Process Dump Reference 20 Dec 2007 ---------------------- Steven H. Levine steve53@earthlink.net 0. Introduction. ------------- This note explains how to set up and record process dumps and how to do basic dump analysis. 1. Setup. ------ There are several ways to enable and configure the process dump facility. Pick the one that best suits you needs. To enable the process dump facility permanently, add the following line to CONFIG.SYS DUMPPROCESS=x where x is the drive where you want to store the dump files. After you reboot, the kernel will generate a dump file for every program that traps. The file will be stored in the root of the drive. Depending on how the program uses memory, dump files can be very large so pick a drive with sufficient free space. For best results pick the JFS or HPFS drive with the most free space. For more information on the DUMPPROCESS command, type view cmdref dumpprocess from the command line. To activate the process dump facility from the command line issue the command procdump on /l:pathname where pathname is the drive or directory where you want to store the dump files. After you enter this command, the kernel will generate a dump file for every program that traps. This setting will stay in effect until you turn it off or until you reboot. To create a onetime dump of a running program, issue the commands procdump on /l:pathname procdump force /proc:processname procdump off where pathname is the drive or directory where you want to store the dump file and processname is the name of the process to be dumped. After you enter these commands, the kernel will generate a single dump file for the named process. The default settings generate a relatively small dump file, but might not include sufficient detail for analysis. If you need all the process memory included in the dump file, enter the command pdumpusr private,update before generating the dump file. If you need all the shared memory included in the dump, enter the command pdumpusr private,shared,update before generating the dump file. There are cases where the dump facility will hang or trap when attempting to dump shared memory. If this occurs, try the alternate command pdumpusr paddr(all),update This will generate a dump file slightly larger than the quantity of physical memory, but it is usually sufficient to generate a usable dump file. For more information on the pdumpusr options, see \os2\system\ras\procdump.doc 2. Recording Process Dumps. ------------------------ Once you have enabled the dump facility, the kernel will a generate dump file for each trap that occurs or whenever you use the FORCE option. If you are getting frequent traps, turn off the dump facility as soon as possible to avoid filling up your disk. To turn off the dump facility from the command line enter the command procdump off If you enabled the dump facility from CONFIG.SYS, don't forget to REM out the line in CONFIG.SYS when you no longer need to collect the dump files. 3. Preparing to Use the PM Dump Facility. -------------------------------------- The PM Dump Facility (PMDF) is a generic tool which needs to be configured to understand the dump files generated by a specific kernel version. PMDF is configured with a set of files known as Dump Symbols and are a combination of programs and data files. The programs (DF_RET.EXE and DF_DEB.EXE) are invoked by PMDF to retrieve data from the dump file. These programs understand the layout of the kernel data structures and are typically specific to a range of kernel revisions. The data files include System Definition files and Symbol files. The System Definition files (.sdf) configure df_ret and df_deb and are typically specific to a single kernel revision. The Symbol files (.sym) contain data used to translate binary addresses within an executable to symbolic labels. These files are typically specific to a specific version of the executable. This means that a set of Dump Symbols is typically kernel version specific, FixPak specific and application version specific. Assuming a standard install of the Dump Facility, the Dump Symbols sets are stored in subdirectories of \os2\pdpsi\pmdf on your boot drive and indentified by the index file \os2\pdpsi\pmdf\pmdfvers.lst. Depending on what components you installed when you installed eCS or Warp on the box where the trap occurred, you might already have the files needed to examine the dump installed. If not, the following examples describe how to get the files you need and how to install the files so that PMDF can use them. The examples that follow assume FixPak 15 Kernel version 14.062 Boot drive f: Be sure to replace the values shown in the examples with values that match your specific system. This applies to kernel revision numbers, boot drive letters and other values that are specific to your system. Add pathname prefixes as needed to match where you have stored the files. If you are not sure of what kernel revision you are running, use your favorite hex editor and search for the string "Internal revision" without the quotes. The string that follows is the value PMDF uses to match the kernel to the symbols. You can find the same information with bldlevel f:\os2krnl but the revision number reported in the build level string is usually not in exactly the format that PMDF is looking for. Go to ftp.software.ibm.com/ps/products/os2/fixes/debug and download m015dmp.zip. This is the symbols zip file for FP15 and kernel version 14.062. If you are using a testcase kernel, the dump symbols will be at testcase.boulder.ibm.com/ps/fromibm/os2 and will be named dfyyyymmdd.zip where mmdd matches the testcase kernel date. When you download a testcase kernel, be sure to download the symbols file at the same time. Otherwise, it might be gone when you need it. There are a few sites that archive copies of the test test case kernels. One such site is http://www.os2site.com/sw/upgrades/kernel/ If you are running eComStation 1.2 or newer, the dump symbols may already be installed. If not, they are available on your installation CD in the \os2image\debug and the \os2image\fi\sysmgmt directories. They are also available from your account at the eComStation website (www.ecomstation.com). Create the subdirectory f:\os2\pdpsi\pmdf\warp4.15 Unzip the contents of m015dmp.zip into this subdirectory with unzip -j m015pmd.zip -d f:\os2\pdpsi\pmdf\warp4.15 This will put all the files in m015pmd.zip into a single directory. This is important. If you don't use the -j option or it's equivalent, the Dump Facility will not be able to find the symbols. The zip files are structured for use with either the Kernel Debugger or PMDF. The Kernel Debugger requires that the .sym files be in the same directory as the associated executable. The zip files contain subdirectory information so that the files will be placed in the correct directory when unzipped. PMDF has different requirements. For PMDF, the .sym file sets must be in a subdirectory named in pmdfvers.lst. The Kernel Debugger runs on the Machine Under Test (MUT). The only set of symbols it needs, or can use, are the one's for the installed kernel revision. PMDF is often run on a system other than the MUT which generated the dump. Using pmdfvers.lst to locate the correct symbols for the dump allows PMDF to be used to analyze dumps even when the systems have different kernel revisions installed. Pmdfvers.lst will have a line defining where to find the symbols for each specific kernel revision to be analyzed. Add the following line to pmdfvers.lst warp4.15:14.062:OS/2 Warp 4 FP15 warp4.15 is the directory you unzipped the symbols. 14.062 is the internal version string for the kernel. The version string is case sensitive and must match exactly. The rest is a comment to help you remember what the other two values mean. This will allow PMDF to automatically find the symbols for the 14.062 kernel in the warp4.15 subdirectory. If your application came with .sym files copy them into this directory. If your application came with .map files, use the mapsym utility to create the .sym files. Mapsym is available with most compilers. If you need usage help, just type mapsym from the command line. If you don't have either .map or .sym files for your application, it may be a bit more difficult to analyze the dump file, but PMDF will still work. When editing pmdfvers.lst, make sure you don't have any blank lines otherwise pmdf will trap on startup. Thanks go to Lars Erdmann for this tip. 4. Analyzing Process Dumps. ------------------------ This is the bare bones. Start PMDF. For Warp4, you should have an object in the Problem Determination Tools folder. For eCS, the object should be in the Utilities folder. If you don't have an object, you can start PMDF from the command line. If you didn't install PMDF, you will need to run Selective Install and install the Problem Determination Tools and reapply the last FixPak you installed (i.e. FP15 or whatever). Open the dump file from the PMDF File menu. PMDF should find the matching .sym files, using the data in pmdfvers.lst. If for some reason PMDF cannot match up the dump file and the .sym files, it will prompt you to select a .sym file set from the sets defined in pmdfvers.lst. Since you configured PMDF above, this should not occur. You can try selecting one of the available .sym file sets, but the results will be unpredictable. PMDF may misinterpret the dump file content or it may even trap. Select Synopsis -> Trap Screen from the Analyze menu. Select Thread -> Call Gate from the Analyze menu. If the thread is in a Call Gate, select Thread -> Ring 0 Stack Trace from the Analyze menu. Select Thread -> Ring 3 Stack Trace from the Analyze menu. Enter the commands r ln u eip-20 eip k dd ebp dd esp db esp in the PMDF command line at the bottom of the window. Press the Enter key after each command. If the r command does not report the same cs:eip as shown on the Trap Screen, repeat the above commands substituting the numeric cs:eip value from the Trap Screen for eip, the numeric ss:ebp value from the Trap Screen for ebp and the numeric ss:esp value from the Trap Screen for esp. Select Process -> Open files the from Analyze menus Select Process -> Module Table from the Analyze menu. Select Save Output from the File menu and save the window contents to a file. If you don't understand what you are seeing, you will have to find someone to help you interpret the content of the dump file. Ask questions and let your helper guide you. Be prepared to spend some time working with your helper to understand the cause of the trap. The bare bones information you generated is just a starting point. It may or may not be sufficient to identify the source of the trap. Unless you are lucky, your helper will request additional output and may ask you generate another dump file using different settings. Often, your helper will to want you to send a copy of the dump file and the debug symbols. It is a good idea to keep notes describing how each dump file was generated and what you were doing when the dump file was generated. This is especially important for intermittent failures where one is looking for a pattern. Save the dump file, the debug symbols and your notes until analysis is complete. 5. Interpreting Process Dumps. --------------------------- This too is just the bare bones. The goal is to understand what the code is trying to do when the exception occurred, the data it was operating and what went wrong. Start with the Call Gate information. If the code is in a Call Gate, the kernel was processing some request when the exception occurred. This may or may not represent a kernel defect. The kernel attempts to validate any data is it passed, but in practice this is impossible. There are just too many variables. If the code is not in a Call Gate, the problem is within the application. The most common causes of application exceptions are buffer overruns or indexing errors. If the exception is in the kernel, look first at the Ring 0 Stack Trace for useful clues. The function names are often a good hint as to what the code is trying to do. If the exception is in the application, look first at the Ring3 Stack Trace for useful clues. Sometimes seeing the stack content in other formats can be useful. The dd command displays memory content as double words The db command displays memory content as bytes and characters. The ds command displays memory content as 0 terminated strings. Look for recognizable strings. At times you might want to display larger amounts of data. For example dd esp l400 displays 400 hex (i.e. 1024) double words starting at esp. 6. E-mailing Process Dumps. ------------------------ In general, don't e-mail a dump file to someone without asking them if they want you to send it. If you do need to e-mail the file, zip it up first. This will save transmit time and protect the dump file from corruption. It's always a good idea to give the zip file a useful name. Something like DavesTrapE_20040501.zip will help everyone remember what the zip file contains. It's also a good idea to include a note in the zip file describing how and why the trap occurred along with the bldlevel output. The zip file may get separated from the e-mail message. You should include the output of procdump query in your note. This will describe the type of data recorded in the dump file. If the zip file is over 5MB or so, check with your helper before sending the e-mail. You may need to use a file splitter and send each chunk in a separate e-mail giving your helper a chance to delete each e-mail from the server before you send the next chunk. Most ISPs limit e-mail inboxes to 10MB and unless you send the zip file in chunks, it will never get to your helper. If you can arrange to FTP the zip file to your helper, this is often a better solution. Be careful to send the e-mail containing the dump file only to the intended addressee. Sending a large, unexpected e-mail to all the members of a mailing list, some of whom may still be one dial up, is sure of upset someone. Good luck. Steven