perfmon.c 169 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1
2
3
4
5
6
7
8
9
10
11
12
13
/*
 * This file implements the perfmon-2 subsystem which is used
 * to program the IA-64 Performance Monitoring Unit (PMU).
 *
 * The initial version of perfmon.c was written by
 * Ganesh Venkitachalam, IBM Corp.
 *
 * Then it was modified for perfmon-1.x by Stephane Eranian and
 * David Mosberger, Hewlett Packard Co.
 *
 * Version Perfmon-2.x is a rewrite of perfmon-1.x
 * by Stephane Eranian, Hewlett Packard Co.
 *
Tony Luck's avatar
Tony Luck committed
14
 * Copyright (C) 1999-2005  Hewlett Packard Co
Linus Torvalds's avatar
Linus Torvalds committed
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
 *               Stephane Eranian <eranian@hpl.hp.com>
 *               David Mosberger-Tang <davidm@hpl.hp.com>
 *
 * More information about perfmon available at:
 * 	http://www.hpl.hp.com/research/linux/perfmon
 */

#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/interrupt.h>
#include <linux/proc_fs.h>
#include <linux/seq_file.h>
#include <linux/init.h>
#include <linux/vmalloc.h>
#include <linux/mm.h>
#include <linux/sysctl.h>
#include <linux/list.h>
#include <linux/file.h>
#include <linux/poll.h>
#include <linux/vfs.h>
36
#include <linux/smp.h>
Linus Torvalds's avatar
Linus Torvalds committed
37
38
39
#include <linux/pagemap.h>
#include <linux/mount.h>
#include <linux/bitops.h>
40
#include <linux/capability.h>
41
#include <linux/rcupdate.h>
42
#include <linux/completion.h>
43
#include <linux/tracehook.h>
Linus Torvalds's avatar
Linus Torvalds committed
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65

#include <asm/errno.h>
#include <asm/intrinsics.h>
#include <asm/page.h>
#include <asm/perfmon.h>
#include <asm/processor.h>
#include <asm/signal.h>
#include <asm/system.h>
#include <asm/uaccess.h>
#include <asm/delay.h>

#ifdef CONFIG_PERFMON
/*
 * perfmon context state
 */
#define PFM_CTX_UNLOADED	1	/* context is not loaded onto any task */
#define PFM_CTX_LOADED		2	/* context is loaded onto a task */
#define PFM_CTX_MASKED		3	/* context is loaded but monitoring is masked due to overflow */
#define PFM_CTX_ZOMBIE		4	/* owner of the context is closing it */

#define PFM_INVALID_ACTIVATION	(~0UL)

66
67
68
#define PFM_NUM_PMC_REGS	64	/* PMC save area for ctxsw */
#define PFM_NUM_PMD_REGS	64	/* PMD save area for ctxsw */

Linus Torvalds's avatar
Linus Torvalds committed
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
/*
 * depth of message queue
 */
#define PFM_MAX_MSGS		32
#define PFM_CTXQ_EMPTY(g)	((g)->ctx_msgq_head == (g)->ctx_msgq_tail)

/*
 * type of a PMU register (bitmask).
 * bitmask structure:
 * 	bit0   : register implemented
 * 	bit1   : end marker
 * 	bit2-3 : reserved
 * 	bit4   : pmc has pmc.pm
 * 	bit5   : pmc controls a counter (has pmc.oi), pmd is used as counter
 * 	bit6-7 : register type
 * 	bit8-31: reserved
 */
#define PFM_REG_NOTIMPL		0x0 /* not implemented at all */
#define PFM_REG_IMPL		0x1 /* register implemented */
#define PFM_REG_END		0x2 /* end marker */
#define PFM_REG_MONITOR		(0x1<<4|PFM_REG_IMPL) /* a PMC with a pmc.pm field only */
#define PFM_REG_COUNTING	(0x2<<4|PFM_REG_MONITOR) /* a monitor + pmc.oi+ PMD used as a counter */
#define PFM_REG_CONTROL		(0x4<<4|PFM_REG_IMPL) /* PMU control register */
#define	PFM_REG_CONFIG		(0x8<<4|PFM_REG_IMPL) /* configuration register */
#define PFM_REG_BUFFER	 	(0xc<<4|PFM_REG_IMPL) /* PMD used as buffer */

#define PMC_IS_LAST(i)	(pmu_conf->pmc_desc[i].type & PFM_REG_END)
#define PMD_IS_LAST(i)	(pmu_conf->pmd_desc[i].type & PFM_REG_END)

#define PMC_OVFL_NOTIFY(ctx, i)	((ctx)->ctx_pmds[i].flags &  PFM_REGFL_OVFL_NOTIFY)

/* i assumed unsigned */
#define PMC_IS_IMPL(i)	  (i< PMU_MAX_PMCS && (pmu_conf->pmc_desc[i].type & PFM_REG_IMPL))
#define PMD_IS_IMPL(i)	  (i< PMU_MAX_PMDS && (pmu_conf->pmd_desc[i].type & PFM_REG_IMPL))

/* XXX: these assume that register i is implemented */
#define PMD_IS_COUNTING(i) ((pmu_conf->pmd_desc[i].type & PFM_REG_COUNTING) == PFM_REG_COUNTING)
#define PMC_IS_COUNTING(i) ((pmu_conf->pmc_desc[i].type & PFM_REG_COUNTING) == PFM_REG_COUNTING)
#define PMC_IS_MONITOR(i)  ((pmu_conf->pmc_desc[i].type & PFM_REG_MONITOR)  == PFM_REG_MONITOR)
#define PMC_IS_CONTROL(i)  ((pmu_conf->pmc_desc[i].type & PFM_REG_CONTROL)  == PFM_REG_CONTROL)

#define PMC_DFL_VAL(i)     pmu_conf->pmc_desc[i].default_value
#define PMC_RSVD_MASK(i)   pmu_conf->pmc_desc[i].reserved_mask
#define PMD_PMD_DEP(i)	   pmu_conf->pmd_desc[i].dep_pmd[0]
#define PMC_PMD_DEP(i)	   pmu_conf->pmc_desc[i].dep_pmd[0]

#define PFM_NUM_IBRS	  IA64_NUM_DBG_REGS
#define PFM_NUM_DBRS	  IA64_NUM_DBG_REGS

#define CTX_OVFL_NOBLOCK(c)	((c)->ctx_fl_block == 0)
#define CTX_HAS_SMPL(c)		((c)->ctx_fl_is_sampling)
#define PFM_CTX_TASK(h)		(h)->ctx_task

#define PMU_PMC_OI		5 /* position of pmc.oi bit */

/* XXX: does not support more than 64 PMDs */
#define CTX_USED_PMD(ctx, mask) (ctx)->ctx_used_pmds[0] |= (mask)
#define CTX_IS_USED_PMD(ctx, c) (((ctx)->ctx_used_pmds[0] & (1UL << (c))) != 0UL)

#define CTX_USED_MONITOR(ctx, mask) (ctx)->ctx_used_monitors[0] |= (mask)

#define CTX_USED_IBR(ctx,n) 	(ctx)->ctx_used_ibrs[(n)>>6] |= 1UL<< ((n) % 64)
#define CTX_USED_DBR(ctx,n) 	(ctx)->ctx_used_dbrs[(n)>>6] |= 1UL<< ((n) % 64)
#define CTX_USES_DBREGS(ctx)	(((pfm_context_t *)(ctx))->ctx_fl_using_dbreg==1)
#define PFM_CODE_RR	0	/* requesting code range restriction */
#define PFM_DATA_RR	1	/* requestion data range restriction */

#define PFM_CPUINFO_CLEAR(v)	pfm_get_cpu_var(pfm_syst_info) &= ~(v)
#define PFM_CPUINFO_SET(v)	pfm_get_cpu_var(pfm_syst_info) |= (v)
#define PFM_CPUINFO_GET()	pfm_get_cpu_var(pfm_syst_info)

#define RDEP(x)	(1UL<<(x))

/*
 * context protection macros
 * in SMP:
 * 	- we need to protect against CPU concurrency (spin_lock)
 * 	- we need to protect against PMU overflow interrupts (local_irq_disable)
 * in UP:
 * 	- we need to protect against PMU overflow interrupts (local_irq_disable)
 *
150
 * spin_lock_irqsave()/spin_unlock_irqrestore():
Linus Torvalds's avatar
Linus Torvalds committed
151
152
153
154
155
156
157
158
159
160
161
 * 	in SMP: local_irq_disable + spin_lock
 * 	in UP : local_irq_disable
 *
 * spin_lock()/spin_lock():
 * 	in UP : removed automatically
 * 	in SMP: protect against context accesses from other CPU. interrupts
 * 	        are not masked. This is useful for the PMU interrupt handler
 * 	        because we know we will not get PMU concurrency in that code.
 */
#define PROTECT_CTX(c, f) \
	do {  \
162
		DPRINT(("spinlock_irq_save ctx %p by [%d]\n", c, task_pid_nr(current))); \
Linus Torvalds's avatar
Linus Torvalds committed
163
		spin_lock_irqsave(&(c)->ctx_lock, f); \
164
		DPRINT(("spinlocked ctx %p  by [%d]\n", c, task_pid_nr(current))); \
Linus Torvalds's avatar
Linus Torvalds committed
165
166
167
168
	} while(0)

#define UNPROTECT_CTX(c, f) \
	do { \
169
		DPRINT(("spinlock_irq_restore ctx %p by [%d]\n", c, task_pid_nr(current))); \
Linus Torvalds's avatar
Linus Torvalds committed
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
		spin_unlock_irqrestore(&(c)->ctx_lock, f); \
	} while(0)

#define PROTECT_CTX_NOPRINT(c, f) \
	do {  \
		spin_lock_irqsave(&(c)->ctx_lock, f); \
	} while(0)


#define UNPROTECT_CTX_NOPRINT(c, f) \
	do { \
		spin_unlock_irqrestore(&(c)->ctx_lock, f); \
	} while(0)


#define PROTECT_CTX_NOIRQ(c) \
	do {  \
		spin_lock(&(c)->ctx_lock); \
	} while(0)

#define UNPROTECT_CTX_NOIRQ(c) \
	do { \
		spin_unlock(&(c)->ctx_lock); \
	} while(0)


#ifdef CONFIG_SMP

#define GET_ACTIVATION()	pfm_get_cpu_var(pmu_activation_number)
#define INC_ACTIVATION()	pfm_get_cpu_var(pmu_activation_number)++
#define SET_ACTIVATION(c)	(c)->ctx_last_activation = GET_ACTIVATION()

#else /* !CONFIG_SMP */
#define SET_ACTIVATION(t) 	do {} while(0)
#define GET_ACTIVATION(t) 	do {} while(0)
#define INC_ACTIVATION(t) 	do {} while(0)
#endif /* CONFIG_SMP */

#define SET_PMU_OWNER(t, c)	do { pfm_get_cpu_var(pmu_owner) = (t); pfm_get_cpu_var(pmu_ctx) = (c); } while(0)
#define GET_PMU_OWNER()		pfm_get_cpu_var(pmu_owner)
#define GET_PMU_CTX()		pfm_get_cpu_var(pmu_ctx)

#define LOCK_PFS(g)	    	spin_lock_irqsave(&pfm_sessions.pfs_lock, g)
#define UNLOCK_PFS(g)	    	spin_unlock_irqrestore(&pfm_sessions.pfs_lock, g)

#define PFM_REG_RETFLAG_SET(flags, val)	do { flags &= ~PFM_REG_RETFL_MASK; flags |= (val); } while(0)

/*
 * cmp0 must be the value of pmc0
 */
#define PMC0_HAS_OVFL(cmp0)  (cmp0 & ~0x1UL)

#define PFMFS_MAGIC 0xa0b4d889

/*
 * debugging
 */
#define PFM_DEBUGGING 1
#ifdef PFM_DEBUGGING
#define DPRINT(a) \
	do { \
231
		if (unlikely(pfm_sysctl.debug >0)) { printk("%s.%d: CPU%d [%d] ", __func__, __LINE__, smp_processor_id(), task_pid_nr(current)); printk a; } \
Linus Torvalds's avatar
Linus Torvalds committed
232
233
234
235
	} while (0)

#define DPRINT_ovfl(a) \
	do { \
236
		if (unlikely(pfm_sysctl.debug > 0 && pfm_sysctl.debug_ovfl >0)) { printk("%s.%d: CPU%d [%d] ", __func__, __LINE__, smp_processor_id(), task_pid_nr(current)); printk a; } \
Linus Torvalds's avatar
Linus Torvalds committed
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
	} while (0)
#endif

/*
 * 64-bit software counter structure
 *
 * the next_reset_type is applied to the next call to pfm_reset_regs()
 */
typedef struct {
	unsigned long	val;		/* virtual 64bit counter value */
	unsigned long	lval;		/* last reset value */
	unsigned long	long_reset;	/* reset value on sampling overflow */
	unsigned long	short_reset;    /* reset value on overflow */
	unsigned long	reset_pmds[4];  /* which other pmds to reset when this counter overflows */
	unsigned long	smpl_pmds[4];   /* which pmds are accessed when counter overflow */
	unsigned long	seed;		/* seed for random-number generator */
	unsigned long	mask;		/* mask for random-number generator */
	unsigned int 	flags;		/* notify/do not notify */
	unsigned long	eventid;	/* overflow event identifier */
} pfm_counter_t;

/*
 * context flags
 */
typedef struct {
	unsigned int block:1;		/* when 1, task will blocked on user notifications */
	unsigned int system:1;		/* do system wide monitoring */
	unsigned int using_dbreg:1;	/* using range restrictions (debug registers) */
	unsigned int is_sampling:1;	/* true if using a custom format */
	unsigned int excl_idle:1;	/* exclude idle task in system wide session */
	unsigned int going_zombie:1;	/* context is zombie (MASKED+blocking) */
	unsigned int trap_reason:2;	/* reason for going into pfm_handle_work() */
	unsigned int no_msg:1;		/* no message sent on overflow */
	unsigned int can_restart:1;	/* allowed to issue a PFM_RESTART */
	unsigned int reserved:22;
} pfm_context_flags_t;

#define PFM_TRAP_REASON_NONE		0x0	/* default value */
#define PFM_TRAP_REASON_BLOCK		0x1	/* we need to block on overflow */
#define PFM_TRAP_REASON_RESET		0x2	/* we need to reset PMDs */


/*
 * perfmon context: encapsulates all the state of a monitoring session
 */

typedef struct pfm_context {
	spinlock_t		ctx_lock;		/* context protection */

	pfm_context_flags_t	ctx_flags;		/* bitmask of flags  (block reason incl.) */
	unsigned int		ctx_state;		/* state: active/inactive (no bitfield) */

	struct task_struct 	*ctx_task;		/* task to which context is attached */

	unsigned long		ctx_ovfl_regs[4];	/* which registers overflowed (notification) */

293
	struct completion	ctx_restart_done;  	/* use for blocking notification mode */
Linus Torvalds's avatar
Linus Torvalds committed
294
295
296
297
298
299
300
301
302

	unsigned long		ctx_used_pmds[4];	/* bitmask of PMD used            */
	unsigned long		ctx_all_pmds[4];	/* bitmask of all accessible PMDs */
	unsigned long		ctx_reload_pmds[4];	/* bitmask of force reload PMD on ctxsw in */

	unsigned long		ctx_all_pmcs[4];	/* bitmask of all accessible PMCs */
	unsigned long		ctx_reload_pmcs[4];	/* bitmask of force reload PMC on ctxsw in */
	unsigned long		ctx_used_monitors[4];	/* bitmask of monitor PMC being used */

303
	unsigned long		ctx_pmcs[PFM_NUM_PMC_REGS];	/*  saved copies of PMC values */
Linus Torvalds's avatar
Linus Torvalds committed
304
305
306
307
308
309

	unsigned int		ctx_used_ibrs[1];		/* bitmask of used IBR (speedup ctxsw in) */
	unsigned int		ctx_used_dbrs[1];		/* bitmask of used DBR (speedup ctxsw in) */
	unsigned long		ctx_dbrs[IA64_NUM_DBG_REGS];	/* DBR values (cache) when not loaded */
	unsigned long		ctx_ibrs[IA64_NUM_DBG_REGS];	/* IBR values (cache) when not loaded */

310
311
312
313
	pfm_counter_t		ctx_pmds[PFM_NUM_PMD_REGS]; /* software state for PMDS */

	unsigned long		th_pmcs[PFM_NUM_PMC_REGS];	/* PMC thread save state */
	unsigned long		th_pmds[PFM_NUM_PMD_REGS];	/* PMD thread save state */
Linus Torvalds's avatar
Linus Torvalds committed
314

315
	unsigned long		ctx_saved_psr_up;	/* only contains psr.up value */
Linus Torvalds's avatar
Linus Torvalds committed
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507

	unsigned long		ctx_last_activation;	/* context last activation number for last_cpu */
	unsigned int		ctx_last_cpu;		/* CPU id of current or last CPU used (SMP only) */
	unsigned int		ctx_cpu;		/* cpu to which perfmon is applied (system wide) */

	int			ctx_fd;			/* file descriptor used my this context */
	pfm_ovfl_arg_t		ctx_ovfl_arg;		/* argument to custom buffer format handler */

	pfm_buffer_fmt_t	*ctx_buf_fmt;		/* buffer format callbacks */
	void			*ctx_smpl_hdr;		/* points to sampling buffer header kernel vaddr */
	unsigned long		ctx_smpl_size;		/* size of sampling buffer */
	void			*ctx_smpl_vaddr;	/* user level virtual address of smpl buffer */

	wait_queue_head_t 	ctx_msgq_wait;
	pfm_msg_t		ctx_msgq[PFM_MAX_MSGS];
	int			ctx_msgq_head;
	int			ctx_msgq_tail;
	struct fasync_struct	*ctx_async_queue;

	wait_queue_head_t 	ctx_zombieq;		/* termination cleanup wait queue */
} pfm_context_t;

/*
 * magic number used to verify that structure is really
 * a perfmon context
 */
#define PFM_IS_FILE(f)		((f)->f_op == &pfm_file_ops)

#define PFM_GET_CTX(t)	 	((pfm_context_t *)(t)->thread.pfm_context)

#ifdef CONFIG_SMP
#define SET_LAST_CPU(ctx, v)	(ctx)->ctx_last_cpu = (v)
#define GET_LAST_CPU(ctx)	(ctx)->ctx_last_cpu
#else
#define SET_LAST_CPU(ctx, v)	do {} while(0)
#define GET_LAST_CPU(ctx)	do {} while(0)
#endif


#define ctx_fl_block		ctx_flags.block
#define ctx_fl_system		ctx_flags.system
#define ctx_fl_using_dbreg	ctx_flags.using_dbreg
#define ctx_fl_is_sampling	ctx_flags.is_sampling
#define ctx_fl_excl_idle	ctx_flags.excl_idle
#define ctx_fl_going_zombie	ctx_flags.going_zombie
#define ctx_fl_trap_reason	ctx_flags.trap_reason
#define ctx_fl_no_msg		ctx_flags.no_msg
#define ctx_fl_can_restart	ctx_flags.can_restart

#define PFM_SET_WORK_PENDING(t, v)	do { (t)->thread.pfm_needs_checking = v; } while(0);
#define PFM_GET_WORK_PENDING(t)		(t)->thread.pfm_needs_checking

/*
 * global information about all sessions
 * mostly used to synchronize between system wide and per-process
 */
typedef struct {
	spinlock_t		pfs_lock;		   /* lock the structure */

	unsigned int		pfs_task_sessions;	   /* number of per task sessions */
	unsigned int		pfs_sys_sessions;	   /* number of per system wide sessions */
	unsigned int		pfs_sys_use_dbregs;	   /* incremented when a system wide session uses debug regs */
	unsigned int		pfs_ptrace_use_dbregs;	   /* incremented when a process uses debug regs */
	struct task_struct	*pfs_sys_session[NR_CPUS]; /* point to task owning a system-wide session */
} pfm_session_t;

/*
 * information about a PMC or PMD.
 * dep_pmd[]: a bitmask of dependent PMD registers
 * dep_pmc[]: a bitmask of dependent PMC registers
 */
typedef int (*pfm_reg_check_t)(struct task_struct *task, pfm_context_t *ctx, unsigned int cnum, unsigned long *val, struct pt_regs *regs);
typedef struct {
	unsigned int		type;
	int			pm_pos;
	unsigned long		default_value;	/* power-on default value */
	unsigned long		reserved_mask;	/* bitmask of reserved bits */
	pfm_reg_check_t		read_check;
	pfm_reg_check_t		write_check;
	unsigned long		dep_pmd[4];
	unsigned long		dep_pmc[4];
} pfm_reg_desc_t;

/* assume cnum is a valid monitor */
#define PMC_PM(cnum, val)	(((val) >> (pmu_conf->pmc_desc[cnum].pm_pos)) & 0x1)

/*
 * This structure is initialized at boot time and contains
 * a description of the PMU main characteristics.
 *
 * If the probe function is defined, detection is based
 * on its return value: 
 * 	- 0 means recognized PMU
 * 	- anything else means not supported
 * When the probe function is not defined, then the pmu_family field
 * is used and it must match the host CPU family such that:
 * 	- cpu->family & config->pmu_family != 0
 */
typedef struct {
	unsigned long  ovfl_val;	/* overflow value for counters */

	pfm_reg_desc_t *pmc_desc;	/* detailed PMC register dependencies descriptions */
	pfm_reg_desc_t *pmd_desc;	/* detailed PMD register dependencies descriptions */

	unsigned int   num_pmcs;	/* number of PMCS: computed at init time */
	unsigned int   num_pmds;	/* number of PMDS: computed at init time */
	unsigned long  impl_pmcs[4];	/* bitmask of implemented PMCS */
	unsigned long  impl_pmds[4];	/* bitmask of implemented PMDS */

	char	      *pmu_name;	/* PMU family name */
	unsigned int  pmu_family;	/* cpuid family pattern used to identify pmu */
	unsigned int  flags;		/* pmu specific flags */
	unsigned int  num_ibrs;		/* number of IBRS: computed at init time */
	unsigned int  num_dbrs;		/* number of DBRS: computed at init time */
	unsigned int  num_counters;	/* PMC/PMD counting pairs : computed at init time */
	int           (*probe)(void);   /* customized probe routine */
	unsigned int  use_rr_dbregs:1;	/* set if debug registers used for range restriction */
} pmu_config_t;
/*
 * PMU specific flags
 */
#define PFM_PMU_IRQ_RESEND	1	/* PMU needs explicit IRQ resend */

/*
 * debug register related type definitions
 */
typedef struct {
	unsigned long ibr_mask:56;
	unsigned long ibr_plm:4;
	unsigned long ibr_ig:3;
	unsigned long ibr_x:1;
} ibr_mask_reg_t;

typedef struct {
	unsigned long dbr_mask:56;
	unsigned long dbr_plm:4;
	unsigned long dbr_ig:2;
	unsigned long dbr_w:1;
	unsigned long dbr_r:1;
} dbr_mask_reg_t;

typedef union {
	unsigned long  val;
	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 */

508
static DEFINE_SPINLOCK(pfm_alt_install_check);
Tony Luck's avatar
Tony Luck committed
509
510
static pfm_intr_handler_desc_t  *pfm_alt_intr_handler;

Linus Torvalds's avatar
Linus Torvalds committed
511
512
513
514
515
516
517
518
519
static struct proc_dir_entry 	*perfmon_dir;
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 */
520
521
pfm_sysctl_t pfm_sysctl;
EXPORT_SYMBOL(pfm_sysctl);
Linus Torvalds's avatar
Linus Torvalds committed
522
523

static ctl_table pfm_ctl_table[]={
524
525
526
527
528
	{
		.procname	= "debug",
		.data		= &pfm_sysctl.debug,
		.maxlen		= sizeof(int),
		.mode		= 0666,
529
		.proc_handler	= proc_dointvec,
530
531
532
533
534
535
	},
	{
		.procname	= "debug_ovfl",
		.data		= &pfm_sysctl.debug_ovfl,
		.maxlen		= sizeof(int),
		.mode		= 0666,
536
		.proc_handler	= proc_dointvec,
537
538
539
540
541
542
	},
	{
		.procname	= "fastctxsw",
		.data		= &pfm_sysctl.fastctxsw,
		.maxlen		= sizeof(int),
		.mode		= 0600,
543
		.proc_handler	= proc_dointvec,
544
545
546
547
548
549
	},
	{
		.procname	= "expert_mode",
		.data		= &pfm_sysctl.expert_mode,
		.maxlen		= sizeof(int),
		.mode		= 0600,
550
		.proc_handler	= proc_dointvec,
551
552
	},
	{}
Linus Torvalds's avatar
Linus Torvalds committed
553
554
};
static ctl_table pfm_sysctl_dir[] = {
555
556
	{
		.procname	= "perfmon",
557
		.mode		= 0555,
558
559
560
		.child		= pfm_ctl_table,
	},
 	{}
Linus Torvalds's avatar
Linus Torvalds committed
561
562
};
static ctl_table pfm_sysctl_root[] = {
563
564
	{
		.procname	= "kernel",
565
		.mode		= 0555,
566
567
568
		.child		= pfm_sysctl_dir,
	},
 	{}
Linus Torvalds's avatar
Linus Torvalds committed
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
};
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);

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

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

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));
}

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

601
static inline void
Linus Torvalds's avatar
Linus Torvalds committed
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
pfm_unprotect_ctx_ctxsw(pfm_context_t *x, unsigned long f)
{
	spin_unlock(&(x)->ctx_lock);
}

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);
}


620
621
622
static int
pfmfs_get_sb(struct file_system_type *fs_type, int flags, const char *dev_name, void *data,
	     struct vfsmount *mnt)
Linus Torvalds's avatar
Linus Torvalds committed
623
{
624
	return get_sb_pseudo(fs_type, "pfm:", NULL, PFMFS_MAGIC, mnt);
Linus Torvalds's avatar
Linus Torvalds committed
625
626
627
628
629
630
631
632
633
634
635
636
}

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

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);
637
EXPORT_PER_CPU_SYMBOL_GPL(pfm_syst_info);
Linus Torvalds's avatar
Linus Torvalds committed
638
639
640


/* forward declaration */
641
static const struct file_operations pfm_file_ops;
Linus Torvalds's avatar
Linus Torvalds committed
642
643
644
645
646
647
648
649
650
651
652
653
654

/*
 * 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"
655
#include "perfmon_montecito.h"
Linus Torvalds's avatar
Linus Torvalds committed
656
657
658
#include "perfmon_generic.h"

static pmu_config_t *pmu_confs[]={
659
	&pmu_conf_mont,
Linus Torvalds's avatar
Linus Torvalds committed
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
	&pmu_conf_mck,
	&pmu_conf_ita,
	&pmu_conf_gen, /* must be last */
	NULL
};


static int pfm_end_notify_user(pfm_context_t *ctx);

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);
}

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;
}

static pfm_context_t *
865
pfm_context_alloc(int ctx_flags)
Linus Torvalds's avatar
Linus Torvalds committed
866
867
868
869
870
871
872
{
	pfm_context_t *ctx;

	/* 
	 * allocate context descriptor 
	 * must be able to free with interrupts disabled
	 */
873
	ctx = kzalloc(sizeof(pfm_context_t), GFP_KERNEL);
Linus Torvalds's avatar
Linus Torvalds committed
874
875
	if (ctx) {
		DPRINT(("alloc ctx @%p\n", ctx));
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915

		/*
		 * init context protection lock
		 */
		spin_lock_init(&ctx->ctx_lock);

		/*
		 * context is unloaded
		 */
		ctx->ctx_state = PFM_CTX_UNLOADED;

		/*
		 * initialization of context's flags
		 */
		ctx->ctx_fl_block       = (ctx_flags & PFM_FL_NOTIFY_BLOCK) ? 1 : 0;
		ctx->ctx_fl_system      = (ctx_flags & PFM_FL_SYSTEM_WIDE) ? 1: 0;
		ctx->ctx_fl_no_msg      = (ctx_flags & PFM_FL_OVFL_NO_MSG) ? 1: 0;
		/*
		 * will move to set properties
		 * ctx->ctx_fl_excl_idle   = (ctx_flags & PFM_FL_EXCL_IDLE) ? 1: 0;
		 */

		/*
		 * init restart semaphore to locked
		 */
		init_completion(&ctx->ctx_restart_done);

		/*
		 * activation is used in SMP only
		 */
		ctx->ctx_last_activation = PFM_INVALID_ACTIVATION;
		SET_LAST_CPU(ctx, -1);

		/*
		 * initialize notification message queue
		 */
		ctx->ctx_msgq_head = ctx->ctx_msgq_tail = 0;
		init_waitqueue_head(&ctx->ctx_msgq_wait);
		init_waitqueue_head(&ctx->ctx_zombieq);

Linus Torvalds's avatar
Linus Torvalds committed
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
	}
	return ctx;
}

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

static void
pfm_mask_monitoring(struct task_struct *task)
{
	pfm_context_t *ctx = PFM_GET_CTX(task);
	unsigned long mask, val, ovfl_mask;
	int i;

936
	DPRINT_ovfl(("masking monitoring for [%d]\n", task_pid_nr(task)));
Linus Torvalds's avatar
Linus Torvalds committed
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952

	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.
	 *
953
	 * As a consequence to this call, the ctx->th_pmds[] array
Linus Torvalds's avatar
Linus Torvalds committed
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
	 * 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;
988
989
990
		ia64_set_pmc(i, ctx->th_pmcs[i] & ~0xfUL);
		ctx->th_pmcs[i] &= ~0xfUL;
		DPRINT_ovfl(("pmc[%d]=0x%lx\n", i, ctx->th_pmcs[i]));
Linus Torvalds's avatar
Linus Torvalds committed
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
	}
	/*
	 * 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
pfm_restore_monitoring(struct task_struct *task)
{
	pfm_context_t *ctx = PFM_GET_CTX(task);
	unsigned long mask, ovfl_mask;
	unsigned long psr, val;
	int i, is_system;

	is_system = ctx->ctx_fl_system;
	ovfl_mask = pmu_conf->ovfl_val;

	if (task != current) {
1015
		printk(KERN_ERR "perfmon.%d: invalid task[%d] current[%d]\n", __LINE__, task_pid_nr(task), task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1016
1017
1018
1019
		return;
	}
	if (ctx->ctx_state != PFM_CTX_MASKED) {
		printk(KERN_ERR "perfmon.%d: task[%d] current[%d] invalid state=%d\n", __LINE__,
1020
			task_pid_nr(task), task_pid_nr(current), ctx->ctx_state);
Linus Torvalds's avatar
Linus Torvalds committed
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
		return;
	}
	psr = pfm_get_psr();
	/*
	 * monitoring is masked via the PMC.
	 * As we restore their value, we do not want each counter to
	 * restart right away. We stop monitoring using the PSR,
	 * restore the PMC (and PMD) and then re-establish the psr
	 * as it was. Note that there can be no pending overflow at
	 * this point, because monitoring was MASKED.
	 *
	 * system-wide session are pinned and self-monitoring
	 */
	if (is_system && (PFM_CPUINFO_GET() & PFM_CPUINFO_DCR_PP)) {
		/* disable dcr pp */
		ia64_setreg(_IA64_REG_CR_DCR, ia64_getreg(_IA64_REG_CR_DCR) & ~IA64_DCR_PP);
		pfm_clear_psr_pp();
	} else {
		pfm_clear_psr_up();
	}
	/*
	 * first, we restore the PMD
	 */
	mask = ctx->ctx_used_pmds[0];
	for (i = 0; mask; i++, mask>>=1) {
		/* skip non used pmds */
		if ((mask & 0x1) == 0) continue;

		if (PMD_IS_COUNTING(i)) {
			/*
			 * we split the 64bit value according to
			 * counter width
			 */
			val = ctx->ctx_pmds[i].val & ovfl_mask;
			ctx->ctx_pmds[i].val &= ~ovfl_mask;
		} else {
			val = ctx->ctx_pmds[i].val;
		}
		ia64_set_pmd(i, val);

		DPRINT(("pmd[%d]=0x%lx hw_pmd=0x%lx\n",
			i,
			ctx->ctx_pmds[i].val,
			val));
	}
	/*
	 * restore the PMCs
	 */
	mask = ctx->ctx_used_monitors[0] >> PMU_FIRST_COUNTER;
	for(i= PMU_FIRST_COUNTER; mask; i++, mask>>=1) {
		if ((mask & 0x1) == 0UL) continue;
1072
1073
		ctx->th_pmcs[i] = ctx->ctx_pmcs[i];
		ia64_set_pmc(i, ctx->th_pmcs[i]);
1074
1075
		DPRINT(("[%d] pmc[%d]=0x%lx\n",
					task_pid_nr(task), i, ctx->th_pmcs[i]));
Linus Torvalds's avatar
Linus Torvalds committed
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
	}
	ia64_srlz_d();

	/*
	 * must restore DBR/IBR because could be modified while masked
	 * XXX: need to optimize 
	 */
	if (ctx->ctx_fl_using_dbreg) {
		pfm_restore_ibrs(ctx->ctx_ibrs, pmu_conf->num_ibrs);
		pfm_restore_dbrs(ctx->ctx_dbrs, pmu_conf->num_dbrs);
	}

	/*
	 * now restore PSR
	 */
	if (is_system && (PFM_CPUINFO_GET() & PFM_CPUINFO_DCR_PP)) {
		/* enable dcr pp */
		ia64_setreg(_IA64_REG_CR_DCR, ia64_getreg(_IA64_REG_CR_DCR) | IA64_DCR_PP);
		ia64_srlz_i();
	}
	pfm_set_psr_l(psr);
}

static inline void
pfm_save_pmds(unsigned long *pmds, unsigned long mask)
{
	int i;

	ia64_srlz_d();

	for (i=0; mask; i++, mask>>=1) {
		if (mask & 0x1) pmds[i] = ia64_get_pmd(i);
	}
}

/*
 * reload from thread state (used for ctxw only)
 */
static inline void
pfm_restore_pmds(unsigned long *pmds, unsigned long mask)
{
	int i;
	unsigned long val, ovfl_val = pmu_conf->ovfl_val;

	for (i=0; mask; i++, mask>>=1) {
		if ((mask & 0x1) == 0) continue;
		val = PMD_IS_COUNTING(i) ? pmds[i] & ovfl_val : pmds[i];
		ia64_set_pmd(i, val);
	}
	ia64_srlz_d();
}

/*
 * propagate PMD from context to thread-state
 */
static inline void
pfm_copy_pmds(struct task_struct *task, pfm_context_t *ctx)
{
	unsigned long ovfl_val = pmu_conf->ovfl_val;
	unsigned long mask = ctx->ctx_all_pmds[0];
	unsigned long val;
	int i;

	DPRINT(("mask=0x%lx\n", mask));

	for (i=0; mask; i++, mask>>=1) {

		val = ctx->ctx_pmds[i].val;

		/*
		 * We break up the 64 bit value into 2 pieces
		 * the lower bits go to the machine state in the
		 * thread (will be reloaded on ctxsw in).
		 * The upper part stays in the soft-counter.
		 */
		if (PMD_IS_COUNTING(i)) {
			ctx->ctx_pmds[i].val = val & ~ovfl_val;
			 val &= ovfl_val;
		}
1155
		ctx->th_pmds[i] = val;
Linus Torvalds's avatar
Linus Torvalds committed
1156
1157
1158

		DPRINT(("pmd[%d]=0x%lx soft_val=0x%lx\n",
			i,
1159
			ctx->th_pmds[i],
Linus Torvalds's avatar
Linus Torvalds committed
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
			ctx->ctx_pmds[i].val));
	}
}

/*
 * propagate PMC from context to thread-state
 */
static inline void
pfm_copy_pmcs(struct task_struct *task, pfm_context_t *ctx)
{
	unsigned long mask = ctx->ctx_all_pmcs[0];
	int i;

	DPRINT(("mask=0x%lx\n", mask));

	for (i=0; mask; i++, mask>>=1) {
		/* masking 0 with ovfl_val yields 0 */
1177
1178
		ctx->th_pmcs[i] = ctx->ctx_pmcs[i];
		DPRINT(("pmc[%d]=0x%lx\n", i, ctx->th_pmcs[i]));
Linus Torvalds's avatar
Linus Torvalds committed
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
	}
}



static inline void
pfm_restore_pmcs(unsigned long *pmcs, unsigned long mask)
{
	int i;

	for (i=0; mask; i++, mask>>=1) {
		if ((mask & 0x1) == 0) continue;
		ia64_set_pmc(i, pmcs[i]);
	}
	ia64_srlz_d();
}

static inline int
pfm_uuid_cmp(pfm_uuid_t a, pfm_uuid_t b)
{
	return memcmp(a, b, sizeof(pfm_uuid_t));
}

static inline int
pfm_buf_fmt_exit(pfm_buffer_fmt_t *fmt, struct task_struct *task, void *buf, struct pt_regs *regs)
{
	int ret = 0;
	if (fmt->fmt_exit) ret = (*fmt->fmt_exit)(task, buf, regs);
	return ret;
}

static inline int
pfm_buf_fmt_getsize(pfm_buffer_fmt_t *fmt, struct task_struct *task, unsigned int flags, int cpu, void *arg, unsigned long *size)
{
	int ret = 0;
	if (fmt->fmt_getsize) ret = (*fmt->fmt_getsize)(task, flags, cpu, arg, size);
	return ret;
}


static inline int
pfm_buf_fmt_validate(pfm_buffer_fmt_t *fmt, struct task_struct *task, unsigned int flags,
		     int cpu, void *arg)
{
	int ret = 0;
	if (fmt->fmt_validate) ret = (*fmt->fmt_validate)(task, flags, cpu, arg);
	return ret;
}

static inline int
pfm_buf_fmt_init(pfm_buffer_fmt_t *fmt, struct task_struct *task, void *buf, unsigned int flags,
		     int cpu, void *arg)
{
	int ret = 0;
	if (fmt->fmt_init) ret = (*fmt->fmt_init)(task, buf, flags, cpu, arg);
	return ret;
}

static inline int
pfm_buf_fmt_restart(pfm_buffer_fmt_t *fmt, struct task_struct *task, pfm_ovfl_ctrl_t *ctrl, void *buf, struct pt_regs *regs)
{
	int ret = 0;
	if (fmt->fmt_restart) ret = (*fmt->fmt_restart)(task, ctrl, buf, regs);
	return ret;
}

static inline int
pfm_buf_fmt_restart_active(pfm_buffer_fmt_t *fmt, struct task_struct *task, pfm_ovfl_ctrl_t *ctrl, void *buf, struct pt_regs *regs)
{
	int ret = 0;
	if (fmt->fmt_restart_active) ret = (*fmt->fmt_restart_active)(task, ctrl, buf, regs);
	return ret;
}

static pfm_buffer_fmt_t *
__pfm_find_buffer_fmt(pfm_uuid_t uuid)
{
	struct list_head * pos;
	pfm_buffer_fmt_t * entry;

	list_for_each(pos, &pfm_buffer_fmt_list) {
		entry = list_entry(pos, pfm_buffer_fmt_t, fmt_list);
		if (pfm_uuid_cmp(uuid, entry->fmt_uuid) == 0)
			return entry;
	}
	return NULL;
}
 
/*
 * find a buffer format based on its uuid
 */
static pfm_buffer_fmt_t *
pfm_find_buffer_fmt(pfm_uuid_t uuid)
{
	pfm_buffer_fmt_t * fmt;
	spin_lock(&pfm_buffer_fmt_lock);
	fmt = __pfm_find_buffer_fmt(uuid);
	spin_unlock(&pfm_buffer_fmt_lock);
	return fmt;
}
 
int
pfm_register_buffer_fmt(pfm_buffer_fmt_t *fmt)
{
	int ret = 0;

	/* some sanity checks */
	if (fmt == NULL || fmt->fmt_name == NULL) return -EINVAL;

	/* we need at least a handler */
	if (fmt->fmt_handler == NULL) return -EINVAL;

	/*
	 * XXX: need check validity of fmt_arg_size
	 */

	spin_lock(&pfm_buffer_fmt_lock);

	if (__pfm_find_buffer_fmt(fmt->fmt_uuid)) {
		printk(KERN_ERR "perfmon: duplicate sampling format: %s\n", fmt->fmt_name);
		ret = -EBUSY;
		goto out;
	} 
	list_add(&fmt->fmt_list, &pfm_buffer_fmt_list);
	printk(KERN_INFO "perfmon: added sampling format %s\n", fmt->fmt_name);

out:
	spin_unlock(&pfm_buffer_fmt_lock);
 	return ret;
}
EXPORT_SYMBOL(pfm_register_buffer_fmt);

int
pfm_unregister_buffer_fmt(pfm_uuid_t uuid)
{
	pfm_buffer_fmt_t *fmt;
	int ret = 0;

	spin_lock(&pfm_buffer_fmt_lock);

	fmt = __pfm_find_buffer_fmt(uuid);
	if (!fmt) {
		printk(KERN_ERR "perfmon: cannot unregister format, not found\n");
		ret = -EINVAL;
		goto out;
	}
	list_del_init(&fmt->fmt_list);
	printk(KERN_INFO "perfmon: removed sampling format: %s\n", fmt->fmt_name);

out:
	spin_unlock(&pfm_buffer_fmt_lock);
	return ret;

}
EXPORT_SYMBOL(pfm_unregister_buffer_fmt);

1335
1336
extern void update_pal_halt_status(int);

Linus Torvalds's avatar
Linus Torvalds committed
1337
1338
1339
1340
1341
static int
pfm_reserve_session(struct task_struct *task, int is_syswide, unsigned int cpu)
{
	unsigned long flags;
	/*
1342
	 * validity checks on cpu_mask have been done upstream
Linus Torvalds's avatar
Linus Torvalds committed
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
	 */
	LOCK_PFS(flags);

	DPRINT(("in sys_sessions=%u task_sessions=%u dbregs=%u syswide=%d cpu=%u\n",
		pfm_sessions.pfs_sys_sessions,
		pfm_sessions.pfs_task_sessions,
		pfm_sessions.pfs_sys_use_dbregs,
		is_syswide,
		cpu));

	if (is_syswide) {
		/*
		 * cannot mix system wide and per-task sessions
		 */
		if (pfm_sessions.pfs_task_sessions > 0UL) {
			DPRINT(("system wide not possible, %u conflicting task_sessions\n",
			  	pfm_sessions.pfs_task_sessions));
			goto abort;
		}

		if (pfm_sessions.pfs_sys_session[cpu]) goto error_conflict;

		DPRINT(("reserving system wide session on CPU%u currently on CPU%u\n", cpu, smp_processor_id()));

		pfm_sessions.pfs_sys_session[cpu] = task;

		pfm_sessions.pfs_sys_sessions++ ;

	} else {
		if (pfm_sessions.pfs_sys_sessions) goto abort;
		pfm_sessions.pfs_task_sessions++;
	}

	DPRINT(("out sys_sessions=%u task_sessions=%u dbregs=%u syswide=%d cpu=%u\n",
		pfm_sessions.pfs_sys_sessions,
		pfm_sessions.pfs_task_sessions,
		pfm_sessions.pfs_sys_use_dbregs,
		is_syswide,
		cpu));

1383
1384
1385
1386
1387
	/*
	 * disable default_idle() to go to PAL_HALT
	 */
	update_pal_halt_status(0);

Linus Torvalds's avatar
Linus Torvalds committed
1388
1389
1390
1391
1392
1393
	UNLOCK_PFS(flags);

	return 0;

error_conflict:
	DPRINT(("system wide not possible, conflicting session [%d] on CPU%d\n",
1394
  		task_pid_nr(pfm_sessions.pfs_sys_session[cpu]),
Tony Luck's avatar
Tony Luck committed
1395
		cpu));
Linus Torvalds's avatar
Linus Torvalds committed
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
abort:
	UNLOCK_PFS(flags);

	return -EBUSY;

}

static int
pfm_unreserve_session(pfm_context_t *ctx, int is_syswide, unsigned int cpu)
{
	unsigned long flags;
	/*
1408
	 * validity checks on cpu_mask have been done upstream
Linus Torvalds's avatar
Linus Torvalds committed
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
	 */
	LOCK_PFS(flags);

	DPRINT(("in sys_sessions=%u task_sessions=%u dbregs=%u syswide=%d cpu=%u\n",
		pfm_sessions.pfs_sys_sessions,
		pfm_sessions.pfs_task_sessions,
		pfm_sessions.pfs_sys_use_dbregs,
		is_syswide,
		cpu));


	if (is_syswide) {
		pfm_sessions.pfs_sys_session[cpu] = NULL;
		/*
		 * would not work with perfmon+more than one bit in cpu_mask
		 */
		if (ctx && ctx->ctx_fl_using_dbreg) {
			if (pfm_sessions.pfs_sys_use_dbregs == 0) {
				printk(KERN_ERR "perfmon: invalid release for ctx %p sys_use_dbregs=0\n", ctx);
			} else {
				pfm_sessions.pfs_sys_use_dbregs--;
			}
		}
		pfm_sessions.pfs_sys_sessions--;
	} else {
		pfm_sessions.pfs_task_sessions--;
	}
	DPRINT(("out sys_sessions=%u task_sessions=%u dbregs=%u syswide=%d cpu=%u\n",
		pfm_sessions.pfs_sys_sessions,
		pfm_sessions.pfs_task_sessions,
		pfm_sessions.pfs_sys_use_dbregs,
		is_syswide,
		cpu));

1443
1444
1445
1446
1447
1448
	/*
	 * if possible, enable default_idle() to go into PAL_HALT
	 */
	if (pfm_sessions.pfs_task_sessions == 0 && pfm_sessions.pfs_sys_sessions == 0)
		update_pal_halt_status(1);

Linus Torvalds's avatar
Linus Torvalds committed
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
	UNLOCK_PFS(flags);

	return 0;
}

/*
 * removes virtual mapping of the sampling buffer.
 * IMPORTANT: cannot be called with interrupts disable, e.g. inside
 * a PROTECT_CTX() section.
 */
static int
pfm_remove_smpl_mapping(struct task_struct *task, void *vaddr, unsigned long size)
{
	int r;

	/* sanity checks */
	if (task->mm == NULL || size == 0UL || vaddr == NULL) {
1466
		printk(KERN_ERR "perfmon: pfm_remove_smpl_mapping [%d] invalid context mm=%p\n", task_pid_nr(task), task->mm);
Linus Torvalds's avatar
Linus Torvalds committed
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
		return -EINVAL;
	}

	DPRINT(("smpl_vaddr=%p size=%lu\n", vaddr, size));

	/*
	 * does the actual unmapping
	 */
	down_write(&task->mm->mmap_sem);

	DPRINT(("down_write done smpl_vaddr=%p size=%lu\n", vaddr, size));

	r = pfm_do_munmap(task->mm, (unsigned long)vaddr, size, 0);

	up_write(&task->mm->mmap_sem);
	if (r !=0) {
1483
		printk(KERN_ERR "perfmon: [%d] unable to unmap sampling buffer @%p size=%lu\n", task_pid_nr(task), vaddr, size);
Linus Torvalds's avatar
Linus Torvalds committed
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
	}

	DPRINT(("do_unmap(%p, %lu)=%d\n", vaddr, size, r));

	return 0;
}

/*
 * free actual physical storage used by sampling buffer
 */
#if 0
static int
pfm_free_smpl_buffer(pfm_context_t *ctx)
{
	pfm_buffer_fmt_t *fmt;

	if (ctx->ctx_smpl_hdr == NULL) goto invalid_free;

	/*
	 * we won't use the buffer format anymore
	 */
	fmt = ctx->ctx_buf_fmt;

	DPRINT(("sampling buffer @%p size %lu vaddr=%p\n",
		ctx->ctx_smpl_hdr,
		ctx->ctx_smpl_size,
		ctx->ctx_smpl_vaddr));

	pfm_buf_fmt_exit(fmt, current, NULL, NULL);

	/*
	 * free the buffer
	 */
	pfm_rvfree(ctx->ctx_smpl_hdr, ctx->ctx_smpl_size);

	ctx->ctx_smpl_hdr  = NULL;
	ctx->ctx_smpl_size = 0UL;

	return 0;

invalid_free:
1525
	printk(KERN_ERR "perfmon: pfm_free_smpl_buffer [%d] no buffer\n", task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
	return -EINVAL;
}
#endif

static inline void
pfm_exit_smpl_buffer(pfm_buffer_fmt_t *fmt)
{
	if (fmt == NULL) return;

	pfm_buf_fmt_exit(fmt, current, NULL, NULL);

}

/*
 * pfmfs should _never_ be mounted by userland - too much of security hassle,
 * no real gain from having the whole whorehouse mounted. So we don't need
 * any operations on the root directory. However, we need a non-trivial
 * d_name - pfm: will go nicely and kill the special-casing in procfs.
 */
static struct vfsmount *pfmfs_mnt;

static int __init
init_pfm_fs(void)
{
	int err = register_filesystem(&pfm_fs_type);
	if (!err) {
		pfmfs_mnt = kern_mount(&pfm_fs_type);
		err = PTR_ERR(pfmfs_mnt);
		if (IS_ERR(pfmfs_mnt))
			unregister_filesystem(&pfm_fs_type);
		else
			err = 0;
	}
	return err;
}

static ssize_t
pfm_read(struct file *filp, char __user *buf, size_t size, loff_t *ppos)
{
	pfm_context_t *ctx;
	pfm_msg_t *msg;
	ssize_t ret;
	unsigned long flags;
  	DECLARE_WAITQUEUE(wait, current);
	if (PFM_IS_FILE(filp) == 0) {
1571
		printk(KERN_ERR "perfmon: pfm_poll: bad magic [%d]\n", task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1572
1573
1574
1575
1576
		return -EINVAL;
	}

	ctx = (pfm_context_t *)filp->private_data;
	if (ctx == NULL) {
1577
		printk(KERN_ERR "perfmon: pfm_read: NULL ctx [%d]\n", task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
		return -EINVAL;
	}

	/*
	 * check even when there is no message
	 */
	if (size < sizeof(pfm_msg_t)) {
		DPRINT(("message is too small ctx=%p (>=%ld)\n", ctx, sizeof(pfm_msg_t)));
		return -EINVAL;
	}

	PROTECT_CTX(ctx, flags);

  	/*
	 * put ourselves on the wait queue
	 */
  	add_wait_queue(&ctx->ctx_msgq_wait, &wait);


  	for(;;) {
		/*
		 * check wait queue
		 */

  		set_current_state(TASK_INTERRUPTIBLE);

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

		ret = 0;
		if(PFM_CTXQ_EMPTY(ctx) == 0) break;

		UNPROTECT_CTX(ctx, flags);

		/*
		 * check non-blocking read
		 */
      		ret = -EAGAIN;
		if(filp->f_flags & O_NONBLOCK) break;

		/*
		 * check pending signals
		 */
		if(signal_pending(current)) {
			ret = -EINTR;
			break;
		}
      		/*
		 * no message, so wait
		 */
      		schedule();

		PROTECT_CTX(ctx, flags);
	}
1631
	DPRINT(("[%d] back to running ret=%ld\n", task_pid_nr(current), ret));
Linus Torvalds's avatar
Linus Torvalds committed
1632
1633
1634
1635
1636
1637
1638
1639
  	set_current_state(TASK_RUNNING);
	remove_wait_queue(&ctx->ctx_msgq_wait, &wait);

	if (ret < 0) goto abort;

	ret = -EINVAL;
	msg = pfm_get_next_msg(ctx);
	if (msg == NULL) {
1640
		printk(KERN_ERR "perfmon: pfm_read no msg for ctx=%p [%d]\n", ctx, task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1641
1642
1643
		goto abort_locked;
	}

1644
	DPRINT(("fd=%d type=%d\n", msg->pfm_gen_msg.msg_ctx_fd, msg->pfm_gen_msg.msg_type));
Linus Torvalds's avatar
Linus Torvalds committed
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670

	ret = -EFAULT;
  	if(copy_to_user(buf, msg, sizeof(pfm_msg_t)) == 0) ret = sizeof(pfm_msg_t);

abort_locked:
	UNPROTECT_CTX(ctx, flags);
abort:
	return ret;
}

static ssize_t
pfm_write(struct file *file, const char __user *ubuf,
			  size_t size, loff_t *ppos)
{
	DPRINT(("pfm_write called\n"));
	return -EINVAL;
}

static unsigned int
pfm_poll(struct file *filp, poll_table * wait)
{
	pfm_context_t *ctx;
	unsigned long flags;
	unsigned int mask = 0;

	if (PFM_IS_FILE(filp) == 0) {
1671
		printk(KERN_ERR "perfmon: pfm_poll: bad magic [%d]\n", task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1672
1673
1674
1675
1676
		return 0;
	}

	ctx = (pfm_context_t *)filp->private_data;
	if (ctx == NULL) {
1677
		printk(KERN_ERR "perfmon: pfm_poll: NULL ctx [%d]\n", task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
		return 0;
	}


	DPRINT(("pfm_poll ctx_fd=%d before poll_wait\n", ctx->ctx_fd));

	poll_wait(filp, &ctx->ctx_msgq_wait, wait);

	PROTECT_CTX(ctx, flags);

	if (PFM_CTXQ_EMPTY(ctx) == 0)
		mask =  POLLIN | POLLRDNORM;

	UNPROTECT_CTX(ctx, flags);

	DPRINT(("pfm_poll ctx_fd=%d mask=0x%x\n", ctx->ctx_fd, mask));

	return mask;
}

static int
pfm_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg)
{
	DPRINT(("pfm_ioctl called\n"));
	return -EINVAL;
}

/*
 * interrupt cannot be masked when coming here
 */
static inline int
pfm_do_fasync(int fd, struct file *filp, pfm_context_t *ctx, int on)
{
	int ret;

	ret = fasync_helper (fd, filp, on, &ctx->ctx_async_queue);

	DPRINT(("pfm_fasync called by [%d] on ctx_fd=%d on=%d async_queue=%p ret=%d\n",
1716
		task_pid_nr(current),
Linus Torvalds's avatar
Linus Torvalds committed
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
		fd,
		on,
		ctx->ctx_async_queue, ret));

	return ret;
}

static int
pfm_fasync(int fd, struct file *filp, int on)
{
	pfm_context_t *ctx;
	int ret;

	if (PFM_IS_FILE(filp) == 0) {
1731
		printk(KERN_ERR "perfmon: pfm_fasync bad magic [%d]\n", task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1732
1733
1734
1735
1736
		return -EBADF;
	}

	ctx = (pfm_context_t *)filp->private_data;
	if (ctx == NULL) {
1737
		printk(KERN_ERR "perfmon: pfm_fasync NULL ctx [%d]\n", task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
		return -EBADF;
	}
	/*
	 * we cannot mask interrupts during this call because this may
	 * may go to sleep if memory is not readily avalaible.
	 *
	 * We are protected from the conetxt disappearing by the get_fd()/put_fd()
	 * done in caller. Serialization of this function is ensured by caller.
	 */
	ret = pfm_do_fasync(fd, filp, ctx, on);


	DPRINT(("pfm_fasync called on ctx_fd=%d on=%d async_queue=%p ret=%d\n",
		fd,
		on,
		ctx->ctx_async_queue, ret));

	return ret;
}

#ifdef CONFIG_SMP
/*
 * this function is exclusively called from pfm_close().
 * The context is not protected at that time, nor are interrupts
 * on the remote CPU. That's necessary to avoid deadlocks.
 */
static void
pfm_syswide_force_stop(void *info)
{
	pfm_context_t   *ctx = (pfm_context_t *)info;
Al Viro's avatar
Al Viro committed
1768
	struct pt_regs *regs = task_pt_regs(current);
Linus Torvalds's avatar
Linus Torvalds committed
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
	struct task_struct *owner;
	unsigned long flags;
	int ret;

	if (ctx->ctx_cpu != smp_processor_id()) {
		printk(KERN_ERR "perfmon: pfm_syswide_force_stop for CPU%d  but on CPU%d\n",
			ctx->ctx_cpu,
			smp_processor_id());
		return;
	}
	owner = GET_PMU_OWNER();
	if (owner != ctx->ctx_task) {
		printk(KERN_ERR "perfmon: pfm_syswide_force_stop CPU%d unexpected owner [%d] instead of [%d]\n",
			smp_processor_id(),
1783
			task_pid_nr(owner), task_pid_nr(ctx->ctx_task));
Linus Torvalds's avatar
Linus Torvalds committed
1784
1785
1786
1787
1788
1789
1790
1791
1792
		return;
	}
	if (GET_PMU_CTX() != ctx) {
		printk(KERN_ERR "perfmon: pfm_syswide_force_stop CPU%d unexpected ctx %p instead of %p\n",
			smp_processor_id(),
			GET_PMU_CTX(), ctx);
		return;
	}

1793
	DPRINT(("on CPU%d forcing system wide stop for [%d]\n", smp_processor_id(), task_pid_nr(ctx->ctx_task)));
Linus Torvalds's avatar
Linus Torvalds committed
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
	/*
	 * the context is already protected in pfm_close(), we simply
	 * need to mask interrupts to avoid a PMU interrupt race on
	 * this CPU
	 */
	local_irq_save(flags);

	ret = pfm_context_unload(ctx, NULL, 0, regs);
	if (ret) {
		DPRINT(("context_unload returned %d\n", ret));
	}

	/*
	 * unmask interrupts, PMU interrupts are now spurious here
	 */
	local_irq_restore(flags);
}

static void
pfm_syswide_cleanup_other_cpu(pfm_context_t *ctx)
{
	int ret;

	DPRINT(("calling CPU%d for cleanup\n", ctx->ctx_cpu));
1818
	ret = smp_call_function_single(ctx->ctx_cpu, pfm_syswide_force_stop, ctx, 1);
Linus Torvalds's avatar
Linus Torvalds committed
1819
1820
1821
1822
1823
1824
1825
1826
1827
	DPRINT(("called CPU%d for cleanup ret=%d\n", ctx->ctx_cpu, ret));
}
#endif /* CONFIG_SMP */

/*
 * called for each close(). Partially free resources.
 * When caller is self-monitoring, the context is unloaded.
 */
static int
1828
pfm_flush(struct file *filp, fl_owner_t id)
Linus Torvalds's avatar
Linus Torvalds committed
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
{
	pfm_context_t *ctx;
	struct task_struct *task;
	struct pt_regs *regs;
	unsigned long flags;
	unsigned long smpl_buf_size = 0UL;
	void *smpl_buf_vaddr = NULL;
	int state, is_system;

	if (PFM_IS_FILE(filp) == 0) {
		DPRINT(("bad magic for\n"));
		return -EBADF;
	}

	ctx = (pfm_context_t *)filp->private_data;
	if (ctx == NULL) {
1845
		printk(KERN_ERR "perfmon: pfm_flush: NULL ctx [%d]\n", task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1846
1847
1848
1849
1850
1851
		return -EBADF;
	}

	/*
	 * remove our file from the async queue, if we use this mode.
	 * This can be done without the context being protected. We come
1852
	 * here when the context has become unreachable by other tasks.
Linus Torvalds's avatar
Linus Torvalds committed
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
	 *
	 * We may still have active monitoring at this point and we may
	 * end up in pfm_overflow_handler(). However, fasync_helper()
	 * operates with interrupts disabled and it cleans up the
	 * queue. If the PMU handler is called prior to entering
	 * fasync_helper() then it will send a signal. If it is
	 * invoked after, it will find an empty queue and no
	 * signal will be sent. In both case, we are safe
	 */
	PROTECT_CTX(ctx, flags);

	state     = ctx->ctx_state;
	is_system = ctx->ctx_fl_system;

	task = PFM_CTX_TASK(ctx);
Al Viro's avatar
Al Viro committed
1868
	regs = task_pt_regs(task);
Linus Torvalds's avatar
Linus Torvalds committed
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987

	DPRINT(("ctx_state=%d is_current=%d\n",
		state,
		task == current ? 1 : 0));

	/*
	 * if state == UNLOADED, then task is NULL
	 */

	/*
	 * we must stop and unload because we are losing access to the context.
	 */
	if (task == current) {
#ifdef CONFIG_SMP
		/*
		 * the task IS the owner but it migrated to another CPU: that's bad
		 * but we must handle this cleanly. Unfortunately, the kernel does
		 * not provide a mechanism to block migration (while the context is loaded).
		 *
		 * We need to release the resource on the ORIGINAL cpu.
		 */
		if (is_system && ctx->ctx_cpu != smp_processor_id()) {

			DPRINT(("should be running on CPU%d\n", ctx->ctx_cpu));
			/*
			 * keep context protected but unmask interrupt for IPI
			 */
			local_irq_restore(flags);

			pfm_syswide_cleanup_other_cpu(ctx);

			/*
			 * restore interrupt masking
			 */
			local_irq_save(flags);

			/*
			 * context is unloaded at this point
			 */
		} else
#endif /* CONFIG_SMP */
		{

			DPRINT(("forcing unload\n"));
			/*
		 	* stop and unload, returning with state UNLOADED
		 	* and session unreserved.
		 	*/
			pfm_context_unload(ctx, NULL, 0, regs);

			DPRINT(("ctx_state=%d\n", ctx->ctx_state));
		}
	}

	/*
	 * remove virtual mapping, if any, for the calling task.
	 * cannot reset ctx field until last user is calling close().
	 *
	 * ctx_smpl_vaddr must never be cleared because it is needed
	 * by every task with access to the context
	 *
	 * When called from do_exit(), the mm context is gone already, therefore
	 * mm is NULL, i.e., the VMA is already gone  and we do not have to
	 * do anything here
	 */
	if (ctx->ctx_smpl_vaddr && current->mm) {
		smpl_buf_vaddr = ctx->ctx_smpl_vaddr;
		smpl_buf_size  = ctx->ctx_smpl_size;
	}

	UNPROTECT_CTX(ctx, flags);

	/*
	 * if there was a mapping, then we systematically remove it
	 * at this point. Cannot be done inside critical section
	 * because some VM function reenables interrupts.
	 *
	 */
	if (smpl_buf_vaddr) pfm_remove_smpl_mapping(current, smpl_buf_vaddr, smpl_buf_size);

	return 0;
}
/*
 * called either on explicit close() or from exit_files(). 
 * Only the LAST user of the file gets to this point, i.e., it is
 * called only ONCE.
 *
 * IMPORTANT: we get called ONLY when the refcnt on the file gets to zero 
 * (fput()),i.e, last task to access the file. Nobody else can access the 
 * file at this point.
 *
 * When called from exit_files(), the VMA has been freed because exit_mm()
 * is executed before exit_files().
 *
 * When called from exit_files(), the current task is not yet ZOMBIE but we
 * flush the PMU state to the context. 
 */
static int
pfm_close(struct inode *inode, struct file *filp)
{
	pfm_context_t *ctx;
	struct task_struct *task;
	struct pt_regs *regs;
  	DECLARE_WAITQUEUE(wait, current);
	unsigned long flags;
	unsigned long smpl_buf_size = 0UL;
	void *smpl_buf_addr = NULL;
	int free_possible = 1;
	int state, is_system;

	DPRINT(("pfm_close called private=%p\n", filp->private_data));

	if (PFM_IS_FILE(filp) == 0) {
		DPRINT(("bad magic\n"));
		return -EBADF;
	}
	
	ctx = (pfm_context_t *)filp->private_data;
	if (ctx == NULL) {
1988
		printk(KERN_ERR "perfmon: pfm_close: NULL ctx [%d]\n", task_pid_nr(current));
Linus Torvalds's avatar
Linus Torvalds committed
1989
1990
1991
1992
1993
1994
1995
1996
1997
		return -EBADF;
	}

	PROTECT_CTX(ctx, flags);

	state     = ctx->ctx_state;
	is_system = ctx->ctx_fl_system;

	task = PFM_CTX_TASK(ctx);
Al Viro's avatar
Al Viro committed
1998
	regs = task_pt_regs(task);
Linus Torvalds's avatar
Linus Torvalds committed
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041

	DPRINT(("ctx_state=%d is_current=%d\n", 
		state,
		task == current ? 1 : 0));

	/*
	 * if task == current, then pfm_flush() unloaded the context
	 */
	if (state == PFM_CTX_UNLOADED) goto doit;

	/*
	 * context is loaded/masked and task != current, we need to
	 * either force an unload or go zombie
	 */

	/*
	 * The task is currently blocked or will block after an overflow.
	 * we must force it to wakeup to get out of the
	 * MASKED state and transition to the unloaded state by itself.
	 *
	 * This situation is only possible for per-task mode
	 */
	if (state == PFM_CTX_MASKED && CTX_OVFL_NOBLOCK(ctx) == 0) {

		/*
		 * set a "partial" zombie state to be checked
		 * upon return from down() in pfm_handle_work().
		 *
		 * We cannot use the ZOMBIE state, because it is checked
		 * by pfm_load_regs() which is called upon wakeup from down().
		 * In such case, it would free the context and then we would
		 * return to pfm_handle_work() which would access the
		 * stale context. Instead, we set a flag invisible to pfm_load_regs()
		 * but visible to pfm_handle_work().
		 *
		 * For some window of time, we have a zombie context with
		 * ctx_state = MASKED  and not ZOMBIE
		 */
		ctx->ctx_fl_going_zombie = 1;

		/*
		 * force task to wake up from MASKED state
		 */
2042
		complete(&ctx->ctx_restart_done);
Linus Torvalds's avatar
Linus Torvalds committed
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084

		DPRINT(("waking up ctx_state=%d\n", state));

		/*
		 * put ourself to sleep waiting for the other
		 * task to report completion
		 *
		 * the context is protected by mutex, therefore there
		 * is no risk of being notified of completion before
		 * begin actually on the waitq.
		 */
  		set_current_state(TASK_INTERRUPTIBLE);
  		add_wait_queue(&ctx->ctx_zombieq, &wait);

		UNPROTECT_CTX(ctx, flags);

		/*
		 * XXX: check for signals :
		 * 	- ok for explicit close
		 * 	- not ok when coming from exit_files()
		 */
      		schedule();


		PROTECT_CTX(ctx, flags);


		remove_wait_queue(&ctx->ctx_zombieq, &wait);
  		set_current_state(TASK_RUNNING);

		/*
		 * context is unloaded at this point
		 */
		DPRINT(("after zombie wakeup ctx_state=%d for\n", state));
	}
	else if (task != current) {
#ifdef CONFIG_SMP
		/*
	 	 * switch context to zombie state
	 	 */
		ctx->ctx_state = PFM_CTX_ZOMBIE;

2085
		DPRINT(("zombie ctx for [%d]\n", task_pid_nr(task)));
Linus Torvalds's avatar
Linus Torvalds committed
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
		/*
		 * cannot free the context on the spot. deferred until
		 * the task notices the ZOMBIE state
		 */
		free_possible = 0;
#else
		pfm_context_unload(ctx, NULL, 0, regs);
#endif
	}

doit:
	/* reload state, may have changed during  opening of critical section */
	state = ctx->ctx_state;

	/*
	 * the context is still attached to a task (possibly current)
	 * we cannot destroy it right now
	 */

	/*
	 * we must free the sampling buffer right here because
	 * we cannot rely on it being cleaned up later by the
	 * monitored task. It is not possible to free vmalloc'ed
	 * memory in pfm_load_regs(). Instead, we remove the buffer
	 * now. should there be subsequent PMU overflow originally
	 * meant for sampling, the will be converted to spurious
	 * and that's fine because the monitoring tools is gone anyway.
	 */
	if (ctx->ctx_smpl_hdr) {
		smpl_buf_addr = ctx->ctx_smpl_hdr;
		smpl_buf_size = ctx->ctx_smpl_size;
		/* no more sampling */
		ctx->ctx_smpl_hdr = NULL;
		ctx->ctx_fl_is_sampling = 0;
	}

	DPRINT(("ctx_state=%d free_possible=%d addr=%p size=%lu\n",
		state,
		free_possible,
		smpl_buf_addr,
		smpl_buf_size));

	if (smpl_buf_addr) pfm_exit_smpl_buffer(ctx->ctx_buf_fmt);

	/*
	 * UNLOADED that the session has already been unreserved.
	 */
	if (state == PFM_CTX_ZOMBIE) {
		pfm_unreserve_session(ctx, ctx->ctx_fl_system , ctx->ctx_cpu);
	}

	/*
	 * disconnect file descriptor from context must be done
	 * before we unlock.
	 */
	filp->private_data = NULL;

	/*
2144
	 * if we free on the spot, the context is now completely unreachable
Linus Torvalds's avatar
Linus Torvalds committed
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
	 * from the callers side. The monitored task side is also cut, so we
	 * can freely cut.
	 *
	 * If we have a deferred free, only the caller side is disconnected.
	 */
	UNPROTECT_CTX(ctx, flags);

	/*
	 * All memory free operations (especially for vmalloc'ed memory)
	 * MUST be done with interrupts ENABLED.
	 */
	if (smpl_buf_addr)  pfm_rvfree(smpl_buf_addr, smpl_buf_size);

	/*
	 * return the memory used by the context
	 */
	if (free_possible) pfm_context_free(ctx);

	return 0;
}

static int
pfm_no_open(struct inode *irrelevant, struct file *dontcare)
{
	DPRINT(("pfm_no_open called\n"));
	return -ENXIO;
}



2175
static const struct file_operations pfm_file_ops = {
Linus Torvalds's avatar
Linus Torvalds committed
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
	.llseek   = no_llseek,
	.read     = pfm_read,
	.write    = pfm_write,
	.poll     = pfm_poll,
	.ioctl    = pfm_ioctl,
	.open     = pfm_no_open,	/* special open code to disallow open via /proc */
	.fasync   = pfm_fasync,
	.release  = pfm_close,
	.flush	  = pfm_flush
};

static int
pfmfs_delete_dentry(struct dentry *dentry)
{
	return 1;
}

Al Viro's avatar
Al Viro committed
2193
static const struct dentry_operations pfmfs_dentry_operations = {
Linus Torvalds's avatar
Linus Torvalds committed
2194
2195
2196
2197
	.d_delete = pfmfs_delete_dentry,
};


2198
2199
static struct file *
pfm_alloc_file(pfm_context_t *ctx)
Linus Torvalds's avatar
Linus Torvalds committed
2200
{
2201
2202
2203
	struct file *file;
	struct inode *inode;
	struct dentry *dentry;
Linus Torvalds's avatar
Linus Torvalds committed
2204
2205
2206
2207
2208
2209
2210
	char name[32];
	struct qstr this;

	/*
	 * allocate a new inode
	 */
	inode = new_inode(pfmfs_mnt->mnt_sb);
2211
2212
	if (!inode)
		return ERR_PTR(-ENOMEM);
Linus Torvalds's avatar
Linus Torvalds committed
2213
2214
2215
2216

	DPRINT(("new inode ino=%ld @%p\n", inode->i_ino, inode));

	inode->i_mode = S_IFCHR|S_IRUGO;
2217
2218
	inode->i_uid  = current_fsuid();
	inode->i_gid  = current_fsgid();
Linus Torvalds's avatar
Linus Torvalds committed
2219
2220
2221
2222
2223
2224
2225
2226
2227

	sprintf(name, "[%lu]", inode->i_ino);
	this.name = name;
	this.len  = strlen(name);
	this.hash = inode->i_ino;

	/*
	 * allocate a new dcache entry
	 */
2228
2229
2230
2231
2232
	dentry = d_alloc(pfmfs_mnt->mnt_sb->s_root, &this);
	if (!dentry) {
		iput(inode);
		return ERR_PTR(-ENOMEM);
	}
Linus Torvalds's avatar
Linus Torvalds committed
2233

2234
2235
	dentry->d_op = &pfmfs_dentry_operations;
	d_add(dentry, inode);
Linus Torvalds's avatar
Linus Torvalds committed
2236

2237
2238
2239
2240
2241
	file = alloc_file(pfmfs_mnt, dentry, FMODE_READ, &pfm_file_ops);
	if (!file) {
		dput(dentry);
		return ERR_PTR(-ENFILE);
	}
Linus Torvalds's avatar
Linus Torvalds committed
2242
2243

	file->f_flags = O_RDONLY;
2244
	file->private_data = ctx;
Linus Torvalds's avatar
Linus Torvalds committed
2245

2246
	return file;
Linus Torvalds's avatar
Linus Torvalds committed
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
}

static int
pfm_remap_buffer(struct vm_area_struct *vma, unsigned long buf, unsigned long addr, unsigned long size)
{
	DPRINT(("CPU%d buf=0x%lx addr=0x%lx size=%ld\n", smp_processor_id(), buf, addr, size));

	while (size > 0) {
		unsigned long pfn = ia64_tpa(buf) >> PAGE_SHIFT;


		if (remap_pfn_range(vma, addr, pfn, PAGE_SIZE, PAGE_READONLY))
			return -ENOMEM;

		addr  += PAGE_SIZE;
		buf   += PAGE_SIZE;
		size  -= PAGE_SIZE;
	}
	return 0;
}

/*
 * allocate a sampling buffer and remaps it into the user address space of the task
 */
static int
Nick Piggin's avatar
Nick Piggin committed
2272
pfm_smpl_buffer_alloc(struct task_struct *task, struct file *filp, pfm_context_t *ctx, unsigned long rsize, void **user_vaddr)
Linus Torvalds's avatar
Linus Torvalds committed
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
{
	struct mm_struct *mm = task->mm;
	struct vm_area_struct *vma = NULL;
	unsigned long size;
	void *smpl_buf;


	/*
	 * the fixed header + requested size and align to page boundary
	 */
	size = PAGE_ALIGN(rsize);

	DPRINT(("sampling buffer rsize=%lu size=%lu bytes\n", rsize, size));

	/*
	 * check requested size to avoid Denial-of-service attacks
	 * XXX: may have to refine this test
	 * Check against address space limit.
	 *
	 * if ((mm->total_vm << PAGE_SHIFT) + len> task->rlim[RLIMIT_AS].rlim_cur)
	 * 	return -ENOMEM;
	 */
	if (size > task->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
		return -ENOMEM;

	/*
	 * We do the easy to undo allocations first.
 	 *
	 * pfm_rvmalloc(), clears the buffer, so there is no leak
	 */
	smpl_buf = pfm_rvmalloc(size);
	if (smpl_buf == NULL) {
		DPRINT(("Can't allocate sampling buffer\n"));
		return -ENOMEM;
	}

	DPRINT(("smpl_buf @%p\n", smpl_buf));

	/* allocate vma */
2312
	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
Linus Torvalds's avatar
Linus Torvalds committed
2313
2314
2315
2316
2317
2318
2319
2320
2321
	if (!vma) {
		DPRINT(("Cannot allocate vma\n"));
		goto error_kmem;
	}

	/*
	 * partially initialize the vma for the sampling buffer
	 */
	vma->vm_mm	     = mm;
Nick Piggin's avatar
Nick Piggin committed
2322
	vma->vm_file	     = filp;
Linus Torvalds's avatar
Linus Torvalds committed
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
	vma->vm_flags	     = VM_READ| VM_MAYREAD |VM_RESERVED;
	vma->vm_page_prot    = PAGE_READONLY; /* XXX may need to change */

	/*
	 * Now we have everything we need and we can initialize
	 * and connect all the data structures
	 */

	ctx->ctx_smpl_hdr   = smpl_buf;
	ctx->ctx_smpl_size  = size; /* aligned size */

	/*
	 * Let's do the difficult operations next.
	 *
	 * now we atomically find some area in the address space and
	 * remap the buffer in it.
	 */
	down_write(&task->mm->mmap_sem);

	/* find some free area in address space, must have mmap sem held */
	vma->vm_start = pfm_get_unmapped_area(NULL, 0, size, 0, MAP_PRIVATE|MAP_ANONYMOUS, 0);
	if (vma->vm_start == 0UL) {
		DPRINT(("Cannot find unmapped area for size %ld\n", size));
		up_write(&task->mm->mmap_sem);
		goto error;
	}
	vma->vm_end = vma->vm_start + size;
	vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;

	DPRINT(("aligned size=%ld, hdr=%p mapped @0x%lx\n", size, ctx->ctx_smpl_hdr, vma->vm_start));

	/* can only be applied to current task, need to have the mm semaphore held when called */
	if (pfm_remap_buffer(vma, (unsigned long)smpl_buf, vma->vm_start, size)) {
		DPRINT(("Can't remap buffer\n"));
		up_write(&task->mm->mmap_sem);
		goto error;
	}

Nick Piggin's avatar
Nick Piggin committed
2361
2362
	get_file(filp);

Linus Torvalds's avatar
Linus Torvalds committed
2363
2364
2365
2366
2367
2368
2369
	/*
	 * now insert the vma in the vm list for the process, must be
	 * done with mmap lock held
	 */
	insert_vm_struct(mm, vma);

	mm->total_vm  += size >> PAGE_SHIFT;
2370
2371
	vm_stat_account(vma->vm_mm, vma->vm_flags, vma->vm_file,
							vma_pages(vma));
Linus Torvalds's avatar
Linus Torvalds committed
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
	up_write(&task->mm->mmap_sem);

	/*
	 * keep track of user level virtual address
	 */
	ctx->ctx_smpl_vaddr = (void *)vma->vm_start;
	*(unsigned long *)user_vaddr = vma->vm_start;

	return 0;

error:
	kmem_cache_free(vm_area_cachep, vma);