chengshan / rpms / kernel

Forked from rpms/kernel 2 years ago
Clone
Blob Blame History Raw
centosplus kernel patch [bug#12818]

From ac68351410df683a149e7eea09df178884cf2419 Mon Sep 17 00:00:00 2001
From: Marcus Sundberg <marcus.sundberg@aptilo.com>
Date: Mon, 13 Feb 2017 22:57:11 +0100
Subject: [PATCH] [x86] perf: uncore: Avoid kernel panic on missing topology

The uncore code is heavily dependent on the CPU topology mapping code,
so if we do not even have a working topology for the boot CPU we just
give up.
This happens for example on HPE DL360 Gen9 servers with x2APIC enabled
in the firmware settings, where the kernel only finds a single core.
---
 arch/x86/events/intel/uncore.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/arch/x86/events/intel/uncore.c b/arch/x86/events/intel/uncore.c
index 77ec6c0..88d79f6 100644
--- a/arch/x86/events/intel/uncore.c
+++ b/arch/x86/events/intel/uncore.c
@@ -1394,6 +1394,20 @@ static int __init intel_uncore_init(void)
 	if (is_kdump_kernel())
 		return -ENODEV;
 
+	/* Sanity check - do not even try to proceed if we have no
+	   working topology mapping for the boot CPU. */
+	if (rh_boot_cpu_data.logical_proc_id < 0 ||
+	    topology_phys_to_logical_pkg(boot_cpu_data.phys_proc_id) < 0) {
+		printk(KERN_ERR
+		       "no mapping between physical and logical CPU, "
+		       "boot CPU phys: %d, logical: %d/%d\n",
+		       (int)boot_cpu_data.phys_proc_id,
+		       (int)rh_boot_cpu_data.logical_proc_id,
+		       (int)topology_phys_to_logical_pkg(
+			       boot_cpu_data.phys_proc_id));
+		return -ENODEV;
+	}
+
 	max_packages = topology_max_packages() * topology_max_die_per_package();
 
 	pret = uncore_pci_init();
-- 
2.5.0