diff -Nru linux-2.6.18.5.ori/arch/i386/Kconfig linux-2.6.18.5/arch/i386/Kconfig --- linux-2.6.18.5.ori/arch/i386/Kconfig 2006-12-02 01:13:05.000000000 +0100 +++ linux-2.6.18.5/arch/i386/Kconfig 2006-12-05 03:20:51.000000000 +0100 @@ -737,6 +737,39 @@ If unsure, say Y. Only embedded should say N here. +config EXECTIME_X86 + bool "ExecTime drivers" + depends on EXECTIME + default y + help + Choose Y here to show a list of available x86 hardware drivers + for ExecTime performance measurement system. + +config EXECTIME_X86_P4 + tristate "Pentium4 performance counters" + depends on EXECTIME_X86 && !SMP + default y + help + ExecTime hardware driver using Pentium4 performance counters. + +config EXECTIME_X86_I686 + tristate "Intel P6 family performance counters" + depends on EXECTIME_X86 && !SMP + default y + help + ExecTime hardware driver using Intel P6 family processor + performance counters. + +config EXECTIME_X86_AMD + tristate "AMD K7/K8 performance counters" + depends on EXECTIME_X86 && !SMP + default y + help + ExecTime hardware driver using AMD K7/K8 performance counters. + +comment "x86 ExecTime drivers requires !SMP" + depends on EXECTIME_X86 && SMP + source kernel/Kconfig.hz config KEXEC diff -Nru linux-2.6.18.5.ori/arch/i386/kernel/Makefile linux-2.6.18.5/arch/i386/kernel/Makefile --- linux-2.6.18.5.ori/arch/i386/kernel/Makefile 2006-12-02 01:13:05.000000000 +0100 +++ linux-2.6.18.5/arch/i386/kernel/Makefile 2006-12-05 03:21:30.000000000 +0100 @@ -39,6 +39,9 @@ obj-$(CONFIG_EARLY_PRINTK) += early_printk.o obj-$(CONFIG_HPET_TIMER) += hpet.o obj-$(CONFIG_K8_NB) += k8.o +obj-$(CONFIG_EXECTIME_X86_P4) += exectime-p4.o +obj-$(CONFIG_EXECTIME_X86_I686) += exectime-i686.o +obj-$(CONFIG_EXECTIME_X86_AMD) += exectime-amd.o EXTRA_AFLAGS := -traditional diff -Nru linux-2.6.18.5.ori/arch/i386/kernel/exectime-amd.c linux-2.6.18.5/arch/i386/kernel/exectime-amd.c --- linux-2.6.18.5.ori/arch/i386/kernel/exectime-amd.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.18.5/arch/i386/kernel/exectime-amd.c 2006-12-05 03:20:51.000000000 +0100 @@ -0,0 +1,81 @@ +/* + * linux/arch/i386/kernel/exectime-amd.c + * + * Copyright (C) 2006 Szymon Acedanski + * + * ExecTime driver using P6 family performance counters. + */ + +#include +#include +#include +#include +#include + +MODULE_AUTHOR("Szymon Acedanski"); +MODULE_DESCRIPTION("AMD K7/K8 ExecTime driver"); +MODULE_LICENSE("GPL"); + +static int method; +module_param(method, int, 0444); +MODULE_PARM_DESC(method, "Counting method: 1 - instr_retired, 2 - uop_retired"); + +static int setup(void) +{ + struct cpuinfo_x86 *c = cpu_data; + + if (!cpu_has(c, X86_FEATURE_K7) && !cpu_has(c, X86_FEATURE_K8)) + goto notavail; + + if (method == 2) { + printk(KERN_INFO "exectime-amd: using UOPS_RETIRED event counter\n"); + wrmsr(MSR_K7_EVNTSEL0, 0x004100c1UL, 0); + } else { + printk(KERN_INFO "exectime-amd: using INST_RETIRED event counter\n"); + wrmsr(MSR_K7_EVNTSEL0, 0x004100c0UL, 0); + } + + return 0; + +notavail: + printk(KERN_INFO + "exectime-amd: AMD performance monitoring not available\n"); + return -ENODEV; +} + +static unsigned long long read_counter(void) +{ + unsigned long a[2]; + rdmsr(MSR_K7_PERFCTR0, a[0], a[1]); + return *(unsigned long long*)a; +} + +static void on_switch(struct task_struct* prev, struct task_struct* next) +{ + unsigned long long now = read_counter(); + prev->exectime += now - prev->exectime_u0; + /* XXX: I'm not sure if the counter is 40-bit on AMD processors. */ + if (now < prev->exectime_u0) /* 40-bit counter overflow */ + prev->exectime += 0x00000100000000000ULL; + next->exectime_u0 = now; +} + +static struct exectime_module module = { + .name = "amd", + .setup = setup, + .on_switch = on_switch, +}; + +int __init exectime_amd_init(void) +{ + return exectime_register_module(&module); +} + +void __exit exectime_amd_exit(void) +{ + exectime_unregister_module(); +} + +module_init(exectime_amd_init); +module_exit(exectime_amd_exit); + diff -Nru linux-2.6.18.5.ori/arch/i386/kernel/exectime-i686.c linux-2.6.18.5/arch/i386/kernel/exectime-i686.c --- linux-2.6.18.5.ori/arch/i386/kernel/exectime-i686.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.18.5/arch/i386/kernel/exectime-i686.c 2006-12-05 03:20:51.000000000 +0100 @@ -0,0 +1,80 @@ +/* + * linux/arch/i386/kernel/exectime-i686.c + * + * Copyright (C) 2006 Szymon Acedanski + * + * ExecTime driver using P6 family performance counters. + */ + +#include +#include +#include +#include +#include + +MODULE_AUTHOR("Szymon Acedanski"); +MODULE_DESCRIPTION("Intel P6 family ExecTime driver"); +MODULE_LICENSE("GPL"); + +static int method; +module_param(method, int, 0444); +MODULE_PARM_DESC(method, "Counting method: 1 - instr_retired, 2 - uop_retired"); + +static int setup(void) +{ + struct cpuinfo_x86 *c = cpu_data; + + if (c->x86_vendor != X86_VENDOR_INTEL || c->x86 != 6) + goto notavail; + + if (method == 2) { + printk(KERN_INFO "exectime-i686: using UOPS_RETIRED event counter\n"); + wrmsr(MSR_P6_EVNTSEL0, 0x004100c2UL, 0); + } else { + printk(KERN_INFO "exectime-i686: using INST_RETIRED event counter\n"); + wrmsr(MSR_P6_EVNTSEL0, 0x004100c0UL, 0); + } + + return 0; + +notavail: + printk(KERN_INFO + "exectime-i686: P6 family performance monitoring not available\n"); + return -ENODEV; +} + +static unsigned long long read_counter(void) +{ + unsigned long a[2]; + rdmsr(MSR_P6_PERFCTR0, a[0], a[1]); + return *(unsigned long long*)a; +} + +static void on_switch(struct task_struct* prev, struct task_struct* next) +{ + unsigned long long now = read_counter(); + prev->exectime += now - prev->exectime_u0; + if (now < prev->exectime_u0) /* 40-bit counter overflow */ + prev->exectime += 0x00000100000000000ULL; + next->exectime_u0 = now; +} + +static struct exectime_module module = { + .name = "i686", + .setup = setup, + .on_switch = on_switch, +}; + +int __init exectime_i686_init(void) +{ + return exectime_register_module(&module); +} + +void __exit exectime_i686_exit(void) +{ + exectime_unregister_module(); +} + +module_init(exectime_i686_init); +module_exit(exectime_i686_exit); + diff -Nru linux-2.6.18.5.ori/arch/i386/kernel/exectime-p4.c linux-2.6.18.5/arch/i386/kernel/exectime-p4.c --- linux-2.6.18.5.ori/arch/i386/kernel/exectime-p4.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.18.5/arch/i386/kernel/exectime-p4.c 2006-12-05 03:20:51.000000000 +0100 @@ -0,0 +1,95 @@ +/* + * linux/arch/i386/kernel/exectime-p4.c + * + * Copyright (C) 2006 Szymon Acedanski + * + * ExecTime driver using P4 performance counters. + */ + +#include +#include +#include +#include +#include + +MODULE_AUTHOR("Szymon Acedanski"); +MODULE_DESCRIPTION("Pentium4 ExecTime driver"); +MODULE_LICENSE("GPL"); + +static int method; +module_param(method, int, 0444); +MODULE_PARM_DESC(method, "Counting method: 0 - instr_completed, 1 - instr_retired, 2 - uop_retired"); + +static int setup(void) +{ + struct cpuinfo_x86 *c = cpu_data; + unsigned long feat; + + if (c->x86_vendor != X86_VENDOR_INTEL || c->x86 < 15) + goto notavail; + + rdmsrl(MSR_IA32_MISC_ENABLE, feat); + if (!(feat & (1UL << 7))) + goto notavail; + + if ((c->x86_model == 3 || c->x86_model == 4) && method == 0) { + printk(KERN_INFO "exectime-p4: using instr_completed event counter\n"); + wrmsr(MSR_P4_CRU_ESCR0, 0x07000204UL, 0); + wrmsr(MSR_P4_IQ_CCCR0, 0x00039000UL, 0); + } else if (method == 2) { + printk(KERN_INFO "exectime-p4: using uop_retired event counter\n"); + wrmsr(MSR_P4_CRU_ESCR0, 0x02000204UL, 0); + wrmsr(MSR_P4_IQ_CCCR0, 0x00039000UL, 0); + + } else { + printk(KERN_INFO "exectime-p4: using instr_retired event counter\n"); + + /* program instr_retired counter for non-bogus non-tagged uops, + USR mode counting. */ + wrmsr(MSR_P4_CRU_ESCR0, 0x04000204UL, 0); + wrmsr(MSR_P4_IQ_CCCR0, 0x00039000UL, 0); + } + + return 0; + +notavail: + printk(KERN_INFO + "exectime-p4: P4 performance monitoring not available\n"); + return -ENODEV; +} + +static unsigned long long read_counter(void) +{ + unsigned long a[2]; + rdmsr(MSR_P4_IQ_PERFCTR0, a[0], a[1]); + return *(unsigned long long*)a; +} + +static void on_switch(struct task_struct* prev, struct task_struct* next) +{ + unsigned long long now = read_counter(); + prev->exectime += now - prev->exectime_u0; + if (now < prev->exectime_u0) /* 40-bit counter overflow */ + prev->exectime += 0x00000100000000000ULL; + next->exectime_u0 = now; +} + +static struct exectime_module module = { + .name = "pentium4", + .setup = setup, + .on_switch = on_switch, +}; + +int __init exectime_p4_init(void) +{ + return exectime_register_module(&module); +} + +void __exit exectime_p4_exit(void) +{ + exectime_unregister_module(); +} + +module_init(exectime_p4_init); +module_exit(exectime_p4_exit); + diff -Nru linux-2.6.18.5.ori/fs/proc/base.c linux-2.6.18.5/fs/proc/base.c --- linux-2.6.18.5.ori/fs/proc/base.c 2006-12-02 01:13:05.000000000 +0100 +++ linux-2.6.18.5/fs/proc/base.c 2006-12-05 03:20:51.000000000 +0100 @@ -138,6 +138,9 @@ #endif PROC_TGID_OOM_SCORE, PROC_TGID_OOM_ADJUST, +#ifdef CONFIG_EXECTIME + PROC_TGID_EXECTIME, +#endif PROC_TID_INO, PROC_TID_STATUS, PROC_TID_MEM, @@ -181,9 +184,12 @@ #endif PROC_TID_OOM_SCORE, PROC_TID_OOM_ADJUST, - - /* Add new entries before this */ - PROC_TID_FD_DIR = 0x8000, /* 0x8000-0xffff */ +#ifdef CONFIG_EXECTIME + PROC_TID_EXECTIME, +#endif + + /* Add new entries before this */ + PROC_TID_FD_DIR = 0x8000, /* 0x8000-0xffff */ }; /* Worst case buffer size needed for holding an integer. */ @@ -240,6 +246,9 @@ #ifdef CONFIG_AUDITSYSCALL E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO), #endif +#ifdef CONFIG_EXECTIME + E(PROC_TGID_EXECTIME, "exectime", S_IFREG|S_IRUGO), +#endif {0,0,NULL,0} }; static struct pid_entry tid_base_stuff[] = { @@ -282,6 +291,10 @@ #ifdef CONFIG_AUDITSYSCALL E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO), #endif +#ifdef CONFIG_EXECTIME + E(PROC_TID_EXECTIME, "exectime", S_IFREG|S_IRUGO), +#endif + {0,0,NULL,0} }; @@ -1064,6 +1077,32 @@ }; #endif /* CONFIG_SECCOMP */ +#ifdef CONFIG_EXECTIME +static ssize_t exectime_read(struct file *file, char __user *buf, + size_t count, loff_t *ppos) +{ + struct task_struct *tsk = get_proc_task(file->f_dentry->d_inode); + char __buf[20]; + loff_t __ppos = *ppos; + size_t len; + + /* no need to print the trailing zero, so use only len */ + len = sprintf(__buf, "%llu\n", tsk->exectime); + if (__ppos >= len) + return 0; + if (count > len - __ppos) + count = len - __ppos; + if (copy_to_user(buf, __buf + __ppos, count)) + return -EFAULT; + *ppos = __ppos + count; + return count; +} + +static struct file_operations proc_exectime_operations = { + .read = exectime_read, +}; +#endif /* CONFIG_EXECTIME */ + static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd) { struct inode *inode = dentry->d_inode; @@ -1834,6 +1873,12 @@ inode->i_fop = &proc_loginuid_operations; break; #endif +#ifdef CONFIG_EXECTIME + case PROC_TGID_EXECTIME: + case PROC_TID_EXECTIME: + inode->i_fop = &proc_exectime_operations; + break; +#endif default: printk("procfs: impossible type (%d)",p->type); iput(inode); diff -Nru linux-2.6.18.5.ori/include/linux/exectime.h linux-2.6.18.5/include/linux/exectime.h --- linux-2.6.18.5.ori/include/linux/exectime.h 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.18.5/include/linux/exectime.h 2006-12-05 03:20:51.000000000 +0100 @@ -0,0 +1,18 @@ +#ifndef __LINUX_EXECTIME_H +#define __LINUX_EXECTIME_H + +struct task_struct; + +struct exectime_module { + const char* name; + int (*setup)(void); + void (*on_switch)(struct task_struct* prev, struct task_struct* next); + void (*on_fork)(struct task_struct* p); +}; + +int exectime_register_module(struct exectime_module* m); +void exectime_unregister_module(void); +void exectime_on_switch(struct task_struct* prev, struct task_struct* next); +void exectime_on_fork(struct task_struct* p); + +#endif diff -Nru linux-2.6.18.5.ori/include/linux/sched.h linux-2.6.18.5/include/linux/sched.h --- linux-2.6.18.5.ori/include/linux/sched.h 2006-12-02 01:13:05.000000000 +0100 +++ linux-2.6.18.5/include/linux/sched.h 2006-12-05 03:20:51.000000000 +0100 @@ -915,6 +915,11 @@ /* Deadlock detection and priority inheritance handling */ struct rt_mutex_waiter *pi_blocked_on; #endif +#ifdef CONFIG_EXECTIME + u64 exectime; + u64 exectime_u0; // for module use + u64 exectime_u1; // for module use +#endif #ifdef CONFIG_DEBUG_MUTEXES /* mutex deadlock detection */ diff -Nru linux-2.6.18.5.ori/init/Kconfig linux-2.6.18.5/init/Kconfig --- linux-2.6.18.5.ori/init/Kconfig 2006-12-02 01:13:05.000000000 +0100 +++ linux-2.6.18.5/init/Kconfig 2006-12-05 03:20:51.000000000 +0100 @@ -273,6 +273,15 @@ help This enables the legacy 16-bit UID syscall wrappers. +config EXECTIME + bool "ExecTime Process Accounting" + default n + help + If you say Y here, each process's execution time will be measured + using the new ExecTime technology. The result may be used by + userspace applications for accurate evaluation of contestants + programs of a programming competition. + config SYSCTL bool "Sysctl support" if EMBEDDED default y diff -Nru linux-2.6.18.5.ori/kernel/Makefile linux-2.6.18.5/kernel/Makefile --- linux-2.6.18.5.ori/kernel/Makefile 2006-12-02 01:13:05.000000000 +0100 +++ linux-2.6.18.5/kernel/Makefile 2006-12-05 03:20:51.000000000 +0100 @@ -46,6 +46,7 @@ obj-$(CONFIG_DETECT_SOFTLOCKUP) += softlockup.o obj-$(CONFIG_GENERIC_HARDIRQS) += irq/ obj-$(CONFIG_SECCOMP) += seccomp.o +obj-$(CONFIG_EXECTIME) += exectime.o obj-$(CONFIG_RCU_TORTURE_TEST) += rcutorture.o obj-$(CONFIG_RELAY) += relay.o obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o diff -Nru linux-2.6.18.5.ori/kernel/exectime.c linux-2.6.18.5/kernel/exectime.c --- linux-2.6.18.5.ori/kernel/exectime.c 1970-01-01 01:00:00.000000000 +0100 +++ linux-2.6.18.5/kernel/exectime.c 2006-12-05 03:20:51.000000000 +0100 @@ -0,0 +1,70 @@ +/* + * linux/kernel/exectime.c + * + * Copyright (C) 2006 Szymon Acedanski + * + * Precise ExecTime application performance measurement. + */ + +#include +#include +#include +#include + + +/* Dummy module */ + +static void dummy_on_switch(struct task_struct* prev, struct task_struct* next) +{ +} + +static struct exectime_module dummy_module = { + .on_switch = dummy_on_switch, +}; + +/* Current module */ + +static struct exectime_module* current_module = &dummy_module; +static int is_active; + +/* Hooks */ + +void exectime_on_switch(struct task_struct* prev, struct task_struct* next) +{ + current_module->on_switch(prev, next); +} + +void exectime_on_fork(struct task_struct* p) +{ + p->exectime = 0; + if (current_module->on_fork) + current_module->on_fork(p); +} + +int exectime_register_module(struct exectime_module* mod) +{ + int ret; + if (is_active) + return -EBUSY; + ret = mod->setup(); + if (ret < 0) + return ret; + + /* NOTE: this is racy, but we do not want to be too pedantic. */ + + is_active = 1; + current_module = mod; + printk(KERN_INFO "exectime: activated module '%s'\n", mod->name); + return 0; +} + +void exectime_unregister_module(void) +{ + current_module = &dummy_module; + is_active = 0; + printk(KERN_INFO "exectime: deactivated\n"); +} + +EXPORT_SYMBOL(exectime_register_module); +EXPORT_SYMBOL(exectime_unregister_module); + diff -Nru linux-2.6.18.5.ori/kernel/sched.c linux-2.6.18.5/kernel/sched.c --- linux-2.6.18.5.ori/kernel/sched.c 2006-12-02 01:13:05.000000000 +0100 +++ linux-2.6.18.5/kernel/sched.c 2006-12-05 03:20:51.000000000 +0100 @@ -53,6 +53,7 @@ #include #include #include +#include #include @@ -1836,6 +1837,10 @@ spin_release(&rq->lock.dep_map, 1, _THIS_IP_); #endif +#ifdef CONFIG_EXECTIME + exectime_on_switch(prev, next); +#endif + /* Here we just switch the register state and the stack. */ switch_to(prev, next, prev);