[PATCH v2 13/18] x86/asm/64: Use a percpu trampoline stack for IDT entries

[Andy Lutomirski <luto@kernel.org>]

Historically, IDT entries from usermode have always gone directly
to the running task's kernel stack.  Rearrange it so that we enter on
a percpu trampoline stack and then manually switch to the task's stack.
This touches a couple of extra cachelines, but it gives us a chance
to run some code before we touch the kernel stack.

The asm isn't exactly beautiful, but I think that fully refactoring
it can wait.

Signed-off-by: Andy Lutomirski <luto@kernel.org>
---
 arch/x86/entry/entry_64.S        | 61 ++++++++++++++++++++++++++++++++--------
 arch/x86/entry/entry_64_compat.S |  4 +++
 arch/x86/include/asm/switch_to.h |  2 +-
 arch/x86/include/asm/traps.h     |  1 -
 arch/x86/kernel/cpu/common.c     |  6 ++--
 arch/x86/kernel/traps.c          | 15 +++++-----
 6 files changed, 67 insertions(+), 22 deletions(-)

diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
index 5063ed1214dd..d55785e9dc8e 100644
--- a/arch/x86/entry/entry_64.S
+++ b/arch/x86/entry/entry_64.S
@@ -560,6 +560,14 @@ END(irq_entries_start)
 	.macro interrupt func
 	cld
 	ALLOC_PT_GPREGS_ON_STACK
+
+	testb	$3, CS(%rsp)
+	jz	1f
+	SWAPGS
+	call	switch_to_thread_stack
+	SWAPGS
+1:
+
 	SAVE_C_REGS
 	SAVE_EXTRA_REGS
 	ENCODE_FRAME_POINTER
@@ -827,6 +835,33 @@ apicinterrupt IRQ_WORK_VECTOR			irq_work_interrupt		smp_irq_work_interrupt
  */
 #define CPU_TSS_IST(x) PER_CPU_VAR(cpu_tss) + (TSS_ist + ((x) - 1) * 8)
 
+/*
+ * Switch to the thread stack.  This is called with the IRET frame and
+ * orig_ax in pt_regs and the rest of pt_regs allocated, but with all GPRs
+ * in the CPU registers.
+ */
+ENTRY(switch_to_thread_stack)
+	UNWIND_HINT_IRET_REGS offset=17*8
+
+	movq	%rdi, RDI+8(%rsp)
+	movq	%rsp, %rdi
+	movq	PER_CPU_VAR(cpu_current_top_of_stack), %rsp
+	UNWIND_HINT_IRET_REGS offset=17*8 base=%rdi
+
+	pushq	SS+8(%rdi)		/* regs->ss */
+	pushq	RSP+8(%rdi)		/* regs->rsp */
+	pushq	EFLAGS+8(%rdi)		/* regs->eflags */
+	pushq	CS+8(%rdi)		/* regs->cs */
+	pushq	RIP+8(%rdi)		/* regs->ip */
+	pushq	ORIG_RAX+8(%rdi)	/* regs->orig_ax */
+	ALLOC_PT_GPREGS_ON_STACK	/* allocate the rest of regs */
+	pushq	(%rdi)			/* return address */
+
+	movq	RDI+8(%rdi), %rdi
+	UNWIND_HINT_IRET_REGS offset=17*8
+	ret
+END(switch_to_thread_stack)
+
 .macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1
 ENTRY(\sym)
 	UNWIND_HINT_IRET_REGS offset=\has_error_code*8
@@ -844,11 +879,12 @@ ENTRY(\sym)
 
 	ALLOC_PT_GPREGS_ON_STACK
 
-	.if \paranoid
-	.if \paranoid == 1
+	.if \paranoid < 2
 	testb	$3, CS(%rsp)			/* If coming from userspace, switch stacks */
-	jnz	1f
+	jnz	.Lfrom_usermode_switch_stack_\@
 	.endif
+
+	.if \paranoid
 	call	paranoid_entry
 	.else
 	call	error_entry
@@ -890,20 +926,15 @@ ENTRY(\sym)
 	jmp	error_exit
 	.endif
 
-	.if \paranoid == 1
+	.if \paranoid < 2
 	/*
-	 * Paranoid entry from userspace.  Switch stacks and treat it
+	 * Entry from userspace.  Switch stacks and treat it
 	 * as a normal entry.  This means that paranoid handlers
 	 * run in real process context if user_mode(regs).
 	 */
-1:
+.Lfrom_usermode_switch_stack_\@:
 	call	error_entry
 
-
-	movq	%rsp, %rdi			/* pt_regs pointer */
-	call	sync_regs
-	movq	%rax, %rsp			/* switch stack */
-
 	movq	%rsp, %rdi			/* pt_regs pointer */
 
 	.if \has_error_code
@@ -1164,6 +1195,14 @@ ENTRY(error_entry)
 	SWAPGS
 
 .Lerror_entry_from_usermode_after_swapgs:
+	/* Put us onto the real thread stack. */
+	leaq	8(%rsp), %rdi			/* pt_regs pointer */
+	movq	(%rsp), %r12
+	call	sync_regs
+	movq	%rax, %rsp			/* switch stack */
+	ENCODE_FRAME_POINTER
+	pushq	%r12
+
 	/*
 	 * We need to tell lockdep that IRQs are off.  We can't do this until
 	 * we fix gsbase, and we should do it before enter_from_user_mode
diff --git a/arch/x86/entry/entry_64_compat.S b/arch/x86/entry/entry_64_compat.S
index dcc6987f9bae..20398e3c5e2e 100644
--- a/arch/x86/entry/entry_64_compat.S
+++ b/arch/x86/entry/entry_64_compat.S
@@ -297,6 +297,10 @@ ENTRY(entry_INT80_compat)
 	ASM_CLAC			/* Do this early to minimize exposure */
 	SWAPGS
 
+	subq	$16*8, %rsp
+	call	switch_to_thread_stack
+	addq	$16*8, %rsp
+
 	/*
 	 * User tracing code (ptrace or signal handlers) might assume that
 	 * the saved RAX contains a 32-bit number when we're invoking a 32-bit
diff --git a/arch/x86/include/asm/switch_to.h b/arch/x86/include/asm/switch_to.h
index 8c6bd6863db9..a6796ac8d311 100644
--- a/arch/x86/include/asm/switch_to.h
+++ b/arch/x86/include/asm/switch_to.h
@@ -93,7 +93,7 @@ static inline void update_sp0(struct task_struct *task)
 #ifdef CONFIG_X86_32
 	load_sp0(task->thread.sp0);
 #else
-	load_sp0(task_top_of_stack(task));
+	/* On x86_64, sp0 always points to the entry trampoline stack. */
 #endif
 }
 
diff --git a/arch/x86/include/asm/traps.h b/arch/x86/include/asm/traps.h
index 1fadd310ff68..31051f35cbb7 100644
--- a/arch/x86/include/asm/traps.h
+++ b/arch/x86/include/asm/traps.h
@@ -75,7 +75,6 @@ dotraplinkage void do_segment_not_present(struct pt_regs *, long);
 dotraplinkage void do_stack_segment(struct pt_regs *, long);
 #ifdef CONFIG_X86_64
 dotraplinkage void do_double_fault(struct pt_regs *, long);
-asmlinkage struct pt_regs *sync_regs(struct pt_regs *);
 #endif
 dotraplinkage void do_general_protection(struct pt_regs *, long);
 dotraplinkage void do_page_fault(struct pt_regs *, unsigned long);
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index d551bb4339ec..ddf99d50c26c 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -1646,11 +1646,13 @@ void cpu_init(void)
 	setup_cpu_entry_area(cpu);
 
 	/*
-	 * Initialize the TSS.  Don't bother initializing sp0, as the initial
-	 * task never enters user mode.
+	 * Initialize the TSS.  sp0 points to the entry trampoline stack
+	 * regardless of what task is running.
 	 */
 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
 	load_TR_desc();
+	load_sp0((unsigned long)&get_cpu_entry_area(cpu)->tss +
+		 offsetofend(struct tss_struct, SYSENTER_stack));
 
 	load_mm_ldt(&init_mm);
 
diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index 1ea03027a4a9..1995d5d32521 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -610,9 +610,10 @@ NOKPROBE_SYMBOL(do_int3);
  * interrupted code was in user mode. The actual stack switch is done in
  * entry_64.S
  */
-asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
+asmlinkage __visible notrace
+struct pt_regs *sync_regs(struct pt_regs *eregs)
 {
-	struct pt_regs *regs = task_pt_regs(current);
+	struct pt_regs *regs = (struct pt_regs *)this_cpu_read(cpu_current_top_of_stack) - 1;
 	*regs = *eregs;
 	return regs;
 }
@@ -629,13 +630,13 @@ struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
 	/*
 	 * This is called from entry_64.S early in handling a fault
 	 * caused by a bad iret to user mode.  To handle the fault
-	 * correctly, we want move our stack frame to task_pt_regs
-	 * and we want to pretend that the exception came from the
-	 * iret target.
+	 * correctly, we want move our stack frame to where it would
+	 * be had we entered directly on the entry stack (rather than
+	 * just below the IRET frame) and we want to pretend that the
+	 * exception came from the iret target.
 	 */
 	struct bad_iret_stack *new_stack =
-		container_of(task_pt_regs(current),
-			     struct bad_iret_stack, regs);
+		(struct bad_iret_stack *)this_cpu_read(cpu_tss.x86_tss.sp0) - 1;
 
 	/* Copy the IRET target to the new stack. */
 	memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
-- 
2.13.6