Take into account the operand and offset size as needed, otherwise
we'd potentially use the wrong value for our calculations.
Gets return values somewhat working on x86-64, though in some cases
looking up a PLT slot's destination fails, for as yet undetermined
reasons.
If the subroutine address in question required us to in a different
image, that lookup would overwrite our pointer to the starting image.
If said lookup then also failed, a crash would occur when performing
operations to look up the next return value, since the image pointer
would then be NULL.
DisassemblerX8664::GetNextInstructionInfo() was mistakenly truncating the
instruction address due to a copy-and-paste bug when it was initially
spawned off DisassemblerX86. This would result in the temporary breakpoints
for skipping over function calls never getting set properly, leading to us
stepping into the PLT on a step over.
Completely non-functional. Should be replaced by something visually
more unified. The selected package's title and icon should be always
visible above the three sections, and the sections should not be
tabs but something more subtle.
BColumnListView derived class with code-duplication from
DriveSetup that should be refactored eventually (BIconStringColumn
and BColumn implementation to show it). Does nothing except create
some columns.
devfs_io() can't fall back to calling vfs_synchronous_io(), if the
device driver doesn't support handling requests asynchronously. The
presence of the io() hook leads the VFS (do_iterative_fd_io()) to
believe that asynchronous handling is supported and set a
finished-callback on the request which calls the io() hook to start the
next chunk. Thus, instead of iterating through the request in a loop
the iteration happens recursively. For sufficiently fragmented requests
the stack may overflow (ticket #9900).
* Introduce a new vnode operation supports_operation(). It can be called
by the VFS to determine whether a present hook is actually currently
supported for a given vnode.
* devfs: implement the new hook and remove the fallback handling in
devfs_io().
* vfs_request_io.cpp: use the new hook to determine whether the io()
hook is really supported.
This is for the most part a direct copy of the equivalent x86 function
with adjustments to deal with the difference in address sizes and
registers. This gets architectural (aka unassisted) backtraces working
on x86-64, assuming the code in question has been built with frame pointers
enabled.
Thanks to Alex Smith for his assistance in implementing the
function prologue and syscall iframe detection.
Although syscalls are done through SYSCALL and therefore don't actually
have an interrupt number, set it to 99 (the syscall vector on 32-bit)
in the iframe so that a syscall frame can be identified. Also added
vector/error_code to x86_64_debug_cpu_state for Debugger to use, not
sure why I didn't put them there in the first place.
* The client should enter state REBINDING only when RENEWING times
out.
* When in RENEWING or REBINDING state do not give up retrying
unless the lease expires.
* Fix bug sending 2^n DHCP requests at n-th lease renewal.
* Use timeout values and renewal/rebinding times suggested by the
RFC.
* Use different XIDs in subsequent transactions.
This is a GNU extension for identifying the case of a template parameter
that is itself a template. We don't currently make use of it, but this
allows us to parse/skip it correctly.
With this change, webcore is loadable/debuggable on x86-64 when built
with -gdwarf-4, with the caveat that ~3-4GB of RAM are currently required.
- DIECompileUnit can also be marked with DW_AT_main_subprogram to
indicate that the main program resides within that particular unit.
- DIEBaseType and DIEMember now have the additional attribute
DW_AT_data_bit_offset which defines the offset in a somewhat different
manner than DWARF2/3's DW_AT_bit_offset, and deprecates the latter.
Unused/untested for the moment, as gcc 4.7.3 doesn't yet appear to be
generating either of the above attributes.
For return values too large to fit into EAX:EDX, the address at which
the value is located is stored in EAX. We were incorrectly assuming
they were at the current stack pointer, which isn't necessarily the
case depending on what optimization/code generation options were
used, leading to us sometimes showing the wrong values for such case.
The same issue showed up with debug binaries generated by Clang.
For DWARF4, gcc now sometimes emits member locations as constant
offsets from their parent rather than as location expressions.
We weren't handling this case properly, rather we were always
assuming that a constant offset implied we were dealing with a
bit field.
Fixes struct/class members not always having correct values on
version 4.
DWARF4 adds two new expression operands for values that are known, but
not located anywhere in memory (e.g. a known constant or a value computed
from a composite of other values/locations in the program). These are
DW_OP_implicit_value, which supplies the value directly as an LEB128 +
data block, and DW_OP_stack_value, which indicates the top of the
expression stack is the value.
Implements corresponding support in DwarfExpressionEvaluator,
ValueLocation and ValueLoader.
DIETypeUnit now inherits from DIECompileUnitBase in order to handle
things like setting the language info properly. This was previously
missing and would lead to occasional crashes with DWARF4.
Depending on how the type unit was actually set up hierarchically,
we sometimes ended up with the unit set to the wrong actual type
object. Instead, we now simply let the unit parse, and then look up
the entry by the offset specified in the header after all is said
and done. Fixes class/struct objects sometimes winding up resolving
to one of their members instead.
Some problems still remain with resolving variable values for such
types though, needs further investigation.
The DIEs generated by gcc sometimes pass a modified type for
DIEPointerToMember's containing type. As such, we need to resolve
the passed in type down to its base type before verifying that it's
indeed a compound type.
Type units need to be finished before compilation units to ensure that e.g.
modified types have their base types already set by the time we get to
compilation unit attributes.
Stephan Aßmus