xtensa: implement fake NMI

In case perf IRQ is the highest of the medium-level IRQs, and is alone
on its level, it may be treated as NMI:
- LOCKLEVEL is defined to be one level less than EXCM level,
- IRQ masking never lowers current IRQ level,
- new fake exception cause code, EXCCAUSE_MAPPED_NMI is assigned to that
  IRQ; new second level exception handler, do_nmi, assigned to it
  handles it as NMI,
- atomic operations in configurations without s32c1i still need to mask
  all interrupts.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>

5 files changed, 126 insertions, 17 deletions

diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S
index 5703bc13a9ba..504130357597 100644
--- a/arch/xtensa/kernel/entry.S
+++ b/arch/xtensa/kernel/entry.S
@@ -1,6 +1,4 @@
 /*
- * arch/xtensa/kernel/entry.S
- *
  * Low-level exception handling
  *
  * This file is subject to the terms and conditions of the GNU General Public
@@ -8,6 +6,7 @@
  * for more details.
  *
  * Copyright (C) 2004 - 2008 by Tensilica Inc.
+ * Copyright (C) 2015 Cadence Design Systems Inc.
  *
  * Chris Zankel <chris@zankel.net>
  *
@@ -75,6 +74,27 @@
 #endif
 	.endm
 
+
+	.macro	irq_save flags tmp
+#if XTENSA_FAKE_NMI
+#if defined(CONFIG_DEBUG_KERNEL) && (LOCKLEVEL | TOPLEVEL) >= XCHAL_DEBUGLEVEL
+	rsr	\flags, ps
+	extui	\tmp, \flags, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH
+	bgei	\tmp, LOCKLEVEL, 99f
+	rsil	\tmp, LOCKLEVEL
+99:
+#else
+	movi	\tmp, LOCKLEVEL
+	rsr	\flags, ps
+	or	\flags, \flags, \tmp
+	xsr	\flags, ps
+	rsync
+#endif
+#else
+	rsil	\flags, LOCKLEVEL
+#endif
+	.endm
+
 /* ----------------- DEFAULT FIRST LEVEL EXCEPTION HANDLERS ----------------- */
 
 /*
@@ -352,11 +372,11 @@ common_exception:
 
 	/* It is now save to restore the EXC_TABLE_FIXUP variable. */
 
-	rsr	a0, exccause
+	rsr	a2, exccause
 	movi	a3, 0
-	rsr	a2, excsave1
-	s32i	a0, a1, PT_EXCCAUSE
-	s32i	a3, a2, EXC_TABLE_FIXUP
+	rsr	a0, excsave1
+	s32i	a2, a1, PT_EXCCAUSE
+	s32i	a3, a0, EXC_TABLE_FIXUP
 
 	/* All unrecoverable states are saved on stack, now, and a1 is valid.
 	 * Now we can allow exceptions again. In case we've got an interrupt
@@ -367,19 +387,46 @@ common_exception:
 	 */
 
 	rsr	a3, ps
-	addi	a0, a0, -EXCCAUSE_LEVEL1_INTERRUPT
-	movi	a2, LOCKLEVEL
+	s32i	a3, a1, PT_PS		# save ps
+
+#if XTENSA_FAKE_NMI
+	/* Correct PS needs to be saved in the PT_PS:
+	 * - in case of exception or level-1 interrupt it's in the PS,
+	 *   and is already saved.
+	 * - in case of medium level interrupt it's in the excsave2.
+	 */
+	movi	a0, EXCCAUSE_MAPPED_NMI
+	extui	a3, a3, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH
+	beq	a2, a0, .Lmedium_level_irq
+	bnei	a2, EXCCAUSE_LEVEL1_INTERRUPT, .Lexception
+	beqz	a3, .Llevel1_irq	# level-1 IRQ sets ps.intlevel to 0
+
+.Lmedium_level_irq:
+	rsr	a0, excsave2
+	s32i	a0, a1, PT_PS		# save medium-level interrupt ps
+	bgei	a3, LOCKLEVEL, .Lexception
+
+.Llevel1_irq:
+	movi	a3, LOCKLEVEL
+
+.Lexception:
+	movi	a0, 1 << PS_WOE_BIT
+	or	a3, a3, a0
+#else
+	addi	a2, a2, -EXCCAUSE_LEVEL1_INTERRUPT
+	movi	a0, LOCKLEVEL
 	extui	a3, a3, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH
 					# a3 = PS.INTLEVEL
-	moveqz	a3, a2, a0		# a3 = LOCKLEVEL iff interrupt
+	moveqz	a3, a0, a2		# a3 = LOCKLEVEL iff interrupt
 	movi	a2, 1 << PS_WOE_BIT
 	or	a3, a3, a2
 	rsr	a2, exccause
+#endif
+
 	/* restore return address (or 0 if return to userspace) */
 	rsr	a0, depc
-	xsr	a3, ps
-
-	s32i	a3, a1, PT_PS		# save ps
+	wsr	a3, ps
+	rsync				# PS.WOE => rsync => overflow
 
 	/* Save lbeg, lend */
 
@@ -417,8 +464,13 @@ common_exception:
 	.global common_exception_return
 common_exception_return:
 
+#if XTENSA_FAKE_NMI
+	l32i	a2, a1, PT_EXCCAUSE
+	movi	a3, EXCCAUSE_MAPPED_NMI
+	beq	a2, a3, .LNMIexit
+#endif
 1:
-	rsil	a2, LOCKLEVEL
+	irq_save a2, a3
 #ifdef CONFIG_TRACE_IRQFLAGS
 	movi	a4, trace_hardirqs_off
 	callx4	a4
@@ -481,6 +533,12 @@ common_exception_return:
 	j	1b
 #endif
 
+#if XTENSA_FAKE_NMI
+.LNMIexit:
+	l32i	a3, a1, PT_PS
+	_bbci.l	a3, PS_UM_BIT, 4f
+#endif
+
 5:
 #ifdef CONFIG_DEBUG_TLB_SANITY
 	l32i	a4, a1, PT_DEPC
@@ -1564,6 +1622,13 @@ ENTRY(fast_second_level_miss)
 	rfde
 
 9:	l32i	a0, a1, TASK_ACTIVE_MM	# unlikely case mm == 0
+	bnez	a0, 8b
+
+	/* Even more unlikely case active_mm == 0.
+	 * We can get here with NMI in the middle of context_switch that
+	 * touches vmalloc area.
+	 */
+	movi	a0, init_mm
 	j	8b
 
 #if (DCACHE_WAY_SIZE > PAGE_SIZE)
@@ -1867,7 +1932,7 @@ ENTRY(_switch_to)
 
 	/* Disable ints while we manipulate the stack pointer. */
 
-	rsil	a14, LOCKLEVEL
+	irq_save a14, a3
 	rsync
 
 	/* Switch CPENABLE */
diff --git a/arch/xtensa/kernel/irq.c b/arch/xtensa/kernel/irq.c
index 32b60568ee45..91ba60563021 100644
--- a/arch/xtensa/kernel/irq.c
+++ b/arch/xtensa/kernel/irq.c
@@ -29,6 +29,7 @@
 #include <asm/platform.h>
 
 atomic_t irq_err_count;
+DECLARE_PER_CPU(unsigned long, nmi_count);
 
 asmlinkage void do_IRQ(int hwirq, struct pt_regs *regs)
 {
@@ -57,11 +58,18 @@ asmlinkage void do_IRQ(int hwirq, struct pt_regs *regs)
 
 int arch_show_interrupts(struct seq_file *p, int prec)
 {
+	unsigned cpu __maybe_unused;
 #ifdef CONFIG_SMP
 	show_ipi_list(p, prec);
 #endif
 	seq_printf(p, "%*s: ", prec, "ERR");
 	seq_printf(p, "%10u\n", atomic_read(&irq_err_count));
+#if XTENSA_FAKE_NMI
+	seq_printf(p, "%*s:", prec, "NMI");
+	for_each_online_cpu(cpu)
+		seq_printf(p, " %10lu", per_cpu(nmi_count, cpu));
+	seq_puts(p, "   Non-maskable interrupts\n");
+#endif
 	return 0;
 }
 
diff --git a/arch/xtensa/kernel/perf_event.c b/arch/xtensa/kernel/perf_event.c
index b44df3c8198f..54f01188c29c 100644
--- a/arch/xtensa/kernel/perf_event.c
+++ b/arch/xtensa/kernel/perf_event.c
@@ -359,7 +359,7 @@ void perf_event_print_debug(void)
 	local_irq_restore(flags);
 }
 
-static irqreturn_t xtensa_pmu_irq_handler(int irq, void *dev_id)
+irqreturn_t xtensa_pmu_irq_handler(int irq, void *dev_id)
 {
 	irqreturn_t rc = IRQ_NONE;
 	struct xtensa_pmu_events *ev = this_cpu_ptr(&xtensa_pmu_events);
@@ -436,10 +436,14 @@ static int __init xtensa_pmu_init(void)
 	int irq = irq_create_mapping(NULL, XCHAL_PROFILING_INTERRUPT);
 
 	perf_cpu_notifier(xtensa_pmu_notifier);
+#if XTENSA_FAKE_NMI
+	enable_irq(irq);
+#else
 	ret = request_irq(irq, xtensa_pmu_irq_handler, IRQF_PERCPU,
 			  "pmu", NULL);
 	if (ret < 0)
 		return ret;
+#endif
 
 	ret = perf_pmu_register(&xtensa_pmu, "cpu", PERF_TYPE_RAW);
 	if (ret)
diff --git a/arch/xtensa/kernel/traps.c b/arch/xtensa/kernel/traps.c
index a1b5bd237c71..42d441f7898b 100644
--- a/arch/xtensa/kernel/traps.c
+++ b/arch/xtensa/kernel/traps.c
@@ -62,6 +62,7 @@ extern void fast_coprocessor(void);
 
 extern void do_illegal_instruction (struct pt_regs*);
 extern void do_interrupt (struct pt_regs*);
+extern void do_nmi(struct pt_regs *);
 extern void do_unaligned_user (struct pt_regs*);
 extern void do_multihit (struct pt_regs*, unsigned long);
 extern void do_page_fault (struct pt_regs*, unsigned long);
@@ -146,6 +147,9 @@ COPROCESSOR(6),
 #if XTENSA_HAVE_COPROCESSOR(7)
 COPROCESSOR(7),
 #endif
+#if XTENSA_FAKE_NMI
+{ EXCCAUSE_MAPPED_NMI,			0,		do_nmi },
+#endif
 { EXCCAUSE_MAPPED_DEBUG,		0,		do_debug },
 { -1, -1, 0 }
 
@@ -199,6 +203,28 @@ void do_multihit(struct pt_regs *regs, unsigned long exccause)
 
 extern void do_IRQ(int, struct pt_regs *);
 
+#if XTENSA_FAKE_NMI
+
+irqreturn_t xtensa_pmu_irq_handler(int irq, void *dev_id);
+
+DEFINE_PER_CPU(unsigned long, nmi_count);
+
+void do_nmi(struct pt_regs *regs)
+{
+	struct pt_regs *old_regs;
+
+	if ((regs->ps & PS_INTLEVEL_MASK) < LOCKLEVEL)
+		trace_hardirqs_off();
+
+	old_regs = set_irq_regs(regs);
+	nmi_enter();
+	++*this_cpu_ptr(&nmi_count);
+	xtensa_pmu_irq_handler(0, NULL);
+	nmi_exit();
+	set_irq_regs(old_regs);
+}
+#endif
+
 void do_interrupt(struct pt_regs *regs)
 {
 	static const unsigned int_level_mask[] = {
diff --git a/arch/xtensa/kernel/vectors.S b/arch/xtensa/kernel/vectors.S
index 1b397a902292..abcdb527f18a 100644
--- a/arch/xtensa/kernel/vectors.S
+++ b/arch/xtensa/kernel/vectors.S
@@ -627,7 +627,11 @@ ENTRY(_Level\level\()InterruptVector)
 	wsr	a0, excsave2
 	rsr	a0, epc\level
 	wsr	a0, epc1
+	.if	\level <= LOCKLEVEL
 	movi	a0, EXCCAUSE_LEVEL1_INTERRUPT
+	.else
+	movi	a0, EXCCAUSE_MAPPED_NMI
+	.endif
 	wsr	a0, exccause
 	rsr	a0, eps\level
 					# branch to user or kernel vector
@@ -682,11 +686,13 @@ ENDPROC(_WindowOverflow4)
 	.align 4
 _SimulateUserKernelVectorException:
 	addi	a0, a0, (1 << PS_EXCM_BIT)
+#if !XTENSA_FAKE_NMI
 	wsr	a0, ps
+#endif
 	bbsi.l	a0, PS_UM_BIT, 1f	# branch if user mode
-	rsr	a0, excsave2		# restore a0
+	xsr	a0, excsave2		# restore a0
 	j	_KernelExceptionVector	# simulate kernel vector exception
-1:	rsr	a0, excsave2		# restore a0
+1:	xsr	a0, excsave2		# restore a0
 	j	_UserExceptionVector	# simulate user vector exception
 #endif