---

yaml
---
r: 217607
b: refs/heads/master
c: d6f0f22
h: refs/heads/master
i:
  217605: c4a8fc4
  217603: aae5721
  217599: 478c209
v: v3

[refs]

@@ -1,2 +1,2 @@
 ---
-refs/heads/master: 5592840ba83c281b4df45345863b224bb290020c
+refs/heads/master: d6f0f22c3c409f9ed88e64f881f8308730be76f0

trunk/arch/tile/kernel/intvec_32.S

@@ -1553,6 +1553,8 @@ STD_ENTRY(_sys_clone)
  * to be available to it on entry.  It does not modify any callee-save
  * registers (including "lr").  It does not check what PL it is being
  * called at, so you'd better not call it other than at PL0.
+ * The <atomic.h> wrapper assumes it only clobbers r20-r29, so if
+ * it ever is necessary to use more registers, be aware.
  *
  * It does not use the stack, but since it might be re-interrupted by
  * a page fault which would assume the stack was valid, it does

trunk/arch/tile/kernel/process.c

@@ -214,9 +214,10 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
 	/*
 	 * Copy the callee-saved registers from the passed pt_regs struct
 	 * into the context-switch callee-saved registers area.
-	 * We have to restore the callee-saved registers since we may
-	 * be cloning a userspace task with userspace register state,
-	 * and we won't be unwinding the same kernel frames to restore them.
+	 * This way when we start the interrupt-return sequence, the
+	 * callee-save registers will be correctly in registers, which
+	 * is how we assume the compiler leaves them as we start doing
+	 * the normal return-from-interrupt path after calling C code.
 	 * Zero out the C ABI save area to mark the top of the stack.
 	 */
 	ksp = (unsigned long) childregs;