Hi Vinod,
EDC was written to support up to Dwarf version 3. At the time version 4 had not been released. I’m not sure if the bug you’re seeing is related specifically to version 4, or if it’s just something unexpected in the general dwarf info. My guess would be the latter. Which compiler are you using? Some build tools let you choose which debug format to generate with a switch. That could be a workaround.
Thanks,
Warren
Hi Warren,
It’s very sad to say that myself again stuck with the Dwarf stuff. This time it’s related to the Dwarf version 4. I found that my code, which using EDC Dwarf reader not working properly with Dwarf 4 (it’s working perfectly with Dwarf version 2). Especially it’s not getting any info about ‘C’ files but it’s working fine with ‘asm’ files. So my question is simple, does the EDC Dwarf reader support Dwarf version 4. If so, is there any API change specific to Dwarf 4.
I believe this is because GCC generates absolute paths in Dwarf on Windows without the drive letter in many cases. You could argue that it shouldn’t when not generated on Windows, or be controlled by a pref. You can probably log a bug for it, but currently there are no committers and it’s not being built as part of CDT. For now you’ll have to fix it locally and do your own build.
Thanks,
Warren
Hey Warren,
The line info provider stuff worked fine. Think that we can proceed with it. Thanks for your help. J
Here I would like to discuss about one more issue related to dwarf reader. What I have done so far to get this issue is described below.
=>
I have a vmlinux elf, which was compiled in the target file system (/home/somedirectory). I have used it to get the source file information using DwarfDebugInfoProvider. getSourceFiles API on a windows machine (copied the vmlinux.elf to some folder in the D drive), I got source file path like D:/home/somedirectory. Which is not expected.
=>
On further investigation, I found that it’s the problem with DwarfFileHelper class and its normalizeFilePath API. In this API, there were some checks like HostOS.IS_WIN32, and which makes the issue.
Is it a valid scenario?
Is there any further developments ongoing with EDC dwarf reader?
Can I report a bug and submit a patch?
It’s possible there’s a memory leak in there somewhere. I don’t see how this snippet is doing what you listed as your requirements though. I think you’d want to get the line info provider for each compile unit for 1/2/3. For 4 you can simply get symbol at address (if any). That will work for code, not data. There is no correspondence between variable declaration/definition and source line.
Hi Warren,
I have changed my VM args as you mentioned below. But the code snippet given below is still causing the same problem L.
Can you please check whether the usage of API’s in the snippet is correct or not (Please take a trial run).
Hi Vinod,
It’s strange that you’d run out of memory parsing a 92MB file, as we were parsing files ~1GB. Check your vm heap arguments. I assume you’re either running/debugging this from Eclipse. If so, edit the VM args in the arguments tab of your launch config to have something like –Xms256m -Xmx512m.
Thanks,
Warren
Hi Warren,
Sorry to say that still I got some trouble with dwarf reader stuff. This time we need to parse VMLINUX image (92 MB) as a part of adding Linux support.
Actually I am interested on following things.
1. Source file names
2. Address of each source line inside each source file
3. Line number
4. Symbol information in that line , if it is present ( function, global variable etc.) and its type
I am using the repository @ eclipse/cdt.edc in GITHUB. Still I am doing some tests with LineInfoProvider and other stuffs, which I got from the Test cases in the repository.
Please see the code snippet I have used .
Path modulePath = new Path("D:/vinod/vmlinux");//Linux kernal path
ElfExecutableSymbolicsReaderFactory fact = new ElfExecutableSymbolicsReaderFactory();
IExecutableSymbolicsReader exeReader = fact.createExecutableSymbolicsReader(modulePath);
DwarfDebugInfoProvider dip = new DwarfDebugInfoProvider(exeReader);
Map<String, List<PublicNameInfo>> variables = dip.getPublicVariables();
final Iterator<Entry<String, List<PublicNameInfo>>> it = variables.entrySet().iterator();
while (it.hasNext()) {
IPath file = null;
Map.Entry pairs = (Map.Entry) it.next();
final String variableName = (String) pairs.getKey();
IModuleScope moduleScope = dip.getModuleScope();
if(null != moduleScope){
moduleScope.getVariablesByName(variableName, false);
//moduleScope.get
Collection<ISymbol> unMangledSymbols = dip.getExecutableSymbolicsReader().findUnmangledSymbols(variableName);
for(ISymbol symbol: unMangledSymbols){
System.out.println(symbol.getName() + symbol.getAddress());
}
}
}
I got the following Exception ,
Exception in thread "main" java.lang.OutOfMemoryError: Java heap space
at org.eclipse.cdt.debug.edc.internal.symbols.dwarf.DwarfInfoReader.parseCompilationUnitForAddressesPrivate(DwarfInfoReader.java:853)
at org.eclipse.cdt.debug.edc.internal.symbols.dwarf.DwarfInfoReader.parseCompilationUnitForAddresses(DwarfInfoReader.java:870)
at org.eclipse.cdt.debug.edc.internal.symbols.dwarf.DwarfDebugInfoProvider.getVariablesByName(DwarfDebugInfoProvider.java:961)
at org.eclipse.cdt.debug.edc.internal.symbols.dwarf.DwarfModuleScope.getVariablesByName(DwarfModuleScope.java:134)
at MemmoryLeakMain.main(MemmoryLeakMain.java:45)
Expecting a swift response. J
No problem. I’m glad it helped.
This might be a good time to ask the list if anyone else out there is using EDC? It sort of died when the Nokia group fell apart, but we’ve been quietly resurrecting it here at Silicon Labs (with some of the same team from Nokia and former CDT committers). At some point we’re planning on contributing back our changes and possibly volunteering some committers to help maintain it going forward.
Doug, I’ve been meaning to talk to you about this but have just been too swamped.
Anyway, I’m just gauging interest at this point.
Thanks,
Warren
Hi Warren,
Thanks for the response, it’s a cool one and help me to solve the problems.
Thanks,
Vinod
Hi Vinod,
If this is a pure embedded target where the executable will not be relocated, then you can use the symbol table. See org.eclipse.cdt.debug.edc.symbols.IExecutableSymbolicsReader.findSymbols(String) or org.eclipse.cdt.debug.edc.symbols.IExecutableSymbolicsReader.findUnmangledSymbols(String).
Otherwise you’ll the module to get the relocated address. See org.eclipse.cdt.debug.edc.symbols.IModuleScope.getVariablesByName(String, boolean). Once you have the IVariable you’re looking for, you can resolve the runtime address by getting the ILocationProvider, then the IVariableLocation, then the IAddress.
I hope that helps.
Thanks,
Warren
Hi All,
We are facing a problem with getting address of global variables using Dwarf Reader (from EDC project).
I can’t find a API which gives the constant address for a global variable (I have checked with Variable.java ).
Hope somebody can give a better option to get the same without any side effects.
Thanks in Advance,
Vinod
