Skip to contentSkip to conversation

powerpc/icp-opal: Fix missing KVM case and harden replay

PATCH In review Benjamin Herrenschmidt 2 participants 3 messages Jan 16, 2017 PowerPC

lore
Share & tools
lore

Benjamin Herrenschmidt

author
#0 lore raw

PATCH 1/1 powerpc/icp-opal: Fix missing KVM case and harden replay

The icp-opal call is missing the code from icp-native to recover interrupts snatched by KVM. Without that, when running KVM, we can get into a situation where an interrupt is lost and the CPU stuck with an elevated CPPR.

Also harden replay by always checking the return from opal_int_eoi

Signed-off-by Benjamin Herrenschmidt <[email protected]>
---
 arch/powerpc/sysdev/xics/icp-opal.c | 31 ++++++++++++++++++++++++-------
 1 file changed, 24 insertions(+), 7 deletions(-)
arch/powerpc/sysdev/xics/icp-opal.c +24 -7
index d38e86f..60c5765 100644
@@ -20,6 +20,7 @@
 #include 
 #include 
 #include 
+#include 
 
 static void icp_opal_teardown_cpu(void)
 {

@@ -39,7 +40,26 @@ static void icp_opal_flush_ipi(void)
 	 * Should we be flagging idle loop instead?
 	 * Or creating some task to be scheduled?
 	 */
-	opal_int_eoi((0x00 << 24) | XICS_IPI);
+	if (opal_int_eoi((0x00 << 24) | XICS_IPI) > 0)
+		force_external_irq_replay();
+}
+
+static unsigned int icp_opal_get_xirr(void)
+{
+	unsigned int kvm_xirr;
+	__be32 hw_xirr;
+	int64_t rc;
+
+	/* Handle an interrupt latched by KVM first */
+	kvm_xirr = kvmppc_get_xics_latch();
+	if (kvm_xirr)
+		return kvm_xirr;
+
+	/* Then ask OPAL */
+	rc = opal_int_get_xirr(&hw_xirr, false);
+	if (rc < 0)
+		return 0;
+	return be32_to_cpu(hw_xirr);
 }
 
 static unsigned int icp_opal_get_irq(void)

@@ -47,12 +67,8 @@ static unsigned int icp_opal_get_irq(void)
 	unsigned int xirr;
 	unsigned int vec;
 	unsigned int irq;
-	int64_t rc;
 
-	rc = opal_int_get_xirr(&xirr, false);
-	if (rc < 0)
-		return 0;
-	xirr = be32_to_cpu(xirr);
+	xirr = icp_opal_get_xirr();
 	vec = xirr & 0x00ffffff;
 	if (vec == XICS_IRQ_SPURIOUS)
 		return 0;

@@ -67,7 +83,8 @@ static unsigned int icp_opal_get_irq(void)
 	xics_mask_unknown_vec(vec);
 
 	/* We might learn about it later, so EOI it */
-	opal_int_eoi(xirr);
+	if (opal_int_eoi(xirr) > 0)
+		force_external_irq_replay();
 
 	return 0;
 }
#1 lore raw
Quoted 17 lines · Benjamin Herrenschmidt

The icp-opal call is missing the code from icp-native to recover interrupts snatched by KVM. Without that, when running KVM, we can get into a situation where an interrupt is lost and the CPU stuck with an elevated CPPR.

Also harden replay by always checking the return from opal_int_eoi

Signed-off-by Benjamin Herrenschmidt <[email protected]>
---
 arch/powerpc/sysdev/xics/icp-opal.c | 31 ++++++++++++++++++++++++-------
 1 file changed, 24 insertions(+), 7 deletions(-)
arch/powerpc/sysdev/xics/icp-opal.c +0 -0
index d38e86f..60c5765 100644

arch/powerpc/sysdev/xics/icp-opal.c line 24

Quoted 5 lines
arch/powerpc/sysdev/xics/icp-opal.c +1 -0
@@ -20,6 +20,7 @@
 #include 
 #include 
 #include 
+#include

??

<snip>

arch/powerpc/sysdev/xics/icp-opal.c static void icp_opal_flush_ipi(void) · line 45

Quoted 8 lines
arch/powerpc/sysdev/xics/icp-opal.c +3 -1
@@ -39,7 +40,26 @@ static void icp_opal_flush_ipi(void)
 	 * Should we be flagging idle loop instead?
 	 * Or creating some task to be scheduled?
 	 */
-	opal_int_eoi((0x00 << 24) | XICS_IPI);
+	if (opal_int_eoi((0x00 << 24) | XICS_IPI) > 0)
+		force_external_irq_replay();
+}

Shouldn't we also update kvm bits on icp_native_ipi_action and icp_native_cause_ipi?

Balbir Singh.

author
#2 lore raw
Quoted 5 lines · Balbir Singh

??

<snip>

Quoted 1 line

@@ -39,7 +40,2

Ah evolution bug ... fixed upstream but fix hasn't made it to Fedora yet.

I'll attach in the meantime.

Cheers,
Ben.