[PATCH] irq: clarify logic calculating bogus irqreturn_t values

[PATCH] irq: clarify logic calculating bogus irqreturn_t values

[Jeremy Kerr]

Although irqreturn_t is an enum, we treat it (and its enumeration
constants) as a bitmask.

However, bad_action_ret() uses a less-than operator to determine whether
an irqreturn_t falls within allowable bit values, which means we need to
know the signededness of an enum type to read the logic, which is
implementation-dependent.

This change explicitly uses an unsigned type for the comparison. We do
this instead of changing to a bitwise test, as the latter compiles to
increased instructions in this hot path.

It looks like we get the correct behaviour currently (bad_action_ret(-1)
returns 1), so this is purely a readability fix.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
---
 kernel/irq/spurious.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/kernel/irq/spurious.c b/kernel/irq/spurious.c
index 5707f97..061ba7e 100644
--- a/kernel/irq/spurious.c
+++ b/kernel/irq/spurious.c
@@ -175,7 +175,9 @@ static void poll_spurious_irqs(unsigned long dummy)
 
 static inline int bad_action_ret(irqreturn_t action_ret)
 {
-	if (likely(action_ret <= (IRQ_HANDLED | IRQ_WAKE_THREAD)))
+	unsigned int r = action_ret;
+
+	if (likely(r <= (IRQ_HANDLED | IRQ_WAKE_THREAD)))
 		return 0;
 	return 1;
 }
-- 
2.7.4

[tip:irq/core] genirq: Clarify logic calculating bogus irqreturn_t values

[tip-bot for Jeremy Kerr]

Commit-ID:  5d4bac9a5f4ef24b2482529bda6661a58e5b5b65
Gitweb:     http://git.kernel.org/tip/5d4bac9a5f4ef24b2482529bda6661a58e5b5b65
Author:     Jeremy Kerr <jk@ozlabs.org>
AuthorDate: Thu, 16 Feb 2017 12:24:09 +0800
Committer:  Thomas Gleixner <tglx@linutronix.de>
CommitDate: Thu, 16 Feb 2017 15:32:19 +0100

genirq: Clarify logic calculating bogus irqreturn_t values

Although irqreturn_t is an enum, we treat it (and its enumeration
constants) as a bitmask.

However, bad_action_ret() uses a less-than operator to determine whether
an irqreturn_t falls within allowable bit values, which means we need to
know the signededness of an enum type to read the logic, which is
implementation-dependent.

This change explicitly uses an unsigned type for the comparison. We do
this instead of changing to a bitwise test, as the latter compiles to
increased instructions in this hot path.

It looks like we get the correct behaviour currently (bad_action_ret(-1)
returns 1), so this is purely a readability fix.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
Link: http://lkml.kernel.org/r/1487219049-4061-1-git-send-email-jk@ozlabs.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

---
 kernel/irq/spurious.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/kernel/irq/spurious.c b/kernel/irq/spurious.c
index 5707f97..061ba7e 100644
--- a/kernel/irq/spurious.c
+++ b/kernel/irq/spurious.c
@@ -175,7 +175,9 @@ out:
 
 static inline int bad_action_ret(irqreturn_t action_ret)
 {
-	if (likely(action_ret <= (IRQ_HANDLED | IRQ_WAKE_THREAD)))
+	unsigned int r = action_ret;
+
+	if (likely(r <= (IRQ_HANDLED | IRQ_WAKE_THREAD)))
 		return 0;
 	return 1;
 }