x86-64, espfix: Don't leak bits 31:16 of %esp returning to 16-bit stack
diff mbox series

Message ID 1398816946-3351-1-git-send-email-hpa@linux.intel.com
State New, archived
Headers show
Series
  • x86-64, espfix: Don't leak bits 31:16 of %esp returning to 16-bit stack
Related show

Commit Message

H. Peter Anvin April 30, 2014, 12:15 a.m. UTC
This is intended as a final RFC for testing and flames^Wcomments
before I add this to -tip.

Linus, this is technically a (functionality) regression fix.  Will you
want this for 3.15?

---------->8--------

The IRET instruction, when returning to a 16-bit segment, only
restores the bottom 16 bits of the user space stack pointer.  This
causes some 16-bit software to break, but it also leaks kernel state
to user space.  We have a software workaround for that ("espfix") for
the 32-bit kernel, but it relies on a nonzero stack segment base which
is not available in 32-bit mode.

In checkin:

    b3b42ac2cbae x86-64, modify_ldt: Ban 16-bit segments on 64-bit kernels

we "solved" this by forbidding 16-bit segments on 64-bit kernels, with
the logic that 16-bit support is crippled on 64-bit kernels anyway (no
V86 support), but it turns out that people are doing stuff like
running old Win16 binaries under Wine and expect it to work.

This works around this by creating percpu "ministacks", each of which
is mapped 2^16 times 64K apart.  When we detect that the return SS is
on the LDT, we copy the IRET frame to the ministack and use the
relevant alias to return to userspace.  The ministacks are mapped
readonly, so if the IRET fault we promote #GP to #DF which is an IST
vector and thus has its own stack; we then do the fixup in the #DF
handler.

(Making #GP an IST exception would make the msr_safe functions unsafe
in NMI/MC context, and quite possibly have other effects.)

Special thanks to:

- Andy Lutomirski, for the suggestion of using very small stack slots
  and copy (as opposed to map) the IRET frame there, and for the
  suggestion to mark them readonly and let the fault promote to #DF.
- Konrad Wilk for paravirt fixup and testing.
- Borislav Petkov for testing help and useful comments.

This is technically a fix for a functionality regression.

Reported-by: Brian Gerst <brgerst@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Andrew Lutomriski <amluto@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Dirk Hohndel <dirk@hohndel.org>
Cc: Arjan van de Ven <arjan.van.de.ven@intel.com>
Cc: comex <comexk@gmail.com>
Cc: Alexander van Heukelum <heukelum@fastmail.fm>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
---
 Documentation/x86/x86_64/mm.txt         |   2 +
 arch/x86/include/asm/pgtable_64_types.h |   2 +
 arch/x86/include/asm/setup.h            |   3 +
 arch/x86/kernel/Makefile                |   1 +
 arch/x86/kernel/entry_64.S              |  73 +++++++++++-
 arch/x86/kernel/espfix_64.c             | 203 ++++++++++++++++++++++++++++++++
 arch/x86/kernel/ldt.c                   |  11 --
 arch/x86/kernel/smpboot.c               |   7 ++
 arch/x86/mm/dump_pagetables.c           |  44 +++++--
 init/main.c                             |   4 +
 10 files changed, 324 insertions(+), 26 deletions(-)
 create mode 100644 arch/x86/kernel/espfix_64.c

Comments

Linus Torvalds April 30, 2014, 12:57 a.m. UTC | #1
On Tue, Apr 29, 2014 at 5:15 PM, H. Peter Anvin <hpa@linux.intel.com> wrote:
>
> Linus, this is technically a (functionality) regression fix.  Will you
> want this for 3.15?

Too scary. If people complain, I think we should just replace the

        if (!ldt_info.seg_32bit) {

with something like

        if (!ldt_info.seg_32bit && !sysctl_allow_16bit_segments) {

and handle the regression that way.

Then, after way more testing, maybe we can backport the espfix thing.

            Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Borislav Petkov April 30, 2014, 10:38 a.m. UTC | #2
On Tue, Apr 29, 2014 at 05:15:46PM -0700, H. Peter Anvin wrote:
> This is intended as a final RFC for testing and flames^Wcomments
> before I add this to -tip.
> 
> Linus, this is technically a (functionality) regression fix.  Will you
> want this for 3.15?
> 
> ---------->8--------
> 
> The IRET instruction, when returning to a 16-bit segment, only
> restores the bottom 16 bits of the user space stack pointer.  This
> causes some 16-bit software to break, but it also leaks kernel state
> to user space.  We have a software workaround for that ("espfix") for
> the 32-bit kernel, but it relies on a nonzero stack segment base which
> is not available in 32-bit mode.

s/32-bit/64-bit/

> 
> In checkin:
> 
>     b3b42ac2cbae x86-64, modify_ldt: Ban 16-bit segments on 64-bit kernels
> 
> we "solved" this by forbidding 16-bit segments on 64-bit kernels, with
> the logic that 16-bit support is crippled on 64-bit kernels anyway (no
> V86 support), but it turns out that people are doing stuff like
> running old Win16 binaries under Wine and expect it to work.
> 
> This works around this by creating percpu "ministacks", each of which
> is mapped 2^16 times 64K apart.  When we detect that the return SS is
> on the LDT, we copy the IRET frame to the ministack and use the
> relevant alias to return to userspace.  The ministacks are mapped
> readonly, so if the IRET fault we promote #GP to #DF which is an IST

"... so if IRET faults... "

> vector and thus has its own stack; we then do the fixup in the #DF
> handler.
> 
> (Making #GP an IST exception would make the msr_safe functions unsafe
> in NMI/MC context, and quite possibly have other effects.)
> 
> Special thanks to:
> 
> - Andy Lutomirski, for the suggestion of using very small stack slots
>   and copy (as opposed to map) the IRET frame there, and for the
>   suggestion to mark them readonly and let the fault promote to #DF.
> - Konrad Wilk for paravirt fixup and testing.
> - Borislav Petkov for testing help and useful comments.
> 
> This is technically a fix for a functionality regression.
> 
> Reported-by: Brian Gerst <brgerst@gmail.com>
> Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
> Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
> Cc: Borislav Petkov <bp@alien8.de>
> Cc: Andrew Lutomriski <amluto@gmail.com>
> Cc: Linus Torvalds <torvalds@linux-foundation.org>
> Cc: Dirk Hohndel <dirk@hohndel.org>
> Cc: Arjan van de Ven <arjan.van.de.ven@intel.com>
> Cc: comex <comexk@gmail.com>
> Cc: Alexander van Heukelum <heukelum@fastmail.fm>
> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> ---

...

> @@ -1110,9 +1143,41 @@ ENTRY(retint_kernel)
>  	call preempt_schedule_irq
>  	jmp exit_intr
>  #endif
> -
>  	CFI_ENDPROC
>  END(common_interrupt)
> +
> +	/*
> +	 * If IRET takes a fault on the espfix stack, then we
> +	 * end up promoting it to a doublefault.  In that case,
> +	 * modify the stack to make it look like we just entered
> +	 * the #GP handler from user space, similar to bad_iret.
> +	 */
> +	ALIGN
> +__do_double_fault:
> +	XCPT_FRAME 1 RDI+8
> +	movq RSP(%rdi),%rax		/* Trap on the espfix stack? */
> +	sarq $PGDIR_SHIFT,%rax
> +	cmpl $ESPFIX_PGD_ENTRY,%eax
> +	jne do_double_fault		/* No, just deliver the fault */
> +	cmpl $__KERNEL_CS,CS(%rdi)

What will happen more likely and thus more often - our "simulated" #DF
or a real one? Judging by the order of the tests, you're saying: the
simulated one. :-)

Otherwise, push the __KERNEL_CS test up?

> +	jne do_double_fault
> +	movq RIP(%rdi),%rax
> +	cmpq $irq_return_iret,%rax
> +#ifdef CONFIG_PARAVIRT
> +	je 1f
> +	cmpq $native_iret,%rax
> +#endif
> +	jne do_double_fault		/* This shouldn't happen... */
> +1:
> +	movq PER_CPU_VAR(kernel_stack),%rax
> +	subq $(6*8-KERNEL_STACK_OFFSET),%rax	/* Reset to original stack */
> +	movq %rax,RSP(%rdi)
> +	movq $0,(%rax)			/* Missing (lost) #GP error code */
> +	movq $general_protection,RIP(%rdi)
> +	retq
> +	CFI_ENDPROC
> +END(__do_double_fault)
> +
>  /*
>   * End of kprobes section
>   */

...

> diff --git a/arch/x86/kernel/espfix_64.c b/arch/x86/kernel/espfix_64.c
> new file mode 100644
> index 000000000000..b1b5ae21a73e
> --- /dev/null
> +++ b/arch/x86/kernel/espfix_64.c
> @@ -0,0 +1,203 @@
> +/* ----------------------------------------------------------------------- *
> + *
> + *   Copyright 2014 Intel Corporation; author: H. Peter Anvin
> + *
> + *   This file is part of the Linux kernel, and is made available under
> + *   the terms of the GNU General Public License version 2 or (at your
> + *   option) any later version; incorporated herein by reference.
> + *
> + * ----------------------------------------------------------------------- */

Yah, this is how big a licensing boilerplate should be! Maybe other
companies would look at this here and learn.

> +
> +/*
> + * The IRET instruction, when returning to a 16-bit segment, only
> + * restores the bottom 16 bits of the user space stack pointer.  This
> + * causes some 16-bit software to break, but it also leaks kernel state
> + * to user space.
> + *
> + * This works around this by creating percpu "ministacks", each of which
> + * is mapped 2^16 times 64K apart.  When we detect that the return SS is
> + * on the LDT, we copy the IRET frame to the ministack and use the
> + * relevant alias to return to userspace.  The ministacks are mapped
> + * readonly, so if the IRET fault we promote #GP to #DF which is an IST

"... so if IRET faults... "

> + * vector and thus has its own stack; we then do the fixup in the #DF
> + * handler.
> + *
> + * This file sets up the ministacks and the related page tables.  The
> + * actual ministack invocation is in entry_64.S.
> + */

Yep, this is how you write an explanation of the issue!

...

> @@ -208,17 +213,24 @@ static void note_page(struct seq_file *m, struct pg_state *st,
>  		/*
>  		 * Now print the actual finished series
>  		 */
> -		pt_dump_seq_printf(m, st->to_dmesg,  "0x%0*lx-0x%0*lx   ",
> -				   width, st->start_address,
> -				   width, st->current_address);
> -
> -		delta = (st->current_address - st->start_address) >> 10;
> -		while (!(delta & 1023) && unit[1]) {
> -			delta >>= 10;
> -			unit++;
> +		if (!st->marker->max_lines ||
> +		    st->lines < st->marker->max_lines) {
> +			pt_dump_seq_printf(m, st->to_dmesg, 
> +					   "0x%0*lx-0x%0*lx   ",
> +					   width, st->start_address,
> +					   width, st->current_address);
> +
> +			delta = st->current_address - st->start_address;
> +			while (!(delta & 1023) && unit[1]) {
> +				delta >>= 10;
> +				unit++;
> +			}
> +			pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
> +					    delta, *unit);
> +			printk_prot(m, st->current_prot, st->level,
> +				    st->to_dmesg);
>  		}
> -		pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ", delta, *unit);
> -		printk_prot(m, st->current_prot, st->level, st->to_dmesg);
> +		st->lines++;
>  
>  		/*
>  		 * We print markers for special areas of address space,
> @@ -226,7 +238,17 @@ static void note_page(struct seq_file *m, struct pg_state *st,
>  		 * This helps in the interpretation.
>  		 */
>  		if (st->current_address >= st->marker[1].start_address) {
> +			if (st->marker->max_lines &&
> +			    st->lines > st->marker->max_lines) {
> +				unsigned long nskip =
> +					st->lines - st->marker->max_lines;
> +				pt_dump_seq_printf(m, st->to_dmesg,
> +						   "... %lu entr%s skipped ... \n",
> +						   nskip,
> +						   nskip == 1 ? "y" : "ies");
> +			}
>  			st->marker++;
> +			st->lines = 0;
>  			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
>  					   st->marker->name);

Nice:

---[ ESPfix Area ]---
0xffffff0000000000-0xffffff1100000000          68G                           pud
0xffffff1100000000-0xffffff110000d000          52K                           pte
0xffffff110000d000-0xffffff110000e000           4K     ro             GLB NX pte
0xffffff110000e000-0xffffff110001d000          60K                           pte
0xffffff110001d000-0xffffff110001e000           4K     ro             GLB NX pte
0xffffff110001e000-0xffffff110002d000          60K                           pte
0xffffff110002d000-0xffffff110002e000           4K     ro             GLB NX pte
0xffffff110002e000-0xffffff110003d000          60K                           pte
0xffffff110003d000-0xffffff110003e000           4K     ro             GLB NX pte
0xffffff110003e000-0xffffff110004d000          60K                           pte
0xffffff110004d000-0xffffff110004e000           4K     ro             GLB NX pte
0xffffff110004e000-0xffffff110005d000          60K                           pte
0xffffff110005d000-0xffffff110005e000           4K     ro             GLB NX pte
0xffffff110005e000-0xffffff110006d000          60K                           pte
0xffffff110006d000-0xffffff110006e000           4K     ro             GLB NX pte
0xffffff110006e000-0xffffff110007d000          60K                           pte
... 131059 entries skipped ...
H. Peter Anvin April 30, 2014, 4:18 p.m. UTC | #3
On 04/30/2014 03:38 AM, Borislav Petkov wrote:
> 
> What will happen more likely and thus more often - our "simulated" #DF
> or a real one? Judging by the order of the tests, you're saying: the
> simulated one. :-)
> 
> Otherwise, push the __KERNEL_CS test up?
> 

Doesn't matter, really: do_double_fault is unconditionally fatal.

	-hpa


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
H. Peter Anvin April 30, 2014, 4:33 p.m. UTC | #4
On 04/30/2014 03:38 AM, Borislav Petkov wrote:
> 
> Yah, this is how big a licensing boilerplate should be! Maybe other
> companies would look at this here and learn.
> 

Heh.  I think I might have used the wrong one, though... GPLv2+ vs GPLv2.

	-hpa

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Borislav Petkov April 30, 2014, 5:16 p.m. UTC | #5
On Wed, Apr 30, 2014 at 09:33:55AM -0700, H. Peter Anvin wrote:
> Heh. I think I might have used the wrong one, though... GPLv2+ vs
> GPLv2.

Why is GPLv2+ wrong?

I thought as long as v2 is explicitly mentioned we're fine.
H. Peter Anvin April 30, 2014, 5:25 p.m. UTC | #6
On 04/30/2014 10:16 AM, Borislav Petkov wrote:
> On Wed, Apr 30, 2014 at 09:33:55AM -0700, H. Peter Anvin wrote:
>> Heh. I think I might have used the wrong one, though... GPLv2+ vs
>> GPLv2.
> 
> Why is GPLv2+ wrong?
> 
> I thought as long as v2 is explicitly mentioned we're fine.
> 

Fine from a Linux kernel perspective, but it is a matter of which rights
the licensor (here, Intel) wants to grant.

	-hpa

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Patch
diff mbox series

diff --git a/Documentation/x86/x86_64/mm.txt b/Documentation/x86/x86_64/mm.txt
index c584a51add15..afe68ddbe6a4 100644
--- a/Documentation/x86/x86_64/mm.txt
+++ b/Documentation/x86/x86_64/mm.txt
@@ -12,6 +12,8 @@  ffffc90000000000 - ffffe8ffffffffff (=45 bits) vmalloc/ioremap space
 ffffe90000000000 - ffffe9ffffffffff (=40 bits) hole
 ffffea0000000000 - ffffeaffffffffff (=40 bits) virtual memory map (1TB)
 ... unused hole ...
+ffffff0000000000 - ffffff7fffffffff (=39 bits) %esp fixup stacks
+... unused hole ...
 ffffffff80000000 - ffffffffa0000000 (=512 MB)  kernel text mapping, from phys 0
 ffffffffa0000000 - ffffffffff5fffff (=1525 MB) module mapping space
 ffffffffff600000 - ffffffffffdfffff (=8 MB) vsyscalls
diff --git a/arch/x86/include/asm/pgtable_64_types.h b/arch/x86/include/asm/pgtable_64_types.h
index c883bf726398..7166e25ecb57 100644
--- a/arch/x86/include/asm/pgtable_64_types.h
+++ b/arch/x86/include/asm/pgtable_64_types.h
@@ -61,6 +61,8 @@  typedef struct { pteval_t pte; } pte_t;
 #define MODULES_VADDR    (__START_KERNEL_map + KERNEL_IMAGE_SIZE)
 #define MODULES_END      _AC(0xffffffffff000000, UL)
 #define MODULES_LEN   (MODULES_END - MODULES_VADDR)
+#define ESPFIX_PGD_ENTRY _AC(-2, UL)
+#define ESPFIX_BASE_ADDR (ESPFIX_PGD_ENTRY << PGDIR_SHIFT)
 
 #define EARLY_DYNAMIC_PAGE_TABLES	64
 
diff --git a/arch/x86/include/asm/setup.h b/arch/x86/include/asm/setup.h
index 9264f04a4c55..9e3be3329a7e 100644
--- a/arch/x86/include/asm/setup.h
+++ b/arch/x86/include/asm/setup.h
@@ -57,6 +57,9 @@  extern void x86_ce4100_early_setup(void);
 static inline void x86_ce4100_early_setup(void) { }
 #endif
 
+extern void init_espfix_bsp(void);
+extern void init_espfix_ap(void);
+
 #ifndef _SETUP
 
 /*
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index f4d96000d33a..1cc3789d99d9 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -29,6 +29,7 @@  obj-$(CONFIG_X86_64)	+= sys_x86_64.o x8664_ksyms_64.o
 obj-y			+= syscall_$(BITS).o vsyscall_gtod.o
 obj-$(CONFIG_X86_64)	+= vsyscall_64.o
 obj-$(CONFIG_X86_64)	+= vsyscall_emu_64.o
+obj-$(CONFIG_X86_64)	+= espfix_64.o
 obj-$(CONFIG_SYSFS)	+= ksysfs.o
 obj-y			+= bootflag.o e820.o
 obj-y			+= pci-dma.o quirks.o topology.o kdebugfs.o
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index 1e96c3628bf2..bffaa986cafc 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -58,6 +58,7 @@ 
 #include <asm/asm.h>
 #include <asm/context_tracking.h>
 #include <asm/smap.h>
+#include <asm/pgtable_types.h>
 #include <linux/err.h>
 
 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this.  */
@@ -1040,8 +1041,16 @@  restore_args:
 	RESTORE_ARGS 1,8,1
 
 irq_return:
+	/*
+	 * Are we returning to a stack segment from the LDT?  Note: in
+	 * 64-bit mode SS:RSP on the exception stack is always valid.
+	 */
+	testb $4,(SS-RIP)(%rsp)
+	jnz irq_return_ldt
+
+irq_return_iret:
 	INTERRUPT_RETURN
-	_ASM_EXTABLE(irq_return, bad_iret)
+	_ASM_EXTABLE(irq_return_iret, bad_iret)
 
 #ifdef CONFIG_PARAVIRT
 ENTRY(native_iret)
@@ -1049,6 +1058,30 @@  ENTRY(native_iret)
 	_ASM_EXTABLE(native_iret, bad_iret)
 #endif
 
+irq_return_ldt:
+	pushq_cfi %rax
+	pushq_cfi %rdi
+	SWAPGS
+	movq PER_CPU_VAR(espfix_waddr),%rdi
+	movq %rax,(0*8)(%rdi)	/* RAX */
+	movq (2*8)(%rsp),%rax	/* RIP */
+	movq %rax,(1*8)(%rdi)
+	movq (3*8)(%rsp),%rax	/* CS */
+	movq %rax,(2*8)(%rdi)
+	movq (4*8)(%rsp),%rax	/* RFLAGS */
+	movq %rax,(3*8)(%rdi)
+	movq (6*8)(%rsp),%rax	/* SS */
+	movq %rax,(5*8)(%rdi)
+	movq (5*8)(%rsp),%rax	/* RSP */
+	movq %rax,(4*8)(%rdi)
+	andl $0xffff0000,%eax
+	popq_cfi %rdi
+	orq PER_CPU_VAR(espfix_stack),%rax
+	SWAPGS
+	movq %rax,%rsp
+	popq_cfi %rax
+	jmp irq_return_iret
+
 	.section .fixup,"ax"
 bad_iret:
 	/*
@@ -1110,9 +1143,41 @@  ENTRY(retint_kernel)
 	call preempt_schedule_irq
 	jmp exit_intr
 #endif
-
 	CFI_ENDPROC
 END(common_interrupt)
+
+	/*
+	 * If IRET takes a fault on the espfix stack, then we
+	 * end up promoting it to a doublefault.  In that case,
+	 * modify the stack to make it look like we just entered
+	 * the #GP handler from user space, similar to bad_iret.
+	 */
+	ALIGN
+__do_double_fault:
+	XCPT_FRAME 1 RDI+8
+	movq RSP(%rdi),%rax		/* Trap on the espfix stack? */
+	sarq $PGDIR_SHIFT,%rax
+	cmpl $ESPFIX_PGD_ENTRY,%eax
+	jne do_double_fault		/* No, just deliver the fault */
+	cmpl $__KERNEL_CS,CS(%rdi)
+	jne do_double_fault
+	movq RIP(%rdi),%rax
+	cmpq $irq_return_iret,%rax
+#ifdef CONFIG_PARAVIRT
+	je 1f
+	cmpq $native_iret,%rax
+#endif
+	jne do_double_fault		/* This shouldn't happen... */
+1:
+	movq PER_CPU_VAR(kernel_stack),%rax
+	subq $(6*8-KERNEL_STACK_OFFSET),%rax	/* Reset to original stack */
+	movq %rax,RSP(%rdi)
+	movq $0,(%rax)			/* Missing (lost) #GP error code */
+	movq $general_protection,RIP(%rdi)
+	retq
+	CFI_ENDPROC
+END(__do_double_fault)
+
 /*
  * End of kprobes section
  */
@@ -1314,7 +1379,7 @@  zeroentry overflow do_overflow
 zeroentry bounds do_bounds
 zeroentry invalid_op do_invalid_op
 zeroentry device_not_available do_device_not_available
-paranoiderrorentry double_fault do_double_fault
+paranoiderrorentry double_fault __do_double_fault
 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
 errorentry invalid_TSS do_invalid_TSS
 errorentry segment_not_present do_segment_not_present
@@ -1601,7 +1666,7 @@  error_sti:
  */
 error_kernelspace:
 	incl %ebx
-	leaq irq_return(%rip),%rcx
+	leaq irq_return_iret(%rip),%rcx
 	cmpq %rcx,RIP+8(%rsp)
 	je error_swapgs
 	movl %ecx,%eax	/* zero extend */
diff --git a/arch/x86/kernel/espfix_64.c b/arch/x86/kernel/espfix_64.c
new file mode 100644
index 000000000000..b1b5ae21a73e
--- /dev/null
+++ b/arch/x86/kernel/espfix_64.c
@@ -0,0 +1,203 @@ 
+/* ----------------------------------------------------------------------- *
+ *
+ *   Copyright 2014 Intel Corporation; author: H. Peter Anvin
+ *
+ *   This file is part of the Linux kernel, and is made available under
+ *   the terms of the GNU General Public License version 2 or (at your
+ *   option) any later version; incorporated herein by reference.
+ *
+ * ----------------------------------------------------------------------- */
+
+/*
+ * The IRET instruction, when returning to a 16-bit segment, only
+ * restores the bottom 16 bits of the user space stack pointer.  This
+ * causes some 16-bit software to break, but it also leaks kernel state
+ * to user space.
+ *
+ * This works around this by creating percpu "ministacks", each of which
+ * is mapped 2^16 times 64K apart.  When we detect that the return SS is
+ * on the LDT, we copy the IRET frame to the ministack and use the
+ * relevant alias to return to userspace.  The ministacks are mapped
+ * readonly, so if the IRET fault we promote #GP to #DF which is an IST
+ * vector and thus has its own stack; we then do the fixup in the #DF
+ * handler.
+ *
+ * This file sets up the ministacks and the related page tables.  The
+ * actual ministack invocation is in entry_64.S.
+ */
+
+#include <linux/init.h>
+#include <linux/init_task.h>
+#include <linux/kernel.h>
+#include <linux/percpu.h>
+#include <linux/gfp.h>
+#include <linux/random.h>
+#include <asm/pgtable.h>
+#include <asm/pgalloc.h>
+#include <asm/setup.h>
+
+/*
+ * Note: we only need 6*8 = 48 bytes for the espfix stack, but round
+ * it up to a cache line to avoid unnecessary sharing.
+ */
+#define ESPFIX_STACK_SIZE	(8*8UL)
+#define ESPFIX_STACKS_PER_PAGE	(PAGE_SIZE/ESPFIX_STACK_SIZE)
+
+/* There is address space for how many espfix pages? */
+#define ESPFIX_PAGE_SPACE	(1UL << (PGDIR_SHIFT-PAGE_SHIFT-16))
+
+#define ESPFIX_MAX_CPUS		(ESPFIX_STACKS_PER_PAGE * ESPFIX_PAGE_SPACE)
+#if CONFIG_NR_CPUS > ESPFIX_MAX_CPUS
+# error "Need more than one PGD for the ESPFIX hack"
+#endif
+
+#define PGALLOC_GFP (GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO)
+
+/* This contains the *bottom* address of the espfix stack */
+DEFINE_PER_CPU_READ_MOSTLY(unsigned long, espfix_stack);
+DEFINE_PER_CPU_READ_MOSTLY(unsigned long, espfix_waddr);
+
+/* Initialization mutex - should this be a spinlock? */
+static DEFINE_MUTEX(espfix_init_mutex);
+
+/* Page allocation bitmap - each page serves ESPFIX_STACKS_PER_PAGE CPUs */
+#define ESPFIX_MAX_PAGES  DIV_ROUND_UP(CONFIG_NR_CPUS, ESPFIX_STACKS_PER_PAGE)
+static void *espfix_pages[ESPFIX_MAX_PAGES];
+
+static __page_aligned_bss pud_t espfix_pud_page[PTRS_PER_PUD]
+	__aligned(PAGE_SIZE);
+
+static unsigned int page_random, slot_random;
+
+/*
+ * This returns the bottom address of the espfix stack for a specific CPU.
+ * The math allows for a non-power-of-two ESPFIX_STACK_SIZE, in which case
+ * we have to account for some amount of padding at the end of each page.
+ */
+static inline unsigned long espfix_base_addr(unsigned int cpu)
+{
+	unsigned long page, slot;
+	unsigned long addr;
+
+	page = (cpu / ESPFIX_STACKS_PER_PAGE) ^ page_random;
+	slot = (cpu + slot_random) % ESPFIX_STACKS_PER_PAGE;
+	addr = (page << PAGE_SHIFT) + (slot * ESPFIX_STACK_SIZE);
+	addr = (addr & 0xffffUL) | ((addr & ~0xffffUL) << 16);
+	addr += ESPFIX_BASE_ADDR;
+	return addr;
+}
+
+#define PTE_STRIDE        (65536/PAGE_SIZE)
+#define ESPFIX_PTE_CLONES (PTRS_PER_PTE/PTE_STRIDE)
+#define ESPFIX_PMD_CLONES PTRS_PER_PMD
+#define ESPFIX_PUD_CLONES (65536/(ESPFIX_PTE_CLONES*ESPFIX_PMD_CLONES))
+
+#define PGTABLE_PROT	  ((_KERNPG_TABLE & ~_PAGE_RW) | _PAGE_NX)
+
+static void init_espfix_random(void)
+{
+	unsigned long rand;
+
+	/*
+	 * This is run before the entropy pools are initialized,
+	 * but this is hopefully better than nothing.
+	 */
+	if (!arch_get_random_long(&rand)) {
+		/* The constant is an arbitrary large prime */
+		rdtscll(rand);
+		rand *= 0xc345c6b72fd16123UL;
+	}
+
+	slot_random = rand % ESPFIX_STACKS_PER_PAGE;
+	page_random = (rand / ESPFIX_STACKS_PER_PAGE)
+		& (ESPFIX_PAGE_SPACE - 1);
+}
+
+void __init init_espfix_bsp(void)
+{
+	pgd_t *pgd_p;
+	pteval_t ptemask;
+
+	ptemask = __supported_pte_mask;
+
+	/* Install the espfix pud into the kernel page directory */
+	pgd_p = &init_level4_pgt[pgd_index(ESPFIX_BASE_ADDR)];
+	pgd_populate(&init_mm, pgd_p, (pud_t *)espfix_pud_page);
+
+	/* Randomize the locations */
+	init_espfix_random();
+
+	/* The rest is the same as for any other processor */
+	init_espfix_ap();
+}
+
+void init_espfix_ap(void)
+{
+	unsigned int cpu, page;
+	unsigned long addr;
+	pud_t pud, *pud_p;
+	pmd_t pmd, *pmd_p;
+	pte_t pte, *pte_p;
+	int n;
+	void *stack_page;
+	pteval_t ptemask;
+
+	/* We only have to do this once... */
+	if (likely(this_cpu_read(espfix_stack)))
+		return;		/* Already initialized */
+
+	cpu = smp_processor_id();
+	addr = espfix_base_addr(cpu);
+	page = cpu/ESPFIX_STACKS_PER_PAGE;
+
+	/* Did another CPU already set this up? */
+	stack_page = ACCESS_ONCE(espfix_pages[page]);
+	if (likely(stack_page))
+		goto done;
+
+	mutex_lock(&espfix_init_mutex);
+
+	/* Did we race on the lock? */
+	stack_page = ACCESS_ONCE(espfix_pages[page]);
+	if (stack_page)
+		goto unlock_done;
+
+	ptemask = __supported_pte_mask;
+
+	pud_p = &espfix_pud_page[pud_index(addr)];
+	pud = *pud_p;
+	if (!pud_present(pud)) {
+		pmd_p = (pmd_t *)__get_free_page(PGALLOC_GFP);
+		pud = __pud(__pa(pmd_p) | (PGTABLE_PROT & ptemask));
+		paravirt_alloc_pud(&init_mm, __pa(pmd_p) >> PAGE_SHIFT);
+		for (n = 0; n < ESPFIX_PUD_CLONES; n++)
+			set_pud(&pud_p[n], pud);
+	}
+
+	pmd_p = pmd_offset(&pud, addr);
+	pmd = *pmd_p;
+	if (!pmd_present(pmd)) {
+		pte_p = (pte_t *)__get_free_page(PGALLOC_GFP);
+		pmd = __pmd(__pa(pte_p) | (PGTABLE_PROT & ptemask));
+		paravirt_alloc_pmd(&init_mm, __pa(pte_p) >> PAGE_SHIFT);
+		for (n = 0; n < ESPFIX_PMD_CLONES; n++)
+			set_pmd(&pmd_p[n], pmd);
+	}
+
+	pte_p = pte_offset_kernel(&pmd, addr);
+	stack_page = (void *)__get_free_page(GFP_KERNEL);
+	pte = __pte(__pa(stack_page) | (__PAGE_KERNEL_RO & ptemask));
+	paravirt_alloc_pte(&init_mm, __pa(stack_page) >> PAGE_SHIFT);
+	for (n = 0; n < ESPFIX_PTE_CLONES; n++)
+		set_pte(&pte_p[n*PTE_STRIDE], pte);
+
+	/* Job is done for this CPU and any CPU which shares this page */
+	ACCESS_ONCE(espfix_pages[page]) = stack_page;
+
+unlock_done:
+	mutex_unlock(&espfix_init_mutex);
+done:
+	this_cpu_write(espfix_stack, addr);
+	this_cpu_write(espfix_waddr, (unsigned long)stack_page
+		       + (addr & ~PAGE_MASK));
+}
diff --git a/arch/x86/kernel/ldt.c b/arch/x86/kernel/ldt.c
index af1d14a9ebda..ebc987398923 100644
--- a/arch/x86/kernel/ldt.c
+++ b/arch/x86/kernel/ldt.c
@@ -229,17 +229,6 @@  static int write_ldt(void __user *ptr, unsigned long bytecount, int oldmode)
 		}
 	}
 
-	/*
-	 * On x86-64 we do not support 16-bit segments due to
-	 * IRET leaking the high bits of the kernel stack address.
-	 */
-#ifdef CONFIG_X86_64
-	if (!ldt_info.seg_32bit) {
-		error = -EINVAL;
-		goto out_unlock;
-	}
-#endif
-
 	fill_ldt(&ldt, &ldt_info);
 	if (oldmode)
 		ldt.avl = 0;
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 34826934d4a7..61a5350850fb 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -244,6 +244,13 @@  static void notrace start_secondary(void *unused)
 	check_tsc_sync_target();
 
 	/*
+	 * Enable the espfix hack for this CPU
+	 */
+#ifdef CONFIG_X86_64
+	init_espfix_ap();
+#endif
+
+	/*
 	 * We need to hold vector_lock so there the set of online cpus
 	 * does not change while we are assigning vectors to cpus.  Holding
 	 * this lock ensures we don't half assign or remove an irq from a cpu.
diff --git a/arch/x86/mm/dump_pagetables.c b/arch/x86/mm/dump_pagetables.c
index 20621d753d5f..c6d4c1fe7004 100644
--- a/arch/x86/mm/dump_pagetables.c
+++ b/arch/x86/mm/dump_pagetables.c
@@ -30,12 +30,14 @@  struct pg_state {
 	unsigned long start_address;
 	unsigned long current_address;
 	const struct addr_marker *marker;
+	unsigned long lines;
 	bool to_dmesg;
 };
 
 struct addr_marker {
 	unsigned long start_address;
 	const char *name;
+	unsigned long max_lines;
 };
 
 /* indices for address_markers; keep sync'd w/ address_markers below */
@@ -46,6 +48,7 @@  enum address_markers_idx {
 	LOW_KERNEL_NR,
 	VMALLOC_START_NR,
 	VMEMMAP_START_NR,
+	ESPFIX_START_NR,
 	HIGH_KERNEL_NR,
 	MODULES_VADDR_NR,
 	MODULES_END_NR,
@@ -68,6 +71,7 @@  static struct addr_marker address_markers[] = {
 	{ PAGE_OFFSET,		"Low Kernel Mapping" },
 	{ VMALLOC_START,        "vmalloc() Area" },
 	{ VMEMMAP_START,        "Vmemmap" },
+	{ ESPFIX_BASE_ADDR,	"ESPfix Area", 16 },
 	{ __START_KERNEL_map,   "High Kernel Mapping" },
 	{ MODULES_VADDR,        "Modules" },
 	{ MODULES_END,          "End Modules" },
@@ -182,7 +186,7 @@  static void note_page(struct seq_file *m, struct pg_state *st,
 		      pgprot_t new_prot, int level)
 {
 	pgprotval_t prot, cur;
-	static const char units[] = "KMGTPE";
+	static const char units[] = "BKMGTPE";
 
 	/*
 	 * If we have a "break" in the series, we need to flush the state that
@@ -197,6 +201,7 @@  static void note_page(struct seq_file *m, struct pg_state *st,
 		st->current_prot = new_prot;
 		st->level = level;
 		st->marker = address_markers;
+		st->lines = 0;
 		pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
 				   st->marker->name);
 	} else if (prot != cur || level != st->level ||
@@ -208,17 +213,24 @@  static void note_page(struct seq_file *m, struct pg_state *st,
 		/*
 		 * Now print the actual finished series
 		 */
-		pt_dump_seq_printf(m, st->to_dmesg,  "0x%0*lx-0x%0*lx   ",
-				   width, st->start_address,
-				   width, st->current_address);
-
-		delta = (st->current_address - st->start_address) >> 10;
-		while (!(delta & 1023) && unit[1]) {
-			delta >>= 10;
-			unit++;
+		if (!st->marker->max_lines ||
+		    st->lines < st->marker->max_lines) {
+			pt_dump_seq_printf(m, st->to_dmesg, 
+					   "0x%0*lx-0x%0*lx   ",
+					   width, st->start_address,
+					   width, st->current_address);
+
+			delta = st->current_address - st->start_address;
+			while (!(delta & 1023) && unit[1]) {
+				delta >>= 10;
+				unit++;
+			}
+			pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
+					    delta, *unit);
+			printk_prot(m, st->current_prot, st->level,
+				    st->to_dmesg);
 		}
-		pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ", delta, *unit);
-		printk_prot(m, st->current_prot, st->level, st->to_dmesg);
+		st->lines++;
 
 		/*
 		 * We print markers for special areas of address space,
@@ -226,7 +238,17 @@  static void note_page(struct seq_file *m, struct pg_state *st,
 		 * This helps in the interpretation.
 		 */
 		if (st->current_address >= st->marker[1].start_address) {
+			if (st->marker->max_lines &&
+			    st->lines > st->marker->max_lines) {
+				unsigned long nskip =
+					st->lines - st->marker->max_lines;
+				pt_dump_seq_printf(m, st->to_dmesg,
+						   "... %lu entr%s skipped ... \n",
+						   nskip,
+						   nskip == 1 ? "y" : "ies");
+			}
 			st->marker++;
+			st->lines = 0;
 			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
 					   st->marker->name);
 		}
diff --git a/init/main.c b/init/main.c
index 9c7fd4c9249f..70fc00e7db06 100644
--- a/init/main.c
+++ b/init/main.c
@@ -617,6 +617,10 @@  asmlinkage void __init start_kernel(void)
 	if (efi_enabled(EFI_RUNTIME_SERVICES))
 		efi_enter_virtual_mode();
 #endif
+#ifdef CONFIG_X86_64
+	/* Should be run before the first non-init thread is created */
+	init_espfix_bsp();
+#endif
 	thread_info_cache_init();
 	cred_init();
 	fork_init(totalram_pages);