srec_examples

NAME
DESCRIPTION
CONVERTING FILE FORMATS
BINARY FILES
JOINING FILES TOGETHER
CROPPING THE DATA
MOVING THINGS AROUND
FILLING THE BLANKS
INSERTING CONSTANT DATA
DATA ABOUT THE DATA
MANGLING THE DATA
COPYRIGHT
MAINTAINER

NAME

srec_examples - examples of how to use SRecord

DESCRIPTION

The srec_cat command is very powerful, due to the ability to combine the the input filters in almost unlimited ways. This manual page describes a few of them.

This manual page describes how to use the various input files, input filters and input generators. But these are only examples, for more complete details, see the srec_input(1) manual page.

The Commands Lines Are Too Long
If you are marooned on an operating system with absurdly short command line length limits, some of the commands which follow may be too long. You can get around this handicap by placing your command line in a file, say fred.txt, and then tell srec_cat(1) to read this file for the rest of its command line, like this

srec_cat @fred.txt

This also has the advantage of allowing comments, allowing you to write your command line options over several lines, and even indenting to make the command more clear. Comments start at a “#” and extend to the end of the line. Blank lines are ignored.

Of course, you could always upgrade to Linux, which has been sucking less for over 30 years now.

Your Examples Wanted
If you have a clever way of using SRecord, or have solved a difficult problem with SRecord, you could contribute to this manual page, making it more useful for everyone. Send your example in an email to the email address at the end of this manual page.

CONVERTING FILE FORMATS

The simplest of the things srec_cat(1) can do is convert from one EPROM file format to another. Please keep in mind, as you read this section, that you can do many of these things simultaneously in one command. They are only broken out separately to make them easier to understand.

Intel to Motorola
One of the simplest examples is converting files from Intel hex format to Motorola S-Record format:

srec_cat intel-file -intel -o srec-file

Note that the format specifier immediately follows the name of the file it is describing. Pick any two formats that SRecord understands, and it can convert between all of them. (Except the assembler, BASIC, C and FPGA outputs which are write only.)

Motorola to Intel
Converting the other way is just as simple:

srec_cat srec-file -o intel-file -intel

The default format is Motorola S-Record format, so it does not need to be specified after the file name.

Different Shapes of the Same Format
It is regrettably common that some addle-pated EPROM programmers only implement a portion of the specification used to represent their hex files. For example, some compilers produce “s19” Motorola data (that is, S1 data records with S9 start records, 16 bit address fields) which would be OK except that some blockhead EPROM programmers insist on “s37” Motorola data (that is, S3 data records with S7 start records, 32 bit address fields).

It is possible to convert from one Motorola shape to another using the -Address-Length option:

srec_cat short.srec -o long.srec -address-length=4

This command says to use four byte (32-bit) addresses on output.

This section also applies to Intel hex files, as they, too, have the ability to select from a variety of address widths. To convert from one Intel shape to another using the same -Address-Length option:

srec_cat i32.hex -o i16.hex -address-length=3

This command says to use “i16hex” 20-bit segmented addresses on output. An address length of 4 is the default (“i32hex” 32-bit linear addressing), and an address length of 2 would request “i8hex” 16-bit addressing.

Line Lengths
From time to time you will come across a feeble-minded EPROM programmer that can’t cope with long text lines, they assume that there will only ever be 46 characters per line and barf when they see the default line lengths that srec_cat(1) writes (or worse, get a stack scribble and crash).

The Motorola S-record format definition permits up to 255 bytes of payload, or lines of 514 characters, plus the line termination. All EPROM programmers should have sufficiently large line buffers to cope with records this big. Few do.

The -line-length option may be used to specify the maximum line length (not including the newline) to be used on output. For example, 16 byte payloads for Motorola hex

srec_cat long.srec -o short.s19 -line-length=46

The line length option interacts with the address length option, so some tinkering to optimize for your particular situation many be necessary.

Output Block Size
Every once in a while you will come across an ancient daft EPROM programmer that can’t cope with long data records, they assume that there will only ever be at most 16 bytes of data per record, and barf when they see the default 32 byte payloads that srec_cat(1) writes (or worse, the buffer over-run causes a tall grass walk that scribbles on your EPROM).

The Intel hex format definition permits up to 255 bytes of payload data per record. All EPROM programmers should have sufficiently large data buffers to cope with records this big. Good luck with that.

The -Output-Block-Size option may be used to specify the record data size to be used on output. For example, Intel hex with 16 byte payloads:

srec_cat long.srec -o short.hex -intel -obs=16

Be careful not to put the -obs option between the output file name and the format specifier.

Just the Data, Please
There are some bonehead EPROM programmers which can only cope with data records, and are unable to cope with header records or execution start address records. If you have this problem, the -data-only option can be used to suppress just about everything except the data. The actual effect depends on the format, of course, because some don’t have these features anyway.

The -data-only option is short hand. There are four properties which may be -disabled or -enabled separately. See the srec_cat(1) man page for a description of the -disabled and -enabled options.

For example, your neanderthal EPROM programmer requires Motorola hex with header records (S0), but without data count (S5) records. Not using the -data-only option has it barf on the data count record, but using the -data-only option has it barf on the missing header record. Using the -disable=data-count option would leave the header record intact while suppressing the data count record.

Data Headers
The srec_cat(1) command always tries to pass through header records unchanged, whenever they are present. It even tries preserve them across file format changes, to the limit the file formats are capable of.

If there is no file header record and you would like to add one, or you wish to override an existing file header record, use the -header=string option. You will need to quote the string (to insulate it from the shell) if it contains spaces or shell meta-characters.

Execution Start Addresses
The srec_cat(1) command always tries to pass through execution start addresses (typically occurring at the end of the file), whenever they are present. They are adjusted along with the data records by the -offset filter. It even tries preserve them across file format changes, to the limit the file formats are capable of.

If there is no execution start address record and you would like to add one, or you wish to override an existing execution start address record, use the -execution-start-address=number option.

Please note: the execution start address is a different concept than the first address in memory of your data. Think of it as a “goto” address to be jumped to by the monitor when the hex load is complete. If you want to change where your data starts in memory, use the -offset filter.

Fixing Checksums
Some embedded firmware developers are saddled with featherbrained tools which produce incorrect checksums, which the more vigilant models of EPROM programmer will not accept.

To fix the checksums on a file, use the -ignore-checksums option. For example:

srec_cat broken.srec -ignore-checksums -o fixed.srec

The checksums in broken.srec are parsed (it is still and error if they are absent) but are not checked. The resulting fixed.srec file has correct checksums. The -ignore-checksums option only applies to input.

This option may be used on any file format which has checksums, including Intel hex.

Discovering Mystery Formats
See the What Format Is This? section, below, for how to discover and convert mystery EPROM load file formats.

BINARY FILES

It is possible to convert to and from binary files. You can even mix binary files and other formats together in the same srec_cat(1) command.

Writing Binary Files
The simplest way of reading a hex file and converting it to a binary file looks like this:

srec_cat fred.hex -o fred.bin -binary

This reads the Motorola hex file fred.srec and writes it out to the fred.bin as raw binary.

Note that the data is placed into the binary file at the byte offset specified by the addresses in the hex file. If there are holes in the data they are filled with zero. This is, of course, common with linker output where the code is placed starting at a particular place in memory. For example, when you have an image that starts at 0x100000, the first 1MB of the output binary file will be zero.

You can automatically cancel this offset using a command like

srec_cat fred.hex -offset − -minimum-addr fred.hex -o fred.bin

The above command works by offsetting the fred.hex file lower in memory by the least address in the fred.hex file’s data.

See also the srec_binary(5) man page for additional detail.

Reading Binary Files
The simplest way of reading a binary file and converting it looks like this

srec_cat fred.bin -binary -o fred.srec

This reads the binary file fred.bin and writes all of its data back out again as a Motorola S-Record file.

Often, this binary isn’t exactly where you want it in the address space, because it is assumed to reside at address zero. If you need to move it around use the -offset filter.

srec_cat fred.bin -binary -offset 0x10000 -o fred.srec

You also need to avoid file “holes” which are filled with zero. You can use the -crop filter, of you could use the -unfill filter if you don’t know exactly where the data is.

srec_cat fred.bin -binary -unfill 0x00 512 -o fred.srec

The above command removes runs of zero bytes that are 512 bytes long or longer. If your file contains 1GB of leading zero bytes, this is going to be slow, it may be better to use the dd(1) command to slice and dice first.

JOINING FILES TOGETHER

The srec_cat command takes its name from the UNIX cat(1) command, which is short for “catenate” or “to join”. The srec_cat command joins EPROM load files together.

All In One
Joining EPROM load files together into a single file is simple, just name as many files on the command line as you need:

srec_cat infile1 infile2 -o outfile

This example is all Motorola S-Record files, because that’s the default format. You can have multiple formats in the one command, and srec_cat(1) will still work. You don’t even have to output the same format:

srec_cat infile1 -spectrum infile2 -needham \
-o outfile -signetics

These are all ancient formats, however it isn’t uncommon to have to mix and match Intel and Motorola formats in the one project.

Overlaying two data files
It is common to want to “join” two hex files together, without any changes of address. on the assumption neither file intersects with the other. This is a simple “layers”, it is quite common for linkers to output the main code, and then a whole bunch of relocation and jump destination, by writing a two layered files.

srec_cat one.he two.hex -o three.hex

Almost always you see an error

srec_cat: two.srec: 49282: contradictory 0x00000000 value (previous = 0x00, this one = 0x80)

This means that the files actually intersect, they try to set the same location. You can turn the error into a warning, using the -contradictory-bytes=warning command line option. But this will probably generate a bazillion warnings.

The necessary step is to crop the first file, to avoid the regions the second file is going o be overwriting.

srec_cat \
one.srec -exclude -within two.srec \
two.srec -exclude -within one.srec \
-o three.hex

Depending on your linker this will have no errors (but if it wants another layer, more jiggery-pokery is required).

Filtering After Joining
There are times when you want to join two sets of data together, and then apply a filter to the joined result. To do this you use parentheses.

srec_cat \
’(’ \
infile
-exclude 0xFFF0 0x10000 \
-generate 0xFFF0 0xFFF8 -repeat-string ’Bananas ’ \
’)’ \
-length-b-e 0xFFF8 4 \
-checksum-neg-b-e 0xFFFC 4 4 \
-o outfile

The above example command catenate an input file (with the generated data area excluded) with a constant string. This catenated input is then filtered to add a 4-byte length, and a 4-byte checksum.

Joining End-to-End
All too often the address ranges in the EPROM load files will overlap. You will get an error if they do. If both files start from address zero, because each goes into a separate EPROM, you may need to use the offset filter:

srec_cat infile1 \
infile2
-offset 0x80000 \
-o outfile

Sometimes you want the two files to follow each other exactly, but you don’t know the offset in advance:

srec_cat infile1 \
infile2
-offset -maximum-addr infile1 \
-o outfile

Notice that where the was a number (0x80000) before, there is now a calculation (-maximum-addr infile1). This is possible most places a number may be used (also -minimum-addr and -range).

CROPPING THE DATA

It is possible to copy an EPROM load file, selecting addresses to keep and addresses to discard.

What To Keep
A common activity is to crop your data to match your EPROM location. Your linker may add other junk that you are not interested in, e.g. at the RAM location. In this example, there is a 1MB EPROM at the 2MB boundary:

srec_cat infile -crop 0x200000 0x300000 \
-o outfile

The lower bound for all address ranges is inclusive, the upper bound is exclusive. If you subtract them, you get the number of bytes.

Address Offset
Just possibly, you have a moronic EPROM programmer, and it barfs if the EPROM image doesn’t start at zero. To find out just where is does start in memory, use the srec_info(1) command:

$ srec_info example.srec
Format: Motorola S-Record
Header: extra-whizz tool chain linker
Execution Start Address: 0x00200000
Data: 0x200000 - 0x32AAEF
$

Rather than butcher the linker command file, just offset the addresses:

srec_cat infile -crop 0x200000 0x300000 -offset −0x200000 \
-o outfile

Note that the offset given is negative, it has the effect of subtracting that value from all addresses in the input records, to form the output record addresses. In this case, shifting the image back to zero.

This example also demonstrates how the input filters may be chained together: first the crop and then the offset, all in one command, without the need for temporary files.

If all you want to do is offset the data to start from address zero, this can be automated, so you don’t have to know the minimum address in advance, by using srec_cat’s ability to calculate some things on the command line:

srec_cat infile -offset − -minimum-addr infile \
-o outfile

Note the spaces either side of the minus sign, they are mandatory.

What To Throw Away
There are times when you need to exclude an small address range from an EPROM load file, rather than wanting to keep a small address range. The -exclude filter may be used for this purpose.

For example, if you wish to exclude the address range where the serial number of an embedded device is kept, say 0x20 bytes at 0x100, you would use a command like this:

srec_cat input.srec -exclude 0x100 0x120 -o output.srec

The output.srec file will have a hole in the data at the necessary locations.

Note that you can have both -crop and -exclude on the same command line, whichever works more naturally for your situation.

Discontinuous Address Ranges
Address ranges don’t have to be a single range, you can build up an address range using more than a single pair.

srec_cat infile -crop 0x100 0x200 0x1000 0x1200 \
-o outfile

This filter results in data from 0x100..0x1FF and data from 0x1000..0x1200 to pass through, the rest is dropped. This is is more efficient than chaining a -crop and an -exclude filter together.

MOVING THINGS AROUND

It is also possible to change the address of data records, both forwards and backwards. It is also possible rearrange where data records are placed in memory.

Offset Filter
The -offset=number filter operates on the addresses of records. If the number is positive the addresses move that many bytes higher in memory, negative values move lower.

srec_cat infile -crop 0x200000 0x300000 -offset −0x200000 \
-o outfile

The above example moves the 1MB block of data at 0x200000 down to zero (the offset is negative) and discards the rest of the data.

Byte Swapping
There are times when the bytes in the data need to be swapped, converting between big-endian and little-endian data usually.

srec_cat infile -byte-swap 4 -o outfile

This reverses bytes in 32 bit values (4 bytes). The default, if you don’t supply a width, is to reverse bytes in 16 bit values (2 bytes). You can actually use any weird value you like, it doesn’t even have to be a power of 2. Perhaps 64 bits (8 bytes) may be useful one day.

Binary Output
You need to watch out for binary files on output, because the holes are filled with zeros. Your 100kB program at the top of 32-bit addressed memory will make a 4GB file. See srec_binary(5) for how understand and avoid this problem, usually with the -offset filter.

Splitting an Image
If you have a 16-bit data bus, but you are using two 8-bit EPROMs to hold your firmware, you can generate the even and odd images by using the -SPlit filter. Assuming your firmware is in the firmware.hex file, use the following:

srec_cat firmware.hex -split 2 0 -o firmware.even.hex
srec_cat firmware.hex -split 2 1 -o firmware.odd.hex

This will result in the two necessary EPROM images. Note that the output addresses are divided by the split multiple, so if your EPROM images are at a particular offset (say 0x10000, in the following example), you need to remove the offset, and then replace it...

srec_cat firmware.hex \
-offset −0x10000 -split 2 0 \
-offset 0x10000 -o firmware.even.hex
srec_cat firmware.hex \
-offset −0x10000 -split 2 1 \
-offset 0x10000 -o firmware.odd.hex

Note how the ability to apply multiple filters simplifies what would otherwise be a much longer script.

Striping
A second use for the -SPlit filter is memory striping.

You don’t have to split into byte-wide parts, you can choose other sizes. It is common to want to convert 32-bit wide data into two set of 16-bit wide data.

srec_cat firmware.hex -split 4 0 2 -o firmware.01.hex
srec_cat firmware.hex -split 4 2 2 -o firmware.23.hex

This is relatively simple to understand, but you can use even wider stripes.

In this next example, the hardware requires that 512-byte blocks alternate between 4 EPROMs. Generating the 4 images would be done as follows:

srec_cat firmware.hex -split 0x800 0x000 0x200 -o firmware.0.hex
srec_cat firmware.hex -split 0x800 0x200 0x200 -o firmware.1.hex
srec_cat firmware.hex -split 0x800 0x400 0x200 -o firmware.2.hex
srec_cat firmware.hex -split 0x800 0x600 0x200 -o firmware.3.hex

Asymmetric Striping
A more peculiar example of striping is the Microchip dsPIC33F microcontroller, that has a weird memory storage pattern and they are able to store 3 bytes in an address that should only contain 2 bytes. The result is a hex file that has zero-filled the top byte (little-endian), and all addresses are doubled from what they are in the chip. Here is an example:

S1130000000102000405060008090A000C0D0E0098
S1130010101112001415160018191A001C1D1E00C8
S1130020202122002425260028292A002C2D2E00F8
S1130030303132003435360038393A003C3D3E0028

To get rid of the 00 padding bytes, leaving only the 3/4 significant bytes, you also use the split filter, with its additional width argument, like this:

srec_cat example.srec -split 4 0 3 -o no_dross.srec

This results in a file with the 00 padding bytes removed. It looks like this:

S113000000010204050608090A0C0D0E1011121451
S1130010151618191A1C1D1E2021222425262829EC
S11300202A2C2D2E30313234353638393A3C3D3E87

Notice how the addresses are 3/4 the size, as well. You can reverse this using the -unsplit and -fill=0 filters.

Unsplit ING Images
The unsplit filter may be used to reverse the effects of the split filter. Note that the address range is expanded leaving holes between the stripes. By using all the stripes, the complete input is reassembled, without any holes.

srec_cat -o firmware.hex \
firmware.even.hex -unsplit 2 0 \
firmware.odd.hex -unsplit 2 1

The above example reverses the previous 16-bit data bus example. In general, you unsplit with the same parameters that you split with.

FILLING THE BLANKS

Often EPROM load files will have “holes” in them, places where the compiler and linker did not put anything. For some purposes this is OK, and for other purposes something has to be done about the holes.

The Fill Filter
It is possible to fill the blanks where your data does not lie. The simplest example of this fills the entire EPROM:

srec_cat infile -fill 0x00 0x200000 0x300000 -o outfile

This example fills the holes, if any, with zeros. You must specify a range - with a 32-bit address space, filling everything generates huge load files.

If you only want to fill the gaps in your data, and don’t want to fill the entire EPROM, try:

srec_cat infile -fill 0x00 -over infile -o outfile

This example demonstrates the fact that wherever an address range may be specified, the -over and -within options may be used.

Unfilling the Blanks
It is common to need to “unfill” an EPROM image after you read it out of a chip. Usually, it will have had all the holes filled with 0xFF (areas of the EPROM you don’t program show as 0xFF when you read them back).

To get rid of all the 0xFF bytes in the data, use this filter:

srec_cat infile -unfill 0xFF -o outfile

This will get rid of all the 0xFF bytes, including the ones you actually wanted in there. There are two ways to deal with this. First, you can specify a minimum run length to the un-fill:

srec_cat infile -unfill 0xFF 5 -o outfile

This says that runs of 1 to 4 bytes of 0xFF are OK, and that a hole should only be created for runs of 5 or more 0xFF bytes in a row. The second method is to re-fill over the intermediate gaps:

srec_cat outfile -fill 0xFF -over outfile \
-o outfile2

Which method you choose depends on your needs, and the shape of the data in your EPROM. You may need to combine both techniques.

Address Range Padding
Some data formats are 16 bits wide, and automatically fill with 0xFF bytes if it is necessary to fill out the other half of a word which is not in the data. If you need to fill with a different value, you can use a command like this:

srec_cat infile -fill 0x0A \
-within infile -range-padding 2 \
-o outfile

This gives the fill filter an address range calculated from details of the input file. The address range is all the address ranges covered by data in the infile, extended downwards (if necessary) at the start of each sub-range to a 2 byte multiple and extended upwards (if necessary) at the end of each sub-range to a 2 byte multiple. This also works for larger multiples, like 1kB page boundaries of flash chips. This address range padding works anywhere an address range is required.

Fill with Copyright
It is possible to fill unused portions of your EPROM with a repeating copyright message. Anyone trying to reverse engineer your EPROMs is going to see the copyright notice in their hex editor.

This is accomplished with two input sources, one from a data file, and one which is generated on-the-fly.

srec_cat infile \
-generate ’(’ 0 0x100000 -minus -within infile ’)’ \
-repeat-string ’Copyright (C) 1812 Tchaikovsky. ’ \
-o outfile

Notice the address range for the data generation: it takes the address range of your EPROM, in this case 1MB starting from 0, and subtracts from it the address ranges used by the input file.

If you want to script this with the current year (because 1812 is a bit out of date) use the shell’s output substitution (back ticks) ability:

srec_cat infile \
-generate ’(’ 0 0x100000 -minus -within infile ’)’ \
-repeat-string "Copyright (C) ’date +%Y’ Tchaikovsky. " \
-o outfile

The string specified is repeated over and over again, until it has filled all the holes.

Obfuscating with Noise
Sometimes you want to fill your EPROM images with noise, to conceal where the real data stops and starts. You can do this with the -random-fill filter.

srec_cat infile -random-fill 0x200000 0x300000 \
-o outfile

It works just like the -fill filter, but uses random numbers instead of a constant byte value.

Fill With 16-bit Words
When filling the image with a constant byte value doesn’t work, and you need a constant 16-bit word value instead, use the -repeat-data generator, which takes an arbitrarily long sequence of bytes to use as the fill pattern:

srec_cat infile \
-generator ’(’ 0x200000 0x300000 -minus -within infile ’)’ \
-repeat-data 0x1B 0x08 \
-o outfile

Notice how the generator’s address range once again avoids the address ranges occupied by the infile’s data. You have to get the endian-ness right yourself.

INSERTING CONSTANT DATA

From time to time you will want to insert constant data, or data not produced by your compiler or assembler, into your EPROM load images.

Binary Means Literal
One simple way is to have the desired information in a file. To insert the file’s contents literally, with no format interpretation, use the binary input format:

srec_cat infile -binary -o outfile

It will probably be necessary to use an offset filter to move the data to where you actually want it within the image:

srec_cat infile -binary -offset 0x1234 -o outfile

It is also possible to use the standard input as a data source, which lends itself to being scripted. For example, to insert the current date and time into an EPROM load file, you could use a pipe:

date | srec_cat - -bin -offset 0xFFE3 -o outfile

The special file name “-” means to read from the standard input. The output of the date command is always 29 characters long, and the offset shown will place it at the top of a 64KB EPROM image.

Repeating Once
The Fill with Copyright section, above, shows how to repeat a string over and over. We can use a single repeat to insert a string just once.

srec_cat -generate 0xFFE3 0x10000 -repeat-string "’date’" \
-o outfile

Notice how the address range for the data generation exactly matches the length of the date(1) output size. You can, of course, add your input file to the above srec_cat(1) command to catenate your EPROM image together with the date and time.

Inserting A Long
Another possibility is to add the Subversion commit number to your EPROM image. In this example, we are inserting it a a 4-byte little-endian value at address 0x0008. The Subversion commit number is in the $version shell variable in this example:

srec_cat -generate 0x0008 0x000C -constant-l-e $version 4 \
infile
-exclude 0x0008 0x000C \
-o outfile

Note that we use a filter to ensure there is a hole in the input where the version number goes, just in case the linker put something there.

DATA ABOUT THE DATA

It is possible to add a variety of data about the data to the output.

Checksums
The -checksum-negative-big-endian filter may be used to sum the data, and then insert the negative of the sum into the data. This has the effect of summing to zero when the checksum itself is summed across, provided the sum width matches the inserted value width.

srec_cat infile \
-crop 0 0xFFFFFC \
-random-fill 0 0xFFFFFC \
-checksum-neg-b-e 0xFFFFFC 4 4 \
-o outfile

In this example, we have an EPROM in the lowest megabyte of memory. The -crop filter ensures we are only summing the data within the EPROM, and not anywhere else. The -random-fill filter fills any holes left in the data with random values. Finally, the -checksum-neg-b-e filter inserts a 32 bit (4 byte) checksum in big-endian format in the last 4 bytes of the EPROM image. Naturally, there is a little-endian version of this filter as well.

Your embedded code can check the EPROM using C code similar to the following:

unsigned long *begin = (unsigned long *)0;
unsigned long *end = (unsigned long *)0x100000;
unsigned long sum = 0;
while (begin < end)
sum += *begin++;
if (sum != 0)
{
Oops

}

The -checksum-bitnot-big-endian filter is similar, except that summing over the checksum should yield a value of all-one-bits (−1). For example, using shorts rather than longs:

srec_cat infile \
-crop 0 0xFFFFFE \
-fill 0xCC 0x00000 0xFFFFFE \
-checksum-neg-b-e 0xFFFFFE 2 2 \
-o outfile

Assuming you chose the correct endian-ness filter, your embedded code can check the EPROM using C code similar to the following:

unsigned short *begin = (unsigned short *)0;
unsigned short *end = (unsigned short *)0x100000;
unsigned short sum = 0;
while (begin < end)
sum += *begin++;
if (sum != 0xFFFF)
{
Oops

}

There is also a -checksum-positive-b-e filter, and a matching little-endian filter, which inserts the simple sum, and which would be checked in C using an equality test.

srec_cat infile \
-crop 0 0xFFFFFF \
-fill 0x00 0x00000 0xFFFFFF \
-checksum-neg-b-e 0xFFFFFF 1 1 \
-o outfile

Assuming you chose the correct endian-ness filter, your embedded code can check the EPROM using C code similar to the following:

unsigned char *begin = (unsigned char *)0;
unsigned char *end = (unsigned char *)0xFFFFF;
unsigned char sum = 0;
while (begin < end)
sum += *begin++;
if (sum != *end)
{
Oops

}

In the 8-bit case, it doesn’t matter whether you use the big-endian or little-endian filter.

Quick Hex-Dump
You can look at the checksum of your data, by using the “hex-dump” output format. This is useful for looking at calculated values, or for debugging an srec_cat(1) command before immortalizing it in a script.

srec_cat infile \
-crop 0 0x10000 \
-fill 0xFF 0x0000 0x10000 \
-checksum-neg-b-e 0x10000 4 \
-crop 0x10000 0x10004 \
-o - -hex-dump

This command reads in the file, checksums the data and places the checksum at 0x10000, crops the result to contain only the checksum, and then prints the checksum on the standard output in a classical hexadecimal dump format. The special file name “-” means “the standard output” in this context.

Cyclic Redundancy Checks
The simple additive checksums have a number of theoretical limitations, to do with errors they can and can’t detect. The CRC methods have fewer problems.

srec_cat infile \
-crop 0 0xFFFFFC \
-fill 0x00 0x00000 0xFFFFFC \
-crc32-b-e 0xFFFFFC \
-o outfile

In the above example, we have an EPROM in the lowest megabyte of memory. The -crop filter ensures we are only summing the data within the EPROM, and not anywhere else. The -fill filter fills any holes left in the data. Finally, the -checksum-neg-b-e filter inserts a 32 bit (4 byte) checksum in big-endian format in the last 4 bytes of the EPROM image. Naturally, there is a little-endian version of this filter as well.

The checksum is calculated using the industry standard 32-bit CRC. Because SRecord is open source, you can always read the source code to see how it works. There are many non-GPL versions of this code available on the Internet, and suitable for embedding in proprietary firmware.

There is also a 16-bit CRC available.

srec_cat infile \
-crop 0 0xFFFFFE \
-fill 0x00 0x00000 0xFFFFFE \
-crc16-b-e 0xFFFFFE \
-o outfile

The checksum is calculated using the CCITT formula. Because SRecord is open source, you can always read the source code to see how it works. There are many non-GPL version of this code available on the Internet, and suitable for embedding in proprietary firmware.

You can look at the CRC of your data, by using the “hex-dump” output format.

srec_cat infile \
-crop 0 0x10000 \
-fill 0xFF 0x0000 0x10000 \
-crc16-b-e 0x10000 \
-crop 0x10000 0x10002 \
-o - -hex-dump

This command reads in the file, calculates the CRC of the data and places the CRC at 0x10000, crops the result to contain only the CRC, and then prints the checksum on the standard output in a classical hexadecimal dump format.

Where Is My Data?
There are several properties of your EPROM image that you may wish to insert into the data.

srec_cat infile -minimum-b-e 0xFFFE 2 -o outfile

The above example inserts the minimum address of the data (low water) into the data, as two bytes in big-endian order at address 0xFFFE. This includes the minimum itself. If the data already contains bytes at the given address, you need to use an exclude filter. The number of bytes defaults to 4.

There is also a -minimum-l-e filter for inserting little-endian values, and two more filters called -exclusive-minimum-b-e and -exclusive-minimum-l-e that do not include the minimum itself in the calculation of the minimum data address.

srec_cat infile -maximum-b-e 0xFFFFFC 4 -o outfile

The above example inserts the maximum address of the data (high water + 1, just like address ranges) into the data, as four bytes in big-endian order at address 0xFFFFFC. This includes the maximum itself. If the data already contains bytes at the given address, you need to use an -exclude filter. The number of bytes defaults to 4.

There is also a -maximum-l-e filter for inserting little-endian values, and two more filters called -exclusive-maximum-b-e and -exclusive-maximum-l-e that do not include the maximum itself in the calculation of the maximum data address.

srec_cat infile -length-b-e 0xFFFFFC 4 -o outfile

The above example inserts the length of the data (high water + 1 − low water) into the data, as four bytes in big-endian order at address 0xFFFFFC. This includes the length itself. If the data already contains bytes at the length location, you need to use an -exclude filter. The number of bytes defaults to 4.

There is also a -length-l-e filter for inserting a little-endian length, and the -exclusive-length-b-e and -exclusive-length-l-e filters that do not include the length itself in the calculation.

What Format Is This?
You can obtain a variety of information about an EPROM load file by using the srec_info(1) command. For example:

$ srec_info example.srec
Format: Motorola S-Record
Header: "http://srecord.sourceforge.net/"
Execution Start Address: 00000000
Data: 0000 - 0122
0456 - 0FFF
$

This example shows that the file is a Motorola S-Record. The text in the file header is printed, along with the execution start address. The final section shows the address ranges containing data (the upper bound of each subrange is inclusive, rather than the exclusive form used on the command line.

$ srec_info some-weird-file.hex -guess
Format: Signetics
Data: 0000 - 0122
0456 - 0FFF
$

The above example guesses the EPROM load file format. It isn’t infallible but it usually gets it right. You can use -guess anywhere you would give an explicit format, but it tends to be slower and for that reason is not recommended. Also, for automated build systems, you want hard errors as early as possible; if a file isn’t in the expected format, you want it to barf.

MANGLING THE DATA

It is possible to change the values of the data bytes in several ways.

srec_cat infile -and 0xF0 -o outfile

The above example performs a bit-wise AND of the data bytes with the 0xF0 mask. The addresses of records are unchanged. I can’t actually think of a use for this filter.

srec_cat infile -or 0x0F -o outfile

The above example performs a bit-wise OR of the data bytes with the 0x0F bits. The addresses of records are unchanged. I can’t actually think of a use for this filter.

srec_cat infile -xor 0xA5 -o outfile

The above example performs a bit-wise exclusive OR of the data bytes with the 0xA5 bits. The addresses of records are unchanged. You could use this to obfuscate the contents of your EPROM.

srec_cat infile -not -o outfile

The above example performs a bit-wise NOT of the data bytes. The addresses of records are unchanged. Security by obscurity?

COPYRIGHT

srec_cat version 1.65
Copyright (C) 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2018, 2019, 2022 Peter Miller

The srec_cat program comes with ABSOLUTELY NO WARRANTY; for details use the ’srec_cat -VERSion License’ command. This is free software and you are welcome to redistribute it under certain conditions; for details use the ’srec_cat -VERSion License’ command.

MAINTAINER

Image srec_examples1.png