[PATCH v2 0/2] x86/entry: Do context tracking with IRQs off (finally!)

[PATCH v2 0/2] x86/entry: Do context tracking with IRQs off (finally!)

[Andy Lutomirski]

This provides no direct benefit by itself, but it finally lets us
apply some of the speedups that Paolo and possible Rik independently
wrote: we won't need an irqsave/irqrestore pair in the context
tracking code.

It should also let us remove the last callers of exception_enter in
x86 kernels, but that removal should probably wait until I carefully
re-read the page fault code.

Changes from v1:
 - Fix a comment
 - Make the ifdeffery less messy

Andy Lutomirski (2):
  x86/entry/32: Change INT80 to be an interrupt gate
  x86/entry: Do enter_from_user_mode with IRQs off

 arch/x86/entry/common.c            | 48 +++++++++++---------------------------
 arch/x86/entry/entry_32.S          |  8 +++----
 arch/x86/entry/entry_64_compat.S   |  2 +-
 arch/x86/include/asm/thread_info.h |  5 +++-
 arch/x86/kernel/traps.c            |  2 +-
 5 files changed, 24 insertions(+), 41 deletions(-)

-- 
2.5.0

[PATCH v2 1/2] x86/entry/32: Change INT80 to be an interrupt gate

[Andy Lutomirski]

I want all of the syscall entries to run with interrupts off so that
I can efficiently run context tracking before enabling interrupts.

This will regress int $0x80 performance on 32-bit kernels by a
couple of cycles.  This shouldn't matter much -- int $0x80 is not a
fast path.

This effectively reverts 657c1eea0019 ("x86/entry/32: Fix
entry_INT80_32() to expect interrupts to be on") and fixes the issue
differently.

Signed-off-by: Andy Lutomirski <luto@kernel.org>
---
 arch/x86/entry/common.c          | 15 +++------------
 arch/x86/entry/entry_32.S        |  8 ++++----
 arch/x86/entry/entry_64_compat.S |  2 +-
 arch/x86/kernel/traps.c          |  2 +-
 4 files changed, 9 insertions(+), 18 deletions(-)

diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
index 1a000f59f04a..9ef3258320ec 100644
--- a/arch/x86/entry/common.c
+++ b/arch/x86/entry/common.c
@@ -381,14 +381,7 @@ __visible void do_syscall_64(struct pt_regs *regs)
  * in workloads that use it, and it's usually called from
  * do_fast_syscall_32, so forcibly inline it to improve performance.
  */
-#ifdef CONFIG_X86_32
-/* 32-bit kernels use a trap gate for INT80, and the asm code calls here. */
-__visible
-#else
-/* 64-bit kernels use do_syscall_32_irqs_off() instead. */
-static
-#endif
-__always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
+static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
 {
 	struct thread_info *ti = pt_regs_to_thread_info(regs);
 	unsigned int nr = (unsigned int)regs->orig_ax;
@@ -423,14 +416,12 @@ __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
 	syscall_return_slowpath(regs);
 }
 
-#ifdef CONFIG_X86_64
-/* Handles INT80 on 64-bit kernels */
-__visible void do_syscall_32_irqs_off(struct pt_regs *regs)
+/* Handles int $0x80 */
+__visible void do_int80_syscall_32(struct pt_regs *regs)
 {
 	local_irq_enable();
 	do_syscall_32_irqs_on(regs);
 }
-#endif
 
 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
 __visible long do_fast_syscall_32(struct pt_regs *regs)
diff --git a/arch/x86/entry/entry_32.S b/arch/x86/entry/entry_32.S
index 66350e6a6ca5..ab710eee4308 100644
--- a/arch/x86/entry/entry_32.S
+++ b/arch/x86/entry/entry_32.S
@@ -348,13 +348,13 @@ ENTRY(entry_INT80_32)
 	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
 
 	/*
-	 * User mode is traced as though IRQs are on.  Unlike the 64-bit
-	 * case, INT80 is a trap gate on 32-bit kernels, so interrupts
-	 * are already on (unless user code is messing around with iopl).
+	 * User mode is traced as though IRQs are on, and the interrupt gate
+	 * turned them off.
 	 */
+	TRACE_IRQS_OFF
 
 	movl	%esp, %eax
-	call	do_syscall_32_irqs_on
+	call	do_int80_syscall_32
 .Lsyscall_32_done:
 
 restore_all:
diff --git a/arch/x86/entry/entry_64_compat.S b/arch/x86/entry/entry_64_compat.S
index 3c990eeee40b..89bcb4979e7a 100644
--- a/arch/x86/entry/entry_64_compat.S
+++ b/arch/x86/entry/entry_64_compat.S
@@ -300,7 +300,7 @@ ENTRY(entry_INT80_compat)
 	TRACE_IRQS_OFF
 
 	movq	%rsp, %rdi
-	call	do_syscall_32_irqs_off
+	call	do_int80_syscall_32
 .Lsyscall_32_done:
 
 	/* Go back to user mode. */
diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index 410e8e2700c5..dd2c2e66c2e1 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -859,7 +859,7 @@ void __init trap_init(void)
 #endif
 
 #ifdef CONFIG_X86_32
-	set_system_trap_gate(IA32_SYSCALL_VECTOR, entry_INT80_32);
+	set_system_intr_gate(IA32_SYSCALL_VECTOR, entry_INT80_32);
 	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 #endif
 
-- 
2.5.0

[PATCH v2 2/2] x86/entry: Do enter_from_user_mode with IRQs off

[Andy Lutomirski]

Now that slow-path syscalls always enter C before enabling
interrupts, it's straightforward to do enter_from_user_mode before
enabling interrupts rather than doing it as part of entry tracing.

With this change, we should finally be able to retire
exception_enter.

This will also enable optimizations based on knowing that we never
change context tracking state with interrupts on.

Signed-off-by: Andy Lutomirski <luto@kernel.org>
---
 arch/x86/entry/common.c            | 33 +++++++++++----------------------
 arch/x86/include/asm/thread_info.h |  5 ++++-
 2 files changed, 15 insertions(+), 23 deletions(-)

diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
index 9ef3258320ec..adea12fb7215 100644
--- a/arch/x86/entry/common.c
+++ b/arch/x86/entry/common.c
@@ -45,6 +45,8 @@ __visible void enter_from_user_mode(void)
 	CT_WARN_ON(ct_state() != CONTEXT_USER);
 	user_exit();
 }
+#else
+static inline void enter_from_user_mode(void) {}
 #endif
 
 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
@@ -85,17 +87,6 @@ unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch)
 
 	work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
 
-#ifdef CONFIG_CONTEXT_TRACKING
-	/*
-	 * If TIF_NOHZ is set, we are required to call user_exit() before
-	 * doing anything that could touch RCU.
-	 */
-	if (work & _TIF_NOHZ) {
-		enter_from_user_mode();
-		work &= ~_TIF_NOHZ;
-	}
-#endif
-
 #ifdef CONFIG_SECCOMP
 	/*
 	 * Do seccomp first -- it should minimize exposure of other
@@ -354,6 +345,7 @@ __visible void do_syscall_64(struct pt_regs *regs)
 	struct thread_info *ti = pt_regs_to_thread_info(regs);
 	unsigned long nr = regs->orig_ax;
 
+	enter_from_user_mode();
 	local_irq_enable();
 
 	if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
@@ -376,9 +368,9 @@ __visible void do_syscall_64(struct pt_regs *regs)
 
 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
 /*
- * Does a 32-bit syscall.  Called with IRQs on and does all entry and
- * exit work and returns with IRQs off.  This function is extremely hot
- * in workloads that use it, and it's usually called from
+ * Does a 32-bit syscall.  Called with IRQs on in CONTEXT_KERNEL.  Does
+ * all entry and exit work and returns with IRQs off.  This function is
+ * extremely hot in workloads that use it, and it's usually called from
  * do_fast_syscall_32, so forcibly inline it to improve performance.
  */
 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
@@ -419,6 +411,7 @@ static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
 /* Handles int $0x80 */
 __visible void do_int80_syscall_32(struct pt_regs *regs)
 {
+	enter_from_user_mode();
 	local_irq_enable();
 	do_syscall_32_irqs_on(regs);
 }
@@ -441,12 +434,11 @@ __visible long do_fast_syscall_32(struct pt_regs *regs)
 	 */
 	regs->ip = landing_pad;
 
-	/*
-	 * Fetch EBP from where the vDSO stashed it.
-	 *
-	 * WARNING: We are in CONTEXT_USER and RCU isn't paying attention!
-	 */
+	enter_from_user_mode();
+
 	local_irq_enable();
+
+	/* Fetch EBP from where the vDSO stashed it. */
 	if (
 #ifdef CONFIG_X86_64
 		/*
@@ -464,9 +456,6 @@ __visible long do_fast_syscall_32(struct pt_regs *regs)
 		/* User code screwed up. */
 		local_irq_disable();
 		regs->ax = -EFAULT;
-#ifdef CONFIG_CONTEXT_TRACKING
-		enter_from_user_mode();
-#endif
 		prepare_exit_to_usermode(regs);
 		return 0;	/* Keep it simple: use IRET. */
 	}
diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
index c0778fcab06d..fbdb24e61835 100644
--- a/arch/x86/include/asm/thread_info.h
+++ b/arch/x86/include/asm/thread_info.h
@@ -134,7 +134,10 @@ struct thread_info {
 #define _TIF_ADDR32		(1 << TIF_ADDR32)
 #define _TIF_X32		(1 << TIF_X32)
 
-/* work to do in syscall_trace_enter() */
+/*
+ * work to do in syscall_trace_enter().  Also includes TIF_NOHZ for
+ * enter_from_user_mode()
+ */
 #define _TIF_WORK_SYSCALL_ENTRY	\
 	(_TIF_SYSCALL_TRACE | _TIF_SYSCALL_EMU | _TIF_SYSCALL_AUDIT |	\
 	 _TIF_SECCOMP | _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT |	\
-- 
2.5.0

[tip:x86/asm] x86/entry/32: Change INT80 to be an interrupt gate

[tip-bot for Andy Lutomirski]

Commit-ID:  a798f091113ef4999277dbe0483d37d04fa35b2e
Gitweb:     http://git.kernel.org/tip/a798f091113ef4999277dbe0483d37d04fa35b2e
Author:     Andy Lutomirski <luto@kernel.org>
AuthorDate: Wed, 9 Mar 2016 13:24:32 -0800
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Thu, 10 Mar 2016 10:53:26 +0100

x86/entry/32: Change INT80 to be an interrupt gate

We want all of the syscall entries to run with interrupts off so that
we can efficiently run context tracking before enabling interrupts.

This will regress int $0x80 performance on 32-bit kernels by a
couple of cycles.  This shouldn't matter much -- int $0x80 is not a
fast path.

This effectively reverts:

  657c1eea0019 ("x86/entry/32: Fix entry_INT80_32() to expect interrupts to be on")

... and fixes the same issue differently.

Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/59b4f90c9ebfccd8c937305dbbbca680bc74b905.1457558566.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/entry/common.c          | 15 +++------------
 arch/x86/entry/entry_32.S        |  8 ++++----
 arch/x86/entry/entry_64_compat.S |  2 +-
 arch/x86/kernel/traps.c          |  2 +-
 4 files changed, 9 insertions(+), 18 deletions(-)

diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
index 209ba33..d69d1b6 100644
--- a/arch/x86/entry/common.c
+++ b/arch/x86/entry/common.c
@@ -371,14 +371,7 @@ __visible void do_syscall_64(struct pt_regs *regs)
  * in workloads that use it, and it's usually called from
  * do_fast_syscall_32, so forcibly inline it to improve performance.
  */
-#ifdef CONFIG_X86_32
-/* 32-bit kernels use a trap gate for INT80, and the asm code calls here. */
-__visible
-#else
-/* 64-bit kernels use do_syscall_32_irqs_off() instead. */
-static
-#endif
-__always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
+static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
 {
 	struct thread_info *ti = pt_regs_to_thread_info(regs);
 	unsigned int nr = (unsigned int)regs->orig_ax;
@@ -413,14 +406,12 @@ __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
 	syscall_return_slowpath(regs);
 }
 
-#ifdef CONFIG_X86_64
-/* Handles INT80 on 64-bit kernels */
-__visible void do_syscall_32_irqs_off(struct pt_regs *regs)
+/* Handles int $0x80 */
+__visible void do_int80_syscall_32(struct pt_regs *regs)
 {
 	local_irq_enable();
 	do_syscall_32_irqs_on(regs);
 }
-#endif
 
 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
 __visible long do_fast_syscall_32(struct pt_regs *regs)
diff --git a/arch/x86/entry/entry_32.S b/arch/x86/entry/entry_32.S
index 286efa3..10868aa 100644
--- a/arch/x86/entry/entry_32.S
+++ b/arch/x86/entry/entry_32.S
@@ -463,13 +463,13 @@ ENTRY(entry_INT80_32)
 	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
 
 	/*
-	 * User mode is traced as though IRQs are on.  Unlike the 64-bit
-	 * case, INT80 is a trap gate on 32-bit kernels, so interrupts
-	 * are already on (unless user code is messing around with iopl).
+	 * User mode is traced as though IRQs are on, and the interrupt gate
+	 * turned them off.
 	 */
+	TRACE_IRQS_OFF
 
 	movl	%esp, %eax
-	call	do_syscall_32_irqs_on
+	call	do_int80_syscall_32
 .Lsyscall_32_done:
 
 restore_all:
diff --git a/arch/x86/entry/entry_64_compat.S b/arch/x86/entry/entry_64_compat.S
index efe0fe3..847f2f0 100644
--- a/arch/x86/entry/entry_64_compat.S
+++ b/arch/x86/entry/entry_64_compat.S
@@ -336,7 +336,7 @@ ENTRY(entry_INT80_compat)
 	TRACE_IRQS_OFF
 
 	movq	%rsp, %rdi
-	call	do_syscall_32_irqs_off
+	call	do_int80_syscall_32
 .Lsyscall_32_done:
 
 	/* Go back to user mode. */
diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index 49e2e77..5fae2d8 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -912,7 +912,7 @@ void __init trap_init(void)
 #endif
 
 #ifdef CONFIG_X86_32
-	set_system_trap_gate(IA32_SYSCALL_VECTOR, entry_INT80_32);
+	set_system_intr_gate(IA32_SYSCALL_VECTOR, entry_INT80_32);
 	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
 #endif
 

[tip:x86/asm] x86/entry: Call enter_from_user_mode() with IRQs off

[tip-bot for Andy Lutomirski]

Commit-ID:  9999c8c01f34c918a57d6e5ba2f5d8b79aa04801
Gitweb:     http://git.kernel.org/tip/9999c8c01f34c918a57d6e5ba2f5d8b79aa04801
Author:     Andy Lutomirski <luto@kernel.org>
AuthorDate: Wed, 9 Mar 2016 13:24:33 -0800
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Thu, 10 Mar 2016 10:53:26 +0100

x86/entry: Call enter_from_user_mode() with IRQs off

Now that slow-path syscalls always enter C before enabling
interrupts, it's straightforward to call enter_from_user_mode() before
enabling interrupts rather than doing it as part of entry tracing.

With this change, we should finally be able to retire exception_enter().

This will also enable optimizations based on knowing that we never
change context tracking state with interrupts on.

Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/bc376ecf87921a495e874ff98139b1ca2f5c5dd7.1457558566.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/entry/common.c            | 33 +++++++++++----------------------
 arch/x86/include/asm/thread_info.h |  5 ++++-
 2 files changed, 15 insertions(+), 23 deletions(-)

diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
index d69d1b6..e79d93d 100644
--- a/arch/x86/entry/common.c
+++ b/arch/x86/entry/common.c
@@ -45,6 +45,8 @@ __visible void enter_from_user_mode(void)
 	CT_WARN_ON(ct_state() != CONTEXT_USER);
 	user_exit();
 }
+#else
+static inline void enter_from_user_mode(void) {}
 #endif
 
 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
@@ -85,17 +87,6 @@ unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch)
 
 	work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
 
-#ifdef CONFIG_CONTEXT_TRACKING
-	/*
-	 * If TIF_NOHZ is set, we are required to call user_exit() before
-	 * doing anything that could touch RCU.
-	 */
-	if (work & _TIF_NOHZ) {
-		enter_from_user_mode();
-		work &= ~_TIF_NOHZ;
-	}
-#endif
-
 #ifdef CONFIG_SECCOMP
 	/*
 	 * Do seccomp first -- it should minimize exposure of other
@@ -344,6 +335,7 @@ __visible void do_syscall_64(struct pt_regs *regs)
 	struct thread_info *ti = pt_regs_to_thread_info(regs);
 	unsigned long nr = regs->orig_ax;
 
+	enter_from_user_mode();
 	local_irq_enable();
 
 	if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
@@ -366,9 +358,9 @@ __visible void do_syscall_64(struct pt_regs *regs)
 
 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
 /*
- * Does a 32-bit syscall.  Called with IRQs on and does all entry and
- * exit work and returns with IRQs off.  This function is extremely hot
- * in workloads that use it, and it's usually called from
+ * Does a 32-bit syscall.  Called with IRQs on in CONTEXT_KERNEL.  Does
+ * all entry and exit work and returns with IRQs off.  This function is
+ * extremely hot in workloads that use it, and it's usually called from
  * do_fast_syscall_32, so forcibly inline it to improve performance.
  */
 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
@@ -409,6 +401,7 @@ static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
 /* Handles int $0x80 */
 __visible void do_int80_syscall_32(struct pt_regs *regs)
 {
+	enter_from_user_mode();
 	local_irq_enable();
 	do_syscall_32_irqs_on(regs);
 }
@@ -431,12 +424,11 @@ __visible long do_fast_syscall_32(struct pt_regs *regs)
 	 */
 	regs->ip = landing_pad;
 
-	/*
-	 * Fetch EBP from where the vDSO stashed it.
-	 *
-	 * WARNING: We are in CONTEXT_USER and RCU isn't paying attention!
-	 */
+	enter_from_user_mode();
+
 	local_irq_enable();
+
+	/* Fetch EBP from where the vDSO stashed it. */
 	if (
 #ifdef CONFIG_X86_64
 		/*
@@ -454,9 +446,6 @@ __visible long do_fast_syscall_32(struct pt_regs *regs)
 		/* User code screwed up. */
 		local_irq_disable();
 		regs->ax = -EFAULT;
-#ifdef CONFIG_CONTEXT_TRACKING
-		enter_from_user_mode();
-#endif
 		prepare_exit_to_usermode(regs);
 		return 0;	/* Keep it simple: use IRET. */
 	}
diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
index f2e2302..8286669 100644
--- a/arch/x86/include/asm/thread_info.h
+++ b/arch/x86/include/asm/thread_info.h
@@ -134,7 +134,10 @@ struct thread_info {
 #define _TIF_ADDR32		(1 << TIF_ADDR32)
 #define _TIF_X32		(1 << TIF_X32)
 
-/* work to do in syscall_trace_enter() */
+/*
+ * work to do in syscall_trace_enter().  Also includes TIF_NOHZ for
+ * enter_from_user_mode()
+ */
 #define _TIF_WORK_SYSCALL_ENTRY	\
 	(_TIF_SYSCALL_TRACE | _TIF_SYSCALL_EMU | _TIF_SYSCALL_AUDIT |	\
 	 _TIF_SECCOMP | _TIF_SYSCALL_TRACEPOINT |	\