/*--------------------------------------------------------------------------------*\
\*--------------------------------------------------------------------------------*/
#include <linux/version.h>
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/mm.h>
#ifdef CONFIG_AVM_POWERMETER
#include <linux/avm_power.h>
#include <linux/vmstat.h>
#endif/*--- #ifdef CONFIG_AVM_POWERMETER ---*/
#if defined(CONFIG_NMI_ARBITER_WORKAROUND)
#include <atheros.h>
#endif/*--- #if defined(CONFIG_NMI_ARBITER_WORKAROUND) ---*/
#if defined(CONFIG_AVM_SIMPLE_PROFILING)
#include <linux/avm_profile.h>
#endif/*--- #if defined(CONFIG_AVM_SIMPLE_PROFILING) ---*/
#include <asm/mach_avm.h>

#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
/**
 * get_avenrun - get the load average array
 * @loads:	pointer to dest load array
 * @offset:	offset to add
 * @shift:	shift count to shift the result left
 *
 * These values are estimates at best, so no need for locking.
 */
void get_avenrun(unsigned long *loads, unsigned long offset, int shift) {
	loads[0] = (avenrun[0] + offset) << shift;
	loads[1] = (avenrun[1] + offset) << shift;
	loads[2] = (avenrun[2] + offset) << shift;
}
#endif/*--- #if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32) ---*/

#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 8, 0)
#define task_cputime_adjusted(a,b,c) task_times(a,b, c)
#endif/*--- #if LINUX_VERSION_CODE < KERNEL_VERSION(3, 8, 0) ---*/

#if defined(CONFIG_X86)
#define CPU_COUNT_SHIFT     4
#else/*--- #if defined(CONFIG_X86) ---*/
#define CPU_COUNT_SHIFT     0
#endif/*--- #else ---*//*--- #if defined(CONFIG_X86) ---*/
atomic_t avm_page_faultlink[NR_CPUS];
int avm_page_faultlink_max[NR_CPUS];

static unsigned long long gCycle_per_10sec = 250ULL * 1000ULL * 1000ULL * 10ULL;

/*--------------------------------------------------------------------------------*\
\*--------------------------------------------------------------------------------*/
static unsigned long vmevents[NR_VM_EVENT_ITEMS];
static unsigned long last_pgalloc_normal;
static unsigned long last_pg_fault_cnt;

/*--------------------------------------------------------------------------------------*\
\*--------------------------------------------------------------------------------------*/
static struct _cpu_idlecount {
    unsigned int last_value;
    unsigned long last_value_done;
    unsigned long long run_summary;
    unsigned int wait_count;
    unsigned int run_count;
    unsigned int sum_count;
    unsigned int show;
    unsigned int fullrun;
    unsigned int actrun;
    unsigned int fulldisp_jiffies;
} cpu_idlecount[NR_CPUS];

#define COUNT_DIFF(act, last)   ((act) - (last))
/*--------------------------------------------------------------------------------*\
\*--------------------------------------------------------------------------------*/
static inline unsigned check_10second_without_idle(unsigned long act_cycle, struct _cpu_idlecount *pcpuidle) {
    pcpuidle->run_summary += COUNT_DIFF(act_cycle, pcpuidle->last_value);
    if(pcpuidle->run_summary > gCycle_per_10sec) {
        pcpuidle->run_summary = 0;
        return 1;
    }
    return 0;
}

/*--------------------------------------------------------------------------------*\
 * liefert ob diese CPU im IDLE-Mode
\*--------------------------------------------------------------------------------*/
unsigned int avm_cpu_idle_state(int vpe) {
    struct _cpu_idlecount *pcpuidle = &cpu_idlecount[vpe];
    return !pcpuidle->last_value_done;
}
/*--------------------------------------------------------------------------------*\
\*--------------------------------------------------------------------------------*/
void avm_cpu_wait_start(void) {
    unsigned int count, tmp;
    struct _cpu_idlecount *pcpuidle = &cpu_idlecount[raw_smp_processor_id()];

#if defined(CONFIG_AVM_SIMPLE_PROFILING) 
    avm_simple_profiling_enter_idle((unsigned int)avm_cpu_wait_start);
#endif/*--- #if defined(CONFIG_AVM_SIMPLE_PROFILING)  ---*/
    count = avm_get_cycles();
    if(pcpuidle->last_value) {
        tmp = COUNT_DIFF(count, pcpuidle->last_value) >> CPU_COUNT_SHIFT;
        pcpuidle->run_count += tmp;
        pcpuidle->sum_count += tmp;
    }
    pcpuidle->last_value      = count;
    pcpuidle->last_value_done = 0;
    pcpuidle->run_summary     = 0;
    if(unlikely(pcpuidle->fulldisp_jiffies == 0)) {
        pcpuidle->fulldisp_jiffies = jiffies;
    }
}
#ifdef CONFIG_AVM_POWERMETER
static struct _hist_task_times {
    unsigned long time_last;
    unsigned int pid;
    unsigned int mark;
} hist_task_times[180];

#define ARRAY_EL(a) (sizeof(a) / sizeof(a[0]))
/*--------------------------------------------------------------------------------*\
\*--------------------------------------------------------------------------------*/
static struct _hist_task_times *get_histtask_by_pid(unsigned int pid, unsigned int *cache_idx) {
    register unsigned int i;
    if(*cache_idx >= ARRAY_EL(hist_task_times)) {
        *cache_idx = 0;
    }
    for(i = *cache_idx; i < ARRAY_EL(hist_task_times); i++) {
        if(hist_task_times[i].pid == pid) {
            *cache_idx = i + 1;
            hist_task_times[i].mark = 2;
            return &hist_task_times[i];
        }
    }
    for(i = 0; i < *cache_idx; i++) {
        if(hist_task_times[i].pid == pid) {
            *cache_idx = i + 1;
            hist_task_times[i].mark = 2;
            return &hist_task_times[i];
        }
    }
    /*--- Create ---*/
    for(i = 0; i < ARRAY_EL(hist_task_times); i++) {
        if(hist_task_times[i].mark == 0) {
            hist_task_times[i].mark = 2;
            hist_task_times[i].time_last = 0;
            hist_task_times[i].pid  = pid;
            return &hist_task_times[i];
        } 
    }
    /*--- printk("<error pid %d>", pid); ---*/
    return NULL;
}
struct _accumulate_output {
    unsigned long sum_time;
    unsigned long curr_time;
    unsigned long maxrun_time;
    struct task_struct *maxrun_task;
    unsigned int tasks;
};
/*--------------------------------------------------------------------------------*\
\*--------------------------------------------------------------------------------*/
static void accumulate_time(struct _accumulate_output *ao) {
    struct task_struct *g, *p, *maxrun_task = current;
    register unsigned int i;
    unsigned int cache_idx = 0;
    register unsigned long time_sum = 0;
    register unsigned long maxrun_time = 0;
    unsigned int tsks = 0;
    ao->maxrun_task = current;
    ao->curr_time   = 0;
	do_each_thread(g, p) {
        struct _hist_task_times *pht = get_histtask_by_pid(p->pid, &cache_idx);
        tsks++;
        if(pht) {
            unsigned long time_act;
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 33)
            unsigned long ut, st;
            task_cputime_adjusted(p, &ut, &st);
            time_act = ut + st;
#else/*--- #if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 33) ---*/
            time_act = task_stime(p) + task_utime(p);
#endif/*--- #else ---*//*--- #if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 33) ---*/
            time_sum += time_act - pht->time_last;
            if(maxrun_time < (time_act - pht->time_last)) {
                maxrun_time = (time_act - pht->time_last);
                maxrun_task = p;
            }
            if(unlikely(p == current)){
                ao->curr_time = time_act - pht->time_last;
            }
            pht->time_last = time_act;
        } else {
            ao->tasks    = tsks;
            ao->sum_time = 0;
            return;
        }
	} while_each_thread(g, p);

    for(i = 0; i < ARRAY_EL(hist_task_times); i++) {
        struct _hist_task_times *pht = &hist_task_times[i];
        if(pht->mark) {
            pht->mark--; /*--- um geloeschte Eintraege frei zu bekommen  ---*/
        }
    }
    ao->maxrun_task = maxrun_task;
    ao->maxrun_time = maxrun_time;
    ao->sum_time    = time_sum | 1;
    ao->tasks       = tsks;
}
#endif/*--- #ifdef CONFIG_AVM_POWERMETER ---*/
/*--------------------------------------------------------------------------------------*\
 * for statistic function 
 * idle stopps
 * call at the end of idle in wait_irq_off-function or direct in the irq-function
\*--------------------------------------------------------------------------------------*/
int avm_cpu_wait_end(void) {
    unsigned int cpu =  raw_smp_processor_id();
    unsigned int cpu_run = 0;
    struct _cpu_idlecount *pcpuidle = &cpu_idlecount[cpu];

    if(test_and_set_bit(0, &pcpuidle->last_value_done) == 0) {
        int count, tmp;
        count = avm_get_cycles();
        tmp = COUNT_DIFF(count, pcpuidle->last_value) >> CPU_COUNT_SHIFT;
        pcpuidle->wait_count += tmp;
        pcpuidle->sum_count  += tmp;
#if defined(CONFIG_NMI_ARBITER_WORKAROUND)
        {
            tmp =  ath_workaround_nmi_wastetime();
            if(pcpuidle->wait_count > tmp) {
                pcpuidle->wait_count -= tmp;
            } else {
                pcpuidle->wait_count = 0;
            }
        }
#endif/*--- #if defined(CONFIG_NMI_ARBITER_WORKAROUND) ---*/
        pcpuidle->last_value  = count;
        return 0;
    }
#ifdef CONFIG_AVM_POWERMETER
#define LOAD_INT(x) ((x) >> FSHIFT)
#define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100)
    if(check_10second_without_idle(avm_get_cycles(), pcpuidle)) {
        register unsigned long j_diff;
        unsigned long curr_percent = 0, max_percent = 0, sum_percent = 0, pg_fault_cnt, pg_fault_max; 
        pcpuidle->last_value       = avm_get_cycles();
        if(pcpuidle->fullrun == 0) {
            unsigned int i;
            pcpuidle->actrun = 100;
            for(i = 0; i < NR_CPUS; i++) {
                struct _cpu_idlecount *pcpuidle = &cpu_idlecount[i];
                cpu_run |= pcpuidle->actrun << (i * 8);
            }
            PowerManagmentRessourceInfo(powerdevice_loadrate, cpu_run);
        }
        pcpuidle->fullrun++;
        j_diff = COUNT_DIFF(jiffies, pcpuidle->fulldisp_jiffies);
        if(j_diff >= (10 * HZ)) {
            struct _accumulate_output ao;
            unsigned long avnrun[3];
            unsigned long pgalloc_normal, pg_alloc_diff, sum_pages;

            accumulate_time(&ao);
            get_avenrun(avnrun, FIXED_1/200, 0);
            if(ao.sum_time) {
                /*--- Promillewert im Verhaeltnis zu allen Tasks ---*/
                curr_percent = (ao.curr_time * 100)   / j_diff;
                max_percent  = (ao.maxrun_time * 100) / j_diff;
                sum_percent  = (ao.sum_time * 100)    / j_diff;
            }
            pg_fault_max = avm_page_faultlink_max[cpu];
            avm_page_faultlink_max[cpu] = 0;
#ifdef CONFIG_VM_EVENT_COUNTERS
            __all_vm_events(vmevents);
#endif/*--- #ifdef CONFIG_VM_EVENT_COUNTERS ---*/

            pgalloc_normal = vmevents[PGALLOC_NORMAL];
            pg_fault_cnt   = vmevents[PGFAULT];

            pg_alloc_diff = (pgalloc_normal - last_pgalloc_normal);

            sum_pages = global_page_state(NR_ACTIVE_ANON)+
                        global_page_state(NR_INACTIVE_ANON)+
                        global_page_state(NR_ACTIVE_FILE)+
                        global_page_state(NR_INACTIVE_FILE)+
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32)
                        global_page_state(NR_ISOLATED_ANON)+
                        global_page_state(NR_ISOLATED_FILE)+
                        global_page_state(NR_SHMEM)+
#endif/*--- #if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) ---*/
                        global_page_state(NR_UNEVICTABLE)+
                        global_page_state(NR_FILE_DIRTY)+
                        global_page_state(NR_WRITEBACK)+
                        global_page_state(NR_UNSTABLE_NFS)+
                        global_page_state(NR_FREE_PAGES)+
                        global_page_state(NR_SLAB_RECLAIMABLE)+
                        global_page_state(NR_SLAB_UNRECLAIMABLE)+
                        global_page_state(NR_FILE_MAPPED)+
                        global_page_state(NR_PAGETABLE)+
                        global_page_state(NR_BOUNCE);

            printk(KERN_WARNING"system-load %d%s loadavg %lu.%lu %lu.%lu %lu.%lu - %d tasks:%lu %% curr:%s(%lu %%) max:%s(%lu %%, pid:%d) "
                               "pgstat: sum=%lu free=%lu slab=%lu alloc=%lu/s fault=%lu/s (sleep %lu)"
                    "\n", 
                    pcpuidle->fullrun >= 10 ? 100 : pcpuidle->fullrun,
                    pcpuidle->fullrun >= 10 ? "%" : "",
                    LOAD_INT(avnrun[0]), LOAD_FRAC(avnrun[0]),
                    LOAD_INT(avnrun[1]), LOAD_FRAC(avnrun[1]),
                    LOAD_INT(avnrun[2]), LOAD_FRAC(avnrun[2]),
                    ao.tasks,
                    sum_percent,
                    current->comm,
                    curr_percent,
                    ao.maxrun_task->comm,
                    max_percent,
                    ao.maxrun_task->pid,
                    sum_pages,
                    global_page_state(NR_FREE_PAGES),
                    global_page_state(NR_SLAB_UNRECLAIMABLE) + global_page_state(NR_SLAB_UNRECLAIMABLE),
                    pg_alloc_diff / (j_diff / HZ),
                    (pg_fault_cnt - last_pg_fault_cnt) / (j_diff / HZ),
                    pg_fault_max
                    );

            last_pgalloc_normal = pgalloc_normal;
            last_pg_fault_cnt   = pg_fault_cnt;

            pcpuidle->fullrun = 0;
            pcpuidle->fulldisp_jiffies = jiffies;
        }
        /*--- printk("<R>"); ---*/
    }
#endif/*--- #ifdef CONFIG_AVM_POWERMETER ---*/
    return 1;
}
/*--------------------------------------------------------------------------------*\
 * trigger powermanagment to inform about idle-rate
\*--------------------------------------------------------------------------------*/
void avm_cpu_wait_info(void) {
    unsigned int cpu =  raw_smp_processor_id();
    struct _cpu_idlecount *pcpuidle = &cpu_idlecount[cpu];
    register unsigned int sum_count =  pcpuidle->sum_count;
    unsigned int cpu_run = 0;
    if(sum_count > (1 << 29)) {
#if defined(CONFIG_AVM_POWERMETER)
        unsigned int i;
        int p_idle = ((pcpuidle->wait_count >> 8) * 100) / (pcpuidle->sum_count >> 8);
        pcpuidle->actrun = max(0, 100 - p_idle);

        for(i = 0; i < NR_CPUS; i++) {
            struct _cpu_idlecount *pcpuidle = &cpu_idlecount[i];
            cpu_run |= pcpuidle->actrun << (i * 8);
        }
        PowerManagmentRessourceInfo(powerdevice_loadrate, cpu_run);
#endif/*--- #ifdef CONFIG_AVM_POWERMETER ---*/
#if defined(CONFIG_SMP) & 0
        {
        int p_run  = ((pcpuidle->run_count >> 8) * 100) / (sum_count >> 8);
        int p_idle = ((pcpuidle->wait_count >> 8) * 100) / (pcpuidle->sum_count >> 8);
        printk(KERN_INFO"[idle:cpu%u]: %u%% (run %u%%) cpu_run %x\n", cpu, p_idle, p_run, cpu_run);
        }
#endif/*--- #if defined(CONFIG_SMP) ---*/
        pcpuidle->sum_count  >>= 8;
        pcpuidle->wait_count >>= 8;
        pcpuidle->run_count  >>= 8;
    }
}
/*--------------------------------------------------------------------------------*\
\*--------------------------------------------------------------------------------*/
int __init avm_cpu_process_init(void) {
    gCycle_per_10sec = (unsigned long long)avm_get_cyclefreq() * 10ULL;
    return 0;
}
__initcall(avm_cpu_process_init);