ia64: Improve layout of cpuinfo_ia64

Stephane made some measurements on the access-pattern of the cpuinfo_ia64
members and based on those results, this patch reorganizes the structure
for better cache-line sharing.

arch/ia64/kernel/smp.c

@@ -91,9 +91,6 @@ handle_IPI (int irq, void *dev_id, struct pt_regs *regs)
 	unsigned long *pending_ipis = &__ia64_per_cpu_var(ipi_operation);
 	unsigned long ops;
 
-	/* Count this now; we may make a call that never returns. */
-	local_cpu_data->ipi_count++;
-
 	mb();	/* Order interrupt and bit testing. */
 	while ((ops = xchg(pending_ipis, 0)) != 0) {
 		mb();	/* Order bit clearing and data access. */
@@ -338,17 +335,6 @@ smp_call_function (void (*func) (void *info), void *info, int nonatomic, int wai
 }
 EXPORT_SYMBOL(smp_call_function);
 
-void
-smp_do_timer (struct pt_regs *regs)
-{
-	int user = user_mode(regs);
-
-	if (--local_cpu_data->prof_counter <= 0) {
-		local_cpu_data->prof_counter = local_cpu_data->prof_multiplier;
-		update_process_times(user);
-	}
-}
-
 /*
  * this function calls the 'stop' function on all other CPUs in the system.
  */

arch/ia64/kernel/smpboot.c

@@ -265,8 +265,6 @@ ia64_sync_itc (unsigned int master)
 static inline void __init
 smp_setup_percpu_timer (void)
 {
-	local_cpu_data->prof_counter = 1;
-	local_cpu_data->prof_multiplier = 1;
 }
 
 static void __init

arch/ia64/kernel/time.c

@@ -255,9 +255,13 @@ timer_interrupt (int irq, void *dev_id, struct pt_regs *regs)
 	ia64_do_profile(regs);
 
 	while (1) {
-
 #ifdef CONFIG_SMP
-		smp_do_timer(regs);
+		/*
+		 * For UP, this is done in do_timer().  Weird, but
+		 * fixing that would require updates to all
+		 * platforms.
+		 */
+		update_process_times(user_mode(regs));
 #endif
 		new_itm += local_cpu_data->itm_delta;
 
@@ -280,17 +284,19 @@ timer_interrupt (int irq, void *dev_id, struct pt_regs *regs)
 	}
 
 	do {
-	    /*
-	     * If we're too close to the next clock tick for comfort, we increase the
-	     * safety margin by intentionally dropping the next tick(s).  We do NOT update
-	     * itm.next because that would force us to call do_timer() which in turn would
-	     * let our clock run too fast (with the potentially devastating effect of
-	     * losing monotony of time).
-	     */
-	    while (!time_after(new_itm, ia64_get_itc() + local_cpu_data->itm_delta/2))
-	      new_itm += local_cpu_data->itm_delta;
-	    ia64_set_itm(new_itm);
-	    /* double check, in case we got hit by a (slow) PMI: */
+		/*
+		 * If we're too close to the next clock tick for
+		 * comfort, we increase the safety margin by
+		 * intentionally dropping the next tick(s).  We do NOT
+		 * update itm.next because that would force us to call
+		 * do_timer() which in turn would let our clock run
+		 * too fast (with the potentially devastating effect
+		 * of losing monotony of time).
+		 */
+		while (!time_after(new_itm, ia64_get_itc() + local_cpu_data->itm_delta/2))
+			new_itm += local_cpu_data->itm_delta;
+		ia64_set_itm(new_itm);
+		/* double check, in case we got hit by a (slow) PMI: */
 	} while (time_after_eq(ia64_get_itc(), new_itm));
 	return IRQ_HANDLED;
 }

include/asm-ia64/processor.h

@@ -148,35 +148,35 @@ struct cpuinfo_ia64 {
 	__u32 softirq_pending;
 	__u64 itm_delta;	/* # of clock cycles between clock ticks */
 	__u64 itm_next;		/* interval timer mask value to use for next clock tick */
+	__u64 nsec_per_cyc;	/* (1000000000<<IA64_NSEC_PER_CYC_SHIFT)/itc_freq */
+	__u64 unimpl_va_mask;	/* mask of unimplemented virtual address bits (from PAL) */
+	__u64 unimpl_pa_mask;	/* mask of unimplemented physical address bits (from PAL) */
 	__u64 *pgd_quick;
 	__u64 *pmd_quick;
 	__u64 pgtable_cache_sz;
-	/* CPUID-derived information: */
-	__u64 ppn;
-	__u64 features;
-	__u8 number;
-	__u8 revision;
-	__u8 model;
-	__u8 family;
-	__u8 archrev;
-	char vendor[16];
 	__u64 itc_freq;		/* frequency of ITC counter */
 	__u64 proc_freq;	/* frequency of processor */
 	__u64 cyc_per_usec;	/* itc_freq/1000000 */
-	__u64 nsec_per_cyc;	/* (1000000000<<IA64_NSEC_PER_CYC_SHIFT)/itc_freq */
-	__u64 unimpl_va_mask;	/* mask of unimplemented virtual address bits (from PAL) */
-	__u64 unimpl_pa_mask;	/* mask of unimplemented physical address bits (from PAL) */
 	__u64 ptce_base;
 	__u32 ptce_count[2];
 	__u32 ptce_stride[2];
 	struct task_struct *ksoftirqd;	/* kernel softirq daemon for this CPU */
+
 #ifdef CONFIG_SMP
-	int cpu;
 	__u64 loops_per_jiffy;
-	__u64 ipi_count;
-	__u64 prof_counter;
-	__u64 prof_multiplier;
+	int cpu;
 #endif
+
+	/* CPUID-derived information: */
+	__u64 ppn;
+	__u64 features;
+	__u8 number;
+	__u8 revision;
+	__u8 model;
+	__u8 family;
+	__u8 archrev;
+	char vendor[16];
+
 #ifdef CONFIG_NUMA
 	struct ia64_node_data *node_data;
 #endif