XenServer has recently had a support case where this bugframe in
context_switch() was hit, presumably from a corrupt function pointer as the
vcpu pointer was fine.
On balance, it is better to leave the bugframe around for peace of mind in
exceptional circumstances, than to use the optimisations provided by noreturn.
At any meaningful levels of optimisation, the noreturn causes the bugframe to
be optimised out, meaning that any exceptional returns fall into unlikely
branches, which will result in very weird behaviour.
The unreachable() in BUG() does the useful part of noreturn for us, allowing
the compiler not to mess about restoring stack frames etc, but causes a ud2
instruction to be present.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
unreachable(); \
})
-#define schedule_tail(vcpu) (((vcpu)->arch.schedule_tail)(vcpu))
+/*
+ * Schedule tail *should* be a terminal function pointer, but leave a bugframe
+ * around just incase it returns, to save going back into the context
+ * switching code and leaving a far more subtle crash to diagnose.
+ */
+#define schedule_tail(vcpu) do { \
+ (((vcpu)->arch.schedule_tail)(vcpu)); \
+ BUG(); \
+ } while (0)
/*
* Which VCPU's state is currently running on each CPU?
unsigned long flags; /* TF_ */
- void noreturn (*schedule_tail) (struct vcpu *);
+ void (*schedule_tail) (struct vcpu *);
void (*ctxt_switch_from) (struct vcpu *);
void (*ctxt_switch_to) (struct vcpu *);