powerpc: Use trap metadata to prevent double restart rather than zeroing trap

It's not very nice to zero trap for this, because then system calls no
longer have trap_is_syscall(regs) invariant, and we can't distinguish
between sc and scv system calls (in a later patch).

Take one last unused bit from the low bits of the pt_regs.trap word
for this instead. There is not a really good reason why it should be
in trap as opposed to another field, but trap has some concept of
flags and it exists. Ideally I think we would move trap to 2-byte
field and have 2 more bytes available independently.

Add a selftests case for this, which can be seen to fail if
trap_norestart() is changed to return false.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Make them static inlines]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200507121332.2233629-4-mpe@ellerman.id.au

2 files changed, 175 insertions, 1 deletions

diff --git a/tools/testing/selftests/powerpc/signal/Makefile b/tools/testing/selftests/powerpc/signal/Makefile
index 932a032bf036..d6ae54663aed 100644
--- a/tools/testing/selftests/powerpc/signal/Makefile
+++ b/tools/testing/selftests/powerpc/signal/Makefile
@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: GPL-2.0
-TEST_GEN_PROGS := signal signal_tm sigfuz sigreturn_vdso
+TEST_GEN_PROGS := signal signal_tm sigfuz sigreturn_vdso sig_sc_double_restart
 
 CFLAGS += -maltivec
 $(OUTPUT)/signal_tm: CFLAGS += -mhtm
diff --git a/tools/testing/selftests/powerpc/signal/sig_sc_double_restart.c b/tools/testing/selftests/powerpc/signal/sig_sc_double_restart.c
new file mode 100644
index 000000000000..e3972264615b
--- /dev/null
+++ b/tools/testing/selftests/powerpc/signal/sig_sc_double_restart.c
@@ -0,0 +1,174 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Test that a syscall does not get restarted twice, handled by trap_norestart()
+ *
+ * Based on Al's description, and a test for the bug fixed in this commit:
+ *
+ * commit 9a81c16b527528ad307843be5571111aa8d35a80
+ * Author: Al Viro <viro@zeniv.linux.org.uk>
+ * Date:   Mon Sep 20 21:48:57 2010 +0100
+ *
+ *  powerpc: fix double syscall restarts
+ *
+ *  Make sigreturn zero regs->trap, make do_signal() do the same on all
+ *  paths.  As it is, signal interrupting e.g. read() from fd 512 (==
+ *  ERESTARTSYS) with another signal getting unblocked when the first
+ *  handler finishes will lead to restart one insn earlier than it ought
+ *  to.  Same for multiple signals with in-kernel handlers interrupting
+ *  that sucker at the same time.  Same for multiple signals of any kind
+ *  interrupting that sucker on 64bit...
+ */
+#define _GNU_SOURCE
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+#include <signal.h>
+#include <errno.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "utils.h"
+
+static void SIGUSR1_handler(int sig)
+{
+	kill(getpid(), SIGUSR2);
+	/*
+	 * SIGUSR2 is blocked until the handler exits, at which point it will
+	 * be raised again and think there is a restart to be done because the
+	 * pending restarted syscall has 512 (ERESTARTSYS) in r3. The second
+	 * restart will retreat NIP another 4 bytes to fail case branch.
+	 */
+}
+
+static void SIGUSR2_handler(int sig)
+{
+}
+
+static ssize_t raw_read(int fd, void *buf, size_t count)
+{
+	register long nr asm("r0") = __NR_read;
+	register long _fd asm("r3") = fd;
+	register void *_buf asm("r4") = buf;
+	register size_t _count asm("r5") = count;
+
+	asm volatile(
+"		b	0f		\n"
+"		b	1f		\n"
+"	0:	sc	0		\n"
+"		bns	2f		\n"
+"		neg	%0,%0		\n"
+"		b	2f		\n"
+"	1:				\n"
+"		li	%0,%4		\n"
+"	2:				\n"
+		: "+r"(_fd), "+r"(nr), "+r"(_buf), "+r"(_count)
+		: "i"(-ENOANO)
+		: "memory", "r6", "r7", "r8", "r9", "r10", "r11", "r12", "ctr", "cr0");
+
+	if (_fd < 0) {
+		errno = -_fd;
+		_fd = -1;
+	}
+
+	return _fd;
+}
+
+#define DATA "test 123"
+#define DLEN (strlen(DATA)+1)
+
+int test_restart(void)
+{
+	int pipefd[2];
+	pid_t pid;
+	char buf[512];
+
+	if (pipe(pipefd) == -1) {
+		perror("pipe");
+		exit(EXIT_FAILURE);
+	}
+
+	pid = fork();
+	if (pid == -1) {
+		perror("fork");
+		exit(EXIT_FAILURE);
+	}
+
+	if (pid == 0) { /* Child reads from pipe */
+		struct sigaction act;
+		int fd;
+
+		memset(&act, 0, sizeof(act));
+		sigaddset(&act.sa_mask, SIGUSR2);
+		act.sa_handler = SIGUSR1_handler;
+		act.sa_flags = SA_RESTART;
+		if (sigaction(SIGUSR1, &act, NULL) == -1) {
+			perror("sigaction");
+			exit(EXIT_FAILURE);
+		}
+
+		memset(&act, 0, sizeof(act));
+		act.sa_handler = SIGUSR2_handler;
+		act.sa_flags = SA_RESTART;
+		if (sigaction(SIGUSR2, &act, NULL) == -1) {
+			perror("sigaction");
+			exit(EXIT_FAILURE);
+		}
+
+		/* Let's get ERESTARTSYS into r3 */
+		while ((fd = dup(pipefd[0])) != 512) {
+			if (fd == -1) {
+				perror("dup");
+				exit(EXIT_FAILURE);
+			}
+		}
+
+		if (raw_read(fd, buf, 512) == -1) {
+			if (errno == ENOANO) {
+				fprintf(stderr, "Double restart moved restart before sc instruction.\n");
+				_exit(EXIT_FAILURE);
+			}
+			perror("read");
+			exit(EXIT_FAILURE);
+		}
+
+		if (strncmp(buf, DATA, DLEN)) {
+			fprintf(stderr, "bad test string %s\n", buf);
+			exit(EXIT_FAILURE);
+		}
+
+		return 0;
+
+	} else {
+		int wstatus;
+
+		usleep(100000);		/* Hack to get reader waiting */
+		kill(pid, SIGUSR1);
+		usleep(100000);
+		if (write(pipefd[1], DATA, DLEN) != DLEN) {
+			perror("write");
+			exit(EXIT_FAILURE);
+		}
+		close(pipefd[0]);
+		close(pipefd[1]);
+		if (wait(&wstatus) == -1) {
+			perror("wait");
+			exit(EXIT_FAILURE);
+		}
+		if (!WIFEXITED(wstatus)) {
+			fprintf(stderr, "child exited abnormally\n");
+			exit(EXIT_FAILURE);
+		}
+
+		FAIL_IF(WEXITSTATUS(wstatus) != EXIT_SUCCESS);
+
+		return 0;
+	}
+}
+
+int main(void)
+{
+	test_harness_set_timeout(10);
+	return test_harness(test_restart, "sig sys restart");
+}