7..000000000000
--- a/drivers/edac/edac_mce.c
+++ /dev/null
@@ -1,61 +0,0 @@
1/* Provides edac interface to mcelog events
2 *
3 * This file may be distributed under the terms of the
4 * GNU General Public License version 2.
5 *
6 * Copyright (c) 2009 by:
7 * Mauro Carvalho Chehab <mchehab@redhat.com>
8 *
9 * Red Hat Inc. http://www.redhat.com
10 */
11
12#include <linux/module.h>
13#include <linux/edac_mce.h>
14#include <asm/mce.h>
15
16int edac_mce_enabled;
17EXPORT_SYMBOL_GPL(edac_mce_enabled);
18
19
20/*
21 * Extension interface
22 */
23
24static LIST_HEAD(edac_mce_list);
25static DEFINE_MUTEX(edac_mce_lock);
26
27int edac_mce_register(struct edac_mce *edac_mce)
28{
29 mutex_lock(&edac_mce_lock);
30 list_add_tail(&edac_mce->list, &edac_mce_list);
31 mutex_unlock(&edac_mce_lock);
32 return 0;
33}
34EXPORT_SYMBOL(edac_mce_register);
35
36void edac_mce_unregister(struct edac_mce *edac_mce)
37{
38 mutex_lock(&edac_mce_lock);
39 list_del(&edac_mce->list);
40 mutex_unlock(&edac_mce_lock);
41}
42EXPORT_SYMBOL(edac_mce_unregister);
43
44int edac_mce_parse(struct mce *mce)
45{
46 struct edac_mce *edac_mce;
47
48 list_for_each_entry(edac_mce, &edac_mce_list, list) {
49 if (edac_mce->check_error(edac_mce->priv, mce))
50 return 1;
51 }
52
53 /* Nobody queued the error */
54 return 0;
55}
56EXPORT_SYMBOL_GPL(edac_mce_parse);
57
58MODULE_LICENSE("GPL");
59MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
60MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
61MODULE_DESCRIPTION("EDAC Driver for mcelog captured errors");
diff --git a/drivers/edac/i7300_edac.c b/drivers/edac/i7300_edac.c
index a76fe8366b68..6104dba380b6 100644
--- a/drivers/edac/i7300_edac.c
+++ b/drivers/edac/i7300_edac.c
@@ -372,7 +372,7 @@ static const char *get_err_from_table(const char *table[], int size, int pos)
372static void i7300_process_error_global(struct mem_ctl_info *mci) 372static void i7300_process_error_global(struct mem_ctl_info *mci)
373{ 373{
374 struct i7300_pvt *pvt; 374 struct i7300_pvt *pvt;
375 u32 errnum, value; 375 u32 errnum, error_reg;
376 unsigned long errors; 376 unsigned long errors;
377 const char *specific; 377 const char *specific;
378 bool is_fatal; 378 bool is_fatal;
@@ -381,9 +381,9 @@ static void i7300_process_error_global(struct mem_ctl_info *mci)
381 381
382 /* read in the 1st FATAL error register */ 382 /* read in the 1st FATAL error register */
383 pci_read_config_dword(pvt->pci_dev_16_2_fsb_err_regs, 383 pci_read_config_dword(pvt->pci_dev_16_2_fsb_err_regs,
384 FERR_GLOBAL_HI, &value); 384 FERR_GLOBAL_HI, &error_reg);
385 if (unlikely(value)) { 385 if (unlikely(error_reg)) {
386 errors = value; 386 errors = error_reg;
387 errnum = find_first_bit(&errors, 387 errnum = find_first_bit(&errors,
388 ARRAY_SIZE(ferr_global_hi_name)); 388 ARRAY_SIZE(ferr_global_hi_name));
389 specific = GET_ERR_FROM_TABLE(ferr_global_hi_name, errnum); 389 specific = GET_ERR_FROM_TABLE(ferr_global_hi_name, errnum);
@@ -391,15 +391,15 @@ static void i7300_process_error_global(struct mem_ctl_info *mci)
391 391
392 /* Clear the error bit */ 392 /* Clear the error bit */
393 pci_write_config_dword(pvt->pci_dev_16_2_fsb_err_regs, 393 pci_write_config_dword(pvt->pci_dev_16_2_fsb_err_regs,
394 FERR_GLOBAL_HI, value); 394 FERR_GLOBAL_HI, error_reg);
395 395
396 goto error_global; 396 goto error_global;
397 } 397 }
398 398
399 pci_read_config_dword(pvt->pci_dev_16_2_fsb_err_regs, 399 pci_read_config_dword(pvt->pci_dev_16_2_fsb_err_regs,
400 FERR_GLOBAL_LO, &value); 400 FERR_GLOBAL_LO, &error_reg);
401 if (unlikely(value)) { 401 if (unlikely(error_reg)) {
402 errors = value; 402 errors = error_reg;
403 errnum = find_first_bit(&errors, 403 errnum = find_first_bit(&errors,
404 ARRAY_SIZE(ferr_global_lo_name)); 404 ARRAY_SIZE(ferr_global_lo_name));
405 specific = GET_ERR_FROM_TABLE(ferr_global_lo_name, errnum); 405 specific = GET_ERR_FROM_TABLE(ferr_global_lo_name, errnum);
@@ -407,7 +407,7 @@ static void i7300_process_error_global(struct mem_ctl_info *mci)
407 407
408 /* Clear the error bit */ 408 /* Clear the error bit */
409 pci_write_config_dword(pvt->pci_dev_16_2_fsb_err_regs, 409 pci_write_config_dword(pvt->pci_dev_16_2_fsb_err_regs,
410 FERR_GLOBAL_LO, value); 410 FERR_GLOBAL_LO, error_reg);
411 411
412 goto error_global; 412 goto error_global;
413 } 413 }
@@ -427,7 +427,7 @@ error_global:
427static void i7300_process_fbd_error(struct mem_ctl_info *mci) 427static void i7300_process_fbd_error(struct mem_ctl_info *mci)
428{ 428{
429 struct i7300_pvt *pvt; 429 struct i7300_pvt *pvt;
430 u32 errnum, value; 430 u32 errnum, value, error_reg;
431 u16 val16; 431 u16 val16;
432 unsigned branch, channel, bank, rank, cas, ras; 432 unsigned branch, channel, bank, rank, cas, ras;
433 u32 syndrome; 433 u32 syndrome;
@@ -440,14 +440,14 @@ static void i7300_process_fbd_error(struct mem_ctl_info *mci)
440 440
441 /* read in the 1st FATAL error register */ 441 /* read in the 1st FATAL error register */
442 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, 442 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map,
443 FERR_FAT_FBD, &value); 443 FERR_FAT_FBD, &error_reg);
444 if (unlikely(value & FERR_FAT_FBD_ERR_MASK)) { 444 if (unlikely(error_reg & FERR_FAT_FBD_ERR_MASK)) {
445 errors = value & FERR_FAT_FBD_ERR_MASK ; 445 errors = error_reg & FERR_FAT_FBD_ERR_MASK ;
446 errnum = find_first_bit(&errors, 446 errnum = find_first_bit(&errors,
447 ARRAY_SIZE(ferr_fat_fbd_name)); 447 ARRAY_SIZE(ferr_fat_fbd_name));
448 specific = GET_ERR_FROM_TABLE(ferr_fat_fbd_name, errnum); 448 specific = GET_ERR_FROM_TABLE(ferr_fat_fbd_name, errnum);
449 branch = (GET_FBD_FAT_IDX(error_reg) == 2) ? 1 : 0;
449 450
450 branch = (GET_FBD_FAT_IDX(value) == 2) ? 1 : 0;
451 pci_read_config_word(pvt->pci_dev_16_1_fsb_addr_map, 451 pci_read_config_word(pvt->pci_dev_16_1_fsb_addr_map,
452 NRECMEMA, &val16); 452 NRECMEMA, &val16);
453 bank = NRECMEMA_BANK(val16); 453 bank = NRECMEMA_BANK(val16);
@@ -455,11 +455,14 @@ static void i7300_process_fbd_error(struct mem_ctl_info *mci)
455 455
456 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, 456 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map,
457 NRECMEMB, &value); 457 NRECMEMB, &value);
458
459 is_wr = NRECMEMB_IS_WR(value); 458 is_wr = NRECMEMB_IS_WR(value);
460 cas = NRECMEMB_CAS(value); 459 cas = NRECMEMB_CAS(value);
461 ras = NRECMEMB_RAS(value); 460 ras = NRECMEMB_RAS(value);
462 461
462 /* Clean the error register */
463 pci_write_config_dword(pvt->pci_dev_16_1_fsb_addr_map,
464 FERR_FAT_FBD, error_reg);
465
463 snprintf(pvt->tmp_prt_buffer, PAGE_SIZE, 466 snprintf(pvt->tmp_prt_buffer, PAGE_SIZE,
464 "FATAL (Branch=%d DRAM-Bank=%d %s " 467 "FATAL (Branch=%d DRAM-Bank=%d %s "
465 "RAS=%d CAS=%d Err=0x%lx (%s))", 468 "RAS=%d CAS=%d Err=0x%lx (%s))",
@@ -476,21 +479,17 @@ static void i7300_process_fbd_error(struct mem_ctl_info *mci)
476 479
477 /* read in the 1st NON-FATAL error register */ 480 /* read in the 1st NON-FATAL error register */
478 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, 481 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map,
479 FERR_NF_FBD, &value); 482 FERR_NF_FBD, &error_reg);
480 if (unlikely(value & FERR_NF_FBD_ERR_MASK)) { 483 if (unlikely(error_reg & FERR_NF_FBD_ERR_MASK)) {
481 errors = value & FERR_NF_FBD_ERR_MASK; 484 errors = error_reg & FERR_NF_FBD_ERR_MASK;
482 errnum = find_first_bit(&errors, 485 errnum = find_first_bit(&errors,
483 ARRAY_SIZE(ferr_nf_fbd_name)); 486 ARRAY_SIZE(ferr_nf_fbd_name));
484 specific = GET_ERR_FROM_TABLE(ferr_nf_fbd_name, errnum); 487 specific = GET_ERR_FROM_TABLE(ferr_nf_fbd_name, errnum);
485 488 branch = (GET_FBD_FAT_IDX(error_reg) == 2) ? 1 : 0;
486 /* Clear the error bit */
487 pci_write_config_dword(pvt->pci_dev_16_2_fsb_err_regs,
488 FERR_GLOBAL_LO, value);
489 489
490 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, 490 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map,
491 REDMEMA, &syndrome); 491 REDMEMA, &syndrome);
492 492
493 branch = (GET_FBD_FAT_IDX(value) == 2) ? 1 : 0;
494 pci_read_config_word(pvt->pci_dev_16_1_fsb_addr_map, 493 pci_read_config_word(pvt->pci_dev_16_1_fsb_addr_map,
495 RECMEMA, &val16); 494 RECMEMA, &val16);
496 bank = RECMEMA_BANK(val16); 495 bank = RECMEMA_BANK(val16);
@@ -498,18 +497,20 @@ static void i7300_process_fbd_error(struct mem_ctl_info *mci)
498 497
499 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, 498 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map,
500 RECMEMB, &value); 499 RECMEMB, &value);
501
502 is_wr = RECMEMB_IS_WR(value); 500 is_wr = RECMEMB_IS_WR(value);
503 cas = RECMEMB_CAS(value); 501 cas = RECMEMB_CAS(value);
504 ras = RECMEMB_RAS(value); 502 ras = RECMEMB_RAS(value);
505 503
506 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, 504 pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map,
507 REDMEMB, &value); 505 REDMEMB, &value);
508
509 channel = (branch << 1); 506 channel = (branch << 1);
510 if (IS_SECOND_CH(value)) 507 if (IS_SECOND_CH(value))
511 channel++; 508 channel++;
512 509
510 /* Clear the error bit */
511 pci_write_config_dword(pvt->pci_dev_16_1_fsb_addr_map,
512 FERR_NF_FBD, error_reg);
513
513 /* Form out message */ 514 /* Form out message */
514 snprintf(pvt->tmp_prt_buffer, PAGE_SIZE, 515 snprintf(pvt->tmp_prt_buffer, PAGE_SIZE,
515 "Corrected error (Branch=%d, Channel %d), " 516 "Corrected error (Branch=%d, Channel %d), "
diff --git a/drivers/edac/i7core_edac.c b/drivers/edac/i7core_edac.c
index f6cf448d69b4..70ad8923f1d7 100644
--- a/drivers/edac/i7core_edac.c
+++ b/drivers/edac/i7core_edac.c
@@ -31,11 +31,13 @@
31#include <linux/pci_ids.h> 31#include <linux/pci_ids.h>
32#include <linux/slab.h> 32#include <linux/slab.h>
33#include <linux/delay.h> 33#include <linux/delay.h>
34#include <linux/dmi.h>
34#include <linux/edac.h> 35#include <linux/edac.h>
35#include <linux/mmzone.h> 36#include <linux/mmzone.h>
36#include <linux/edac_mce.h>
37#include <linux/smp.h> 37#include <linux/smp.h>
38#include <asm/mce.h>
38#include <asm/processor.h> 39#include <asm/processor.h>
40#include <asm/div64.h>
39 41
40#include "edac_core.h" 42#include "edac_core.h"
41 43
@@ -78,6 +80,8 @@ MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices");
78 /* OFFSETS for Device 0 Function 0 */ 80 /* OFFSETS for Device 0 Function 0 */
79 81
80#define MC_CFG_CONTROL 0x90 82#define MC_CFG_CONTROL 0x90
83 #define MC_CFG_UNLOCK 0x02
84 #define MC_CFG_LOCK 0x00
81 85
82 /* OFFSETS for Device 3 Function 0 */ 86 /* OFFSETS for Device 3 Function 0 */
83 87
@@ -98,6 +102,15 @@ MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices");
98 #define DIMM0_COR_ERR(r) ((r) & 0x7fff) 102 #define DIMM0_COR_ERR(r) ((r) & 0x7fff)
99 103
100/* OFFSETS for Device 3 Function 2, as inicated on Xeon 5500 datasheet */ 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
109
110#define MC_SCRUB_CONTROL 0x4c
111 #define STARTSCRUB (1 << 24)
112 #define SCRUBINTERVAL_MASK 0xffffff
113
101#define MC_COR_ECC_CNT_0 0x80 114#define MC_COR_ECC_CNT_0 0x80
102#define MC_COR_ECC_CNT_1 0x84 115#define MC_COR_ECC_CNT_1 0x84
103#define MC_COR_ECC_CNT_2 0x88 116#define MC_COR_ECC_CNT_2 0x88
@@ -253,10 +266,7 @@ struct i7core_pvt {
253 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS]; 266 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
254 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS]; 267 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
255 268
256 unsigned int is_registered; 269 bool is_registered, enable_scrub;
257
258 /* mcelog glue */
259 struct edac_mce edac_mce;
260 270
261 /* Fifo double buffers */ 271 /* Fifo double buffers */
262 struct mce mce_entry[MCE_LOG_LEN]; 272 struct mce mce_entry[MCE_LOG_LEN];
@@ -268,6 +278,9 @@ struct i7core_pvt {
268 /* Count indicator to show errors not got */ 278 /* Count indicator to show errors not got */
269 unsigned mce_overrun; 279 unsigned mce_overrun;
270 280
281 /* DCLK Frequency used for computing scrub rate */
282 int dclk_freq;
283
271 /* Struct to control EDAC polling */ 284 /* Struct to control EDAC polling */
272 struct edac_pci_ctl_info *i7core_pci; 285 struct edac_pci_ctl_info *i7core_pci;
273}; 286};
@@ -281,8 +294,7 @@ static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
281 /* Memory controller */ 294 /* Memory controller */
282 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) }, 295 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
283 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) }, 296 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
284 297 /* Exists only for RDIMM */
285 /* Exists only for RDIMM */
286 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 }, 298 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 },
287 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) }, 299 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
288 300
@@ -303,6 +315,16 @@ static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
303 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) }, 315 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
304 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) }, 316 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
305 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) }, 317 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
318
319 /* Generic Non-core registers */
320 /*
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
325 */
326 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) },
327
306}; 328};
307 329
308static const struct pci_id_descr pci_dev_descr_lynnfield[] = { 330static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
@@ -319,6 +341,12 @@ static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
319 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) }, 341 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
320 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) }, 342 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
321 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) }, 343 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) },
344
345 /*
346 * This is the PCI device has an alternate address on some
347 * processors like Core i7 860
348 */
349 { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) },
322}; 350};
323 351
324static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = { 352static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
@@ -346,6 +374,10 @@ static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
346 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) }, 374 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
347 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) }, 375 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
348 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) }, 376 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) },
377
378 /* Generic Non-core registers */
379 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) },
380
349}; 381};
350 382
351#define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) } 383#define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
@@ -714,6 +746,10 @@ static int get_dimm_config(const struct mem_ctl_info *mci)
714 746
715 csr->edac_mode = mode; 747 csr->edac_mode = mode;
716 csr->mtype = mtype; 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);
717 753
718 csrow++; 754 csrow++;
719 } 755 }
@@ -731,7 +767,7 @@ static int get_dimm_config(const struct mem_ctl_info *mci)
731 debugf1("\t\t%#x\t%#x\t%#x\n", 767 debugf1("\t\t%#x\t%#x\t%#x\n",
732 (value[j] >> 27) & 0x1, 768 (value[j] >> 27) & 0x1,
733 (value[j] >> 24) & 0x7, 769 (value[j] >> 24) & 0x7,
734 (value[j] && ((1 << 24) - 1))); 770 (value[j] & ((1 << 24) - 1)));
735 } 771 }
736 772
737 return 0; 773 return 0;
@@ -1324,6 +1360,20 @@ static int i7core_get_onedevice(struct pci_dev **prev,
1324 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 1360 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1325 dev_descr->dev_id, *prev); 1361 dev_descr->dev_id, *prev);
1326 1362
1363 /*
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
1367 */
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);
1371
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);
1376
1327 if (!pdev) { 1377 if (!pdev) {
1328 if (*prev) { 1378 if (*prev) {
1329 *prev = pdev; 1379 *prev = pdev;
@@ -1444,8 +1494,10 @@ static int mci_bind_devs(struct mem_ctl_info *mci,
1444 struct i7core_pvt *pvt = mci->pvt_info; 1494 struct i7core_pvt *pvt = mci->pvt_info;
1445 struct pci_dev *pdev; 1495 struct pci_dev *pdev;
1446 int i, func, slot; 1496 int i, func, slot;
1497 char *family;
1447 1498
1448 pvt->is_registered = 0; 1499 pvt->is_registered = false;
1500 pvt->enable_scrub = false;
1449 for (i = 0; i < i7core_dev->n_devs; i++) { 1501 for (i = 0; i < i7core_dev->n_devs; i++) {
1450 pdev = i7core_dev->pdev[i]; 1502 pdev = i7core_dev->pdev[i];
1451 if (!pdev) 1503 if (!pdev)
@@ -1461,9 +1513,37 @@ static int mci_bind_devs(struct mem_ctl_info *mci,
1461 if (unlikely(func > MAX_CHAN_FUNC)) 1513 if (unlikely(func > MAX_CHAN_FUNC))
1462 goto error; 1514 goto error;
1463 pvt->pci_ch[slot - 4][func] = pdev; 1515 pvt->pci_ch[slot - 4][func] = pdev;
1464 } else if (!slot && !func) 1516 } else if (!slot && !func) {
1465 pvt->pci_noncore = pdev; 1517 pvt->pci_noncore = pdev;
1466 else 1518
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;
1544 }
1545 debugf0("Detected a processor type %s\n", family);
1546 } else
1467 goto error; 1547 goto error;
1468 1548
1469 debugf0("Associated fn %d.%d, dev = %p, socket %d\n", 1549 debugf0("Associated fn %d.%d, dev = %p, socket %d\n",
@@ -1472,7 +1552,7 @@ static int mci_bind_devs(struct mem_ctl_info *mci,
1472 1552
1473 if (PCI_SLOT(pdev->devfn) == 3 && 1553 if (PCI_SLOT(pdev->devfn) == 3 &&
1474 PCI_FUNC(pdev->devfn) == 2) 1554 PCI_FUNC(pdev->devfn) == 2)
1475 pvt->is_registered = 1; 1555 pvt->is_registered = true;
1476 } 1556 }
1477 1557
1478 return 0; 1558 return 0;
@@ -1826,33 +1906,43 @@ check_ce_error:
1826 * WARNING: As this routine should be called at NMI time, extra care should 1906 * WARNING: As this routine should be called at NMI time, extra care should
1827 * be taken to avoid deadlocks, and to be as fast as possible. 1907 * be taken to avoid deadlocks, and to be as fast as possible.
1828 */ 1908 */
1829static int i7core_mce_check_error(void *priv, struct mce *mce) 1909static int i7core_mce_check_error(struct notifier_block *nb, unsigned long val,
1910 void *data)
1830{ 1911{
1831 struct mem_ctl_info *mci = priv; 1912 struct mce *mce = (struct mce *)data;
1832 struct i7core_pvt *pvt = mci->pvt_info; 1913 struct i7core_dev *i7_dev;
1914 struct mem_ctl_info *mci;
1915 struct i7core_pvt *pvt;
1916
1917 i7_dev = get_i7core_dev(mce->socketid);
1918 if (!i7_dev)
1919 return NOTIFY_BAD;
1920
1921 mci = i7_dev->mci;
1922 pvt = mci->pvt_info;
1833 1923
1834 /* 1924 /*
1835 * Just let mcelog handle it if the error is 1925 * Just let mcelog handle it if the error is
1836 * outside the memory controller 1926 * outside the memory controller
1837 */ 1927 */
1838 if (((mce->status & 0xffff) >> 7) != 1) 1928 if (((mce->status & 0xffff) >> 7) != 1)
1839 return 0; 1929 return NOTIFY_DONE;
1840 1930
1841 /* Bank 8 registers are the only ones that we know how to handle */ 1931 /* Bank 8 registers are the only ones that we know how to handle */
1842 if (mce->bank != 8) 1932 if (mce->bank != 8)
1843 return 0; 1933 return NOTIFY_DONE;
1844 1934
1845#ifdef CONFIG_SMP 1935#ifdef CONFIG_SMP
1846 /* Only handle if it is the right mc controller */ 1936 /* Only handle if it is the right mc controller */
1847 if (cpu_data(mce->cpu).phys_proc_id != pvt->i7core_dev->socket) 1937 if (mce->socketid != pvt->i7core_dev->socket)
1848 return 0; 1938 return NOTIFY_DONE;
1849#endif 1939#endif
1850 1940
1851 smp_rmb(); 1941 smp_rmb();
1852 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) { 1942 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1853 smp_wmb(); 1943 smp_wmb();
1854 pvt->mce_overrun++; 1944 pvt->mce_overrun++;
1855 return 0; 1945 return NOTIFY_DONE;
1856 } 1946 }
1857 1947
1858 /* Copy memory error at the ringbuffer */ 1948 /* Copy memory error at the ringbuffer */
@@ -1865,7 +1955,240 @@ static int i7core_mce_check_error(void *priv, struct mce *mce)
1865 i7core_check_error(mci); 1955 i7core_check_error(mci);
1866 1956
1867 /* Advise mcelog that the errors were handled */ 1957 /* Advise mcelog that the errors were handled */
1868 return 1; 1958 return NOTIFY_STOP;
1959}
1960
1961static struct notifier_block i7_mce_dec = {
1962 .notifier_call = i7core_mce_check_error,
1963};
1964
1965struct 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__));
1989
1990
1991/*
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.
1995 */
1996static void decode_dclk(const struct dmi_header *dh, void *_dclk_freq)
1997{
1998 int *dclk_freq = _dclk_freq;
1999 u16 dmi_mem_clk_speed;
2000
2001 if (*dclk_freq == -1)
2002 return;
2003
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;
2013
2014 /* Check that a DIMM is present */
2015 if (memdev_dmi_entry->size == 0)
2016 return;
2017
2018 /*
2019 * Pick the configured speed if it's available, otherwise
2020 * pick the DIMM speed, or we don't have a speed.
2021 */
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;
2030 }
2031
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;
2040 }
2041 } else if (*dclk_freq > 0 &&
2042 *dclk_freq != dmi_mem_clk_speed) {
2043 /*
2044 * If we have a speed, check that all DIMMS are the same
2045 * speed, otherwise set the speed as invalid.
2046 */
2047 *dclk_freq = -1;
2048 }
2049 }
2050}
2051
2052/*
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.
2056 */
2057#define DEFAULT_DCLK_FREQ 800
2058
2059static int get_dclk_freq(void)
2060{
2061 int dclk_freq = 0;
2062
2063 dmi_walk(decode_dclk, (void *)&dclk_freq);
2064
2065 if (dclk_freq < 1)
2066 return DEFAULT_DCLK_FREQ;
2067
2068 return dclk_freq;
2069}
2070
2071/*
2072 * set_sdram_scrub_rate This routine sets byte/sec bandwidth scrub rate
2073 * to hardware according to SCRUBINTERVAL formula
2074 * found in datasheet.
2075 */
2076static int set_sdram_scrub_rate(struct mem_ctl_info *mci, u32 new_bw)
2077{
2078 struct i7core_pvt *pvt = mci->pvt_info;
2079 struct pci_dev *pdev;
2080 u32 dw_scrub;
2081 u32 dw_ssr;
2082
2083 /* Get data from the MC register, function 2 */
2084 pdev = pvt->pci_mcr[2];
2085 if (!pdev)
2086 return -ENODEV;
2087
2088 pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &dw_scrub);
2089
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);
2096
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;
2105 /*
2106 * Translate the desired scrub rate to a register value and
2107 * program the corresponding register value.
2108 */
2109 scrub_interval = (unsigned long long)freq_dclk_mhz *
2110 cache_line_size * 1000000;
2111 do_div(scrub_interval, new_bw);
2112
2113 if (!scrub_interval || scrub_interval > SCRUBINTERVAL_MASK)
2114 return -EINVAL;
2115
2116 dw_scrub = SCRUBINTERVAL_MASK & scrub_interval;
2117
2118 /* Start the patrol scrub engine */
2119 pci_write_config_dword(pdev, MC_SCRUB_CONTROL,
2120 STARTSCRUB | dw_scrub);
2121
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;
2126 }
2127 /* Disable or enable scrubbing */
2128 pci_write_config_dword(pdev, MC_SSRCONTROL, dw_ssr);
2129
2130 return new_bw;
2131}
2132
2133/*
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.
2137 */
2138static int get_sdram_scrub_rate(struct mem_ctl_info *mci)
2139{
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;
2146
2147 /* Get data from the MC register, function 2 */
2148 pdev = pvt->pci_mcr[2];
2149 if (!pdev)
2150 return -ENODEV;
2151
2152 /* Get current scrub control data */
2153 pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &scrubval);
2154
2155 /* Mask highest 8-bits to 0 */
2156 scrubval &= SCRUBINTERVAL_MASK;
2157 if (!scrubval)
2158 return 0;
2159
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;
2165}
2166
2167static void enable_sdram_scrub_setting(struct mem_ctl_info *mci)
2168{
2169 struct i7core_pvt *pvt = mci->pvt_info;
2170 u32 pci_lock;
2171
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);
2177
2178 mci->set_sdram_scrub_rate = set_sdram_scrub_rate;
2179 mci->get_sdram_scrub_rate = get_sdram_scrub_rate;
2180}
2181
2182static void disable_sdram_scrub_setting(struct mem_ctl_info *mci)
2183{
2184 struct i7core_pvt *pvt = mci->pvt_info;
2185 u32 pci_lock;
2186
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);
1869} 2192}
1870 2193
1871static void i7core_pci_ctl_create(struct i7core_pvt *pvt) 2194static void i7core_pci_ctl_create(struct i7core_pvt *pvt)
@@ -1874,7 +2197,8 @@ static void i7core_pci_ctl_create(struct i7core_pvt *pvt)
1874 &pvt->i7core_dev->pdev[0]->dev, 2197 &pvt->i7core_dev->pdev[0]->dev,
1875 EDAC_MOD_STR); 2198 EDAC_MOD_STR);
1876 if (unlikely(!pvt->i7core_pci)) 2199 if (unlikely(!pvt->i7core_pci))
1877 pr_warn("Unable to setup PCI error report via EDAC\n"); 2200 i7core_printk(KERN_WARNING,
2201 "Unable to setup PCI error report via EDAC\n");
1878} 2202}
1879 2203
1880static void i7core_pci_ctl_release(struct i7core_pvt *pvt) 2204static void i7core_pci_ctl_release(struct i7core_pvt *pvt)
@@ -1906,8 +2230,11 @@ static void i7core_unregister_mci(struct i7core_dev *i7core_dev)
1906 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n", 2230 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1907 __func__, mci, &i7core_dev->pdev[0]->dev); 2231 __func__, mci, &i7core_dev->pdev[0]->dev);
1908 2232
1909 /* Disable MCE NMI handler */ 2233 /* Disable scrubrate setting */
1910 edac_mce_unregister(&pvt->edac_mce); 2234 if (pvt->enable_scrub)
2235 disable_sdram_scrub_setting(mci);
2236
2237 atomic_notifier_chain_unregister(&x86_mce_decoder_chain, &i7_mce_dec);
1911 2238
1912 /* Disable EDAC polling */ 2239 /* Disable EDAC polling */
1913 i7core_pci_ctl_release(pvt); 2240 i7core_pci_ctl_release(pvt);
@@ -1979,6 +2306,10 @@ static int i7core_register_mci(struct i7core_dev *i7core_dev)
1979 /* Set the function pointer to an actual operation function */ 2306 /* Set the function pointer to an actual operation function */
1980 mci->edac_check = i7core_check_error; 2307 mci->edac_check = i7core_check_error;
1981 2308
2309 /* Enable scrubrate setting */
2310 if (pvt->enable_scrub)
2311 enable_sdram_scrub_setting(mci);
2312
1982 /* add this new MC control structure to EDAC's list of MCs */ 2313 /* add this new MC control structure to EDAC's list of MCs */
1983 if (unlikely(edac_mc_add_mc(mci))) { 2314 if (unlikely(edac_mc_add_mc(mci))) {
1984 debugf0("MC: " __FILE__ 2315 debugf0("MC: " __FILE__
@@ -2002,21 +2333,13 @@ static int i7core_register_mci(struct i7core_dev *i7core_dev)
2002 /* allocating generic PCI control info */ 2333 /* allocating generic PCI control info */
2003 i7core_pci_ctl_create(pvt); 2334 i7core_pci_ctl_create(pvt);
2004 2335
2005 /* Registers on edac_mce in order to receive memory errors */ 2336 /* DCLK for scrub rate setting */
2006 pvt->edac_mce.priv = mci; 2337 pvt->dclk_freq = get_dclk_freq();
2007 pvt->edac_mce.check_error = i7core_mce_check_error; 2338
2008 rc = edac_mce_register(&pvt->edac_mce); 2339 atomic_notifier_chain_register(&x86_mce_decoder_chain, &i7_mce_dec);
2009 if (unlikely(rc < 0)) {
2010 debugf0("MC: " __FILE__
2011 ": %s(): failed edac_mce_register()\n", __func__);
2012 goto fail1;
2013 }
2014 2340
2015 return 0; 2341 return 0;
2016 2342
2017fail1:
2018 i7core_pci_ctl_release(pvt);
2019 edac_mc_del_mc(mci->dev);
2020fail0: 2343fail0:
2021 kfree(mci->ctl_name); 2344 kfree(mci->ctl_name);
2022 edac_mc_free(mci); 2345 edac_mc_free(mci);
@@ -2035,7 +2358,7 @@ fail0:
2035static int __devinit i7core_probe(struct pci_dev *pdev, 2358static int __devinit i7core_probe(struct pci_dev *pdev,
2036 const struct pci_device_id *id) 2359 const struct pci_device_id *id)
2037{ 2360{
2038 int rc; 2361 int rc, count = 0;
2039 struct i7core_dev *i7core_dev; 2362 struct i7core_dev *i7core_dev;
2040 2363
2041 /* get the pci devices we want to reserve for our use */ 2364 /* get the pci devices we want to reserve for our use */
@@ -2055,12 +2378,28 @@ static int __devinit i7core_probe(struct pci_dev *pdev,
2055 goto fail0; 2378 goto fail0;
2056 2379
2057 list_for_each_entry(i7core_dev, &i7core_edac_list, list) { 2380 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
2381 count++;
2058 rc = i7core_register_mci(i7core_dev); 2382 rc = i7core_register_mci(i7core_dev);
2059 if (unlikely(rc < 0)) 2383 if (unlikely(rc < 0))
2060 goto fail1; 2384 goto fail1;
2061 } 2385 }
2062 2386
2063 i7core_printk(KERN_INFO, "Driver loaded.\n"); 2387 /*
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.
2394 */
2395 if (!count) {
2396 rc = -ENODEV;
2397 goto fail1;
2398 }
2399
2400 i7core_printk(KERN_INFO,
2401 "Driver loaded, %d memory controller(s) found.\n",
2402 count);
2064 2403
2065 mutex_unlock(&i7core_edac_lock); 2404 mutex_unlock(&i7core_edac_lock);
2066 return 0; 2405 return 0;
diff --git a/drivers/edac/sb_edac.c b/drivers/edac/sb_edac.c
new file mode 100644
index 000000000000..7a402bfbee7d
--- /dev/null
+++ b/drivers/edac/sb_edac.c
@@ -0,0 +1,1893 @@
1/* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module
2 *
3 * This driver supports the memory controllers found on the Intel
4 * processor family Sandy Bridge.
5 *
6 * This file may be distributed under the terms of the
7 * GNU General Public License version 2 only.
8 *
9 * Copyright (c) 2011 by:
10 * Mauro Carvalho Chehab <mchehab@redhat.com>
11 */
12
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/pci.h>
16#include <linux/pci_ids.h>
17#include <linux/slab.h>
18#include <linux/delay.h>
19#include <linux/edac.h>
20#include <linux/mmzone.h>
21#include <linux/smp.h>
22#include <linux/bitmap.h>
23#include <asm/processor.h>
24#include <asm/mce.h>
25
26#include "edac_core.h"
27
28/* Static vars */
29static LIST_HEAD(sbridge_edac_list);
30static DEFINE_MUTEX(sbridge_edac_lock);
31static int probed;
32
33/*
34 * Alter this version for the module when modifications are made
35 */
36#define SBRIDGE_REVISION " Ver: 1.0.0 "
37#define EDAC_MOD_STR "sbridge_edac"
38
39/*
40 * Debug macros
41 */
42#define sbridge_printk(level, fmt, arg...) \
43 edac_printk(level, "sbridge", fmt, ##arg)
44
45#define sbridge_mc_printk(mci, level, fmt, arg...) \
46 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
47
48/*
49 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
50 */
51#define GET_BITFIELD(v, lo, hi) \
52 (((v) & ((1ULL << ((hi) - (lo) + 1)) - 1) << (lo)) >> (lo))
53
54/*
55 * sbridge Memory Controller Registers
56 */
57
58/*
59 * FIXME: For now, let's order by device function, as it makes
60 * easier for driver's development proccess. This table should be
61 * moved to pci_id.h when submitted upstream
62 */
63#define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0 0x3cf4 /* 12.6 */
64#define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1 0x3cf6 /* 12.7 */
65#define PCI_DEVICE_ID_INTEL_SBRIDGE_BR 0x3cf5 /* 13.6 */
66#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0 0x3ca0 /* 14.0 */
67#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA 0x3ca8 /* 15.0 */
68#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS 0x3c71 /* 15.1 */
69#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0 0x3caa /* 15.2 */
70#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1 0x3cab /* 15.3 */
71#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2 0x3cac /* 15.4 */
72#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3 0x3cad /* 15.5 */
73#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO 0x3cb8 /* 17.0 */
74
75 /*
76 * Currently, unused, but will be needed in the future
77 * implementations, as they hold the error counters
78 */
79#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR0 0x3c72 /* 16.2 */
80#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR1 0x3c73 /* 16.3 */
81#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR2 0x3c76 /* 16.6 */
82#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR3 0x3c77 /* 16.7 */
83
84/* Devices 12 Function 6, Offsets 0x80 to 0xcc */
85static const u32 dram_rule[] = {
86 0x80, 0x88, 0x90, 0x98, 0xa0,
87 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
88};
89#define MAX_SAD ARRAY_SIZE(dram_rule)
90
91#define SAD_LIMIT(reg) ((GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff)
92#define DRAM_ATTR(reg) GET_BITFIELD(reg, 2, 3)
93#define INTERLEAVE_MODE(reg) GET_BITFIELD(reg, 1, 1)
94#define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
95
96static char *get_dram_attr(u32 reg)
97{
98 switch(DRAM_ATTR(reg)) {
99 case 0:
100 return "DRAM";
101 case 1:
102 return "MMCFG";
103 case 2:
104 return "NXM";
105 default:
106 return "unknown";
107 }
108}
109
110static const u32 interleave_list[] = {
111 0x84, 0x8c, 0x94, 0x9c, 0xa4,
112 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
113};
114#define MAX_INTERLEAVE ARRAY_SIZE(interleave_list)
115
116#define SAD_PKG0(reg) GET_BITFIELD(reg, 0, 2)
117#define SAD_PKG1(reg) GET_BITFIELD(reg, 3, 5)
118#define SAD_PKG2(reg) GET_BITFIELD(reg, 8, 10)
119#define SAD_PKG3(reg) GET_BITFIELD(reg, 11, 13)
120#define SAD_PKG4(reg) GET_BITFIELD(reg, 16, 18)
121#define SAD_PKG5(reg) GET_BITFIELD(reg, 19, 21)
122#define SAD_PKG6(reg) GET_BITFIELD(reg, 24, 26)
123#define SAD_PKG7(reg) GET_BITFIELD(reg, 27, 29)
124
125static inline int sad_pkg(u32 reg, int interleave)
126{
127 switch (interleave) {
128 case 0:
129 return SAD_PKG0(reg);
130 case 1:
131 return SAD_PKG1(reg);
132 case 2:
133 return SAD_PKG2(reg);
134 case 3:
135 return SAD_PKG3(reg);
136 case 4:
137 return SAD_PKG4(reg);
138 case 5:
139 return SAD_PKG5(reg);
140 case 6:
141 return SAD_PKG6(reg);
142 case 7:
143 return SAD_PKG7(reg);
144 default:
145 return -EINVAL;
146 }
147}
148
149/* Devices 12 Function 7 */
150
151#define TOLM 0x80
152#define TOHM 0x84
153
154#define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
155#define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
156
157/* Device 13 Function 6 */
158
159#define SAD_TARGET 0xf0
160
161#define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
162
163#define SAD_CONTROL 0xf4
164
165#define NODE_ID(reg) GET_BITFIELD(reg, 0, 2)
166
167/* Device 14 function 0 */
168
169static const u32 tad_dram_rule[] = {
170 0x40, 0x44, 0x48, 0x4c,
171 0x50, 0x54, 0x58, 0x5c,
172 0x60, 0x64, 0x68, 0x6c,
173};
174#define MAX_TAD ARRAY_SIZE(tad_dram_rule)
175
176#define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
177#define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
178#define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
179#define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
180#define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
181#define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
182#define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
183
184/* Device 15, function 0 */
185
186#define MCMTR 0x7c
187
188#define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
189#define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
190#define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
191
192/* Device 15, function 1 */
193
194#define RASENABLES 0xac
195#define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
196
197/* Device 15, functions 2-5 */
198
199static const int mtr_regs[] = {
200 0x80, 0x84, 0x88,
201};
202
203#define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
204#define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
205#define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
206#define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
207#define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
208
209static const u32 tad_ch_nilv_offset[] = {
210 0x90, 0x94, 0x98, 0x9c,
211 0xa0, 0xa4, 0xa8, 0xac,
212 0xb0, 0xb4, 0xb8, 0xbc,
213};
214#define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
215#define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
216
217static const u32 rir_way_limit[] = {
218 0x108, 0x10c, 0x110, 0x114, 0x118,
219};
220#define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
221
222#define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
223#define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
224#define RIR_LIMIT(reg) ((GET_BITFIELD(reg, 1, 10) << 29)| 0x1fffffff)
225
226#define MAX_RIR_WAY 8
227
228static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
229 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
230 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
231 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
232 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
233 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
234};
235
236#define RIR_RNK_TGT(reg) GET_BITFIELD(reg, 16, 19)
237#define RIR_OFFSET(reg) GET_BITFIELD(reg, 2, 14)
238
239/* Device 16, functions 2-7 */
240
241/*
242 * FIXME: Implement the error count reads directly
243 */
244
245static const u32 correrrcnt[] = {
246 0x104, 0x108, 0x10c, 0x110,
247};
248
249#define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
250#define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
251#define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
252#define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
253
254static const u32 correrrthrsld[] = {
255 0x11c, 0x120, 0x124, 0x128,
256};
257
258#define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
259#define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
260
261
262/* Device 17, function 0 */
263
264#define RANK_CFG_A 0x0328
265
266#define IS_RDIMM_ENABLED(reg) GET_BITFIELD(reg, 11, 11)
267
268/*
269 * sbridge structs
270 */
271
272#define NUM_CHANNELS 4
273#define MAX_DIMMS 3 /* Max DIMMS per channel */
274
275struct sbridge_info {
276 u32 mcmtr;
277};
278
279struct sbridge_channel {
280 u32 ranks;
281 u32 dimms;
282};
283
284struct pci_id_descr {
285 int dev;
286 int func;
287 int dev_id;
288 int optional;
289};
290
291struct pci_id_table {
292 const struct pci_id_descr *descr;
293 int n_devs;
294};
295
296struct sbridge_dev {
297 struct list_head list;
298 u8 bus, mc;
299 u8 node_id, source_id;
300 struct pci_dev **pdev;
301 int n_devs;
302 struct mem_ctl_info *mci;
303};
304
305struct sbridge_pvt {
306 struct pci_dev *pci_ta, *pci_ddrio, *pci_ras;
307 struct pci_dev *pci_sad0, *pci_sad1, *pci_ha0;
308 struct pci_dev *pci_br;
309 struct pci_dev *pci_tad[NUM_CHANNELS];
310
311 struct sbridge_dev *sbridge_dev;
312
313 struct sbridge_info info;
314 struct sbridge_channel channel[NUM_CHANNELS];
315
316 int csrow_map[NUM_CHANNELS][MAX_DIMMS];
317
318 /* Memory type detection */
319 bool is_mirrored, is_lockstep, is_close_pg;
320
321 /* Fifo double buffers */
322 struct mce mce_entry[MCE_LOG_LEN];
323 struct mce mce_outentry[MCE_LOG_LEN];
324
325 /* Fifo in/out counters */
326 unsigned mce_in, mce_out;
327
328 /* Count indicator to show errors not got */
329 unsigned mce_overrun;
330
331 /* Memory description */
332 u64 tolm, tohm;
333};
334
335#define PCI_DESCR(device, function, device_id) \
336 .dev = (device), \
337 .func = (function), \
338 .dev_id = (device_id)
339
340static const struct pci_id_descr pci_dev_descr_sbridge[] = {
341 /* Processor Home Agent */
342 { PCI_DESCR(14, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0) },
343
344 /* Memory controller */
345 { PCI_DESCR(15, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA) },
346 { PCI_DESCR(15, 1, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS) },
347 { PCI_DESCR(15, 2, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0) },
348 { PCI_DESCR(15, 3, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1) },
349 { PCI_DESCR(15, 4, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2) },
350 { PCI_DESCR(15, 5, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3) },
351 { PCI_DESCR(17, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO) },
352
353 /* System Address Decoder */
354 { PCI_DESCR(12, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0) },
355 { PCI_DESCR(12, 7, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1) },
356
357 /* Broadcast Registers */
358 { PCI_DESCR(13, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_BR) },
359};
360
361#define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
362static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
363 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge),
364 {0,} /* 0 terminated list. */
365};
366
367/*
368 * pci_device_id table for which devices we are looking for
369 */
370static const struct pci_device_id sbridge_pci_tbl[] __devinitdata = {
371 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA)},
372 {0,} /* 0 terminated list. */
373};
374
375
376/****************************************************************************
377 Anciliary status routines
378 ****************************************************************************/
379
380static inline int numrank(u32 mtr)
381{
382 int ranks = (1 << RANK_CNT_BITS(mtr));
383
384 if (ranks > 4) {
385 debugf0("Invalid number of ranks: %d (max = 4) raw value = %x (%04x)",
386 ranks, (unsigned int)RANK_CNT_BITS(mtr), mtr);
387 return -EINVAL;
388 }
389
390 return ranks;
391}
392
393static inline int numrow(u32 mtr)
394{
395 int rows = (RANK_WIDTH_BITS(mtr) + 12);
396
397 if (rows < 13 || rows > 18) {
398 debugf0("Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)",
399 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
400 return -EINVAL;
401 }
402
403 return 1 << rows;
404}
405
406static inline int numcol(u32 mtr)
407{
408 int cols = (COL_WIDTH_BITS(mtr) + 10);
409
410 if (cols > 12) {
411 debugf0("Invalid number of cols: %d (max = 4) raw value = %x (%04x)",
412 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
413 return -EINVAL;
414 }
415
416 return 1 << cols;
417}
418
419static struct sbridge_dev *get_sbridge_dev(u8 bus)
420{
421 struct sbridge_dev *sbridge_dev;
422
423 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
424 if (sbridge_dev->bus == bus)
425 return sbridge_dev;
426 }
427
428 return NULL;
429}
430
431static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
432 const struct pci_id_table *table)
433{
434 struct sbridge_dev *sbridge_dev;
435
436 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
437 if (!sbridge_dev)
438 return NULL;
439
440 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
441 GFP_KERNEL);
442 if (!sbridge_dev->pdev) {
443 kfree(sbridge_dev);
444 return NULL;
445 }
446
447 sbridge_dev->bus = bus;
448 sbridge_dev->n_devs = table->n_devs;
449 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
450
451 return sbridge_dev;
452}
453
454static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
455{
456 list_del(&sbridge_dev->list);
457 kfree(sbridge_dev->pdev);
458 kfree(sbridge_dev);
459}
460
461/****************************************************************************
462 Memory check routines
463 ****************************************************************************/
464static struct pci_dev *get_pdev_slot_func(u8 bus, unsigned slot,
465 unsigned func)
466{
467 struct sbridge_dev *sbridge_dev = get_sbridge_dev(bus);
468 int i;
469
470 if (!sbridge_dev)
471 return NULL;
472
473 for (i = 0; i < sbridge_dev->n_devs; i++) {
474 if (!sbridge_dev->pdev[i])
475 continue;
476
477 if (PCI_SLOT(sbridge_dev->pdev[i]->devfn) == slot &&
478 PCI_FUNC(sbridge_dev->pdev[i]->devfn) == func) {
479 debugf1("Associated %02x.%02x.%d with %p\n",
480 bus, slot, func, sbridge_dev->pdev[i]);
481 return sbridge_dev->pdev[i];
482 }
483 }
484
485 return NULL;
486}
487
488/**
489 * sbridge_get_active_channels() - gets the number of channels and csrows
490 * bus: Device bus
491 * @channels: Number of channels that will be returned
492 * @csrows: Number of csrows found
493 *
494 * Since EDAC core needs to know in advance the number of available channels
495 * and csrows, in order to allocate memory for csrows/channels, it is needed
496 * to run two similar steps. At the first step, implemented on this function,
497 * it checks the number of csrows/channels present at one socket, identified
498 * by the associated PCI bus.
499 * this is used in order to properly allocate the size of mci components.
500 * Note: one csrow is one dimm.
501 */
502static int sbridge_get_active_channels(const u8 bus, unsigned *channels,
503 unsigned *csrows)
504{
505 struct pci_dev *pdev = NULL;
506 int i, j;
507 u32 mcmtr;
508
509 *channels = 0;
510 *csrows = 0;
511
512 pdev = get_pdev_slot_func(bus, 15, 0);
513 if (!pdev) {
514 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
515 "%2x.%02d.%d!!!\n",
516 bus, 15, 0);
517 return -ENODEV;
518 }
519
520 pci_read_config_dword(pdev, MCMTR, &mcmtr);
521 if (!IS_ECC_ENABLED(mcmtr)) {
522 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
523 return -ENODEV;
524 }
525
526 for (i = 0; i < NUM_CHANNELS; i++) {
527 u32 mtr;
528
529 /* Device 15 functions 2 - 5 */
530 pdev = get_pdev_slot_func(bus, 15, 2 + i);
531 if (!pdev) {
532 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
533 "%2x.%02d.%d!!!\n",
534 bus, 15, 2 + i);
535 return -ENODEV;
536 }
537 (*channels)++;
538
539 for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
540 pci_read_config_dword(pdev, mtr_regs[j], &mtr);
541 debugf1("Bus#%02x channel #%d MTR%d = %x\n", bus, i, j, mtr);
542 if (IS_DIMM_PRESENT(mtr))
543 (*csrows)++;
544 }
545 }
546
547 debugf0("Number of active channels: %d, number of active dimms: %d\n",
548 *channels, *csrows);
549
550 return 0;
551}
552
553static int get_dimm_config(const struct mem_ctl_info *mci)
554{
555 struct sbridge_pvt *pvt = mci->pvt_info;
556 struct csrow_info *csr;
557 int i, j, banks, ranks, rows, cols, size, npages;
558 int csrow = 0;
559 unsigned long last_page = 0;
560 u32 reg;
561 enum edac_type mode;
562 enum mem_type mtype;
563
564 pci_read_config_dword(pvt->pci_br, SAD_TARGET, &reg);
565 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
566
567 pci_read_config_dword(pvt->pci_br, SAD_CONTROL, &reg);
568 pvt->sbridge_dev->node_id = NODE_ID(reg);
569 debugf0("mc#%d: Node ID: %d, source ID: %d\n",
570 pvt->sbridge_dev->mc,
571 pvt->sbridge_dev->node_id,
572 pvt->sbridge_dev->source_id);
573
574 pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg);
575 if (IS_MIRROR_ENABLED(reg)) {
576 debugf0("Memory mirror is enabled\n");
577 pvt->is_mirrored = true;
578 } else {
579 debugf0("Memory mirror is disabled\n");
580 pvt->is_mirrored = false;
581 }
582
583 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
584 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
585 debugf0("Lockstep is enabled\n");
586 mode = EDAC_S8ECD8ED;
587 pvt->is_lockstep = true;
588 } else {
589 debugf0("Lockstep is disabled\n");
590 mode = EDAC_S4ECD4ED;
591 pvt->is_lockstep = false;
592 }
593 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
594 debugf0("address map is on closed page mode\n");
595 pvt->is_close_pg = true;
596 } else {
597 debugf0("address map is on open page mode\n");
598 pvt->is_close_pg = false;
599 }
600
601 pci_read_config_dword(pvt->pci_ta, RANK_CFG_A, &reg);
602 if (IS_RDIMM_ENABLED(reg)) {
603 /* FIXME: Can also be LRDIMM */
604 debugf0("Memory is registered\n");
605 mtype = MEM_RDDR3;
606 } else {
607 debugf0("Memory is unregistered\n");
608 mtype = MEM_DDR3;
609 }
610
611 /* On all supported DDR3 DIMM types, there are 8 banks available */
612 banks = 8;
613
614 for (i = 0; i < NUM_CHANNELS; i++) {
615 u32 mtr;
616
617 for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
618 pci_read_config_dword(pvt->pci_tad[i],
619 mtr_regs[j], &mtr);
620 debugf4("Channel #%d MTR%d = %x\n", i, j, mtr);
621 if (IS_DIMM_PRESENT(mtr)) {
622 pvt->channel[i].dimms++;
623
624 ranks = numrank(mtr);
625 rows = numrow(mtr);
626 cols = numcol(mtr);
627
628 /* DDR3 has 8 I/O banks */
629 size = (rows * cols * banks * ranks) >> (20 - 3);
630 npages = MiB_TO_PAGES(size);
631
632 debugf0("mc#%d: channel %d, dimm %d, %d Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
633 pvt->sbridge_dev->mc, i, j,
634 size, npages,
635 banks, ranks, rows, cols);
636 csr = &mci->csrows[csrow];
637
638 csr->first_page = last_page;
639 csr->last_page = last_page + npages - 1;
640 csr->page_mask = 0UL; /* Unused */
641 csr->nr_pages = npages;
642 csr->grain = 32;
643 csr->csrow_idx = csrow;
644 csr->dtype = (banks == 8) ? DEV_X8 : DEV_X4;
645 csr->ce_count = 0;
646 csr->ue_count = 0;
647 csr->mtype = mtype;
648 csr->edac_mode = mode;
649 csr->nr_channels = 1;
650 csr->channels[0].chan_idx = i;
651 csr->channels[0].ce_count = 0;
652 pvt->csrow_map[i][j] = csrow;
653 snprintf(csr->channels[0].label,
654 sizeof(csr->channels[0].label),
655 "CPU_SrcID#%u_Channel#%u_DIMM#%u",
656 pvt->sbridge_dev->source_id, i, j);
657 last_page += npages;
658 csrow++;
659 }
660 }
661 }
662
663 return 0;
664}
665
666static void get_memory_layout(const struct mem_ctl_info *mci)
667{
668 struct sbridge_pvt *pvt = mci->pvt_info;
669 int i, j, k, n_sads, n_tads, sad_interl;
670 u32 reg;
671 u64 limit, prv = 0;
672 u64 tmp_mb;
673 u32 rir_way;
674
675 /*
676 * Step 1) Get TOLM/TOHM ranges
677 */
678
679 /* Address range is 32:28 */
680 pci_read_config_dword(pvt->pci_sad1, TOLM,
681 &reg);
682 pvt->tolm = GET_TOLM(reg);
683 tmp_mb = (1 + pvt->tolm) >> 20;
684
685 debugf0("TOLM: %Lu.%03Lu GB (0x%016Lx)\n",
686 tmp_mb / 1000, tmp_mb % 1000, (u64)pvt->tolm);
687
688 /* Address range is already 45:25 */
689 pci_read_config_dword(pvt->pci_sad1, TOHM,
690 &reg);
691 pvt->tohm = GET_TOHM(reg);
692 tmp_mb = (1 + pvt->tohm) >> 20;
693
694 debugf0("TOHM: %Lu.%03Lu GB (0x%016Lx)",
695 tmp_mb / 1000, tmp_mb % 1000, (u64)pvt->tohm);
696
697 /*
698 * Step 2) Get SAD range and SAD Interleave list
699 * TAD registers contain the interleave wayness. However, it
700 * seems simpler to just discover it indirectly, with the
701 * algorithm bellow.
702 */
703 prv = 0;
704 for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
705 /* SAD_LIMIT Address range is 45:26 */
706 pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
707 &reg);
708 limit = SAD_LIMIT(reg);
709
710 if (!DRAM_RULE_ENABLE(reg))
711 continue;
712
713 if (limit <= prv)
714 break;
715
716 tmp_mb = (limit + 1) >> 20;
717 debugf0("SAD#%d %s up to %Lu.%03Lu GB (0x%016Lx) %s reg=0x%08x\n",
718 n_sads,
719 get_dram_attr(reg),
720 tmp_mb / 1000, tmp_mb % 1000,
721 ((u64)tmp_mb) << 20L,
722 INTERLEAVE_MODE(reg) ? "Interleave: 8:6" : "Interleave: [8:6]XOR[18:16]",
723 reg);
724 prv = limit;
725
726 pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
727 &reg);
728 sad_interl = sad_pkg(reg, 0);
729 for (j = 0; j < 8; j++) {
730 if (j > 0 && sad_interl == sad_pkg(reg, j))
731 break;
732
733 debugf0("SAD#%d, interleave #%d: %d\n",
734 n_sads, j, sad_pkg(reg, j));
735 }
736 }
737
738 /*
739 * Step 3) Get TAD range
740 */
741 prv = 0;
742 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
743 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
744 &reg);
745 limit = TAD_LIMIT(reg);
746 if (limit <= prv)
747 break;
748 tmp_mb = (limit + 1) >> 20;
749
750 debugf0("TAD#%d: up to %Lu.%03Lu GB (0x%016Lx), socket interleave %d, memory interleave %d, TGT: %d, %d, %d, %d, reg=0x%08x\n",
751 n_tads, tmp_mb / 1000, tmp_mb % 1000,
752 ((u64)tmp_mb) << 20L,
753 (u32)TAD_SOCK(reg),
754 (u32)TAD_CH(reg),
755 (u32)TAD_TGT0(reg),
756 (u32)TAD_TGT1(reg),
757 (u32)TAD_TGT2(reg),
758 (u32)TAD_TGT3(reg),
759 reg);
760 prv = tmp_mb;
761 }
762
763 /*
764 * Step 4) Get TAD offsets, per each channel
765 */
766 for (i = 0; i < NUM_CHANNELS; i++) {
767 if (!pvt->channel[i].dimms)
768 continue;
769 for (j = 0; j < n_tads; j++) {
770 pci_read_config_dword(pvt->pci_tad[i],
771 tad_ch_nilv_offset[j],
772 &reg);
773 tmp_mb = TAD_OFFSET(reg) >> 20;
774 debugf0("TAD CH#%d, offset #%d: %Lu.%03Lu GB (0x%016Lx), reg=0x%08x\n",
775 i, j,
776 tmp_mb / 1000, tmp_mb % 1000,
777 ((u64)tmp_mb) << 20L,
778 reg);
779 }
780 }
781
782 /*
783 * Step 6) Get RIR Wayness/Limit, per each channel
784 */
785 for (i = 0; i < NUM_CHANNELS; i++) {
786 if (!pvt->channel[i].dimms)
787 continue;
788 for (j = 0; j < MAX_RIR_RANGES; j++) {
789 pci_read_config_dword(pvt->pci_tad[i],
790 rir_way_limit[j],
791 &reg);
792
793 if (!IS_RIR_VALID(reg))
794 continue;
795
796 tmp_mb = RIR_LIMIT(reg) >> 20;
797 rir_way = 1 << RIR_WAY(reg);
798 debugf0("CH#%d RIR#%d, limit: %Lu.%03Lu GB (0x%016Lx), way: %d, reg=0x%08x\n",
799 i, j,
800 tmp_mb / 1000, tmp_mb % 1000,
801 ((u64)tmp_mb) << 20L,
802 rir_way,
803 reg);
804
805 for (k = 0; k < rir_way; k++) {
806 pci_read_config_dword(pvt->pci_tad[i],
807 rir_offset[j][k],
808 &reg);
809 tmp_mb = RIR_OFFSET(reg) << 6;
810
811 debugf0("CH#%d RIR#%d INTL#%d, offset %Lu.%03Lu GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
812 i, j, k,
813 tmp_mb / 1000, tmp_mb % 1000,
814 ((u64)tmp_mb) << 20L,
815 (u32)RIR_RNK_TGT(reg),
816 reg);
817 }
818 }
819 }
820}
821
822struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
823{
824 struct sbridge_dev *sbridge_dev;
825
826 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
827 if (sbridge_dev->node_id == node_id)
828 return sbridge_dev->mci;
829 }
830 return NULL;
831}
832
833static int get_memory_error_data(struct mem_ctl_info *mci,
834 u64 addr,
835 u8 *socket,
836 long *channel_mask,
837 u8 *rank,
838 char *area_type)
839{
840 struct mem_ctl_info *new_mci;
841 struct sbridge_pvt *pvt = mci->pvt_info;
842 char msg[256];
843 int n_rir, n_sads, n_tads, sad_way, sck_xch;
844 int sad_interl, idx, base_ch;
845 int interleave_mode;
846 unsigned sad_interleave[MAX_INTERLEAVE];
847 u32 reg;
848 u8 ch_way,sck_way;
849 u32 tad_offset;
850 u32 rir_way;
851 u64 ch_addr, offset, limit, prv = 0;
852
853
854 /*
855 * Step 0) Check if the address is at special memory ranges
856 * The check bellow is probably enough to fill all cases where
857 * the error is not inside a memory, except for the legacy
858 * range (e. g. VGA addresses). It is unlikely, however, that the
859 * memory controller would generate an error on that range.
860 */
861 if ((addr > (u64) pvt->tolm) && (addr < (1L << 32))) {
862 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
863 edac_mc_handle_ce_no_info(mci, msg);
864 return -EINVAL;
865 }
866 if (addr >= (u64)pvt->tohm) {
867 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
868 edac_mc_handle_ce_no_info(mci, msg);
869 return -EINVAL;
870 }
871
872 /*
873 * Step 1) Get socket
874 */
875 for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
876 pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
877 &reg);
878
879 if (!DRAM_RULE_ENABLE(reg))
880 continue;
881
882 limit = SAD_LIMIT(reg);
883 if (limit <= prv) {
884 sprintf(msg, "Can't discover the memory socket");
885 edac_mc_handle_ce_no_info(mci, msg);
886 return -EINVAL;
887 }
888 if (addr <= limit)
889 break;
890 prv = limit;
891 }
892 if (n_sads == MAX_SAD) {
893 sprintf(msg, "Can't discover the memory socket");
894 edac_mc_handle_ce_no_info(mci, msg);
895 return -EINVAL;
896 }
897 area_type = get_dram_attr(reg);
898 interleave_mode = INTERLEAVE_MODE(reg);
899
900 pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
901 &reg);
902 sad_interl = sad_pkg(reg, 0);
903 for (sad_way = 0; sad_way < 8; sad_way++) {
904 if (sad_way > 0 && sad_interl == sad_pkg(reg, sad_way))
905 break;
906 sad_interleave[sad_way] = sad_pkg(reg, sad_way);
907 debugf0("SAD interleave #%d: %d\n",
908 sad_way, sad_interleave[sad_way]);
909 }
910 debugf0("mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
911 pvt->sbridge_dev->mc,
912 n_sads,
913 addr,
914 limit,
915 sad_way + 7,
916 INTERLEAVE_MODE(reg) ? "" : "XOR[18:16]");
917 if (interleave_mode)
918 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
919 else
920 idx = (addr >> 6) & 7;
921 switch (sad_way) {
922 case 1:
923 idx = 0;
924 break;
925 case 2:
926 idx = idx & 1;
927 break;
928 case 4:
929 idx = idx & 3;
930 break;
931 case 8:
932 break;
933 default:
934 sprintf(msg, "Can't discover socket interleave");
935 edac_mc_handle_ce_no_info(mci, msg);
936 return -EINVAL;
937 }
938 *socket = sad_interleave[idx];
939 debugf0("SAD interleave index: %d (wayness %d) = CPU socket %d\n",
940 idx, sad_way, *socket);
941
942 /*
943 * Move to the proper node structure, in order to access the
944 * right PCI registers
945 */
946 new_mci = get_mci_for_node_id(*socket);
947 if (!new_mci) {
948 sprintf(msg, "Struct for socket #%u wasn't initialized",
949 *socket);
950 edac_mc_handle_ce_no_info(mci, msg);
951 return -EINVAL;
952 }
953 mci = new_mci;
954 pvt = mci->pvt_info;
955
956 /*
957 * Step 2) Get memory channel
958 */
959 prv = 0;
960 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
961 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
962 &reg);
963 limit = TAD_LIMIT(reg);
964 if (limit <= prv) {
965 sprintf(msg, "Can't discover the memory channel");
966 edac_mc_handle_ce_no_info(mci, msg);
967 return -EINVAL;
968 }
969 if (addr <= limit)
970 break;
971 prv = limit;
972 }
973 ch_way = TAD_CH(reg) + 1;
974 sck_way = TAD_SOCK(reg) + 1;
975 /*
976 * FIXME: Is it right to always use channel 0 for offsets?
977 */
978 pci_read_config_dword(pvt->pci_tad[0],
979 tad_ch_nilv_offset[n_tads],
980 &tad_offset);
981
982 if (ch_way == 3)
983 idx = addr >> 6;
984 else
985 idx = addr >> (6 + sck_way);
986 idx = idx % ch_way;
987
988 /*
989 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
990 */
991 switch (idx) {
992 case 0:
993 base_ch = TAD_TGT0(reg);
994 break;
995 case 1:
996 base_ch = TAD_TGT1(reg);
997 break;
998 case 2:
999 base_ch = TAD_TGT2(reg);
1000 break;
1001 case 3:
1002 base_ch = TAD_TGT3(reg);
1003 break;
1004 default:
1005 sprintf(msg, "Can't discover the TAD target");
1006 edac_mc_handle_ce_no_info(mci, msg);
1007 return -EINVAL;
1008 }
1009 *channel_mask = 1 << base_ch;
1010
1011 if (pvt->is_mirrored) {
1012 *channel_mask |= 1 << ((base_ch + 2) % 4);
1013 switch(ch_way) {
1014 case 2:
1015 case 4:
1016 sck_xch = 1 << sck_way * (ch_way >> 1);
1017 break;
1018 default:
1019 sprintf(msg, "Invalid mirror set. Can't decode addr");
1020 edac_mc_handle_ce_no_info(mci, msg);
1021 return -EINVAL;
1022 }
1023 } else
1024 sck_xch = (1 << sck_way) * ch_way;
1025
1026 if (pvt->is_lockstep)
1027 *channel_mask |= 1 << ((base_ch + 1) % 4);
1028
1029 offset = TAD_OFFSET(tad_offset);
1030
1031 debugf0("TAD#%d: address 0x%016Lx < 0x%016Lx, socket interleave %d, channel interleave %d (offset 0x%08Lx), index %d, base ch: %d, ch mask: 0x%02lx\n",
1032 n_tads,
1033 addr,
1034 limit,
1035 (u32)TAD_SOCK(reg),
1036 ch_way,
1037 offset,
1038 idx,
1039 base_ch,
1040 *channel_mask);
1041
1042 /* Calculate channel address */
1043 /* Remove the TAD offset */
1044
1045 if (offset > addr) {
1046 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
1047 offset, addr);
1048 edac_mc_handle_ce_no_info(mci, msg);
1049 return -EINVAL;
1050 }
1051 addr -= offset;
1052 /* Store the low bits [0:6] of the addr */
1053 ch_addr = addr & 0x7f;
1054 /* Remove socket wayness and remove 6 bits */
1055 addr >>= 6;
1056 addr /= sck_xch;
1057#if 0
1058 /* Divide by channel way */
1059 addr = addr / ch_way;
1060#endif
1061 /* Recover the last 6 bits */
1062 ch_addr |= addr << 6;
1063
1064 /*
1065 * Step 3) Decode rank
1066 */
1067 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
1068 pci_read_config_dword(pvt->pci_tad[base_ch],
1069 rir_way_limit[n_rir],
1070 &reg);
1071
1072 if (!IS_RIR_VALID(reg))
1073 continue;
1074
1075 limit = RIR_LIMIT(reg);
1076
1077 debugf0("RIR#%d, limit: %Lu.%03Lu GB (0x%016Lx), way: %d\n",
1078 n_rir,
1079 (limit >> 20) / 1000, (limit >> 20) % 1000,
1080 limit,
1081 1 << RIR_WAY(reg));
1082 if (ch_addr <= limit)
1083 break;
1084 }
1085 if (n_rir == MAX_RIR_RANGES) {
1086 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
1087 ch_addr);
1088 edac_mc_handle_ce_no_info(mci, msg);
1089 return -EINVAL;
1090 }
1091 rir_way = RIR_WAY(reg);
1092 if (pvt->is_close_pg)
1093 idx = (ch_addr >> 6);
1094 else
1095 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
1096 idx %= 1 << rir_way;
1097
1098 pci_read_config_dword(pvt->pci_tad[base_ch],
1099 rir_offset[n_rir][idx],
1100 &reg);
1101 *rank = RIR_RNK_TGT(reg);
1102
1103 debugf0("RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
1104 n_rir,
1105 ch_addr,
1106 limit,
1107 rir_way,
1108 idx);
1109
1110 return 0;
1111}
1112
1113/****************************************************************************
1114 Device initialization routines: put/get, init/exit
1115 ****************************************************************************/
1116
1117/*
1118 * sbridge_put_all_devices 'put' all the devices that we have
1119 * reserved via 'get'
1120 */
1121static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
1122{
1123 int i;
1124
1125 debugf0(__FILE__ ": %s()\n", __func__);
1126 for (i = 0; i < sbridge_dev->n_devs; i++) {
1127 struct pci_dev *pdev = sbridge_dev->pdev[i];
1128 if (!pdev)
1129 continue;
1130 debugf0("Removing dev %02x:%02x.%d\n",
1131 pdev->bus->number,
1132 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1133 pci_dev_put(pdev);
1134 }
1135}
1136
1137static void sbridge_put_all_devices(void)
1138{
1139 struct sbridge_dev *sbridge_dev, *tmp;
1140
1141 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
1142 sbridge_put_devices(sbridge_dev);
1143 free_sbridge_dev(sbridge_dev);
1144 }
1145}
1146
1147/*
1148 * sbridge_get_all_devices Find and perform 'get' operation on the MCH's
1149 * device/functions we want to reference for this driver
1150 *
1151 * Need to 'get' device 16 func 1 and func 2
1152 */
1153static int sbridge_get_onedevice(struct pci_dev **prev,
1154 u8 *num_mc,
1155 const struct pci_id_table *table,
1156 const unsigned devno)
1157{
1158 struct sbridge_dev *sbridge_dev;
1159 const struct pci_id_descr *dev_descr = &table->descr[devno];
1160
1161 struct pci_dev *pdev = NULL;
1162 u8 bus = 0;
1163
1164 sbridge_printk(KERN_INFO,
1165 "Seeking for: dev %02x.%d PCI ID %04x:%04x\n",
1166 dev_descr->dev, dev_descr->func,
1167 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1168
1169 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1170 dev_descr->dev_id, *prev);
1171
1172 if (!pdev) {
1173 if (*prev) {
1174 *prev = pdev;
1175 return 0;
1176 }
1177
1178 if (dev_descr->optional)
1179 return 0;
1180
1181 if (devno == 0)
1182 return -ENODEV;
1183
1184 sbridge_printk(KERN_INFO,
1185 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1186 dev_descr->dev, dev_descr->func,
1187 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1188
1189 /* End of list, leave */
1190 return -ENODEV;
1191 }
1192 bus = pdev->bus->number;
1193
1194 sbridge_dev = get_sbridge_dev(bus);
1195 if (!sbridge_dev) {
1196 sbridge_dev = alloc_sbridge_dev(bus, table);
1197 if (!sbridge_dev) {
1198 pci_dev_put(pdev);
1199 return -ENOMEM;
1200 }
1201 (*num_mc)++;
1202 }
1203
1204 if (sbridge_dev->pdev[devno]) {
1205 sbridge_printk(KERN_ERR,
1206 "Duplicated device for "
1207 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1208 bus, dev_descr->dev, dev_descr->func,
1209 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1210 pci_dev_put(pdev);
1211 return -ENODEV;
1212 }
1213
1214 sbridge_dev->pdev[devno] = pdev;
1215
1216 /* Sanity check */
1217 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1218 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1219 sbridge_printk(KERN_ERR,
1220 "Device PCI ID %04x:%04x "
1221 "has dev %02x:%d.%d instead of dev %02x:%02x.%d\n",
1222 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1223 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1224 bus, dev_descr->dev, dev_descr->func);
1225 return -ENODEV;
1226 }
1227
1228 /* Be sure that the device is enabled */
1229 if (unlikely(pci_enable_device(pdev) < 0)) {
1230 sbridge_printk(KERN_ERR,
1231 "Couldn't enable "
1232 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1233 bus, dev_descr->dev, dev_descr->func,
1234 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1235 return -ENODEV;
1236 }
1237
1238 debugf0("Detected dev %02x:%d.%d PCI ID %04x:%04x\n",
1239 bus, dev_descr->dev,
1240 dev_descr->func,
1241 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1242
1243 /*
1244 * As stated on drivers/pci/search.c, the reference count for
1245 * @from is always decremented if it is not %NULL. So, as we need
1246 * to get all devices up to null, we need to do a get for the device
1247 */
1248 pci_dev_get(pdev);
1249
1250 *prev = pdev;
1251
1252 return 0;
1253}
1254
1255static int sbridge_get_all_devices(u8 *num_mc)
1256{
1257 int i, rc;
1258 struct pci_dev *pdev = NULL;
1259 const struct pci_id_table *table = pci_dev_descr_sbridge_table;
1260
1261 while (table && table->descr) {
1262 for (i = 0; i < table->n_devs; i++) {
1263 pdev = NULL;
1264 do {
1265 rc = sbridge_get_onedevice(&pdev, num_mc,
1266 table, i);
1267 if (rc < 0) {
1268 if (i == 0) {
1269 i = table->n_devs;
1270 break;
1271 }
1272 sbridge_put_all_devices();
1273 return -ENODEV;
1274 }
1275 } while (pdev);
1276 }
1277 table++;
1278 }
1279
1280 return 0;
1281}
1282
1283static int mci_bind_devs(struct mem_ctl_info *mci,
1284 struct sbridge_dev *sbridge_dev)
1285{
1286 struct sbridge_pvt *pvt = mci->pvt_info;
1287 struct pci_dev *pdev;
1288 int i, func, slot;
1289
1290 for (i = 0; i < sbridge_dev->n_devs; i++) {
1291 pdev = sbridge_dev->pdev[i];
1292 if (!pdev)
1293 continue;
1294 slot = PCI_SLOT(pdev->devfn);
1295 func = PCI_FUNC(pdev->devfn);
1296 switch (slot) {
1297 case 12:
1298 switch (func) {
1299 case 6:
1300 pvt->pci_sad0 = pdev;
1301 break;
1302 case 7:
1303 pvt->pci_sad1 = pdev;
1304 break;
1305 default:
1306 goto error;
1307 }
1308 break;
1309 case 13:
1310 switch (func) {
1311 case 6:
1312 pvt->pci_br = pdev;
1313 break;
1314 default:
1315 goto error;
1316 }
1317 break;
1318 case 14:
1319 switch (func) {
1320 case 0:
1321 pvt->pci_ha0 = pdev;
1322 break;
1323 default:
1324 goto error;
1325 }
1326 break;
1327 case 15:
1328 switch (func) {
1329 case 0:
1330 pvt->pci_ta = pdev;
1331 break;
1332 case 1:
1333 pvt->pci_ras = pdev;
1334 break;
1335 case 2:
1336 case 3:
1337 case 4:
1338 case 5:
1339 pvt->pci_tad[func - 2] = pdev;
1340 break;
1341 default:
1342 goto error;
1343 }
1344 break;
1345 case 17:
1346 switch (func) {
1347 case 0:
1348 pvt->pci_ddrio = pdev;
1349 break;
1350 default:
1351 goto error;
1352 }
1353 break;
1354 default:
1355 goto error;
1356 }
1357
1358 debugf0("Associated PCI %02x.%02d.%d with dev = %p\n",
1359 sbridge_dev->bus,
1360 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1361 pdev);
1362 }
1363
1364 /* Check if everything were registered */
1365 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
1366 !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta ||
1367 !pvt->pci_ddrio)
1368 goto enodev;
1369
1370 for (i = 0; i < NUM_CHANNELS; i++) {
1371 if (!pvt->pci_tad[i])
1372 goto enodev;
1373 }
1374 return 0;
1375
1376enodev:
1377 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1378 return -ENODEV;
1379
1380error:
1381 sbridge_printk(KERN_ERR, "Device %d, function %d "
1382 "is out of the expected range\n",
1383 slot, func);
1384 return -EINVAL;
1385}
1386
1387/****************************************************************************
1388 Error check routines
1389 ****************************************************************************/
1390
1391/*
1392 * While Sandy Bridge has error count registers, SMI BIOS read values from
1393 * and resets the counters. So, they are not reliable for the OS to read
1394 * from them. So, we have no option but to just trust on whatever MCE is
1395 * telling us about the errors.
1396 */
1397static void sbridge_mce_output_error(struct mem_ctl_info *mci,
1398 const struct mce *m)
1399{
1400 struct mem_ctl_info *new_mci;
1401 struct sbridge_pvt *pvt = mci->pvt_info;
1402 char *type, *optype, *msg, *recoverable_msg;
1403 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
1404 bool overflow = GET_BITFIELD(m->status, 62, 62);
1405 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
1406 bool recoverable = GET_BITFIELD(m->status, 56, 56);
1407 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
1408 u32 mscod = GET_BITFIELD(m->status, 16, 31);
1409 u32 errcode = GET_BITFIELD(m->status, 0, 15);
1410 u32 channel = GET_BITFIELD(m->status, 0, 3);
1411 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
1412 long channel_mask, first_channel;
1413 u8 rank, socket;
1414 int csrow, rc, dimm;
1415 char *area_type = "Unknown";
1416
1417 if (ripv)
1418 type = "NON_FATAL";
1419 else
1420 type = "FATAL";
1421
1422 /*
1423 * According with Table 15-9 of the Intel Archictecture spec vol 3A,
1424 * memory errors should fit in this mask:
1425 * 000f 0000 1mmm cccc (binary)
1426 * where:
1427 * f = Correction Report Filtering Bit. If 1, subsequent errors
1428 * won't be shown
1429 * mmm = error type
1430 * cccc = channel
1431 * If the mask doesn't match, report an error to the parsing logic
1432 */
1433 if (! ((errcode & 0xef80) == 0x80)) {
1434 optype = "Can't parse: it is not a mem";
1435 } else {
1436 switch (optypenum) {
1437 case 0:
1438 optype = "generic undef request";
1439 break;
1440 case 1:
1441 optype = "memory read";
1442 break;
1443 case 2:
1444 optype = "memory write";
1445 break;
1446 case 3:
1447 optype = "addr/cmd";
1448 break;
1449 case 4:
1450 optype = "memory scrubbing";
1451 break;
1452 default:
1453 optype = "reserved";
1454 break;
1455 }
1456 }
1457
1458 rc = get_memory_error_data(mci, m->addr, &socket,
1459 &channel_mask, &rank, area_type);
1460 if (rc < 0)
1461 return;
1462 new_mci = get_mci_for_node_id(socket);
1463 if (!new_mci) {
1464 edac_mc_handle_ce_no_info(mci, "Error: socket got corrupted!");
1465 return;
1466 }
1467 mci = new_mci;
1468 pvt = mci->pvt_info;
1469
1470 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
1471
1472 if (rank < 4)
1473 dimm = 0;
1474 else if (rank < 8)
1475 dimm = 1;
1476 else
1477 dimm = 2;
1478
1479 csrow = pvt->csrow_map[first_channel][dimm];
1480
1481 if (uncorrected_error && recoverable)
1482 recoverable_msg = " recoverable";
1483 else
1484 recoverable_msg = "";
1485
1486 /*
1487 * FIXME: What should we do with "channel" information on mcelog?
1488 * Probably, we can just discard it, as the channel information
1489 * comes from the get_memory_error_data() address decoding
1490 */
1491 msg = kasprintf(GFP_ATOMIC,
1492 "%d %s error(s): %s on %s area %s%s: cpu=%d Err=%04x:%04x (ch=%d), "
1493 "addr = 0x%08llx => socket=%d, Channel=%ld(mask=%ld), rank=%d\n",
1494 core_err_cnt,
1495 area_type,
1496 optype,
1497 type,
1498 recoverable_msg,
1499 overflow ? "OVERFLOW" : "",
1500 m->cpu,
1501 mscod, errcode,
1502 channel, /* 1111b means not specified */
1503 (long long) m->addr,
1504 socket,
1505 first_channel, /* This is the real channel on SB */
1506 channel_mask,
1507 rank);
1508
1509 debugf0("%s", msg);
1510
1511 /* Call the helper to output message */
1512 if (uncorrected_error)
1513 edac_mc_handle_fbd_ue(mci, csrow, 0, 0, msg);
1514 else
1515 edac_mc_handle_fbd_ce(mci, csrow, 0, msg);
1516
1517 kfree(msg);
1518}
1519
1520/*
1521 * sbridge_check_error Retrieve and process errors reported by the
1522 * hardware. Called by the Core module.
1523 */
1524static void sbridge_check_error(struct mem_ctl_info *mci)
1525{
1526 struct sbridge_pvt *pvt = mci->pvt_info;
1527 int i;
1528 unsigned count = 0;
1529 struct mce *m;
1530
1531 /*
1532 * MCE first step: Copy all mce errors into a temporary buffer
1533 * We use a double buffering here, to reduce the risk of
1534 * loosing an error.
1535 */
1536 smp_rmb();
1537 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1538 % MCE_LOG_LEN;
1539 if (!count)
1540 return;
1541
1542 m = pvt->mce_outentry;
1543 if (pvt->mce_in + count > MCE_LOG_LEN) {
1544 unsigned l = MCE_LOG_LEN - pvt->mce_in;
1545
1546 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1547 smp_wmb();
1548 pvt->mce_in = 0;
1549 count -= l;
1550 m += l;
1551 }
1552 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1553 smp_wmb();
1554 pvt->mce_in += count;
1555
1556 smp_rmb();
1557 if (pvt->mce_overrun) {
1558 sbridge_printk(KERN_ERR, "Lost %d memory errors\n",
1559 pvt->mce_overrun);
1560 smp_wmb();
1561 pvt->mce_overrun = 0;
1562 }
1563
1564 /*
1565 * MCE second step: parse errors and display
1566 */
1567 for (i = 0; i < count; i++)
1568 sbridge_mce_output_error(mci, &pvt->mce_outentry[i]);
1569}
1570
1571/*
1572 * sbridge_mce_check_error Replicates mcelog routine to get errors
1573 * This routine simply queues mcelog errors, and
1574 * return. The error itself should be handled later
1575 * by sbridge_check_error.
1576 * WARNING: As this routine should be called at NMI time, extra care should
1577 * be taken to avoid deadlocks, and to be as fast as possible.
1578 */
1579static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
1580 void *data)
1581{
1582 struct mce *mce = (struct mce *)data;
1583 struct mem_ctl_info *mci;
1584 struct sbridge_pvt *pvt;
1585
1586 mci = get_mci_for_node_id(mce->socketid);
1587 if (!mci)
1588 return NOTIFY_BAD;
1589 pvt = mci->pvt_info;
1590
1591 /*
1592 * Just let mcelog handle it if the error is
1593 * outside the memory controller. A memory error
1594 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
1595 * bit 12 has an special meaning.
1596 */
1597 if ((mce->status & 0xefff) >> 7 != 1)
1598 return NOTIFY_DONE;
1599
1600 printk("sbridge: HANDLING MCE MEMORY ERROR\n");
1601
1602 printk("CPU %d: Machine Check Exception: %Lx Bank %d: %016Lx\n",
1603 mce->extcpu, mce->mcgstatus, mce->bank, mce->status);
1604 printk("TSC %llx ", mce->tsc);
1605 printk("ADDR %llx ", mce->addr);
1606 printk("MISC %llx ", mce->misc);
1607
1608 printk("PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x\n",
1609 mce->cpuvendor, mce->cpuid, mce->time,
1610 mce->socketid, mce->apicid);
1611
1612#ifdef CONFIG_SMP
1613 /* Only handle if it is the right mc controller */
1614 if (cpu_data(mce->cpu).phys_proc_id != pvt->sbridge_dev->mc)
1615 return NOTIFY_DONE;
1616#endif
1617
1618 smp_rmb();
1619 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1620 smp_wmb();
1621 pvt->mce_overrun++;
1622 return NOTIFY_DONE;
1623 }
1624
1625 /* Copy memory error at the ringbuffer */
1626 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1627 smp_wmb();
1628 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1629
1630 /* Handle fatal errors immediately */
1631 if (mce->mcgstatus & 1)
1632 sbridge_check_error(mci);
1633
1634 /* Advice mcelog that the error were handled */
1635 return NOTIFY_STOP;
1636}
1637
1638static struct notifier_block sbridge_mce_dec = {
1639 .notifier_call = sbridge_mce_check_error,
1640};
1641
1642/****************************************************************************
1643 EDAC register/unregister logic
1644 ****************************************************************************/
1645
1646static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
1647{
1648 struct mem_ctl_info *mci = sbridge_dev->mci;
1649 struct sbridge_pvt *pvt;
1650
1651 if (unlikely(!mci || !mci->pvt_info)) {
1652 debugf0("MC: " __FILE__ ": %s(): dev = %p\n",
1653 __func__, &sbridge_dev->pdev[0]->dev);
1654
1655 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
1656 return;
1657 }
1658
1659 pvt = mci->pvt_info;
1660
1661 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1662 __func__, mci, &sbridge_dev->pdev[0]->dev);
1663
1664 atomic_notifier_chain_unregister(&x86_mce_decoder_chain,
1665 &sbridge_mce_dec);
1666
1667 /* Remove MC sysfs nodes */
1668 edac_mc_del_mc(mci->dev);
1669
1670 debugf1("%s: free mci struct\n", mci->ctl_name);
1671 kfree(mci->ctl_name);
1672 edac_mc_free(mci);
1673 sbridge_dev->mci = NULL;
1674}
1675
1676static int sbridge_register_mci(struct sbridge_dev *sbridge_dev)
1677{
1678 struct mem_ctl_info *mci;
1679 struct sbridge_pvt *pvt;
1680 int rc, channels, csrows;
1681
1682 /* Check the number of active and not disabled channels */
1683 rc = sbridge_get_active_channels(sbridge_dev->bus, &channels, &csrows);
1684 if (unlikely(rc < 0))
1685 return rc;
1686
1687 /* allocate a new MC control structure */
1688 mci = edac_mc_alloc(sizeof(*pvt), csrows, channels, sbridge_dev->mc);
1689 if (unlikely(!mci))
1690 return -ENOMEM;
1691
1692 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1693 __func__, mci, &sbridge_dev->pdev[0]->dev);
1694
1695 pvt = mci->pvt_info;
1696 memset(pvt, 0, sizeof(*pvt));
1697
1698 /* Associate sbridge_dev and mci for future usage */
1699 pvt->sbridge_dev = sbridge_dev;
1700 sbridge_dev->mci = mci;
1701
1702 mci->mtype_cap = MEM_FLAG_DDR3;
1703 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1704 mci->edac_cap = EDAC_FLAG_NONE;
1705 mci->mod_name = "sbridge_edac.c";
1706 mci->mod_ver = SBRIDGE_REVISION;
1707 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
1708 mci->dev_name = pci_name(sbridge_dev->pdev[0]);
1709 mci->ctl_page_to_phys = NULL;
1710
1711 /* Set the function pointer to an actual operation function */
1712 mci->edac_check = sbridge_check_error;
1713
1714 /* Store pci devices at mci for faster access */
1715 rc = mci_bind_devs(mci, sbridge_dev);
1716 if (unlikely(rc < 0))
1717 goto fail0;
1718
1719 /* Get dimm basic config and the memory layout */
1720 get_dimm_config(mci);
1721 get_memory_layout(mci);
1722
1723 /* record ptr to the generic device */
1724 mci->dev = &sbridge_dev->pdev[0]->dev;
1725
1726 /* add this new MC control structure to EDAC's list of MCs */
1727 if (unlikely(edac_mc_add_mc(mci))) {
1728 debugf0("MC: " __FILE__
1729 ": %s(): failed edac_mc_add_mc()\n", __func__);
1730 rc = -EINVAL;
1731 goto fail0;
1732 }
1733
1734 atomic_notifier_chain_register(&x86_mce_decoder_chain,
1735 &sbridge_mce_dec);
1736 return 0;
1737
1738fail0:
1739 kfree(mci->ctl_name);
1740 edac_mc_free(mci);
1741 sbridge_dev->mci = NULL;
1742 return rc;
1743}
1744
1745/*
1746 * sbridge_probe Probe for ONE instance of device to see if it is
1747 * present.
1748 * return:
1749 * 0 for FOUND a device
1750 * < 0 for error code
1751 */
1752
1753static int __devinit sbridge_probe(struct pci_dev *pdev,
1754 const struct pci_device_id *id)
1755{
1756 int rc;
1757 u8 mc, num_mc = 0;
1758 struct sbridge_dev *sbridge_dev;
1759
1760 /* get the pci devices we want to reserve for our use */
1761 mutex_lock(&sbridge_edac_lock);
1762
1763 /*
1764 * All memory controllers are allocated at the first pass.
1765 */
1766 if (unlikely(probed >= 1)) {
1767 mutex_unlock(&sbridge_edac_lock);
1768 return -ENODEV;
1769 }
1770 probed++;
1771
1772 rc = sbridge_get_all_devices(&num_mc);
1773 if (unlikely(rc < 0))
1774 goto fail0;
1775 mc = 0;
1776
1777 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1778 debugf0("Registering MC#%d (%d of %d)\n", mc, mc + 1, num_mc);
1779 sbridge_dev->mc = mc++;
1780 rc = sbridge_register_mci(sbridge_dev);
1781 if (unlikely(rc < 0))
1782 goto fail1;
1783 }
1784
1785 sbridge_printk(KERN_INFO, "Driver loaded.\n");
1786
1787 mutex_unlock(&sbridge_edac_lock);
1788 return 0;
1789
1790fail1:
1791 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1792 sbridge_unregister_mci(sbridge_dev);
1793
1794 sbridge_put_all_devices();
1795fail0:
1796 mutex_unlock(&sbridge_edac_lock);
1797 return rc;
1798}
1799
1800/*
1801 * sbridge_remove destructor for one instance of device
1802 *
1803 */
1804static void __devexit sbridge_remove(struct pci_dev *pdev)
1805{
1806 struct sbridge_dev *sbridge_dev;
1807
1808 debugf0(__FILE__ ": %s()\n", __func__);
1809
1810 /*
1811 * we have a trouble here: pdev value for removal will be wrong, since
1812 * it will point to the X58 register used to detect that the machine
1813 * is a Nehalem or upper design. However, due to the way several PCI
1814 * devices are grouped together to provide MC functionality, we need
1815 * to use a different method for releasing the devices
1816 */
1817
1818 mutex_lock(&sbridge_edac_lock);
1819
1820 if (unlikely(!probed)) {
1821 mutex_unlock(&sbridge_edac_lock);
1822 return;
1823 }
1824
1825 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1826 sbridge_unregister_mci(sbridge_dev);
1827
1828 /* Release PCI resources */
1829 sbridge_put_all_devices();
1830
1831 probed--;
1832
1833 mutex_unlock(&sbridge_edac_lock);
1834}
1835
1836MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl);
1837
1838/*
1839 * sbridge_driver pci_driver structure for this module
1840 *
1841 */
1842static struct pci_driver sbridge_driver = {
1843 .name = "sbridge_edac",
1844 .probe = sbridge_probe,
1845 .remove = __devexit_p(sbridge_remove),
1846 .id_table = sbridge_pci_tbl,
1847};
1848
1849/*
1850 * sbridge_init Module entry function
1851 * Try to initialize this module for its devices
1852 */
1853static int __init sbridge_init(void)
1854{
1855 int pci_rc;
1856
1857 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1858
1859 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1860 opstate_init();
1861
1862 pci_rc = pci_register_driver(&sbridge_driver);
1863
1864 if (pci_rc >= 0)
1865 return 0;
1866
1867 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
1868 pci_rc);
1869
1870 return pci_rc;
1871}
1872
1873/*
1874 * sbridge_exit() Module exit function
1875 * Unregister the driver
1876 */
1877static void __exit sbridge_exit(void)
1878{
1879 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1880 pci_unregister_driver(&sbridge_driver);
1881}
1882
1883module_init(sbridge_init);
1884module_exit(sbridge_exit);
1885
1886module_param(edac_op_state, int, 0444);
1887MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
1888
1889MODULE_LICENSE("GPL");
1890MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
1891MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
1892MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge memory controllers - "
1893 SBRIDGE_REVISION);
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-01 14:00:46 -0500