[PATCH v16 5/9] isolation: Introduce managed_irq_strict isolcpus type

[Aaron Tomlin]

From: Daniel Wagner <wagi@xxxxxxxxxx>

Multiqueue drivers spread I/O queues and managed interrupts across all
CPUs for optimal performance. However, these drivers are not aware of
CPU isolation requirements and will distribute queues and interrupt
vectors without considering the isolcpus configuration.

The existing isolcpus=managed_irq parameter attempts post-allocation
affinity steering, but this is best-effort and fails on modern multiqueue
storage where the number of queues meets or exceeds the CPU count.

Introduce a new isolcpus flag, "managed_irq_strict", that allows users
to strictly define which CPUs should have multiqueue hardware queues
and managed interrupts assigned. This enforces pre-allocation queue
bounding and mask exclusion, preventing managed hardware interrupts
from ever targeting isolated cores.

Because managed_irq_strict is a strict superset of managed_irq, specifying
managed_irq_strict automatically enables HK_FLAG_MANAGED_IRQ as well. This
ensures subsystems that rely on managed interrupt isolation (e.g. Hyper-V
VMBus channel distribution) continue to function seamlessly.

Signed-off-by: Daniel Wagner <wagi@xxxxxxxxxx>
Reviewed-by: Martin K. Petersen <martin.petersen@xxxxxxxxxx>
Reviewed-by: Hannes Reinecke <hare@xxxxxxx>
Signed-off-by: Aaron Tomlin <atomlin@xxxxxxxxxxx>
---
 include/linux/sched/isolation.h |  1 +
 kernel/sched/isolation.c        | 16 ++++++++++++++++
 2 files changed, 17 insertions(+)

diff --git a/include/linux/sched/isolation.h b/include/linux/sched/isolation.h
index cf0fd03dd7a2..bda3003ce86f 100644
--- a/include/linux/sched/isolation.h
+++ b/include/linux/sched/isolation.h
@@ -18,6 +18,7 @@ enum hk_type {
 	HK_TYPE_MANAGED_IRQ,
 	/* Inverse of boot-time nohz_full= or isolcpus=nohz arguments */
 	HK_TYPE_KERNEL_NOISE,
+	HK_TYPE_MANAGED_IRQ_STRICT,
 	HK_TYPE_MAX,
 
 	/*
diff --git a/kernel/sched/isolation.c b/kernel/sched/isolation.c
index 9f02e9264a3f..7e9e78a18fb9 100644
--- a/kernel/sched/isolation.c
+++ b/kernel/sched/isolation.c
@@ -17,6 +17,7 @@ enum hk_flags {
 	HK_FLAG_DOMAIN		= BIT(HK_TYPE_DOMAIN),
 	HK_FLAG_MANAGED_IRQ	= BIT(HK_TYPE_MANAGED_IRQ),
 	HK_FLAG_KERNEL_NOISE	= BIT(HK_TYPE_KERNEL_NOISE),
+	HK_FLAG_MANAGED_IRQ_STRICT = BIT(HK_TYPE_MANAGED_IRQ_STRICT),
 };
 
 DEFINE_STATIC_KEY_FALSE(housekeeping_overridden);
@@ -354,6 +355,12 @@ static int __init housekeeping_isolcpus_setup(char *str)
 			continue;
 		}
 
+		if (!strncmp(str, "managed_irq_strict,", 19)) {
+			str += 19;
+			flags |= HK_FLAG_MANAGED_IRQ_STRICT;
+			continue;
+		}
+
 		/*
 		 * Skip unknown sub-parameter and validate that it is not
 		 * containing an invalid character.
@@ -374,6 +381,15 @@ static int __init housekeeping_isolcpus_setup(char *str)
 		str++;
 	}
 
+	/*
+	 * managed_irq_strict is a strict superset of managed_irq.
+	 * Ensure that HK_FLAG_MANAGED_IRQ is also set so that subsystems
+	 * relying on managed interrupt isolation (e.g. Hyper-V VMBus channel
+	 * distribution and CPU hotplug IRQ restoration) continue to function.
+	 */
+	if (flags & HK_FLAG_MANAGED_IRQ_STRICT)
+		flags |= HK_FLAG_MANAGED_IRQ;
+
 	/* Default behaviour for isolcpus without flags */
 	if (!flags)
 		flags |= HK_FLAG_DOMAIN | HK_FLAG_DOMAIN_BOOT;
-- 
2.55.0