[lldb][swift] Fix unwinding of Q funclets by comparing PC to continuation ptr #9574
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…tion ptr Unwinding these funclets is tricky because they change, halfway through the function, the meaning of the async register. In particular, the meaning changes from: a) "the async context to be freed" [of the async function that just finished executing] To: b) "the async context of the current async function" [which has just been resumed]. LLDB has no way of identifying the instruction in which this transition happened. This patch improves the situation slightly by employing a heuristic: if the async register has a continuation pointer that is equal to the currently executing funclet, then the stop is _before_ the transition point. This heuristic fails on some recursive async functions. An alternative approach involves assuming we will never stop between the end of the prologue and the transition point, so that LLDB may always unwind assuming it is past the transition point if the stop is outside the prologue. This has the benefit that it should "always work", including in recursive funclets. However, it is difficult to reason about where a debugger may stop; in fact, very often the transition point is the second instruction after the prologue, which will trigger the fail point in a common scenario. Breakpoints are often placed in the _first_ instruction after the prologue, and any kind of step operation will first "instruction step" over the breakpoint location, placing the debugger exactly in the "incorrect unwinding" zone. A correct implementation will likely require more guarantees from the compiler. See rdar://139676623
felipepiovezan
force-pushed
the
felipe/improve_q_funclet_unwinding
branch
from
62d943a to
2451d19
Compare
|
@swift-ci test |
adrian-prantl
approved these changes
Nov 12, 2024
| /// FIXME: this fails in some recursive async functions. See: rdar://139676623 | ||
| static llvm::Expected<bool> IsIndirectContext(Process &process, | ||
| StringRef mangled_name, | ||
| addr_t async_reg, |
There was a problem hiding this comment.
Just out of curiosity: what is this? Is it a DWARF register number stored in an addr_t?
There was a problem hiding this comment.
it is the contents of the async reg. I hesitated to call it "async context" because we don't know whose context this is (this is what the function is indirectly determining)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Unwinding these funclets is tricky because they change, halfway through the function, the meaning of the async register. In particular, the meaning changes from:
a) "the async context to be freed" [of the async function that just finished executing]
To:
b) "the async context of the current async function" [which has just been resumed].
LLDB has no way of identifying the instruction in which this transition happened.
This patch improves the situation slightly by employing a heuristic: if the async register has a continuation pointer that is equal to the currently executing funclet, then the stop is before the transition point. This heuristic fails on some recursive async functions.
An alternative approach involves assuming we will never stop between the end of the prologue and the transition point, so that LLDB may always unwind assuming it is past the transition point if the stop is outside the prologue. This has the benefit that it should "always work", including in recursive funclets.
However, it is difficult to reason about where a debugger may stop; in fact, very often the transition point is the second instruction after the prologue, which will trigger the fail point in a common scenario. Breakpoints are often placed in the first instruction after the prologue, and any kind of step operation will first "instruction step" over the breakpoint location, placing the debugger exactly in the "incorrect unwinding" zone.
A correct implementation will likely require more guarantees from the compiler. See rdar://139676623