For unwinding purposes, we don't care about unsupported registers. Yet because
we added these rules to the cache entry, we'd later try to evaluate them and
thus fail the unwind attempt for no good reason. They'd also take up cache rule
slots that would be better spent on actually relevant registers.
Note that any attempt to read unsupported registers during unwinding will still
fail the unwind attempt as expected.
There were only a few dozen lines of common logic, and they frankly
introduced more complexity than they eliminated. Instead, let's accept
that the implementations of `SelfInfo` are all pretty different and want
to track different state. This probably fixes some synchronization and
memory bugs by simplifying a bunch of stuff. It also improves the DWARF
unwind cache, making it around twice as fast in a debug build with the
self-hosted x86_64 backend, because we no longer have to redundantly go
through the hashmap lookup logic to find the module. Unwinding on
Windows will also see a slight performance boost from this change,
because `RtlVirtualUnwind` does not need to know the module whatsoever,
so the old `SelfInfo` implementation was doing redundant work. Lastly,
this makes it even easier to implement `SelfInfo` on freestanding
targets; there is no longer a need to emulate a real module system,
since the user controls the whole implementation!
There are various other small refactors here in the `SelfInfo`
implementations as well as in the DWARF unwinding logic. This change
turned out to make a lot of stuff simpler!
Apparently the `__eh_frame` in Mach-O binaries doesn't include the
terminator entry, but in all other respects it acts like `.eh_frame`
rather than `.debug_frame`. I have no idea.
By my estimation, these changes speed up DWARF unwinding when using the
self-hosted x86_64 backend by around 7x. There are two very significant
enhancements: we no longer iterate frames which don't fit in the stack
trace buffer, and we cache register rules (in a fixed buffer) to avoid
re-parsing and evaluating CFI instructions in most cases. Alongside this
are a bunch of smaller enhancements, such as pre-caching the result of
evaluating the CIE's initial instructions, avoiding re-parsing of CIEs,
and big simplifications to the `Dwarf.Unwind.VirtualMachine` logic.
This logic was causing some occasional infinite looping on ARM, where
the `.debug_frame` section is often incomplete since the `.exidx`
section is used for unwind information. But the information we're
getting from the compiler is totally *valid*: it's leaving the rule as
the default, which is (as with most architectures) equivalent to
`.undefined`!
This has been a TODO for ages, but in the past it didn't really matter
because stack traces are typically printed to stderr for which a mutex
is held so in practice there was a mutex guarding usage of `SelfInfo`.
However, now that `SelfInfo` is also used for simply capturing traces,
thread safety is needed. Instead of just a single mutex, though, there
are a couple of different mutexes involved; this helps make critical
sections smaller, particularly when unwinding the stack as `unwindFrame`
doesn't typically need to hold any lock at all.
Calling `current` here causes compilation failures as the C backend
currently does not emit valid MSVC inline assembly. This change means
that when building for MSVC with the self-hosted C backend, only FP
unwinding can be used.
...and just deal with signal handlers by adding 1 to create a fake
"return address". The system I tried out where the addresses returned by
`StackIterator` were pre-subtracted didn't play nicely with error
traces, which in hindsight, makes perfect sense. This definition also
removes some ugly off-by-one issues in matching `first_address`, so I do
think this is a better approach.
This crash exists on master, and seems to have existed since 2019; I
think it's just very rare and depends on the exact binary generated. In
theory, a stream block should always be a "data" block rather than a FPM
block; the FPMs use blocks `1, 4097, 8193, ...` and `2, 4097, 8194, ...`
respectively. However, I have observed LLVM emitting an otherwise valid
PDB which maps FPM blocks into streams. This is not a bug in
`std.debug.Pdb`, because `llvm-pdbutil` agrees with our stream indices.
I think this is arguably an LLVM bug; however, we don't really lose
anything from just weakening this check. To be fair, MSF doesn't have an
explicit specification, and LLVM's documentation (which is the closest
thing we have) does not explicitly state that FPM blocks cannot be
mapped into streams, so perhaps this is actually valid.
In the rare case that LLVM emits this, previously, stack traces would
have been completely useless; now, stack traces will work okay.
Mostly on macOS, since Loris showed me a not-great stack trace, and I
spent 8 hours trying to make it better. The dyld shared cache is
designed in a way which makes this really hard to do right, and
documentation is non-existent, but this *seems* to work pretty well.
I'll leave the ruling on whether I did a good job to CI and our users.
Our usage of `ucontext_t` in the standard library was kind of
problematic. We unnecessarily mimiced libc-specific structures, and our
`getcontext` implementation was overkill for our use case of stack
tracing.
This commit introduces a new namespace, `std.debug.cpu_context`, which
contains "context" types for various architectures (currently x86,
x86_64, ARM, and AARCH64) containing the general-purpose CPU registers;
the ones needed in practice for stack unwinding. Each implementation has
a function `current` which populates the structure using inline
assembly. The structure is user-overrideable, though that should only be
necessary if the standard library does not have an implementation for
the *architecture*: that is to say, none of this is OS-dependent.
Of course, in POSIX signal handlers, we get a `ucontext_t` from the
kernel. The function `std.debug.cpu_context.fromPosixSignalContext`
converts this to a `std.debug.cpu_context.Native` with a big ol' target
switch.
This functionality is not exposed from `std.c` or `std.posix`, and
neither are `ucontext_t`, `mcontext_t`, or `getcontext`. The rationale
is that these types and functions do not conform to a specific ABI, and
in fact tend to get updated over time based on CPU features and
extensions; in addition, different libcs use different structures which
are "partially compatible" with the kernel structure. Overall, it's a
mess, but all we need is the kernel context, so we can just define a
kernel-compatible structure as long as we don't claim C compatibility by
putting it in `std.c` or `std.posix`.
This change resulted in a few nice `std.debug` simplifications, but
nothing too noteworthy. However, the main benefit of this change is that
DWARF unwinding---sometimes necessary for collecting stack traces
reliably---now requires far less target-specific integration.
Also fix a bug I noticed in `PageAllocator` (I found this due to a bug
in my distro's QEMU distribution; thanks, broken QEMU patch!) and I
think a couple of minor bugs in `std.debug`.
Resolves: #23801Resolves: #23802
Turns out that RtlCaptureStackBackTrace is actually just doing FP (ebp)
unwinding under the hood, making this logic completely redundant with
our own FP-walking implementation; see added comment for details.
Previously, the `test-stack-traces` step was essentially just testing
error traces, and even there we didn't have much coverage. This commit
solves that by splitting the "stack trace" tests into two separate
harnesses: the "stack trace" tests are for actual stack traces (i.e.
involving stack unwinding), while the "error trace" tests are
specifically for error return traces.
The "stack trace" tests will test different configurations of:
* `-lc`
* `-fPIE`
* `-fomit-frame-pointer`
* `-fllvm`
* unwind tables (currently disabled)
* strip debug info (currently disabled)
The main goal there is to test *stack unwinding* under different
conditions. Meanwhile, the "error trace" tests will test different
configurations of `-O` and `-fllvm`; the main goal here, aside from
checking that error traces themselves do not miscompile, is to check
whether debug info is still working even in optimized builds. Of course,
aggressive optimizations *can* thwart debug info no matter what, so as
before, there is a way to disable cases for specific targets / optimize
modes.
The program which converts stack traces into a more validatable format
by removing things like addresses (previously `check-stack-trace.zig`,
now `convert-stack-trace.zig`) has been rewritten and simplified. Also,
thanks to various fixes in this branch, several workarounds have become
unnecessary: for instance, we don't need to ignore the function name
printed in stack traces in release modes, because `std.debug.Dwarf` now
uses the correct DIE for inlined functions!
Neither `test-stack-traces` nor `test-error-traces` does general foreign
architecture testing, because it seems that (at least for now) external
executors often aren't particularly good at handling stack tracing
correctly (looking at you, Wine). Generally, they just test the native
target (this matches the old behavior of `test-stack-traces`). However,
there is one exception: when on an x86_64 or aarch64 host, we will also
test the 32-bit version (x86 or arm) if the OS supports it, because such
executables can be trivially tested without an external executor.
Oh, also, I wrote a bunch of stack trace tests. Previously there was,
erm, *one* test in `test-stack-traces` which wasn't for error traces.
Now there are a good few!
Alex Rønne Petersen