[PATCH] ia64: gettimeoffset hooks

Make it possible to plug in alternate time-offset sources, such as external
(e.g., chipset) timers or High-Performance Event Timer (HPET) etc. This
is needed on platforms where the cycle-counters on different CPUs may
drift apart from each other.
This patch contains the ia64-specific portion only.

arch/ia64/kernel/fsys.S

@@ -14,6 +14,7 @@
 #include <asm/offsets.h>
 #include <asm/percpu.h>
 #include <asm/thread_info.h>
+#include <asm/sal.h>
 
 /*
  * See Documentation/ia64/fsys.txt for details on fsyscalls.
@@ -142,23 +143,30 @@ END(fsys_set_tid_address)
  *	   we ought to either skip the ITC-based interpolation or run an ntp-like
  *	   daemon to keep the ITCs from drifting too far apart.
  */
+
 ENTRY(fsys_gettimeofday)
 	add r9=TI_FLAGS+IA64_TASK_SIZE,r16
 	movl r3=THIS_CPU(cpu_info)
 
 	mov.m r31=ar.itc		// put time stamp into r31 (ITC) == now		(35 cyc)
-	movl r19=xtime			// xtime is a timespec struct
-	;;
-
 #ifdef CONFIG_SMP
 	movl r10=__per_cpu_offset
+	movl r2=sal_platform_features
 	;;
+
+	ld8 r2=[r2]
+	movl r19=xtime			// xtime is a timespec struct
+
 	ld8 r10=[r10]			// r10 <- __per_cpu_offset[0]
 	movl r21=cpu_info__per_cpu
 	;;
 	add r10=r21, r10		// r10 <- &cpu_data(time_keeper_id)
+	tbit.nz p8,p0 = r2, IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT_BIT
+(p8)	br.spnt.many fsys_fallback_syscall
 #else
+	;;
 	mov r10=r3
+	movl r19=xtime			// xtime is a timespec struct
 #endif
 	ld4 r9=[r9]
 	movl r17=xtime_lock

arch/ia64/kernel/time.c

@@ -17,6 +17,7 @@
 #include <linux/time.h>
 #include <linux/interrupt.h>
 #include <linux/efi.h>
+#include <linux/timex.h>
 
 #include <asm/delay.h>
 #include <asm/hw_irq.h>
@@ -26,9 +27,14 @@
 
 extern unsigned long wall_jiffies;
 extern unsigned long last_nsec_offset;
+static unsigned long __ia64_gettimeoffset (void);
+
+unsigned long (*gettimeoffset)(void) = &__ia64_gettimeoffset;
 
 u64 jiffies_64 = INITIAL_JIFFIES;
 
+#define TIME_KEEPER_ID	0	/* smp_processor_id() of time-keeper */
+
 #ifdef CONFIG_IA64_DEBUG_IRQ
 
 unsigned long last_cli_ip;
@@ -63,15 +69,14 @@ do_profile (unsigned long ip)
  * Return the number of nano-seconds that elapsed since the last update to jiffy.  The
  * xtime_lock must be at least read-locked when calling this routine.
  */
-static inline unsigned long
-gettimeoffset (void)
+static unsigned long
+__ia64_gettimeoffset (void)
 {
 	unsigned long elapsed_cycles, lost = jiffies - wall_jiffies;
 	unsigned long now, last_tick;
-#	define time_keeper_id	0	/* smp_processor_id() of time-keeper */
 
-	last_tick = (cpu_data(time_keeper_id)->itm_next
-		     - (lost + 1)*cpu_data(time_keeper_id)->itm_delta);
+	last_tick = (cpu_data(TIME_KEEPER_ID)->itm_next
+		     - (lost + 1)*cpu_data(TIME_KEEPER_ID)->itm_delta);
 
 	now = ia64_get_itc();
 	if (unlikely((long) (now - last_tick) < 0)) {
@@ -112,7 +117,7 @@ do_settimeofday (struct timeval *tv)
 		 * Discover what correction gettimeofday would have done, and then undo
 		 * it!
 		 */
-		nsec -= gettimeoffset();
+		nsec -= (*gettimeoffset)();
 
 		wtm_sec  = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
 		wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
@@ -124,6 +129,9 @@ do_settimeofday (struct timeval *tv)
 		time_status |= STA_UNSYNC;
 		time_maxerror = NTP_PHASE_LIMIT;
 		time_esterror = NTP_PHASE_LIMIT;
+		if (update_wall_time_hook)
+			(*update_wall_time_hook)();
+
 	}
 	write_sequnlock_irq(&xtime_lock);
 	clock_was_set();
@@ -138,7 +146,7 @@ do_gettimeofday (struct timeval *tv)
 		seq = read_seqbegin(&xtime_lock);
 		{
 			old = last_nsec_offset;
-			offset = gettimeoffset();
+			offset = (*gettimeoffset)();
 			sec = xtime.tv_sec;
 			nsec = xtime.tv_nsec;
 		}
@@ -214,7 +222,7 @@ timer_interrupt (int irq, void *dev_id, struct pt_regs *regs)
 #endif
 		new_itm += local_cpu_data->itm_delta;
 
-		if (smp_processor_id() == 0) {
+		if (smp_processor_id() == TIME_KEEPER_ID) {
 			/*
 			 * Here we are in the timer irq handler. We have irqs locally
 			 * disabled, but we don't know if the timer_bh is running on

arch/ia64/sn/kernel/setup.c

@@ -78,9 +78,9 @@ DEFINE_PER_CPU(struct pda_s, pda_percpu);
 
 extern void bte_init_node (nodepda_t *, cnodeid_t);
 extern void bte_init_cpu (void);
+extern void sn_timer_init (void);
 
 unsigned long sn_rtc_cycles_per_second;   
-unsigned long sn_rtc_usec_per_cyc;
 
 partid_t sn_partid = -1;
 char sn_system_serial_number_string[128];
@@ -263,13 +263,6 @@ sn_setup(char **cmdline_p)
 	else
 		sn_rtc_cycles_per_second = ticks_per_sec;
 
-#ifdef CONFIG_IA64_SGI_SN1
-	/* PROM has wrong value on SN1 */
-	sn_rtc_cycles_per_second = 990177;
-#endif
-	sn_rtc_usec_per_cyc = ((1000000000UL<<IA64_NSEC_PER_CYC_SHIFT)
-			       + sn_rtc_cycles_per_second/2) / sn_rtc_cycles_per_second;
-		
 	for (i=0;i<NR_CPUS;i++)
 		_sn_irq_desc[i] = _irq_desc;
 
@@ -309,6 +302,8 @@ sn_setup(char **cmdline_p)
 	 * Turn off "floating-point assist fault" warnings by default.
 	 */
 	current->thread.flags |= IA64_THREAD_FPEMU_NOPRINT;
+
+	sn_timer_init();
 }
 
 /**

arch/ia64/sn/kernel/sn2/Makefile

@@ -11,4 +11,4 @@
 
 EXTRA_CFLAGS := -DLITTLE_ENDIAN
 
-obj-y += cache.o iomv.o ptc_deadlock.o sn2_smp.o sn_proc_fs.o
+obj-y += cache.o iomv.o ptc_deadlock.o sn2_smp.o sn_proc_fs.o timer.o

arch/ia64/sn/kernel/sn2/timer.c

+/*
+ * linux/arch/ia64/sn/kernel/sn2/timer.c
+ *
+ * Copyright (C) 2003 Silicon Graphics, Inc.
+ */
+
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/time.h>
+#include <linux/interrupt.h>
+
+#include <asm/hw_irq.h>
+#include <asm/system.h>
+
+#include <asm/sn/leds.h>
+#include <asm/sn/clksupport.h>
+
+
+extern unsigned long sn_rtc_cycles_per_second;
+static volatile unsigned long last_wall_rtc;
+
+/**
+ * gettimeoffset - number of nsecs elapsed since &xtime was last updated
+ *
+ * This function is used by do_gettimeofday() to determine the number
+ * of nsecs that have elapsed since the last update to &xtime.  On SN
+ * this is accomplished using the RTC built in to each Hub chip; each
+ * is guaranteed to be synchronized by the PROM, so a local read will
+ * suffice (GET_RTC_COUNTER() does this for us).  A snapshot of the RTC
+ * value is taken every time wall_jiffies is updated by the
+ * update_wall_time_hook (sn2_update_wall_time) which means we don't
+ * have to adjust for lost jiffies ticks or anything like that.
+ */
+
+static volatile long rtc_offset;
+static long rtc_nsecs_per_cycle;
+static long rtc_per_timer_tick;
+
+unsigned long
+sn_gettimeoffset(void)
+{
+	long current_rtc, elapsed_rtc, old, new_offset;
+
+	do {
+		old = rtc_offset;
+		current_rtc = GET_RTC_COUNTER();
+
+		/*
+		 * Need to address wrapping here!
+		 */
+		elapsed_rtc = (long)(current_rtc - last_wall_rtc);
+
+		new_offset = max(elapsed_rtc, old);
+	} while (cmpxchg(&rtc_offset, old, new_offset) != old);
+
+	return new_offset * rtc_nsecs_per_cycle;
+}
+
+
+void sn2_update_wall_time(void)
+{
+	rtc_offset -= min(rtc_offset, rtc_per_timer_tick);
+	last_wall_rtc = GET_RTC_COUNTER();
+}
+
+
+void sn2_reset_wall_time(void)
+{
+	rtc_offset = 0;
+	last_wall_rtc = GET_RTC_COUNTER();
+}
+
+
+void __init
+sn_timer_init(void)
+{
+	rtc_per_timer_tick = sn_rtc_cycles_per_second / HZ;
+	rtc_nsecs_per_cycle = 1000000000 / sn_rtc_cycles_per_second;
+
+	last_wall_rtc = GET_RTC_COUNTER();
+	update_wall_time_hook = sn2_update_wall_time;
+	reset_wall_time_hook = sn2_reset_wall_time;
+	gettimeoffset = sn_gettimeoffset;
+}

include/asm-ia64/sal.h

@@ -23,6 +23,18 @@
  *                      (plus examples of platform error info structures from smariset @ Intel)
  */
 
+#define IA64_SAL_PLATFORM_FEATURE_BUS_LOCK_BIT		0
+#define IA64_SAL_PLATFORM_FEATURE_IRQ_REDIR_HINT_BIT	1
+#define IA64_SAL_PLATFORM_FEATURE_IPI_REDIR_HINT_BIT	2
+#define IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT_BIT	 	3
+
+#define IA64_SAL_PLATFORM_FEATURE_BUS_LOCK	  (1<<IA64_SAL_PLATFORM_FEATURE_BUS_LOCK_BIT)
+#define IA64_SAL_PLATFORM_FEATURE_IRQ_REDIR_HINT (1<<IA64_SAL_PLATFORM_FEATURE_IRQ_REDIR_HINT_BIT)
+#define IA64_SAL_PLATFORM_FEATURE_IPI_REDIR_HINT (1<<IA64_SAL_PLATFORM_FEATURE_IPI_REDIR_HINT_BIT)
+#define IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT	  (1<<IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT_BIT)
+
+#ifndef __ASSEMBLY__
+
 #include <linux/spinlock.h>
 #include <linux/efi.h>
 
@@ -162,11 +174,6 @@ typedef struct ia64_sal_desc_memory {
 	u8 oem_reserved[8];
 } ia64_sal_desc_memory_t;
 
-#define IA64_SAL_PLATFORM_FEATURE_BUS_LOCK		(1 << 0)
-#define IA64_SAL_PLATFORM_FEATURE_IRQ_REDIR_HINT	(1 << 1)
-#define IA64_SAL_PLATFORM_FEATURE_IPI_REDIR_HINT	(1 << 2)
-#define IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT	 	(1 << 3)
-
 typedef struct ia64_sal_desc_platform_feature {
 	u8 type;
 	u8 feature_mask;
@@ -790,4 +797,6 @@ ia64_sal_update_pal (u64 param_buf, u64 scratch_buf, u64 scratch_buf_size,
 
 extern unsigned long sal_platform_features;
 
+#endif /* __ASSEMBLY__ */
+
 #endif /* _ASM_IA64_PAL_H */

include/asm-ia64/sn/clksupport.h

@@ -38,7 +38,7 @@ extern unsigned long sn_rtc_cycles_per_second;
 
 extern nasid_t master_nasid;
 
-#define RTC_MASK		(0x007fffffffffffff)
+#define RTC_MASK		0x007fffffffffffffUL
 /* clocks are not synchronized yet on SN1  - used node 0 (problem if no NASID 0) */
 #define RTC_COUNTER_ADDR	((clkreg_t*)REMOTE_HUB_ADDR(master_nasid, PI_RT_COUNTER))
 #define RTC_COMPARE_A_ADDR      ((clkreg_t*)REMOTE_HUB_ADDR(master_nasid, PI_RT_COMPARE_A))
@@ -52,7 +52,7 @@ extern nasid_t master_nasid;
 #include <asm/sn/sn2/addrs.h>
 #include <asm/sn/sn2/shubio.h>
 #include <asm/sn/sn2/shub_mmr.h>
-#define RTC_MASK		(SH_RTC_MASK)
+#define RTC_MASK		SH_RTC_MASK
 #define RTC_COUNTER_ADDR	((clkreg_t*)LOCAL_MMR_ADDR(SH_RTC))
 #define RTC_COMPARE_A_ADDR      ((clkreg_t*)LOCAL_MMR_ADDR(SH_RTC))
 #define RTC_COMPARE_B_ADDR      ((clkreg_t*)LOCAL_MMR_ADDR(SH_RTC))
@@ -62,7 +62,6 @@ extern nasid_t master_nasid;
 #define RTC_INT_ENABLED_B_ADDR  ((clkreg_t*)LOCAL_MMR_ADDR(SH_RTC))
 #endif	/* CONFIG_IA64_SGI_SN1 */
 
-
 #define GET_RTC_COUNTER()	(*RTC_COUNTER_ADDR)
 #define rtc_time()		GET_RTC_COUNTER()
 

include/asm-ia64/timex.h

@@ -25,4 +25,6 @@ get_cycles (void)
 	return ret;
 }
 
+extern unsigned long (*gettimeoffset)(void);
+
 #endif /* _ASM_IA64_TIMEX_H */