File kdump-treat-KDUMP_CPUS-0-as-all-available-CPUs.... of Package kdump

Overview Repositories Revisions Requests Users Attributes Meta

File kdump-treat-KDUMP_CPUS-0-as-all-available-CPUs.patch of Package kdump (Revision 4cef3c678fc05e46b25db82dd173dc19)

Currently displaying revision 4cef3c678fc05e46b25db82dd173dc19 , Show latest

From: Petr Tesarik <ptesarik@suse.com>
Date: Fri Jun 23 13:43:15 2017 +0200
Subject: Treat KDUMP_CPUS=0 as "all available CPUs"
References: bsc#1036223
Upstream: v0.8.17
Git-commit: 34192abe9a0f5ddbcd580fa52b7fa4400533b9a4

Since SMP after crash is now considered stable, remove the warnings and
make the feature more useful by not requiring a fixed number of CPUs.

Signed-off-by: Petr Tesarik <ptesarik@suse.com>

--- a/doc/man/kdump.5.txt.in
+++ b/doc/man/kdump.5.txt.in
@@ -136,13 +136,12 @@ the number of CPUs in your system.
 This parameter modifies the _maxcpus_ parameter of the kdump kernel and
 the number of *makedumpfile*(8) processes created.
 
+If the value is zero, all available CPUs are used, i.e. the _maxcpus_
+parameter is not added to the kdump kernel command line.
+
 *Note:* This parameter does not work properly for the _ELF_ format,
 because *makedumpfile*(8) does not support split _ELF_ dump files.
 
-*WARNING:* SETTING THIS NUMBER TO A NUMBER HIGHER THAN 1 IS EXPERIMENTAL!
-SOME MACHINES ARE KNOWN TO HANG OR REBOOT RANDOMLY DURING INITIALISATION
-OF THE DUMP KERNEL IF KDUMP_CPUS > 1.
-
 Default is 1.
 
 
--- a/init/load.sh
+++ b/init/load.sh
@@ -63,14 +63,17 @@ function build_kdump_commandline()
     local commandline="$KDUMP_COMMANDLINE"
 
     if [ -z "$commandline" ] ; then
-        local nr_cpus=$(cpus_param "$kdump_kernel")
+        local nr_cpus
         commandline=$(
             remove_from_commandline \
                 'root|resume|crashkernel|splash|mem|BOOT_IMAGE|showopts|zfcp\.allow_lun_scan|hugepages|acpi_no_memhotplug|cgroup_disable|unknown_nmi_panic|rd\.udev\.children-max' \
                 < /proc/cmdline)
+        if [ ${KDUMP_CPUS:-1} -ne 0 ] ; then
+            nr_cpus=$(cpus_param "$kdump_kernel")=${KDUMP_CPUS:-1}
+        fi
         # Use deadline for saving the memory footprint
         commandline="$commandline elevator=deadline sysrq=yes reset_devices acpi_no_memhotplug cgroup_disable=memory"
-        commandline="$commandline irqpoll ${nr_cpus}=${KDUMP_CPUS:-1}"
+        commandline="$commandline irqpoll ${nr_cpus}"
         commandline="$commandline root=kdump rootflags=bind rd.udev.children-max=8"
         case $(uname -i) in
         i?86|x86_64)
--- a/kdumptool/calibrate.cc
+++ b/kdumptool/calibrate.cc
@@ -905,10 +905,10 @@ void Calibrate::execute()
 	}
 
 	// Add memory based on CPU count
-	unsigned long cpus;
-	if (CAN_REDUCE_CPUS) {
+	unsigned long cpus = 0;
+	if (CAN_REDUCE_CPUS)
 	    cpus = config->KDUMP_CPUS.value();
-	} else {
+        if (!cpus) {
 	    SystemCPU syscpu;
 	    unsigned long online = syscpu.numOnline();
 	    unsigned long offline = syscpu.numOffline();
--- a/kdumptool/savedump.cc
+++ b/kdumptool/savedump.cc
@@ -41,6 +41,7 @@
 #include "identifykernel.h"
 #include "email.h"
 #include "routable.h"
+#include "calibrate.h"
 
 using std::string;
 using std::list;
@@ -60,7 +61,7 @@ using std::ifstream;
 SaveDump::SaveDump()
     throw ()
     : m_dump(DEFAULT_DUMP), m_transfer(NULL), m_usedDirectSave(false),
-      m_useMakedumpfile(false), m_useSplit(false), m_nomail(false)
+      m_useMakedumpfile(false), m_split(0), m_nomail(false)
 {
     Debug::debug()->trace("SaveDump::SaveDump()");
 
@@ -284,10 +285,15 @@ void SaveDump::saveDump(const RootDirURL
     if (noDump)
 	return;			// nothing to be done
 
+    unsigned long cpus = config->KDUMP_CPUS.value();
+    if (cpus) {
+        SystemCPU syscpu;
+        cpus = syscpu.numOnline();
+    }
     if (!config->kdumptoolContainsFlag("NOSPLIT") &&
-        config->KDUMP_CPUS.value() > 1) {
+        cpus > 1) {
 	if (!useElf)
-	    m_useSplit = true;
+	    m_split = cpus;
 	else
 	    cerr << "Splitting ELF dumps is not supported." << endl;
     }
@@ -305,7 +311,7 @@ void SaveDump::saveDump(const RootDirURL
         // use makedumpfile
         ostringstream cmdline;
         cmdline << "makedumpfile ";
-	if (m_useSplit)
+	if (m_split)
 	    cmdline << "--split ";
         cmdline << config->MAKEDUMPFILE_OPTIONS.value() << " ";
         cmdline << "-d " << config->KDUMP_DUMPLEVEL.value() << " ";
@@ -340,10 +346,9 @@ void SaveDump::saveDump(const RootDirURL
             provider->setProgress(&progress);
         else
             cout << "Saving dump ..." << endl;
-	if (m_useSplit) {
+	if (m_split) {
 	    StringVector targets;
-            const int cpus = config->KDUMP_CPUS.value();
-	    for (int i = 1; i <= cpus; ++i) {
+	    for (unsigned long i = 1; i <= m_split; ++i) {
 		ostringstream ss;
 		ss << "vmcore" << i;
 		targets.push_back(ss.str());
@@ -482,8 +487,8 @@ void SaveDump::generateInfo()
     ss << "Dump level     : "
        << Stringutil::number2string(config->KDUMP_DUMPLEVEL.value()) << endl;
     ss << "Dump format    : " << config->KDUMP_DUMPFORMAT.value() << endl;
-    if (m_useSplit && m_usedDirectSave)
-        ss << "Split parts    : " << config->KDUMP_CPUS.value() << endl;
+    if (m_split && m_usedDirectSave)
+        ss << "Split parts    : " << m_split << endl;
     ss << endl;
 
 
--- a/kdumptool/savedump.h
+++ b/kdumptool/savedump.h
@@ -112,7 +112,7 @@ class SaveDump : public Subcommand {
         Transfer *m_transfer;
         bool m_usedDirectSave;
         bool m_useMakedumpfile;
-	bool m_useSplit;
+	unsigned long m_split;
         std::string m_crashtime;
         std::string m_crashrelease;
         std::string m_rootdir;
--- a/sysconfig.kdump.in
+++ b/sysconfig.kdump.in
@@ -18,13 +18,8 @@ KDUMP_KERNELVER=""
 #
 # Number of CPUs to be used in the kdump environment. You may want to
 # increase the number if computing power is the bottleneck in your setup.
-# This value is used as the maxcpus parameter of the secondary kernel, so
-# DON'T CHANGE IT unless your system can reliably re-initialize SMP after
-# a kernel crash.
 #
-# WARNING: SETTING THIS NUMBER TO A NUMBER HIGHER THAN 1 IS EXPERIMENTAL!
-# SOME MACHINES ARE KNOWN TO HANG OR REBOOT RANDOMLY DURING INITIALISATION
-# OF THE DUMP KERNEL IF KDUMP_CPUS > 1.
+# If the value is zero, use all available CPUs.
 #
 # See also: kdump(5).
 #

Places

File kdump-treat-KDUMP_CPUS-0-as-all-available-CPUs.patch of Package kdump (Revision 4cef3c678fc05e46b25db82dd173dc19)

Places