ia64: perfmon update.

Here is a ChangeLog for the patch:

        - The perfmon core will invoke the sampling module handler
          routine once for each overflowed PMD. When multiple PMDs
          overflow at the same time (with the same PMU interrupt),
          then up to 64 distinct calls can happen. A common timestamp
          parameter allows the module to identify this kind of
          entries.

        - Changed the module ovfl_ctrl arguments to simplify the reset
          field. Now it is a simple boolean.

        - Updated perfmon.h to convert the "set" field to ushort from
          uint.  Other structure updates to get better layout.

        - Update perfmon_default_smpl.h to reflect the change in
          overflow processing mentioned above.

        - Cleanup some state checking code to use switch-case instead
          of if-then with macros. Make the code more readable and
          easier to optmize for gcc. Thanks to David for the
          suggestion.

        - Added extra safety checks on pfm_context_load() to verify
          that the task actually exists.

        - The default sampling format module now supports the
          fmt_restart_active callbacks. Patch from David.

arch/ia64/kernel/perfmon_default_smpl.c

@@ -109,21 +109,15 @@ default_init(struct task_struct *task, void *buf, unsigned int flags, int cpu, v
 }
 
 static int
-default_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct pt_regs *regs)
+default_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct pt_regs *regs, unsigned long stamp)
 {
 	pfm_default_smpl_hdr_t *hdr;
 	pfm_default_smpl_entry_t *ent;
 	void *cur, *last;
 	unsigned long *e;
-	unsigned long ovfl_mask;
-	unsigned long ovfl_notify;
-	unsigned long stamp;
 	unsigned int npmds, i;
-
-	/*
-	 * some time stamp
-	 */
-	stamp = ia64_get_itc();
+	unsigned char ovfl_pmd;
+	unsigned char ovfl_notify;
 
 	if (unlikely(buf == NULL || arg == NULL|| regs == NULL || task == NULL)) {
 		DPRINT(("[%d] invalid arguments buf=%p arg=%p\n", task->pid, buf, arg));
@@ -133,8 +127,8 @@ default_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct
 	hdr         = (pfm_default_smpl_hdr_t *)buf;
 	cur         = hdr->hdr_cur_pos;
 	last        = hdr->hdr_last_pos;
-	ovfl_mask   = arg->ovfl_pmds[0];
-	ovfl_notify = arg->ovfl_notify[0];
+	ovfl_pmd    = arg->ovfl_pmd;
+	ovfl_notify = arg->ovfl_notify;
 
 	/*
 	 * check for space against largest possibly entry.
@@ -153,12 +147,12 @@ default_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct
 
 	hdr->hdr_count++;
 
-	DPRINT_ovfl(("[%d] count=%lu cur=%p last=%p free_bytes=%lu ovfl_pmds=0x%lx ovfl_notify=0x%lx npmds=%u\n",
+	DPRINT_ovfl(("[%d] count=%lu cur=%p last=%p free_bytes=%lu ovfl_pmd=%d ovfl_notify=%d npmds=%u\n",
 			task->pid,
 			hdr->hdr_count,
 			cur, last,
 			last-cur,
-			ovfl_mask,
+			ovfl_pmd,
 			ovfl_notify, npmds));
 
 	/*
@@ -172,7 +166,7 @@ default_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct
 	 * 	- this is not necessarily the task controlling the session
 	 */
 	ent->pid            = current->pid;
-	ent->cpu            = smp_processor_id();
+	ent->ovfl_pmd  	    = ovfl_pmd;
 	ent->last_reset_val = arg->pmd_last_reset; //pmd[0].reg_last_reset_val;
 
 	/*
@@ -180,13 +174,9 @@ default_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct
 	 */
 	ent->ip = regs->cr_iip | ((regs->cr_ipsr >> 41) & 0x3);
 
-	/*
-	 * which registers overflowed
-	 */
-	ent->ovfl_pmds = ovfl_mask;
 	ent->tstamp    = stamp;
+	ent->cpu       = smp_processor_id();
 	ent->set       = arg->active_set;
-	ent->reserved1 = 0;
 
 	/*
 	 * selectively store PMDs in increasing index number
@@ -206,14 +196,14 @@ default_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct
 	/*
 	 * keep same ovfl_pmds, ovfl_notify
 	 */
-	arg->ovfl_ctrl.notify_user     = 0;
-	arg->ovfl_ctrl.block           = 0;
-	arg->ovfl_ctrl.stop_monitoring = 0;
-	arg->ovfl_ctrl.reset_pmds      = 1;
+	arg->ovfl_ctrl.bits.notify_user     = 0;
+	arg->ovfl_ctrl.bits.block_task      = 0;
+	arg->ovfl_ctrl.bits.mask_monitoring = 0;
+	arg->ovfl_ctrl.bits.reset_ovfl_pmds = 1; /* reset before returning from interrupt handler */
 
 	return 0;
 full:
-	DPRINT_ovfl(("sampling buffer full free=%lu, count=%lu, ovfl_notify=0x%lx\n", last-cur, hdr->hdr_count, ovfl_notify));
+	DPRINT_ovfl(("sampling buffer full free=%lu, count=%lu, ovfl_notify=%d\n", last-cur, hdr->hdr_count, ovfl_notify));
 
 	/*
 	 * increment number of buffer overflow.
@@ -222,22 +212,21 @@ default_handler(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct
 	hdr->hdr_overflows++;
 
 	/*
-	 * if no notification is needed, then we just reset the buffer index.
+	 * if no notification is needed, then we saturate the buffer
 	 */
-	if (ovfl_notify == 0UL) {
+	if (ovfl_notify == 0) {
 		hdr->hdr_count = 0UL;
-		arg->ovfl_ctrl.notify_user     = 0;
-		arg->ovfl_ctrl.block           = 0;
-		arg->ovfl_ctrl.stop_monitoring = 0;
-		arg->ovfl_ctrl.reset_pmds      = 1;
+		arg->ovfl_ctrl.bits.notify_user     = 0;
+		arg->ovfl_ctrl.bits.block_task      = 0;
+		arg->ovfl_ctrl.bits.mask_monitoring = 1;
+		arg->ovfl_ctrl.bits.reset_ovfl_pmds = 0;
 	} else {
-		/* keep same ovfl_pmds, ovfl_notify */
-		arg->ovfl_ctrl.notify_user     = 1;
-		arg->ovfl_ctrl.block           = 1;
-		arg->ovfl_ctrl.stop_monitoring = 1;
-		arg->ovfl_ctrl.reset_pmds      = 0;
+		arg->ovfl_ctrl.bits.notify_user     = 1;
+		arg->ovfl_ctrl.bits.block_task      = 1; /* ignored for non-blocking context */
+		arg->ovfl_ctrl.bits.mask_monitoring = 1;
+		arg->ovfl_ctrl.bits.reset_ovfl_pmds = 0; /* no reset now */
 	}
-	return 0;
+	return -1; /* we are full, sorry */
 }
 
 static int
@@ -250,8 +239,8 @@ default_restart(struct task_struct *task, pfm_ovfl_ctrl_t *ctrl, void *buf, stru
 	hdr->hdr_count   = 0UL;
 	hdr->hdr_cur_pos = (void *)((unsigned long)buf)+sizeof(*hdr);
 
-	ctrl->stop_monitoring = 0;
-	ctrl->reset_pmds      = PFM_PMD_LONG_RESET;
+	ctrl->bits.mask_monitoring = 0;
+	ctrl->bits.reset_ovfl_pmds = 1; /* uses long-reset values */
 
 	return 0;
 }
@@ -264,15 +253,16 @@ default_exit(struct task_struct *task, void *buf, struct pt_regs *regs)
 }
 
 static pfm_buffer_fmt_t default_fmt={
-	.fmt_name 	= "default_format",
-	.fmt_uuid	= PFM_DEFAULT_SMPL_UUID,
-	.fmt_arg_size	= sizeof(pfm_default_smpl_arg_t),
-	.fmt_validate	= default_validate,
-	.fmt_getsize	= default_get_size,
-	.fmt_init	= default_init,
-	.fmt_handler	= default_handler,
-	.fmt_restart	= default_restart,
-	.fmt_exit	= default_exit,
+ 	.fmt_name 	    = "default_format",
+ 	.fmt_uuid	    = PFM_DEFAULT_SMPL_UUID,
+ 	.fmt_arg_size	    = sizeof(pfm_default_smpl_arg_t),
+ 	.fmt_validate	    = default_validate,
+ 	.fmt_getsize	    = default_get_size,
+ 	.fmt_init	    = default_init,
+ 	.fmt_handler	    = default_handler,
+ 	.fmt_restart	    = default_restart,
+ 	.fmt_restart_active = default_restart,
+ 	.fmt_exit	    = default_exit,
 };
 
 static int __init

include/asm-ia64/perfmon.h

@@ -70,64 +70,70 @@ typedef unsigned char pfm_uuid_t[16];	/* custom sampling buffer identifier type
  * Request structure used to define a context
  */
 typedef struct {
-	pfm_uuid_t    ctx_smpl_buf_id;	/* which buffer format to use (if needed) */
-	unsigned long ctx_flags;	/* noblock/block */
-	unsigned int  ctx_nextra_sets;	/* number of extra event sets (you always get 1) */
-	int	      ctx_fd;		/* return arg: unique identification for context */
-	void	      *ctx_smpl_vaddr;	/* return arg: virtual address of sampling buffer, is used */
-	unsigned long ctx_reserved[11];	/* for future use */
+	pfm_uuid_t     ctx_smpl_buf_id;	 /* which buffer format to use (if needed) */
+	unsigned long  ctx_flags;	 /* noblock/block */
+	unsigned short ctx_nextra_sets;	 /* number of extra event sets (you always get 1) */
+	unsigned short ctx_reserved1;	 /* for future use */
+	int	       ctx_fd;		 /* return arg: unique identification for context */
+	void	       *ctx_smpl_vaddr;	 /* return arg: virtual address of sampling buffer, is used */
+	unsigned long  ctx_reserved2[11];/* for future use */
 } pfarg_context_t;
 
 /*
  * Request structure used to write/read a PMC or PMD
  */
 typedef struct {
-	unsigned int	reg_num;	   /* which register                             */
-	unsigned int	reg_set;	   /* event set for this register                */
+	unsigned int	reg_num;	   /* which register */
+	unsigned short	reg_set;	   /* event set for this register */
+	unsigned short	reg_reserved1;	   /* for future use */
 
-	unsigned long	reg_value;	   /* initial pmc/pmd value                      */
-	unsigned long	reg_flags;	   /* input: pmc/pmd flags, return: reg error    */
+	unsigned long	reg_value;	   /* initial pmc/pmd value */
+	unsigned long	reg_flags;	   /* input: pmc/pmd flags, return: reg error */
 
-	unsigned long	reg_long_reset;	   /* reset after buffer overflow notification   */
-	unsigned long	reg_short_reset;   /* reset after counter overflow               */
+	unsigned long	reg_long_reset;	   /* reset after buffer overflow notification */
+	unsigned long	reg_short_reset;   /* reset after counter overflow */
 
-	unsigned long	reg_reset_pmds[4]; /* which other counters to reset on overflow  */
-	unsigned long	reg_random_seed;   /* seed value when randomization is used      */
-	unsigned long	reg_random_mask;   /* bitmask used to limit random value         */
-	unsigned long   reg_last_reset_val;/* return: PMD last reset value               */
+	unsigned long	reg_reset_pmds[4]; /* which other counters to reset on overflow */
+	unsigned long	reg_random_seed;   /* seed value when randomization is used */
+	unsigned long	reg_random_mask;   /* bitmask used to limit random value */
+	unsigned long   reg_last_reset_val;/* return: PMD last reset value */
 
 	unsigned long	reg_smpl_pmds[4];  /* which pmds are accessed when PMC overflows */
-	unsigned long	reg_smpl_eventid;  /* opaque sampling event identifier           */
+	unsigned long	reg_smpl_eventid;  /* opaque sampling event identifier */
 
-	unsigned long   reserved[3];	   /* for future use                             */
+	unsigned long   reg_reserved2[3];   /* for future use */
 } pfarg_reg_t;
 
 typedef struct {
-	unsigned int	dbreg_num;		/* which debug register        */
-	unsigned int 	dbreg_set;		/* event set for this register */
-	unsigned long	dbreg_value;		/* value for debug register    */
-	unsigned long	dbreg_flags;		/* return: dbreg error         */
-	unsigned long	dbreg_reserved[1];	/* for future use              */
+	unsigned int	dbreg_num;		/* which debug register */
+	unsigned short	dbreg_set;		/* event set for this register */
+	unsigned short	dbreg_reserved1;	/* for future use */
+	unsigned long	dbreg_value;		/* value for debug register */
+	unsigned long	dbreg_flags;		/* return: dbreg error */
+	unsigned long	dbreg_reserved2[1];	/* for future use */
 } pfarg_dbreg_t;
 
 typedef struct {
 	unsigned int	ft_version;	/* perfmon: major [16-31], minor [0-15] */
-	unsigned int	ft_reserved;	/* reserved for future use              */
-	unsigned long	reserved[4];	/* for future use                       */
+	unsigned int	ft_reserved;	/* reserved for future use */
+	unsigned long	reserved[4];	/* for future use */
 } pfarg_features_t;
 
 typedef struct {
-	pid_t		load_pid;	  /* process to load the context into */
-	unsigned int	load_set;	  /* first event set to load          */
-	unsigned long	load_reserved[2]; /* for future use                   */
+	pid_t		load_pid;	   /* process to load the context into */
+	unsigned short	load_set;	   /* first event set to load */
+	unsigned short	load_reserved1;	   /* for future use */
+	unsigned long	load_reserved2[3]; /* for future use */
 } pfarg_load_t;
 
 typedef struct {
 	int		msg_type;		/* generic message header */
 	int		msg_ctx_fd;		/* generic message header */
-	unsigned long	msg_tstamp;		/* for perf tuning */
-	unsigned int	msg_active_set;		/* active set at the time of overflow */
 	unsigned long	msg_ovfl_pmds[4];	/* which PMDs overflowed */
+	unsigned short  msg_active_set;		/* active set at the time of overflow */
+	unsigned short  msg_reserved1;		/* for future use */
+	unsigned int    msg_reserved2;		/* for future use */
+	unsigned long	msg_tstamp;		/* for perf tuning/debug */
 } pfm_ovfl_msg_t;
 
 typedef struct {
@@ -192,25 +198,28 @@ extern void pfm_handle_work(void);
 #define PFM_PMD_LONG_RESET	1
 #define PFM_PMD_SHORT_RESET	2
 
-typedef struct {
-	unsigned int notify_user:1;	/* notify user program of overflow                           */
-	unsigned int reset_pmds :2;	/* PFM_PMD_NO_RESET, PFM_PMD_LONG_RESET, PFM_PMD_SHORT_RESET */
-	unsigned int block:1;		/* block monitored task on kernel exit                       */
-	unsigned int stop_monitoring:1; /* will mask monitoring via PMCx.plm                         */
-	unsigned int reserved:26;	/* for future use                                            */
+typedef union {
+	unsigned int val;
+	struct {
+		unsigned int notify_user:1;	/* notify user program of overflow */
+		unsigned int reset_ovfl_pmds:1;	/* reset overflowed PMDs */
+		unsigned int block_task:1;	/* block monitored task on kernel exit */
+		unsigned int mask_monitoring:1; /* mask monitors via PMCx.plm */
+		unsigned int reserved:28;	/* for future use */
+	} bits;
 } pfm_ovfl_ctrl_t;
 
 typedef struct {
-	unsigned long   ovfl_pmds[4];	/* bitmask of overflowed pmds                            */
-	unsigned long   ovfl_notify[4];	/* bitmask of overflow pmds which asked for notification */
-	unsigned long   pmd_value;	/* current 64-bit value of 1st pmd which overflowed      */
-	unsigned long   pmd_last_reset;	/* last reset value of 1st pmd which overflowed          */
-	unsigned long	pmd_eventid;	/* eventid associated with 1st pmd which overflowed      */
-	unsigned int    active_set;	/* event set active at the time of the overflow          */
-	unsigned int    reserved1;
-	unsigned long	smpl_pmds[4];
-	unsigned long   smpl_pmds_values[PMU_MAX_PMDS];
-	pfm_ovfl_ctrl_t ovfl_ctrl;	/* return: perfmon controls to set by handler            */
+	unsigned char	ovfl_pmd;			/* index of overflowed PMD  */
+	unsigned char   ovfl_notify;			/* =1 if monitor requested overflow notification */
+	unsigned short  active_set;			/* event set active at the time of the overflow */
+	pfm_ovfl_ctrl_t ovfl_ctrl;			/* return: perfmon controls to set by handler */
+
+	unsigned long   pmd_last_reset;			/* last reset value of of the PMD */
+	unsigned long	smpl_pmds[4];			/* bitmask of other PMD of interest on overflow */
+	unsigned long   smpl_pmds_values[PMU_MAX_PMDS]; /* values for the other PMDs of interest */
+	unsigned long   pmd_value;			/* current 64-bit value of the PMD */
+	unsigned long	pmd_eventid;			/* eventid associated with PMD */
 } pfm_ovfl_arg_t;
 
 
@@ -223,7 +232,7 @@ typedef struct _pfm_buffer_fmt_t {
 	int		(*fmt_validate)(struct task_struct *task, unsigned int flags, int cpu, void *arg);
 	int		(*fmt_getsize)(struct task_struct *task, unsigned int flags, int cpu, void *arg, unsigned long *size);
 	int 		(*fmt_init)(struct task_struct *task, void *buf, unsigned int flags, int cpu, void *arg);
-	int		(*fmt_handler)(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct pt_regs *regs);
+	int		(*fmt_handler)(struct task_struct *task, void *buf, pfm_ovfl_arg_t *arg, struct pt_regs *regs, unsigned long stamp);
 	int		(*fmt_restart)(struct task_struct *task, pfm_ovfl_ctrl_t *ctrl, void *buf, struct pt_regs *regs);
 	int		(*fmt_restart_active)(struct task_struct *task, pfm_ovfl_ctrl_t *ctrl, void *buf, struct pt_regs *regs);
 	int		(*fmt_exit)(struct task_struct *task, void *buf, struct pt_regs *regs);

include/asm-ia64/perfmon_default_smpl.h

@@ -16,7 +16,9 @@
  */
 typedef struct {
 	unsigned long buf_size;		/* size of the buffer in bytes */
-	unsigned long reserved[3];	/* for future use */
+	unsigned int  flags;		/* buffer specific flags */
+	unsigned int  res1;		/* for future use */
+	unsigned long reserved[2];	/* for future use */
 } pfm_default_smpl_arg_t;
 
 /*
@@ -46,28 +48,27 @@ typedef struct {
 
 /*
  * Entry header in the sampling buffer.  The header is directly followed
- * with the PMDs saved in increasing index order: PMD4, PMD5, .... How
- * many PMDs are present depends on how the session was programmed.
+ * with the values of the PMD registers of interest saved in increasing 
+ * index order: PMD4, PMD5, and so on. How many PMDs are present depends 
+ * on how the session was programmed.
  *
- * XXX: in this version of the entry, only up to 64 registers can be
- * recorded. This should be enough for quite some time. Always check
- * sampling format before parsing entries!
+ * In the case where multiple counters overflow at the same time, multiple
+ * entries are written consecutively.
  *
- * In the case where multiple counters overflow at the same time, the
- * last_reset_value member indicates the initial value of the
- * overflowed PMD with the smallest index.  For instance, if PMD2 and
- * PMD5 have overflowed, the last_reset_value member contains the
- * initial value of PMD2.
+ * last_reset_value member indicates the initial value of the overflowed PMD. 
  */
 typedef struct {
-	int		pid;			/* current process at PMU interrupt point */
-	int		cpu;			/* cpu on which the overfow occured */
-	unsigned long	last_reset_val;		/* initial value of 1st overflowed PMD */
-	unsigned long	ip;			/* where did the overflow interrupt happened */
-	unsigned long	ovfl_pmds;		/* which PMDS registers overflowed (64 max) */
-	unsigned long   tstamp;			/* ar.itc on the CPU that took the overflow */
-	unsigned int	set;			/* event set active when overflow ocurred   */
-	unsigned int	reserved1;		/* for future use */
+        int             pid;                    /* active process at PMU interrupt point */
+        unsigned char   reserved1[3];           /* reserved for future use */
+        unsigned char   ovfl_pmd;               /* index of overflowed PMD */
+
+        unsigned long   last_reset_val;         /* initial value of overflowed PMD */
+        unsigned long   ip;                     /* where did the overflow interrupt happened  */
+        unsigned long   tstamp;                 /* ar.itc when entering perfmon intr. handler */
+
+        unsigned short  cpu;                    /* cpu on which the overfow occured */
+        unsigned short  set;                    /* event set active when overflow ocurred   */
+        unsigned int    reserved2;              /* for future use */
 } pfm_default_smpl_entry_t;
 
 #define PFM_DEFAULT_MAX_PMDS		64 /* how many pmds supported by data structures (sizeof(unsigned long) */