The kernel profiles were recently changed to have a `strings` array
as part of the profile objects. The `ProfileViewer` now checks for
that during startup and declares the profile invalid if the array
is not present.
The UserspaceEmulator doesn't use the API which the kernel exposed
the string array for, so just fake it by always adding an empty array
to the generated profiles.
The fact that profiles are json on one giant line makes them very
difficult to debug when things go wrong. Instead make sure to wrap
each event or sample on a newline so you can easily grep/heap/tail
the profile files.
It was fragile to use the address of the body of the memory management
functions to disable memory auditing within them. Functions called from
these did not get exempted from the audits, so in some cases
UserspaceEmulator reported bogus heap buffer overflows.
Memory auditing did not work at all on Clang because when querying the
addresses, their offset was taken relative to the base of `.text` which
is not the first segment in the `R/RX/RW(RELRO)/RW(non-RELRO)` layout
produced by LLD.
Similarly to when setting metadata about the allocations, we now use the
`emuctl` system call to selectively suppress auditing when we reach
these functions. This ensures that functions called from `malloc` are
affected too, and no issues occur because of the inconsistency between
Clang and GCC memory layouts.
Previously, we pushed the old `eip` on the stack before reading the new
address, which made us jump to the wrong place if the destination was
relative to the `esp`.
When printing a backtrace, each library's base address is found by
walking through all memory regions in the coredump, and selecting the
address of the first region whose name begins with the library's soname.
This is done to support the Clang toolchain, where .text is not at
offset 0.
However, because the libraries loaded by the emulated process used the
same names, we could not distinguish those with the ones used by
UserspaceEmulator, so the backtrace ended up being garbage.
Using the libraries mapped by UE would not be a sufficient, as the
running application could ask for other libraries too, and doing away
with setting names would make debugging issues within UE code more
difficult.
... segment
This happens with binaries build with Clang or with a custom linker
script. If this is the case, offsets should be calculated not from the
base address of `.text`, but from the first section loaded for the
library.
This commit moves all UserspaceEmulator symbolication into a common
helper function and fixes a FIXME.
`ue --profile --profile-file ~/some-file.profile id` can now generate a
full profile (instruction-by-instruction, if needed), at the cost of not
being able to see past the syscall boundary (a.la. callgrind).
This makes it significantly easier to profile seemingly fast userspace
things, like Loader.so :^)
This implements almost all instructions related to the FPU, including
all MMX instructions as well.
A lot of these were copied and adjusted from the SoftCPU implementation.
The next big milestone would be QNan detection and ShadowValue handling.
Depending on the values it might be difficult to figure out whether a
value is decimal or hexadecimal. So let's make this more obvious. Also
this allows copying and pasting those numbers into GNOME calculator and
probably also other apps which auto-detect the base.
`fpu_get` returns a long double and `fpu_set` expects a long double as
its parameter, and the X87 FPU uses long doubles as its internal
storage, meaning the `FABS` operates on them. This means the correct
intrinsic function for implementing it is `__builtin_fabsl`.
The LexicalPath instance methods dirname(), basename(), title() and
extension() will be changed to return StringView const& in a further
commit. Due to this, users creating temporary LexicalPath objects just
to call one of those getters will recieve a StringView const& pointing
to a possible freed buffer.
To avoid this, static methods for those APIs have been added, which will
return a String by value to avoid those problems. All cases where
temporary LexicalPath objects have been used as described above haven
been changed to use the static APIs.
For now this only allows us to single-step through execution and inspect
part of the execution environment for debugging
This also allows to run to function return and sending signals to the VM
This changes the behavior of SIGINT for UE to pause execution and then
terminate if already paused
A way of setting a watchpoint for a function would be a good addition in
the future, the scaffold for this is already present, we only need to
figure out a way to find the address of a function
On a side note I have changed all occurences of west-const to east const
Doing these as custom classes might be faster, especially when writing
them in SSE, but this would cause a lot of Code duplication and due to
the nature of constexprs and the intelligence of the compiler they might
be using SSE/MMX either way
This adds __attribute__((used)) to the function declaration so the
compiler doesn't discard it. It also makes the function NEVER_INLINE
so that we don't end up with multiple copies of the function. This
is necessary because the function uses inline assembly to define some
unique labels.
We had two functions for doing mostly the same thing. Combine both
of them into String::find() and use that everywhere.
Also add some tests to cover basic behavior.
Previously each malloc size class would keep around a limited number of
unused blocks which were marked with MADV_SET_VOLATILE which could then
be reinitialized when additional blocks were needed.
This changes malloc() so that it also keeps around a number of blocks
without marking them with MADV_SET_VOLATILE. I termed these "hot"
blocks whereas blocks which were marked as MADV_SET_VOLATILE are called
"cold" blocks because they're more expensive to reinitialize.
In the worst case this could increase memory usage per process by
1MB when a program requests a bunch of memory and frees all of it.
Also, in order to make more efficient use of these unused blocks
they're now shared between size classes.
Problem:
- Default destructors (and constructors) are in `.cpp` files. This
prevents the compiler's optimizer from inlining them when it thinks
inlining is appropriate (unless LTO is used).
- Forward declarations can prevent some optimizations, such as
inlining of constructors and destructors.
Solution:
- Remove them or set them to `= default` and let the compiler handle
the generation of them.
- Remove unneeded forward declarations.
Problem:
- `size_classes` is a C-style array which makes it difficult to use in
algorithms.
- `all_of` algorithm is re-written for the specific implementation.
Solution:
- Change `size_classes` to be an `Array`.
- Directly use the generic `all_of` algorithm instead of
reimplementing.
When the user specifies a path such as ./test we'd incorrectly look for
the binary in the PATH environment variable and end up executing an
incorrect binary (e.g. /bin/test). We should only look up binaries in
PATH if the user-specified path does not contain a slash.
