perfmon.c

xen虚拟机源代码安装包

	ibr_mask_reg_t ibr;
	dbr_mask_reg_t dbr;
} dbreg_t;


/*
 * perfmon command descriptions
 */
typedef struct {
	int		(*cmd_func)(pfm_context_t *ctx, void *arg, int count, struct pt_regs *regs);
	char		*cmd_name;
	int		cmd_flags;
	unsigned int	cmd_narg;
	size_t		cmd_argsize;
	int		(*cmd_getsize)(void *arg, size_t *sz);
} pfm_cmd_desc_t;

#define PFM_CMD_FD		0x01	/* command requires a file descriptor */
#define PFM_CMD_ARG_READ	0x02	/* command must read argument(s) */
#define PFM_CMD_ARG_RW		0x04	/* command must read/write argument(s) */
#define PFM_CMD_STOP		0x08	/* command does not work on zombie context */


#define PFM_CMD_NAME(cmd)	pfm_cmd_tab[(cmd)].cmd_name
#define PFM_CMD_READ_ARG(cmd)	(pfm_cmd_tab[(cmd)].cmd_flags & PFM_CMD_ARG_READ)
#define PFM_CMD_RW_ARG(cmd)	(pfm_cmd_tab[(cmd)].cmd_flags & PFM_CMD_ARG_RW)
#define PFM_CMD_USE_FD(cmd)	(pfm_cmd_tab[(cmd)].cmd_flags & PFM_CMD_FD)
#define PFM_CMD_STOPPED(cmd)	(pfm_cmd_tab[(cmd)].cmd_flags & PFM_CMD_STOP)

#define PFM_CMD_ARG_MANY	-1 /* cannot be zero */

typedef struct {
	unsigned long pfm_spurious_ovfl_intr_count;	/* keep track of spurious ovfl interrupts */
	unsigned long pfm_replay_ovfl_intr_count;	/* keep track of replayed ovfl interrupts */
	unsigned long pfm_ovfl_intr_count; 		/* keep track of ovfl interrupts */
	unsigned long pfm_ovfl_intr_cycles;		/* cycles spent processing ovfl interrupts */
	unsigned long pfm_ovfl_intr_cycles_min;		/* min cycles spent processing ovfl interrupts */
	unsigned long pfm_ovfl_intr_cycles_max;		/* max cycles spent processing ovfl interrupts */
	unsigned long pfm_smpl_handler_calls;
	unsigned long pfm_smpl_handler_cycles;
	char pad[SMP_CACHE_BYTES] ____cacheline_aligned;
} pfm_stats_t;

/*
 * perfmon internal variables
 */
static pfm_stats_t		pfm_stats[NR_CPUS];
static pfm_session_t		pfm_sessions;	/* global sessions information */

#ifndef XEN
static DEFINE_SPINLOCK(pfm_alt_install_check);
#endif
static pfm_intr_handler_desc_t  *pfm_alt_intr_handler;

#ifndef XEN
static struct proc_dir_entry 	*perfmon_dir;
#endif
static pfm_uuid_t		pfm_null_uuid = {0,};

static spinlock_t		pfm_buffer_fmt_lock;
static LIST_HEAD(pfm_buffer_fmt_list);

static pmu_config_t		*pmu_conf;

/* sysctl() controls */
pfm_sysctl_t pfm_sysctl;
EXPORT_SYMBOL(pfm_sysctl);

#ifndef XEN
static ctl_table pfm_ctl_table[]={
	{1, "debug", &pfm_sysctl.debug, sizeof(int), 0666, NULL, &proc_dointvec, NULL,},
	{2, "debug_ovfl", &pfm_sysctl.debug_ovfl, sizeof(int), 0666, NULL, &proc_dointvec, NULL,},
	{3, "fastctxsw", &pfm_sysctl.fastctxsw, sizeof(int), 0600, NULL, &proc_dointvec, NULL,},
	{4, "expert_mode", &pfm_sysctl.expert_mode, sizeof(int), 0600, NULL, &proc_dointvec, NULL,},
	{ 0, },
};
static ctl_table pfm_sysctl_dir[] = {
	{1, "perfmon", NULL, 0, 0755, pfm_ctl_table, },
 	{0,},
};
static ctl_table pfm_sysctl_root[] = {
	{1, "kernel", NULL, 0, 0755, pfm_sysctl_dir, },
 	{0,},
};
static struct ctl_table_header *pfm_sysctl_header;

static int pfm_context_unload(pfm_context_t *ctx, void *arg, int count, struct pt_regs *regs);
static int pfm_flush(struct file *filp);
#endif

#define pfm_get_cpu_var(v)		__ia64_per_cpu_var(v)
#define pfm_get_cpu_data(a,b)		per_cpu(a, b)

#ifndef XEN
static inline void
pfm_put_task(struct task_struct *task)
{
	if (task != current) put_task_struct(task);
}

static inline void
pfm_set_task_notify(struct task_struct *task)
{
	struct thread_info *info;

	info = (struct thread_info *) ((char *) task + IA64_TASK_SIZE);
	set_bit(TIF_NOTIFY_RESUME, &info->flags);
}

static inline void
pfm_clear_task_notify(void)
{
	clear_thread_flag(TIF_NOTIFY_RESUME);
}

static inline void
pfm_reserve_page(unsigned long a)
{
	SetPageReserved(vmalloc_to_page((void *)a));
}
static inline void
pfm_unreserve_page(unsigned long a)
{
	ClearPageReserved(vmalloc_to_page((void*)a));
}
#endif

static inline unsigned long
pfm_protect_ctx_ctxsw(pfm_context_t *x)
{
	spin_lock(&(x)->ctx_lock);
	return 0UL;
}

static inline void
pfm_unprotect_ctx_ctxsw(pfm_context_t *x, unsigned long f)
{
	spin_unlock(&(x)->ctx_lock);
}

#ifndef XEN
static inline unsigned int
pfm_do_munmap(struct mm_struct *mm, unsigned long addr, size_t len, int acct)
{
	return do_munmap(mm, addr, len);
}

static inline unsigned long 
pfm_get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, unsigned long pgoff, unsigned long flags, unsigned long exec)
{
	return get_unmapped_area(file, addr, len, pgoff, flags);
}


static struct super_block *
pfmfs_get_sb(struct file_system_type *fs_type, int flags, const char *dev_name, void *data)
{
	return get_sb_pseudo(fs_type, "pfm:", NULL, PFMFS_MAGIC);
}

static struct file_system_type pfm_fs_type = {
	.name     = "pfmfs",
	.get_sb   = pfmfs_get_sb,
	.kill_sb  = kill_anon_super,
};
#endif

DEFINE_PER_CPU(unsigned long, pfm_syst_info);
DEFINE_PER_CPU(struct task_struct *, pmu_owner);
DEFINE_PER_CPU(pfm_context_t  *, pmu_ctx);
DEFINE_PER_CPU(unsigned long, pmu_activation_number);
#ifndef XEN
EXPORT_PER_CPU_SYMBOL_GPL(pfm_syst_info);


/* forward declaration */
static struct file_operations pfm_file_ops;
#endif

/*
 * forward declarations
 */
#ifndef CONFIG_SMP
static void pfm_lazy_save_regs (struct task_struct *ta);
#endif

void dump_pmu_state(const char *);
static int pfm_write_ibr_dbr(int mode, pfm_context_t *ctx, void *arg, int count, struct pt_regs *regs);

#include "perfmon_itanium.h"
#include "perfmon_mckinley.h"
#include "perfmon_montecito.h"
#include "perfmon_generic.h"

static pmu_config_t *pmu_confs[]={
	&pmu_conf_mont,
	&pmu_conf_mck,
	&pmu_conf_ita,
	&pmu_conf_gen, /* must be last */
	NULL
};


#ifndef XEN
static int pfm_end_notify_user(pfm_context_t *ctx);
#endif

static inline void
pfm_clear_psr_pp(void)
{
	ia64_rsm(IA64_PSR_PP);
	ia64_srlz_i();
}

static inline void
pfm_set_psr_pp(void)
{
	ia64_ssm(IA64_PSR_PP);
	ia64_srlz_i();
}

static inline void
pfm_clear_psr_up(void)
{
	ia64_rsm(IA64_PSR_UP);
	ia64_srlz_i();
}

static inline void
pfm_set_psr_up(void)
{
	ia64_ssm(IA64_PSR_UP);
	ia64_srlz_i();
}

static inline unsigned long
pfm_get_psr(void)
{
	unsigned long tmp;
	tmp = ia64_getreg(_IA64_REG_PSR);
	ia64_srlz_i();
	return tmp;
}

static inline void
pfm_set_psr_l(unsigned long val)
{
	ia64_setreg(_IA64_REG_PSR_L, val);
	ia64_srlz_i();
}

static inline void
pfm_freeze_pmu(void)
{
	ia64_set_pmc(0,1UL);
	ia64_srlz_d();
}

static inline void
pfm_unfreeze_pmu(void)
{
	ia64_set_pmc(0,0UL);
	ia64_srlz_d();
}

static inline void
pfm_restore_ibrs(unsigned long *ibrs, unsigned int nibrs)
{
	int i;

	for (i=0; i < nibrs; i++) {
		ia64_set_ibr(i, ibrs[i]);
		ia64_dv_serialize_instruction();
	}
	ia64_srlz_i();
}

static inline void
pfm_restore_dbrs(unsigned long *dbrs, unsigned int ndbrs)
{
	int i;

	for (i=0; i < ndbrs; i++) {
		ia64_set_dbr(i, dbrs[i]);
		ia64_dv_serialize_data();
	}
	ia64_srlz_d();
}

/*
 * PMD[i] must be a counter. no check is made
 */
static inline unsigned long
pfm_read_soft_counter(pfm_context_t *ctx, int i)
{
	return ctx->ctx_pmds[i].val + (ia64_get_pmd(i) & pmu_conf->ovfl_val);
}

/*
 * PMD[i] must be a counter. no check is made
 */
static inline void
pfm_write_soft_counter(pfm_context_t *ctx, int i, unsigned long val)
{
	unsigned long ovfl_val = pmu_conf->ovfl_val;

	ctx->ctx_pmds[i].val = val  & ~ovfl_val;
	/*
	 * writing to unimplemented part is ignore, so we do not need to
	 * mask off top part
	 */
	ia64_set_pmd(i, val & ovfl_val);
}

#ifndef XEN
static pfm_msg_t *
pfm_get_new_msg(pfm_context_t *ctx)
{
	int idx, next;

	next = (ctx->ctx_msgq_tail+1) % PFM_MAX_MSGS;

	DPRINT(("ctx_fd=%p head=%d tail=%d\n", ctx, ctx->ctx_msgq_head, ctx->ctx_msgq_tail));
	if (next == ctx->ctx_msgq_head) return NULL;

 	idx = 	ctx->ctx_msgq_tail;
	ctx->ctx_msgq_tail = next;

	DPRINT(("ctx=%p head=%d tail=%d msg=%d\n", ctx, ctx->ctx_msgq_head, ctx->ctx_msgq_tail, idx));

	return ctx->ctx_msgq+idx;
}

static pfm_msg_t *
pfm_get_next_msg(pfm_context_t *ctx)
{
	pfm_msg_t *msg;

	DPRINT(("ctx=%p head=%d tail=%d\n", ctx, ctx->ctx_msgq_head, ctx->ctx_msgq_tail));

	if (PFM_CTXQ_EMPTY(ctx)) return NULL;

	/*
	 * get oldest message
	 */
	msg = ctx->ctx_msgq+ctx->ctx_msgq_head;

	/*
	 * and move forward
	 */
	ctx->ctx_msgq_head = (ctx->ctx_msgq_head+1) % PFM_MAX_MSGS;

	DPRINT(("ctx=%p head=%d tail=%d type=%d\n", ctx, ctx->ctx_msgq_head, ctx->ctx_msgq_tail, msg->pfm_gen_msg.msg_type));

	return msg;
}

static void
pfm_reset_msgq(pfm_context_t *ctx)
{
	ctx->ctx_msgq_head = ctx->ctx_msgq_tail = 0;
	DPRINT(("ctx=%p msgq reset\n", ctx));
}

static void *
pfm_rvmalloc(unsigned long size)
{
	void *mem;
	unsigned long addr;

	size = PAGE_ALIGN(size);
	mem  = vmalloc(size);
	if (mem) {
		//printk("perfmon: CPU%d pfm_rvmalloc(%ld)=%p\n", smp_processor_id(), size, mem);
		memset(mem, 0, size);
		addr = (unsigned long)mem;
		while (size > 0) {
			pfm_reserve_page(addr);
			addr+=PAGE_SIZE;
			size-=PAGE_SIZE;
		}
	}
	return mem;
}

static void
pfm_rvfree(void *mem, unsigned long size)
{
	unsigned long addr;

	if (mem) {
		DPRINT(("freeing physical buffer @%p size=%lu\n", mem, size));
		addr = (unsigned long) mem;
		while ((long) size > 0) {
			pfm_unreserve_page(addr);
			addr+=PAGE_SIZE;
			size-=PAGE_SIZE;
		}
		vfree(mem);
	}
	return;
}
#endif

static pfm_context_t *
pfm_context_alloc(void)
{
	pfm_context_t *ctx;

	/* 
	 * allocate context descriptor 
	 * must be able to free with interrupts disabled
	 */
	ctx = kmalloc(sizeof(pfm_context_t), GFP_KERNEL);
	if (ctx) {
		memset(ctx, 0, sizeof(pfm_context_t));
		DPRINT(("alloc ctx @%p\n", ctx));
	}
	return ctx;
}

static void
pfm_context_free(pfm_context_t *ctx)
{
	if (ctx) {
		DPRINT(("free ctx @%p\n", ctx));
		kfree(ctx);
	}
}

#ifndef XEN
static void
pfm_mask_monitoring(struct task_struct *task)
{
	pfm_context_t *ctx = PFM_GET_CTX(task);
	struct thread_struct *th = &task->thread;
	unsigned long mask, val, ovfl_mask;
	int i;

	DPRINT_ovfl(("masking monitoring for [%d]\n", task->pid));

	ovfl_mask = pmu_conf->ovfl_val;
	/*
	 * monitoring can only be masked as a result of a valid
	 * counter overflow. In UP, it means that the PMU still
	 * has an owner. Note that the owner can be different
	 * from the current task. However the PMU state belongs
	 * to the owner.
	 * In SMP, a valid overflow only happens when task is
	 * current. Therefore if we come here, we know that
	 * the PMU state belongs to the current task, therefore
	 * we can access the live registers.
	 *
	 * So in both cases, the live register contains the owner's
	 * state. We can ONLY touch the PMU registers and NOT the PSR.
	 *
	 * As a consequence to this call, the thread->pmds[] array
	 * contains stale information which must be ignored
	 * when context is reloaded AND monitoring is active (see
	 * pfm_restart).
	 */
	mask = ctx->ctx_used_pmds[0];
	for (i = 0; mask; i++, mask>>=1) {
		/* skip non used pmds */
		if ((mask & 0x1) == 0) continue;
		val = ia64_get_pmd(i);

		if (PMD_IS_COUNTING(i)) {
			/*
		 	 * we rebuild the full 64 bit value of the counter
		 	 */
			ctx->ctx_pmds[i].val += (val & ovfl_mask);
		} else {
			ctx->ctx_pmds[i].val = val;
		}
		DPRINT_ovfl(("pmd[%d]=0x%lx hw_pmd=0x%lx\n",
			i,
			ctx->ctx_pmds[i].val,
			val & ovfl_mask));
	}
	/*
	 * mask monitoring by setting the privilege level to 0
	 * we cannot use psr.pp/psr.up for this, it is controlled by
	 * the user
	 *
	 * if task is current, modify actual registers, otherwise modify
	 * thread save state, i.e., what will be restored in pfm_load_regs()
	 */
	mask = ctx->ctx_used_monitors[0] >> PMU_FIRST_COUNTER;
	for(i= PMU_FIRST_COUNTER; mask; i++, mask>>=1) {
		if ((mask & 0x1) == 0UL) continue;
		ia64_set_pmc(i, th->pmcs[i] & ~0xfUL);
		th->pmcs[i] &= ~0xfUL;
		DPRINT_ovfl(("pmc[%d]=0x%lx\n", i, th->pmcs[i]));
	}
	/*
	 * make all of this visible
	 */
	ia64_srlz_d();
}

/*
 * must always be done with task == current
 *
 * context must be in MASKED state when calling
 */
static void