Signed-off-by: Isaku Yamahata <yamahata@xxxxxxxxxxxxx>
---
arch/ia64/kernel/Makefile | 58 +++++++++++++++++++++++++++++++++++++++
arch/ia64/kernel/gate.S | 55 +++++++++++++++++++++++++++++++++++++
arch/ia64/kernel/gate.lds.S | 18 ++++++++++++
arch/ia64/kernel/patch.c | 28 +++++++++++++++++++
arch/ia64/kernel/vmlinux.lds.S | 6 ++++
arch/ia64/kernel/xengate-data.S | 3 ++
arch/ia64/mm/init.c | 22 +++++++++++++-
7 files changed, 188 insertions(+), 2 deletions(-)
create mode 100644 arch/ia64/kernel/xengate-data.S
diff --git a/arch/ia64/kernel/Makefile b/arch/ia64/kernel/Makefile
index d0e22bb..661619c 100644
--- a/arch/ia64/kernel/Makefile
+++ b/arch/ia64/kernel/Makefile
@@ -75,3 +75,61 @@ $(obj)/gate-syms.o: $(obj)/gate.lds $(obj)/gate.o FORCE
# 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
+
+ifeq ($(CONFIG_XEN), y)
+# The gate DSO image is built using a special linker script.
+targets += xengate.so xengate-syms.o
+endif
+
+extra-$(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: $(obj)/xengate.so
+
+#
+# .tmp_gate.o to calculate padding size for __kernel_syscall_via_epc
+#
+extra-$(CONFIG_XEN) += 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 --git a/arch/ia64/kernel/gate.S b/arch/ia64/kernel/gate.S
index 74b1ccc..7c62417 100644
--- a/arch/ia64/kernel/gate.S
+++ b/arch/ia64/kernel/gate.S
@@ -77,15 +77,40 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
epc // B causes split-issue
}
;;
+#ifdef __XEN_IA64_VDSO_PARAVIRT
+ // r20 = 1
+ // r22 = &vcpu->vcpu_info->evtchn_upcall_mask
+ // r24 = &vcpu->vcpu_info->evtchn_upcall_pending
+ // r25 = tmp
+ // r31 = tmp
+ // p11 = tmp
+ // p14 = tmp
+ mov r20=1
+ movl r22=XSI_PSR_I_ADDR
+ ;;
+ 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
;;
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
+ mov r29=r8
+#else
;;
lfetch [r18] // M0|1
mov r29=psr // M2 (12 cyc)
+#endif
// If r17 is a NaT, p6 will be zero
cmp.geu p6,p7=r19,r17 // A (sysnr > 0 && sysnr < 1024+NR_syscalls)?
;;
@@ -99,9 +124,27 @@ 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 __XEN_IA64_VDSO_PARAVIRT
+
+#define XEN_SET_PSR_I(pred) \
+(pred) ld1 r31=[r22]; \
+ ;; ; \
+(pred) st1 [r22]=r0; \
+(pred) cmp.ne.unc p14,p0=r0,r31; \
+ ;; ; \
+(p14) ld1 r25=[r24]; \
+ ;; ; \
+(p14) cmp.ne.unc p11,p0=r0,r25; \
+ ;; ; \
+(p11) XEN_HYPER_SSM_I;
+
+ ;;
+ XEN_SET_PSR_I(p8)
+#else
nop.i 0
;;
(p8) ssm psr.i
+#endif
(p6) mov b7=r18 // I0
(p8) br.dptk.many b7 // B
@@ -122,11 +165,23 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
#else
BRL_COND_FSYS_BUBBLE_DOWN(p6)
#endif
+#ifdef __XEN_IA64_VDSO_PARAVIRT
+ XEN_SET_PSR_I(p0)
+#else
ssm psr.i
+#endif
mov r10=-1
(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 --git a/arch/ia64/kernel/gate.lds.S b/arch/ia64/kernel/gate.lds.S
index 3cb1abc..b9615cf 100644
--- a/arch/ia64/kernel/gate.lds.S
+++ b/arch/ia64/kernel/gate.lds.S
@@ -33,6 +33,24 @@ SECTIONS
. = GATE_ADDR + 0x600;
.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 = .;
diff --git a/arch/ia64/kernel/patch.c b/arch/ia64/kernel/patch.c
index 2cb9425..4676ac4 100644
--- a/arch/ia64/kernel/patch.c
+++ b/arch/ia64/kernel/patch.c
@@ -181,9 +181,37 @@ patch_brl_fsys_bubble_down (unsigned long start, unsigned long end)
ia64_srlz_i();
}
+#ifdef CONFIG_XEN
+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 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)
diff --git a/arch/ia64/kernel/vmlinux.lds.S b/arch/ia64/kernel/vmlinux.lds.S
index 0cbe0a1..0cedc30 100644
--- a/arch/ia64/kernel/vmlinux.lds.S
+++ b/arch/ia64/kernel/vmlinux.lds.S
@@ -230,6 +230,12 @@ SECTIONS
__start_gate_section = .;
*(.data.gate)
__stop_gate_section = .;
+#if defined(CONFIG_XEN)
+ . = ALIGN(PAGE_SIZE);
+ __start_xen_gate_section = .;
+ *(.data.gate.xen)
+ __stop_xen_gate_section = .;
+#endif
}
. = ALIGN(PAGE_SIZE); /* make sure the gate page doesn't expose
* kernel data
diff --git a/arch/ia64/kernel/xengate-data.S b/arch/ia64/kernel/xengate-data.S
new file mode 100644
index 0000000..122bb5f
--- /dev/null
+++ b/arch/ia64/kernel/xengate-data.S
@@ -0,0 +1,3 @@
+ .section .data.gate.xen, "aw"
+
+ .incbin "arch/ia64/kernel/xengate.so"
diff --git a/arch/ia64/mm/init.c b/arch/ia64/mm/init.c
index 25aef62..5e327ac 100644
--- a/arch/ia64/mm/init.c
+++ b/arch/ia64/mm/init.c
@@ -267,16 +267,34 @@ static void __init
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);
--
1.5.3
--
yamahata
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/virtualization
