x86/entry: Add kernel IBRS implementation

Implement Kernel IBRS - currently the only known option to mitigate RSB
underflow speculation issues on Skylake hardware.

Note: since IBRS_ENTER requires fuller context established than
UNTRAIN_RET, it must be placed after it. However, since UNTRAIN_RET
itself implies a RET, it must come after IBRS_ENTER. This means
IBRS_ENTER needs to also move UNTRAIN_RET.

Note 2: KERNEL_IBRS is sub-optimal for XenPV.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Josh Poimboeuf <jpoimboe@kernel.org>
Signed-off-by: Borislav Petkov <bp@suse.de>

arch/x86/entry/calling.h

@@ -7,6 +7,8 @@
 #include <asm/asm-offsets.h>
 #include <asm/processor-flags.h>
 #include <asm/ptrace-abi.h>
+#include <asm/msr.h>
+#include <asm/nospec-branch.h>
 
 /*
 
@@ -282,6 +284,62 @@ For 32-bit we have the following conventions - kernel is built with
 
 #endif
 
+/*
+ * IBRS kernel mitigation for Spectre_v2.
+ *
+ * Assumes full context is established (PUSH_REGS, CR3 and GS) and it clobbers
+ * the regs it uses (AX, CX, DX). Must be called before the first RET
+ * instruction (NOTE! UNTRAIN_RET includes a RET instruction)
+ *
+ * The optional argument is used to save/restore the current value,
+ * which is used on the paranoid paths.
+ *
+ * Assumes x86_spec_ctrl_{base,current} to have SPEC_CTRL_IBRS set.
+ */
+.macro IBRS_ENTER save_reg
+	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_KERNEL_IBRS
+	movl	$MSR_IA32_SPEC_CTRL, %ecx
+
+.ifnb \save_reg
+	rdmsr
+	shl	$32, %rdx
+	or	%rdx, %rax
+	mov	%rax, \save_reg
+	test	$SPEC_CTRL_IBRS, %eax
+	jz	.Ldo_wrmsr_\@
+	lfence
+	jmp	.Lend_\@
+.Ldo_wrmsr_\@:
+.endif
+
+	movq	PER_CPU_VAR(x86_spec_ctrl_current), %rdx
+	movl	%edx, %eax
+	shr	$32, %rdx
+	wrmsr
+.Lend_\@:
+.endm
+
+/*
+ * Similar to IBRS_ENTER, requires KERNEL GS,CR3 and clobbers (AX, CX, DX)
+ * regs. Must be called after the last RET.
+ */
+.macro IBRS_EXIT save_reg
+	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_KERNEL_IBRS
+	movl	$MSR_IA32_SPEC_CTRL, %ecx
+
+.ifnb \save_reg
+	mov	\save_reg, %rdx
+.else
+	movq	PER_CPU_VAR(x86_spec_ctrl_current), %rdx
+	andl	$(~SPEC_CTRL_IBRS), %edx
+.endif
+
+	movl	%edx, %eax
+	shr	$32, %rdx
+	wrmsr
+.Lend_\@:
+.endm
+
 /*
  * Mitigate Spectre v1 for conditional swapgs code paths.
  *

arch/x86/entry/entry_64.S

@@ -96,7 +96,6 @@ SYM_CODE_START(entry_SYSCALL_64)
 
 SYM_INNER_LABEL(entry_SYSCALL_64_safe_stack, SYM_L_GLOBAL)
 	ANNOTATE_NOENDBR
-	UNTRAIN_RET
 
 	/* Construct struct pt_regs on stack */
 	pushq	$__USER_DS				/* pt_regs->ss */
@@ -113,6 +112,11 @@ SYM_INNER_LABEL(entry_SYSCALL_64_after_hwframe, SYM_L_GLOBAL)
 	movq	%rsp, %rdi
 	/* Sign extend the lower 32bit as syscall numbers are treated as int */
 	movslq	%eax, %rsi
+
+	/* clobbers %rax, make sure it is after saving the syscall nr */
+	IBRS_ENTER
+	UNTRAIN_RET
+
 	call	do_syscall_64		/* returns with IRQs disabled */
 
 	/*
@@ -192,6 +196,7 @@ SYM_INNER_LABEL(entry_SYSCALL_64_after_hwframe, SYM_L_GLOBAL)
 	 * perf profiles. Nothing jumps here.
 	 */
 syscall_return_via_sysret:
+	IBRS_EXIT
 	POP_REGS pop_rdi=0
 
 	/*
@@ -605,6 +610,7 @@ __irqentry_text_end:
 
 SYM_CODE_START_LOCAL(common_interrupt_return)
 SYM_INNER_LABEL(swapgs_restore_regs_and_return_to_usermode, SYM_L_GLOBAL)
+	IBRS_EXIT
 #ifdef CONFIG_DEBUG_ENTRY
 	/* Assert that pt_regs indicates user mode. */
 	testb	$3, CS(%rsp)
@@ -891,6 +897,9 @@ SYM_CODE_END(xen_failsafe_callback)
  *              1 -> no SWAPGS on exit
  *
  *     Y        GSBASE value at entry, must be restored in paranoid_exit
+ *
+ * R14 - old CR3
+ * R15 - old SPEC_CTRL
  */
 SYM_CODE_START_LOCAL(paranoid_entry)
 	UNWIND_HINT_FUNC
@@ -913,7 +922,6 @@ SYM_CODE_START_LOCAL(paranoid_entry)
 	 * be retrieved from a kernel internal table.
 	 */
 	SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg=%rax save_reg=%r14
-	UNTRAIN_RET
 
 	/*
 	 * Handling GSBASE depends on the availability of FSGSBASE.
@@ -935,7 +943,7 @@ SYM_CODE_START_LOCAL(paranoid_entry)
 	 * is needed here.
 	 */
 	SAVE_AND_SET_GSBASE scratch_reg=%rax save_reg=%rbx
-	RET
+	jmp .Lparanoid_gsbase_done
 
 .Lparanoid_entry_checkgs:
 	/* EBX = 1 -> kernel GSBASE active, no restore required */
@@ -954,8 +962,16 @@ SYM_CODE_START_LOCAL(paranoid_entry)
 	xorl	%ebx, %ebx
 	swapgs
 .Lparanoid_kernel_gsbase:
-
 	FENCE_SWAPGS_KERNEL_ENTRY
+.Lparanoid_gsbase_done:
+
+	/*
+	 * Once we have CR3 and %GS setup save and set SPEC_CTRL. Just like
+	 * CR3 above, keep the old value in a callee saved register.
+	 */
+	IBRS_ENTER save_reg=%r15
+	UNTRAIN_RET
+
 	RET
 SYM_CODE_END(paranoid_entry)
 
@@ -977,9 +993,19 @@ SYM_CODE_END(paranoid_entry)
  *              1 -> no SWAPGS on exit
  *
  *     Y        User space GSBASE, must be restored unconditionally
+ *
+ * R14 - old CR3
+ * R15 - old SPEC_CTRL
  */
 SYM_CODE_START_LOCAL(paranoid_exit)
 	UNWIND_HINT_REGS
+
+	/*
+	 * Must restore IBRS state before both CR3 and %GS since we need access
+	 * to the per-CPU x86_spec_ctrl_shadow variable.
+	 */
+	IBRS_EXIT save_reg=%r15
+
 	/*
 	 * The order of operations is important. RESTORE_CR3 requires
 	 * kernel GSBASE.
@@ -1023,10 +1049,12 @@ SYM_CODE_START_LOCAL(error_entry)
 	FENCE_SWAPGS_USER_ENTRY
 	/* We have user CR3.  Change to kernel CR3. */
 	SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
+	IBRS_ENTER
 	UNTRAIN_RET
 
 	leaq	8(%rsp), %rdi			/* arg0 = pt_regs pointer */
 .Lerror_entry_from_usermode_after_swapgs:
+
 	/* Put us onto the real thread stack. */
 	call	sync_regs
 	RET
@@ -1076,6 +1104,7 @@ SYM_CODE_START_LOCAL(error_entry)
 	swapgs
 	FENCE_SWAPGS_USER_ENTRY
 	SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
+	IBRS_ENTER
 	UNTRAIN_RET
 
 	/*
@@ -1172,7 +1201,6 @@ SYM_CODE_START(asm_exc_nmi)
 	movq	%rsp, %rdx
 	movq	PER_CPU_VAR(cpu_current_top_of_stack), %rsp
 	UNWIND_HINT_IRET_REGS base=%rdx offset=8
-	UNTRAIN_RET
 	pushq	5*8(%rdx)	/* pt_regs->ss */
 	pushq	4*8(%rdx)	/* pt_regs->rsp */
 	pushq	3*8(%rdx)	/* pt_regs->flags */
@@ -1183,6 +1211,9 @@ SYM_CODE_START(asm_exc_nmi)
 	PUSH_AND_CLEAR_REGS rdx=(%rdx)
 	ENCODE_FRAME_POINTER
 
+	IBRS_ENTER
+	UNTRAIN_RET
+
 	/*
 	 * At this point we no longer need to worry about stack damage
 	 * due to nesting -- we're on the normal thread stack and we're
@@ -1407,6 +1438,9 @@ end_repeat_nmi:
 	movq	$-1, %rsi
 	call	exc_nmi
 
+	/* Always restore stashed SPEC_CTRL value (see paranoid_entry) */
+	IBRS_EXIT save_reg=%r15
+
 	/* Always restore stashed CR3 value (see paranoid_entry) */
 	RESTORE_CR3 scratch_reg=%r15 save_reg=%r14
 

arch/x86/entry/entry_64_compat.S

@@ -4,7 +4,6 @@
  *
  * Copyright 2000-2002 Andi Kleen, SuSE Labs.
  */
-#include "calling.h"
 #include <asm/asm-offsets.h>
 #include <asm/current.h>
 #include <asm/errno.h>
@@ -18,6 +17,8 @@
 #include <linux/linkage.h>
 #include <linux/err.h>
 
+#include "calling.h"
+
 	.section .entry.text, "ax"
 
 /*
@@ -73,7 +74,6 @@ SYM_CODE_START(entry_SYSENTER_compat)
 	pushq	$__USER32_CS		/* pt_regs->cs */
 	pushq	$0			/* pt_regs->ip = 0 (placeholder) */
 SYM_INNER_LABEL(entry_SYSENTER_compat_after_hwframe, SYM_L_GLOBAL)
-	UNTRAIN_RET
 
 	/*
 	 * User tracing code (ptrace or signal handlers) might assume that
@@ -90,6 +90,9 @@ SYM_INNER_LABEL(entry_SYSENTER_compat_after_hwframe, SYM_L_GLOBAL)
 
 	cld
 
+	IBRS_ENTER
+	UNTRAIN_RET
+
 	/*
 	 * SYSENTER doesn't filter flags, so we need to clear NT and AC
 	 * ourselves.  To save a few cycles, we can check whether
@@ -192,7 +195,6 @@ SYM_CODE_START(entry_SYSCALL_compat)
 
 SYM_INNER_LABEL(entry_SYSCALL_compat_safe_stack, SYM_L_GLOBAL)
 	ANNOTATE_NOENDBR
-	UNTRAIN_RET
 
 	/* Construct struct pt_regs on stack */
 	pushq	$__USER32_DS		/* pt_regs->ss */
@@ -206,6 +208,9 @@ SYM_INNER_LABEL(entry_SYSCALL_compat_after_hwframe, SYM_L_GLOBAL)
 	PUSH_AND_CLEAR_REGS rcx=%rbp rax=$-ENOSYS
 	UNWIND_HINT_REGS
 
+	IBRS_ENTER
+	UNTRAIN_RET
+
 	movq	%rsp, %rdi
 	call	do_fast_syscall_32
 	/* XEN PV guests always use IRET path */
@@ -220,6 +225,8 @@ sysret32_from_system_call:
 	 */
 	STACKLEAK_ERASE
 
+	IBRS_EXIT
+
 	movq	RBX(%rsp), %rbx		/* pt_regs->rbx */
 	movq	RBP(%rsp), %rbp		/* pt_regs->rbp */
 	movq	EFLAGS(%rsp), %r11	/* pt_regs->flags (in r11) */
@@ -335,12 +342,14 @@ SYM_CODE_START(entry_INT80_compat)
 	pushq	0*8(%rax)		/* regs->orig_ax */
 .Lint80_keep_stack:
 
-	UNTRAIN_RET
 	PUSH_AND_CLEAR_REGS rax=$-ENOSYS
 	UNWIND_HINT_REGS
 
 	cld
 
+	IBRS_ENTER
+	UNTRAIN_RET
+
 	movq	%rsp, %rdi
 	call	do_int80_syscall_32
 	jmp	swapgs_restore_regs_and_return_to_usermode

arch/x86/include/asm/cpufeatures.h

@@ -203,7 +203,7 @@
 #define X86_FEATURE_PROC_FEEDBACK	( 7*32+ 9) /* AMD ProcFeedbackInterface */
 #define X86_FEATURE_XCOMPACTED		( 7*32+10) /* "" Use compacted XSTATE (XSAVES or XSAVEC) */
 #define X86_FEATURE_PTI			( 7*32+11) /* Kernel Page Table Isolation enabled */
-/* FREE!				( 7*32+12) */
+#define X86_FEATURE_KERNEL_IBRS		( 7*32+12) /* "" Set/clear IBRS on kernel entry/exit */
 /* FREE!				( 7*32+13) */
 #define X86_FEATURE_INTEL_PPIN		( 7*32+14) /* Intel Processor Inventory Number */
 #define X86_FEATURE_CDP_L2		( 7*32+15) /* Code and Data Prioritization L2 */