OS-5322 lxbrand use comm page for vdso

Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Reviewed by: Joshua M. Clulow <jmc@joyent.com>

17 files changed, 499 insertions, 240 deletions

diff --git a/manifest b/manifest
index 59c6dcd894..7ca1ae3197 100644
--- a/manifest
+++ b/manifest
@@ -5078,6 +5078,7 @@ f usr/lib/brand/lx/lx_boot_zone_ubuntu 0755 root root
 f usr/lib/brand/lx/lxinit 0755 root root
 f usr/lib/brand/lx/lx_librtld_db.so.1 0755 root root
 f usr/lib/brand/lx/lx_support 0755 root root
+f usr/lib/brand/lx/lx_vdso.so.1 0755 root root
 f usr/lib/brand/lx/platform.xml 0444 root root
 d usr/lib/brand/shared 0755 root sys
 f usr/lib/brand/shared/common.ksh 0444 root bin
diff --git a/usr/src/common/brand/lx/lx_auxv.c b/usr/src/common/brand/lx/lx_auxv.c
index bd7b588ac8..8994ea31a5 100644
--- a/usr/src/common/brand/lx/lx_auxv.c
+++ b/usr/src/common/brand/lx/lx_auxv.c
@@ -37,14 +37,9 @@ lx_auxv_stol(const auxv_t *ap, auxv_t *oap, const lx_elf_data_t *edp)
 		oap->a_un.a_val = edp->ed_phnum;
 		break;
 	case AT_SUN_BRAND_LX_SYSINFO_EHDR:
-		if (edp->ed_vdso != 0) {
-			oap->a_type = AT_SYSINFO_EHDR;
-			oap->a_un.a_val = edp->ed_vdso;
-			return (0);
-		} else {
-			/* No vDSO for i386 */
-			return (1);
-		}
+		oap->a_type = AT_SYSINFO_EHDR;
+		oap->a_un.a_val = ap->a_un.a_val;
+		return (0);
 	case AT_SUN_BRAND_LX_CLKTCK:
 		oap->a_type = AT_CLKTCK;
 		oap->a_un.a_val = ap->a_un.a_val;
diff --git a/usr/src/lib/brand/lx/lx_brand/common/lx_brand.c b/usr/src/lib/brand/lx/lx_brand/common/lx_brand.c
index bce525a78e..75fba80b07 100644
--- a/usr/src/lib/brand/lx/lx_brand/common/lx_brand.c
+++ b/usr/src/lib/brand/lx/lx_brand/common/lx_brand.c
@@ -422,29 +422,6 @@ lx_close_fh(FILE *file)
 
 extern int set_l10n_alternate_root(char *path);
 
-#if defined(_LP64)
-static void *
-map_vdso()
-{
-	int fd;
-	mmapobj_result_t	mpp[10]; /* we know the size of our lib */
-	mmapobj_result_t	*smpp = mpp;
-	uint_t			mapnum = 10;
-
-	if ((fd = open("/native/usr/lib/brand/lx/amd64/lx_vdso.so.1",
-	    O_RDONLY)) == -1)
-		lx_err_fatal("couldn't open lx_vdso.so.1");
-
-	if (mmapobj(fd, MMOBJ_INTERPRET, smpp, &mapnum, NULL) == -1)
-		lx_err_fatal("couldn't mmapobj lx_vdso.so.1");
-
-	(void) close(fd);
-
-	/* assume first segment is the base of the mapping */
-	return (smpp->mr_addr);
-}
-#endif
-
 /*
  * Initialize the thread specific data for this thread.
  */
@@ -592,12 +569,8 @@ lx_init(int argc, char *argv[], char *envp[])
 	lx_elf_data_t	edp;
 	lx_brand_registration_t reg;
 	lx_tsd_t	*lxtsd;
-#if defined(_LP64)
-	void		*vdso_hdr;
-#endif
 
 	bzero(&reg, sizeof (reg));
-
 	stack_size = 2 * sysconf(_SC_PAGESIZE);
 
 	/*
@@ -686,18 +659,6 @@ lx_init(int argc, char *argv[], char *envp[])
 	if (lx_statfs_init() != 0)
 		lx_err_fatal("failed to setup the statfs translator");
 
-#if defined(_LP64)
-	vdso_hdr = map_vdso();
-	edp.ed_vdso = (uintptr_t)vdso_hdr;
-	/*
-	 * Notify the kernel of this mapping location to keep its
-	 * representation of the auxv consistent with reality.
-	 */
-	(void) syscall(SYS_brand, B_NOTIFY_VDSO_LOC, (void *)vdso_hdr);
-#else
-	edp.ed_vdso = 0;
-#endif
-
 	/*
 	 * Find the aux vector on the stack.
 	 */
diff --git a/usr/src/lib/brand/lx/lx_vdso/Makefile b/usr/src/lib/brand/lx/lx_vdso/Makefile
index b71a8a1aeb..56dd0a7a3c 100644
--- a/usr/src/lib/brand/lx/lx_vdso/Makefile
+++ b/usr/src/lib/brand/lx/lx_vdso/Makefile
@@ -10,17 +10,14 @@
 #
 
 #
-# Copyright 2015 Joyent, Inc.
+# Copyright 2016 Joyent, Inc.
 #
 
 include ../../../Makefile.lib
 
-SUBDIRS = tools
+SUBDIRS = tools $(MACH)
 $(BUILD64)SUBDIRS += $(MACH64)
 
-LINT_SUBDIRS = 
-$(BUILD64)LINT_SUBDIRS += $(MACH64)
-
 all :=          TARGET= all
 clean :=        TARGET= clean
 clobber :=      TARGET= clobber
@@ -33,6 +30,7 @@ all install clean clobber: $(SUBDIRS)
 
 lint: $(LINT_SUBDIRS)
 
+$(MACH): tools
 $(MACH64): tools
 
 $(SUBDIRS): FRC
diff --git a/usr/src/lib/brand/lx/lx_vdso/Makefile.com b/usr/src/lib/brand/lx/lx_vdso/Makefile.com
index 3839260345..0cdb1aaf70 100644
--- a/usr/src/lib/brand/lx/lx_vdso/Makefile.com
+++ b/usr/src/lib/brand/lx/lx_vdso/Makefile.com
@@ -10,14 +10,16 @@
 #
 
 #
-# Copyright 2015 Joyent, Inc.
+# Copyright 2016 Joyent, Inc.
 #
 
 LIBRARY	=	lx_vdso.a
 VERS =		.1
 
-COBJS =		lx_vdso.o
-OBJECTS =	$(COBJS)
+include $(SRC)/lib/commpage/Makefile.shared.com
+
+COBJS =		vdso_main.o vdso_subr.o
+OBJECTS =	$(COBJS) $(COMMPAGE_OBJS)
 
 include ../../../../Makefile.lib
 include ../../Makefile.lx
@@ -31,14 +33,15 @@ LIBNAME =	lx_vdso
 MAPFILES =	../common/mapfile-vers
 MAPOPTS =	$(MAPFILES:%=-M%)
 
-ASOBJS  =	lx_vdso.o
-OBJECTS =	$(ASOBJS)
-
-ASSRCS =	$(ASOBJS:%o=$(ISASRCDIR)/%s)
-SRCS =		$(ASSRCS)
+ASOBJS  =	vdso_subr.o
+COBJS  =	vdso_main.o
+OBJECTS =	$(ASOBJS) $(COBJS) $(COMMPAGE_OBJS)
 
 SRCDIR =	../common
-UTSBASE	=	../../../../../uts
+
+ASSRCS =	$(ASOBJS:%.o=$(ISASRCDIR)/%.s)
+CSRCS =		$(COBJS:%.o=$(SRCDIR)/%.c)
+SRCS =		$(ASSRCS) $(CSRCS)
 
 LIBS =		$(DYNLIB)
 DYNFLAGS +=	$(DYNFLAGS_$(CLASS))
@@ -72,7 +75,11 @@ all: $(LIBS)
 lint: $(LINTLIB) lintcheck
 
 include ../../../../Makefile.targ
+include $(SRC)/lib/commpage/Makefile.shared.targ
 
 pics/%.o: $(ISASRCDIR)/%.s
 	$(COMPILE.s) -o $@ $<
 	$(POST_PROCESS_O)
+
+pics/vdso_main.o := CPPFLAGS += $(COMMPAGE_CPPFLAGS)
+pics/vdso_subr.o := ASFLAGS += -I$(SRC)/uts/common/brand/lx
diff --git a/usr/src/lib/brand/lx/lx_vdso/amd64/Makefile b/usr/src/lib/brand/lx/lx_vdso/amd64/Makefile
index cdf6eaa62d..f1c17dcd91 100644
--- a/usr/src/lib/brand/lx/lx_vdso/amd64/Makefile
+++ b/usr/src/lib/brand/lx/lx_vdso/amd64/Makefile
@@ -10,14 +10,19 @@
 #
 
 #
-# Copyright 2014 Joyent, Inc.  All rights reserved.
+# Copyright 2016 Joyent, Inc.
 #
 
 ISASRCDIR=.
+TARGET_ARCH=$(MACH64)
 
 include ../Makefile.com
 include $(SRC)/lib/Makefile.lib.64
 
+ASFLAGS += -D__$(MACH64)
+
+SONAME = linux-vdso.so.1
+
 #
 # You might ask, why aren't we overriding BUILD.SO in Makefile.com.
 # That's a sad story. The answer is that Makefile.lib.64 includes
diff --git a/usr/src/lib/brand/lx/lx_vdso/amd64/lx_vdso.s b/usr/src/lib/brand/lx/lx_vdso/amd64/lx_vdso.s
deleted file mode 100644
index 039f9b95b3..0000000000
--- a/usr/src/lib/brand/lx/lx_vdso/amd64/lx_vdso.s
+++ /dev/null
@@ -1,132 +0,0 @@
-/*
- *
- * 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 2015 Joyent, Inc.
- */
-
-#include <sys/asm_linkage.h>
-#include <sys/trap.h>
-
-/*
- * lx vDSO emulation library
- *
- * This so needs to look like the correct Linux vDSO elf library. We cannot
- * use any native symbols or link with any native libraries, particularly libc.
- */
-
-#define	LX_SYS_gettimeofday	96
-#define	LX_SYS_time		201
-#define	LX_SYS_clock_gettime	228
-#define	LX_SYS_getcpu		309
-
-#if defined(lint)
-int
-__vdso_gettimeofday(void *tp, void *tz)
-{}
-
-time_t
-__vdso_time(void *tp)
-{}
-
-time_t
-__vdso_clock_gettime(uintptr_t id, void *tp)
-{}
-
-int
-__vdso_getcpu(void *cpu, void *np, void *cp)
-{}
-
-#else /* lint */
-
-	/*
-	 * We know the arguments are already in the correct registers (e.g. arg0
-	 * already in %rdi, arg1 already in %rsi, etc.). %rax has result of
-	 * call.
-	 */
-
-	/*
-	 * Uses fasttrap, based on lib/libc/amd64/sys/gettimeofday.s
-	 */
-	ENTRY_NP(__vdso_gettimeofday)
-	pushq	%rdi		/* pointer to timeval */
-	movl	$T_GETHRESTIME, %eax
-	int	$T_FASTTRAP
-	/*
-	 *	gethrestime trap returns seconds in %rax, nsecs in %edx
-	 *	need to convert nsecs to usecs & store into area pointed
-	 *	to by struct timeval * argument.
-	 */
-	popq	%rcx		/* pointer to timeval */
-	jrcxz	1f		/* bail if we get a null pointer */
-	movq	%rax, (%rcx)	/* store seconds into timeval ptr */
-	movl	$274877907, %eax /* divide by 1000 as impl. by gcc */
-	imull	%edx		/* See Hacker's Delight pg 162 */
-	sarl	$6, %edx	/* simplified by 0 <= nsec <= 1e9 */
-	movq	%rdx, 8(%rcx)	/* store usecs into timeval ptr + 8. */
-1:
-	xorq	%rax, %rax	/* return 0 */
-	ret
-	SET_SIZE(__vdso_gettimeofday)
-
-	/*
-	 * Uses fasttrap, based on lib/libc/amd64/sys/gettimeofday.s, but only
-	 * returns seconds. This is based on what the kernel's gtime function
-	 * will do.
-	 */
-	ENTRY_NP(__vdso_time)
-	pushq	%rdi		/* pointer to time_t */
-	movl	$T_GETHRESTIME, %eax
-	int	$T_FASTTRAP
-	/*
-	 *	gethrestime trap returns seconds in %rax
-	 *	store secs into area pointed by time_t * argument.
-	 */
-	popq	%rcx		/* pointer to time_t */
-	jrcxz	1f		/* don't save if we get a null pointer */
-	movq	%rax, (%rcx)	/* store seconds into time_t ptr */
-1:
-	ret			/* return seconds in %rax */
-	SET_SIZE(__vdso_time)
-
-	/*
-	 * Does not use fasttrap since there more work to emulate than we can
-	 * do with a fasttrap.
-	 */
-	ENTRY_NP(__vdso_clock_gettime)
-	movq $LX_SYS_clock_gettime, %rax
-	syscall
-	ret
-	SET_SIZE(__vdso_clock_gettime)
-
-	/*
-	 * Uses fasttrap.
-	 * getcpu takes 3 pointers but we only support saving the cpu ID into
-	 * the first pointer.
-	 */
-	ENTRY_NP(__vdso_getcpu)
-	pushq	%rdi		/* pointer to int */
-	movl	$T_GETLGRP, %eax
-	int	$T_FASTTRAP
-	/*
-	 *	getlgrp trap returns CPU ID in %eax
-	 *	store it into area pointed by int * argument.
-	 */
-	popq	%rcx		/* pointer to int */
-	jrcxz	1f		/* don't save if we get a null pointer */
-	movl	%eax, (%rcx)	/* store CPU ID into int ptr */
-1:
-	xorq	%rax, %rax	/* return 0 */
-	ret
-	SET_SIZE(__vdso_getcpu)
-#endif
diff --git a/usr/src/lib/brand/lx/lx_vdso/amd64/vdso_subr.s b/usr/src/lib/brand/lx/lx_vdso/amd64/vdso_subr.s
new file mode 100644
index 0000000000..bf066600aa
--- /dev/null
+++ b/usr/src/lib/brand/lx/lx_vdso/amd64/vdso_subr.s
@@ -0,0 +1,68 @@
+/*
+ *
+ * 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 2016 Joyent, Inc.
+ */
+
+
+#include <sys/asm_linkage.h>
+#include <sys/lx_syscalls.h>
+
+
+#if defined(lint)
+
+comm_page_t *
+__vdso_find_commpage()
+{}
+
+long
+__vdso_sys_clock_gettime(uint_t clock_id, timespec_t *tp)
+{}
+
+int
+__vdso_sys_gettimeofday(timespec_t *tp, struct lx_timezone *tz)
+{}
+
+time_t
+__vdso_sys_time(timespec_t *tp)
+{}
+
+#else /* lint */
+
+	ENTRY_NP(__vdso_find_commpage)
+	leaq	0x0(%rip), %rax
+	andq	$LX_VDSO_ADDR_MASK, %rax
+	addq	$LX_VDSO_SIZE, %rax
+	ret
+	SET_SIZE(__vdso_find_commpage)
+
+	ENTRY_NP(__vdso_sys_clock_gettime)
+	movl	$LX_SYS_clock_gettime, %eax
+	syscall
+	ret
+	SET_SIZE(__vdso_sys_clock_gettime)
+
+	ENTRY_NP(__vdso_sys_gettimeofday)
+	movl	$LX_SYS_gettimeofday, %eax
+	syscall
+	ret
+	SET_SIZE(__vdso_sys_gettimeofday)
+
+	ENTRY_NP(__vdso_sys_time)
+	movl	$LX_SYS_time, %eax
+	syscall
+	ret
+	SET_SIZE(__vdso_sys_time)
+
+#endif /* lint */
diff --git a/usr/src/lib/brand/lx/lx_vdso/common/vdso_main.c b/usr/src/lib/brand/lx/lx_vdso/common/vdso_main.c
new file mode 100644
index 0000000000..a0474bf26e
--- /dev/null
+++ b/usr/src/lib/brand/lx/lx_vdso/common/vdso_main.c
@@ -0,0 +1,160 @@
+/*
+ * 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 2016 Joyent, Inc.
+ */
+
+#include <cp_defs.h>
+
+
+struct lx_timezone {
+	int tz_minuteswest;	/* minutes W of Greenwich */
+	int tz_dsttime;		/* type of dst correction */
+};
+
+extern comm_page_t *__vdso_find_commpage();
+extern int __vdso_sys_gettimeofday(timespec_t *, struct lx_timezone *);
+extern time_t __vdso_sys_time(timespec_t *);
+extern long __vdso_sys_clock_gettime(uint_t, timespec_t *);
+
+#define	LX_CLOCK_REALTIME		0	/* CLOCK_REALTIME	*/
+#define	LX_CLOCK_MONOTONIC		1	/* CLOCK_HIGHRES	*/
+#define	LX_CLOCK_PROCESS_CPUTIME_ID	2	/* Emulated		*/
+#define	LX_CLOCK_THREAD_CPUTIME_ID	3	/* Emulated		*/
+#define	LX_CLOCK_MONOTONIC_RAW		4	/* CLOCK_HIGHRES	*/
+#define	LX_CLOCK_REALTIME_COARSE	5	/* CLOCK_REALTIME	*/
+#define	LX_CLOCK_MONOTONIC_COARSE	6	/* CLOCK_HIGHRES	*/
+
+
+void
+__hrt2ts(hrtime_t hrt, timespec_t *tsp)
+{
+	uint32_t sec, nsec, tmp;
+
+	tmp = (uint32_t)(hrt >> 30);
+	sec = tmp - (tmp >> 2);
+	sec = tmp - (sec >> 5);
+	sec = tmp + (sec >> 1);
+	sec = tmp - (sec >> 6) + 7;
+	sec = tmp - (sec >> 3);
+	sec = tmp + (sec >> 1);
+	sec = tmp + (sec >> 3);
+	sec = tmp + (sec >> 4);
+	tmp = (sec << 7) - sec - sec - sec;
+	tmp = (tmp << 7) - tmp - tmp - tmp;
+	tmp = (tmp << 7) - tmp - tmp - tmp;
+	nsec = (uint32_t)hrt - (tmp << 9);
+	while (nsec >= NANOSEC) {
+		nsec -= NANOSEC;
+		sec++;
+	}
+	tsp->tv_sec = (time_t)sec;
+	tsp->tv_nsec = nsec;
+}
+
+int
+__vdso_gettimeofday(timespec_t *tp, struct lx_timezone *tz)
+{
+	comm_page_t *cp = __vdso_find_commpage();
+
+	if (__cp_can_gettime(cp) != 0) {
+		return (__vdso_sys_gettimeofday(tp, tz));
+	}
+
+	if (tp != NULL) {
+		long usec, nsec;
+
+		__cp_clock_gettime_realtime(cp, tp);
+
+		nsec = tp->tv_nsec;
+		usec = nsec + (nsec >> 2);
+		usec = nsec + (usec >> 1);
+		usec = nsec + (usec >> 2);
+		usec = nsec + (usec >> 4);
+		usec = nsec - (usec >> 3);
+		usec = nsec + (usec >> 2);
+		usec = nsec + (usec >> 3);
+		usec = nsec + (usec >> 4);
+		usec = nsec + (usec >> 1);
+		usec = nsec + (usec >> 6);
+		usec = usec >> 10;
+		tp->tv_nsec = usec;
+	}
+
+	if (tz != NULL) {
+		tz->tz_minuteswest = 0;
+		tz->tz_dsttime = 0;
+	}
+
+	return (0);
+}
+
+time_t
+__vdso_time(timespec_t *tp)
+{
+	comm_page_t *cp = __vdso_find_commpage();
+	timespec_t ts;
+
+	if (__cp_can_gettime(cp) != 0) {
+		return (__vdso_sys_time(tp));
+	}
+
+	__cp_clock_gettime_realtime(cp, &ts);
+	if (tp != NULL) {
+		tp->tv_sec = ts.tv_sec;
+		tp->tv_nsec = 0;
+	}
+	return (ts.tv_sec);
+}
+
+long
+__vdso_clock_gettime(uint_t clock_id, timespec_t *tp)
+{
+	comm_page_t *cp = __vdso_find_commpage();
+
+	if (__cp_can_gettime(cp) != 0) {
+		return (__vdso_sys_clock_gettime(clock_id, tp));
+	}
+
+	switch (clock_id) {
+	case LX_CLOCK_REALTIME:
+	case LX_CLOCK_REALTIME_COARSE:
+		__cp_clock_gettime_realtime(cp, tp);
+		return (0);
+
+	case LX_CLOCK_MONOTONIC:
+	case LX_CLOCK_MONOTONIC_RAW:
+	case LX_CLOCK_MONOTONIC_COARSE:
+		__hrt2ts(__cp_gethrtime(cp), tp);
+		return (0);
+
+	case LX_CLOCK_PROCESS_CPUTIME_ID:
+	case LX_CLOCK_THREAD_CPUTIME_ID:
+	default:
+		break;
+	}
+	return (__vdso_sys_clock_gettime(clock_id, tp));
+}
+
+long
+__vdso_getcpu(uint_t *cpu, uint_t *node, void *tcache)
+{
+	comm_page_t *cp = __vdso_find_commpage();
+
+	if (cpu != NULL) {
+		*cpu = __cp_getcpu(cp);
+	}
+	if (node != NULL) {
+		*node = 0;
+	}
+	return (0);
+}
diff --git a/usr/src/lib/brand/lx/lx_vdso/i386/Makefile b/usr/src/lib/brand/lx/lx_vdso/i386/Makefile
new file mode 100644
index 0000000000..7f9a6a13e6
--- /dev/null
+++ b/usr/src/lib/brand/lx/lx_vdso/i386/Makefile
@@ -0,0 +1,43 @@
+#
+# 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 2016 Joyent, Inc.
+#
+
+ISASRCDIR=.
+TARGET_ARCH=$(MACH)
+
+include ../Makefile.com
+
+ASFLAGS += -D__$(MACH)
+
+SONAME = linux-gate.so.1
+
+#
+# You might ask, why aren't we overriding BUILD.SO in Makefile.com.
+# See the amd64 Makefile for more answers
+#
+BUILD.SO =	$(LD) -o $@ $(GSHARED) $(DYNFLAGS) $(PICS) $(LDLIBS)
+
+ASSYMDEP_OBJS = lx_vdso.o
+
+CLOBBERFILES	= $(ROOTLIBDIR)/$(DYNLIB) $(ROOTLIBDIR)/$(LINTLIB)
+
+# Set the object entry point for __vsyscall-ers
+entryfix: $(DYNLIB)
+	$(ELFEDIT) -e "ehdr:e_entry \
+	    $$($(ELFEDIT) -re 'sym:st_value -osimple __vsyscall' $(DYNLIB))" \
+	    $(DYNLIB)
+
+all: entryfix
+
+install: all $(ROOTLIBS)
diff --git a/usr/src/lib/brand/lx/lx_vdso/i386/vdso_subr.s b/usr/src/lib/brand/lx/lx_vdso/i386/vdso_subr.s
new file mode 100644
index 0000000000..ed7be8bb23
--- /dev/null
+++ b/usr/src/lib/brand/lx/lx_vdso/i386/vdso_subr.s
@@ -0,0 +1,92 @@
+/*
+ *
+ * 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 2016 Joyent, Inc.
+ */
+
+
+#include <sys/asm_linkage.h>
+#include <sys/lx_syscalls.h>
+
+
+#if defined(lint)
+
+comm_page_t *
+__vdso_find_commpage()
+{}
+
+long
+__vdso_sys_clock_gettime(uint_t clock_id, timespec_t *tp)
+{}
+
+int
+__vdso_sys_gettimeofday(timespec_t *tp, struct lx_timezone *tz)
+{}
+
+time_t
+__vdso_sys_time(timespec_t *tp)
+{}
+
+#else /* lint */
+
+	ENTRY_NP(__vdso_find_commpage)
+	call	1f
+1:	popl	%eax
+	andl	$LX_VDSO_ADDR_MASK, %eax
+	addl	$LX_VDSO_SIZE, %eax
+	ret
+	SET_SIZE(__vdso_find_commpage)
+
+	ENTRY_NP(__vdso_sys_clock_gettime)
+	movl	$LX_SYS_clock_gettime, %eax
+	movl	0x4(%esp), %ebx
+	movl	0x8(%esp), %ecx
+	int	$0x80
+	ret
+	SET_SIZE(__vdso_sys_clock_gettime)
+
+	ENTRY_NP(__vdso_sys_gettimeofday)
+	movl	$LX_SYS_gettimeofday, %eax
+	movl	0x4(%esp), %ebx
+	movl	0x8(%esp), %ecx
+	int	$0x80
+	ret
+	SET_SIZE(__vdso_sys_gettimeofday)
+
+	ENTRY_NP(__vdso_sys_time)
+	movl	$LX_SYS_time, %eax
+	movl	0x4(%esp), %ebx
+	int	$0x80
+	ret
+	SET_SIZE(__vdso_sys_time)
+
+	ENTRY_NP(__vsyscall)
+	/*
+	 * On 32-bit Linux, the VDSO entry point (specified by e_entry)
+	 * provides a potentially accelerated means to vector into the kernel.
+	 * Normally this means using 'sysenter' with a Linux-custom calling
+	 * convention so programs expecting int80 behavior are not required to
+	 * change how arguments are passed.
+	 *
+	 * The SunOS sysenter entry point does _not_ tolerate such a departure
+	 * from convention, so if this function is updated to use sysenter, it
+	 * must properly marshal arguments onto the stack from the int80 style.
+	 * Such an enhancement can only occur once sysenter receives the same
+	 * branding hooks as syscall and int80.
+	 */
+	int	$0x80
+	ret
+	SET_SIZE(__vsyscall)
+
+#endif /* lint */
diff --git a/usr/src/uts/common/brand/lx/os/lx_brand.c b/usr/src/uts/common/brand/lx/os/lx_brand.c
index 17ce72dbf3..6cff045a80 100644
--- a/usr/src/uts/common/brand/lx/os/lx_brand.c
+++ b/usr/src/uts/common/brand/lx/os/lx_brand.c
@@ -1358,7 +1358,6 @@ lx_brandsys(int cmd, int64_t *rval, uintptr_t arg1, uintptr_t arg2,
 			led32.ed_entry = (int)pd->l_elf_data.ed_entry;
 			led32.ed_base = (int)pd->l_elf_data.ed_base;
 			led32.ed_ldentry = (int)pd->l_elf_data.ed_ldentry;
-			led32.ed_vdso = 0;
 			mutex_exit(&p->p_lock);
 
 			if (copyout(&led32, (void *)arg1,
@@ -1792,30 +1791,6 @@ lx_brandsys(int cmd, int64_t *rval, uintptr_t arg1, uintptr_t arg2,
 		return (0);
 	}
 
-	case B_NOTIFY_VDSO_LOC: {
-#if defined(_LP64)
-		if (get_udatamodel() == DATAMODEL_NATIVE) {
-			int i;
-
-			mutex_enter(&p->p_lock);
-			pd = ptolxproc(p);
-			pd->l_elf_data.ed_vdso = arg1;
-			/* overwrite the auxv data too */
-			for (i = 0; i < __KERN_NAUXV_IMPL; i++) {
-				if (p->p_user.u_auxv[i].a_type ==
-				    AT_SUN_BRAND_LX_SYSINFO_EHDR) {
-					p->p_user.u_auxv[i].a_un.a_val = arg1;
-					break;
-				}
-			}
-			mutex_exit(&p->p_lock);
-			return (0);
-		}
-#endif /* defined(_LP64) */
-		/* This is not valid for 32bit processes */
-		return (EINVAL);
-	}
-
 	case B_GET_PERSONALITY: {
 		unsigned int result;
 
@@ -1967,6 +1942,63 @@ extern int elfexec(vnode_t *, execa_t *, uarg_t *, intpdata_t *, int,
 extern int elf32exec(struct vnode *, execa_t *, uarg_t *, intpdata_t *, int,
     long *, int, caddr_t, cred_t *, int *);
 
+static uintptr_t
+lx_map_vdso(struct uarg *args, struct cred *cred)
+{
+	int err;
+	char *fpath = LX_VDSO_PATH;
+	vnode_t *vp;
+	vattr_t attr;
+	caddr_t addr;
+
+#if defined(_LP64)
+	if (args->to_model != DATAMODEL_NATIVE) {
+		fpath = LX_VDSO_PATH32;
+	}
+#endif
+
+	/*
+	 * The comm page should have been mapped in already.
+	 */
+	if (args->commpage == NULL) {
+		return (NULL);
+	}
+
+	/*
+	 * Ensure the VDSO library is present and appropriately sized.
+	 * This lookup is started at the zone root to avoid complications for
+	 * processes which have chrooted.  For the specified lookup root to be
+	 * used, the leading slash must be dropped from the path.
+	 */
+	ASSERT(fpath[0] == '/');
+	fpath++;
+	if (lookupnameat(fpath, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp,
+	    curzone->zone_rootvp) != 0) {
+		return (NULL);
+	}
+
+	/*
+	 * The VDSO requires data exposed via the comm page in order to
+	 * function properly.  The VDSO is always mapped in at a fixed known
+	 * offset from the comm page, providing an easy means to locate it.
+	 */
+	addr = (caddr_t)(args->commpage - LX_VDSO_SIZE);
+	attr.va_mask = AT_SIZE;
+	if (VOP_GETATTR(vp, &attr, 0, cred, NULL) != 0 ||
+	    attr.va_size > LX_VDSO_SIZE) {
+		VN_RELE(vp);
+		return (NULL);
+	}
+
+	err = execmap(vp, addr, attr.va_size, 0, 0,
+	    PROT_USER|PROT_READ|PROT_EXEC, 1, 0);
+	VN_RELE(vp);
+	if (err != 0) {
+		return (NULL);
+	}
+	return ((uintptr_t)addr);
+}
+
 /*
  * Exec routine called by elfexec() to load either 32-bit or 64-bit Linux
  * binaries.
@@ -2159,6 +2191,12 @@ lx_elfexec(struct vnode *vp, struct execa *uap, struct uarg *args,
 	 * brand emulation library and its linker.
 	 */
 
+	/*
+	 * After execing the brand library (which should have implicitly mapped
+	 * in the comm page), map the VDSO into the approprate place in the AS.
+	 */
+	lxpd->l_vdso = lx_map_vdso(args, cred);
+
 	bzero(&env, sizeof (env));
 
 	/*
@@ -2347,11 +2385,7 @@ lx_elfexec(struct vnode *vp, struct execa *uap, struct uarg *args,
 		phdr_auxv[0].a_un.a_val = edp.ed_phdr;
 		phdr_auxv[1].a_un.a_val = ldaddr;
 		phdr_auxv[2].a_un.a_val = hz;
-		/*
-		 * The userspace brand library will map in the vDSO and notify
-		 * the kernel of its location during lx_init.
-		 */
-		phdr_auxv[3].a_un.a_val = 1;
+		phdr_auxv[3].a_un.a_val = lxpd->l_vdso;
 
 		if (copyout(&phdr_auxv, args->auxp_brand,
 		    sizeof (phdr_auxv)) == -1)
@@ -2368,11 +2402,7 @@ lx_elfexec(struct vnode *vp, struct execa *uap, struct uarg *args,
 		phdr_auxv32[0].a_un.a_val = edp.ed_phdr;
 		phdr_auxv32[1].a_un.a_val = ldaddr;
 		phdr_auxv32[2].a_un.a_val = hz;
-		/*
-		 * Unused on i386 due to lack of vDSO.
-		 * It will be cleaned up during lx_init.
-		 */
-		phdr_auxv32[3].a_un.a_val = 0;
+		phdr_auxv32[3].a_un.a_val = lxpd->l_vdso;
 
 		if (copyout(&phdr_auxv32, args->auxp_brand,
 		    sizeof (phdr_auxv32)) == -1)
diff --git a/usr/src/uts/common/brand/lx/procfs/lx_prvnops.c b/usr/src/uts/common/brand/lx/procfs/lx_prvnops.c
index b74df9d818..262339c31c 100644
--- a/usr/src/uts/common/brand/lx/procfs/lx_prvnops.c
+++ b/usr/src/uts/common/brand/lx/procfs/lx_prvnops.c
@@ -1568,6 +1568,7 @@ static void
 lxpr_read_pid_maps(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
 {
 	proc_t *p;
+	lx_proc_data_t *lxpd;
 	struct as *as;
 	struct seg *seg;
 	char *buf;
@@ -1579,6 +1580,7 @@ lxpr_read_pid_maps(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
 		char prot[5];
 		uintptr_t offset;
 		vnode_t *vp;
+		char *name_override;
 		struct print_data *next;
 	} *print_head = NULL;
 	struct print_data **print_tail = &print_head;
@@ -1593,6 +1595,7 @@ lxpr_read_pid_maps(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
 	}
 
 	as = p->p_as;
+	lxpd = ptolxproc(p);
 
 	if (as == &kas) {
 		lxpr_unlock(p);
@@ -1637,6 +1640,15 @@ lxpr_read_pid_maps(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
 
 		pbuf->offset = SEGOP_GETOFFSET(seg, (caddr_t)pbuf->saddr);
 
+		pbuf->name_override = NULL;
+		if (lxpd != NULL) {
+			if (pbuf->saddr == lxpd->l_vdso) {
+				pbuf->name_override = "[vdso]";
+			} else if (pbuf->saddr == p->p_user.u_commpagep) {
+				pbuf->name_override = "[vvar]";
+			}
+		}
+
 		pbuf->next = NULL;
 		*print_tail = pbuf;
 		print_tail = &pbuf->next;
@@ -1658,7 +1670,9 @@ lxpr_read_pid_maps(lxpr_node_t *lxpnp, lxpr_uiobuf_t *uiobuf)
 		ino_t inode = 0;
 
 		*buf = '\0';
-		if (pbuf->vp != NULL) {
+		if (pbuf->name_override != NULL) {
+			(void) strncpy(buf, pbuf->name_override, buflen);
+		} else if (pbuf->vp != NULL) {
 			vattr.va_mask = AT_FSID | AT_NODEID;
 			if (VOP_GETATTR(pbuf->vp, &vattr, 0, CRED(),
 			    NULL) == 0) {
diff --git a/usr/src/uts/common/brand/lx/sys/lx_brand.h b/usr/src/uts/common/brand/lx/sys/lx_brand.h
index 034b87c758..4906e444f1 100644
--- a/usr/src/uts/common/brand/lx/sys/lx_brand.h
+++ b/usr/src/uts/common/brand/lx/sys/lx_brand.h
@@ -59,12 +59,17 @@ extern "C" {
 #define	LX_LIB_PATH32	"/native/usr/lib/lx_brand.so.1"
 #define	LX_LIB_PATH64	"/native/usr/lib/amd64/lx_brand.so.1"
 
+#define	LX_VDSO_PATH32	"/native/usr/lib/brand/lx/lx_vdso.so.1"
+#define	LX_VDSO_PATH64	"/native/usr/lib/brand/lx/amd64/lx_vdso.so.1"
+
 #if defined(_LP64)
 #define	LX_LIB_PATH		LX_LIB_PATH64
 #define	LX_UNAME_MACHINE	LX_UNAME_MACHINE64
+#define	LX_VDSO_PATH		LX_VDSO_PATH64
 #else
 #define	LX_LIB_PATH		LX_LIB_PATH32
 #define	LX_UNAME_MACHINE	LX_UNAME_MACHINE32
+#define	LX_VDSO_PATH		LX_VDSO_PATH32
 #endif
 
 /*
@@ -103,7 +108,7 @@ extern "C" {
 #define	B_SET_NATIVE_STACK	147
 #define	B_SIGEV_THREAD_ID	148
 #define	B_OVERRIDE_KERN_VER	149
-#define	B_NOTIFY_VDSO_LOC	150
+/* formerly B_NOTIFY_VDSO_LOC	150 */
 #define	B_GET_PERSONALITY	151
 
 #ifndef _ASM
@@ -229,7 +234,6 @@ typedef struct lx_elf_data64 {
 	uintptr_t	ed_entry;
 	uintptr_t	ed_base;
 	uintptr_t	ed_ldentry;
-	uintptr_t	ed_vdso;
 } lx_elf_data64_t;
 
 typedef struct lx_elf_data32 {
@@ -239,7 +243,6 @@ typedef struct lx_elf_data32 {
 	uint32_t	ed_entry;
 	uint32_t	ed_base;
 	uint32_t	ed_ldentry;
-	uint32_t	ed_vdso;
 } lx_elf_data32_t;
 
 #if defined(_LP64)
@@ -313,6 +316,9 @@ typedef struct lx_proc_data {
 
 	/* Linux process personality */
 	unsigned int l_personality;
+
+	/* VDSO location */
+	uintptr_t l_vdso;
 } lx_proc_data_t;
 
 #endif	/* _KERNEL */
diff --git a/usr/src/uts/common/brand/lx/sys/lx_syscalls.h b/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
index 483f30545b..64084b77f1 100644
--- a/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
+++ b/usr/src/uts/common/brand/lx/sys/lx_syscalls.h
@@ -184,12 +184,13 @@ extern long lx_writev();
 #define	LX_VSYSCALL_SIZE		(uintptr_t)0x1000
 #endif
 
+#endif	/* _KERNEL */
 
 /*
  * System call numbers for revectoring:
  */
 
-#if defined(_LP64)
+#if defined(__amd64)
 #define	LX_SYS_close		3
 #define	LX_SYS_gettimeofday	96
 #define	LX_SYS_time		201
@@ -203,17 +204,26 @@ extern long lx_writev();
 #define	LX_SYS32_clock_gettime	265
 #define	LX_SYS32_io_setup	245
 #define	LX_SYS32_getcpu		318
-#else
+#elif defined(__i386)
 #define	LX_SYS_close		6
 #define	LX_SYS_gettimeofday	78
 #define	LX_SYS_time		13
 #define	LX_SYS_clock_gettime	265
 #define	LX_SYS_io_setup		245
 #define	LX_SYS_getcpu		318
-#endif
-
+#else
+#error "Architecture not supported"
+#endif /* defined(__amd64) */
 
-#endif	/* _KERNEL */
+/*
+ * The current code in the VDSO operates under the expectation that it will be
+ * mapped at a fixed offset from the comm page.  This simplifies the act of
+ * locating said page without any other reference.  The VDSO must fit within
+ * this offset, matching the same value as COMM_PAGE_ALIGN.
+ * See: uts/i86pc/sys/comm_page.h
+ */
+#define	LX_VDSO_SIZE		0x4000
+#define	LX_VDSO_ADDR_MASK	~(LX_VDSO_SIZE - 1)
 
 #ifdef	__cplusplus
 }
diff --git a/usr/src/uts/i86pc/os/comm_page_util.c b/usr/src/uts/i86pc/os/comm_page_util.c
index 3c635fe79b..1c8c9f8afd 100644
--- a/usr/src/uts/i86pc/os/comm_page_util.c
+++ b/usr/src/uts/i86pc/os/comm_page_util.c
@@ -39,12 +39,12 @@ comm_page_mapin()
 {
 #if defined(__amd64) && !defined(__xpv)
 	proc_t *p = curproc;
-	caddr_t addr = NULL;
+	caddr_t addr = (caddr_t)COMM_PAGE_ALIGN;
 	size_t len = COMM_PAGE_SIZE;
 	uint_t prot = PROT_USER | PROT_READ;
 	segumap_crargs_t suarg;
 
-	map_addr(&addr, len, (offset_t)0, 1, 0);
+	map_addr(&addr, len, (offset_t)0, 1, MAP_ALIGN);
 	if (addr == NULL || valid_usr_range(addr, len, prot, p->p_as,
 	    p->p_as->a_userlimit) != RANGE_OKAY) {
 		return (NULL);
diff --git a/usr/src/uts/i86pc/sys/comm_page.h b/usr/src/uts/i86pc/sys/comm_page.h
index 9d94a27763..dbf00bc7a7 100644
--- a/usr/src/uts/i86pc/sys/comm_page.h
+++ b/usr/src/uts/i86pc/sys/comm_page.h
@@ -27,6 +27,7 @@ extern "C" {
 #endif
 
 #define	COMM_PAGE_SIZE	PAGESIZE
+#define	COMM_PAGE_ALIGN	0x4000
 
 #ifndef _ASM