Main Page

From EdacWiki
Jump to navigation Jump to search

EDAC Wiki

This is a wiki for the Linux EDAC project

What is it?

EDAC Stands for "Error Detection and Correction". The Linux EDAC project comprises a series of Linux kernel modules, which make use of error detection facilities of computer hardware, currently hardware which detects the following errors is supported:

  • System RAM errors (this is the original, and most mature part of the project) - many computers support RAM EDAC, (especially for chipsets which are aimed at high-reliability applications), but RAM which has extra storage capacity ("ECC RAM") is needed for these facilities to operate
  • RAM scrubbing - some memory controllers support "scrubbing" DRAM during normal operation. Continuously scrubbing DRAM allows for actively detecting and correcting ECC errors.
  • PCI bus transfer errors - the majority of PCI bridges, and peripherals support such error detection
  • Cache ECC errors

Why do I need it?

Without the EDAC modules, on most current Linux systems:

  • You may be experiencing PCI data corruption (e.g. your data is being corrupted between whilst travelling to/from your NIC/storage adapter, whilst on the PCI bus), and not know about it, as most systems do not check PCI devices for reported PCI parity errors (some may trigger an NMI, but you have no more info about what caused the NMI).
  • If you have ECC memory, and you are experiencing correctable ECC errors, you probably won't know anything about it. With the EDAC modules installed on your system, you will get to know about bad memory modules before the errors become uncorrectable, and you have potentially corrupted data (and a crashed machine) - this includes finding out about memory modules which are bad as-shipped, before such systems are put into service (saving you time and hassle).
  • If you have a motherboard which claims to support ECC, but the BIOS is not correctly enabling ECC mode, you won't know anything about it (until your machine crashes with unexplained memory errors - you won't even get an NMI, and the extra money spent on ECC memory will be wasted).

How to turn it on

  • PCI error checking can be enabled with:
dougal:~# modprobe edac_mc
dougal:~# cd /sys/devices/system/edac/pci/
dougal:/sys/devices/system/edac/pci# cat check_pci_parity
0
dougal:/sys/devices/system/edac/pci# echo 1 > check_pci_parity
dougal:/sys/devices/system/edac/pci# cat pci_parity_count
1
dougal:/sys/devices/system/edac/pci# dmesg | tail -4
usb0: rxqlen 0 --> 4
usb0: no IPv6 routers present
EDAC MC: Ver: 2.0.1 May  9 2007
EDAC PCI: Detected Parity Error on 0000:00:09.0
dougal:/sys/devices/system/edac/pci# lspci -s 0000:00:09.0
00:09.0 Multimedia audio controller: Yamaha Corporation YMF-744B [DS-1S Audio Controller] (rev 02)
dougal:/sys/devices/system/edac/pci# arecord > /dev/null
Recording WAVE 'stdin' : Unsigned 8 bit, Rate 8000 Hz, Mono
Aborted by signal Interrupt...
dougal:/sys/devices/system/edac/pci# cat pci_parity_count
15
dougal:/sys/devices/system/edac/pci# dmesg | tail -4
EDAC PCI: Detected Parity Error on 0000:00:09.0
EDAC PCI: Detected Parity Error on 0000:00:09.0
EDAC PCI: Detected Parity Error on 0000:00:09.0
EDAC PCI: Detected Parity Error on 0000:00:09.0

Oh dear, my laptop sound device seems to be broken! At the moment the PCI checking is built into the edac_mc (memory controller) kernel module, in time this will be split out... As you can see from the above, PCI error checking is turned off by default, and needs to be turned on (using the "echo" statement above).

Help!

About the Errors that EDAC generates

If the EDAC subsystem is reporting errors on your system, please see WhyAmIgettingMemoryErrors, and WhyAmIgettingPciErrors. Please try and check out the possibilities listed here, and elsewhere on this wiki, before you either open a new bug report, or post to the mailing list.

The EDAC Bug Database

If you think you've found a bug, please search the EDAC Bugzilla to see if it has already been reported (you can then add yourself to the cc list for that bug, so that you are automatically informed of updates etc.), if it hasn't, then please create a new bug report.

In-kernel documentation

There is some documentation in the kernel in Documentation/edac.txt .

The EDAC Mailing List

Most of the EDAC developers keep an eye on the EDAC mailing list (hosted by Sourceforge) to a greater or lesser extent, but please remember that not many of them work on EDAC as part of their job, (and if they do, then they are paid to keep their employer's systems running), so check the Wiki, the bug database, and the mailing list archives (for both the current and the previous mailing lists) for your problem first.

If you have exhausted these possibilities, then by all means post to the mailing list...

  • Be polite
  • Please make sure you give all information which might be relevant e.g. your (exact) kernel version
  • Be patient

If you get a reply, or find things out which weren't known about before, please add the information to this Wiki, in order to help others.

Userspace Tools

There are userspace tools in development at http://sourceforge.net/projects/edac-utils

The userspace needs some help, please get involved and help out!

Status

The EDAC code is in Linux Kernel version 2.6.16. There is a userspace API (via sysfs) in 2.6.18 and above.

Getting the code

If you want a more recent version than the version in your current kernel, you can download a quilt stack from the sourceforge download page (see below), or by anonymous SVN checkout: 

$ cd mydev-dir
$ svn checkout https://bluesmoke.svn.sourceforge.net/svnroot/bluesmoke/trunk edac-trunk/
$ less bluesmoke/edac/patches/README

Prior to May 2007, things can be found in CVS. See the sourceforge main page for CVS information.

You will need a recent Linux kernel tree to apply the patches to. Or, if you just want to have a look at the recent changes, you can browse the SVN at:

[1]

History

The EDAC project was renamed from the "bluesmoke" prior to submission to the mainline Linux kernel. The Bluesmoke code was created by Thayne Harbaugh. The Linux-ECC project was EDAC's predecessor and its major inspiration. Developed by Dan Hollis and others, the Linux-ECC project is no longer maintained.

Supported Hardware

System Main Memory EDAC

Supported Memory Controllers

Please see the individual driver pages for information on supported revisions, motherboard-specific information etc.

Manufacturer Model EDAC Driver Tech Docs Controller Capabilities Status
AMCC 4xx ppc4xx_edac.c Supported (Linux 2.6.30)
AMD Opteron amd64_edac.c AMD EDAC, ErrorScrub, BackgroundScrub Supported Development Tree
AMD Athlon64 amd64_edac.c AMD EDAC, ErrorScrub, BackgroundScrub Supported Development Tree
AMD AthlonFX amd64_edac.c AMD EDAC, ErrorScrub, BackgroundScrub Supported Development Tree
AMD 760 amd76x_edac.c AMD Supported (Linux 2.6.16)
AMD 762 amd76x_edac.c AMD Supported (Linux 2.6.16)
AMD 768 amd76x_edac.c AMD Supported (Linux 2.6.16)
AMD 8111 amd8111_edac.c Supported (Linux 2.6.30)
Freescale MPC83xx mpc85xx_edac.c Supported (Linux 2.6.32)
Freescale MPC85xx mpc85xx_edac.c Supported (Linux 2.6.25)
Freescale P2020 mpc85xx_edac.c Supported (Linux 2.6.32)
Intel e7500 e7xxx_edac.c Supported (Linux 2.6.16)
Intel e7501 e7xxx_edac.c Supported (Linux 2.6.16)
Intel e7505 e7xxx_edac.c Supported (Linux 2.6.16)
Intel e7520 e752x_edac.c Supported (Linux 2.6.16)
Intel e7525 e752x_edac.c Supported (Linux 2.6.16)
Intel i3100 e752x_edac.c Supported (Linux 2.6.26)
Intel i3200 and i3210 i3200_edac.c Supported (Linux 2.6.??)
Intel 82875p i82875p_edac.c EDAC Supported (Linux 2.6.16)
Intel e7210 Supported (Linux 2.6.16)
Intel 82860 i82860_edac.c Supported (Linux 2.6.16)
Intel 5000(P/V/X) i5000_edac.c Patch in CVS
Intel 82443BX/GX(440BX/GX) i82443bxgx_edac.c Intel EDAC, ErrorScrub Patch in SVN
Radisys 82600 r82600_edac.c Radisys EDAC, ErrorScrub Supported (Linux 2.6.16)
Via VT82c693/694(Pro133) Local EDAC Author Needed
Intel 3000/3010 i3000_edac.c Intel EDAC Supported (Linux 2.6.25)
Intel X38 x38_edac.c Intel EDAC Supported (Linux 2.6.28)

Customisation for your Hardware

For many chipsets and motherboards, there is no consistant relationship between the memory banks/slots as made available to the EDAC driver, and the physical labels present next to the memory module socket. You can help by working out the relationship for your hardware, and adding the info to the MemorySlotLabels page.

PCI Error Reporting

PCI Parity error reporting facilities are included in the PCI specification, and the majority of add-in cards (and chips which are capable of being included in either add-in, or on-motherboard designs) support the PCI parity error detection, and reporting functionality. Some "fake" PCI devices which are not physically connected by a PCI bus (such as e.g. some ATA host adaptors which are built-in to a motherboard chipset) typically do not include the functionality.

Error Detection Overhead

The driver currently only support error detection via polling. Polling all of the PCI devices' error status registers can be time consuming, especially on machines which have many devices. You may wish to slow the error polling rate, or disable it altogether on such systems.

Faulty Hardware

Some PCI devices (or just particular revisions of those devices) are broken with respect to PCI parity detection, and display false positives. You can check (and add to) the list of broken devices on the PCIDevicesWithBrokenParityDetection page.

Help Wanted!

We need your help:

  • Improve this documentation
  • HowToWriteNewMemoryControllerDrivers
  • HardwareWanted
  • Test the code
  • Report broken hardware for the blacklists
  • Create memory slot entries for your hardware
  • Create some user-space code (e.g. scripts to go in a cron job, extensions to SNMP daemons etc. etc.)
  • Create a script to generate dimm labels, whitelists from the WIKI contents

Other Resources

Sourceforge project page [2]

An overview of EDAC technologies on Wikipedia [3]

The original Linux ECC project (Dan Hollis et al) - [4]

A talk delivered by Tim Small at UKUUG 2006 - [5]

Mailing list [6]

How to use this site

A Wiki is a collaborative site, anyone can contribute and share:

  • Edit any page by pressing Edit at the top of the page
  • MediaWiki FAQ
Retrieved from ‘https://buttersideup.com/mediawiki/index.php?title=Main_Page&oldid=185’