[PATCH] ia64: perfmon update

Here is the patch. It is rather big because there is some renaming and
cleanups.  This patch bring 2.5 in line with 2.4.20: perfmon-1.3

It adds:
        - idle task exclusion
        - less ctxsw overhead in system wide
        - cleanups most of the inline asm
        - don't use PAL anymore to determine PMU features
        - added temporary hooks for custom overflow handlers (VTUNE/Oprofile)
        - renaming of the perfmon init functions

Thanks.

arch/ia64/kernel/ia64_ksyms.c

@@ -142,4 +142,8 @@ EXPORT_SYMBOL(efi_dir);
 EXPORT_SYMBOL(ia64_mv);
 #endif
 EXPORT_SYMBOL(machvec_noop);
-
+#ifdef CONFIG_PERFMON
+#include <asm/perfmon.h>
+EXPORT_SYMBOL(pfm_install_alternate_syswide_subsystem);
+EXPORT_SYMBOL(pfm_remove_alternate_syswide_subsystem);
+#endif

arch/ia64/kernel/irq_ia64.c

@@ -178,7 +178,7 @@ init_IRQ (void)
 	register_percpu_irq(IA64_IPI_VECTOR, &ipi_irqaction);
 #endif
 #ifdef CONFIG_PERFMON
-	perfmon_init_percpu();
+	pfm_init_percpu();
 #endif
 	platform_irq_init();
 }

arch/ia64/kernel/perfmon_generic.h

+/*
+ * This file contains the architected PMU register description tables
+ * and pmc checker used by perfmon.c.
+ *
+ * Copyright (C) 2002  Hewlett Packard Co
+ *               Stephane Eranian <eranian@hpl.hp.com>
+ */
 #define RDEP(x)	(1UL<<(x))
 
-#if defined(CONFIG_ITANIUM) || defined(CONFIG_MCKINLEY)
-#error "This file should only be used when CONFIG_ITANIUM and CONFIG_MCKINLEY are not defined"
+#if defined(CONFIG_ITANIUM) || defined (CONFIG_MCKINLEY)
+#error "This file should not be used when CONFIG_ITANIUM or CONFIG_MCKINLEY is defined"
 #endif
 
-static pfm_reg_desc_t pmc_desc[PMU_MAX_PMCS]={
+static pfm_reg_desc_t pmc_gen_desc[PMU_MAX_PMCS]={
 /* pmc0  */ { PFM_REG_CONTROL , 0, 0x1UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
 /* pmc1  */ { PFM_REG_CONTROL , 0, 0x0UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
 /* pmc2  */ { PFM_REG_CONTROL , 0, 0x0UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
@@ -13,10 +20,10 @@ static pfm_reg_desc_t pmc_desc[PMU_MAX_PMCS]={
 /* pmc5  */ { PFM_REG_COUNTING, 0, 0x0UL, -1UL, NULL, NULL, {RDEP(5),0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
 /* pmc6  */ { PFM_REG_COUNTING, 0, 0x0UL, -1UL, NULL, NULL, {RDEP(6),0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
 /* pmc7  */ { PFM_REG_COUNTING, 0, 0x0UL, -1UL, NULL, NULL, {RDEP(7),0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
-	     { PFM_REG_END     , 0, 0x0UL, -1UL, NULL, NULL, {0,}, {0,}}, /* end marker */
+	    { PFM_REG_END     , 0, 0x0UL, -1UL, NULL, NULL, {0,}, {0,}}, /* end marker */
 };
 
-static pfm_reg_desc_t pmd_desc[PMU_MAX_PMDS]={
+static pfm_reg_desc_t pmd_gen_desc[PMU_MAX_PMDS]={
 /* pmd0  */ { PFM_REG_NOTIMPL , 0, 0x0UL, -1UL, NULL, NULL, {0,}, {0,}},
 /* pmd1  */ { PFM_REG_NOTIMPL , 0, 0x0UL, -1UL, NULL, NULL, {0,}, {0,}},
 /* pmd2  */ { PFM_REG_NOTIMPL , 0, 0x0UL, -1UL, NULL, NULL, {0,}, {0,}},
@@ -25,5 +32,17 @@ static pfm_reg_desc_t pmd_desc[PMU_MAX_PMDS]={
 /* pmd5  */ { PFM_REG_COUNTING, 0, 0x0UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {RDEP(5),0UL, 0UL, 0UL}},
 /* pmd6  */ { PFM_REG_COUNTING, 0, 0x0UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {RDEP(6),0UL, 0UL, 0UL}},
 /* pmd7  */ { PFM_REG_COUNTING, 0, 0x0UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {RDEP(7),0UL, 0UL, 0UL}},
-	     { PFM_REG_END     , 0, 0x0UL, -1UL, NULL, NULL, {0,}, {0,}}, /* end marker */
+	    { PFM_REG_END     , 0, 0x0UL, -1UL, NULL, NULL, {0,}, {0,}}, /* end marker */
+};
+
+/*
+ * impl_pmcs, impl_pmds are computed at runtime to minimize errors!
+ */
+static pmu_config_t pmu_conf={
+	disabled:	1,
+	ovfl_val:	(1UL << 32) - 1,
+	num_ibrs:	8,
+	num_dbrs:	8,
+	pmd_desc:	pfm_gen_pmd_desc,
+	pmc_desc:	pfm_gen_pmc_desc
 };

arch/ia64/kernel/perfmon_itanium.h

@@ -15,7 +15,7 @@
 static int pfm_ita_pmc_check(struct task_struct *task, unsigned int cnum, unsigned long *val, struct pt_regs *regs);
 static int pfm_write_ibr_dbr(int mode, struct task_struct *task, void *arg, int count, struct pt_regs *regs);
 
-static pfm_reg_desc_t pfm_pmc_desc[PMU_MAX_PMCS]={
+static pfm_reg_desc_t pfm_ita_pmc_desc[PMU_MAX_PMCS]={
 /* pmc0  */ { PFM_REG_CONTROL , 0, 0x1UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
 /* pmc1  */ { PFM_REG_CONTROL , 0, 0x0UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
 /* pmc2  */ { PFM_REG_CONTROL , 0, 0x0UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
@@ -33,7 +33,7 @@ static pfm_reg_desc_t pfm_pmc_desc[PMU_MAX_PMCS]={
 	    { PFM_REG_END     , 0, 0x0UL, -1UL, NULL, NULL, {0,}, {0,}}, /* end marker */
 };
 
-static pfm_reg_desc_t pfm_pmd_desc[PMU_MAX_PMDS]={
+static pfm_reg_desc_t pfm_ita_pmd_desc[PMU_MAX_PMDS]={
 /* pmd0  */ { PFM_REG_BUFFER  , 0, 0UL, -1UL, NULL, NULL, {RDEP(1),0UL, 0UL, 0UL}, {RDEP(10),0UL, 0UL, 0UL}},
 /* pmd1  */ { PFM_REG_BUFFER  , 0, 0UL, -1UL, NULL, NULL, {RDEP(0),0UL, 0UL, 0UL}, {RDEP(10),0UL, 0UL, 0UL}},
 /* pmd2  */ { PFM_REG_BUFFER  , 0, 0UL, -1UL, NULL, NULL, {RDEP(3)|RDEP(17),0UL, 0UL, 0UL}, {RDEP(11),0UL, 0UL, 0UL}},
@@ -55,6 +55,19 @@ static pfm_reg_desc_t pfm_pmd_desc[PMU_MAX_PMDS]={
 	    { PFM_REG_END     , 0, 0UL, -1UL, NULL, NULL, {0,}, {0,}}, /* end marker */
 };
 
+/*
+ * impl_pmcs, impl_pmds are computed at runtime to minimize errors!
+ */
+static pmu_config_t pmu_conf={
+	disabled:	1,
+	ovfl_val:	(1UL << 32) - 1,
+	num_ibrs:	8,
+	num_dbrs:	8,
+	pmd_desc:	pfm_ita_pmd_desc,
+	pmc_desc:	pfm_ita_pmc_desc
+};
+
+
 static int
 pfm_ita_pmc_check(struct task_struct *task, unsigned int cnum, unsigned long *val, struct pt_regs *regs)
 {

arch/ia64/kernel/perfmon_mckinley.h

@@ -16,7 +16,7 @@ static int pfm_mck_reserved(struct task_struct *task, unsigned int cnum, unsigne
 static int pfm_mck_pmc_check(struct task_struct *task, unsigned int cnum, unsigned long *val, struct pt_regs *regs);
 static int pfm_write_ibr_dbr(int mode, struct task_struct *task, void *arg, int count, struct pt_regs *regs);
 
-static pfm_reg_desc_t pfm_pmc_desc[PMU_MAX_PMCS]={
+static pfm_reg_desc_t pfm_mck_pmc_desc[PMU_MAX_PMCS]={
 /* pmc0  */ { PFM_REG_CONTROL , 0, 0x1UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
 /* pmc1  */ { PFM_REG_CONTROL , 0, 0x0UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
 /* pmc2  */ { PFM_REG_CONTROL , 0, 0x0UL, -1UL, NULL, NULL, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
@@ -36,7 +36,7 @@ static pfm_reg_desc_t pfm_pmc_desc[PMU_MAX_PMCS]={
 	    { PFM_REG_END     , 0, 0x0UL, -1UL, NULL, NULL, {0,}, {0,}}, /* end marker */
 };
 
-static pfm_reg_desc_t pfm_pmd_desc[PMU_MAX_PMDS]={
+static pfm_reg_desc_t pfm_mck_pmd_desc[PMU_MAX_PMDS]={
 /* pmd0  */ { PFM_REG_BUFFER  , 0, 0x0UL, -1UL, NULL, NULL, {RDEP(1),0UL, 0UL, 0UL}, {RDEP(10),0UL, 0UL, 0UL}},
 /* pmd1  */ { PFM_REG_BUFFER  , 0, 0x0UL, -1UL, NULL, NULL, {RDEP(0),0UL, 0UL, 0UL}, {RDEP(10),0UL, 0UL, 0UL}},
 /* pmd2  */ { PFM_REG_BUFFER  , 0, 0x0UL, -1UL, NULL, NULL, {RDEP(3)|RDEP(17),0UL, 0UL, 0UL}, {RDEP(11),0UL, 0UL, 0UL}},
@@ -58,6 +58,19 @@ static pfm_reg_desc_t pfm_pmd_desc[PMU_MAX_PMDS]={
 	    { PFM_REG_END     , 0, 0x0UL, -1UL, NULL, NULL, {0,}, {0,}}, /* end marker */
 };
 
+/*
+ * impl_pmcs, impl_pmds are computed at runtime to minimize errors!
+ */
+static pmu_config_t pmu_conf={
+	disabled:	1,
+	ovfl_val:	(1UL << 47) - 1,
+	num_ibrs:	8,
+	num_dbrs:	8,
+	pmd_desc:	pfm_mck_pmd_desc,
+	pmc_desc:	pfm_mck_pmc_desc
+};
+
+
 /*
  * PMC reserved fields must have their power-up values preserved
  */

arch/ia64/kernel/process.c

@@ -205,6 +205,10 @@ cpu_idle (void *unused)
 void
 ia64_save_extra (struct task_struct *task)
 {
+#ifdef CONFIG_PERFMON
+	unsigned long info;
+#endif
+
 	if ((task->thread.flags & IA64_THREAD_DBG_VALID) != 0)
 		ia64_save_debug_regs(&task->thread.dbr[0]);
 
@@ -212,8 +216,9 @@ ia64_save_extra (struct task_struct *task)
 	if ((task->thread.flags & IA64_THREAD_PM_VALID) != 0)
 		pfm_save_regs(task);
 
-	if (__get_cpu_var(pfm_syst_wide))
-		pfm_syst_wide_update_task(task, 0);
+	info = __get_cpu_var(pfm_syst_info);
+	if (info & PFM_CPUINFO_SYST_WIDE)
+		pfm_syst_wide_update_task(task, info, 0);
 #endif
 
 #ifdef CONFIG_IA32_SUPPORT
@@ -225,6 +230,10 @@ ia64_save_extra (struct task_struct *task)
 void
 ia64_load_extra (struct task_struct *task)
 {
+#ifdef CONFIG_PERFMON
+	unsigned long info;
+#endif
+
 	if ((task->thread.flags & IA64_THREAD_DBG_VALID) != 0)
 		ia64_load_debug_regs(&task->thread.dbr[0]);
 
@@ -232,8 +241,9 @@ ia64_load_extra (struct task_struct *task)
 	if ((task->thread.flags & IA64_THREAD_PM_VALID) != 0)
 		pfm_load_regs(task);
 
-	if (__get_cpu_var(pfm_syst_wide)) 
-		pfm_syst_wide_update_task(task, 1);
+	info = __get_cpu_var(pfm_syst_info);
+	if (info & PFM_CPUINFO_SYST_WIDE) 
+		pfm_syst_wide_update_task(task, info, 1);
 #endif
 
 #ifdef CONFIG_IA32_SUPPORT

arch/ia64/kernel/smpboot.c

@@ -265,7 +265,7 @@ smp_callin (void)
 	extern void ia64_init_itm(void);
 
 #ifdef CONFIG_PERFMON
-	extern void perfmon_init_percpu(void);
+	extern void pfm_init_percpu(void);
 #endif
 
 	cpuid = smp_processor_id();
@@ -300,7 +300,7 @@ smp_callin (void)
 #endif
 
 #ifdef CONFIG_PERFMON
-	perfmon_init_percpu();
+	pfm_init_percpu();
 #endif
 
 	local_irq_enable();

include/asm-ia64/perfmon.h

@@ -40,6 +40,7 @@
 #define PFM_FL_INHERIT_ALL	 0x02	/* always clone pfm_context across fork() */
 #define PFM_FL_NOTIFY_BLOCK    	 0x04	/* block task on user level notifications */
 #define PFM_FL_SYSTEM_WIDE	 0x08	/* create a system wide context */
+#define PFM_FL_EXCL_IDLE         0x20   /* exclude idle task from system wide session */
 
 /*
  * PMC flags
@@ -86,11 +87,12 @@ typedef struct {
 	unsigned long	reg_long_reset;	/* reset after sampling buffer overflow (large) */
 	unsigned long	reg_short_reset;/* reset after counter overflow (small) */
 
-	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_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_value;/* last value used to reset the PMD (PFM_READ_PMDS) */
 
-	unsigned long   reserved[14];	/* for future use */
+	unsigned long   reserved[13];	/* for future use */
 } pfarg_reg_t;
 
 typedef struct {
@@ -123,7 +125,7 @@ typedef struct {
  * Define the version numbers for both perfmon as a whole and the sampling buffer format.
  */
 #define PFM_VERSION_MAJ		1U
-#define PFM_VERSION_MIN		1U
+#define PFM_VERSION_MIN		3U
 #define PFM_VERSION		(((PFM_VERSION_MAJ&0xffff)<<16)|(PFM_VERSION_MIN & 0xffff))
 
 #define PFM_SMPL_VERSION_MAJ	1U
@@ -156,13 +158,17 @@ typedef struct {
 	unsigned long	stamp;			/* timestamp */
 	unsigned long	ip;			/* where did the overflow interrupt happened */
 	unsigned long	regs;			/* bitmask of which registers overflowed */
-	unsigned long   period;			/* unused */
+	unsigned long   reserved;		/* unused */
 } perfmon_smpl_entry_t;
 
 extern int perfmonctl(pid_t pid, int cmd, void *arg, int narg);
 
 #ifdef __KERNEL__
 
+typedef struct {
+	void (*handler)(int irq, void *arg, struct pt_regs *regs);
+} pfm_intr_handler_desc_t;
+
 extern void pfm_save_regs (struct task_struct *);
 extern void pfm_load_regs (struct task_struct *);
 
@@ -174,9 +180,24 @@ extern void pfm_cleanup_owners (struct task_struct *);
 extern int  pfm_use_debug_registers(struct task_struct *);
 extern int  pfm_release_debug_registers(struct task_struct *);
 extern int  pfm_cleanup_smpl_buf(struct task_struct *);
-extern void pfm_syst_wide_update_task(struct task_struct *, int);
+extern void pfm_syst_wide_update_task(struct task_struct *, unsigned long info, int is_ctxswin);
 extern void pfm_ovfl_block_reset(void);
-extern void perfmon_init_percpu(void);
+extern void pfm_init_percpu(void);
+
+/* 
+ * hooks to allow VTune/Prospect to cooperate with perfmon.
+ * (reserved for system wide monitoring modules only)
+ */
+extern int pfm_install_alternate_syswide_subsystem(pfm_intr_handler_desc_t *h);
+extern int pfm_remove_alternate_syswide_subsystem(pfm_intr_handler_desc_t *h);
+
+/*
+ * describe the content of the local_cpu_date->pfm_syst_info field
+ */
+#define PFM_CPUINFO_SYST_WIDE	0x1	/* if set a system wide session exist */
+#define PFM_CPUINFO_DCR_PP	0x2	/* if set the system wide session has started */
+#define PFM_CPUINFO_EXCL_IDLE	0x4	/* the system wide session excludes the idle task */
+
 
 #endif /* __KERNEL__ */
 

include/asm-ia64/system.h

@@ -205,8 +205,8 @@ extern void ia64_save_extra (struct task_struct *task);
 extern void ia64_load_extra (struct task_struct *task);
 
 #ifdef CONFIG_PERFMON
-  DECLARE_PER_CPU(int, pfm_syst_wide);
-# define PERFMON_IS_SYSWIDE() (get_cpu_var(pfm_syst_wide) != 0)
+  DECLARE_PER_CPU(unsigned long, pfm_syst_info);
+# define PERFMON_IS_SYSWIDE() (get_cpu_var(pfm_syst_info) & 0x1)
 #else
 # define PERFMON_IS_SYSWIDE() (0)
 #endif