From: Nadav Har'El <[email protected]>
Committer: Nadav Har'El <[email protected]>
Branch: master
signal: remove "nested signals" protection
This reverts commit 4af3677117a5bb2f4660d750fa4eddc6ef44e7f2 which
aborted on a "nested signal" (handling a signal handler while inside
a signal handler).
Nested signals were never strictly impossible, but even more importantly,
if a signal handler is exited with a longjmp or siglongjmp, our test code
would think it is still inside the handler, and the next signal handler
called by thi thread will abort on a supposedly "nested signal".
Perhaps this test can be redeemed if siglongjmp() would also clear the
signal_nesting flag, but it's easier just to remove this test.
Signed-off-by: Nadav Har'El <[email protected]>
Message-Id: <[email protected]>
---
diff --git a/arch/x64/signal.cc b/arch/x64/signal.cc
--- a/arch/x64/signal.cc
+++ b/arch/x64/signal.cc
@@ -43,24 +43,16 @@ void build_signal_frame(exception_frame* ef,
}
-static unsigned __thread signal_nesting;
-
extern "C"
void call_signal_handler(arch::signal_frame* frame)
{
- if (signal_nesting) {
- // Note: nested signals are legal, but rarely used, so they usually
- // indicate trouble
- abort("nested signals");
- }
// The user's signal handler might use the FPU, so save its current
state.
// FIXME: this fpu saving is not necessary if the callers already save
the
// FPU state. Currently, only divide_error() is missing FPU saving, and
// callers (such as page_fault()) already save it. If we fix
// divide_error(), we can probably get rid of the fpu saving here.
sched::fpu_lock fpu;
SCOPE_LOCK(fpu);
- ++signal_nesting;
if (frame->sa.sa_flags & SA_SIGINFO) {
ucontext_t uc = {};
auto& regs = uc.uc_mcontext.gregs;
@@ -103,7 +95,6 @@ void call_signal_handler(arch::signal_frame* frame)
} else {
frame->sa.sa_handler(frame->si.si_signo);
}
- --signal_nesting;
}
--
You received this message because you are subscribed to the Google Groups "OSv
Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
