13508 door_layout() should align the stack to 16 bytes for i386 processes

Reviewed by: Robert Mustacchi <rm+illumos@fingolfin.org>
Reviewed by: Andy Fiddaman <andy@omniosce.org>
Reviewed by: Gordon Ross <gordon.ross@tintri.com>
Approved by: Dan McDonald <danmcd@joyent.com>

9 files changed, 424 insertions, 2 deletions

diff --git a/usr/src/lib/libc/i386/sys/door.s b/usr/src/lib/libc/i386/sys/door.s
index 1e5561c387..d0785af1bc 100644
--- a/usr/src/lib/libc/i386/sys/door.s
+++ b/usr/src/lib/libc/i386/sys/door.s
@@ -22,6 +22,8 @@
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright 2021 Tintri by DDN, Inc. All rights reserved.
  */
 
 	.file	"door.s"
@@ -110,7 +112,7 @@
 /*
  * int
  * __door_return(
- *	void 			*data_ptr,
+ *	void			*data_ptr,
  *	size_t			data_size,	(in bytes)
  *	door_return_desc_t	*door_ptr,	(holds returned desc info)
  *	caddr_t			stack_base,
@@ -142,6 +144,8 @@ door_restart:
 	 *		data (if any)
 	 *	 sp->	struct door_results
 	 *
+	 * The stack will be aligned to 16 bytes; we must maintain that
+	 * alignment prior to any call instruction.
 	 * struct door_results has the arguments in place for the server proc,
 	 * so we just call it directly.
 	 */
@@ -152,14 +156,16 @@ door_restart:
 	 * this is the last server thread - call creation func for more
 	 */
 	movl	DOOR_INFO_PTR(%esp), %eax
+	subl	$12, %esp
 	pushl	%eax		/* door_info_t * */
 	call	door_depletion_cb@PLT
-	addl	$4, %esp
+	addl	$16, %esp
 1:
 	/* Call the door server function now */
 	movl	DOOR_PC(%esp), %eax
 	call	*%eax
 	/* Exit the thread if we return here */
+	subl	$12, %esp
 	pushl	$0
 	call	_thrp_terminate
 	/* NOTREACHED */
diff --git a/usr/src/pkg/manifests/system-test-ostest.mf b/usr/src/pkg/manifests/system-test-ostest.mf
index e65c7b654f..596c19fb5d 100644
--- a/usr/src/pkg/manifests/system-test-ostest.mf
+++ b/usr/src/pkg/manifests/system-test-ostest.mf
@@ -14,6 +14,7 @@
 # Copyright 2014, OmniTI Computer Consulting, Inc. All rights reserved.
 # Copyright 2020 Joyent, Inc.
 # Copyright 2020 OmniOS Community Edition (OmniOSce) Association.
+# Copyright 2021 Tintri by DDN, Inc. All rights reserved.
 #
 
 set name=pkg.fmri value=pkg:/system/test/ostest@$(PKGVERS)
@@ -38,6 +39,7 @@ dir path=opt/os-tests/tests/sdevfs
 dir path=opt/os-tests/tests/secflags
 dir path=opt/os-tests/tests/sigqueue
 dir path=opt/os-tests/tests/sockfs
+dir path=opt/os-tests/tests/stackalign
 dir path=opt/os-tests/tests/stress
 dir path=opt/os-tests/tests/syscall
 dir path=opt/os-tests/tests/timer
@@ -110,6 +112,8 @@ file path=opt/os-tests/tests/sockfs/rights.32 mode=0555
 file path=opt/os-tests/tests/sockfs/rights.64 mode=0555
 file path=opt/os-tests/tests/sockfs/sockpair mode=0555
 file path=opt/os-tests/tests/spoof-ras mode=0555
+file path=opt/os-tests/tests/stackalign/stackalign.32 mode=0555
+file path=opt/os-tests/tests/stackalign/stackalign.64 mode=0555
 file path=opt/os-tests/tests/stress/dladm-kstat mode=0555
 file path=opt/os-tests/tests/syscall/open.32 mode=0555
 file path=opt/os-tests/tests/syscall/open.64 mode=0555
diff --git a/usr/src/test/os-tests/runfiles/default.run b/usr/src/test/os-tests/runfiles/default.run
index 0370c425a9..e7b44684db 100644
--- a/usr/src/test/os-tests/runfiles/default.run
+++ b/usr/src/test/os-tests/runfiles/default.run
@@ -13,6 +13,7 @@
 # Copyright (c) 2012 by Delphix. All rights reserved.
 # Copyright 2020 Joyent, Inc.
 # Copyright 2020 OmniOS Community Edition (OmniOSce) Association.
+# Copyright 2021 Tintri by DDN, Inc. All rights reserved.
 #
 
 [DEFAULT]
@@ -116,3 +117,6 @@ pre = ksensor_init
 tests = [ 'ksensor_basic.32','ksensor_basic.64', 'ksensor_err.32',
     'ksensor_err.64' ]
 post = ksensor_fini
+
+[/opt/os-tests/tests/stackalign]
+tests = ['stackalign.32', 'stackalign.64']
diff --git a/usr/src/test/os-tests/tests/Makefile b/usr/src/test/os-tests/tests/Makefile
index c9f3391576..2783408243 100644
--- a/usr/src/test/os-tests/tests/Makefile
+++ b/usr/src/test/os-tests/tests/Makefile
@@ -12,6 +12,7 @@
 #
 # Copyright (c) 2012, 2016 by Delphix. All rights reserved.
 # Copyright 2020 Joyent, Inc.
+# Copyright 2021 Tintri by DDN, Inc. All rights reserved.
 #
 
 SUBDIRS_i386 = i386 imc
@@ -28,6 +29,7 @@ SUBDIRS =       \
 		sigqueue \
 		sockfs \
 		spoof-ras \
+		stackalign \
 		stress \
 		syscall \
 		timer \
diff --git a/usr/src/test/os-tests/tests/stackalign/Makefile b/usr/src/test/os-tests/tests/stackalign/Makefile
new file mode 100644
index 0000000000..4af7cb7352
--- /dev/null
+++ b/usr/src/test/os-tests/tests/stackalign/Makefile
@@ -0,0 +1,78 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright 2020 Oxide Computer Company
+# Copyright 2021 Tintri by DDN, Inc. All rights reserved.
+#
+
+PROGS = stackalign
+
+PROGS32 = $(PROGS:%=%.32)
+PROGS64 = $(PROGS:%=%.64)
+
+OBJS = $(PROGS32:%=%.o) $(PROGS64:%=%.o)
+$(INTEL_BLD)OBJS += stack_i386.o stack_amd64.o
+
+ROOTOPTDIR = $(ROOT)/opt/os-tests/tests
+ROOTOPTSTACK = $(ROOTOPTDIR)/stackalign
+ROOTOPTPROGS = $(PROGS32:%=$(ROOTOPTSTACK)/%) $(PROGS64:%=$(ROOTOPTSTACK)/%)
+
+include $(SRC)/cmd/Makefile.cmd
+
+ASFLAGS += -P -D_ASM
+CFLAGS += -D_REENTRANT
+$(INTEL_BLD)LDFLAGS += -Wl,-zinitarray=get_stack_at_init
+
+.KEEP_STATE:
+
+all: $(PROGS32) $(PROGS64)
+
+install: $(ROOTOPTPROGS)
+
+clean:
+	-$(RM) $(OBJS)
+
+$(ROOTOPTPROGS): $(PROGS32) $(PROGS64) $(ROOTOPTSTACK)
+
+$(ROOTOPTDIR):
+	$(INS.dir)
+
+$(ROOTOPTSTACK): $(ROOTOPTDIR)
+	$(INS.dir)
+
+$(ROOTOPTSTACK)/%: %
+	$(INS.file)
+
+%.64.o: %.c
+	$(COMPILE64.c) $< -o $@
+
+%.32.o: %.c
+	$(COMPILE.c) $< -o $@
+
+stack_$(MACH).o: stack_$(MACH).s
+	$(COMPILE.s) $< -o $@
+
+stack_$(MACH64).o: stack_$(MACH64).s
+	$(COMPILE64.s) $< -o $@
+
+%.64: %.64.o $(INTEL_BLD)stack_$(MACH64).o
+	$(LINK64.c) -o $@ $^ $(LDLIBS64)
+	$(POST_PROCESS)
+
+%.32: %.32.o $(INTEL_BLD)stack_$(MACH).o
+	$(LINK.c) -o $@ $^ $(LDLIBS)
+	$(POST_PROCESS)
+
+clobber: clean
+	$(RM) $(PROGS32) $(PROGS64)
+
+FRC:
diff --git a/usr/src/test/os-tests/tests/stackalign/stack_amd64.s b/usr/src/test/os-tests/tests/stackalign/stack_amd64.s
new file mode 100644
index 0000000000..7fb4be88e3
--- /dev/null
+++ b/usr/src/test/os-tests/tests/stackalign/stack_amd64.s
@@ -0,0 +1,69 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2021 Tintri by DDN, Inc. All rights reserved.
+ */
+
+/*
+ * Get the stack at entry and call a function with it as an argument.
+ */
+
+	.file	"stack_amd64.s"
+
+#include <sys/asm_linkage.h>
+
+/*
+ * void
+ * get_stack_at_entry(test_ctx_t *ctx)
+ *
+ * ctx+0 is void (*)(uintptr_t stack, char *text),
+ * and ctx+8 is the 'text' argument.
+ *
+ * Passes the stack pointer prior to the invoking call instruction
+ * to the specified function.
+ */
+	ENTRY(get_stack_at_entry)
+	pushq	%rbp
+	movq	%rsp, %rbp
+	movq	%rdi, %rax
+	leaq	16(%rbp), %rdi
+	movq	8(%rax), %rsi
+	call	*(%rax)
+	popq	%rbp
+	ret
+	SET_SIZE(get_stack_at_entry)
+
+/*
+ * void
+ * get_stack_at_init(void)
+ *
+ * Passes the stack pointer prior to the invoking call instruction
+ * to initarray() (defined elsewhere).
+ * Tests alignment in section .init_array.
+ */
+	ENTRY(get_stack_at_init)
+	pushq	%rbp
+	movq	%rsp, %rbp
+	leaq	16(%rbp), %rdi
+	call	initarray@PLT
+	popq	%rbp
+	ret
+	SET_SIZE(get_stack_at_init)
+
+/*
+ * Passes the stack pointer during init to initmain() (defined elsewhere).
+ * Tests alignment in section .init.
+ */
+	.section ".init"
+	movq	%rsp, %rdi
+	call	initmain@PLT
+
diff --git a/usr/src/test/os-tests/tests/stackalign/stack_i386.s b/usr/src/test/os-tests/tests/stackalign/stack_i386.s
new file mode 100644
index 0000000000..f8fb525cad
--- /dev/null
+++ b/usr/src/test/os-tests/tests/stackalign/stack_i386.s
@@ -0,0 +1,75 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2021 Tintri by DDN, Inc. All rights reserved.
+ */
+
+/*
+ * Get the stack at entry and call a function with it as an argument.
+ */
+
+	.file	"stack_i386.s"
+
+#include <sys/asm_linkage.h>
+
+/*
+ * void
+ * get_stack_at_entry(test_ctx_t *ctx)
+ *
+ * ctx+0 is void (*)(uintptr_t stack, char *text),
+ * and ctx+4 is the 'text' argument.
+ *
+ * Passes the stack pointer prior to the invoking call instruction
+ * to the specified function.
+ */
+	ENTRY(get_stack_at_entry)
+	pushl	%ebp
+	movl	%esp, %ebp
+	leal	8(%ebp), %eax
+	movl	8(%ebp), %ecx
+	pushl	4(%ecx)
+	pushl	%eax
+	call	*(%ecx)
+	addl	$8, %esp
+	popl	%ebp
+	ret
+	SET_SIZE(get_stack_at_entry)
+
+/*
+ * void
+ * get_stack_at_init(void)
+ *
+ * Passes the stack pointer prior to the invoking call instruction
+ * to initarray() (defined elsewhere).
+ * Tests alignment in section .init_array.
+ */
+	ENTRY(get_stack_at_init)
+	pushl	%ebp
+	movl	%esp, %ebp
+	leal	8(%ebp), %eax
+	subl	$8, %esp
+	movl	%eax, (%esp)
+	call	initarray@PLT
+	addl	$8, %esp
+	popl	%ebp
+	ret
+	SET_SIZE(get_stack_at_init)
+
+/*
+ * Passes the stack pointer during init to initmain() (defined elsewhere).
+ * Tests alignment in section .init.
+ */
+	.section ".init"
+	movl	%esp, %eax
+	pushl	%eax
+	call	initmain@PLT
+	addl	$4, %esp
diff --git a/usr/src/test/os-tests/tests/stackalign/stackalign.c b/usr/src/test/os-tests/tests/stackalign/stackalign.c
new file mode 100644
index 0000000000..2ad77ea39d
--- /dev/null
+++ b/usr/src/test/os-tests/tests/stackalign/stackalign.c
@@ -0,0 +1,170 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2021 Tintri by DDN, Inc. All rights reserved.
+ */
+
+/*
+ * Test that the stack is aligned to expected values.
+ */
+
+#include <stdio.h>
+#include <pthread.h>
+#include <thread.h>
+#include <door.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <ucontext.h>
+
+#include <sys/stack.h>
+
+/*
+ * The introduction of SSE led to the IA32 ABI changing the required stack
+ * alignment from 4 bytes to 16 bytes. Compilers assume this when using SSE.
+ */
+#if defined(__i386)
+#undef STACK_ALIGN
+#define	STACK_ALIGN 16
+#endif
+
+#define	ALIGN_ERR_IMPL(align, text)				\
+	"stack was not aligned to " #align " on " text "\n"
+#define	ALIGN_ERR_HELP(align, text) ALIGN_ERR_IMPL(align, text)
+#define	ALIGN_ERR(text) ALIGN_ERR_HELP(STACK_ALIGN, text)
+
+#define	STACK_SIZE 16*1024
+
+typedef struct test_ctx {
+	void (*func)(uintptr_t, char *);
+	char *text;
+} test_ctx_t;
+
+extern void get_stack_at_entry(test_ctx_t *);
+
+void
+teststack(uintptr_t stack, char *arg)
+{
+	if ((stack & (STACK_ALIGN - 1)) != 0) {
+		fprintf(stderr, ALIGN_ERR("%s"), (char *)arg);
+		exit(1);
+	}
+}
+
+void
+initmain(uintptr_t stack)
+{
+	teststack(stack, "section .init");
+}
+
+void
+initarray(uintptr_t stack)
+{
+	teststack(stack, "section .init_array");
+}
+
+void
+doorstack(uintptr_t stack, char *arg)
+{
+	teststack(stack, arg);
+	(void) door_return(NULL, 0, NULL, 0);
+}
+
+char door_arg[] = "DOOR ARG";
+
+int
+main(int argc, char *argv[])
+{
+	door_arg_t da = {
+	    .data_ptr = (void *)door_arg,
+	    .data_size = sizeof (door_arg)
+	};
+	test_ctx_t arg = {
+	    .func = teststack,
+	    .text = "pthread_create()"
+	};
+	ucontext_t back, uc;
+	pthread_t tid;
+	int door_fd, rc;
+
+#if defined(__sparc)
+	/*
+	 * This hasn't been implemented for SPARC, so skip.
+	 */
+	fprintf(stderr, "No SPARC implementation of get_stack_at_entry\n");
+	return (3);
+#else
+	if (pthread_create(&tid, NULL,
+	    (void *(*)(void *))get_stack_at_entry, &arg) != 0) {
+		perror("pthread_create() failed:");
+		exit(-2);
+	}
+	(void) pthread_join(tid, NULL);
+
+	arg.text = "thr_create()";
+
+	if (thr_create(NULL, 0, (void *(*)(void *))get_stack_at_entry,
+	    &arg, 0, &tid) != 0) {
+		perror("thr_create() failed:");
+		exit(-3);
+	}
+	(void) thr_join(tid, NULL, NULL);
+
+	if (getcontext(&uc) < 0) {
+		perror("getcontext() failed");
+		exit(-4);
+	}
+
+	uc.uc_link = &back;
+	uc.uc_stack.ss_size = STACK_SIZE;
+	uc.uc_stack.ss_flags = 0;
+	if ((uc.uc_stack.ss_sp = malloc(STACK_SIZE)) == NULL) {
+		perror("failed to allocate stack");
+		exit(-5);
+	}
+
+	arg.text = "swapcontext()";
+	makecontext(&uc, (void (*)(void *))get_stack_at_entry, 1, &arg);
+	if (swapcontext(&back, &uc) < 0) {
+		perror("swapcontext() failed");
+		exit(-6);
+	}
+
+	arg.func = doorstack;
+	arg.text = "door_call()";
+
+	if ((door_fd = door_create(
+	    (door_server_procedure_t *)get_stack_at_entry,
+	    &arg, 0)) < 0) {
+		perror("failed to create door");
+		exit(-7);
+	}
+
+	rc = door_call(door_fd, &da);
+
+	if (rc < 0) {
+		perror("door call #1 failed");
+		exit(-8);
+	}
+
+	da.data_size += 5;
+	rc = door_call(door_fd, &da);
+
+	if (rc < 0) {
+		perror("door call #2 failed");
+		exit(-9);
+	}
+
+	(void) close(door_fd);
+
+	return (0);
+#endif
+}
diff --git a/usr/src/uts/common/fs/doorfs/door_sys.c b/usr/src/uts/common/fs/doorfs/door_sys.c
index 68a7a11d82..a2d3812938 100644
--- a/usr/src/uts/common/fs/doorfs/door_sys.c
+++ b/usr/src/uts/common/fs/doorfs/door_sys.c
@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2016 by Delphix. All rights reserved.
+ * Copyright 2021 Tintri by DDN, Inc. All rights reserved.
  */
 
 /*
@@ -1114,6 +1115,19 @@ door_stack_copyout(const void *kaddr, void *uaddr, size_t count)
 }
 
 /*
+ * The IA32 ABI supplement 1.0 changed the required stack alignment to
+ * 16 bytes (from 4 bytes), so that code can make use of SSE instructions.
+ * This is already done for process entry, thread entry, and makecontext();
+ * We need to do this for door_return as well. The stack will be aligned to
+ * whatever the door_results is aligned.
+ * See: usr/src/lib/libc/i386/gen/makectxt.c for more details.
+ */
+#if defined(__amd64)
+#undef STACK_ALIGN32
+#define	STACK_ALIGN32 16
+#endif
+
+/*
  * Writes the stack layout for door_return() into the door_server_t of the
  * server thread.
  */