memtest86plus/README.md

# PCMemTest

PCMemTest is a stand-alone memory tester for x86 and x86-64 architecture
computers. It provides a more thorough memory check than that provided by
BIOS memory tests.

PCMemTest can be loaded and run either directly by a PC BIOS (legacy or UEFI)
or via an intermediate bootloader that supports the Linux 16-bit, 32-bit,
64-bit, or EFI handover boot protocol. It should work on any Pentium class or
later 32-bit or 64-bit CPU.

## Table of Contents

  * [Origins](#origins)
  * [Licensing](#licensing)
  * [Build and Installation](#build-and-installation)
  * [Boot Options](#boot-options)
  * [Keyboard Selection](#keyboard-selection)
  * [Operation](#operation)
  * [Error Display](#error-display)
  * [Trouble-shooting Memory Errors](#trouble-shooting-memory-errors)
  * [Execution Time](#execution-time)
  * [Memory Testing Philosophy](#memory-testing-philosophy)
  * [PCMemTest Test Algorithms](#pcmemtest-test-algorithms)
  * [Individual Test Descriptions](#individual-test-descriptions)
  * [Known Limitations and Bugs](#known-limitations-and-bugs)
  * [Acknowledgments](#acknowledgments)

## Origins

PCMemTest is a fork and rewrite of Memtest86+, which in turn was a fork of
Memtest86. The purpose of the rewrite was to:

  * make the code more readable and easier to maintain
  * make the code 64-bit clean and support UEFI boot
  * fix failures seen when building with newer versions of GCC

In the process, a number of features of Memtest86+ that are not required for
the main purpose of PCMemTest (testing the system memory) have been dropped.
In particular, no attempt is made to measure the cache and main memory speed,
or to identify and report the DRAM type. This should allow PCMemTest to work
without modification on future hardware.

PCMemTest is based on the v5.01 release of Memtest86+, which was the last
public release at the time of the fork.

## Licensing

PCMemTest is released under the terms of the GNU General Public License version
2 (GPLv2). Other than the provisions of the GPL there are no restrictions for
use, private or commercial. See the LICENSE file for details.

## Build and Installation

Build is only tested on a Linux system, but should be possible on any system
using the GNU toolchain and the ELF file format. The tools required are:

  * GCC
  * binutils
  * make
  * dosfstools and mtools (optional)
  * xorrisofs (optional)

To build a 32-bit image, change directory into the `build32` directory and
type `make`. The result is a `memtest.bin` binary image file which can be
booted directly by a legacy BIOS (in floppy mode) or by an intermediate
bootloader using the Linux 16-bit boot protocol and a `memtest.efi` binary
image file which can be booted directly by a 32-bit UEFI BIOS. Either image
can be booted by an intermediate bootloader using the Linux 32-bit or 32-bit
EFI handover boot protocols. 

To build a 64-bit image, change directory into the `build64` directory and
type `make`. The result is a `memtest.bin` binary image file which can be
booted directly by a legacy BIOS (in floppy mode) or by an intermediate
bootloader using the Linux 16-bit boot protocol and a `memtest.efi` binary
image file which can be booted directly by a 64-bit UEFI BIOS. Either image
can be booted by an intermediate bootloader using the Linux 32-bit, 64-bit,
or 64-bit EFI handover boot protocols.

In either case, to build an ISO image that can be used to create a bootable
CD, DVD, or USB Flash drive, type `make iso`, The result is a `memtest.iso`
ISO image file. This can then be written directly to a blank CD or DVD, or
to a USB Flash drive, which can then be booted directly by a legacy or UEFI
PC BIOS.

Note that when writing to a USB Flash drive, the ISO image must be written
directly ('dumped') to the raw device, either by using the `dd` command or
by using a utility that provides the same functionality.

When using an intermediate bootloader, either the `memtest.bin` file or the
`memtest.efi` file should be stored in a disk partition the bootloader can
access, and the bootloader configuration should be updated to boot from
that file as if it were a Linux kernel with no initial RAM disk. Several
boot command line options are recognised, as described below. If using the
16-bit boot protocol, PCMemTest will use the display in text mode (640x400).
If using the 32-bit or 64-bit boot protocols, PCMemTest will use the display
in either text mode or graphics mode, as specified in the `boot_params` struct
passed to it by the bootloader. If in graphics mode, the supplied framebuffer
must be at least 640x400 pixels; if larger, the display will be centred. If
the system was booted in UEFI mode, graphics mode must be used.

For test purposes, there is also an option to build an ISO image that uses
GRUB as an intermediate bootloader. See the `Makefile` in the `build32` or
`build64` directory for details. The ISO image is both legacy and UEFI
bootable, so you need GRUB modules for both legacy and EFI boot installed
on your build system. You may need to adjust some path and file names in
the make file to match the naming on your system.

The GRUB configuration files contained in the `grub` directory are there for
use on the test ISO, but also serve as an example of how to boot PCMemTest
from GRUB.

## Boot Options

An intermediate bootloader may pass a boot command line to PCMemTest. The
command line may contain one or more options, separated by spaces. Each
option consists of an option name, optionally followed by an `=` sign and
one or more parameters, separated by commas. The following options are
recognised:

  * smp
    * enables the use of multiple CPU cores at startup
  * nopause
    * skips the pause for configuration at startup
  * keyboard=*type*
    * where *type* is one of
      * legacy
      * usb
      * buggy-usb

## Keyboard Selection

PCMemTest supports both the legacy keyboard interface (using I/O ports 0x60
and 0x64) and USB keyboards (using its own USB device drivers). One or the
other can be selected via the boot command line, If neither is selected, the
default is to use both. An additional option on the boot command line is
`buggy-usb` which enables the longer initialisation sequence required by some
older USB devices.

Older BIOSs usually support USB legacy keyboard emulation, which makes USB
keyboards act like legacy keyboards connected to ports 0x60 and 0x64. This
can often be enabled or disabled in the BIOS setup menus. If PCMemTest's
USB device drivers are enabled, they will override this and access any USB
keyboards directly. The downside of that is that the USB controllers and
device drivers require some memory to be reserved for their private use,
which means that memory can't then be covered by the memory tests. So to
maximise test coverage, if it is supported, enable USB legacy keyboard
emulation and add `keyboard=legacy` on the boot command line.

**NOTE**: Some UEFI BIOSs only support USB legacy keyboard emulation when
you enable the Compatibility System Module (CSM) in the BIOS setup. Others
only support it when actually booting in legacy mode.

**NOTE**: PCMemTest's USB device drivers are work in progress. Not all USB
devices are supported yet.

## Operation

Once booted, PCMemTest will initialise its display, then pause for a few
seconds to allow the user to configure its operation. If no key is pressed,
it will automatically start running all tests using a single CPU core,
continuing indefinitely until the user reboots or halts the machine.

At startup, and when running tests, PCMemTest responds to the following keys:

  * F1
    * enters the configuration menu
  * F2
    * toggles detection and use of multiple CPU cores (SMP)
  * Space
    * toggles scroll lock (stops/starts error message scrolling)
  * Enter
    * single message scroll (only when scroll lock enabled)
  * Escape
    * exits the test and reboots the machine

Note that testing is stalled when scroll lock is enabled and the scroll region
is full.

The configuration menu allows the user to:

  * select which tests are run (default: all tests)
  * limit the address range over which tests are performed (default: all memory)
  * select the CPU sequencing mode (default: parallel)
    * parallel
      * each CPU core works in parallel on a subset of the memory region being
        tested
    * sequential
      * each CPU core works in turn on the full memory region being tested
    * round robin
      * a single CPU core works on the full memory region being tested, with a
        new CPU core being selected (in round-robin fashion) for each test
  * select the error reporting mode (default: individual errors)
    * error counts only
    * error summary
    * individual errors
    * BadRAM patterns
  * select which of the available CPU cores are used (at startup only)
    * a maximum of 32 CPU cores can be selected, due to display limits
    * the bootstrap processor (BSP) cannot be deselected
  * enable or disable the temperature display (at startup only)
  * enable or disable boot tracing for debug (at startup only)
  * skip to the next test (when running tests)

In all cases, the number keys may be used as alternatives to the function keys
(1 = F1, 2 = F2, ... 0 = F10).

## Error Reporting

The error reporting mode may be changed at any time without disrupting the
current test sequence. Error statistics are collected regardless of the
current error reporting mode (so switching to error summary mode will show
the accumulated statistics since the current test sequence started). BadRAM
patterns are only accumulated when in BadRAM mode.

Any change to the selected tests, address range, or CPU sequencing mode will
start a new test sequence and reset the error statistics.

### Error Counts Only

The error counts only mode just displays the total number of errors found since
the current test sequence started.

### Error Summary

The error summary mode displays the following information:

  * Lowest Error Address
    * the lowest address that where an error has been reported
  * Highest Error Address
    * the highest address that where an error has been reported
  * Bits in Error Mask
    * a hexadecimal mask of all bits that have been in error
  * Bits in Error
    * total bits in error for all error instances and the min, max and average
      number of bits in error across each error instance
  * Max Contiguous Errors
    * the maximum of contiguous addresses with errors
  * Test Errors
     * the total number of errors for each individual test

### Individual Errors

The individual error mode displays the following information for each error
instance:

  * pCPU
    * the physical CPU core number that detected the error
  * Pass
    * the test pass number where the error occurred (a test pass is a single
      run over all the currently selected tests)
  * Test
    * the individual test number where the error occurred
  * Failing Address
    * the memory address where the error occurred
  * Expected
    * the hexadecimal data pattern expected to be found
  * Found
    * the hexadecimal data pattern read from the failing address
  * Err Bits (only in 32-bit builds)
    * a hexadecimal mask showing the bits in error

### BadRAM Patterns

The BadRAM patterns mode accumulates and displays error patterns for use with
the [Linux BadRAM feature](http://rick.vanrein.org/linux/badram/). Lines are
printed in the form `badram=F1,M1,F2,M2...` In each `F,M` pair, the `F`
represents a fault address and the `M` is a bitmask for that address. These
patterns state that faults have occurred in addresses that equal F on all `1`
bits in M. Such a pattern may capture more errors that actually exist, but
at least all the errors are captured. These patterns have been designed to
capture regular patterns of errors caused by the hardware structure in a terse
syntax.

The BadRAM patterns are grown incrementally rather than calculated from an
overview of all errors. The number of pairs is constrained to ten for a
number of practical reasons. As a result, handcrafting patterns from the
output in address printing mode may, in exceptional cases, yield better
results.

## Trouble-shooting Memory Errors

Please be aware that not all errors reported by PCMemTest are due to bad
memory. The test implicitly tests the CPU, caches, and motherboard. It is
impossible for the test to determine what causes the failure to occur. Most
failures will be due to a problem with memory. When it is not, the only option
is to replace parts until the failure is corrected.

Once a memory error has been detected, determining the failing module is not a
clear cut procedure. With the large number of motherboard vendors and possible
combinations of memory slots it would be difficult if not impossible to assemble
complete information about how a particular error would map to a failing memory
module. However, there are steps that may be taken to determine the failing
module. Here are some techniques that you may wish to use:

* Removing modules
  * This is the simplest method for isolating a failing modules, but may only
    be employed when one or more modules can be removed from the system. By
    selectively removing modules from the system and then running the test
    you will be able to find the bad module(s). Be sure to note exactly which
    modules are in the system when the test passes and when the test fails.

* Rotating modules
  * When none of the modules can be removed then you may wish to rotate modules
    to find the failing one. This technique can only be used if there are three
    or more modules in the system. Change the location of two modules at a time.
    For example put the module from slot 1 into slot 2 and put the module from
    slot 2 in slot 1. Run the test and if either the failing bit or address
    changes then you know that the failing module is one of the ones just moved.
    By using several combinations of module movement you should be able to
    determine which module is failing.

* Replacing modules
  * If you are unable to use either of the previous techniques then you are
    left to selective replacement of modules to find the failure.

Sometimes memory errors show up due to component incompatibility. A memory
module may work fine in one system and not in another. This is not uncommon
and is a source of confusion. The components are not necessarily bad but
certain combinations may need to be avoided.

In the vast majority of cases errors reported by PCMemTest are valid. There
are some systems that cause PCMemTest to be confused about the size of memory
and it will try to test non-existent memory. This will cause a large number of
consecutive addresses to be reported as bad and generally there will be many
bits in error. If you have a relatively small number of failing addresses and
only one or two bits in error you can be certain that the errors are valid.
Also intermittent errors are always valid.

All valid memory errors should be corrected. It is possible that a particular
error will never show up in normal operation. However, operating with marginal
memory is risky and can result in data loss and even disk corruption.

PCMemTest can not diagnose many types of PC failures. For example a faulty CPU
that causes your OS to crash will most likely just cause PCMemTest to crash in
the same way.

## Execution Time

The time required for a complete pass of PCMemTest will vary greatly depending
on CPU speed, memory speed, and memory size. PCMemTest executes indefinitely.
The pass counter increments each time that all of the selected tests have been
run. Generally a single pass is sufficient to catch all but the most obscure
errors. However, for complete confidence when intermittent errors are suspected
testing for a longer period is advised.

## Memory Testing Philosophy

There are many good approaches for testing memory. However, many tests simply
throw some patterns at memory without much thought or knowledge of memory
architecture or how errors can best be detected. This works fine for hard
memory failures but does little to find intermittent errors. BIOS based memory
tests are useless for finding intermittent memory errors.

Memory chips consist of a large array of tightly packed memory cells, one for
each bit of data. The vast majority of the intermittent failures are a result
of interaction between these memory cells. Often writing a memory cell can
cause one of the adjacent cells to be written with the same data. An effective
memory test attempts to test for this condition. Therefore, an ideal strategy
for testing memory would be the following:

  1. write a cell with a zero
  2. write all of the adjacent cells with a one, one or more times
  3. check that the first cell still has a zero

It should be obvious that this strategy requires an exact knowledge of how the
memory cells are laid out on the chip. In addition there is a never ending
number of possible chip layouts for different chip types and manufacturers
making this strategy impractical. However, there are testing algorithms that
can approximate this ideal strategy.

## PCMemTest Test Algorithms

PCMemTest uses two algorithms that provide a reasonable approximation of the
ideal test strategy above. The first of these strategies is called moving
inversions. The moving inversion tests work as follows:

  1. Fill memory with a pattern
  2. Starting at the lowest address
     1. check that the pattern has not changed
     2. write the pattern's complement
     3. increment the address
     4. repeat 2.i to 2.iii
  3. Starting at the highest address
     1. check that the pattern has not changed
     2. write the pattern's complement
     3. decrement the address
     4. repeat 3.i - 3.iii

This algorithm is a good approximation of an ideal memory test but there are
some limitations. Most high density chips today store data 4 to 16 bits wide.
With chips that are more than one bit wide it is impossible to selectively
read or write just one bit. This means that we cannot guarantee that all
adjacent cells have been tested for interaction. In this case the best we can
do is to use some patterns to ensure that all adjacent cells have at least
been written with all possible one and zero combinations.

It can also be seen that caching, buffering, and out of order execution will
interfere with the moving inversions algorithm and make it less effective.
It is possible to turn off caching but the memory buffering in new high
performance chips cannot be disabled. To address this limitation a new
algorithm called Modulo-20 was created. This algorithm is not affected by
caching or buffering. The algorithm works as follows:

  1. For starting offsets of 0 - 19 do
     1. write every 20th location with a pattern
     2. write all other locations with the pattern's complement
     3. repeat 1.ii one or more times
     4. check every 20th location for the pattern

This algorithm accomplishes nearly the same level of adjacency testing as
moving inversions but is not affected by caching or buffering. Since separate
write passes (1.i, 1.ii) and the read pass (1.iv) are done for all of memory we
can be assured that all of the buffers and cache have been flushed between
passes. The selection of 20 as the stride size was somewhat arbitrary. Larger
strides may be more effective but would take longer to execute. The choice of
20 seemed to be a reasonable compromise between speed and thoroughness.

## Individual Test Descriptions

PCMemTest executes a series of numbered tests to check for errors. These tests
consist of a combination of test algorithm, data pattern and caching. The
execution order for these tests were arranged so that errors will be detected
as rapidly as possible. A description of each test follows.

To allow testing of more than 4GB of memory on 32-bit CPUs, the physical
address range is split into 1GB windows which are be mapped one at a time
into a virtual memory window. Each 1GB window may contain one or more
contiguous memory regions. For most tests, the test is performed on each
memory region in turn. Caching is enabled for all but the first test.

### Test 0 : Address test, walking ones, no cache

In each memory region in turn, tests all address bits by using a walking
ones address pattern. Errors from this test are not used to calculate BadRAM
patterns.

### Test 1 : Address test, own address in window

In each memory region in turn, each address is written with its own address
and then each address is checked for consistency. This test is performed
sequentially with each available CPU, regardless of the CPU sequencing mode
selected by the user.

### Test 2 : Address test, own address + window

Across all memory regions, each address is written with its own address plus
the window number and then each address is checked for consistency. This
catches any errors in the high order address bits that would be missed when
testing each window in turn. This test is performed sequentially with each
available CPU, regardless of the CPU sequencing mode selected by the user.

### Test 3 : Moving inversions, ones & zeros

In each memory region in turn, and for each pattern in turn, uses the moving
inversions algorithm with patterns of all ones and all zeros.

### Test 4 : Moving inversions, 8 bit pattern

In each memory region in turn, and for each pattern in turn, uses the moving
inversions algorithm with patterns of 8-bit wide walking ones and walking zeros.

### Test 5 : Moving inversions, random pattern

In each memory region in turn, and for each pattern in turn, uses the moving
inversions algorithm with patterns of a random number and its complement. The
random number is different on each test pass so multiple passes increase
effectiveness.

### Test 6 : Moving inversions, 32/64 bit pattern

In each memory region in turn, and for each pattern in turn, uses the moving
inversions algorithm with patterns of 32-bit wide (on 32-bit builds) or 64-bit
wide (on 64-bit builds) walking ones and walking zeros. Unlike previous tests,
the pattern is rotated 1 bit on each successive address.

### Test 7 : Block move, 64 moves

This test stresses memory by using block move (movs) instructions and is based
on Robert Redelmeier's burnBX test.

In each memory region in turn, memory is initialized with shifting patterns
that are inverted every 8 bytes. Then blocks of memory are moved around using
the movs instruction. After the moves are completed the data patterns are
checked. Because the data is checked only after the memory moves are completed
it is not possible to know where the error occurred. The addresses reported
are only for where the bad pattern was found. In consequence, errors from this
test are not used to calculate BadRAM patterns.

### Test 8 : Random number sequence

In each memory region in turn, each address is written with a random number,
then each address is checked for consistency and written with the complement
of the original data, then each address is again checked for consistency.

### Test 9 : Modulo 20, random pattern

In each memory region in turn, and for each pattern in turn, uses the
Modulo-20 algorithm with patterns of a random number and its complement.
The random number is different on each test pass so multiple passes increase
effectiveness.

### Test 10 : Bit fade test, 2 patterns

Across all memory regions, and for each pattern in turn, initialises each
memory location with a pattern, sleeps for a period of time, then checks
each memory location for consistency. The test is performed with patterns
of all zeros and all ones.

## Known Limitations and Bugs

  * The UHCI USB controller is not yet supported.
  * Temperature reporting is currently only supported for Intel CPUs.

## Acknowledgments

PCMemTest was based on Memtest86+, developed by Samuel Demeulemeester, which
in turn was based on Memtest86, developed by Chris Brady with the resources
and assistance listed below:

 * The initial versions of the source files bootsect.S, setup.S, head.S and
   build.c are from the Linux 1.2.1 kernel and have been heavily modified.

 * Doug Sisk provided code to support a console connected via a serial port.
   (not used by PCMemTest)

 * Code to create BadRAM patterns was provided by Rick van Rein.

 * The block move test is based on Robert Redelmeier's burnBX test.

 * Screen buffer code was provided by Jani Averbach.
   (not used by PCMemTest)

 * Eric Biederman provided all of the feature content for version 3.0
   plus many bugfixes and significant code cleanup.

 * Major enhancements to hardware detection and reporting in version 3.2, 3.3
   and 3.4 provided by Samuel Demeulemeester (from Memtest86+ v1.11, v1.60
   and v1.70).

In addition, several bug fixes for Memtest86+ were imported from
[anphsw/memtest86](https://github.com/anphsw/memtest86).