[PATCH] intel_idle: prevent SKX boot failure when C6 & SERIRQ enabled

[Stephen Douthit]

Interrupts are getting misrouted and/or dropped on SKLYLAKE_X based D-2100s
when C6 and SERIRQ are enabled.  I've only seen this issue on systems
using SERIRQs (in my case for a LPC based UART providing the serial
console for a headless server).

One failure mode is "do_IRQ: 8.33 No irq handler for vector" getting
printed in the kernel logs. The core getting the unhandled irq is typically
the one handling the UART SERIRQ.  I've seen it on other cores, but I
haven't confirmed if that's because the UART irq handler was moved to
another core at some point.  The vector varies from 33-36, but it's most
often 33.

The other failure mode is the system hanging.  Sometimes forcing some non
SERIRQ interrupt to fire (by plugging/unplugging a network/USB cable) can
get the system out of this state.  Generating more SERIRQs via the UART
will not unstick the system.

Both failures seemed to occur when transition to a low load state, which
is why I started playing around with power management options and found
that booting with "intel_idle.max_cstate=2" fixed the issue.

This patch only disables C6 if it's able to determine that SERIRQs are
enabled by checking the enable bit in the LPC controllers PCI config space.

Signed-off-by: Stephen Douthit <stephend@xxxxxxxxxxxxxxx>
---
 drivers/idle/intel_idle.c | 35 ++++++++++++++++++++++++++++++++++-
 include/linux/pci_ids.h   |  1 +
 2 files changed, 35 insertions(+), 1 deletion(-)

diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c
index b8647b5c3d4d..353f6a9b1818 100644
--- a/drivers/idle/intel_idle.c
+++ b/drivers/idle/intel_idle.c
@@ -61,12 +61,13 @@
 #include <linux/notifier.h>
 #include <linux/cpu.h>
 #include <linux/moduleparam.h>
+#include <linux/pci.h>
 #include <asm/cpu_device_id.h>
 #include <asm/intel-family.h>
 #include <asm/mwait.h>
 #include <asm/msr.h>
 
-#define INTEL_IDLE_VERSION "0.4.1"
+#define INTEL_IDLE_VERSION "0.4.2"
 
 static struct cpuidle_driver intel_idle_driver = {
 	.name = "intel_idle",
@@ -1306,6 +1307,35 @@ static void sklh_idle_state_table_update(void)
 	skl_cstates[5].disabled = 1;	/* C8-SKL */
 	skl_cstates[6].disabled = 1;	/* C9-SKL */
 }
+/*
+ * skx_idle_state_table_update()
+ *
+ * On SKX (model 0x55) SoCs disable C6 if SERIRQ is enabled
+ */
+static void skx_idle_state_table_update(void)
+{
+#define SCNT_OFF 0x64
+#define SCNT_EN (1 << 7)
+	struct pci_dev *pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
+					      PCI_DEVICE_ID_INTEL_SKX_LPC,
+					      NULL);
+	u8 reg;
+
+	/*
+	 * Check bit 7 of the Serial IRQ Control (SCNT) register (0x64) in the
+	 * LPC controller.  If it's set serial IRQs are enabled, and we need to
+	 * disable C6 to prevent hangs.
+	 */
+	if (!pdev)
+		return;
+	if (pci_read_config_byte(pdev, SCNT_OFF, &reg))
+		return;
+	if (!(reg & SCNT_EN))
+		return;
+
+	pr_debug("SERIRQ enabled on SKX, disabling C6 to avoid hangs\n");
+	skx_cstates[2].disabled = 1;	/* C6-SKX */
+}
 /*
  * intel_idle_state_table_update()
  *
@@ -1326,6 +1356,9 @@ static void intel_idle_state_table_update(void)
 	case INTEL_FAM6_SKYLAKE_DESKTOP:
 		sklh_idle_state_table_update();
 		break;
+	case INTEL_FAM6_SKYLAKE_X:
+		skx_idle_state_table_update();
+		break;
 	}
 }
 
diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
index 70e86148cb1e..02bac8de03fd 100644
--- a/include/linux/pci_ids.h
+++ b/include/linux/pci_ids.h
@@ -2997,6 +2997,7 @@
 #define PCI_DEVICE_ID_INTEL_84460GX	0x84ea
 #define PCI_DEVICE_ID_INTEL_IXP4XX	0x8500
 #define PCI_DEVICE_ID_INTEL_IXP2800	0x9004
+#define PCI_DEVICE_ID_INTEL_SKX_LPC	0xa1c8
 #define PCI_DEVICE_ID_INTEL_S21152BB	0xb152
 
 #define PCI_VENDOR_ID_SCALEMP		0x8686
-- 
2.21.0