Skip to content

Latest commit

 

History

History
312 lines (179 loc) · 9.77 KB

README.TRACEFILE.md

File metadata and controls

312 lines (179 loc) · 9.77 KB
Raw

header image Return to master README.md

Hercules Trace-to-File Support

Contents

  1. Background
  2. Creating a Tracefile
  3. Using TFPRINT to print a Tracefile
  4. Bug Reports

Background

Hercules debug tracing has traditionally involved using the t+ and/or t+devnum commands, which causes Hercules to print debug messages directly to the Hercules console (and thus to the Hercules logfile too).

Depending on what is being traced, this typically causes many thousands of messages to be generated, literally flooding the Hercules console with messages, therefore slowing Hercules down to a crawl due to the unavoidable overhead of having to format each trace message before displaying it, as well as the inefficiency of Hercules's overall management of console messages.

The trace-to-file feature of Hercules greatly reduces the overhead and inefficiency of performing debug tracing by simply writing the raw unformatted debugging information to a tracefile instead, which can then be printed offline at a later time, thereby allowing Hercules to run much faster than otherwise. That is to say, the impact to your guest's execution speed is greatly reduced, allowing it to run significantly faster than it otherwise would had the messages been formatted and displayed on the Hercules console instead like normal.

The resulting debugging tracefile that is created can then be sent to Hercules support personnel for analysis, which is much more convenient than sending what would otherwise be a very large Hercules logfile.

This also allows Hercules support personnel to select and print only specific information from the tracefile too, which is impossible to do when all you have is a very large text logfile with all messages already formatted in it (even ones they may not be currently interested in).

Creating a Tracefile

To create a tracefile, you use the tf command that defines the file where the debugging information will be written, and then activate debug tracing like normal (e.g. t+ and/or t+devnum, etc):

   tf on "file=path/to/your-tracefile.dat" max=10g
   t+

   (tracing occurs...)

   t-
   tf off

The ON parameter enables tracing to a file. Please note that nothing will actually be written to the file until the first tracing event occurs, which is why after you enter the tf command you must then manually enable debug tracing via the t+ and/or t+devnum commands. Once the first event occurs, the file is opened and logging begins.

The FILE= parameter is required, and defines the path to your desired output tracefile. You can name the file whatever you want. If you path contains any spaces, enclose the entire parameter within double quotes.

The MAX= parameter specifies the desired maximum size (in either megabytes or gigabytes) that the tracefile is allowed to grow to. Specify the value as either nnnM or nnnG. This prevents your tracefile from becoming unacceptably large and filling up your hard drive.

Specify a value that is acceptable to your needs, being aware that once the maximum is reached, all tracing is stopped. So choose a value that is large enough to hold all of the information you wish to capture. Choosing a large value is probably the best course of action. Depending on what you are tracing, the actual resulting tracefile will likely be much smaller.

When tracing begins and the file is opened, you will see a message issued to the Hercules console:

    HHC02383I Trace file tracing begun...

If the tracefile unexpectedly fills up (i.e. if the specified max= is reached), you will see another message appear:

    HHC02379I Trace file MAX= reached; file closed, tracing stopped

To manually stop tracing at any time, simply enter the command to disable (turn off) debug tracing (e.g. t- and/or t-devnum), and then enter the tf off (or tf close) command to stop/disable tracefile tracing. This will close the tracefile allowing you to then print it via the tfprint utility, or .zip it and send it to your Hercules support contact.

Using TFPRINT to print a Tracefile

The Hercules tfprint utility allows you to print all or only some of the information captured in a tracefile. Its many options allow you to select only the specific records needed to research your bug.

    Usage:  tfprint  [options...] tracefile

Options:

Short Long Argument(s)
-i --info
-c --cpu hh[[-hh][,hh]]
-r --traceopt         (Hercules 'TRACEOPT')
-n --count nnnnnn[[-nnnnnn] | [.nnn]]
-e --msg nnnnn[,nnnnn]
-s --storage V|R:hhhhhh[[-hhhhhh] | [.hhh]]
-d --date YYYY/MM/DD-YYYY/MM/DD
-t --time HH:MM:SS.nnnnnn-HH:MM:SS.nnnnnn
-o --opcode hhhhhhhh[,hhxxhhxxhhhh]
-m --msglvl         (Hercules 'msglevel')
-u --unit uuuu[[-uuuu]|[.nnn]]
-p --codepage        (Hercules 'CODEPAGE')

 

Meaning:

--info

Prints only Summary Information from the tracefile and then exits.

--cpu

Prints only the instruction trace records for the specified CPU(s).

--traceopt

Defines how registers will be printed for instruction tracing.

TRADITIONAL (the default), displays the instruction first, followed by the registers.

REGSFIRST displays the registers first, followed by the instruction.

NOREGS does not display registers at all, and displays only the instruction instead.

--count

Print only records nnnnnn to nnnnnn (by count). Allows you to print only a very specific trace record or range of trace records.

--msg

Print only messages with the specified message number. The nnnnn value is the numeric portion of the Hercules message number. For example: to print only HHC01334I messages, specify '1334` for this option.

--storage

Prints only instructions that reference or modify the specified 'V'irtual or 'R'eal storage range.

This option is particularly handy in that it allows you to see only those instructions that access a specific area of storage. This allows you to determine where a possible storage overlay (buffer overrun) problem is occurring for example.

It doesn't only work for instructions either. It also works for I/O. If a CCW for example reads data into, or writes data from, the specified storage, this will identify it.

--date

Allows you to selectively print only trace records according to the date range they were created. This option is usually coupled with the --time option as well, to choose only events that occurred within a very specific date/time range.

--time

Allows you to selectively print only trace records according to the time range they were created. This option is usually coupled with the --date option as well, to choose only events that occurred within a very specific date/time range.

--opcode

This option allows you to print only specific instructions identified by their opcode. The hhhhhh... value is the hexadecimal machine instruction you wish to print. Only as many bytes are checked as are specified. For 2-byte instructions, specify only hhhh. For 4-byte instructions, specify hhhhhhhh, etc.

Wildcards may be used to cause the comparison to ignore specific nibbles. For example: to print only those IIHL (Insert Immediate Halfword Low) instructions that load any value into any register, specify A5x1xxxx.

--msglvl

Allows you to run the tfprint utility with a specific non-default msglevel setting. Refer to help msglevel for more information.

--unit

Prints only CCW trace records for the specified I/O unit(s), where uuuu identifies the specific unit(s) whose trace record(s) you wish to see.

--codepage

Allows you to override Hercules's default CODEPAGE setting used for ASCII/EBCDIC translation. Refer to Hercules documentation for valid CODEPAGE values.

It should be noted that some of the options support multiple arguments separated by commas (most notibly the --cpu, --msg and --opcode options), allowing you to print only trace records for more than one CPU, more than one message number and/or more than one instruction opcode. Simply follow an option's argument with a comma with no intervening space followed by the second or subsequent argument for that same option.

For example: --cpu 00,06 will print only instruction trace records for CPUs 00 or 06. Similarly, specifying --opcode 45Ex,07FE will print only instruction trace record for the BAL R14,xxxx and BR R14 instructions.

Bug Reports

Bug reports (together with your diagnosis of the fault, please!) should be either entered into our Github issue tracker (preferred) at https://github.com/SDL-Hercules-390/hyperion/issues/, or else reported via message to the main hercules-390 discussion group at https://hercules-390.groups.io/g/group.