powerpc: Implement accurate task and CPU time accounting

This implements accurate task and cpu time accounting for 64-bit
powerpc kernels.  Instead of accounting a whole jiffy of time to a
task on a timer interrupt because that task happened to be running at
the time, we now account time in units of timebase ticks according to
the actual time spent by the task in user mode and kernel mode.  We
also count the time spent processing hardware and software interrupts
accurately.  This is conditional on CONFIG_VIRT_CPU_ACCOUNTING.  If
that is not set, we do tick-based approximate accounting as before.

To get this accurate information, we read either the PURR (processor
utilization of resources register) on POWER5 machines, or the timebase
on other machines on

* each entry to the kernel from usermode
* each exit to usermode
* transitions between process context, hard irq context and soft irq
  context in kernel mode
* context switches.

On POWER5 systems with shared-processor logical partitioning we also
read both the PURR and the timebase at each timer interrupt and
context switch in order to determine how much time has been taken by
the hypervisor to run other partitions ("steal" time).  Unfortunately,
since we need values of the PURR on both threads at the same time to
accurately calculate the steal time, and since we can only calculate
steal time on a per-core basis, the apportioning of the steal time
between idle time (time which we ceded to the hypervisor in the idle
loop) and actual stolen time is somewhat approximate at the moment.

This is all based quite heavily on what s390 does, and it uses the
generic interfaces that were added by the s390 developers,
i.e. account_system_time(), account_user_time(), etc.

This patch doesn't add any new interfaces between the kernel and
userspace, and doesn't change the units in which time is reported to
userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(),
times(), etc.  Internally the various task and cpu times are stored in
timebase units, but they are converted to USER_HZ units (1/100th of a
second) when reported to userspace.  Some precision is therefore lost
but there should not be any accumulating error, since the internal
accumulation is at full precision.

Signed-off-by: Paul Mackerras <paulus@samba.org>

8 files changed, 297 insertions, 14 deletions

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index fb0dcb994b84..d112aed2999b 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig
@@ -250,6 +250,21 @@ config PPC_STD_MMU_32
         def_bool y
         depends on PPC_STD_MMU && PPC32
 
+config VIRT_CPU_ACCOUNTING
+        bool "Deterministic task and CPU time accounting"
+        depends on PPC64
+        default y
+        help
+          Select this option to enable more accurate task and CPU time
+          accounting.  This is done by reading a CPU counter on each
+          kernel entry and exit and on transitions within the kernel
+          between system, softirq and hardirq state, so there is a
+          small performance impact.  This also enables accounting of
+          stolen time on logically-partitioned systems running on
+          IBM POWER5-based machines.
+
+          If in doubt, say Y here.
+
 config SMP
         depends on PPC_STD_MMU
         bool "Symmetric multi-processing support"
diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
index 840aad43a98b..18810ac55bcc 100644
--- a/arch/powerpc/kernel/asm-offsets.c
+++ b/arch/powerpc/kernel/asm-offsets.c
@@ -137,6 +137,9 @@ int main(void)
         DEFINE(PACAEMERGSP, offsetof(struct paca_struct, emergency_sp));
         DEFINE(PACALPPACAPTR, offsetof(struct paca_struct, lppaca_ptr));
         DEFINE(PACAHWCPUID, offsetof(struct paca_struct, hw_cpu_id));
+        DEFINE(PACA_STARTPURR, offsetof(struct paca_struct, startpurr));
+        DEFINE(PACA_USER_TIME, offsetof(struct paca_struct, user_time));
+        DEFINE(PACA_SYSTEM_TIME, offsetof(struct paca_struct, system_time));
 
         DEFINE(LPPACASRR0, offsetof(struct lppaca, saved_srr0));
         DEFINE(LPPACASRR1, offsetof(struct lppaca, saved_srr1));
diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
index 79a0c910f0d8..8f606c1889fa 100644
--- a/arch/powerpc/kernel/entry_64.S
+++ b/arch/powerpc/kernel/entry_64.S
@@ -61,6 +61,7 @@ system_call_common:
         std     r12,_MSR(r1)
         std     r0,GPR0(r1)
         std     r10,GPR1(r1)
+        ACCOUNT_CPU_USER_ENTRY(r10, r11)
         std     r2,GPR2(r1)
         std     r3,GPR3(r1)
         std     r4,GPR4(r1)
@@ -168,8 +169,9 @@ syscall_error_cont:
         stdcx.  r0,0,r1                 /* to clear the reservation */
         andi.   r6,r8,MSR_PR
         ld      r4,_LINK(r1)
-        beq-    1f                      /* only restore r13 if */
-        ld      r13,GPR13(r1)           /* returning to usermode */
+        beq-    1f
+        ACCOUNT_CPU_USER_EXIT(r11, r12)
+        ld      r13,GPR13(r1)   /* only restore r13 if returning to usermode */
 1:      ld      r2,GPR2(r1)
         li      r12,MSR_RI
         andc    r11,r10,r12
@@ -536,6 +538,7 @@ restore:
          * userspace
          */
         beq     1f
+        ACCOUNT_CPU_USER_EXIT(r3, r4)
         REST_GPR(13, r1)
 1:
         ld      r3,_CTR(r1)
diff --git a/arch/powerpc/kernel/head_64.S b/arch/powerpc/kernel/head_64.S
index 2b21ec499285..be3ae7733577 100644
--- a/arch/powerpc/kernel/head_64.S
+++ b/arch/powerpc/kernel/head_64.S
@@ -277,6 +277,7 @@ exception_marker:
         std     r10,0(r1);              /* make stack chain pointer     */ \
         std     r0,GPR0(r1);            /* save r0 in stackframe        */ \
         std     r10,GPR1(r1);           /* save r1 in stackframe        */ \
+        ACCOUNT_CPU_USER_ENTRY(r9, r10);                                   \
         std     r2,GPR2(r1);            /* save r2 in stackframe        */ \
         SAVE_4GPRS(3, r1);              /* save r3 - r6 in stackframe   */ \
         SAVE_2GPRS(7, r1);              /* save r7, r8 in stackframe    */ \
@@ -844,6 +845,14 @@ fast_exception_return:
         ld      r11,_NIP(r1)
         andi.   r3,r12,MSR_RI           /* check if RI is set */
         beq-    unrecov_fer
+
+#ifdef CONFIG_VIRT_CPU_ACCOUNTING
+        andi.   r3,r12,MSR_PR
+        beq     2f
+        ACCOUNT_CPU_USER_EXIT(r3, r4)
+2:
+#endif
+
         ld      r3,_CCR(r1)
         ld      r4,_LINK(r1)
         ld      r5,_CTR(r1)
diff --git a/arch/powerpc/kernel/irq.c b/arch/powerpc/kernel/irq.c
index edb2b00edbd2..24dc8117b822 100644
--- a/arch/powerpc/kernel/irq.c
+++ b/arch/powerpc/kernel/irq.c
@@ -369,6 +369,7 @@ unsigned int real_irq_to_virt_slowpath(unsigned int real_irq)
         return NO_IRQ;
 
 }
+#endif /* CONFIG_PPC64 */
 
 #ifdef CONFIG_IRQSTACKS
 struct thread_info *softirq_ctx[NR_CPUS];
@@ -392,10 +393,24 @@ void irq_ctx_init(void)
         }
 }
 
+static inline void do_softirq_onstack(void)
+{
+        struct thread_info *curtp, *irqtp;
+
+        curtp = current_thread_info();
+        irqtp = softirq_ctx[smp_processor_id()];
+        irqtp->task = curtp->task;
+        call_do_softirq(irqtp);
+        irqtp->task = NULL;
+}
+
+#else
+#define do_softirq_onstack()    __do_softirq()
+#endif /* CONFIG_IRQSTACKS */
+
 void do_softirq(void)
 {
         unsigned long flags;
-        struct thread_info *curtp, *irqtp;
 
         if (in_interrupt())
                 return;
@@ -403,19 +418,18 @@ void do_softirq(void)
         local_irq_save(flags);
 
         if (local_softirq_pending()) {
-                curtp = current_thread_info();
-                irqtp = softirq_ctx[smp_processor_id()];
-                irqtp->task = curtp->task;
-                call_do_softirq(irqtp);
-                irqtp->task = NULL;
+                account_system_vtime(current);
+                local_bh_disable();
+                do_softirq_onstack();
+                account_system_vtime(current);
+                __local_bh_enable();
         }
 
         local_irq_restore(flags);
 }
 EXPORT_SYMBOL(do_softirq);
 
-#endif /* CONFIG_IRQSTACKS */
-
+#ifdef CONFIG_PPC64
 static int __init setup_noirqdistrib(char *str)
 {
         distribute_irqs = 0;
diff --git a/arch/powerpc/kernel/process.c b/arch/powerpc/kernel/process.c
index dd774c3c9302..1770a066c217 100644
--- a/arch/powerpc/kernel/process.c
+++ b/arch/powerpc/kernel/process.c
@@ -45,9 +45,9 @@
 #include <asm/mmu.h>
 #include <asm/prom.h>
 #include <asm/machdep.h>
+#include <asm/time.h>
 #ifdef CONFIG_PPC64
 #include <asm/firmware.h>
-#include <asm/time.h>
 #endif
 
 extern unsigned long _get_SP(void);
@@ -328,6 +328,11 @@ struct task_struct *__switch_to(struct task_struct *prev,
 #endif
 
         local_irq_save(flags);
+
+        account_system_vtime(current);
+        account_process_vtime(current);
+        calculate_steal_time();
+
         last = _switch(old_thread, new_thread);
 
         local_irq_restore(flags);
diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
index 13595a64f013..805eaedbc308 100644
--- a/arch/powerpc/kernel/smp.c
+++ b/arch/powerpc/kernel/smp.c
@@ -541,7 +541,7 @@ int __devinit start_secondary(void *unused)
                 smp_ops->take_timebase();
 
         if (system_state > SYSTEM_BOOTING)
-                per_cpu(last_jiffy, cpu) = get_tb();
+                snapshot_timebase();
 
         spin_lock(&call_lock);
         cpu_set(cpu, cpu_online_map);
@@ -573,6 +573,8 @@ void __init smp_cpus_done(unsigned int max_cpus)
 
         set_cpus_allowed(current, old_mask);
 
+        snapshot_timebases();
+
         dump_numa_cpu_topology();
 }
 
diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
index 2a7ddc579379..0b34db28916f 100644
--- a/arch/powerpc/kernel/time.c
+++ b/arch/powerpc/kernel/time.c
@@ -51,6 +51,7 @@
 #include <linux/percpu.h>
 #include <linux/rtc.h>
 #include <linux/jiffies.h>
+#include <linux/posix-timers.h>
 
 #include <asm/io.h>
 #include <asm/processor.h>
@@ -135,6 +136,220 @@ unsigned long tb_last_stamp;
  */
 DEFINE_PER_CPU(unsigned long, last_jiffy);
 
+#ifdef CONFIG_VIRT_CPU_ACCOUNTING
+/*
+ * Factors for converting from cputime_t (timebase ticks) to
+ * jiffies, milliseconds, seconds, and clock_t (1/USER_HZ seconds).
+ * These are all stored as 0.64 fixed-point binary fractions.
+ */
+u64 __cputime_jiffies_factor;
+u64 __cputime_msec_factor;
+u64 __cputime_sec_factor;
+u64 __cputime_clockt_factor;
+
+static void calc_cputime_factors(void)
+{
+        struct div_result res;
+
+        div128_by_32(HZ, 0, tb_ticks_per_sec, &res);
+        __cputime_jiffies_factor = res.result_low;
+        div128_by_32(1000, 0, tb_ticks_per_sec, &res);
+        __cputime_msec_factor = res.result_low;
+        div128_by_32(1, 0, tb_ticks_per_sec, &res);
+        __cputime_sec_factor = res.result_low;
+        div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res);
+        __cputime_clockt_factor = res.result_low;
+}
+
+/*
+ * Read the PURR on systems that have it, otherwise the timebase.
+ */
+static u64 read_purr(void)
+{
+        if (cpu_has_feature(CPU_FTR_PURR))
+                return mfspr(SPRN_PURR);
+        return mftb();
+}
+
+/*
+ * Account time for a transition between system, hard irq
+ * or soft irq state.
+ */
+void account_system_vtime(struct task_struct *tsk)
+{
+        u64 now, delta;
+        unsigned long flags;
+
+        local_irq_save(flags);
+        now = read_purr();
+        delta = now - get_paca()->startpurr;
+        get_paca()->startpurr = now;
+        if (!in_interrupt()) {
+                delta += get_paca()->system_time;
+                get_paca()->system_time = 0;
+        }
+        account_system_time(tsk, 0, delta);
+        local_irq_restore(flags);
+}
+
+/*
+ * Transfer the user and system times accumulated in the paca
+ * by the exception entry and exit code to the generic process
+ * user and system time records.
+ * Must be called with interrupts disabled.
+ */
+void account_process_vtime(struct task_struct *tsk)
+{
+        cputime_t utime;
+
+        utime = get_paca()->user_time;
+        get_paca()->user_time = 0;
+        account_user_time(tsk, utime);
+}
+
+static void account_process_time(struct pt_regs *regs)
+{
+        int cpu = smp_processor_id();
+
+        account_process_vtime(current);
+        run_local_timers();
+        if (rcu_pending(cpu))
+                rcu_check_callbacks(cpu, user_mode(regs));
+        scheduler_tick();
+        run_posix_cpu_timers(current);
+}
+
+#ifdef CONFIG_PPC_SPLPAR
+/*
+ * Stuff for accounting stolen time.
+ */
+struct cpu_purr_data {
+        int     initialized;                    /* thread is running */
+        u64     tb0;                    /* timebase at origin time */
+        u64     purr0;                  /* PURR at origin time */
+        u64     tb;                     /* last TB value read */
+        u64     purr;                   /* last PURR value read */
+        u64     stolen;                 /* stolen time so far */
+        spinlock_t lock;
+};
+
+static DEFINE_PER_CPU(struct cpu_purr_data, cpu_purr_data);
+
+static void snapshot_tb_and_purr(void *data)
+{
+        struct cpu_purr_data *p = &__get_cpu_var(cpu_purr_data);
+
+        p->tb0 = mftb();
+        p->purr0 = mfspr(SPRN_PURR);
+        p->tb = p->tb0;
+        p->purr = 0;
+        wmb();
+        p->initialized = 1;
+}
+
+/*
+ * Called during boot when all cpus have come up.
+ */
+void snapshot_timebases(void)
+{
+        int cpu;
+
+        if (!cpu_has_feature(CPU_FTR_PURR))
+                return;
+        for_each_cpu(cpu)
+                spin_lock_init(&per_cpu(cpu_purr_data, cpu).lock);
+        on_each_cpu(snapshot_tb_and_purr, NULL, 0, 1);
+}
+
+void calculate_steal_time(void)
+{
+        u64 tb, purr, t0;
+        s64 stolen;
+        struct cpu_purr_data *p0, *pme, *phim;
+        int cpu;
+
+        if (!cpu_has_feature(CPU_FTR_PURR))
+                return;
+        cpu = smp_processor_id();
+        pme = &per_cpu(cpu_purr_data, cpu);
+        if (!pme->initialized)
+                return;         /* this can happen in early boot */
+        p0 = &per_cpu(cpu_purr_data, cpu & ~1);
+        phim = &per_cpu(cpu_purr_data, cpu ^ 1);
+        spin_lock(&p0->lock);
+        tb = mftb();
+        purr = mfspr(SPRN_PURR) - pme->purr0;
+        if (!phim->initialized || !cpu_online(cpu ^ 1)) {
+                stolen = (tb - pme->tb) - (purr - pme->purr);
+        } else {
+                t0 = pme->tb0;
+                if (phim->tb0 < t0)
+                        t0 = phim->tb0;
+                stolen = phim->tb - t0 - phim->purr - purr - p0->stolen;
+        }
+        if (stolen > 0) {
+                account_steal_time(current, stolen);
+                p0->stolen += stolen;
+        }
+        pme->tb = tb;
+        pme->purr = purr;
+        spin_unlock(&p0->lock);
+}
+
+/*
+ * Must be called before the cpu is added to the online map when
+ * a cpu is being brought up at runtime.
+ */
+static void snapshot_purr(void)
+{
+        int cpu;
+        u64 purr;
+        struct cpu_purr_data *p0, *pme, *phim;
+        unsigned long flags;
+
+        if (!cpu_has_feature(CPU_FTR_PURR))
+                return;
+        cpu = smp_processor_id();
+        pme = &per_cpu(cpu_purr_data, cpu);
+        p0 = &per_cpu(cpu_purr_data, cpu & ~1);
+        phim = &per_cpu(cpu_purr_data, cpu ^ 1);
+        spin_lock_irqsave(&p0->lock, flags);
+        pme->tb = pme->tb0 = mftb();
+        purr = mfspr(SPRN_PURR);
+        if (!phim->initialized) {
+                pme->purr = 0;
+                pme->purr0 = purr;
+        } else {
+                /* set p->purr and p->purr0 for no change in p0->stolen */
+                pme->purr = phim->tb - phim->tb0 - phim->purr - p0->stolen;
+                pme->purr0 = purr - pme->purr;
+        }
+        pme->initialized = 1;
+        spin_unlock_irqrestore(&p0->lock, flags);
+}
+
+#endif /* CONFIG_PPC_SPLPAR */
+
+#else /* ! CONFIG_VIRT_CPU_ACCOUNTING */
+#define calc_cputime_factors()
+#define account_process_time(regs)      update_process_times(user_mode(regs))
+#define calculate_steal_time()          do { } while (0)
+#endif
+
+#if !(defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR))
+#define snapshot_purr()                 do { } while (0)
+#endif
+
+/*
+ * Called when a cpu comes up after the system has finished booting,
+ * i.e. as a result of a hotplug cpu action.
+ */
+void snapshot_timebase(void)
+{
+        __get_cpu_var(last_jiffy) = get_tb();
+        snapshot_purr();
+}
+
 void __delay(unsigned long loops)
 {
         unsigned long start;
@@ -382,6 +597,7 @@ static void iSeries_tb_recal(void)
                                                 new_tb_ticks_per_jiffy, sign, tick_diff );
                                 tb_ticks_per_jiffy = new_tb_ticks_per_jiffy;
                                 tb_ticks_per_sec   = new_tb_ticks_per_sec;
+                                calc_cputime_factors();
                                 div128_by_32( XSEC_PER_SEC, 0, tb_ticks_per_sec, &divres );
                                 do_gtod.tb_ticks_per_sec = tb_ticks_per_sec;
                                 tb_to_xs = divres.result_low;
@@ -430,6 +646,7 @@ void timer_interrupt(struct pt_regs * regs)
         irq_enter();
 
         profile_tick(CPU_PROFILING, regs);
+        calculate_steal_time();
 
 #ifdef CONFIG_PPC_ISERIES
         get_lppaca()->int_dword.fields.decr_int = 0;
@@ -451,7 +668,7 @@ void timer_interrupt(struct pt_regs * regs)
                  * is the case.
                  */
                 if (!cpu_is_offline(cpu))
-                        update_process_times(user_mode(regs));
+                        account_process_time(regs);
 
                 /*
                  * No need to check whether cpu is offline here; boot_cpuid
@@ -508,13 +725,27 @@ void wakeup_decrementer(void)
 void __init smp_space_timers(unsigned int max_cpus)
 {
         int i;
+        unsigned long half = tb_ticks_per_jiffy / 2;
         unsigned long offset = tb_ticks_per_jiffy / max_cpus;
         unsigned long previous_tb = per_cpu(last_jiffy, boot_cpuid);
 
         /* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */
         previous_tb -= tb_ticks_per_jiffy;
+        /*
+         * The stolen time calculation for POWER5 shared-processor LPAR
+         * systems works better if the two threads' timebase interrupts
+         * are staggered by half a jiffy with respect to each other.
+         */
         for_each_cpu(i) {
-                if (i != boot_cpuid) {
+                if (i == boot_cpuid)
+                        continue;
+                if (i == (boot_cpuid ^ 1))
+                        per_cpu(last_jiffy, i) =
+                                per_cpu(last_jiffy, boot_cpuid) - half;
+                else if (i & 1)
+                        per_cpu(last_jiffy, i) =
+                                per_cpu(last_jiffy, i ^ 1) + half;
+                else {
                         previous_tb += offset;
                         per_cpu(last_jiffy, i) = previous_tb;
                 }
@@ -706,6 +937,7 @@ void __init time_init(void)
         tb_ticks_per_sec = ppc_tb_freq;
         tb_ticks_per_usec = ppc_tb_freq / 1000000;
         tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000);
+        calc_cputime_factors();
 
         /*
          * Calculate the length of each tick in ns.  It will not be