- When we detect that we're stepping over a function call, also store
the CPU state at the time it occurred. This information is needed
in order to correctly reconstruct target addresses later since some
operands may be register-based.
- Add the aforementioned CPU state to CreateFrame()'s arguments and
adjust implementations and callers accordingly.
- ArchitectureX86 now hands off the work for GetInstructionInfo() to
DisassemblerX86, since the latter has all the information we need
to properly classify and evaluate instructions. Correspondingly a
CpuState is passed down to it in order to perform address calculations
for the instruction if it's a jump or call instruction. The latter's
targets are then stored on the thread for later retrieval when
constructing a stack trace. Adjust X86_64 accordingly for the
signature changes. This also fixes a bug where Step Over would
sometimes result in a Step Into instead due to the previous
implementation of GetInstructionInfo() occasionally failing to
classify call instructions correctly.
- Architecture::CreateStackTrace() now takes an argument specifying
the address of the last executed function if applicable. This is used
to decide who/where to decode a return value from. Adjust callers.
- DwarfImageDebugInfo::_CreateReturnValue() uses the above information
in order to know directly who the caller it needs to look up a return
value for is, rather than trying to walk backwards to find them.
Type resolution is now also a bit more sophisticated due to various
cases where the subprogram entry didn't directly contain the return
type but referred to another DIE that did. Retrieving return value
now appears to work properly in all cases except when position
independent code is involved. The latter however will require
resolving the appropriate function address in the PLT, which will
need some additional work.
- Thread now has a data member indicating if a subroutine was
executed during the last set of steps.
- ThreadHandler now sets the aforementioned state appropriate during
Step Over/Step Out.
- Architecture::CreateStackTrace() now takes a parameter indicating
whether return value retrieval is desired (based on aforementioned
thread value). Adjust callers accordingly.
- DwarfImageDebugInfo: If return value retrieval is requested, loop
backwards from the current IP to find the call instruction.
- CreateStackTrace() now takes a parameter indicating whether or not
to try and retrieve full frame information. This in turn is passed
on to SpecificImageDebugInfo, where e.g. DwarfImageDebugInfo can
use it to avoid constructing variables and parameters. This is
used by ThreadHandler since, when it requests the top frame for
its stepping calculations, this additional data/work is completely
unnecessary.
- Architecture::CreateStackTrace() now uses the last frame's
PreviousCpuState() as the basis to continue unwinding when passed
a partial trace to continue from, rather than the (incorrect) actual
cpu state of that frame, which would have resulted in the last frame
being duplicated in the trace.
- Renamed variables to be more clear.
CreateStackTrace() can now optionally be asked to limit the maximum
number of frames it tries to unwind. In conjunction, it can also be
passed an already existing partial stack trace, and be asked to
unwind more frames from it.
class in order to allow us to set up architecture specific default rules
for registers that aren't explicitly set by the CFI program.
- Set up default rule for EIP on x86.
- Theoretically set up a default rule for the stack pointer. Disabled for the
time being though until I determine why that rule's not yielding the expected
values for ESP in anything other than the top frame, though the other location
offset rules work as expected.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39816 a95241bf-73f2-0310-859d-f6bbb57e9c96
- Added a type constant describing the format of the register value.
- Added flag "is callee-preserved" indicating whether the ABI requires the
register to be preserved by a called function.
* CpuState: Added SetRegisterValue(), made GetRegisterValue() const.
* Added RegisterMap interface for mapping register indices between different
models.
* Architecture:
- Added CreateCpuState() version to create a clean CpuState.
- Added GetDwarfRegisterMaps(), which returns to RegisterMaps, converting
from and to DWARF register indices.
- Added ReadValueFromMemory() reading a value from the target team's memory.
The value type is specified by a type constant and the value return via a
BVariant.
- CreateStackTrace: No longer decide whether to adjust the instruction pointer
of the previous CPU state depending on who created the CPU state. Instead
compare it with the return address of the next frame. If they are equal it
obviously has to be adjusted.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31539 a95241bf-73f2-0310-859d-f6bbb57e9c96
FunctionInstance does now also have a (DisassembledCode) source code
attribute. Function keeps its attribute, but it explicitly is a
FileSourceCode now.
* SourceCode:
- Removed GetStatementAtLocation(). Instead DisassembledCode has a
StatementAtLocation() now. As well as a StatementAtAddress() and
StatementAddressRange(). Rather cast to the subclass (in two instances)
instead of having those methods in the base class. In most cases we already
have the subclasses now, anyway.
- Added Lock()/Unlock(), which are implemented in FileSourceCode. The
statement ranges are no longer immutable, so we have to lock.
* TeamDebugModel:
- Revived GetBreakpointsInAddressRange().
- GetBreakpointsForSourceCode(): Optimized for DisassembledCode and fixed
in the FileSourceCode case. We need to compare with the functions' source
file instead of their source code, since they might not have the source
code set yet. Fixed two instances of the same problem in SourceView. Setting
breakpoints in functions that have no associated source code yet, works now.
* Team:
- GetStatementAtAddress(): Optimized by using the DisassembledCode, if
available.
- GetStatementAtSourceLocation(): If the supplied source code is
DisassembledCode, we have to get the statement from it directly, since
we can't get that information from the image debug info.
* TeamDebugInfo: Added LoadSourceCode() and DisassembleFunction(), the new way
to get FileSourceCode respectively DisassembledCode. SpecificTeamDebugInfo
has lost LoadSourceCode() and gained service methods AddSourceCodeInfo() and
ReadCode(). This avoids unnecessary code duplication in the subclasses.
Moreover it allows for joining source location info source files from
different images (and compilation units) -- interesting for inline functions
in headers.
* Adjusted LoadSourceCodeJob and TeamDebugger::FunctionSourceCodeRequested()
accordingly.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31514 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Created "types" subdirectory for basic types and moved Types.h,
SourceLocation, TargetAddressRange there.
* Added TargetAddressRangeList, representing a list of address ranges.
* Array: Added copy constructor and assignment operator.
* Added DwarfFunctionDebugInfo.
* ElfFile: Also read the program headers and provide access to the segment
information.
* DWARF:
- Some work on DIECompileUnitBase and DIESubprogram to handle attributes we
need.
- Added DwarfUtils class which provides static utility methods. Currently some
to get DIE names. Only provisionally implemented yet.
- Read range list attribute values from the .debug_ranges section. Extended
AttributeValue to handle them correctly (ref-counting).
* DwarfImageDebugInfo:
- Implemented GetFunctions() for real, i.e. we return functions for all
subprogram debug info entries we find (those that refer to actual
functions, that is).
- Implemented the fallback part of LoadSourceCode() (reading the code from the
file and disassembling it).
Things should hopefully work as before, just a bit slower and with less accurate
function names, if DWARF debug info is available. Promising, eh? ;-)
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31321 a95241bf-73f2-0310-859d-f6bbb57e9c96
support for temporary breakpoints.
* TeamDebugger: No longer handle debug events in the listener thread. Instead
post a message to the looper thread. Makes the locking a bit easier.
* Architecture: Added virtual GetInstructionInfo() and GetStatement() returning
information on the instruction respectively a statement at a given address.
Implemented for x86.
* DebugInfo: Added virtual GetStatement() and implemented it for
DebuggerDebugInfo by means of using the Architecture.
* Implemented step over/into/out support. Works in principle, but has no
handling for PLTs yet, i.e. stepping into functions of other libraries
requires two steps ATM.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31244 a95241bf-73f2-0310-859d-f6bbb57e9c96
SourceView.
* Fixed the information flow problem in Architecture::CreateStackTrace()/
ArchitectureX86::UpdateStackCpuState() by introducing a virtual
UpdateStackFrameCpuState() which allows the architecture to update the CPU
state it generated before after the function the state belongs to is known.
That's where moving the instruction pointer to the previous instruction
happens now.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31229 a95241bf-73f2-0310-859d-f6bbb57e9c96
identification for x86.
* For non-top stack frames adjust the instruction pointer so that it points to
the calling function instead of the return address.
* SourceView: Consider a syscall frame a non-top frame (the arrow is drawn
differently).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31228 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Added x86 disassembler (via libudis86).
* Added Architecture::DisassembleCode() to disassemble a function to SourceCode.
* Added virtual DebugInfo::LoadSourceCode() to retrieve the source code for a
given function. The implementation in DebuggerDebugInfo disassembles the
function.
* Added source code info to StackFrame. Also added a listener mechanism to get
notified on source code changes.
* Added job to load the source code for a stack frame.
* Added (very basic) source code view and wired everything so that when a stack
frame is selected the source code (respectively disassembly) for the
underlying function is retrieved and shown.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31158 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Added the beginnings of the debug info abstraction. Currently we can only load
the symbols via the debugger.
* Added a job to retrieve debug info for an image. Extended the GetStackTraceJob
to support waiting for image debug info to be loaded.
* Extended ImageInfo by text/data address and size.
* Removed StackFrameX86 and made StackFrame a simple non-polymorphic class
featuring all the needed data. The really architecture-dependent is in the
referenced CpuState already. Added Image* and FunctionDebugInfo* attributes,
referring to the image respectively debug info for the function hit by the
instruction pointer.
* Switched StrackTrace's StackFrame management from DoublyLinkedList to
BObjectList. This makes it more comfortable to use.
* Changed the code for creating stack traces:
- The creation of the StackTrace object and the main loop to collect the
frames are now located in the no longer virtual
Architecture::CreateStackTrace().
- The decision how to create a StackFrame is based on the instruction pointer.
If it hit a function for which debug info is available, the respective
DebugInfo::CreateStackFrame() is used, otherwise we fall back to the new
virtual Architecture::CreateStackFrame().
* Adjusted the stack trace view to also show function names (mangled ATM).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31142 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Added a virtual Architecture::CreateStackTrace() and added a basic
implementation in ArchitectureX86. Fleshed out StackTrace/StackFrame a bit
and added StackFrameX86. This needs to be organized differently, though, so
that we can get the maximum available information for each stack frame,
depending on what info is available for the respective function.
* Added job to get the stack trace for a thread.
* Added stack trace related handling in TeamDebugger. Reorganized the thread
state/CPU state/stack trace change handling a bit -- we're using a
Team::Listener now, and do things asynchronously.
* Added a StackTraceView to display the stack trace of the current thread. No
function name available yet, otherwise working fine.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31126 a95241bf-73f2-0310-859d-f6bbb57e9c96
about what registers the architecture has and how to get them from the CpuState.
Implemented the respective x86 part.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31109 a95241bf-73f2-0310-859d-f6bbb57e9c96
