linux/386: use Xen-friendly ELF TLS instruction sequence

Fixes #465.

R=iant
CC=golang-dev
https://golang.org/cl/1665051

2 files changed, 32 insertions, 4 deletions

diff --git a/src/pkg/runtime/linux/386/sys.s b/src/pkg/runtime/linux/386/sys.s
index 57ffc4aa4f..35c3780cef 100644
--- a/src/pkg/runtime/linux/386/sys.s
+++ b/src/pkg/runtime/linux/386/sys.s
@@ -161,12 +161,12 @@ TEXT clone(SB),7,$0
 	// In child on new stack.  Reload registers (paranoia).
 	MOVL	0(SP), BX	// m
 	MOVL	4(SP), DX	// g
-	MOVL	8(SP), CX	// fn
+	MOVL	8(SP), SI	// fn
 
 	MOVL	AX, m_procid(BX)	// save tid as m->procid
 
 	// set up ldt 7+id to point at m->tls.
-	// m->tls is at m+40.  newosproc left the id in tls[0].
+	// newosproc left the id in tls[0].
 	LEAL	m_tls(BX), BP
 	MOVL	0(BP), DI
 	ADDL	$7, DI	// m0 is LDT#7. count up.
@@ -186,7 +186,7 @@ TEXT clone(SB),7,$0
 	MOVL	DX, g(AX)
 	MOVL	BX, m(AX)
 
-	CALL	stackcheck(SB)	// smashes AX
+	CALL	stackcheck(SB)	// smashes AX, CX
 	MOVL	0(DX), DX	// paranoia; check they are not nil
 	MOVL	0(BX), BX
 
@@ -195,7 +195,7 @@ TEXT clone(SB),7,$0
 	CALL	emptyfunc(SB)
 	POPAL
 
-	CALL	CX	// fn()
+	CALL	SI	// fn()
 	CALL	exit1(SB)
 	MOVL	$0x1234, 0x1005
 	RET
@@ -247,8 +247,11 @@ TEXT setldt(SB),7,$32
 	 * To accommodate that rewrite, we translate
 	 * the address here and bump the limit to 0xffffffff (no limit)
 	 * so that -8(GS) maps to 0(address).
+	 * Also, the final 0(GS) (current 8(CX)) has to point
+	 * to itself, to mimic ELF.
 	 */
 	ADDL	$0x8, CX	// address
+	MOVL	CX, 0(CX)
 
 	// set up user_desc
 	LEAL	16(SP), AX	// struct user_desc
diff --git a/src/pkg/runtime/mkasmh.sh b/src/pkg/runtime/mkasmh.sh
index aae773cfee..df8ad88381 100755
--- a/src/pkg/runtime/mkasmh.sh
+++ b/src/pkg/runtime/mkasmh.sh
@@ -24,6 +24,31 @@ case "$GOARCH" in
 		echo '#define	g(r)	0(r)'
 		echo '#define	m(r)	4(r)'
 		;;
+	linux)
+		# On Linux systems, what we call 0(GS) and 4(GS) for g and m
+		# turn into %gs:-8 and %gs:-4 (using gcc syntax to denote
+		# what the machine sees as opposed to 8l input).
+		# 8l rewrites 0(GS) and 4(GS) into these.
+		#
+		# On Linux Xen, it is not allowed to use %gs:-8 and %gs:-4
+		# directly.  Instead, we have to store %gs:0 into a temporary
+		# register and then use -8(%reg) and -4(%reg).  This kind
+		# of addressing is correct even when not running Xen.
+		#
+		# 8l can rewrite MOVL 0(GS), CX into the appropriate pair
+		# of mov instructions, using CX as the intermediate register
+		# (safe because CX is about to be written to anyway).
+		# But 8l cannot handle other instructions, like storing into 0(GS),
+		# which is where these macros come into play.
+		# get_tls sets up the temporary and then g and r use it.
+		#
+		# The final wrinkle is that get_tls needs to read from %gs:0,
+		# but in 8l input it's called 8(GS), because 8l is going to
+		# subtract 8 from all the offsets, as described above.
+		echo '#define	get_tls(r)	MOVL 8(GS), r'
+		echo '#define	g(r)	-8(r)'
+		echo '#define	m(r)	-4(r)'
+		;;
 	*)
 		echo '#define	get_tls(r)'
 		echo '#define	g(r)	0(GS)'