Name: Gather Stats on per-cpu Usage
Author: Rusty Russell
Status: Experimental

Gathers stats on per-cpu variable usage.  Hack prints them out
every time /proc/modules opened.

Index: linux-2.6.10-rc2-bk13-Percpu/mm/percpu.c
===================================================================
--- linux-2.6.10-rc2-bk13-Percpu.orig/mm/percpu.c	2004-12-02 14:50:04.124129536 +1100
+++ linux-2.6.10-rc2-bk13-Percpu/mm/percpu.c	2004-12-02 14:50:13.980631120 +1100
@@ -62,6 +62,13 @@
 /* All blocks have to be the same size per cpu, otherwise span would differ. */
 static unsigned long reserved_size, percpu_size;
 
+atomic_t percpu_local_count = ATOMIC_INIT(0), percpu_local_ptr_count = ATOMIC_INIT(0), percpu_count = ATOMIC_INIT(0), percpu_ptr_count = ATOMIC_INIT(0), smp_id_count = ATOMIC_INIT(0);
+EXPORT_SYMBOL(percpu_local_count);
+EXPORT_SYMBOL(percpu_local_ptr_count);
+EXPORT_SYMBOL(percpu_count);
+EXPORT_SYMBOL(percpu_ptr_count);
+EXPORT_SYMBOL(smp_id_count);
+
 #ifdef __GENERIC_PER_CPU
 /* Ideally, an arch will sew together pages local to CPUs to form a
  * continuous allocation. */
@@ -518,3 +518,34 @@
 }
 late_initcall(test_percpu);
 #endif
+
+#include <linux/proc_fs.h>
+static int percpu_stats(char *buffer, char **start, off_t offset, int length)
+{
+	int len = 0;
+
+	len += sprintf(buffer+len, "percpu: local = %u, non-local = %u\n",
+		       atomic_read(&percpu_local_count),
+		       atomic_read(&percpu_count)
+		       - atomic_read(&percpu_local_count));
+	len += sprintf(buffer+len,
+		       "percpu pointers: local = %u, non-local = %u\n",
+		       atomic_read(&percpu_local_ptr_count),
+		       atomic_read(&percpu_ptr_count) - 
+		       atomic_read(&percpu_local_ptr_count));
+	len += sprintf(buffer+len, "smp_processor_id(): %u\n",
+		       atomic_read(&smp_id_count)
+		       - atomic_read(&percpu_local_ptr_count)
+		       - atomic_read(&percpu_local_count));
+	return len;
+}
+
+static int procfile_init(void)
+{
+	struct proc_dir_entry *proc;
+
+	proc = create_proc_entry("percpu_stats", 0400, NULL);
+	proc->get_info = percpu_stats;
+	return 0;
+}
+__initcall(procfile_init);
Index: linux-2.6.10-rc2-bk13-Percpu/include/asm-i386/smp.h
===================================================================
--- linux-2.6.10-rc2-bk13-Percpu.orig/include/asm-i386/smp.h	2004-10-19 14:34:18.000000000 +1000
+++ linux-2.6.10-rc2-bk13-Percpu/include/asm-i386/smp.h	2004-12-02 14:49:53.338769160 +1100
@@ -9,6 +9,7 @@
 #include <linux/kernel.h>
 #include <linux/threads.h>
 #include <linux/cpumask.h>
+#include <asm/atomic.h>
 #endif
 
 #ifdef CONFIG_X86_LOCAL_APIC
@@ -45,12 +46,13 @@
 #define MAX_APICID 256
 extern u8 x86_cpu_to_apicid[];
 
+extern atomic_t smp_id_count;
 /*
  * This function is needed by all SMP systems. It must _always_ be valid
  * from the initial startup. We map APIC_BASE very early in page_setup(),
  * so this is correct in the x86 case.
  */
-#define smp_processor_id() (current_thread_info()->cpu)
+#define smp_processor_id() ({ atomic_inc(&smp_id_count); (current_thread_info()->cpu); })
 
 extern cpumask_t cpu_callout_map;
 #define cpu_possible_map cpu_callout_map
Index: linux-2.6.10-rc2-bk13-Percpu/include/linux/percpu.h
===================================================================
--- linux-2.6.10-rc2-bk13-Percpu.orig/include/linux/percpu.h	2004-12-02 14:21:08.716951528 +1100
+++ linux-2.6.10-rc2-bk13-Percpu/include/linux/percpu.h	2004-12-02 14:49:53.338769160 +1100
@@ -16,8 +16,11 @@
 #define put_cpu_var(var) preempt_enable()
 
 #ifdef CONFIG_SMP
+#include <asm/atomic.h>
+extern atomic_t percpu_local_count, percpu_local_ptr_count, percpu_count, percpu_ptr_count;
+
 /* var is in discarded region: offset to particular copy we want */
-#define per_cpu(var, cpu) (*RELOC_HIDE(&per_cpu__##var, __per_cpu_offset[cpu]))
+#define per_cpu(var, cpu) (*({ atomic_inc(&percpu_count); RELOC_HIDE(&per_cpu__##var, __per_cpu_offset[cpu]); }))
 extern unsigned long __per_cpu_offset[NR_CPUS];
 
 /* 
@@ -26,7 +29,7 @@
  * to use get_cpu_ptr... 
  */ 
 #define per_cpu_ptr(ptr, cpu)						\
-        ((__typeof__(ptr))((void *)ptr + __per_cpu_offset[(cpu)]))
+        ({ atomic_inc(&percpu_ptr_count); ((__typeof__(ptr))((void *)ptr + __per_cpu_offset[(cpu)])); })
 
 extern void *__alloc_percpu(unsigned long size, unsigned long align);
 extern void free_percpu(const void *);
Index: linux-2.6.10-rc2-bk13-Percpu/include/asm-generic/percpu.h
===================================================================
--- linux-2.6.10-rc2-bk13-Percpu.orig/include/asm-generic/percpu.h	2004-12-02 14:21:08.718951224 +1100
+++ linux-2.6.10-rc2-bk13-Percpu/include/asm-generic/percpu.h	2004-12-02 14:49:53.338769160 +1100
@@ -1,6 +1,7 @@
 #ifndef _ASM_GENERIC_PERCPU_H_
 #define _ASM_GENERIC_PERCPU_H_
 #include <linux/compiler.h>
+#include <asm/atomic.h>
 
 #define __GENERIC_PER_CPU
 #ifdef CONFIG_SMP
@@ -9,9 +10,9 @@
 #define DEFINE_PER_CPU(type, name) \
     __attribute__((__section__(".data.percpu"))) __typeof__(type) per_cpu__##name
 
-#define __get_cpu_var(var) per_cpu(var, smp_processor_id())
+#define __get_cpu_var(var) (*({ atomic_inc(&percpu_local_count); &per_cpu(var, smp_processor_id()); }))
 #define __get_cpu_ptr(ptr) \
-	((__typeof__(ptr))((void *)ptr + __per_cpu_offset[smp_processor_id()]))
+	({ atomic_inc(&percpu_local_ptr_count); ((__typeof__(ptr))((void *)ptr + __per_cpu_offset[smp_processor_id()])); })
 
 /* A macro to avoid #include hell... */
 #define percpu_modcopy(pcpudst, src, size)			\