[PATCH kernel v2] powerpc/xive: Drop deregistered irqs

Alexey Kardashevskiy aik at ozlabs.ru
Mon Jul 15 17:11:31 AEST 2019


There is a race between releasing an irq on one cpu and fetching it
from XIVE on another cpu as there does not seem to be any locking between
these, probably because xive_irq_chip::irq_shutdown() is supposed to
remove the irq from all queues in the system which it does not do.

As a result, when such released irq appears in a queue, we take it
from the queue but we do not change the current priority on that cpu and
since there is no handler for the irq, EOI is never called and the cpu
current priority remains elevated (7 vs. 0xff==unmasked). If another irq
is assigned to the same cpu, then that device stops working until irq
is moved to another cpu or the device is reset.

This adds a new ppc_md.orphan_irq callback which is called if no irq
descriptor is found. The XIVE implementation drops the current priority
to 0xff which effectively unmasks interrupts in a current CPU.

Signed-off-by: Alexey Kardashevskiy <aik at ozlabs.ru>
---
Changes:
v2:
* added ppc_md.orphan_irq

---

Found it on P9 system with:
- a host with 8 cpus online
- a boot disk on ahci with its msix on cpu#0
- a guest with 2xGPUs + 6xNVLink + 4 cpus
- GPU#0 from the guest is bound to the same cpu#0.

Killing a guest killed ahci and therefore the host because of the race.
Note that VFIO masks interrupts first and only then resets the device.
---
 arch/powerpc/include/asm/machdep.h |  3 +++
 arch/powerpc/kernel/irq.c          |  9 ++++++---
 arch/powerpc/sysdev/xive/common.c  | 10 ++++++++++
 3 files changed, 19 insertions(+), 3 deletions(-)

diff --git a/arch/powerpc/include/asm/machdep.h b/arch/powerpc/include/asm/machdep.h
index c43d6eca9edd..6cc14e28e89a 100644
--- a/arch/powerpc/include/asm/machdep.h
+++ b/arch/powerpc/include/asm/machdep.h
@@ -59,6 +59,9 @@ struct machdep_calls {
 	/* Return an irq, or 0 to indicate there are none pending. */
 	unsigned int	(*get_irq)(void);
 
+	/* Drops irq if it does not have a valid descriptor */
+	void		(*orphan_irq)(unsigned int irq);
+
 	/* PCI stuff */
 	/* Called after allocating resources */
 	void		(*pcibios_fixup)(void);
diff --git a/arch/powerpc/kernel/irq.c b/arch/powerpc/kernel/irq.c
index bc68c53af67c..b4e06d05bdba 100644
--- a/arch/powerpc/kernel/irq.c
+++ b/arch/powerpc/kernel/irq.c
@@ -632,10 +632,13 @@ void __do_irq(struct pt_regs *regs)
 	may_hard_irq_enable();
 
 	/* And finally process it */
-	if (unlikely(!irq))
+	if (unlikely(!irq)) {
 		__this_cpu_inc(irq_stat.spurious_irqs);
-	else
-		generic_handle_irq(irq);
+	} else if (generic_handle_irq(irq)) {
+		if (ppc_md.orphan_irq)
+			ppc_md.orphan_irq(irq);
+		__this_cpu_inc(irq_stat.spurious_irqs);
+	}
 
 	trace_irq_exit(regs);
 
diff --git a/arch/powerpc/sysdev/xive/common.c b/arch/powerpc/sysdev/xive/common.c
index 082c7e1c20f0..b4054091999a 100644
--- a/arch/powerpc/sysdev/xive/common.c
+++ b/arch/powerpc/sysdev/xive/common.c
@@ -283,6 +283,15 @@ static unsigned int xive_get_irq(void)
 	return irq;
 }
 
+static void xive_orphan_irq(unsigned int irq)
+{
+	struct xive_cpu *xc = __this_cpu_read(xive_cpu);
+
+	xc->cppr = 0xff;
+	out_8(xive_tima + xive_tima_offset + TM_CPPR, 0xff);
+	DBG_VERBOSE("orphan_irq: irq %d, adjusting CPPR to 0xff\n", irq);
+}
+
 /*
  * After EOI'ing an interrupt, we need to re-check the queue
  * to see if another interrupt is pending since multiple
@@ -1419,6 +1428,7 @@ bool __init xive_core_init(const struct xive_ops *ops, void __iomem *area, u32 o
 	xive_irq_priority = max_prio;
 
 	ppc_md.get_irq = xive_get_irq;
+	ppc_md.orphan_irq = xive_orphan_irq;
 	__xive_enabled = true;
 
 	pr_devel("Initializing host..\n");
-- 
2.17.1



More information about the Linuxppc-dev mailing list