diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index 9e68cd5..c897b8c 100644
--- a/Documentation/kernel-parameters.txt
+++ b/Documentation/kernel-parameters.txt
@@ -915,6 +915,13 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
 			kexec loader will pass this option to capture kernel.
 			See Documentation/kdump/kdump.txt for details.
 
+	emulated_files=1 [X86_32]
+			if CONFIG_CPU_PROC_EMULATED_FILES is set,
+			start runnung /proc/emulated_files at boot.
+			Normally, it is not run until expcilitly enabled
+			by 'echo 1 > /proc/emulated_files', since the
+			data collection is a very heavy task on CPU.
+			
 	enable_mtrr_cleanup [X86]
 			The kernel tries to adjust MTRR layout from continuous
 			to discrete, to make X server driver able to add WB
diff --git a/arch/x86/Kconfig.cpu b/arch/x86/Kconfig.cpu
index a9f8463..7fa5c5d 100644
--- a/arch/x86/Kconfig.cpu
+++ b/arch/x86/Kconfig.cpu
@@ -400,12 +400,26 @@ config CPU_EMU_SSE2
 config CPU_PROC_EMULATED_OPS
 	bool "/proc/emulated_ops support"
 	depends on X86_32 && CPU_EMU486
+	default y
 	---help---
 	/proc/emulated_ops will show counts of emulated 
 	CMOV, NOPL, FCOMI/FCOMIP, FUCOMI/FUCOMIP and FCMOVcc instructions.
 	
 	If you are not sure, say N.
 
+config CPU_PROC_EMULATED_FILES 
+	bool "/proc/emulated_files support"
+	depends on CPU_PROC_EMULATED_OPS
+	default n
+	---help---
+	/proc/emulated_files will show statistics of what executable uses
+	emulated CMOV, SSE and SSE2 instructions.
+	Since this is heavy on CPU, by default data collecting is off;
+	explicitly enable it by "echo 1 > /proc/emulated_files" or
+	passing "emulated_files=1" kernel command line option.
+
+	If you are not sure, say N.
+
 config CPU_EMU486_DEBUG
 	bool "Emulation debug"
 	depends on X86_32 && CPU_EMU486
diff --git a/arch/x86/configs/i386_defconfig b/arch/x86/configs/i386_defconfig
index c7c8534..ff1443b 100644
--- a/arch/x86/configs/i386_defconfig
+++ b/arch/x86/configs/i386_defconfig
@@ -45,6 +45,7 @@ CONFIG_X86_GENERIC=y
 # CONFIG_GEODE_NOPL is not set
 # CONFIG_CPU_EMU_FUCOMI is not set
 # CONFIG_CPU_PROC_EMULATED_OPS is not set
+# CONFIG_CPU_PROC_EMULATED_FILES is not set
 CONFIG_HPET_TIMER=y
 CONFIG_SCHED_SMT=y
 CONFIG_PREEMPT_VOLUNTARY=y
diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index 3cc89fc..8160d8d 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -930,9 +930,198 @@ static int __init proc_emulated_ops_init(void)
 module_init(proc_emulated_ops_init);
 #endif /* CONFIG_CPU_PROC_EMULATED_OPS */
 
-#ifdef CONFIG_CPU_EMU_SSE2
+#ifdef CONFIG_CPU_PROC_EMULATED_FILES
+/*
+ * /proc/emulated_files support.
+ * Dumps list of "cmov-ops sse-ops sse2-ops /path/of/lib-or-exec" since boot.
+ * Is a heavy process on slow CPUs.
+ */
+#define EMU_OPS_FILES_MAX 1024
+struct {
+	char buf[256];
+	unsigned long	cmov;
+	unsigned long	sse;
+	unsigned long	sse2;
+} emulated_files_list[EMU_OPS_FILES_MAX] = {
+	{ {'\0'}, 0,0,0 },
+};
+enum emulated_op_type {
+	OP_NONE = 0,
+	OP_CMOV,
+	OP_SSE,
+	OP_SSE2,
+};
+static int emulated_files_running = 0;
+/* cmdline: emulated_files=1 to start /proc/emulated_files on boot */
+core_param(emulated_files, emulated_files_running, int, 0644);
+
+static void emulated_files_clear(void)
+{
+	int	i;
+	for (i=0; i<EMU_OPS_FILES_MAX; i++) {
+		 emulated_files_list[i].buf[0]= '\0';
+		 emulated_files_list[i].cmov = 0;
+		 emulated_files_list[i].sse = 0;
+		 emulated_files_list[i].sse2 = 0;
+	}
+}
+/*
+ * Get executable's filename from IP, and add+count into emulated_files_list;
+ * a clone of mm/memory.c:print_vma_addr() with local changes
+ */
+static void __emulated_files_byip(unsigned long ip, enum emulated_op_type op_type);
+static inline void emulated_files_byip(unsigned long ip, enum emulated_op_type op_type)
+{
+	/* do nothing if stopped */
+	if (!emulated_files_running)
+		return;
+	__emulated_files_byip(ip, op_type);
+}
+
+static void __emulated_files_byip(unsigned long ip, enum emulated_op_type op_type)
+{
+	struct mm_struct *mm = current->mm;
+	struct vm_area_struct *vma;
+
+	/*
+	 * Do not print if we are in atomic
+	 * contexts (in exception stacks, etc.):
+	 */
+	if (preempt_count())
+		return;
+
+	down_read(&mm->mmap_sem);
+	vma = find_vma(mm, ip);
+	if (vma && vma->vm_file) {
+		struct file *f = vma->vm_file;
+		char *buf = (char *)__get_free_page(GFP_KERNEL);
+		if (buf) {
+			char *p;
+			int i;
+			char *ph;
+			int h;
+
+			p = d_path(&f->f_path, buf, PAGE_SIZE);
+			if (IS_ERR(p))
+				p = "?";
+
+			/* hash the path */
+			for (h=0,ph=p; *ph!='\0'; ph++)
+				h += (int)*ph;
+			h &= 127;
+			i = h * (EMU_OPS_FILES_MAX / 128);
+			/* linear search in emulated_files_list[] for matching or vacant slot */
+			for (   ;i<EMU_OPS_FILES_MAX; i++) {
+				if(!emulated_files_list[i].buf[0]) {
+					/* reached null entry; use this slot */
+					strncpy(emulated_files_list[i].buf, p, sizeof(emulated_files_list[i].buf)-1);
+					emulated_files_list[i].buf[sizeof(emulated_files_list[i].buf)-1]='\0';
+					emulated_files_list[i].cmov=0;
+					emulated_files_list[i].sse=0;
+					emulated_files_list[i].sse2=0;
+					break;
+				}
+				if(!strncmp(emulated_files_list[i].buf, p, sizeof(emulated_files_list[i].buf)-1)) {
+					/* got match */
+					break;
+				}
+			}
+			if (i<EMU_OPS_FILES_MAX) {
+				switch(op_type) {
+				  case OP_CMOV:
+					emulated_files_list[i].cmov++; break;
+				  case OP_SSE:
+					emulated_files_list[i].sse++; break;
+				  case OP_SSE2:
+					emulated_files_list[i].sse2++; break;
+				  case OP_NONE:
+					;/*nop*/
+				  default:
+					;/*nop*/
+				}
+			}
+			free_page((unsigned long)buf);
+		}
+	}
+	up_read(&mm->mmap_sem);
+}
+
+static int emulated_files_proc_show(struct seq_file *m, void *v)
+{
+	int i;
+	seq_printf(m, "%10s %10s %10s %s%s\n", "cmov","sse","sse2","filename",
+	    (!emulated_files_running)?" (stopped; write \"1\" to start)":""
+	);
+	for (i=0;i<EMU_OPS_FILES_MAX; i++) {	/*todo:partial read*/
+		if (emulated_files_list[i].buf[0]) {
+			seq_printf(m, "%10lu %10lu %10lu %s\n",
+			    emulated_files_list[i].cmov,
+			    emulated_files_list[i].sse,
+			    emulated_files_list[i].sse2,
+			    emulated_files_list[i].buf
+			);
+		}
+	}
+	return 0;
+}
+static int emulated_files_proc_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, emulated_files_proc_show, NULL);
+}
+/*
+ * start by "echo 1 > /proc/emulated_files"
+ * stop by "echo -1 > /proc/emulated_files"
+ * clear by "echo 0 > /proc/emulated_files"
+ */
+static ssize_t emulated_files_proc_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
+{
+	if (count) {
+		char c;
+		if (get_user(c, buf))
+			return -EFAULT;
+
+		switch(c) {
+		  case '-':
+			emulated_files_running = 0; /* false */
+			break;
+		  case '0':
+			/*printk(KERN_NOTICE "clear /proc/emulated_files\n");*/
+			emulated_files_clear();
+			break;
+		  case '1':
+			emulated_files_running = 1; /* true */
+			break;
+		  default:
+			return count;
+		}
+	}
+	return count;
+}
+
+static const struct file_operations emulated_files_proc_fops = {
+	.open		= emulated_files_proc_open,
+	.read		= seq_read,
+	.write		= emulated_files_proc_write,
+	.llseek		= seq_lseek,
+	.release	= single_release,
+};
+static int __init proc_emulated_files_init(void)
+{
+	emulated_files_clear();
+	proc_create("emulated_files", 0644, NULL, &emulated_files_proc_fops);
+	return 0;
+}
+module_init(proc_emulated_files_init);
+#endif /* CONFIG_CPU_PROC_EMULATED_FILES */
+
+#ifdef CONFIG_CPU_PROC_EMULATED_OPS
+# ifdef CONFIG_CPU_PROC_EMULATED_FILES
+#define	EMU_COUNT_SSE(eip)	emulated_files_byip(((unsigned long)(eip)), OP_SSE); emulated_ops_counter.sse++
+#define	EMU_COUNT_SSE2(eip)	emulated_files_byip(((unsigned long)(eip)), OP_SSE2);emulated_ops_counter.sse2++
+# else
 #define	EMU_COUNT_SSE(eip)	emulated_ops_counter.sse++
 #define	EMU_COUNT_SSE2(eip)	emulated_ops_counter.sse2++
+# endif
 #else
 #define	EMU_COUNT_SSE(eip)
 #define	EMU_COUNT_SSE2(eip)
@@ -1411,10 +1600,13 @@ dotraplinkage void do_invalid_op(struct pt_regs *regs, long error_code)
 				/* ncond is now true if the cond matches the opcode */
 
 				modrm = *(eip + 1);
-				eip += 2; /* skips all the opcodes */
 #ifdef CONFIG_CPU_PROC_EMULATED_OPS
 				emulated_ops_counter.cmov++;
 #endif
+#ifdef CONFIG_CPU_PROC_EMULATED_FILES
+				emulated_files_byip((unsigned long)eip, OP_CMOV);
+#endif
+				eip += 2; /* skips all the opcodes */
 
 				if (!ncond) {
 					/* condition is not valid, skip the instruction and do nothing */