blob: 8e102447aa5812b1e68dc47acf48bd54a050e66d [file] [log] [blame]
Martin Roth9b1b3352016-02-24 12:27:06 -08001 ====================
2 = MemTest-86 v4.0 =
3 = 28 Mar, 2011 =
4 = Chris Brady =
5 ====================
6Table of Contents
8 1) Introduction
9 2) Licensing
10 3) Installation
11 4) Serial Port Console
12 5) Online Commands
13 6) Memory Sizing
14 7) Error Display
15 8) Trouble-shooting Memory Errors
16 9) Execution Time
17 10) Memory Testing Philosophy
18 11) Memtest86 Test Algorithms
19 12) Individual Test Descriptions
20 13) Problem Reporting - Contact Information
21 14) Known Problems
22 15) Planned Features List
23 16) Change Log
24 17) Acknowledgments
271) Introduction
29Memtest86 is thorough, stand alone memory test for Intel/AMD x86 architecture
30systems. BIOS based memory tests are only a quick check and often miss
31failures that are detected by Memtest86.
33For updates go to the Memtest86 web page:
382) Licensing
40Memtest86 is released under the terms of the Gnu Public License (GPL). Other
41than the provisions of the GPL there are no restrictions for use, private or
42commercial. See: for details.
453) Linux Installation
47Memtest86 is a stand alone program and can be loaded from either a disk
48partition or from a floppy disk.
50To build Memtest86:
51 1) Review the Makefile and adjust options as needed.
52 2) Type "make"
54This creates a file named "memtest.bin" which is a bootable image. This
55image file may be copied to a floppy disk or may be loaded from a disk
56partition via Lilo or Grub image from a hard disk partition.
58 To create a Memtest86 bootdisk
59 1) Insert a blank write enabled floppy disk.
60 2) As root, Type "make install"
62 To boot from a disk partition via Grub
63 1) Copy the image file to a permanent location (ie. /boot/memtest.bin).
64 2) Add an entry in the Grub config file (/boot/grub/menu.lst) to boot
Martin Roth4dcd13d2016-02-24 13:53:07 -080065 memtest86. Only the title and kernel fields need to be specified.
Martin Roth9b1b3352016-02-24 12:27:06 -080066 The following is a sample Grub entry for booting memtest86:
68 title Memtest86
69 kernel (hd0,0)/memtest.bin
71 To boot from a disk partition via Lilo
72 1) Copy the image file to a permanent location (ie. /boot/memtest.bin).
73 2) Add an entry in the lilo config file (usually /etc/lilo.conf) to boot
74 memtest86. Only the image and label fields need to be specified.
75 The following is a sample Lilo entry for booting memtest86:
77 image = /boot/memtest.bin
78 label = memtest86
80 3) As root, type "lilo"
82If you encounter build problems a binary image has been included (precomp.bin).
83To create a boot-disk with this pre-built image do the following:
84 1) Insert a blank write enabled floppy disk.
85 2) Type "make install-precomp"
884) Serial Console
90Memtest86 can be used on PC's equipped with a serial port for the console.
91By default serial port console support is not enabled since it slows
92down testing. To enable change the SERIAL_CONSOLE_DEFAULT define in
93config.h from a zero to a one. The serial console baud rate may also
94be set in config.h with the SERIAL_BAUD_RATE define. The other serial
95port settings are no parity, 8 data bits, 1 stop bit. All of the features
96used by memtest86 are accessible via the serial console. However, the
97screen sometimes is garbled when the online commands are used.
1005) Online Commands
102Memtest86 has a limited number of online commands. Online commands
103provide control over caching, test selection, address range and error
104scrolling. A help bar is displayed at the bottom of the screen listing
Martin Roth4dcd13d2016-02-24 13:53:07 -0800105the available on-line commands.
Martin Roth9b1b3352016-02-24 12:27:06 -0800106
107 Command Description
109 ESC Exits the test and does a warm restart via the BIOS.
111 c Enters test configuration menu
112 Menu options are:
113 1) Test selection
114 2) Address Range
115 3) Error Report Mode
116 4) CPU Selection Mode
Martin Roth4dcd13d2016-02-24 13:53:07 -0800117 5) Refresh Screen
Martin Roth9b1b3352016-02-24 12:27:06 -0800118
119 SP Set scroll lock (Stops scrolling of error messages)
120 Note: Testing is stalled when the scroll lock is
121 set and the scroll region is full.
123 CR Clear scroll lock (Enables error message scrolling)
1266) Error Information
128Memtest has three options for reporting errors. The default is an an error
129summary that displays the most relevant error information. The second option
130is reporting of individual errors. In BadRAM Patterns mode patterns are
131created for use with the Linux BadRAM feature. This slick feature allows
132Linux to avoid bad memory pages. Details about the BadRAM feature can be
133found at:
137The error summary mode displays the following information:
139 Error Confidence Value:
140 A value that indicates the validity of the errors being reported with
141 larger values indicating greater validity. There is a high probability
142 that all errors reported are valid regardless of this value. However,
143 when this value exceeds 100 it is nearly impossible that the reported
144 errors will be invalid.
146 Lowest Error Address:
147 The lowest address that where an error has been reported.
149 Highest Error Address:
150 The highest address that where an error has been reported.
152 Bits in Error Mask:
153 A mask of all bits that have been in error (hexadecimal).
155 Bits in Error:
156 Total bit in error for all error instances and the min, max and average
157 bit in error of each individual occurrence.
159 Max Contiguous Errors:
160 The maximum of contiguous addresses with errors.
162 ECC Correctable Errors:
163 The number of errors that have been corrected by ECC hardware.
165 Test Errors:
166 On the right hand side of the screen the number of errors for each test
167 are displayed.
169For individual errors the following information is displayed when a memory
170error is detected. An error message is only displayed for errors with a
171different address or failing bit pattern. All displayed values are in
174 Tst: Test number
Martin Roth4dcd13d2016-02-24 13:53:07 -0800175 Failing Address: Failing memory address
176 Good: Expected data pattern
177 Bad: Failing data pattern
Martin Roth9b1b3352016-02-24 12:27:06 -0800178 Err-Bits: Exclusive or of good and bad data (this shows the
179 position of the failing bit(s))
180 Count: Number of consecutive errors with the same address
181 and failing bits
182 CPU: CPU that detected the error
184In BadRAM Patterns mode, Lines are printed in a form badram=F1,M1,F2,M2.
185In each F/M pair, the F represents a fault address, and the corresponding M
186is a bitmask for that address. These patterns state that faults have
187occurred in addresses that equal F on all "1" bits in M. Such a pattern may
188capture more errors that actually exist, but at least all the errors are
189captured. These patterns have been designed to capture regular patterns of
190errors caused by the hardware structure in a terse syntax.
192The BadRAM patterns are `grown' increment-ally rather than `designed' from an
193overview of all errors. The number of pairs is constrained to five for a
194number of practical reasons. As a result, handcrafting patterns from the
195output in address printing mode may, in exceptional cases, yield better
1997) Trouble-shooting Memory Errors
201Please be aware that not all errors reported by Memtest86 are due to
202bad memory. The test implicitly tests the CPU, L1 and L2 caches as well as
203the motherboard. It is impossible for the test to determine what causes
204the failure to occur. Most failures will be due to a problem with memory.
205When it is not, the only option is to replace parts until the failure is
Martin Roth4dcd13d2016-02-24 13:53:07 -0800206corrected.
Martin Roth9b1b3352016-02-24 12:27:06 -0800207
208Once a memory error has been detected, determining the failing
209module is not a clear cut procedure. With the large number of motherboard
210vendors and possible combinations of simm slots it would be difficult if
211not impossible to assemble complete information about how a particular
212error would map to a failing memory module. However, there are steps
213that may be taken to determine the failing module. Here are three
214techniques that you may wish to use:
2161) Removing modules
217This is simplest method for isolating a failing modules, but may only be
218employed when one or more modules can be removed from the system. By
219selectively removing modules from the system and then running the test
220you will be able to find the bad module(s). Be sure to note exactly which
221modules are in the system when the test passes and when the test fails.
2232) Rotating modules
224When none of the modules can be removed then you may wish to rotate modules
225to find the failing one. This technique can only be used if there are
226three or more modules in the system. Change the location of two modules
227at a time. For example put the module from slot 1 into slot 2 and put
228the module from slot 2 in slot 1. Run the test and if either the failing
229bit or address changes then you know that the failing module is one of the
230ones just moved. By using several combinations of module movement you
231should be able to determine which module is failing.
2333) Replacing modules
234If you are unable to use either of the previous techniques then you are
Martin Roth4dcd13d2016-02-24 13:53:07 -0800235left to selective replacement of modules to find the failure.
Martin Roth9b1b3352016-02-24 12:27:06 -0800236
2374) Avoiding allocation
238The printing mode for BadRAM patterns is intended to construct boot time
239parameters for a Linux kernel that is compiled with BadRAM support. This
240work-around makes it possible for Linux to reliably run on defective
241RAM. For more information on BadRAM support
242for Linux, sail to
246Sometimes memory errors show up due to component incompatibility. A memory
247module may work fine in one system and not in another. This is not
248uncommon and is a source of confusion. The components are not necessarily
249bad but certain combinations may need to be avoided.
251I am often asked about the reliability of errors reported by Mestest86.
252In the vast majority of cases errors reported by the test are valid.
253There are some systems that cause Memtest86 to be confused about the size of
254memory and it will try to test non-existent memory. This will cause a large
255number of consecutive addresses to be reported as bad and generally there
256will be many bits in error. If you have a relatively small number of
257failing addresses and only one or two bits in error you can be certain
258that the errors are valid. Also intermittent errors are always valid.
260All valid memory errors should be corrected. It is possible that a
261particular error will never show up in normal operation. However, operating
262with marginal memory is risky and can result in data loss and even
263disk corruption. You can be sure that Murphy will get you if you know
264about a memory error and ignore it.
266Memtest86 can not diagnose many types of PC failures. For example a
267faulty CPU that causes Windows to crash will most likely just cause
268Memtest86 to crash in the same way.
2718) Execution Time
273The time required for a complete pass of Memtest86 will vary greatly
Martin Roth4dcd13d2016-02-24 13:53:07 -0800274depending on CPU speed, memory speed and memory size. Memtest86 executes
275indefinitely. The pass counter increments each time that all of the
276selected tests have been run. Generally a single pass is sufficient to
277catch all but the most obscure errors. However, for complete confidence
Martin Roth9b1b3352016-02-24 12:27:06 -0800278when intermittent errors are suspected testing for a longer period is advised.
2809) Memory Testing Philosophy
282There are many good approaches for testing memory. However, many tests
283simply throw some patterns at memory without much thought or knowledge
284of memory architecture or how errors can best be detected. This
285works fine for hard memory failures but does little to find intermittent
286errors. BIOS based memory tests are useless for finding intermittent
287memory errors.
289Memory chips consist of a large array of tightly packed memory cells,
290one for each bit of data. The vast majority of the intermittent failures
291are a result of interaction between these memory cells. Often writing a
292memory cell can cause one of the adjacent cells to be written with the
293same data. An effective memory test attempts to test for this
294condition. Therefore, an ideal strategy for testing memory would be
295the following:
297 1) write a cell with a zero
298 2) write all of the adjacent cells with a one, one or more times
299 3) check that the first cell still has a zero
301It should be obvious that this strategy requires an exact knowledge
302of how the memory cells are laid out on the chip. In addition there is a
303never ending number of possible chip layouts for different chip types
304and manufacturers making this strategy impractical. However, there
Martin Roth4dcd13d2016-02-24 13:53:07 -0800305are testing algorithms that can approximate this ideal strategy.
Martin Roth9b1b3352016-02-24 12:27:06 -0800306
30811) Memtest86 Test Algorithms
310Memtest86 uses two algorithms that provide a reasonable approximation
311of the ideal test strategy above. The first of these strategies is called
312moving inversions. The moving inversion test works as follows:
314 1) Fill memory with a pattern
315 2) Starting at the lowest address
316 2a check that the pattern has not changed
317 2b write the patterns complement
318 2c increment the address
319 repeat 2a - 2c
320 3) Starting at the highest address
321 3a check that the pattern has not changed
322 3b write the patterns complement
323 3c decrement the address
324 repeat 3a - 3c
326This algorithm is a good approximation of an ideal memory test but
327there are some limitations. Most high density chips today store data
3284 to 16 bits wide. With chips that are more than one bit wide it
329is impossible to selectively read or write just one bit. This means
330that we cannot guarantee that all adjacent cells have been tested
331for interaction. In this case the best we can do is to use some
332patterns to insure that all adjacent cells have at least been written
333with all possible one and zero combinations.
335It can also be seen that caching, buffering and out of order execution
336will interfere with the moving inversions algorithm and make less effective.
337It is possible to turn off cache but the memory buffering in new high
338performance chips can not be disabled. To address this limitation a new
339algorithm I call Modulo-X was created. This algorithm is not affected by
340cache or buffering. The algorithm works as follows:
341 1) For starting offsets of 0 - 20 do
342 1a write every 20th location with a pattern
343 1b write all other locations with the patterns complement
344 repeat 1b one or more times
345 1c check every 20th location for the pattern
347This algorithm accomplishes nearly the same level of adjacency testing
348as moving inversions but is not affected by caching or buffering. Since
349separate write passes (1a, 1b) and the read pass (1c) are done for all of
350memory we can be assured that all of the buffers and cache have been
351flushed between passes. The selection of 20 as the stride size was somewhat
352arbitrary. Larger strides may be more effective but would take longer to
353execute. The choice of 20 seemed to be a reasonable compromise between
354speed and thoroughness.
35711) Individual Test Descriptions
359Memtest86 executes a series of numbered test sections to check for
360errors. These test sections consist of a combination of test
361algorithm, data pattern and caching. The execution order for these tests
362were arranged so that errors will be detected as rapidly as possible.
363A description of each of the test sections follows:
365Test 0 [Address test, walking ones, no cache]
366 Tests all address bits in all memory banks by using a walking ones
367 address pattern. Errors from this test are not used to calculate
368 BadRAM patterns.
370Test 1 [Address test, own address Sequential]
371 Each address is written with its own address and then is checked
372 for consistency. In theory previous tests should have caught any
373 memory addressing problems. This test should catch any addressing
374 errors that somehow were not previously detected. This test is done
375 sequentially with each available CPU.
377Test 2 [Address test, own address Parallel]
378 Same as test 1 but the testing is done in parallel using all CPUs using
Martin Roth4dcd13d2016-02-24 13:53:07 -0800379 overlapping addresses.
Martin Roth9b1b3352016-02-24 12:27:06 -0800380
381Test 3 [Moving inversions, ones&zeros Sequential]
382 This test uses the moving inversions algorithm with patterns of all
383 ones and zeros. Cache is enabled even though it interferes to some
384 degree with the test algorithm. With cache enabled this test does not
385 take long and should quickly find all "hard" errors and some more
386 subtle errors. This test is done sequentially with each available CPU.
388Test 4 [Moving inversions, ones&zeros Parallel]
389 Same as test 3 but the testing is done in parallel using all CPUs.
Martin Roth4dcd13d2016-02-24 13:53:07 -0800390
Martin Roth9b1b3352016-02-24 12:27:06 -0800391Test 5 [Moving inversions, 8 bit pat]
392 This is the same as test 4 but uses a 8 bit wide pattern of
393 "walking" ones and zeros. This test will better detect subtle errors
394 in "wide" memory chips. A total of 20 data patterns are used.
396Test 6 [Moving inversions, random pattern]
397 Test 6 uses the same algorithm as test 4 but the data pattern is a
398 random number and it's complement. This test is particularly effective
Martin Roth4dcd13d2016-02-24 13:53:07 -0800399 in finding difficult to detect data sensitive errors.
400 The random number sequence is different with each pass
401 so multiple passes increase effectiveness.
Martin Roth9b1b3352016-02-24 12:27:06 -0800403Test 7 [Block move, 64 moves]
404 This test stresses memory by using block move (movsl) instructions
405 and is based on Robert Redelmeier's burnBX test. Memory is initialized
406 with shifting patterns that are inverted every 8 bytes. Then 4MB blocks
407 of memory are moved around using the movsl instruction. After the moves
408 are completed the data patterns are checked. Because the data is checked
409 only after the memory moves are completed it is not possible to know
410 where the error occurred. The addresses reported are only for where the
411 bad pattern was found. Since the moves are constrained to a 8MB segment
412 of memory the failing address will always be lest than 8MB away from the
413 reported address. Errors from this test are not used to calculate
414 BadRAM patterns.
416Test 8 [Moving inversions, 32 bit pat]
417 This is a variation of the moving inversions algorithm that shifts the data
418 pattern left one bit for each successive address. The starting bit position
419 is shifted left for each pass. To use all possible data patterns 32 passes
420 are required. This test is quite effective at detecting data sensitive
421 errors but the execution time is long.
Martin Roth4dcd13d2016-02-24 13:53:07 -0800422
Martin Roth9b1b3352016-02-24 12:27:06 -0800423Test 9 [Random number sequence]
424 This test writes a series of random numbers into memory. By resetting the
425 seed for the random number the same sequence of number can be created for
426 a reference. The initial pattern is checked and then complemented and
427 checked again on the next pass. However, unlike the moving inversions test
428 writing and checking can only be done in the forward direction.
430Test 10 [Modulo 20, random pattern]
431 Using the Modulo-X algorithm should uncover errors that are not
432 detected by moving inversions due to cache and buffering interference
433 with the the algorithm. A 32 bit random pattern is used.
435Test 11 [Bit fade test, 2 patterns]
436 The bit fade test initializes all of memory with a pattern and then
437 sleeps for 5 minutes. Then memory is examined to see if any memory bits
438 have changed. All ones and all zero patterns are used.
44112) Problem Reporting - Contact Information
443Due to the growing popularity of Memtest86 (more than 200,000 downloads per
444month) I have been inundated by, questions, feedback, problem reports and
445requests for enhancements. I simply do not have time to respond to ANY Memtest86
446emails. Bug reports and suggestions are welcome but will typically not be
447responded to.
449*** NOTE: *** The Keyword MEM86 must appear in the subject of all emails or
450the message will be automaticly deleted before it gets to me. This thanks to
451spam and viruses!
454Before submitting a problem report please check the Known Problems section
455to see if this problem has already been reported. Be sure to include the
456version number and also any details that may be relevant.
458Chris Brady, Email:
460With some PC's Memtest86 will just die with no hints as to what went wrong.
461Without any details it is impossible to fix these failures. Fixing these
462problems will require debugging on your part. There is no point in reporting
463these failures unless you have a Linux system and would be willing to debug
464the failure.
467If you would like to request an enhancement please see if is already on
468the Planned Features List before sending your request. All requests will
469be considered, but not all can be implemented. If you are be interested in
470contributing code please contact me so that the integration can be
473Chris Brady, Email:
476Unfortunately, I do not have time to respond to any questions or provide
477assistance with troubleshooting problems. Please read the Troubleshooting
478and Known Problems sections for assistance with problems. These sections have
479the answers for the questions that I have answers to. If there is not an
480answer for your problem in these sections it is probably not something I can
481help you with.
48415) Known Problems
486Sometimes when booting from a floppy disk the following messages scroll up
487on the screen:
488 X:8000
489 AX:0212
490 BX:8600
491 CX:0201
492 DX:0000
493This the BIOS reporting floppy disk read errors. Either re-write or toss
494the floppy disk.
496Memtest86 can not diagnose many types of PC failures. For example a
497faulty CPU that causes Windows to crash will most likely just cause
498Memtest86 to crash in the same way.
500There have been numerous reports of errors in only the block move test. Often
501the memory works in a different system or the vendor insists that it is good.
502In these cases the memory is not necessarily bad but is not able to operate
503reliably high speeds. Sometimes more conservative memory timings on the
504motherboard will correct these errors. In other cases the only option is to
505replace the memory with better quality, higher speed memory. Don't buy cheap
506memory and expect it to work at full speed.
508Memtest86 supports all types of memory. If fact the test has absolutely
509no knowledge of the memory type nor does it need to. This not a problem
510or bug but is listed here due to the many questions I get about this issue.
512Changes in the compiler and loader have caused problems with
513Memtest86 resulting in both build failures and errors in execution. A
514binary image (precomp.bin) of the test is included and may be used if
515problems are encountered.
51815) Planned Features List
520This is a list of enhancements planned for future releases of Memtest86.
521There is no timetable for when these will be implemented.
523 - Testing in 64 bit mode with 64 data patterns
524 - Support for reporting ECC errors was removed in the 4.0 release. A
525 simplified implementation of ECC reporting is planned for a future release.
52816) Change Log
530Enhancements in v4.0 (28/Mar/2011)
532 Full support for testing with multiple CPUs. All tests except for #11 (Bit
533 Fade) have been multi-threaded. A maximum of 16 CPUs will be used for testing.
535 CPU detection has been completely re-written to use the brand ID string
536 rather than the cumbersome, difficult to maintain and often out of date
537 CPUID family information. All new processors will now be correctly
Martin Roth4dcd13d2016-02-24 13:53:07 -0800538 identified without requiring code support.
Martin Roth9b1b3352016-02-24 12:27:06 -0800539
540 All code related to controller identification, PCI and DMI has been removed.
541 This may be a controversial decision and was not made lightly. The following
542 are justifications for the decision:
544 1. Controller identification has nothing to do with actual testing of
Martin Roth4dcd13d2016-02-24 13:53:07 -0800545 memory, the core purpose of Memtest86.
Martin Roth9b1b3352016-02-24 12:27:06 -0800546 2. This code needed to be updated with every new chipset. With the ever
547 growing number of chipsets it is not possible to keep up with the
548 changes. The result is that new chipsets were more often than not
549 reported in-correctly. In the authors opinion incorrect information is
550 worse than no information.
551 3. Probing for chipset information carries the risk of making the program
Martin Roth4dcd13d2016-02-24 13:53:07 -0800552 crash.
Martin Roth9b1b3352016-02-24 12:27:06 -0800553 4. The amount of code involved with controller identification was quite
554 large, making support more difficult.
556 Removing this code also had the unfortunate effect of removing reporting of
557 correctable ECC errors. The code to support ECC was hopelessly intertwined
558 the controller identification code. A fresh, streamlined implementation of
559 ECC reporting is planned for a future release.
Martin Roth4dcd13d2016-02-24 13:53:07 -0800560
Martin Roth9b1b3352016-02-24 12:27:06 -0800561 A surprising number of conditions existed that potentially cause problems
562 when testing more than 4 GB of memory. Most if not all of these conditions
563 have been identified and corrected.
565 A number of cases were corrected where not all of memory was being tested.
566 For most tests the last word of each test block was not tested. In addition
567 an error in the paging code was fixed that omitted from testing the last 256
568 bytes of each block above 2 GB.
570 The information display has been simplified and a number of details that were
571 not relevant to testing were removed.
573 Memory speed reporting has been parallelized for more accurate reporting for
574 multi channel memory controllers.
576 This is a major re-write of the Memtest86 with a large number of minor
Martin Roth4dcd13d2016-02-24 13:53:07 -0800577 bug-fixes and substantial cleanup and re-organization of the code.
Martin Roth9b1b3352016-02-24 12:27:06 -0800578
580Enhancements in v3.5 (3/Jan/2008)
582 Limited support for execution with multiple CPUs. CPUs are selected
583 round-robin or sequential for each test.
585 Support for additional chipsets. (from Memtest86+ v2.11).
587 Additions and corrections for CPU detection including reporting of L3 cache.
589 Reworked information display for better readability and new information.
591 Abbreviated iterations for first pass.
593 Enhancements to memory sizing.
595 Misc fixes.
597Enhancements in v3.4 (2/Aug/2007)
599 A new error summary display with error confidence analysis.
601 Support for additional chipsets. (from Memtest86+ v1.70).
603 Additions and corrections for CPU detection.
605 Support for memory module information reporting.
607 Misc bug fixes.
609Enhancements in v3.3 (12/Jan/2007)
611 Added support for additional chipsets. (from Memtest86+ v1.60)
613 Changed Modulo 20 test (#8) to use a more effective random pattern rather
614 than simple ones and zeros.
616 Fixed a bug that prevented testing of low memory.
618 Added an advanced menu option to display SPD info (only for selected
619 chipsets).
621 Updated CPU detection for new CPUs and corrected some bugs.
623 Reworked online command text for better clarity.
625 Added a fix to correct a Badram pattern bug.
628Enhancements in v3.2 (11/Nov/2004)
630 Added two new, highly effective tests that use random number patterns
631 (tests 4 and 6).
633 Reworked the online commands:
634 - Changed wording for better clarity
635 - Dropped Cache Mode menu
637 Updated CPU detection for newer AMD, Intel and Cyrix CPUs.
639 Reworked test sequence:
Martin Roth4dcd13d2016-02-24 13:53:07 -0800640 - Dropped ineffective non cached tests (Numbers 7-11)
641 - Changed cache mode to "cached" for test 2
Martin Roth9b1b3352016-02-24 12:27:06 -0800642 - Fixed bug that did not allow some tests to be skipped
643 - Added bailout for Bit fade test
Martin Roth4dcd13d2016-02-24 13:53:07 -0800645 Error reports are highlighted in red to provide a more vivid error
Martin Roth9b1b3352016-02-24 12:27:06 -0800646 indication.
Martin Roth4dcd13d2016-02-24 13:53:07 -0800647
Martin Roth9b1b3352016-02-24 12:27:06 -0800648 Added support for a large number of additional chipsets. (from Memtest86+
649 v1.30)
651 Added an advanced setup feature that with new chiset allows memory timings
652 to be altered from inside Memtest86. (from Memtest86+ v1.30)
655Enhancements in v3.1 (11/Mar/2004)
657 Added processor detection for newer AMD processors.
659 Added new "Bit Fade" extended test.
661 Fixed a compile time bug with gcc version 3.x.
663 E7500 memory controller ECC support
665 Added support for 16bit ECC syndromes
667 Option to keep the serial port baud rate of the boot loader
670Enhancements in v3.0 (22/May/2002) Provided by Eric Biederman
672 Testing of more than 2gb of memory is at last fixed (tested with 6Gb)
674 The infrastructure is to poll ecc error reporting chipset regisets,
675 and the support has been done for some chipsets.
677 Uses dynamic relocation information records to make itself PIC
678 instead of requiring 2 copies of memtest86 in the binary.
680 The serial console code does not do redundant writes to the serial port
681 Very little slow down at 9600 baud.
683 You can press ^l or just l to get a screen refresh, when you are
684 connecting and UN-connecting a serial cable.
686 Net-booting is working again
Martin Roth61ea5412016-03-07 20:58:18 -0700688 Linux-BIOS (coreboot) support (To get the memory size)
Martin Roth9b1b3352016-02-24 12:27:06 -0800689
690 Many bug-fixes and code cleanup.
692Enhancements in v2.9 (29/Feb/2002)
694 The memory sizing code has been completely rewritten. By default
Martin Roth4dcd13d2016-02-24 13:53:07 -0800695 Memtest86 gets a memory map from the BIOS that is now used to find
Martin Roth9b1b3352016-02-24 12:27:06 -0800696 available memory. A new online configuration option provides three
697 choices for how memory will be sized, including the old "probe" method.
698 The default mode generally will not test all of memory, but should be more
699 stable. See the "Memory Sizing" section for details.
701 Testing of more than 2gb of memory should now work. A number of bugs
702 were found and corrected that prevented testing above 2gb. Testing
703 with more than 2gb has been limited and there could be problems with a
704 full 4gb of memory.
706 Memory is divided into segments for testing. This allow for frequent
707 progress updates and responsiveness to interactive commands. The
708 memory segment size has been increased from 8 to 32mb. This should
709 improve testing effectiveness but progress reports will be less frequent.
711 Minor bug fixes.
713Enhancements in v2.8 (18/Oct/2001)
714 Eric Biederman reworked the build process making it far simpler and also
715 to produce a network bootable ELF image.
717 Re-wrote the memory and cache speed detection code. Previously the
718 reported numbers were inaccurate for Intel CPU's and completely wrong
719 for Athlon/Duron CPU's.
721 By default the serial console is disabled since this was slowing
722 down testing.
724 Added CPU detection for Pentium 4.
Martin Roth4dcd13d2016-02-24 13:53:07 -0800726
Martin Roth9b1b3352016-02-24 12:27:06 -0800727Enhancements in v2.7 (12/Jul/2001)
728 Expanded workaround for errors caused by BIOS USB keyboard support to
729 include test #5.
731 Re-worked L1 / L2 cache detection code to provide clearer reporting.
733 Fixed an obvious bug in the computation of cache and memory speeds.
735 Changed on-line menu to stay in the menu between option selections.
737 Fixed bugs in the test restart and redraw code.
739 Adjusted code size to fix compilation problems with RedHat 7.1.
741 Misc updates to the documentation.
743Enhancements in v2.6 (25/May/2001)
744 Added workaround for errors caused by BIOS USB keyboard support.
746 Fixed problems with reporting of 1 GHZ + processor speeds.
748 Fixed Duron cache detection.
750 Added screen buffer so that menus will work correctly from a serial
751 console.
753 The Memtest86 image is now built in ELF format.
755Enhancements in v2.5 (14/Dec/00)
756 Enhanced CPU and cache detection to correctly identify Duron CPU
757 and K6-III 1MB cache.
759 Added code to report cache-able memory size.
761 Added limited support for parity memory.
763 Support was added to allow use of on-line commands from a serial
764 port.
766 Dropped option for changing refresh rates. This was not useful
767 and did not work on newer motherboards.
769 Improved fatal exception reporting to include a register and stack
770 dump.
772 The pass number is now displayed in the error report.
774 Fixed a bug that crashed the test when selecting one of the extended
775 tests.
777Enhancements in v2.4
778 The error report format was reworked for better clarity and now
779 includes a decimal address in megabytes.
781 A new memory move test was added (from Robert Redelmeier's CPU-Burn)
783 The test sequence and iterations were modified.
785 Fixed scrolling problems with the BadRAM patterns.
788Enhancements in v2.3
789 A progress meter was added to replace the spinner and dots.
Martin Roth4dcd13d2016-02-24 13:53:07 -0800791 Measurement and reporting of memory and cache performance
Martin Roth9b1b3352016-02-24 12:27:06 -0800792 was added.
794 Support for creating BadRAM patterns was added.
796 All of the test routines were rewritten in assembler to
797 improve both test performance and speed.
799 The screen layout was reworked to hopefully be more readable.
801 An error summary option was added to the online commands.
804Enhancements in v2.2
805 Added two new address tests
807 Added an on-line command for setting test address range
809 Optimized test code for faster execution (-O3, -funroll-loops and
810 -fomit-frame-pointer)
812 Added and elapsed time counter.
814 Adjusted menu options for better consistency
817Enhancements in v2.1
818 Fixed a bug in the CPU detection that caused the test to
819 hang or crash with some 486 and Cryrix CPU's
821 Added CPU detection for Cyrix CPU's
823 Extended and improved CPU detection for Intel and AMD CPU's
825 Added a compile time option (BIOS_MEMSZ) for obtaining the last
826 memory address from the BIOS. This should fix problems with memory
827 sizing on certain motherboards. This option is not enabled by default.
828 It may be enabled be default in a future release.
830Enhancements in v2.0
831 Added new Modulo-20 test algorithm.
833 Added a 32 bit shifting pattern to the moving inversions algorithm.
835 Created test sections to specify algorithm, pattern and caching.
837 Improved test progress indicators.
839 Created popup menus for configuration.
841 Added menu for test selection.
843 Added CPU and cache identification.
845 Added a "bail out" feature to quit the current test when it does not
846 fit the test selection parameters.
848 Re-arranged the screen layout and colors.
850 Created local include files for I/O and serial interface definitions
Martin Roth4dcd13d2016-02-24 13:53:07 -0800851 rather than using the sometimes incompatible system include files.
Martin Roth9b1b3352016-02-24 12:27:06 -0800852
853 Broke up the "C" source code into four separate source modules.
855Enhancements in v1.5
856 Some additional changes were made to fix obscure memory sizing
857 problems.
859 The 4 bit wide data pattern was increased to 8 bits since 8 bit
860 wide memory chips are becoming more common.
862 A new test algorithm was added to improve detection of data
Martin Roth4dcd13d2016-02-24 13:53:07 -0800863 pattern sensitive errors.
Martin Roth9b1b3352016-02-24 12:27:06 -0800864
866Enhancements in v1.4
867 Changes to the memory sizing code to avoid problems with some
868 motherboards where memtest would find more memory than actually
869 exists.
871 Added support for a console serial port. (thanks to Doug Sisk)
873 On-line commands are now available for configuring Memtest86 on
874 the fly (see On-line Commands).
Martin Roth4dcd13d2016-02-24 13:53:07 -0800875
Martin Roth9b1b3352016-02-24 12:27:06 -0800876
877Enhancements in v1.3
878 Scrolling of memory errors is now provided. Previously, only one screen
879 of error information was displayed.
881 Memtest86 can now be booted from any disk via lilo.
883 Testing of up to 4gb of memory has been fixed is now enabled by default.
884 This capability was clearly broken in v1.2a and should work correctly
885 now but has not been fully tested (4gb PC's are a bit rare).
887 The maximum memory size supported by the motherboard is now being
888 calculated correctly. In previous versions there were cases where not
889 all of memory would be tested and the maximum memory size supported
890 was incorrect.
892 For some types of failures the good and bad values were reported to be
893 same with an Xor value of 0. This has been fixed by retaining the data
894 read from memory and not re-reading the bad data in the error reporting
895 routine.
897 APM (advanced power management) is now disabled by Memtest86. This
898 keeps the screen from blanking while the test is running.
900 Problems with enabling & disabling cache on some motherboards have been
901 corrected.
90417) Acknowledgments
906Memtest86 was developed by Chris Brady with the resources and assistance
907listed below:
909- The initial versions of the source files bootsect.S, setup.S, head.S and
910 build.c are from the Linux 1.2.1 kernel and have been heavily modified.
912- Doug Sisk provided code to support a console connected via a serial port.
914- Code to create BadRAM patterns was provided by Rick van Rein.
916- Tests 5 and 8 are based on Robert Redelmeier's burnBX test.
918- Screen buffer code was provided by Jani Averbach.
920- Eric Biederman provided all of the feature content for version 3.0
921 plus many bugfixes and significant code cleanup.
923- Major enhancements to hardware detection and reporting in version 3.2,
924 3.3 pnd 3.4 rovided by Samuel Demeulemeester (from Memtest86+ v1.11, v1.60
925 and v1.70).