ARM64, x86 (64-bit), and x86 (32-bit) each have different alignment
requirements, so this change ensures that consistent data layout is
used for packing and unpacking.
Some types (notably size_t on 32-bit) have different sizes on the guest than on
the host. This template function must be aware of these differences, so a
second parameter list with fixed-size types must be provided to describe the
guest types.
Note that this information can't be queried through type traits: To a C++
compiler, size_t is indistuingishable from uint64_t. For this reason, the
correct guest type must indeed be provided externally.
This can be used to allow automatically handling structures that require
special behavior for one member but are automatically repackable otherwise.
The feature is enabled using the new custom_repack annotation and requires
additional repacking functions to be defined in the host file for each
customized member.
Pointer types inherently cause data layout compatibility issues, so they're
worth special-casing here. The wrappers will type-pun pointers to 32-bit or
64-bit integers (matching the guest architecture) to avoid direct host-side
use of guest pointers without consideration.
The guest_layout wrapper provides an architecture-agnostic representation of
the guest data layout of each struct used in a thunked library. A constructor
is added to host_layout to allow conversion of the data to the host layout.
For types that are already fully compatible, both layout wrappers are simple
type aliases to minimize overhead.
This allows future changes to emit interdependent helper structures in the
same order.
The sort algorithm was chosen for simplicity rather than performance. It's
fast enough in practice even for APIs as large as Vulkan.
These annotations allow for a given type or parameter to be treated as
"compatible" even if data layout analysis can't infer this automatically.
assume_compatible_data_layout is more powerful than is_opaque, since it
allows for structs containing members of a certain type to be automatically
inferred as "compatible".
Conversely however, is_opaque enforces that the underlying data is never
accessed directly, since non-pointer uses of the type would still be
detected as "incompatible".
This annotation can be used for data types that can't be repacked
automatically even with custom repack annotations. With ptr_passthrough,
the types are wrapped in guest_layout and passed to the host like that.
Previously, two functions with the same signature would always be wrapped
in the same logic. This change allows customizing one function with
annotations while leaving the other one unchanged.
This runs the data layout analysis pass added in the previous change twice:
Once for the host architecture and once for the guest architecture. This
allows the new DataLayoutCompareAction to query architecture differences for
each type, which can then be used to instruct code generation accordingly.
Currently, type compatibility is classified into 3 categories:
* Fully compatible (same size/alignment for the type itself and any members)
* Repackable (incompatibility can be resolved with emission of automatable
repacking code, e.g. when struct members are located at differing offsets
due to padding bytes)
* Incompatible
1) The host library needs to be loaded in the global namespace.
2) We need to use `RTLD_DEFAULT` instead of querying the object
directly.
We need to load the host library in the global namespace so the symbols
end up in the global symbol table. This follows how all these symbols
/usually/ get loaded. Either by linking directly to the library or how
loaders will end up loading these.
We need to use RTLD_DEFAULT to follow symbol overriding rules that tend
to occur. For example, MangoHUD will LD_PRELOAD a library that provides
GLX and EGL symbols. Which FEX's thunk libraries need to pick up this
override.
If we are querying the host library directly then we fail to pickup
these overrides, thus breaking MangoHUD and other overlays.
ASTVisitor is great for iterating over AST nodes by type, but most of our
analysis is based on symbol names. For this task, a lookup in DeclContexts
after parsing is complete is better suited.
This changes how host trampolines for guest functions are created. Instead
of doing this purely on the guest-side, it's either the host-side that
creates them in a single step *or* a cooperative two-step initialization
process must be used. In the latter, trampolines are allocated and partially
initialized on the guest and must be finalized on the host before use.
sunshineinabox