blob: f89cfd85ae134b1530ef76a46970d4fc044f66a6
1 | EDAC - Error Detection And Correction |
2 | ===================================== |
3 | |
4 | "bluesmoke" was the name for this device driver when it |
5 | was "out-of-tree" and maintained at sourceforge.net - |
6 | bluesmoke.sourceforge.net. That site is mostly archaic now and can be |
7 | used only for historical purposes. |
8 | |
9 | When the subsystem was pushed into 2.6.16 for the first time, it was |
10 | renamed to 'EDAC'. |
11 | |
12 | PURPOSE |
13 | ------- |
14 | |
15 | The 'edac' kernel module's goal is to detect and report hardware errors |
16 | that occur within the computer system running under linux. |
17 | |
18 | MEMORY |
19 | ------ |
20 | |
21 | Memory Correctable Errors (CE) and Uncorrectable Errors (UE) are the |
22 | primary errors being harvested. These types of errors are harvested by |
23 | the 'edac_mc' device. |
24 | |
25 | Detecting CE events, then harvesting those events and reporting them, |
26 | *can* but must not necessarily be a predictor of future UE events. With |
27 | CE events only, the system can and will continue to operate as no data |
28 | has been damaged yet. |
29 | |
30 | However, preventive maintenance and proactive part replacement of memory |
31 | DIMMs exhibiting CEs can reduce the likelihood of the dreaded UE events |
32 | and system panics. |
33 | |
34 | OTHER HARDWARE ELEMENTS |
35 | ----------------------- |
36 | |
37 | A new feature for EDAC, the edac_device class of device, was added in |
38 | the 2.6.23 version of the kernel. |
39 | |
40 | This new device type allows for non-memory type of ECC hardware detectors |
41 | to have their states harvested and presented to userspace via the sysfs |
42 | interface. |
43 | |
44 | Some architectures have ECC detectors for L1, L2 and L3 caches, |
45 | along with DMA engines, fabric switches, main data path switches, |
46 | interconnections, and various other hardware data paths. If the hardware |
47 | reports it, then a edac_device device probably can be constructed to |
48 | harvest and present that to userspace. |
49 | |
50 | |
51 | PCI BUS SCANNING |
52 | ---------------- |
53 | |
54 | In addition, PCI devices are scanned for PCI Bus Parity and SERR Errors |
55 | in order to determine if errors are occurring during data transfers. |
56 | |
57 | The presence of PCI Parity errors must be examined with a grain of salt. |
58 | There are several add-in adapters that do *not* follow the PCI specification |
59 | with regards to Parity generation and reporting. The specification says |
60 | the vendor should tie the parity status bits to 0 if they do not intend |
61 | to generate parity. Some vendors do not do this, and thus the parity bit |
62 | can "float" giving false positives. |
63 | |
64 | There is a PCI device attribute located in sysfs that is checked by |
65 | the EDAC PCI scanning code. If that attribute is set, PCI parity/error |
66 | scanning is skipped for that device. The attribute is: |
67 | |
68 | broken_parity_status |
69 | |
70 | and is located in /sys/devices/pci<XXX>/0000:XX:YY.Z directories for |
71 | PCI devices. |
72 | |
73 | |
74 | VERSIONING |
75 | ---------- |
76 | |
77 | EDAC is composed of a "core" module (edac_core.ko) and several Memory |
78 | Controller (MC) driver modules. On a given system, the CORE is loaded |
79 | and one MC driver will be loaded. Both the CORE and the MC driver (or |
80 | edac_device driver) have individual versions that reflect current |
81 | release level of their respective modules. |
82 | |
83 | Thus, to "report" on what version a system is running, one must report |
84 | both the CORE's and the MC driver's versions. |
85 | |
86 | |
87 | LOADING |
88 | ------- |
89 | |
90 | If 'edac' was statically linked with the kernel then no loading |
91 | is necessary. If 'edac' was built as modules then simply modprobe |
92 | the 'edac' pieces that you need. You should be able to modprobe |
93 | hardware-specific modules and have the dependencies load the necessary |
94 | core modules. |
95 | |
96 | Example: |
97 | |
98 | $> modprobe amd76x_edac |
99 | |
100 | loads both the amd76x_edac.ko memory controller module and the edac_mc.ko |
101 | core module. |
102 | |
103 | |
104 | SYSFS INTERFACE |
105 | --------------- |
106 | |
107 | EDAC presents a 'sysfs' interface for control and reporting purposes. It |
108 | lives in the /sys/devices/system/edac directory. |
109 | |
110 | Within this directory there currently reside 2 components: |
111 | |
112 | mc memory controller(s) system |
113 | pci PCI control and status system |
114 | |
115 | |
116 | |
117 | Memory Controller (mc) Model |
118 | ---------------------------- |
119 | |
120 | Each 'mc' device controls a set of DIMM memory modules. These modules |
121 | are laid out in a Chip-Select Row (csrowX) and Channel table (chX). |
122 | There can be multiple csrows and multiple channels. |
123 | |
124 | Memory controllers allow for several csrows, with 8 csrows being a |
125 | typical value. Yet, the actual number of csrows depends on the layout of |
126 | a given motherboard, memory controller and DIMM characteristics. |
127 | |
128 | Dual channels allows for 128 bit data transfers to/from the CPU from/to |
129 | memory. Some newer chipsets allow for more than 2 channels, like Fully |
130 | Buffered DIMMs (FB-DIMMs). The following example will assume 2 channels: |
131 | |
132 | |
133 | Channel 0 Channel 1 |
134 | =================================== |
135 | csrow0 | DIMM_A0 | DIMM_B0 | |
136 | csrow1 | DIMM_A0 | DIMM_B0 | |
137 | =================================== |
138 | |
139 | =================================== |
140 | csrow2 | DIMM_A1 | DIMM_B1 | |
141 | csrow3 | DIMM_A1 | DIMM_B1 | |
142 | =================================== |
143 | |
144 | In the above example table there are 4 physical slots on the motherboard |
145 | for memory DIMMs: |
146 | |
147 | DIMM_A0 |
148 | DIMM_B0 |
149 | DIMM_A1 |
150 | DIMM_B1 |
151 | |
152 | Labels for these slots are usually silk-screened on the motherboard. |
153 | Slots labeled 'A' are channel 0 in this example. Slots labeled 'B' are |
154 | channel 1. Notice that there are two csrows possible on a physical DIMM. |
155 | These csrows are allocated their csrow assignment based on the slot into |
156 | which the memory DIMM is placed. Thus, when 1 DIMM is placed in each |
157 | Channel, the csrows cross both DIMMs. |
158 | |
159 | Memory DIMMs come single or dual "ranked". A rank is a populated csrow. |
160 | Thus, 2 single ranked DIMMs, placed in slots DIMM_A0 and DIMM_B0 above |
161 | will have 1 csrow, csrow0. csrow1 will be empty. On the other hand, |
162 | when 2 dual ranked DIMMs are similarly placed, then both csrow0 and |
163 | csrow1 will be populated. The pattern repeats itself for csrow2 and |
164 | csrow3. |
165 | |
166 | The representation of the above is reflected in the directory |
167 | tree in EDAC's sysfs interface. Starting in directory |
168 | /sys/devices/system/edac/mc each memory controller will be represented |
169 | by its own 'mcX' directory, where 'X' is the index of the MC. |
170 | |
171 | |
172 | ..../edac/mc/ |
173 | | |
174 | |->mc0 |
175 | |->mc1 |
176 | |->mc2 |
177 | .... |
178 | |
179 | Under each 'mcX' directory each 'csrowX' is again represented by a |
180 | 'csrowX', where 'X' is the csrow index: |
181 | |
182 | |
183 | .../mc/mc0/ |
184 | | |
185 | |->csrow0 |
186 | |->csrow2 |
187 | |->csrow3 |
188 | .... |
189 | |
190 | Notice that there is no csrow1, which indicates that csrow0 is composed |
191 | of a single ranked DIMMs. This should also apply in both Channels, in |
192 | order to have dual-channel mode be operational. Since both csrow2 and |
193 | csrow3 are populated, this indicates a dual ranked set of DIMMs for |
194 | channels 0 and 1. |
195 | |
196 | |
197 | Within each of the 'mcX' and 'csrowX' directories are several EDAC |
198 | control and attribute files. |
199 | |
200 | |
201 | 'mcX' directories |
202 | ----------------- |
203 | |
204 | In 'mcX' directories are EDAC control and attribute files for |
205 | this 'X' instance of the memory controllers. |
206 | |
207 | For a description of the sysfs API, please see: |
208 | Documentation/ABI/testing/sysfs-devices-edac |
209 | |
210 | |
211 | |
212 | 'csrowX' directories |
213 | -------------------- |
214 | |
215 | When CONFIG_EDAC_LEGACY_SYSFS is enabled, sysfs will contain the csrowX |
216 | directories. As this API doesn't work properly for Rambus, FB-DIMMs and |
217 | modern Intel Memory Controllers, this is being deprecated in favor of |
218 | dimmX directories. |
219 | |
220 | In the 'csrowX' directories are EDAC control and attribute files for |
221 | this 'X' instance of csrow: |
222 | |
223 | |
224 | Total Uncorrectable Errors count attribute file: |
225 | |
226 | 'ue_count' |
227 | |
228 | This attribute file displays the total count of uncorrectable |
229 | errors that have occurred on this csrow. If panic_on_ue is set |
230 | this counter will not have a chance to increment, since EDAC |
231 | will panic the system. |
232 | |
233 | |
234 | Total Correctable Errors count attribute file: |
235 | |
236 | 'ce_count' |
237 | |
238 | This attribute file displays the total count of correctable |
239 | errors that have occurred on this csrow. This count is very |
240 | important to examine. CEs provide early indications that a |
241 | DIMM is beginning to fail. This count field should be |
242 | monitored for non-zero values and report such information |
243 | to the system administrator. |
244 | |
245 | |
246 | Total memory managed by this csrow attribute file: |
247 | |
248 | 'size_mb' |
249 | |
250 | This attribute file displays, in count of megabytes, the memory |
251 | that this csrow contains. |
252 | |
253 | |
254 | Memory Type attribute file: |
255 | |
256 | 'mem_type' |
257 | |
258 | This attribute file will display what type of memory is currently |
259 | on this csrow. Normally, either buffered or unbuffered memory. |
260 | Examples: |
261 | Registered-DDR |
262 | Unbuffered-DDR |
263 | |
264 | |
265 | EDAC Mode of operation attribute file: |
266 | |
267 | 'edac_mode' |
268 | |
269 | This attribute file will display what type of Error detection |
270 | and correction is being utilized. |
271 | |
272 | |
273 | Device type attribute file: |
274 | |
275 | 'dev_type' |
276 | |
277 | This attribute file will display what type of DRAM device is |
278 | being utilized on this DIMM. |
279 | Examples: |
280 | x1 |
281 | x2 |
282 | x4 |
283 | x8 |
284 | |
285 | |
286 | Channel 0 CE Count attribute file: |
287 | |
288 | 'ch0_ce_count' |
289 | |
290 | This attribute file will display the count of CEs on this |
291 | DIMM located in channel 0. |
292 | |
293 | |
294 | Channel 0 UE Count attribute file: |
295 | |
296 | 'ch0_ue_count' |
297 | |
298 | This attribute file will display the count of UEs on this |
299 | DIMM located in channel 0. |
300 | |
301 | |
302 | Channel 0 DIMM Label control file: |
303 | |
304 | 'ch0_dimm_label' |
305 | |
306 | This control file allows this DIMM to have a label assigned |
307 | to it. With this label in the module, when errors occur |
308 | the output can provide the DIMM label in the system log. |
309 | This becomes vital for panic events to isolate the |
310 | cause of the UE event. |
311 | |
312 | DIMM Labels must be assigned after booting, with information |
313 | that correctly identifies the physical slot with its |
314 | silk screen label. This information is currently very |
315 | motherboard specific and determination of this information |
316 | must occur in userland at this time. |
317 | |
318 | |
319 | Channel 1 CE Count attribute file: |
320 | |
321 | 'ch1_ce_count' |
322 | |
323 | This attribute file will display the count of CEs on this |
324 | DIMM located in channel 1. |
325 | |
326 | |
327 | Channel 1 UE Count attribute file: |
328 | |
329 | 'ch1_ue_count' |
330 | |
331 | This attribute file will display the count of UEs on this |
332 | DIMM located in channel 0. |
333 | |
334 | |
335 | Channel 1 DIMM Label control file: |
336 | |
337 | 'ch1_dimm_label' |
338 | |
339 | This control file allows this DIMM to have a label assigned |
340 | to it. With this label in the module, when errors occur |
341 | the output can provide the DIMM label in the system log. |
342 | This becomes vital for panic events to isolate the |
343 | cause of the UE event. |
344 | |
345 | DIMM Labels must be assigned after booting, with information |
346 | that correctly identifies the physical slot with its |
347 | silk screen label. This information is currently very |
348 | motherboard specific and determination of this information |
349 | must occur in userland at this time. |
350 | |
351 | |
352 | |
353 | SYSTEM LOGGING |
354 | -------------- |
355 | |
356 | If logging for UEs and CEs is enabled, then system logs will contain |
357 | information indicating that errors have been detected: |
358 | |
359 | EDAC MC0: CE page 0x283, offset 0xce0, grain 8, syndrome 0x6ec3, row 0, |
360 | channel 1 "DIMM_B1": amd76x_edac |
361 | |
362 | EDAC MC0: CE page 0x1e5, offset 0xfb0, grain 8, syndrome 0xb741, row 0, |
363 | channel 1 "DIMM_B1": amd76x_edac |
364 | |
365 | |
366 | The structure of the message is: |
367 | the memory controller (MC0) |
368 | Error type (CE) |
369 | memory page (0x283) |
370 | offset in the page (0xce0) |
371 | the byte granularity (grain 8) |
372 | or resolution of the error |
373 | the error syndrome (0xb741) |
374 | memory row (row 0) |
375 | memory channel (channel 1) |
376 | DIMM label, if set prior (DIMM B1 |
377 | and then an optional, driver-specific message that may |
378 | have additional information. |
379 | |
380 | Both UEs and CEs with no info will lack all but memory controller, error |
381 | type, a notice of "no info" and then an optional, driver-specific error |
382 | message. |
383 | |
384 | |
385 | PCI Bus Parity Detection |
386 | ------------------------ |
387 | |
388 | On Header Type 00 devices, the primary status is looked at for any |
389 | parity error regardless of whether parity is enabled on the device or |
390 | not. (The spec indicates parity is generated in some cases). On Header |
391 | Type 01 bridges, the secondary status register is also looked at to see |
392 | if parity occurred on the bus on the other side of the bridge. |
393 | |
394 | |
395 | SYSFS CONFIGURATION |
396 | ------------------- |
397 | |
398 | Under /sys/devices/system/edac/pci are control and attribute files as follows: |
399 | |
400 | |
401 | Enable/Disable PCI Parity checking control file: |
402 | |
403 | 'check_pci_parity' |
404 | |
405 | |
406 | This control file enables or disables the PCI Bus Parity scanning |
407 | operation. Writing a 1 to this file enables the scanning. Writing |
408 | a 0 to this file disables the scanning. |
409 | |
410 | Enable: |
411 | echo "1" >/sys/devices/system/edac/pci/check_pci_parity |
412 | |
413 | Disable: |
414 | echo "0" >/sys/devices/system/edac/pci/check_pci_parity |
415 | |
416 | |
417 | Parity Count: |
418 | |
419 | 'pci_parity_count' |
420 | |
421 | This attribute file will display the number of parity errors that |
422 | have been detected. |
423 | |
424 | |
425 | |
426 | MODULE PARAMETERS |
427 | ----------------- |
428 | |
429 | Panic on UE control file: |
430 | |
431 | 'edac_mc_panic_on_ue' |
432 | |
433 | An uncorrectable error will cause a machine panic. This is usually |
434 | desirable. It is a bad idea to continue when an uncorrectable error |
435 | occurs - it is indeterminate what was uncorrected and the operating |
436 | system context might be so mangled that continuing will lead to further |
437 | corruption. If the kernel has MCE configured, then EDAC will never |
438 | notice the UE. |
439 | |
440 | LOAD TIME: module/kernel parameter: edac_mc_panic_on_ue=[0|1] |
441 | |
442 | RUN TIME: echo "1" > /sys/module/edac_core/parameters/edac_mc_panic_on_ue |
443 | |
444 | |
445 | Log UE control file: |
446 | |
447 | 'edac_mc_log_ue' |
448 | |
449 | Generate kernel messages describing uncorrectable errors. These errors |
450 | are reported through the system message log system. UE statistics |
451 | will be accumulated even when UE logging is disabled. |
452 | |
453 | LOAD TIME: module/kernel parameter: edac_mc_log_ue=[0|1] |
454 | |
455 | RUN TIME: echo "1" > /sys/module/edac_core/parameters/edac_mc_log_ue |
456 | |
457 | |
458 | Log CE control file: |
459 | |
460 | 'edac_mc_log_ce' |
461 | |
462 | Generate kernel messages describing correctable errors. These |
463 | errors are reported through the system message log system. |
464 | CE statistics will be accumulated even when CE logging is disabled. |
465 | |
466 | LOAD TIME: module/kernel parameter: edac_mc_log_ce=[0|1] |
467 | |
468 | RUN TIME: echo "1" > /sys/module/edac_core/parameters/edac_mc_log_ce |
469 | |
470 | |
471 | Polling period control file: |
472 | |
473 | 'edac_mc_poll_msec' |
474 | |
475 | The time period, in milliseconds, for polling for error information. |
476 | Too small a value wastes resources. Too large a value might delay |
477 | necessary handling of errors and might loose valuable information for |
478 | locating the error. 1000 milliseconds (once each second) is the current |
479 | default. Systems which require all the bandwidth they can get, may |
480 | increase this. |
481 | |
482 | LOAD TIME: module/kernel parameter: edac_mc_poll_msec=[0|1] |
483 | |
484 | RUN TIME: echo "1000" > /sys/module/edac_core/parameters/edac_mc_poll_msec |
485 | |
486 | |
487 | Panic on PCI PARITY Error: |
488 | |
489 | 'panic_on_pci_parity' |
490 | |
491 | |
492 | This control file enables or disables panicking when a parity |
493 | error has been detected. |
494 | |
495 | |
496 | module/kernel parameter: edac_panic_on_pci_pe=[0|1] |
497 | |
498 | Enable: |
499 | echo "1" > /sys/module/edac_core/parameters/edac_panic_on_pci_pe |
500 | |
501 | Disable: |
502 | echo "0" > /sys/module/edac_core/parameters/edac_panic_on_pci_pe |
503 | |
504 | |
505 | |
506 | EDAC device type |
507 | ---------------- |
508 | |
509 | In the header file, edac_core.h, there is a series of edac_device structures |
510 | and APIs for the EDAC_DEVICE. |
511 | |
512 | User space access to an edac_device is through the sysfs interface. |
513 | |
514 | At the location /sys/devices/system/edac (sysfs) new edac_device devices will |
515 | appear. |
516 | |
517 | There is a three level tree beneath the above 'edac' directory. For example, |
518 | the 'test_device_edac' device (found at the bluesmoke.sourceforget.net website) |
519 | installs itself as: |
520 | |
521 | /sys/devices/systm/edac/test-instance |
522 | |
523 | in this directory are various controls, a symlink and one or more 'instance' |
524 | directories. |
525 | |
526 | The standard default controls are: |
527 | |
528 | log_ce boolean to log CE events |
529 | log_ue boolean to log UE events |
530 | panic_on_ue boolean to 'panic' the system if an UE is encountered |
531 | (default off, can be set true via startup script) |
532 | poll_msec time period between POLL cycles for events |
533 | |
534 | The test_device_edac device adds at least one of its own custom control: |
535 | |
536 | test_bits which in the current test driver does nothing but |
537 | show how it is installed. A ported driver can |
538 | add one or more such controls and/or attributes |
539 | for specific uses. |
540 | One out-of-tree driver uses controls here to allow |
541 | for ERROR INJECTION operations to hardware |
542 | injection registers |
543 | |
544 | The symlink points to the 'struct dev' that is registered for this edac_device. |
545 | |
546 | INSTANCES |
547 | --------- |
548 | |
549 | One or more instance directories are present. For the 'test_device_edac' case: |
550 | |
551 | test-instance0 |
552 | |
553 | |
554 | In this directory there are two default counter attributes, which are totals of |
555 | counter in deeper subdirectories. |
556 | |
557 | ce_count total of CE events of subdirectories |
558 | ue_count total of UE events of subdirectories |
559 | |
560 | BLOCKS |
561 | ------ |
562 | |
563 | At the lowest directory level is the 'block' directory. There can be 0, 1 |
564 | or more blocks specified in each instance. |
565 | |
566 | test-block0 |
567 | |
568 | |
569 | In this directory the default attributes are: |
570 | |
571 | ce_count which is counter of CE events for this 'block' |
572 | of hardware being monitored |
573 | ue_count which is counter of UE events for this 'block' |
574 | of hardware being monitored |
575 | |
576 | |
577 | The 'test_device_edac' device adds 4 attributes and 1 control: |
578 | |
579 | test-block-bits-0 for every POLL cycle this counter |
580 | is incremented |
581 | test-block-bits-1 every 10 cycles, this counter is bumped once, |
582 | and test-block-bits-0 is set to 0 |
583 | test-block-bits-2 every 100 cycles, this counter is bumped once, |
584 | and test-block-bits-1 is set to 0 |
585 | test-block-bits-3 every 1000 cycles, this counter is bumped once, |
586 | and test-block-bits-2 is set to 0 |
587 | |
588 | |
589 | reset-counters writing ANY thing to this control will |
590 | reset all the above counters. |
591 | |
592 | |
593 | Use of the 'test_device_edac' driver should enable any others to create their own |
594 | unique drivers for their hardware systems. |
595 | |
596 | The 'test_device_edac' sample driver is located at the |
597 | bluesmoke.sourceforge.net project site for EDAC. |
598 | |
599 | |
600 | NEHALEM USAGE OF EDAC APIs |
601 | -------------------------- |
602 | |
603 | This chapter documents some EXPERIMENTAL mappings for EDAC API to handle |
604 | Nehalem EDAC driver. They will likely be changed on future versions |
605 | of the driver. |
606 | |
607 | Due to the way Nehalem exports Memory Controller data, some adjustments |
608 | were done at i7core_edac driver. This chapter will cover those differences |
609 | |
610 | 1) On Nehalem, there is one Memory Controller per Quick Patch Interconnect |
611 | (QPI). At the driver, the term "socket" means one QPI. This is |
612 | associated with a physical CPU socket. |
613 | |
614 | Each MC have 3 physical read channels, 3 physical write channels and |
615 | 3 logic channels. The driver currently sees it as just 3 channels. |
616 | Each channel can have up to 3 DIMMs. |
617 | |
618 | The minimum known unity is DIMMs. There are no information about csrows. |
619 | As EDAC API maps the minimum unity is csrows, the driver sequentially |
620 | maps channel/dimm into different csrows. |
621 | |
622 | For example, supposing the following layout: |
623 | Ch0 phy rd0, wr0 (0x063f4031): 2 ranks, UDIMMs |
624 | dimm 0 1024 Mb offset: 0, bank: 8, rank: 1, row: 0x4000, col: 0x400 |
625 | dimm 1 1024 Mb offset: 4, bank: 8, rank: 1, row: 0x4000, col: 0x400 |
626 | Ch1 phy rd1, wr1 (0x063f4031): 2 ranks, UDIMMs |
627 | dimm 0 1024 Mb offset: 0, bank: 8, rank: 1, row: 0x4000, col: 0x400 |
628 | Ch2 phy rd3, wr3 (0x063f4031): 2 ranks, UDIMMs |
629 | dimm 0 1024 Mb offset: 0, bank: 8, rank: 1, row: 0x4000, col: 0x400 |
630 | The driver will map it as: |
631 | csrow0: channel 0, dimm0 |
632 | csrow1: channel 0, dimm1 |
633 | csrow2: channel 1, dimm0 |
634 | csrow3: channel 2, dimm0 |
635 | |
636 | exports one |
637 | DIMM per csrow. |
638 | |
639 | Each QPI is exported as a different memory controller. |
640 | |
641 | 2) Nehalem MC has the ability to generate errors. The driver implements this |
642 | functionality via some error injection nodes: |
643 | |
644 | For injecting a memory error, there are some sysfs nodes, under |
645 | /sys/devices/system/edac/mc/mc?/: |
646 | |
647 | inject_addrmatch/*: |
648 | Controls the error injection mask register. It is possible to specify |
649 | several characteristics of the address to match an error code: |
650 | dimm = the affected dimm. Numbers are relative to a channel; |
651 | rank = the memory rank; |
652 | channel = the channel that will generate an error; |
653 | bank = the affected bank; |
654 | page = the page address; |
655 | column (or col) = the address column. |
656 | each of the above values can be set to "any" to match any valid value. |
657 | |
658 | At driver init, all values are set to any. |
659 | |
660 | For example, to generate an error at rank 1 of dimm 2, for any channel, |
661 | any bank, any page, any column: |
662 | echo 2 >/sys/devices/system/edac/mc/mc0/inject_addrmatch/dimm |
663 | echo 1 >/sys/devices/system/edac/mc/mc0/inject_addrmatch/rank |
664 | |
665 | To return to the default behaviour of matching any, you can do: |
666 | echo any >/sys/devices/system/edac/mc/mc0/inject_addrmatch/dimm |
667 | echo any >/sys/devices/system/edac/mc/mc0/inject_addrmatch/rank |
668 | |
669 | inject_eccmask: |
670 | specifies what bits will have troubles, |
671 | |
672 | inject_section: |
673 | specifies what ECC cache section will get the error: |
674 | 3 for both |
675 | 2 for the highest |
676 | 1 for the lowest |
677 | |
678 | inject_type: |
679 | specifies the type of error, being a combination of the following bits: |
680 | bit 0 - repeat |
681 | bit 1 - ecc |
682 | bit 2 - parity |
683 | |
684 | inject_enable starts the error generation when something different |
685 | than 0 is written. |
686 | |
687 | All inject vars can be read. root permission is needed for write. |
688 | |
689 | Datasheet states that the error will only be generated after a write on an |
690 | address that matches inject_addrmatch. It seems, however, that reading will |
691 | also produce an error. |
692 | |
693 | For example, the following code will generate an error for any write access |
694 | at socket 0, on any DIMM/address on channel 2: |
695 | |
696 | echo 2 >/sys/devices/system/edac/mc/mc0/inject_addrmatch/channel |
697 | echo 2 >/sys/devices/system/edac/mc/mc0/inject_type |
698 | echo 64 >/sys/devices/system/edac/mc/mc0/inject_eccmask |
699 | echo 3 >/sys/devices/system/edac/mc/mc0/inject_section |
700 | echo 1 >/sys/devices/system/edac/mc/mc0/inject_enable |
701 | dd if=/dev/mem of=/dev/null seek=16k bs=4k count=1 >& /dev/null |
702 | |
703 | For socket 1, it is needed to replace "mc0" by "mc1" at the above |
704 | commands. |
705 | |
706 | The generated error message will look like: |
707 | |
708 | EDAC MC0: UE row 0, channel-a= 0 channel-b= 0 labels "-": NON_FATAL (addr = 0x0075b980, socket=0, Dimm=0, Channel=2, syndrome=0x00000040, count=1, Err=8c0000400001009f:4000080482 (read error: read ECC error)) |
709 | |
710 | 3) Nehalem specific Corrected Error memory counters |
711 | |
712 | Nehalem have some registers to count memory errors. The driver uses those |
713 | registers to report Corrected Errors on devices with Registered Dimms. |
714 | |
715 | However, those counters don't work with Unregistered Dimms. As the chipset |
716 | offers some counters that also work with UDIMMS (but with a worse level of |
717 | granularity than the default ones), the driver exposes those registers for |
718 | UDIMM memories. |
719 | |
720 | They can be read by looking at the contents of all_channel_counts/ |
721 | |
722 | $ for i in /sys/devices/system/edac/mc/mc0/all_channel_counts/*; do echo $i; cat $i; done |
723 | /sys/devices/system/edac/mc/mc0/all_channel_counts/udimm0 |
724 | 0 |
725 | /sys/devices/system/edac/mc/mc0/all_channel_counts/udimm1 |
726 | 0 |
727 | /sys/devices/system/edac/mc/mc0/all_channel_counts/udimm2 |
728 | 0 |
729 | |
730 | What happens here is that errors on different csrows, but at the same |
731 | dimm number will increment the same counter. |
732 | So, in this memory mapping: |
733 | csrow0: channel 0, dimm0 |
734 | csrow1: channel 0, dimm1 |
735 | csrow2: channel 1, dimm0 |
736 | csrow3: channel 2, dimm0 |
737 | The hardware will increment udimm0 for an error at the first dimm at either |
738 | csrow0, csrow2 or csrow3; |
739 | The hardware will increment udimm1 for an error at the second dimm at either |
740 | csrow0, csrow2 or csrow3; |
741 | The hardware will increment udimm2 for an error at the third dimm at either |
742 | csrow0, csrow2 or csrow3; |
743 | |
744 | 4) Standard error counters |
745 | |
746 | The standard error counters are generated when an mcelog error is received |
747 | by the driver. Since, with udimm, this is counted by software, it is |
748 | possible that some errors could be lost. With rdimm's, they display the |
749 | contents of the registers |
750 | |
751 | AMD64_EDAC REFERENCE DOCUMENTS USED |
752 | ----------------------------------- |
753 | amd64_edac module is based on the following documents |
754 | (available from http://support.amd.com/en-us/search/tech-docs): |
755 | |
756 | 1. Title: BIOS and Kernel Developer's Guide for AMD Athlon 64 and AMD |
757 | Opteron Processors |
758 | AMD publication #: 26094 |
759 | Revision: 3.26 |
760 | Link: http://support.amd.com/TechDocs/26094.PDF |
761 | |
762 | 2. Title: BIOS and Kernel Developer's Guide for AMD NPT Family 0Fh |
763 | Processors |
764 | AMD publication #: 32559 |
765 | Revision: 3.00 |
766 | Issue Date: May 2006 |
767 | Link: http://support.amd.com/TechDocs/32559.pdf |
768 | |
769 | 3. Title: BIOS and Kernel Developer's Guide (BKDG) For AMD Family 10h |
770 | Processors |
771 | AMD publication #: 31116 |
772 | Revision: 3.00 |
773 | Issue Date: September 07, 2007 |
774 | Link: http://support.amd.com/TechDocs/31116.pdf |
775 | |
776 | 4. Title: BIOS and Kernel Developer's Guide (BKDG) for AMD Family 15h |
777 | Models 30h-3Fh Processors |
778 | AMD publication #: 49125 |
779 | Revision: 3.06 |
780 | Issue Date: 2/12/2015 (latest release) |
781 | Link: http://support.amd.com/TechDocs/49125_15h_Models_30h-3Fh_BKDG.pdf |
782 | |
783 | 5. Title: BIOS and Kernel Developer's Guide (BKDG) for AMD Family 15h |
784 | Models 60h-6Fh Processors |
785 | AMD publication #: 50742 |
786 | Revision: 3.01 |
787 | Issue Date: 7/23/2015 (latest release) |
788 | Link: http://support.amd.com/TechDocs/50742_15h_Models_60h-6Fh_BKDG.pdf |
789 | |
790 | 6. Title: BIOS and Kernel Developer's Guide (BKDG) for AMD Family 16h |
791 | Models 00h-0Fh Processors |
792 | AMD publication #: 48751 |
793 | Revision: 3.03 |
794 | Issue Date: 2/23/2015 (latest release) |
795 | Link: http://support.amd.com/TechDocs/48751_16h_bkdg.pdf |
796 | |
797 | CREDITS: |
798 | ======== |
799 | |
800 | Written by Doug Thompson <dougthompson@xmission.com> |
801 | 7 Dec 2005 |
802 | 17 Jul 2007 Updated |
803 | |
804 | (c) Mauro Carvalho Chehab |
805 | 05 Aug 2009 Nehalem interface |
806 | |
807 | EDAC authors/maintainers: |
808 | |
809 | Doug Thompson, Dave Jiang, Dave Peterson et al, |
810 | Mauro Carvalho Chehab |
811 | Borislav Petkov |
812 | original author: Thayne Harbaugh |
813 |