tsrc/tinjpage.c

/*
 * Test program for Linux poison memory error recovery.
 * This injects poison into various mapping cases and triggers the poison
 * handling.  Requires special injection support in the kernel.
 * 
 * Copyright 2009, 2010 Intel Corporation
 *
 * tinjpage is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License as published by the Free Software Foundation; version
 * 2.
 *
 * tinjpage is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should find a copy of v2 of the GNU General Public License somewhere
 * on your Linux system; if not, write to the Free Software Foundation, 
 * Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 
 *
 * Authors: Andi Kleen, Fengguang Wu
 */
#define _GNU_SOURCE 1
#include <stdio.h>
#include <signal.h>
#include <unistd.h>
#include <sys/fcntl.h>
#include <sys/wait.h>
#include <sys/mman.h>
#include <stdlib.h>
#include <setjmp.h>
#include <errno.h>
#include <string.h>
#include <time.h>
#include <pthread.h>
#include <sys/ipc.h>
#include <sys/shm.h>
#include <sys/sem.h>
#include "utils.h"
#include "hugepage.h"

#define MADV_POISON 100

#define TMPDIR "./"
#define PATHBUFLEN 100

#define Perror(x) failure++, perror(x)
#define PAIR(x) x, sizeof(x)-1
#define mb() asm volatile("" ::: "memory")
#if defined(__i386__) || defined(__x86_64__)
#define cpu_relax() asm volatile("rep ; nop" ::: "memory")
#else
#define cpu_relax() mb()
#endif

typedef unsigned long long u64;

int PS;
int failure;
int unexpected;
int early_kill;
int test_hugepage;

void *checked_mmap(void *start, size_t length, int prot, int flags,
                   int fd, off_t offset)
{
	void *map = mmap(start, length, prot, flags, fd, offset);
	if (map == (void*)-1L)
		err("mmap");
	return map;
}

void munmap_reserve(void *page, int size)
{
	if (munmap(page, size) < 0)
		err("munmap");
	if (mmap(page, size, PROT_NONE, MAP_PRIVATE|MAP_FIXED, 0, 0) < 0)
		err("mmap2");
}

void *xmalloc(size_t s)
{
	void *p = malloc(s);
	if (!p)
		exit(ENOMEM);
	return p;
}

static int ilog2(int n)
{
	int r = 0;
	n--;
	while (n) {
		n >>= 1;
		r++;
	}
	return r;
}

int recovercount;
sigjmp_buf recover_ctx;
sigjmp_buf early_recover_ctx;
void *expected_addr;

/* Work around glibc not defining this yet */
struct my_siginfo {
	int si_signo;
	int si_errno;
	int si_code;
	union {
	struct {
		void  *_addr; /* faulting insn/memory ref. */
#ifdef __ARCH_SI_TRAPNO
		int _trapno;	/* TRAP # which caused the signal */
#endif
		short _addr_lsb; /* LSB of the reported address */
	} _sigfault;
	} _sifields;
};
#undef si_addr_lsb
#define si_addr_lsb _sifields._sigfault._addr_lsb

void sighandler(int sig, siginfo_t *si, void *arg)
{
	if (si->si_addr != expected_addr) {
		printf("XXX: Unexpected address in signal %p (expected %p)\n", si->si_addr,
			expected_addr);
		failure++;
	}

	int lsb = ((struct my_siginfo *)si)->si_addr_lsb;
	if (test_hugepage) {
		if (lsb != ilog2(HPS)) {
			printf("LATER: Unexpected addr lsb in siginfo %d\n", lsb);
		}
	} else {
		if (lsb != ilog2(sysconf(_SC_PAGE_SIZE))) {
			printf("LATER: Unexpected addr lsb in siginfo %d\n", lsb);
		}
	}

	printf("\tsignal %d code %d addr %p\n", sig, si->si_code, si->si_addr);

	if (--recovercount == 0) {
		write(1, PAIR("I seem to be in a signal loop. bailing out.\n"));
		exit(1);
	}

	if (si->si_code == 4)
		siglongjmp(recover_ctx, 1);
	else
		siglongjmp(early_recover_ctx, 1);
}

enum rmode {
	MREAD = 0,
	MWRITE = 1,
	MREAD_OK = 2,
	MWRITE_OK = 3,
	MNOTHING = -1,
};

void inject_madvise(char *page)
{
	if (madvise(page, PS, MADV_POISON) != 0) {
		if (errno == EINVAL) {
			printf("Kernel doesn't support poison injection\n");
			exit(0);
		}
		Perror("madvise");
	}
}

u64 page_to_pfn(char *page)
{
	static int pagemap_fd = -1;
	u64 pfn;

	if (pagemap_fd < 0)  {
		pagemap_fd = open("/proc/self/pagemap", O_RDONLY); 
		if (pagemap_fd < 0)
			err("/proc/self/pagemap not supported");
	}

	if (pread(pagemap_fd, &pfn, sizeof(u64), 
		((u64)page / PS)*sizeof(u64)) != sizeof(u64))
		err("Cannot read from pagemap");

	pfn &= (1ULL<<56)-1; 
	return pfn;
}

/* 
 * Inject Action Optional #MC 
 * with mce-inject using the software injector.
 * 
 * This tests the low level machine check handler too.
 * 
 * Slightly racy with page migration because we don't mlock the page.
 */
void inject_mce_inject(char *page)
{
	u64 pfn = page_to_pfn(page);
	FILE *mce_inject;

	mce_inject = popen("mce-inject", "w");
	if (!mce_inject) {
		fprintf(stderr, "Cannot open pipe to mce-inject: %s\n",
				strerror(errno));
		exit(1);
	}

	fprintf(mce_inject, 
		"CPU 0 BANK 3 STATUS UNCORRECTED SRAO 0xc0\n"
		"MCGSTATUS RIPV MCIP\n"
		"ADDR %#llx\n"
		"MISC 0x8c\n"
		"RIP 0x73:0x1eadbabe\n", pfn);

	if (ferror(mce_inject) || fclose(mce_inject) < 0) { 
		fprintf(stderr, "mce-inject failed: %s\n", strerror(errno));
		exit(1);
	} 
}

void (*inject)(char *page) = inject_madvise;

void poison(char *msg, char *page, enum rmode mode)
{
	expected_addr = page;
	recovercount = 5;

	if (sigsetjmp(early_recover_ctx, 1) == 0) {
		inject(page);
		
		if (early_kill && (mode == MWRITE || mode == MREAD)) {
			printf("XXX: %s: process is not early killed\n", msg);
			failure++;
		}

		return;
	}

	if (early_kill) {
		if (mode == MREAD_OK || mode == MWRITE_OK) {
			printf("XXX: %s: killed\n", msg);
			failure++;
		} else
			printf("\trecovered\n");
	}
}

void recover(char *msg, char *page, enum rmode mode)
{
	expected_addr = page;
	recovercount = 5;

	if (sigsetjmp(recover_ctx, 1) == 0) {
		switch (mode) {
		case MWRITE:
			printf("\twriting 2\n");
			*page = 2;
			break;
		case MWRITE_OK:
			printf("\twriting 4\n");
			*page = 4;
			return;
		case MREAD:
			printf("\treading %x\n", *(unsigned char *)page);
			break;
		case MREAD_OK:
			printf("\treading %x\n", *(unsigned char *)page);
			return;
		case MNOTHING:
			return;
		}
		/* signal or kill should have happened */
		printf("XXX: %s: page not poisoned after injection\n", msg);
		failure++;
		return;
	}
	if (mode == MREAD_OK || mode == MWRITE_OK) {
		printf("XXX: %s: killed\n", msg);
		failure++;
	} else
		printf("\trecovered\n");
}

void testmem(char *msg, char *page, enum rmode mode)
{
	printf("\t%s poisoning page %p\n", msg, page);
	poison(msg, page, mode);
	recover(msg, page, mode);
}

void expecterr(char *msg, int err)
{
	if (err) {
		printf("\texpected error %d on %s\n", errno, msg);
	} else {
		failure++;
		printf("XXX: unexpected no error on %s\n", msg);
	}
}

/* 
 * Any optional error is really a deficiency in the kernel VFS error reporting
 * and should be eventually fixed and turned into a expecterr
 */
void optionalerr(char *msg, int err)
{
	if (err) {
		printf("\texpected optional error %d on %s\n", errno, msg);
	} else {
		unexpected++;
		printf("LATER: expected likely incorrect no error on %s\n", msg);
	}
}

static int tmpcount;
int tempfd(void)
{
	int fd;
	char buf[PATHBUFLEN];
	snprintf(buf, sizeof buf, TMPDIR "~poison%d",tmpcount++);
	fd = open(buf, O_CREAT|O_RDWR, 0600);
	if (fd >= 0)
		unlink(buf);
	if (fd < 0)
		err("opening temporary file in " TMPDIR);
	return fd;
}

int playfile(char *buf)
{
	int fd;
	if (buf[0] == 0)
		snprintf(buf, PATHBUFLEN, TMPDIR "~poison%d", tmpcount++);
	fd = open(buf, O_CREAT|O_RDWR|O_TRUNC, 0600);
	if (fd < 0)
		err("opening temporary file in " TMPDIR);

	const int NPAGES = 5;
	char *tmp = xmalloc(PS * NPAGES);
	int i;
	for (i = 0; i < PS*NPAGES; i++)
		tmp[i] = i;
	write(fd, tmp, PS*NPAGES);

	lseek(fd, 0, SEEK_SET);
	return fd;
}

static void dirty_anonymous(void)
{
	char *page;
	page = checked_mmap(NULL, PS, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_POPULATE, 0, 0);
	testmem("dirty", page, MWRITE);
}

static void dirty_anonymous_unmap(void)
{
	char *page;
	page = checked_mmap(NULL, PS, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_POPULATE, 0, 0);
	testmem("dirty", page, MWRITE);
	munmap_reserve(page, PS);
}

static void mlocked_anonymous(void)
{
	char *page;
	page = checked_mmap(NULL, PS, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_LOCKED, 0, 0);
	testmem("mlocked", page, MWRITE);
}

static void do_file_clean(int flags, char *name)
{
	char *page;
	char fn[30];
	snprintf(fn, 30, TMPDIR "~test%d", tmpcount++);
	int fd = open(fn, O_RDWR|O_TRUNC|O_CREAT);
	if (fd < 0)
		err("open temp file");
	write(fd, fn, 4);
	page = checked_mmap(NULL, PS, PROT_READ|PROT_WRITE, MAP_SHARED|flags, 
		fd, 0);
	fsync(fd);
	close(fd);
	testmem(name, page, MREAD_OK);
	 /* reread page from disk */
	printf("\t reading %x\n", *(unsigned char *)page);	
	testmem(name, page, MWRITE_OK);
}

static void file_clean(void)
{
	do_file_clean(0, "file clean");
}

static void file_clean_mlocked(void)
{
	do_file_clean(MAP_LOCKED, "file clean mlocked");
}

static char *ndesc(char *buf, char *name, char *add)
{
	snprintf(buf, 100, "%s %s", name, add);
	return buf;
}

static void do_file_dirty(int flags, char *name)
{
	char nbuf[100];
	char *page;
	char fn[PATHBUFLEN];
	fn[0] = 0;
	int fd = playfile(fn);

	page = checked_mmap(NULL, PS, PROT_READ, 
			MAP_SHARED|MAP_POPULATE|flags, fd, 0);
	testmem(ndesc(nbuf, name, "initial"), page, MREAD);
	expecterr("msync expect error", msync(page, PS, MS_SYNC) < 0);
	close(fd);
	munmap_reserve(page, PS);

	fd = open(fn, O_RDONLY);
	if (fd < 0) err("reopening temp file");
	page = checked_mmap(NULL, PS, PROT_READ, MAP_SHARED|MAP_POPULATE|flags, 
				fd, 0);
	recover(ndesc(nbuf, name, "populated"), page, MREAD_OK);
	close(fd);
	munmap_reserve(page, PS);

	fd = open(fn, O_RDONLY);
	if (fd < 0) err("reopening temp file");
	page = checked_mmap(NULL, PS, PROT_READ, MAP_SHARED|flags, fd, 0);
	recover(ndesc(nbuf, name, "fault"), page, MREAD_OK);
	close(fd);
	munmap_reserve(page, PS);

	fd = open(fn, O_RDWR);
	char buf[128];
	/* the earlier close has eaten the error */
	optionalerr("explicit read after poison", read(fd, buf, sizeof buf) < 0);
	optionalerr("explicit write after poison", write(fd, "foobar", 6) < 0);
	optionalerr("fsync expect error", fsync(fd) < 0);
	close(fd);

	/* should unlink return an error here? */
	if (unlink(fn) < 0)
		perror("unlink");
}

static void file_dirty(void)
{
	do_file_dirty(0, "file dirty");
}

static void file_dirty_mlocked(void)
{
	do_file_dirty(MAP_LOCKED, "file dirty mlocked");
}

/* TBD */
static void file_hole(void)
{
	int fd = tempfd();
	char *page;

	ftruncate(fd, PS);
	page = checked_mmap(NULL, PS, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
	*page = 1;
	testmem("hole file dirty", page, MREAD);
	/* hole error reporting doesn't work in kernel currently, so optional */
	optionalerr("hole fsync expect error", fsync(fd) < 0);
	optionalerr("hole msync expect error", msync(page, PS, MS_SYNC) < 0);
	close(fd);
}

static void nonlinear(void)
{
	int fd;
	const int NPAGES = 10;
	int i;
	char *page;
	char *tmp;

	fd = tempfd();
	tmp = xmalloc(PS);
	for (i = 0; i < NPAGES; i++)  {
		memset(tmp, i, PS);
		write(fd, tmp, PS);
	}
	free(tmp);
	page = checked_mmap(NULL, PS*NPAGES, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
	int k = NPAGES - 1;
	for (i = 0; i < NPAGES; i++, k--) {
		if (remap_file_pages(page + i*PS, PS, 0, k, 0))
			perror("remap_file_pages");
	}
	*page = 1;
	testmem("rfp file dirty", page, MREAD);
	expecterr("rfp fsync expect error", fsync(fd) < 0);
	optionalerr("rfp msync expect error", msync(page, PS, MS_SYNC) < 0);
	close(fd);
}

/* 
 * These tests are currently too racy to be enabled.
 */

/*
 * This is quite timing dependent. The sniper might hit the page
 * before it is dirtied. If that happens tweak the delay
 * (should auto tune)
 */
enum {
	DELAY_NS = 30,
};

volatile enum sstate { START, WAITING, SNIPE } sstate;

void waitfor(enum sstate w, enum sstate s)
{
	sstate = w;
	mb();
	while (sstate != s)
		cpu_relax();
}

struct poison_arg {
	char *msg;
	char *page;
	enum rmode mode;
};

void *sniper(void *p)
{
	struct poison_arg *arg = p;

	waitfor(START, WAITING);
	nanosleep(&((struct timespec) { .tv_nsec = DELAY_NS }), NULL);
	poison(arg->msg, arg->page, arg->mode);
	return NULL;
}

int setup_sniper(struct poison_arg *arg)
{
	if (sysconf(_SC_NPROCESSORS_ONLN) < 2)  {
		printf("%s: Need at least two CPUs. Not tested\n", arg->msg);
		return -1;
	}
	sstate = START;
	mb();
	pthread_t thr;
	if (pthread_create(&thr, NULL, sniper, arg) < 0)
		err("pthread_create");
	pthread_detach(thr);
	return 0;
}

static void under_io_dirty(void)
{
	struct poison_arg arg;
	int fd = tempfd();
	char *page;

	page = checked_mmap(NULL, PS, PROT_READ|PROT_WRITE, MAP_SHARED|MAP_POPULATE, fd, 0);

	arg.page = page;
	arg.msg  = "under io dirty";
	arg.mode = MWRITE;
	if (setup_sniper(&arg) < 0)
		return;

	write(fd, "xyz", 3);
	waitfor(WAITING, WAITING);
	expecterr("write under io", fsync(fd) < 0);
	close(fd);
}

static void under_io_clean(void)
{
	struct poison_arg arg;
	char fn[PATHBUFLEN];
	int fd;
	char *page;
	char buf[10];

 	fd = playfile(fn);
	page = checked_mmap(NULL, PS, PROT_READ|PROT_WRITE, MAP_SHARED|MAP_POPULATE, fd, 0);
	madvise(page, PS, MADV_DONTNEED);

	arg.page = page;
	arg.msg  = "under io clean";
	arg.mode = MREAD_OK;
	if (setup_sniper(&arg) < 0)
		return;

	waitfor(WAITING, WAITING);
	// what is correct here?
	if (pread(fd, buf, 10, 0) != 0)
		perror("pread under io clean");
	close(fd);
}

/*
 * semaphore get/put wrapper
 */
int get_semaphore(int sem_id, struct sembuf *sembuffer)
{
	sembuffer->sem_num = 0;
	sembuffer->sem_op  = -1;
	sembuffer->sem_flg = SEM_UNDO;
	return semop(sem_id, sembuffer, 1);
}

int put_semaphore(int sem_id, struct sembuf *sembuffer)
{
	sembuffer->sem_num = 0;
	sembuffer->sem_op  = 1;
	sembuffer->sem_flg = SEM_UNDO;
	return semop(sem_id, sembuffer, 1);
}

/* memory sharing mode */
enum shared_mode {
	MMAP_SHARED = 0,
	IPV_SHARED  = 1,
};

/*
 * testcase for shared pages, where
 *  if early_kill == 0, parent access the shared page hwpoisoned by child, and
 *  if early_kill == 1, parent will be killed by SIGBUS from child.
 * This testcase checks whether if a shared page is hwpoisoned by one process,
 * another process sharing the page will be killed expectedly.
 */
static void do_shared(int shared_mode)
{
	int shm_id = -1, sem_id = -1, semaphore;
	pid_t pid;
	char *shared_page = NULL;
	struct sembuf sembuffer;

	if (shared_mode == MMAP_SHARED) {
		shared_page = checked_mmap(NULL, PS, PROT_READ|PROT_WRITE,
				MAP_SHARED|MAP_ANONYMOUS|MAP_POPULATE, 0, 0);
	} else if (shared_mode == IPV_SHARED) {
		shm_id = shmget(IPC_PRIVATE, PS, 0666|IPC_CREAT);
		if (shm_id == -1)
			err("shmget");
	} else {
		printf("XXX: invalid shared_mode\n");
		return;
	}

	if (early_kill) {
		sem_id = semget(IPC_PRIVATE, 1, 0666|IPC_CREAT);
		if (sem_id == -1) {
			perror("semget");
			goto cleanup;
		}
		semaphore = semctl(sem_id, 0, SETVAL, 1);
		if (semaphore == -1) {
			perror("semctl");
			goto cleanup;
		}
		if (get_semaphore(sem_id, &sembuffer)) {
			perror("get_semaphore");
			goto cleanup;
		}
	}

	pid = fork();
	if (pid < 0) {
		perror("fork");
		goto cleanup;
	}

	if (shared_mode == IPV_SHARED) {
		shared_page = shmat(shm_id, NULL, 0);
		if (shared_page == (char *)-1) {
			perror("shmat");
			goto cleanup;
		}
	}

	memset(shared_page, 'a', 3);

	if (early_kill) {
		struct sigaction sa = {
			.sa_sigaction = sighandler,
			.sa_flags = SA_SIGINFO
		};
		sigaction(SIGBUS, &sa, NULL);
		expected_addr = shared_page;
	}

	if (pid) {
		siginfo_t sig;

		if (early_kill && sigsetjmp(early_recover_ctx, 1) == 0) {
			if (put_semaphore(sem_id, &sembuffer))
				err("get_semaphore");
			/* waiting for SIGBUS from child */
			sleep(10);
			printf("XXX timeout: child process does not send signal\n");
			failure++;
			goto cleanup;
		}
		waitid(P_PID, pid, &sig, WEXITED);

		/*
		 * check child termination status
		 * late kill       : child should exit
		 * suicide version : child should be killed by signal
		 * early kill      : child should be killed by signal
		 */
		if (!early_kill) {
			struct sigaction sigact;
			sigaction(SIGBUS, NULL, &sigact);

			if (sigact.sa_handler == SIG_DFL) {/* suicide version */
				if (sig.si_code != CLD_KILLED)
					goto child_error;
			} else { /* late kill */
				if (sig.si_code != CLD_EXITED)
					goto child_error;
			}
		} else { /* early kill */
			if (sig.si_code != CLD_EXITED)
				goto child_error;
		}

		if (!early_kill)
			recover("ipv shared page (parent)",
				shared_page, MWRITE);

		if (shared_mode == IPV_SHARED && shmdt(shared_page) == -1) {
			perror("shmdt");
			goto cleanup;
		}
	}

	if (!pid) {
		failure = 0;

		if (early_kill)
			if (get_semaphore(sem_id, &sembuffer))
				err("get_semaphore");
		testmem("ipv shared page", shared_page, MWRITE);

		if (shared_mode == IPV_SHARED && shmdt(shared_page) == -1)
			err("shmdt");

		fflush(stdout);
		_exit(failure);
	}

cleanup:
	if (shared_page) {
		if (shared_mode == IPV_SHARED)
			shmdt(shared_page);
		else
			munmap_reserve(shared_page, PS);
	}
	if (shm_id >= 0 && shmctl(shm_id, IPC_RMID, NULL) < 0) 
		err("shmctl IPC_RMID");
	if (sem_id >= 0 && semctl(sem_id, 0, IPC_RMID) < 0)
		err("semctl IPC_RMID");
	return;

child_error:
	printf("XXX child process was terminated unexpectedly\n");
	failure++;
	goto cleanup;
}

static void mmap_shared(void)
{
	do_shared(MMAP_SHARED);
}

static void ipv_shared(void)
{
	do_shared(IPV_SHARED);
}

static void anonymous_hugepage(void)
{
	char *page;
	/* Hugepage isn't supported. */
	if (!HPS)
		return;
	test_hugepage = 1;
	page = alloc_anonymous_hugepage(HPS, 1);
	/* prefault */
	page[0] = 'a';
	testmem("anonymous hugepage", page, MWRITE);
	free_anonymous_hugepage(page, HPS);
	test_hugepage = 0;
}

static void file_backed_hugepage(void)
{
	char *page;
	char buf[PATHBUFLEN];
	int fd;
	/* Hugepage isn't supported. */
	if (!HPS)
		return;
	test_hugepage = 1;
	snprintf(buf, PATHBUFLEN, "%s/test%d", hugetlbfsdir, tmpcount++);
	page = alloc_filebacked_hugepage(buf, HPS, 0, &fd);
	/* prefault */
	page[0] = 'a';
	testmem("file backed hugepage", page, MWRITE);
	free_filebacked_hugepage(page, HPS, fd, buf);
	test_hugepage = 0;
}

static void shm_hugepage(void)
{
	char *page;
	/* Hugepage isn't supported. */
	if (!HPS)
		return;
	test_hugepage = 1;
	page = alloc_shm_hugepage(&tmpcount, HPS);
	/* prefault */
	page[0] = 'a';
	testmem("shared memory hugepage", page, MWRITE);
	free_shm_hugepage(tmpcount, page);
	tmpcount++;
	test_hugepage = 0;
}

struct testcase {
	void (*f)(void);
	char *name;
	int survivable;
} cases[] = {
	{ dirty_anonymous, "dirty anonymous" },
	{ dirty_anonymous_unmap, "dirty anonymous unmap" },
	{ mlocked_anonymous, "mlocked anonymous" },
	{ file_clean, "file clean", 1 },
	{ file_dirty, "file dirty" },
	{ file_hole, "file hole" },
	{ file_clean_mlocked, "file clean mlocked", 1 },
	{ file_dirty_mlocked, "file dirty mlocked"},
	{ nonlinear, "nonlinear" },
	{ mmap_shared, "mmap shared" },
	{ ipv_shared, "ipv shared" },
	{ anonymous_hugepage, "anonymous hugepage" },
	{ file_backed_hugepage, "file backed hugepage" },
	{ shm_hugepage, "shared memory hugepage" },
	{},	/* dummy 1 for sniper */
	{},	/* dummy 2 for sniper */
	{}
};

struct testcase snipercases[] = {
	{ under_io_dirty, "under io dirty" }, 
	{ under_io_clean, "under io clean" },
};

void usage(void)
{
	fprintf(stderr, "Usage: tinjpage [--sniper]\n"
			"Test hwpoison injection on pages in various states\n"
			"--mce-inject    Use mce-inject for injection\n"
			"--sniper  Enable racy sniper tests (likely broken)\n");
	exit(1);
}

void handle_opts(char **av)
{
	while (*++av) { 
		if (!strcmp(*av, "--sniper")) { 
			struct testcase *t;
			for (t = cases; t->f; t++)
				;
			*t++ = snipercases[0];
			*t++ = snipercases[1];
		}
		else if (!strcmp(*av, "--mce-inject")) { 
			inject = inject_mce_inject;			
		} else 
			usage();
	}
}

int main(int ac, char **av)
{
	if (av[1])
		handle_opts(av);

	PS = getpagesize();
	if (hugetlbfs_root(hugetlbfsdir))
		HPS = gethugepagesize();

	/* don't kill me at poison time, but possibly at page fault time */
	early_kill = 0;
	system("sysctl -w vm.memory_failure_early_kill=0");

	struct sigaction sa = {
		.sa_sigaction = sighandler,
		.sa_flags = SA_SIGINFO
	};

	struct testcase *t;
	/* catch signals */
	sigaction(SIGBUS, &sa, NULL);
	for (t = cases; t->f; t++) { 
		printf("---- testing %s\n", t->name);
		t->f();
	}

	/* suicide version */
	for (t = cases; t->f; t++) {
		printf("---- testing %s in child\n", t->name);
		pid_t child = fork();
		if (child == 0) {
			signal(SIGBUS, SIG_DFL);
			t->f();
			if (t->survivable)
				_exit(2);
			write(1, t->name, strlen(t->name));
			write(1, PAIR(" didn't kill itself?\n"));
			_exit(1);
		} else {
			siginfo_t sig;
			if (waitid(P_PID, child, &sig, WEXITED) < 0)
				perror("waitid");
			else {
				if (t->survivable) {
					if (sig.si_code != CLD_EXITED) {
						printf("XXX: %s: child not survived\n", t->name);
						failure++;
					}
				} else {
					if (sig.si_code != CLD_KILLED || sig.si_status != SIGBUS) {
						printf("XXX: %s: child not killed by SIGBUS\n", t->name);
						failure++;
					}
				}
			}
		}
	}

	/* early kill version */
	early_kill = 1;
	system("sysctl -w vm.memory_failure_early_kill=1");

	sigaction(SIGBUS, &sa, NULL);
	for (t = cases; t->f; t++) {
		printf("---- testing %s (early kill)\n", t->name);
		t->f();
	}

	if (failure > 0) {
		printf("FAILURE -- %d cases broken!\n", failure);
		return 1;
	}
	printf("SUCCESS\n");
	return 0;
}