# HG changeset patch
# User yamahata@xxxxxxxxxxxxx
# Date 1185941538 -32400
# Node ID 9d4bcd10abcb51136dbea5a4c2d3606a830de1bb
# Parent  6e0d898e9dad897b29db1c08be2b8b64c929d779
Introduce xen specific gate page.
With this transparent paravirtualization over head is eliminated.
PATCHNAME: xen_specific_gate_page

Signed-off-by: Isaku Yamahata <yamahata@xxxxxxxxxxxxx>

diff -r 6e0d898e9dad -r 9d4bcd10abcb arch/ia64/Kconfig
--- a/arch/ia64/Kconfig	Mon Jul 30 11:46:14 2007 -0600
+++ b/arch/ia64/Kconfig	Wed Aug 01 13:12:18 2007 +0900
@@ -64,13 +64,6 @@ config XEN
 	help
 	  Enable Xen hypervisor support.  Resulting kernel runs
 	  both as a guest OS on Xen and natively on hardware.
-
-config XEN_IA64_VDSO_PARAVIRT
-	bool "Xen/IA64 vdso area paravirtualization"
-	depends on XEN && !ITANIUM
-	default y
-	help
-	  vDSO paravirtualization
 
 config XEN_IA64_EXPOSE_P2M
 	bool "Xen/IA64 exposure p2m table"
diff -r 6e0d898e9dad -r 9d4bcd10abcb arch/ia64/kernel/Makefile
--- a/arch/ia64/kernel/Makefile	Mon Jul 30 11:46:14 2007 -0600
+++ b/arch/ia64/kernel/Makefile	Wed Aug 01 13:12:18 2007 +0900
@@ -61,3 +61,58 @@ GATECFLAGS_gate-syms.o = -r
 # We must build gate.so before we can assemble it.
 # Note: kbuild does not track this dependency due to usage of .incbin
 $(obj)/gate-data.o: $(obj)/gate.so
+
+
+#
+# gate page paravirtualization for xen
+#
+obj-$(CONFIG_XEN) += xengate-data.o
+
+# The gate DSO image is built using a special linker script.
+targets += xengate.so xengate-syms.o
+
+extra-$(CONFIG_XEN) += xengate.so xengate.lds xengate.o
+
+AFLAGS_xengate.o += -D__XEN_IA64_VDSO_PARAVIRT
+$(obj)/xengate.o: $(src)/gate.S FORCE
+	$(call if_changed_dep,as_o_S)
+
+CPPFLAGS_xengate.lds := -P -C -U$(ARCH) -D__XEN_IA64_VDSO_PARAVIRT
+$(obj)/xengate.lds: $(src)/gate.lds.S
+	$(call if_changed_dep,cpp_lds_S)
+
+GATECFLAGS_xengate.so = -shared -s -Wl,-soname=linux-gate.so.1 \
+		     $(call ld-option, -Wl$(comma)--hash-style=sysv)
+$(obj)/xengate.so: $(obj)/xengate.lds $(obj)/xengate.o FORCE
+	$(call if_changed,gate)
+
+ifeq ($(CONFIG_XEN), y)
+$(obj)/built-in.o: $(obj)/xengate-syms.o
+$(obj)/built-in.o: ld_flags += -R $(obj)/xengate-syms.o
+$(obj)/mca_recovery.o: $(obj)/gate-syms.o $(obj)/xengate-syms.o
+endif
+
+GATECFLAGS_xengate-syms.o = -r
+$(obj)/xengate-syms.o: $(obj)/xengate.lds $(obj)/xengate.o FORCE
+	$(call if_changed,gate)
+$(obj)/xengate-data.o: $(obj)/xengate.so
+
+
+#
+# .tmp_gate.o to calculate padding size for __kernel_syscall_via_epc
+#
+extra-y += gate-skip.s .tmp_gate.o
+
+ifeq ($(CONFIG_XEN), y)
+AFLAGS_gate.o += -D__KERNEL_SYSCALL_VIA_EPC_PADDING
+$(obj)/gate.o: $(obj)/gate-skip.s FORCE
+endif
+$(obj)/.tmp_gate.o: $(src)/gate.S FORCE
+	$(call if_changed_dep,as_o_S)
+quiet_cmd_gate_size = GATE_SIZE	$@
+      cmd_gate_size = $(NM) --extern-only --print-size $(obj)/xengate.o | \
+	$(AWK) '/__kernel_syscall_via_epc/{printf "\t.skip 0x"$$2" - "}' > $@; \
+	$(NM) --extern-only --print-size $(obj)/.tmp_gate.o | \
+	$(AWK) '/__kernel_syscall_via_epc/{printf "0x"$$2"\n"}' >> $@
+$(obj)/gate-skip.s: $(obj)/xengate.o $(obj)/.tmp_gate.o FORCE
+	$(call if_changed,gate_size)
diff -r 6e0d898e9dad -r 9d4bcd10abcb arch/ia64/kernel/gate.S
--- a/arch/ia64/kernel/gate.S	Mon Jul 30 11:46:14 2007 -0600
+++ b/arch/ia64/kernel/gate.S	Wed Aug 01 13:12:18 2007 +0900
@@ -31,17 +31,6 @@
 #define BRL_COND_FSYS_BUBBLE_DOWN(pr)			\
 [1:](pr)brl.cond.sptk 0;				\
 	.xdata4 ".data.patch.brl_fsys_bubble_down", 1b-.
-
-#ifdef CONFIG_XEN_IA64_VDSO_PARAVIRT
-	// Currently is_running_on_xen() is defined as running_on_xen.
-	// If is_running_on_xen() is a real function, we must update
-	// according to it.
-	.section ".data.patch.running_on_xen", "a"
-	.previous
-#define LOAD_RUNNING_ON_XEN(reg)			\
-[1:]	movl reg=0;					\
-	.xdata4 ".data.patch.running_on_xen", 1b-.
-#endif /* CONFIG_XEN_IA64_VDSO_PARAVIRT */
 
 GLOBAL_ENTRY(__kernel_syscall_via_break)
 	.prologue
@@ -86,37 +75,24 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
 	mov r10=0				// A    default to successful syscall execution
 	epc					// B	causes split-issue
 }
-#ifdef CONFIG_XEN_IA64_VDSO_PARAVIRT
+	;;
+#ifdef __XEN_IA64_VDSO_PARAVIRT
 	// r20 = 1
 	// r22 = &vcpu->vcpu_info->evtchn_upcall_mask
 	// r24 = &vcpu->vcpu_info->evtchn_upcall_pending
 	// r25 = tmp
-	// r28 = &running_on_xen
-	// r30 = running_on_xen
 	// r31 = tmp
 	// p11 = tmp
-	// p12 = running_on_xen
-	// p13 = !running_on_xen
 	// p14 = tmp
-	// p15 = tmp
-#define isXen	p12
-#define isRaw	p13
-	LOAD_RUNNING_ON_XEN(r28)
+	mov r20=1
 	movl r22=XSI_PSR_I_ADDR
-	mov r20=1
-	;;
-	ld4 r30=[r28]
-	;;
-	cmp.ne isXen,isRaw=r0,r30
-	;;
-(isXen)	ld8 r22=[r22]
-	;; 
-(isRaw)	rsm psr.be | psr.i
-(isXen)	adds r24=-1,r22
-(isXen)	st1 [r22]=r20
-(isXen)	rum psr.be
-#else
-	;;
+	;;
+	ld8 r22=[r22]
+	;;
+	st1 [r22]=r20
+	rum psr.be
+	adds r24=-1,r22
+#else
 	rsm psr.be | psr.i			// M2 (5 cyc to srlz.d)
 #endif
 	LOAD_FSYSCALL_TABLE(r14)		// X
@@ -124,14 +100,14 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
 	mov r16=IA64_KR(CURRENT)		// M2 (12 cyc)
 	shladd r18=r17,3,r14			// A
 	mov r19=NR_syscalls-1			// A
+#ifdef __XEN_IA64_VDSO_PARAVIRT
+	XEN_HYPER_GET_PSR
 	;;
 	lfetch [r18]				// M0|1
-#ifdef CONFIG_XEN_IA64_VDSO_PARAVIRT
-(isRaw)	mov r29=psr
-(isXen)	XEN_HYPER_GET_PSR
-	;;
-(isXen)	mov r29=r8
-#else
+	mov r29=r8
+#else
+	;;
+	lfetch [r18]				// M0|1
 	mov r29=psr				// M2 (12 cyc)
 #endif
 	// If r17 is a NaT, p6 will be zero
@@ -147,7 +123,7 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
 	;;
 	nop.m 0
 (p6)	tbit.z.unc p8,p0=r18,0			// I0 (dual-issues with "mov b7=r18"!)
-#ifdef CONFIG_XEN_IA64_VDSO_PARAVIRT
+#ifdef __XEN_IA64_VDSO_PARAVIRT
 	
 #define XEN_SET_PSR_I(pred)		\
 (pred)	ld1 r31=[r22];			\
@@ -162,12 +138,7 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
 (p11)	XEN_HYPER_SSM_I;
 
 	;; 
-	// p14 = running_on_xen && p8
-	// p15 = !running_on_xen && p8
-(p8)	cmp.ne.unc p14,p15=r0,r30
-	;;
-(p15)	ssm psr.i
-	XEN_SET_PSR_I(p14)
+	XEN_SET_PSR_I(p8)
 #else
 	nop.i 0
 	;;
@@ -193,9 +164,8 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
 #else
 	BRL_COND_FSYS_BUBBLE_DOWN(p6)
 #endif
-#ifdef CONFIG_XEN_IA64_VDSO_PARAVIRT
-(isRaw)	ssm psr.i
-	XEN_SET_PSR_I(isXen)
+#ifdef __XEN_IA64_VDSO_PARAVIRT
+	XEN_SET_PSR_I(p0)
 #else
 	ssm psr.i
 #endif
@@ -203,6 +173,14 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
 (p10)	mov r8=EINVAL
 (p9)	mov r8=ENOSYS
 	FSYS_RETURN
+#ifdef __KERNEL_SYSCALL_VIA_EPC_PADDING
+	/*
+	 * All values/sizes of __kernel_xxx symbol in gate.so and xengate.so
+	 * must be same to each other.
+	 * Adjust symbol size in gate.so to be same to the one in xengate.so.
+	 */
+.include "arch/ia64/kernel/gate-skip.s"
+#endif
 END(__kernel_syscall_via_epc)
 
 #	define ARG0_OFF		(16 + IA64_SIGFRAME_ARG0_OFFSET)
diff -r 6e0d898e9dad -r 9d4bcd10abcb arch/ia64/kernel/gate.lds.S
--- a/arch/ia64/kernel/gate.lds.S	Mon Jul 30 11:46:14 2007 -0600
+++ b/arch/ia64/kernel/gate.lds.S	Wed Aug 01 13:12:18 2007 +0900
@@ -28,6 +28,24 @@ SECTIONS
   . = GATE_ADDR + 0x500;
 
   .data.patch			: {
+#ifdef __XEN_IA64_VDSO_PARAVIRT
+#define __start_gate_mckinley_e9_patchlist \
+	__start_gate_mckinley_e9_patchlist_xen
+#define __end_gate_mckinley_e9_patchlist \
+	__end_gate_mckinley_e9_patchlist_xen
+#define __start_gate_vtop_patchlist \
+	__start_gate_vtop_patchlist_xen
+#define __end_gate_vtop_patchlist \
+	__end_gate_vtop_patchlist_xen
+#define __start_gate_fsyscall_patchlist \
+	__start_gate_fsyscall_patchlist_xen
+#define __end_gate_fsyscall_patchlist \
+	__end_gate_fsyscall_patchlist_xen
+#define __start_gate_brl_fsys_bubble_down_patchlist \
+	__start_gate_brl_fsys_bubble_down_patchlist_xen
+#define __end_gate_brl_fsys_bubble_down_patchlist \
+	__end_gate_brl_fsys_bubble_down_patchlist_xen
+#endif
 				    __start_gate_mckinley_e9_patchlist = .;
 				    *(.data.patch.mckinley_e9)
 				    __end_gate_mckinley_e9_patchlist = .;
@@ -43,12 +61,6 @@ SECTIONS
 				    __start_gate_brl_fsys_bubble_down_patchlist = .;
 				    *(.data.patch.brl_fsys_bubble_down)
 				    __end_gate_brl_fsys_bubble_down_patchlist = .;
-
-#ifdef CONFIG_XEN_IA64_VDSO_PARAVIRT
-				    __start_gate_running_on_xen_patchlist = .;
-				    *(.data.patch.running_on_xen)
-				    __end_gate_running_on_xen_patchlist = .;
-#endif
   }									:readable
   .IA_64.unwind_info		: { *(.IA_64.unwind_info*) }
   .IA_64.unwind			: { *(.IA_64.unwind*) }			:readable :unwind
diff -r 6e0d898e9dad -r 9d4bcd10abcb arch/ia64/kernel/patch.c
--- a/arch/ia64/kernel/patch.c	Mon Jul 30 11:46:14 2007 -0600
+++ b/arch/ia64/kernel/patch.c	Wed Aug 01 13:12:18 2007 +0900
@@ -185,40 +185,41 @@ patch_brl_fsys_bubble_down (unsigned lon
 }
 
 #ifdef CONFIG_XEN_IA64_VDSO_PARAVIRT
-extern char __start_gate_running_on_xen_patchlist[];
-extern char __end_gate_running_on_xen_patchlist[];
-
-void
-patch_running_on_xen(unsigned long start, unsigned long end)
-{
-	extern const int running_on_xen;
-	s32 *offp = (s32 *)start;
-	u64 ip;
-
-	while (offp < (s32 *)end) {
-		ip = (u64)ia64_imva((char *)offp + *offp);
-		ia64_patch_imm64(ip, (u64)&running_on_xen);
-		ia64_fc((void *)ip);
-		++offp;
-	}
-	ia64_sync_i();
-	ia64_srlz_i();
+void __init
+ia64_patch_gate_xen (void)
+{
+	extern char __start_gate_mckinley_e9_patchlist_xen[], __end_gate_mckinley_e9_patchlist_xen[];
+	extern char __start_gate_vtop_patchlist_xen[], __end_gate_vtop_patchlist_xen[];
+	extern char __start_gate_fsyscall_patchlist_xen[], __end_gate_fsyscall_patchlist_xen[];
+	extern char __start_gate_brl_fsys_bubble_down_patchlist_xen[], __end_gate_brl_fsys_bubble_down_patchlist_xen[];
+#	define START(name)	((unsigned long) __start_gate_##name##_patchlist_xen)
+#	define END(name)	((unsigned long)__end_gate_##name##_patchlist_xen)
+
+	patch_fsyscall_table(START(fsyscall), END(fsyscall));
+	patch_brl_fsys_bubble_down(START(brl_fsys_bubble_down), END(brl_fsys_bubble_down));
+	ia64_patch_vtop(START(vtop), END(vtop));
+	ia64_patch_mckinley_e9(START(mckinley_e9), END(mckinley_e9));
+
+#	undef START
+#	undef END
 }
 #else
-#define patch_running_on_xen(start, end)	do { } while (0)
+#define ia64_patch_gate_xen()	do { } while (0)
 #endif
 
 void __init
 ia64_patch_gate (void)
 {
+	if (is_running_on_xen()) {
+		ia64_patch_gate_xen();
+		return;
+	}
+	
 #	define START(name)	((unsigned long) __start_gate_##name##_patchlist)
 #	define END(name)	((unsigned long)__end_gate_##name##_patchlist)
 
 	patch_fsyscall_table(START(fsyscall), END(fsyscall));
 	patch_brl_fsys_bubble_down(START(brl_fsys_bubble_down), END(brl_fsys_bubble_down));
-#ifdef CONFIG_XEN
-	patch_running_on_xen(START(running_on_xen), END(running_on_xen));
-#endif
 	ia64_patch_vtop(START(vtop), END(vtop));
 	ia64_patch_mckinley_e9(START(mckinley_e9), END(mckinley_e9));
 }
diff -r 6e0d898e9dad -r 9d4bcd10abcb arch/ia64/kernel/vmlinux.lds.S
--- a/arch/ia64/kernel/vmlinux.lds.S	Mon Jul 30 11:46:14 2007 -0600
+++ b/arch/ia64/kernel/vmlinux.lds.S	Wed Aug 01 13:12:18 2007 +0900
@@ -183,6 +183,10 @@ SECTIONS
 	  __start_gate_section = .;
 	  *(.data.gate)
 	  __stop_gate_section = .;
+	  . = ALIGN(PAGE_SIZE);
+	  __start_xen_gate_section = .;
+	  *(.data.gate.xen)
+	  __stop_xen_gate_section = .;
 	}
   . = ALIGN(PAGE_SIZE);		/* make sure the gate page doesn't expose kernel data */
 
diff -r 6e0d898e9dad -r 9d4bcd10abcb arch/ia64/kernel/xengate-data.S
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/arch/ia64/kernel/xengate-data.S	Wed Aug 01 13:12:18 2007 +0900
@@ -0,0 +1,3 @@
+	.section .data.gate.xen, "aw"
+
+	.incbin "arch/ia64/kernel/xengate.so"
diff -r 6e0d898e9dad -r 9d4bcd10abcb arch/ia64/mm/init.c
--- a/arch/ia64/mm/init.c	Mon Jul 30 11:46:14 2007 -0600
+++ b/arch/ia64/mm/init.c	Wed Aug 01 13:12:18 2007 +0900
@@ -303,16 +303,34 @@ setup_gate (void)
 setup_gate (void)
 {
 	struct page *page;
+	void *gate_page_addr = __start_gate_section;
+	
+#ifdef CONFIG_XEN
+	unsigned long unused_gate;
+	extern char __start_xen_gate_section[];
+	if (is_running_on_xen()) {
+		gate_page_addr = __start_xen_gate_section;
+		unused_gate = (unsigned long)ia64_imva(__start_gate_section);
+	} else
+		unused_gate =
+			(unsigned long)ia64_imva(__start_xen_gate_section);
+#ifndef HAVE_BUGGY_SEGREL
+	ClearPageReserved(virt_to_page(unused_gate));
+	init_page_count(virt_to_page(unused_gate));
+	free_page(unused_gate);
+	++totalram_pages;
+#endif
+#endif
 
 	/*
 	 * Map the gate page twice: once read-only to export the ELF
 	 * headers etc. and once execute-only page to enable
 	 * privilege-promotion via "epc":
 	 */
-	page = virt_to_page(ia64_imva(__start_gate_section));
+	page = virt_to_page(ia64_imva(gate_page_addr));
 	put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
 #ifdef HAVE_BUGGY_SEGREL
-	page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
+	page = virt_to_page(ia64_imva(gate_page_addr + PAGE_SIZE));
 	put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
 #else
 	put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);