powerpc/64: Re-fix race condition between going idle and entering guest

Commit 8117ac6 ("powerpc/powernv: Switch off MMU before entering
nap/sleep/rvwinkle mode", 2014-12-10) fixed a race condition where one
thread entering a KVM guest could switch the MMU context to the guest
while another thread was still in host kernel context with the MMU on.
That commit moved the point where a thread entering a power-saving
mode set its kvm_hstate.hwthread_state field in its PACA to
KVM_HWTHREAD_IN_IDLE from a point where the MMU was on to after the
MMU had been switched off.  That commit also added a comment
explaining that we have to switch to real mode before setting
hwthread_state to avoid this race.

Nevertheless, commit 4eae2c9 ("powerpc/powernv: Make
pnv_powersave_common more generic", 2016-07-08) subsequently moved
the setting of hwthread_state back to a point where the MMU is on,
thus reintroducing the race, despite the comment saying that this
should not be done being included in full in the context lines of
the patch that did it.

This fixes the race again and adds a bigger and shoutier comment
explaining the potential race condition.

Cc: stable@vger.kernel.org # v4.8
Fixes: 4eae2c9
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>

arch/powerpc/kernel/idle_book3s.S

@@ -163,12 +163,6 @@ _GLOBAL(pnv_powersave_common)
 	std	r9,_MSR(r1)
 	std	r1,PACAR1(r13)
 
-#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
-	/* Tell KVM we're entering idle */
-	li	r4,KVM_HWTHREAD_IN_IDLE
-	stb	r4,HSTATE_HWTHREAD_STATE(r13)
-#endif
-
 	/*
 	 * Go to real mode to do the nap, as required by the architecture.
 	 * Also, we need to be in real mode before setting hwthread_state,
@@ -185,6 +179,26 @@ _GLOBAL(pnv_powersave_common)
 
 	.globl pnv_enter_arch207_idle_mode
 pnv_enter_arch207_idle_mode:
+#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
+	/* Tell KVM we're entering idle */
+	li	r4,KVM_HWTHREAD_IN_IDLE
+	/******************************************************/
+	/*  N O T E   W E L L    ! ! !    N O T E   W E L L   */
+	/* The following store to HSTATE_HWTHREAD_STATE(r13)  */
+	/* MUST occur in real mode, i.e. with the MMU off,    */
+	/* and the MMU must stay off until we clear this flag */
+	/* and test HSTATE_HWTHREAD_REQ(r13) in the system    */
+	/* reset interrupt vector in exceptions-64s.S.        */
+	/* The reason is that another thread can switch the   */
+	/* MMU to a guest context whenever this flag is set   */
+	/* to KVM_HWTHREAD_IN_IDLE, and if the MMU was on,    */
+	/* that would potentially cause this thread to start  */
+	/* executing instructions from guest memory in        */
+	/* hypervisor mode, leading to a host crash or data   */
+	/* corruption, or worse.                              */
+	/******************************************************/
+	stb	r4,HSTATE_HWTHREAD_STATE(r13)
+#endif
 	stb	r3,PACA_THREAD_IDLE_STATE(r13)
 	cmpwi	cr3,r3,PNV_THREAD_SLEEP
 	bge	cr3,2f
@@ -250,6 +264,12 @@ enter_winkle:
  * r3 - requested stop state
  */
 power_enter_stop:
+#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
+	/* Tell KVM we're entering idle */
+	li	r4,KVM_HWTHREAD_IN_IDLE
+	/* DO THIS IN REAL MODE!  See comment above. */
+	stb	r4,HSTATE_HWTHREAD_STATE(r13)
+#endif
 /*
  * Check if the requested state is a deep idle state.
  */