Skip to content

Commit

Permalink
Merge pull request #15153 from keithc-ca/elf
Browse files Browse the repository at this point in the history 
Only add memory segments from core file to process view
  • Loading branch information
@pshipton
pshipton committed May 31, 2022
2 parents 3a853c2 + c2a1871 commit eeedfa4
Show file tree
Hide file tree
Showing 2 changed files with 71 additions and 93 deletions.
136 changes: 70 additions & 66 deletions debugtools/DDR_VM/src/com/ibm/j9ddr/corereaders/elf/ELFDumpReader.java
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*******************************************************************************
* Copyright (c) 2009, 2021 IBM Corp. and others
* Copyright (c) 2009, 2022 IBM Corp. and others
*
* This program and the accompanying materials are made available under
* the terms of the Eclipse Public License 2.0 which accompanies this
Expand Down Expand Up @@ -457,9 +457,8 @@ private void createAllModules(ELFFileReader executableELF, String executableName
* section header table from that (as well as the symbols as those sections may not be loaded either).
*/
List<IModule> allModules = new LinkedList<>();
long executableBaseAddress = executableELF.getBaseAddress();

Map<Long, ELFFileReader> inCoreReaders = getElfReadersForModulesWithinCoreFile(executableELF, executableName);
Map<Long, ELFFileReader> inCoreReaders = getElfReadersForModulesWithinCoreFile(executableName);
Map<Long, String> libraryNamesFromDebugData = readLibraryNamesFromDebugData(executableELF, inCoreReaders,
executableELF);

Expand Down Expand Up @@ -497,7 +496,7 @@ private void createAllModules(ELFFileReader executableELF, String executableName
// TODO - Sort modules. For the sake of tidiness.
_modules.addAll(allModules);
ELFFileReader readerForExectuableOnDiskOrAppended = getReaderForModuleOnDiskOrAppended(executableName);
_executable = createModuleFromElfReader(executableBaseAddress, executableName, executableELF,
_executable = createModuleFromElfReader(0, executableName, executableELF,
readerForExectuableOnDiskOrAppended);
}

Expand Down Expand Up @@ -531,20 +530,19 @@ private ELFFileReader getELFReaderForExecutableWithinCoreFile() {
*
* @return map BaseAddress -> ELFFileReader
*/
private TreeMap<Long, ELFFileReader> getElfReadersForModulesWithinCoreFile(ELFFileReader executableELF,
String executableName) {
private TreeMap<Long, ELFFileReader> getElfReadersForModulesWithinCoreFile(String executableName) {
TreeMap<Long, ELFFileReader> moduleReadersByBaseAddress = new TreeMap<>();
// System.out.println("dumping program header table");
// for (ProgramHeaderEntry entry : _reader.getProgramHeaderEntries()) {
// System.out.printf("offset %x address %x, type %d\n",entry.fileOffset,entry.virtualAddress, entry._type);
// System.out.printf("offset %x address %x, type %d\n", entry.fileOffset, entry.virtualAddress, entry._type);
// }

for (ProgramHeaderEntry entry : _reader.getProgramHeaderEntries()) {
if (!entry.isLoadable() || entry.fileSize == 0) {
if (entry.fileSize == 0 || !entry.isLoadable()) {
continue;
}
try {
// System.err.printf("Getting internal reader for %s, virtual address 0x%x, file offset 0x%x\n", executableName, entry.virtualAddress, entry.fileOffset );
// System.err.printf("Getting internal reader for %s, virtual address 0x%x, file offset 0x%x\n", executableName, entry.virtualAddress, entry.fileOffset);
ELFFileReader loadedElf = getReaderForSegment(_reader.getStream(), entry);
if (loadedElf != null) {
moduleReadersByBaseAddress.put(entry.virtualAddress, loadedElf);
Expand Down Expand Up @@ -728,12 +726,12 @@ private Map<Long, String> readLibraryNamesFromDebugData(ELFFileReader executable
* and construct a Module object. The ELF reader may point to a segment within the core file
* or may point to a copy of the module on disk or appended to the core file.
*
* @param loadedBaseAddress of the module
* @param loadedBaseAddress of the module or 0 for a non-loaded module
* @param name of the module
* @param elfReader to the module
* @param inCoreReader to the module
* @param diskReader to the module
*
* @return IModule
* @throws IOException
*/
private IModule createModuleFromElfReader(final long loadedBaseAddress, String name, ELFFileReader inCoreReader,
ELFFileReader diskReader) {
Expand All @@ -743,9 +741,9 @@ private IModule createModuleFromElfReader(final long loadedBaseAddress, String n
if (inCoreReader == null) {
return new MissingFileModule(_process, name, Collections.<IMemoryRange>emptyList());
}
List<? extends ISymbol> symbols = null;
Map<Long, String> sectionHeaderStringTable = null;
List<SectionHeaderEntry> sectionHeaderEntries = null;
List<? extends ISymbol> symbols;
Map<Long, String> sectionHeaderStringTable;
List<SectionHeaderEntry> sectionHeaderEntries;
Properties properties;
Collection<? extends IMemorySource> declaredRanges;

Expand Down Expand Up @@ -805,66 +803,72 @@ private IModule createModuleFromElfReader(final long loadedBaseAddress, String n
return null;
}

// Of the declared memory ranges, some will already be in core (i.e. .data) others will have been declared
// in the core, but not backed.
List<IMemoryRange> ranges = new ArrayList<>(declaredRanges.size());

for (IMemorySource source : declaredRanges) {
IMemorySource coreSource = _process.getRangeForAddress(source.getBaseAddress());

/**
* The following test skips sections that originally had an address of 0
* the section header table.
*
* Explanation follows - see also the example section header table at the top
* of SectionHeaderEntry.java
*
* Some of the later sections in the section header table have an address field
* 0. That is a relative address, relative to the base address of the module.
* They are usually the sections from .comment onwards.
*
* When the entry is constructed the code that creates the entry creates it with
* address (base address of the module) + (address field in the SHT) hence those
* that had an address of 0 will have an address == base address of the module.
* These entries are precisely those that are often not in the core file so it is
* not safe to create them as sections - they were there in the on-disk version of
* the library but often not in memory. The code above that called elfReader.getMemoryRanges
* may have been reading the version of the module from disk (which is good, it means
* you get a good section header string table so you get good names for all the
* sections) but not all the sections exist in memory. The danger of adding them
* as memory ranges is that they start to overlay the segments in the core file
* that come immediately afterwards; that is, they cause jdmpview, for example, to
* believe that the contents of these later areas of memory are backed by the
* contents of the library on disk when they are not.
* See CMVC 185753
*
* So, don't add sections that originally had address 0.
*
*/
if (source.getBaseAddress() == loadedBaseAddress) { // must have originally had address 0 in the section header table
continue;
}
List<IMemoryRange> ranges;

if (loadedBaseAddress == 0) {
ranges = new ArrayList<>(declaredRanges);
} else {
ranges = new ArrayList<>(declaredRanges.size());

// Of the declared memory ranges, some will already be in core (i.e. .data),
// others will have been declared in the core, but not backed.
for (IMemorySource source : declaredRanges) {
IMemorySource coreSource = _process.getRangeForAddress(source.getBaseAddress());

/**
* The following test skips sections that originally had an address of 0
* the section header table.
*
* Explanation follows - see also the example section header table at the top
* of SectionHeaderEntry.java
*
* Some of the later sections in the section header table have an address field
* 0. That is a relative address, relative to the base address of the module.
* They are usually the sections from .comment onwards.
*
* When the entry is constructed the code that creates the entry creates it with
* address (base address of the module) + (address field in the SHT) hence those
* that had an address of 0 will have an address == base address of the module.
* These entries are precisely those that are often not in the core file so it is
* not safe to create them as sections - they were there in the on-disk version of
* the library but often not in memory. The code above that called elfReader.getMemoryRanges
* may have been reading the version of the module from disk (which is good, it means
* you get a good section header string table so you get good names for all the
* sections) but not all the sections exist in memory. The danger of adding them
* as memory ranges is that they start to overlay the segments in the core file
* that come immediately afterwards; that is, they cause jdmpview, for example, to
* believe that the contents of these later areas of memory are backed by the
* contents of the library on disk when they are not.
* See CMVC 185753
*
* So, don't add sections that originally had address 0.
*
*/
if (source.getBaseAddress() == loadedBaseAddress) { // must have originally had address 0 in the section header table
continue;
}

if (null != coreSource) {
if (coreSource.isBacked()) {
// Range already exists
if (null != coreSource) {
if (coreSource.isBacked()) {
// Range already exists
} else {
// The range exists but isn't backed. We need to replace the in-core memory range with the one
// from the library
if (source.getSize() > 0) {
_process.removeMemorySource(coreSource);
_process.addMemorySource(source);
}
}
} else {
// The range exists but isn't backed. We need to replace the in-core memory range with the one
// from the library
if (source.getSize() > 0) {
_process.removeMemorySource(coreSource);
_process.addMemorySource(source);
}
}
} else {
if (source.getSize() > 0) {
_process.addMemorySource(source);
}
ranges.add(source);
}
ranges.add(source);
}
return new Module(_process, name, symbols, ranges, loadedBaseAddress, properties);

return new Module(_process, name, symbols, ranges, loadedBaseAddress, properties);
}

private boolean isValidAddress(long address) {
Expand Down
28 changes: 1 addition & 27 deletions debugtools/DDR_VM/src/com/ibm/j9ddr/corereaders/elf/ELFFileReader.java
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*******************************************************************************
* Copyright (c) 2004, 2020 IBM Corp. and others
* Copyright (c) 2004, 2022 IBM Corp. and others
*
* This program and the accompanying materials are made available under
* the terms of the Eclipse Public License 2.0 which accompanies this
Expand Down Expand Up @@ -748,32 +748,6 @@ public String readString() {
}
}

/**
* Get the base address of an executable or library, given the ELF file.
* This is the lowest virtual address of the Program Header Table entries of type PT_LOAD, which
* is probably the first but we search them all just in case.
* The following is summarised from the System V Application Binary Interface,
* gabi41.pdf, section Base Address
*
* "... to compute the base address, one determines the memory address associated with
* the lowest p_vaddr value for a PT_LOAD segment. This address is truncated ..."
*
* The truncation is to do with page sizes and in practice not found to be needed.
*
* @return base address for the executable or library
*/
public long getBaseAddress() {
long lowestVirtualAddressSoFar = Long.MAX_VALUE;
for (ProgramHeaderEntry entry : getProgramHeaderEntries()) {
if (entry.isLoadable()) {
if (entry.virtualAddress < lowestVirtualAddressSoFar) {
lowestVirtualAddressSoFar = entry.virtualAddress;
}
}
}
return lowestVirtualAddressSoFar;
}

/**
* Search the program header table for the dynamic entry. There should be only one of these
* Typically it is within the first few entries, often the third, so this is not expensive
Expand Down

0 comments on commit eeedfa4

Please sign in to comment.