Re: Softirq priority inversion from "softirq: reduce latencies"

[Eric Dumazet]

On lun., 2016-02-29 at 07:54 -0800, Peter Hurley wrote:

>  The current kernel is HZ=250 but this would occur on HZ=1000 as well.

Right. But the problem with HZ=100 and HZ=250 is that the detection can
happens because jiffy granularity is too coarse, since 

msecs_to_jiffies(2) -> 1

Following patch might reduce the probability, but wont really fix your
problem.

Fact that ksoftirqd prio is not what you want is completely orthogonal.

diff --git a/kernel/softirq.c b/kernel/softirq.c
index 479e443..f7cc594 100644
--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -180,7 +180,7 @@ EXPORT_SYMBOL(__local_bh_enable_ip);
 
 /*
  * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times,
- * but break the loop if need_resched() is set or after 2 ms.
+ * but break the loop if need_resched() is set or after 2 ms/ticks.
  * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in
  * certain cases, such as stop_machine(), jiffies may cease to
  * increment and so we need the MAX_SOFTIRQ_RESTART limit as
@@ -191,7 +191,7 @@ EXPORT_SYMBOL(__local_bh_enable_ip);
  * we want to handle softirqs as soon as possible, but they
  * should not be able to lock up the box.
  */
-#define MAX_SOFTIRQ_TIME  msecs_to_jiffies(2)
+#define MAX_SOFTIRQ_TIME  (1 + msecs_to_jiffies(2))
 #define MAX_SOFTIRQ_RESTART 10
 
 #ifdef CONFIG_TRACE_IRQFLAGS