Fix: bump stable kernel version ranges for clock work-around

[lttng-modules.git] / wrapper / trace-clock.h

diff --git a/wrapper/trace-clock.h b/wrapper/trace-clock.h
index 3e8780da167ffa4b3c832eea7e8756c647091c29..496fec4360c7f654c18f454acae564ba43adabe7 100644 (file)
--- a/wrapper/trace-clock.h
+++ b/wrapper/trace-clock.h
@@ -47,7 +47,20 @@
 
 extern struct lttng_trace_clock *lttng_trace_clock;
 
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0))
+/*
+ * Upstream Linux commit 27727df240c7 ("Avoid taking lock in NMI path with
+ * CONFIG_DEBUG_TIMEKEEPING") introduces a buggy ktime_get_mono_fast_ns().
+ * This is fixed by patch "timekeeping: Fix __ktime_get_fast_ns() regression".
+ */
+#if (LTTNG_KERNEL_RANGE(4,8,0, 4,8,2) \
+       || LTTNG_KERNEL_RANGE(4,7,4, 4,7,8) \
+       || LTTNG_KERNEL_RANGE(4,4,20, 4,4,25) \
+       || LTTNG_KERNEL_RANGE(4,1,32, 4,1,35))
+#define LTTNG_CLOCK_NMI_SAFE_BROKEN
+#endif
+
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) \
+       && !defined(LTTNG_CLOCK_NMI_SAFE_BROKEN))
 
 DECLARE_PER_CPU(local_t, lttng_last_tsc);