[ANNOUNCE] v5.0.21-rt16v5.0.21-rt16-patches

Dear RT folks!

I'm pleased to announce the v5.0.21-rt16 patch set.

Changes since v5.0.21-rt15:

  - Do not invoke softirq if ksoftirqd has been scheduled. This change
    aligns softirq handling closer with mainline. A busy network driver
    will now handover further processing to ksoftirqd. This wasn't the
    case since the recent softirq rework.

Known issues
     - rcutorture is currently broken on -RT. Reported by Juri Lelli.

The delta patch against v5.0.21-rt15 is appended below and can be found here:

     https://cdn.kernel.org/pub/linux/kernel/projects/rt/5.0/incr/patch-5.0.21-rt15-rt16.patch.xz

You can get this release via the git tree at:

    git://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git v5.0.21-rt16

The RT patch against v5.0.21 can be found here:

    https://cdn.kernel.org/pub/linux/kernel/projects/rt/5.0/older/patch-5.0.21-rt16.patch.xz

The split quilt queue is available at:

    https://cdn.kernel.org/pub/linux/kernel/projects/rt/5.0/older/patches-5.0.21-rt16.tar.xz

Sebastian

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

3 files changed, 51 insertions, 1 deletions

diff --git a/patches/localversion.patch b/patches/localversion.patch
index 340816c8febc..0cccc7790a5d 100644
--- a/patches/localversion.patch
+++ b/patches/localversion.patch
@@ -10,4 +10,4 @@ Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
 --- /dev/null
 +++ b/localversion-rt
 @@ -0,0 +1 @@
-+-rt15
++-rt16
diff --git a/patches/series b/patches/series
index d1b51ee5d06f..e1ca1ebe5a3d 100644
--- a/patches/series
+++ b/patches/series
@@ -132,6 +132,7 @@ x86-fpu-Update-kernel-s-FPU-state-before-using-for-t.patch
 efi-Allow-efi-runtime.patch
 
 softirq-Add-preemptible-softirq.patch
+softirq-Don-t-do-softirq-if-ksoftird-is-pending.patch
 sched-swait-Add-swait_event_lock_irq.patch
 # WORKQUEUE
 workqueue-use-rcu.patch
diff --git a/patches/softirq-Don-t-do-softirq-if-ksoftird-is-pending.patch b/patches/softirq-Don-t-do-softirq-if-ksoftird-is-pending.patch
new file mode 100644
index 000000000000..a36c5c89c186
--- /dev/null
+++ b/patches/softirq-Don-t-do-softirq-if-ksoftird-is-pending.patch
@@ -0,0 +1,49 @@
+From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
+Date: Wed, 10 Jul 2019 12:32:04 +0200
+Subject: [PATCH] softirq: Don't do softirq if ksoftird is pending
+
+!RT does not invoke softirq on return from interrupt if ksoftirqd is
+scheduled. This for instance for networking: once the budget is used,
+napi will mark NET_RX pending and let ksoftird scheduled. This will
+ensure that further napi processing happens in ksoftirqd and the system
+can do something else.
+In RT we had "shortcut" for this once budget is used to ensure that
+further softirq handling happens in ksoftirqd. After the softirq rework
+I dropped that shortcut assuming that we would gain mainline behaviour
+here. The missing part piece is not to invoke softirqs if ksoftird is
+pending.
+
+Don't invoke softirqs on BH-enable if ksoftirqd is scheduled.
+
+Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
+---
+ kernel/softirq.c |    4 +---
+ 1 file changed, 1 insertion(+), 3 deletions(-)
+
+--- a/kernel/softirq.c
++++ b/kernel/softirq.c
+@@ -78,7 +78,6 @@ static void wakeup_softirqd(void)
+ 		wake_up_process(tsk);
+ }
+ 
+-#ifndef CONFIG_PREEMPT_RT_FULL
+ /*
+  * If ksoftirqd is scheduled, we do not want to process pending softirqs
+  * right now. Let ksoftirqd handle this at its own rate, to get fairness,
+@@ -93,7 +92,6 @@ static bool ksoftirqd_running(unsigned l
+ 		return false;
+ 	return tsk && (tsk->state == TASK_RUNNING);
+ }
+-#endif
+ 
+ /*
+  * preempt_count and SOFTIRQ_OFFSET usage:
+@@ -173,7 +171,7 @@ void __local_bh_enable_ip(unsigned long
+ 
+ 	if (unlikely(count == 1)) {
+ 		pending = local_softirq_pending();
+-		if (pending) {
++		if (pending && !ksoftirqd_running(pending)) {
+ 			if (!in_atomic())
+ 				__do_softirq();
+ 			else