[PATCH] lockdep: Fix TRACE_IRQFLAGS vs NMIs

[PATCH] lockdep: Fix TRACE_IRQFLAGS vs NMIs

[peterz]

Prior to commit 859d069ee1dd ("lockdep: Prepare for NMI IRQ state
tracking") IRQ state tracking was disabled in NMIs due to nmi_enter()
doing lockdep_off() -- with the obvious requirement that NMI entry
call nmi_enter() before trace_hardirqs_off().

[ afaict, PowerPC and SH violate this order on their NMI entry ]

However, that commit explicitly changed lockdep_hardirqs_*() to ignore
lockdep_off() and breaks every architecture that has irq-tracing in
it's NMI entry that hasn't been fixed up (x86 being the only fixed one
at this point).

The reason for this change is that by ignoring lockdep_off() we can:

  - get rid of 'current->lockdep_recursion' in lockdep_assert_irqs*()
    which was going to to give header-recursion issues with the
    seqlock rework.

  - allow these lockdep_assert_*() macros to function in NMI context.

Restore the previous state of things and allow an architecture to
opt-in to the NMI IRQ tracking support, however instead of relying on
lockdep_off(), rely on in_nmi(), both are part of nmi_enter() and so
over-all entry ordering doesn't need to change.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 arch/x86/Kconfig.debug   |    3 +++
 kernel/locking/lockdep.c |    8 +++++++-
 lib/Kconfig.debug        |    6 ++++++
 3 files changed, 16 insertions(+), 1 deletion(-)

--- a/arch/x86/Kconfig.debug
+++ b/arch/x86/Kconfig.debug
@@ -3,6 +3,9 @@
 config TRACE_IRQFLAGS_SUPPORT
 	def_bool y
 
+config TRACE_IRQFLAGS_NMI_SUPPORT
+	def_bool y
+
 config EARLY_PRINTK_USB
 	bool
 
--- a/kernel/locking/lockdep.c
+++ b/kernel/locking/lockdep.c
@@ -3712,6 +3712,9 @@ void noinstr lockdep_hardirqs_on(unsigne
 	 * and not rely on hardware state like normal interrupts.
 	 */
 	if (unlikely(in_nmi())) {
+		if (!IS_ENABLED(CONFIG_TRACE_IRQFLAGS_NMI))
+			return;
+
 		/*
 		 * Skip:
 		 *  - recursion check, because NMI can hit lockdep;
@@ -3773,7 +3776,10 @@ void noinstr lockdep_hardirqs_off(unsign
 	 * they will restore the software state. This ensures the software
 	 * state is consistent inside NMIs as well.
 	 */
-	if (unlikely(!in_nmi() && (current->lockdep_recursion & LOCKDEP_RECURSION_MASK)))
+	if (in_nmi()) {
+		if (!IS_ENABLED(CONFIG_TRACE_IRQFLAGS_NMI))
+			return;
+	} else if (current->lockdep_recursion & LOCKDEP_RECURSION_MASK)
 		return;
 
 	/*
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -1325,11 +1325,17 @@ config WW_MUTEX_SELFTEST
 endmenu # lock debugging
 
 config TRACE_IRQFLAGS
+	depends on TRACE_IRQFLAGS_SUPPORT
 	bool
 	help
 	  Enables hooks to interrupt enabling and disabling for
 	  either tracing or lock debugging.
 
+config TRACE_IRQFLAGS_NMI
+	def_bool y
+	depends on TRACE_IRQFLAGS
+	depends on TRACE_IRQFLAGS_NMI_SUPPORT
+
 config STACKTRACE
 	bool "Stack backtrace support"
 	depends on STACKTRACE_SUPPORT

Re: [PATCH] lockdep: Fix TRACE_IRQFLAGS vs NMIs

[Ingo Molnar]

* peterz@infradead.org <peterz@infradead.org> wrote:

> 
> Prior to commit 859d069ee1dd ("lockdep: Prepare for NMI IRQ state
> tracking") IRQ state tracking was disabled in NMIs due to nmi_enter()
> doing lockdep_off() -- with the obvious requirement that NMI entry
> call nmi_enter() before trace_hardirqs_off().
> 
> [ afaict, PowerPC and SH violate this order on their NMI entry ]
> 
> However, that commit explicitly changed lockdep_hardirqs_*() to ignore
> lockdep_off() and breaks every architecture that has irq-tracing in
> it's NMI entry that hasn't been fixed up (x86 being the only fixed one
> at this point).
> 
> The reason for this change is that by ignoring lockdep_off() we can:
> 
>   - get rid of 'current->lockdep_recursion' in lockdep_assert_irqs*()
>     which was going to to give header-recursion issues with the
>     seqlock rework.
> 
>   - allow these lockdep_assert_*() macros to function in NMI context.
> 
> Restore the previous state of things and allow an architecture to
> opt-in to the NMI IRQ tracking support, however instead of relying on
> lockdep_off(), rely on in_nmi(), both are part of nmi_enter() and so
> over-all entry ordering doesn't need to change.
> 
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
>  arch/x86/Kconfig.debug   |    3 +++
>  kernel/locking/lockdep.c |    8 +++++++-
>  lib/Kconfig.debug        |    6 ++++++
>  3 files changed, 16 insertions(+), 1 deletion(-)

Tree management side note: to apply this I've created a new 
tip:locking/nmi branch, which is based off the existing NMI vs. IRQ 
tracing commits included in locking/core:

ed00495333cc: ("locking/lockdep: Fix TRACE_IRQFLAGS vs. NMIs")
ba1f2b2eaa2a: ("x86/entry: Fix NMI vs IRQ state tracking")
859d069ee1dd: ("lockdep: Prepare for NMI IRQ state tracking")
248591f5d257: ("kcsan: Make KCSAN compatible with new IRQ state tracking")
e1bcad609f5a: ("Merge branch 'tip/x86/entry'")
b037b09b9058: ("x86/entry: Rename idtentry_enter/exit_cond_rcu() to idtentry_enter/exit()")
dcb7fd82c75e: ("Linux 5.8-rc4")

This locking/nmi branch can then be merged into irq/entry (there's a 
bunch of conflicts between them), without coupling all of v5.9's 
locking changes to Thomas's generic entry work.

Thanks,

	Ingo