From cbadd201cbd20b7e44b9f8edea932a9420f67230 Mon Sep 17 00:00:00 2001
From: Adrian Reber <areber@redhat.com>
Date: Thu, 16 Jan 2020 16:41:40 +0100
Subject: [PATCH] s390x: use clone3() if possible

This adds the parasite clone3() with set_tid wrapper for s390x.

In contrast to the x86_64 implementation the thread start address and
arguments are not put on the thread stack but passed via r4 and r5. As
those registers are caller-saved they still contain the correct value
(thread start address and arguments) after returning from the syscall.

Tested on 5.5.0-rc6.

Signed-off-by: Adrian Reber <areber@redhat.com>
---
 criu/arch/s390/include/asm/restorer.h | 40 +++++++++++++++++++++++----
 criu/kerndat.c                        |  4 +--
 2 files changed, 37 insertions(+), 7 deletions(-)

diff --git a/criu/arch/s390/include/asm/restorer.h b/criu/arch/s390/include/asm/restorer.h
index 733f2de33..2fc266535 100644
--- a/criu/arch/s390/include/asm/restorer.h
+++ b/criu/arch/s390/include/asm/restorer.h
@@ -40,11 +40,41 @@
 	: "0", "1", "2", "3", "4", "5", "6", "cc", "memory")
 
 #define RUN_CLONE3_RESTORE_FN(ret, clone_args, size, args, \
-			      clone_restore_fn)	do { \
-	pr_err("This architecture does not support clone3() with set_tid, yet!\n"); \
-	pr_err("Not creating a process with PID: %d\n", ((pid_t *)u64_to_ptr(clone_args.set_tid))[0]); \
-	ret = -1; \
-} while (0)
+			      clone_restore_fn)				\
+	asm volatile(							\
+	/*
+	 * clone3 only needs two arguments (r2, r3), this means
+	 * we can use r4 and r5 for args and thread function.
+	 * r4 and r5 are callee-saved and are not overwritten.
+	 * No need to put these values on the child stack.
+	 */								\
+	"lgr	%%r4,%4\n"	/* Save args in %r4 */			\
+	"lgr	%%r5,%3\n"	/* Save clone_restore_fn in %r5 */	\
+	"lgr	%%r2,%1\n"	/* Parameter 1: clone_args */		\
+	"lgr	%%r3,%2\n"	/* Parameter 2: size */			\
+	/*
+	 * On s390x a syscall is done sc <syscall number>.
+	 * That only works for syscalls < 255. clone3 is 435,
+	 * therefore it is necessary to load the syscall number
+	 * into r1 and do 'svc 0'.
+	 */								\
+	"lghi	%%r1,"__stringify(__NR_clone3)"\n"			\
+	"svc	0\n"							\
+	"ltgr	%0,%%r2\n"	/* Set and check "ret" */		\
+	"jnz	0f\n"		/* ret != 0: Continue caller */		\
+	"lgr	%%r2,%%r4\n"	/* Thread arguments taken from r4. */	\
+	"lgr	%%r1,%%r5\n"	/* Thread function taken from r5. */	\
+	"aghi	%%r15,-160\n"	/* Prepare stack frame */		\
+	"xc	0(8,%%r15),0(%%r15)\n"					\
+	"basr	%%r14,%%r1\n"	/* Jump to clone_restore_fn() */	\
+	"j	.+2\n"		/* BUG(): Force PGM check */		\
+"0:\n"				/* Continue caller */			\
+	: "=d"(ret)							\
+	: "a"(&clone_args),						\
+	  "d"(size),							\
+	  "d"(clone_restore_fn),					\
+	  "d"(args)							\
+	: "0", "1", "2", "3", "4", "5", "cc", "memory")
 
 #define arch_map_vdso(map, compat)		-1
 
diff --git a/criu/kerndat.c b/criu/kerndat.c
index 0772828bc..2261cca60 100644
--- a/criu/kerndat.c
+++ b/criu/kerndat.c
@@ -992,10 +992,10 @@ static bool kerndat_has_clone3_set_tid(void)
 	pid_t pid;
 	struct _clone_args args = {};
 
-#ifndef CONFIG_X86_64
+#if !defined(CONFIG_X86_64) && !defined(CONFIG_S390)
 	/*
 	 * Currently the CRIU PIE assembler clone3() wrapper is
-	 * only implemented for X86_64.
+	 * only implemented for X86_64 and S390X.
 	 */
 	kdat.has_clone3_set_tid = false;
 	return 0;