[PATCHv2] percpu: make this_cpu_generic_read() atomic w.r.t. interrupts

From: Mark Rutland
Date: Tue Sep 26 2017 - 07:43:26 EST


As raw_cpu_generic_read() is a plain read from a raw_cpu_ptr() address,
it's possible (albeit unlikely) that the compiler will split the access
across multiple instructions.

In this_cpu_generic_read() we disable preemption but not interrupts
before calling raw_cpu_generic_read(). Thus, an interrupt could be taken
in the middle of the split load instructions. If a this_cpu_write() or
RMW this_cpu_*() op is made to the same variable in the interrupt
handling path, this_cpu_read() will return a torn value.

For native word types, we can avoid tearing using READ_ONCE(), but this
won't work in all cases (e.g. 64-bit types on most 32-bit platforms).
This patch reworks this_cpu_generic_read() to use READ_ONCE() where
possible, otherwise falling back to disabling interrupts.

Signed-off-by: Mark Rutland <mark.rutland@xxxxxxx>
Cc: Arnd Bergmann <arnd@xxxxxxxx>
Cc: Christoph Lameter <cl@xxxxxxxxx>
Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Cc: Pranith Kumar <bobby.prani@xxxxxxxxx>
Cc: Tejun Heo <tj@xxxxxxxxxx>
Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Cc: linux-arch@xxxxxxxxxxxxxxx
---
include/asm-generic/percpu.h | 24 ++++++++++++++++++++++--
1 file changed, 22 insertions(+), 2 deletions(-)

Since v1 [1]:
* Use READ_ONCE() only for native word types
* Disable IRQs for non-native word types

[1] https://lkml.kernel.org/r/1506345872-30559-1-git-send-email-mark.rutland@xxxxxxx

diff --git a/include/asm-generic/percpu.h b/include/asm-generic/percpu.h
index 0504ef8..976f8ac 100644
--- a/include/asm-generic/percpu.h
+++ b/include/asm-generic/percpu.h
@@ -115,15 +115,35 @@
(__ret); \
})

-#define this_cpu_generic_read(pcp) \
+#define __this_cpu_generic_read_nopreempt(pcp) \
({ \
typeof(pcp) __ret; \
preempt_disable_notrace(); \
- __ret = raw_cpu_generic_read(pcp); \
+ __ret = READ_ONCE(*raw_cpu_ptr(&(pcp))); \
preempt_enable_notrace(); \
__ret; \
})

+#define __this_cpu_generic_read_noirq(pcp) \
+({ \
+ typeof(pcp) __ret; \
+ unsigned long __flags; \
+ raw_local_irq_save(__flags); \
+ __ret = raw_cpu_generic_read(pcp); \
+ raw_local_irq_restore(__flags); \
+ __ret; \
+})
+
+#define this_cpu_generic_read(pcp) \
+({ \
+ typeof(pcp) __ret; \
+ if (__native_word(pcp)) \
+ __ret = __this_cpu_generic_read_nopreempt(pcp); \
+ else \
+ __ret = __this_cpu_generic_read_noirq(pcp); \
+ __ret; \
+})
+
#define this_cpu_generic_to_op(pcp, val, op) \
do { \
unsigned long __flags; \
--
1.9.1