[PATCH 3.16 17/76] x86/cpu/AMD: Make LFENCE a serializing instruction

[Ben Hutchings]

3.16.56-rc1 review patch.  If anyone has any objections, please let me know.

------------------

From: Tom Lendacky <thomas.lendacky@xxxxxxx>

commit e4d0e84e490790798691aaa0f2e598637f1867ec upstream.

To aid in speculation control, make LFENCE a serializing instruction
since it has less overhead than MFENCE.  This is done by setting bit 1
of MSR 0xc0011029 (DE_CFG).  Some families that support LFENCE do not
have this MSR.  For these families, the LFENCE instruction is already
serializing.

Signed-off-by: Tom Lendacky <thomas.lendacky@xxxxxxx>
Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Reviewed-by: Reviewed-by: Borislav Petkov <bp@xxxxxxx>
Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Cc: Tim Chen <tim.c.chen@xxxxxxxxxxxxxxx>
Cc: Dave Hansen <dave.hansen@xxxxxxxxx>
Cc: Borislav Petkov <bp@xxxxxxxxx>
Cc: Dan Williams <dan.j.williams@xxxxxxxxx>
Cc: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxxx>
Cc: David Woodhouse <dwmw@xxxxxxxxxxxx>
Cc: Paul Turner <pjt@xxxxxxxxxx>
Link: https://lkml.kernel.org/r/20180108220921.12580.71694.stgit@xxxxxxxxxxxxxxxxxxxxxxxxx
[bwh: Backported to 3.16: adjust filename, context]
Signed-off-by: Ben Hutchings <ben@xxxxxxxxxxxxxxx>
---
 arch/x86/include/uapi/asm/msr-index.h |  2 ++
 arch/x86/kernel/cpu/amd.c             | 10 ++++++++++
 2 files changed, 12 insertions(+)

--- a/arch/x86/include/uapi/asm/msr-index.h
+++ b/arch/x86/include/uapi/asm/msr-index.h
@@ -223,6 +223,8 @@
 #define FAM10H_MMIO_CONF_BASE_MASK	0xfffffffULL
 #define FAM10H_MMIO_CONF_BASE_SHIFT	20
 #define MSR_FAM10H_NODE_ID		0xc001100c
+#define MSR_F10H_DECFG			0xc0011029
+#define MSR_F10H_DECFG_LFENCE_SERIALIZE_BIT	1
 
 /* K8 MSRs */
 #define MSR_K8_TOP_MEM1			0xc001001a
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -673,6 +673,16 @@ static void init_amd(struct cpuinfo_x86
 		set_cpu_cap(c, X86_FEATURE_K8);
 
 	if (cpu_has_xmm2) {
+		/*
+		 * A serializing LFENCE has less overhead than MFENCE, so
+		 * use it for execution serialization.  On families which
+		 * don't have that MSR, LFENCE is already serializing.
+		 * msr_set_bit() uses the safe accessors, too, even if the MSR
+		 * is not present.
+		 */
+		msr_set_bit(MSR_F10H_DECFG,
+			    MSR_F10H_DECFG_LFENCE_SERIALIZE_BIT);
+
 		/* MFENCE stops RDTSC speculation */
 		set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
 	}