Random soft offline test cases

Signed-off-by: Andi Kleen <ak@linux.intel.com>

4 files changed, 230 insertions, 1 deletions

diff --git a/tsrc/Makefile b/tsrc/Makefile
index 5f384d5..17da18d 100644
--- a/tsrc/Makefile
+++ b/tsrc/Makefile
@@ -5,7 +5,7 @@ CFLAGS += -I ${LSRC}/arch/x86/kernel/cpu/mcheck/ -g -Wall
 
 KFLAGS := -I ./kinclude
 
-EXE := tinjpage tinjpage-working tkillpoison tprctl
+EXE := tinjpage tinjpage-working tkillpoison tprctl tsoft tsoftinj
 EXEKERNEL := tring ttable
 
 OBJ := $(addsuffix .o,${EXE})
diff --git a/tsrc/random_offline b/tsrc/random_offline
new file mode 100755
index 0000000..3386abc
--- /dev/null
+++ b/tsrc/random_offline
@@ -0,0 +1,71 @@
+#!/bin/bash
+# randomly soft offline pages
+
+# fixme: uses time seed, non reproducible
+
+#mount -t debugfs none /debug
+
+THRESH=${1:-1000}
+
+end_of_memory() {
+	for i in /sys/firmware/memmap/* ; do
+		case "$(< $i/type)" in
+		"System RAM") ;;
+		*) continue ;;
+		esac
+	
+		k=$(< $i/end)
+		k=${k/0x/}
+		k=$(echo $k | tr a-z A-Z)
+
+		echo "ibase=16; $k/1000" | bc
+	done | sort -n | tail -n1
+}
+
+E=$(end_of_memory)
+
+echo "soft offlining pages upto $E" 
+
+unpoison() { 
+	if [ ! -f offlined ] ; then
+		return
+	fi
+	
+	echo unpoisioning
+	while read i ; do 
+		#echo -n ,
+		#echo "u $i"
+	 	echo $i | sed 's/000$//' > /debug/hwpoison/unpoison-pfn \
+			|| echo "unpoisioning $i failed: $?"
+	done < offlined
+	echo done
+	echo
+}
+
+trap unpoison 0
+
+(( k = 0 ))
+rm -f offlined
+while true ; do 
+	T=$( 
+	R=$RANDOM
+	X=$(echo "obase=16; ($R%$E)*4096"  | bc)
+	echo 0x$X
+	) 
+	#echo "p $T" 
+	if echo 2>/dev/null $T >/sys/devices/system/memory/soft_offline_page ; then
+		echo $T >> offlined
+	else
+		#echo offlining $T failed $?
+		true
+	fi
+	#echo -n . 
+
+	(( k++ )) 
+	if [ $k -gt $THRESH ] ; then
+		unpoison
+		(( k = 0 ))
+		rm offlined
+	fi
+done	
+
diff --git a/tsrc/tsoft.c b/tsrc/tsoft.c
new file mode 100644
index 0000000..4fbaa15
--- /dev/null
+++ b/tsrc/tsoft.c
@@ -0,0 +1,29 @@
+/* Simplest soft offline testcase */
+#include <stdlib.h>
+#include <sys/mman.h>
+#include <stdio.h>
+#include <unistd.h>
+
+#define err(x) perror(x), exit(1)
+
+#define MADV_SOFT_OFFLINE 101          /* soft offline page for testing */
+
+int PS;
+
+int main(void)
+{
+	PS = getpagesize();
+	char *map = mmap(NULL, PS,  PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANON, 0, 0);
+
+	if (map == (char *)-1L)
+		err("mmap");
+
+	*map = 1;
+
+	if (madvise(map, PS, MADV_SOFT_OFFLINE) < 0) 
+		perror("madvise SOFT_OFFLINE");
+	
+	*map = 2;
+
+	return 0;
+}
diff --git a/tsrc/tsoftinj.c b/tsrc/tsoftinj.c
new file mode 100644
index 0000000..e48c912
--- /dev/null
+++ b/tsrc/tsoftinj.c
@@ -0,0 +1,129 @@
+/* Test soft page offline for process pages using madvise injector */
+#define _GNU_SOURCE 1
+#include <sys/mman.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <sys/fcntl.h>
+#include <stdlib.h>
+#include <errno.h>
+
+#define MADV_SOFT_OFFLINE 101
+
+#define err(x) perror(x), exit(1)
+
+int PS;
+int exitcode;
+char empty[4096];
+int corrupted;
+
+void *checked_mmap(void *addr, size_t length, int prot, int flags,
+                  int fd, off_t offset)
+{
+	void *p = mmap(addr, length, prot, flags, fd, offset);
+	if (p == (void *)-1L)
+		err("mmap");
+	return p;
+}
+
+unsigned meminfo(char *fmt)
+{
+	int found = 0;
+	FILE *f = fopen("/proc/meminfo", "r");
+	if (!f) err("open /proc/meminfo");
+	char *line = NULL;
+	size_t linelen = 0;
+	unsigned val = 0;
+	while (getline(&line, &linelen, f) > 0) {
+		if (sscanf(line, fmt, &val) == 1) { 
+			found = 1;
+			break;
+		}
+	}
+	free(line);
+	fclose(f);
+	if (!found)  {
+		printf("cannot read HardwareCorruptedPages in meminfo\n");
+		exitcode = 1;
+	}
+	return val;
+}
+
+unsigned hardware_corrupted(void)
+{
+	return (meminfo("HardwareCorrupted: %u") * 1024) / PS;
+}
+
+char *ndesc(char *buf, char *a, char *b)
+{
+	snprintf(buf, 100, "%s %s", a, b);
+	return buf;
+}
+
+void offline(char *name, void *p)
+{
+	char buf[100];
+	if (madvise(p, PS, MADV_SOFT_OFFLINE) < 0)
+		err(ndesc(buf, name, "offline"));
+	corrupted++;
+}
+
+void disk_backed(char *name, int flags)
+{
+	char fn[100];
+	snprintf(fn, sizeof fn, "t%u", getpid());
+	printf("shared, diskbacked\n");
+	int fd = open(fn, O_RDWR|O_CREAT|O_TRUNC, 0644);
+	if (fd < 0) err("open tmpfile");
+	write(fd, empty, sizeof empty);
+	char *p = checked_mmap(NULL, PS, PROT_READ|PROT_WRITE, 
+			MAP_SHARED|flags, fd, 0);
+	*(volatile int *)p = 1;
+	offline(ndesc(fn, "disk backed", name), p);
+	munmap(p, PS);
+}
+
+void anonymous(char *name, int flags)
+{
+	char buf[100];
+	char *p = checked_mmap(NULL, PS, PROT_READ|PROT_WRITE, 
+			MAP_PRIVATE|MAP_ANONYMOUS|flags, 0, 0);
+	printf("anonymous\n");
+	*(volatile int *)p = 1;
+	offline(ndesc(buf, "anonymous", name), p);
+	*(volatile int *)p = 1;
+	munmap(p, PS);
+}
+
+void check(unsigned *count, char *name)
+{
+	unsigned count2 = hardware_corrupted();
+	unsigned diff = count2 - *count;
+	if (diff != corrupted) {
+		printf("%s: expected %d corrupted pages, got %u\n", name,
+			corrupted, 
+			diff);	
+		if (diff < corrupted)
+			exitcode = 1;
+	}
+	*count = count2;
+	corrupted = 0;
+}
+
+int main(void)
+{
+	PS = getpagesize();
+
+	unsigned count = hardware_corrupted();
+	
+	anonymous("anonymous", 0);	
+	check(&count, "anonymous");
+	anonymous("anonymous mlock", MAP_LOCKED);	
+	check(&count, "anonymous mlock");
+	disk_backed("disk backed", 0);
+	check(&count, "disk backed");
+	disk_backed("disk backed mlock", 0);
+	check(&count, "disk backed mlock");
+	// add more test cases here
+
+	return exitcode;
+}