summaryrefslogtreecommitdiff
path: root/rts
diff options
context:
space:
mode:
authorÖmer Sinan Ağacan <omeragacan@gmail.com>2019-01-10 11:55:25 +0300
committerÖmer Sinan Ağacan <omeragacan@gmail.com>2019-01-12 10:11:19 +0300
commit74cd4ec5d2f9321aad5db3285cb60d78f2562996 (patch)
tree8c2cec2852dd719a97edf8a8345ea4f84277bda7 /rts
parentcb2349a4233d6bdef2d7090853397e6278a7378c (diff)
downloadhaskell-74cd4ec5d2f9321aad5db3285cb60d78f2562996.tar.gz
Fix raiseAsync() UNDERFLOW_FRAME handling in profiling runtime
UNDERFLOW_FRAMEs don't have profiling headers so we have to use the AP_STACK's function's CCS as the new frame's CCS. Fixes one of the many bugs caught by concprog001 (#15508).
Diffstat (limited to 'rts')
-rw-r--r--rts/RaiseAsync.c13
1 files changed, 6 insertions, 7 deletions
diff --git a/rts/RaiseAsync.c b/rts/RaiseAsync.c
index 72f5dff7f6..f58f9177c8 100644
--- a/rts/RaiseAsync.c
+++ b/rts/RaiseAsync.c
@@ -93,7 +93,7 @@ suspendComputation (Capability *cap, StgTSO *tso, StgUpdateFrame *stop_here)
throwTo().
Note [Throw to self when masked]
-
+
When a StackOverflow occurs when the thread is masked, we want to
defer the exception to when the thread becomes unmasked/hits an
interruptible point. We already have a mechanism for doing this,
@@ -103,26 +103,26 @@ suspendComputation (Capability *cap, StgTSO *tso, StgUpdateFrame *stop_here)
multithreaded nonsense). Morally, a stack overflow should be an
asynchronous exception sent by a thread to itself, and it should
have the same semantics. But there are a few key differences:
-
+
- If you actually tried to send an asynchronous exception to
yourself using throwTo, the exception would actually immediately
be delivered. This is because throwTo itself is considered an
interruptible point, so the exception is always deliverable. Thus,
ordinarily, we never end up with a message to oneself in the
blocked_exceptions queue.
-
+
- In the case of a StackOverflow, we don't actually care about the
wakeup semantics; when an exception is delivered, the thread that
originally threw the exception should be woken up, since throwTo
blocks until the exception is successfully thrown. Fortunately,
it is harmless to wakeup a thread that doesn't actually need waking
up, e.g. ourselves.
-
+
- No synchronization is necessary, because we own the TSO and the
capability. You can observe this by tracing through the execution
of throwTo. We skip synchronizing the message and inter-capability
communication.
-
+
We think this doesn't break any invariants, but do be careful!
-------------------------------------------------------------------------- */
@@ -921,8 +921,7 @@ raiseAsync(Capability *cap, StgTSO *tso, StgClosure *exception,
ap->payload[i] = (StgClosure *)*sp++;
}
- SET_HDR(ap,&stg_AP_STACK_NOUPD_info,
- ((StgClosure *)frame)->header.prof.ccs /* ToDo */);
+ SET_HDR(ap,&stg_AP_STACK_NOUPD_info,stack->header.prof.ccs);
TICK_ALLOC_SE_THK(WDS(words+1),0);
stack->sp = sp;