From: Andi Kleen <andi@firstfloor.org>
To: tglx@linutronix.de
Cc: x86@kernel.org, dwmw@amazon.co.uk, linux-kernel@vger.kernel.org,
        pjt@google.com, torvalds@linux-foundation.org,
        gregkh@linux-foundation.org, peterz@infradead.org,
        luto@amacapital.net, thomas.lendacky@amd.com,
        arjan.van.de.ven@intel.com, Andi Kleen <ak@linux.intel.com>
Subject: [PATCH 4/4] x86/retpoline: Fill return buffer on interrupt return to kernel
Date: Fri, 12 Jan 2018 10:45:50 -0800
Message-Id: <20180112184550.6573-5-andi@firstfloor.org>
In-Reply-To: <20180112184550.6573-1-andi@firstfloor.org>
References: <20180112184550.6573-1-andi@firstfloor.org>
Sender: linux-kernel-owner@vger.kernel.org

From: Andi Kleen <ak@linux.intel.com>

Interrupts can have rather deep call chains on top of the original
call chain. Fill the return buffer on Skylake when returning from
an interrupt to the kernel, to avoid return buffer underflows
later.

This only needs to be done when returning to the kernel,
so interrupts interrupting user space are not impacted.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 arch/x86/entry/entry_32.S | 16 +++++++++++++---
 arch/x86/entry/entry_64.S | 22 ++++++++++++++++++++++
 2 files changed, 35 insertions(+), 3 deletions(-)

diff --git a/arch/x86/entry/entry_32.S b/arch/x86/entry/entry_32.S
index bbecb7c2f6cb..a58b0ae7121c 100644
--- a/arch/x86/entry/entry_32.S
+++ b/arch/x86/entry/entry_32.S
@@ -65,7 +65,6 @@
 # define preempt_stop(clobbers)	DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
 #else
 # define preempt_stop(clobbers)
-# define resume_kernel		restore_all
 #endif
 
 .macro TRACE_IRQS_IRET
@@ -349,8 +348,17 @@ ENTRY(resume_userspace)
 	jmp	restore_all
 END(ret_from_exception)
 
-#ifdef CONFIG_PREEMPT
 ENTRY(resume_kernel)
+	/*
+	 * Interrupts/faults could cause the return buffer of the CPU
+	 * to overflow, which would lead to a underflow later,
+	 * which may lead to a uncontrolled indirect branch.
+	 * Fill the return buffer when returning to the kernel.
+	 */
+	FILL_RETURN_BUFFER %eax, RSB_FILL_LOOPS, X86_FEATURE_RETURN_UNDERFLOW
+5:
+
+#ifdef CONFIG_PREEMPT
 	DISABLE_INTERRUPTS(CLBR_ANY)
 .Lneed_resched:
 	cmpl	$0, PER_CPU_VAR(__preempt_count)
@@ -359,8 +367,10 @@ ENTRY(resume_kernel)
 	jz	restore_all
 	call	preempt_schedule_irq
 	jmp	.Lneed_resched
-END(resume_kernel)
+#else
+	jmp	restore_all
 #endif
+END(resume_kernel)
 
 GLOBAL(__begin_SYSENTER_singlestep_region)
 /*
diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
index 3caac129cd07..36ee97fac6af 100644
--- a/arch/x86/entry/entry_64.S
+++ b/arch/x86/entry/entry_64.S
@@ -792,6 +792,15 @@ retint_kernel:
 	TRACE_IRQS_IRETQ
 
 GLOBAL(restore_regs_and_return_to_kernel)
+	/*
+	 * Interrupts/faults could cause the return buffer of the CPU
+	 * to overflow, which would lead to a underflow later,
+	 * which may lead to a uncontrolled indirect branch.
+	 * Fill the return buffer when returning to the kernel.
+	 */
+	FILL_RETURN_BUFFER %rax, RSB_FILL_LOOPS, X86_FEATURE_RETURN_UNDERFLOW
+4:
+
 #ifdef CONFIG_DEBUG_ENTRY
 	/* Assert that pt_regs indicates kernel mode. */
 	testb	$3, CS(%rsp)
@@ -1660,6 +1669,10 @@ nested_nmi:
 nested_nmi_out:
 	popq	%rdx
 
+	/*
+	 * No need to clear return buffer here because the outter NMI will do it,
+	 * and we assume two NMIs will not overflow the return buffer.
+	 */
 	/* We are returning to kernel mode, so this cannot result in a fault. */
 	iretq
 
@@ -1757,6 +1770,15 @@ end_repeat_nmi:
 nmi_swapgs:
 	SWAPGS_UNSAFE_STACK
 nmi_restore:
+	/*
+	 * NMI could cause the return buffer of the CPU
+	 * to overflow, which would lead to a underflow later,
+	 * which may lead to a uncontrolled indirect branch.
+	 * Fill the return buffer when returning to the kernel.
+	 */
+
+	FILL_RETURN_BUFFER %rax, RSB_FILL_LOOPS, X86_FEATURE_RETURN_UNDERFLOW
+5:
 	POP_EXTRA_REGS
 	POP_C_REGS
 
-- 
2.14.3