[PATCH v5 2/3] x86/mce/threshold: Use atomic bit operations on mce_poll_banks

[Aaron Tomlin]

cmci_storm_begin() and cmci_storm_end() modify the per-CPU bitmap
mce_poll_banks using non-atomic __set_bit() and __clear_bit().

While mce_poll_banks is a per-CPU variable, cmci_storm_end() runs in timer
softirq context with local hardirqs enabled, whereas cmci_storm_begin()
can be invoked from CMCI hardirq context (via intel_threshold_interrupt()).

If a CMCI hardirq fires while a timer softirq is midway through a
non-atomic read-modify-write operation on mce_poll_banks, the hardirq's
bit update will be overwritten and lost when the softirq resumes.

If a bank entering storm mode loses its bit in mce_poll_banks, it will
neither generate interrupts (as its hardware threshold is bumped to
CMCI_STORM_THRESHOLD) nor be polled by mce_timer_fn(). Consequently, the
bank remains unpolled and its error telemetry is permanently lost until
the next reboot.

Fix this by switching to atomic set_bit() and clear_bit() operations in
cmci_storm_begin() and cmci_storm_end().

Fixes: 7eae17c4add5 ("x86/mce: Add per-bank CMCI storm mitigation")
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Aaron Tomlin <atomlin@xxxxxxxxxxx>
---
 arch/x86/kernel/cpu/mce/threshold.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/cpu/mce/threshold.c b/arch/x86/kernel/cpu/mce/threshold.c
index 6c370d5af5bd..af8346f06318 100644
--- a/arch/x86/kernel/cpu/mce/threshold.c
+++ b/arch/x86/kernel/cpu/mce/threshold.c
@@ -86,7 +86,7 @@ void cmci_storm_begin(unsigned int bank)
 {
 	struct mca_storm_desc *storm = this_cpu_ptr(&storm_desc);
 
-	__set_bit(bank, this_cpu_ptr(mce_poll_banks));
+	set_bit(bank, this_cpu_ptr(mce_poll_banks));
 	storm->banks[bank].in_storm_mode = true;
 
 	/*
@@ -102,7 +102,7 @@ void cmci_storm_end(unsigned int bank)
 	struct mca_storm_desc *storm = this_cpu_ptr(&storm_desc);
 
 	if (!mce_flags.amd_threshold)
-		__clear_bit(bank, this_cpu_ptr(mce_poll_banks));
+		clear_bit(bank, this_cpu_ptr(mce_poll_banks));
 	storm->banks[bank].history = 0;
 	storm->banks[bank].in_storm_mode = false;
 
-- 
2.55.0