---

yaml
---
r: 162679
b: refs/heads/master
c: 7a7967d
h: refs/heads/master
i:
  162677: 497ad86
  162675: 9a5f377
  162671: 21c6c68
v: v3

[refs]

@@ -1,2 +1,2 @@
 ---
-refs/heads/master: 70f47202320623270f327e249df0121c09e45c09
+refs/heads/master: 7a7967dc1b606f8c88e33bbec773bf82b4a52e6e

trunk/arch/blackfin/mach-common/interrupt.S

@@ -261,6 +261,31 @@ ENTRY(_evt_system_call)
 ENDPROC(_evt_system_call)
 
 #ifdef CONFIG_IPIPE
+/*
+ * __ipipe_call_irqtail: lowers the current priority level to EVT15
+ * before running a user-defined routine, then raises the priority
+ * level to EVT14 to prepare the caller for a normal interrupt
+ * return through RTI.
+ *
+ * We currently use this facility in two occasions:
+ *
+ * - to branch to __ipipe_irq_tail_hook as requested by a high
+ *   priority domain after the pipeline delivered an interrupt,
+ *   e.g. such as Xenomai, in order to start its rescheduling
+ *   procedure, since we may not switch tasks when IRQ levels are
+ *   nested on the Blackfin, so we have to fake an interrupt return
+ *   so that we may reschedule immediately.
+ *
+ * - to branch to sync_root_irqs, in order to play any interrupt
+ *   pending for the root domain (i.e. the Linux kernel). This lowers
+ *   the core priority level enough so that Linux IRQ handlers may
+ *   never delay interrupts handled by high priority domains; we defer
+ *   those handlers until this point instead. This is a substitute
+ *   to using a threaded interrupt model for the Linux kernel.
+ *
+ * r0: address of user-defined routine
+ * context: caller must have preempted EVT15, hw interrupts must be off.
+ */
 ENTRY(___ipipe_call_irqtail)
 	p0 = r0;
 	r0.l = 1f;