[PATCH 2/3] x86_64: Define 128-bit memory-mapped I/O operations

@@ -372,6 +372,10 @@ config X86_USE_3DNOW
 	def_bool y
 	depends on (MCYRIXIII || MK7 || MGEODE_LX) && !UML
 
+config X86_USE_SSE
+	def_bool y
+	depends on X86_64
+
 config X86_OOSTORE
 	def_bool y
 	depends on (MWINCHIP3D || MWINCHIPC6) && MTRR

@@ -95,6 +95,20 @@ build_mmio_write(writeq, "q", unsigned long, "r", :"memory")
 
 #endif
 
+#ifdef CONFIG_X86_USE_SSE
+
+u128 reado(const volatile void __iomem *addr);
+void writeo(u128 val, volatile void __iomem *addr);
+
+#define __raw_reado(addr) reado(addr)
+#define __raw_writeo(val, addr)	writeo(val, addr)
+
+/* Let people know that we have them */
+#define reado			reado
+#define writeo			writeo
+
+#endif
+
 /**
  *	virt_to_phys	-	map virtual addresses to physical
  *	@address: address to remap

@@ -25,6 +25,7 @@ lib-$(CONFIG_RWSEM_XCHGADD_ALGORITHM) += rwsem.o
 lib-$(CONFIG_INSTRUCTION_DECODER) += insn.o inat.o
 
 obj-y += msr.o msr-reg.o msr-reg-export.o
+obj-$(CONFIG_X86_USE_SSE) += oword_io.o
 
 ifeq ($(CONFIG_X86_32),y)
         obj-y += atomic64_32.o

@@ -0,0 +1,65 @@
+/****************************************************************************
+ * 128-bit MMIO for x86
+ * Copyright 2012 Solarflare Communications Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation, incorporated herein by reference.
+ */
+
+#include <linux/export.h>
+#include <linux/preempt.h>
+#include <linux/types.h>
+#include <asm/io.h>
+
+/*
+ * We copy the data between a memory buffer and the MMIO address via
+ * register xmm0.  We have to save and restore the SSE state, and
+ * disable preemption.  Only the MMIO address is required to be
+ * 128-bit aligned, since the stack generally is not.
+ */
+
+u128 reado(const volatile void __iomem *addr)
+{
+	u128 ret;
+	u64 cr0;
+	u128 xmm0;
+
+	preempt_disable();
+	asm volatile (
+		"movq %%cr0,%0\n\t"
+		"clts\n\t"
+		"movups %%xmm0,%1\n\t"
+		"movaps %3,%%xmm0\n\t"
+		"movups %%xmm0,%2\n\t"
+		"sfence\n\t"
+		"movups %1,%%xmm0\n\t"
+		"movq %0,%%cr0\n\t"
+		: "=r"(cr0), "=m"(xmm0), "=m"(ret)
+		: "m"(*(const volatile u128 __iomem *)addr));
+	preempt_enable();
+
+	return ret;
+}
+EXPORT_SYMBOL(reado);
+
+void writeo(u128 val, volatile void __iomem *addr)
+{
+	u64 cr0;
+	u128 xmm0;
+
+	preempt_disable();
+	asm volatile (
+		"movq %%cr0,%0\n\t"
+		"clts\n\t"
+		"movups %%xmm0,%1\n\t"
+		"movups %3,%%xmm0\n\t"
+		"movaps %%xmm0,%2\n\t"
+		"sfence\n\t"
+		"movups %1,%%xmm0\n\t"
+		"movq %0,%%cr0\n\t"
+		: "=r"(cr0), "=m"(xmm0), "=m"(*(volatile u128 __iomem *)addr)
+		: "m"(val));
+	preempt_enable();
+}
+EXPORT_SYMBOL(writeo);