[IA64] alternate perfmon handler

Patch from Charles Spirakis

Some linux customers want to optimize their applications on the latest
hardware but are not yet willing to upgrade to the latest kernel. This
patch provides a way to plug in an alternate, basic, and GPL'ed PMU
subsystem to help with their monitoring needs or for specialty work. It
can also be used in case of serious unexpected bugs in perfmon. Mutual
exclusion between the two subsystems is guaranteed, hence no conflict
can arise from both subsystem being present.

Acked-by: Stephane Eranian <eranian@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

2 files changed, 168 insertions, 15 deletions

diff --git a/arch/ia64/kernel/perfmon.c b/arch/ia64/kernel/perfmon.c
index 71c101601e3e..ab9682f8b044 100644
--- a/arch/ia64/kernel/perfmon.c
+++ b/arch/ia64/kernel/perfmon.c
@@ -11,7 +11,7 @@
  * Version Perfmon-2.x is a rewrite of perfmon-1.x
  * by Stephane Eranian, Hewlett Packard Co.
  *
- * Copyright (C) 1999-2003, 2005  Hewlett Packard Co
+ * Copyright (C) 1999-2005  Hewlett Packard Co
  *               Stephane Eranian <eranian@hpl.hp.com>
  *               David Mosberger-Tang <davidm@hpl.hp.com>
  *
@@ -497,6 +497,9 @@ typedef struct {
 static pfm_stats_t		pfm_stats[NR_CPUS];
 static pfm_session_t		pfm_sessions;	/* global sessions information */
 
+static spinlock_t pfm_alt_install_check;
+static pfm_intr_handler_desc_t  *pfm_alt_intr_handler;
+
 static struct proc_dir_entry 	*perfmon_dir;
 static pfm_uuid_t		pfm_null_uuid = {0,};
 
@@ -606,6 +609,7 @@ 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);
+EXPORT_SYMBOL_GPL(per_cpu__pfm_syst_info);
 
 
 /* forward declaration */
@@ -1325,7 +1329,7 @@ pfm_reserve_session(struct task_struct *task, int is_syswide, unsigned int cpu)
 error_conflict:
 	DPRINT(("system wide not possible, conflicting session [%d] on CPU%d\n",
   		pfm_sessions.pfs_sys_session[cpu]->pid,
-		smp_processor_id()));
+		cpu));
 abort:
 	UNLOCK_PFS(flags);
 
@@ -5555,26 +5559,32 @@ pfm_interrupt_handler(int irq, void *arg, struct pt_regs *regs)
 	int ret;
 
 	this_cpu = get_cpu();
-	min      = pfm_stats[this_cpu].pfm_ovfl_intr_cycles_min;
-	max      = pfm_stats[this_cpu].pfm_ovfl_intr_cycles_max;
+	if (likely(!pfm_alt_intr_handler)) {
+		min = pfm_stats[this_cpu].pfm_ovfl_intr_cycles_min;
+		max = pfm_stats[this_cpu].pfm_ovfl_intr_cycles_max;
 
-	start_cycles = ia64_get_itc();
+		start_cycles = ia64_get_itc();
 
-	ret = pfm_do_interrupt_handler(irq, arg, regs);
+		ret = pfm_do_interrupt_handler(irq, arg, regs);
 
-	total_cycles = ia64_get_itc();
+		total_cycles = ia64_get_itc();
 
-	/*
-	 * don't measure spurious interrupts
-	 */
-	if (likely(ret == 0)) {
-		total_cycles -= start_cycles;
+		/*
+		 * don't measure spurious interrupts
+		 */
+		if (likely(ret == 0)) {
+			total_cycles -= start_cycles;
 
-		if (total_cycles < min) pfm_stats[this_cpu].pfm_ovfl_intr_cycles_min = total_cycles;
-		if (total_cycles > max) pfm_stats[this_cpu].pfm_ovfl_intr_cycles_max = total_cycles;
+			if (total_cycles < min) pfm_stats[this_cpu].pfm_ovfl_intr_cycles_min = total_cycles;
+			if (total_cycles > max) pfm_stats[this_cpu].pfm_ovfl_intr_cycles_max = total_cycles;
 
-		pfm_stats[this_cpu].pfm_ovfl_intr_cycles += total_cycles;
+			pfm_stats[this_cpu].pfm_ovfl_intr_cycles += total_cycles;
+		}
+	}
+	else {
+		(*pfm_alt_intr_handler->handler)(irq, arg, regs);
 	}
+
 	put_cpu_no_resched();
 	return IRQ_HANDLED;
 }
@@ -6425,6 +6435,141 @@ static struct irqaction perfmon_irqaction = {
 	.name    = "perfmon"
 };
 
+static void
+pfm_alt_save_pmu_state(void *data)
+{
+	struct pt_regs *regs;
+
+	regs = ia64_task_regs(current);
+
+	DPRINT(("called\n"));
+
+	/*
+	 * should not be necessary but
+	 * let's take not risk
+	 */
+	pfm_clear_psr_up();
+	pfm_clear_psr_pp();
+	ia64_psr(regs)->pp = 0;
+
+	/*
+	 * This call is required
+	 * May cause a spurious interrupt on some processors
+	 */
+	pfm_freeze_pmu();
+
+	ia64_srlz_d();
+}
+
+void
+pfm_alt_restore_pmu_state(void *data)
+{
+	struct pt_regs *regs;
+
+	regs = ia64_task_regs(current);
+
+	DPRINT(("called\n"));
+
+	/*
+	 * put PMU back in state expected
+	 * by perfmon
+	 */
+	pfm_clear_psr_up();
+	pfm_clear_psr_pp();
+	ia64_psr(regs)->pp = 0;
+
+	/*
+	 * perfmon runs with PMU unfrozen at all times
+	 */
+	pfm_unfreeze_pmu();
+
+	ia64_srlz_d();
+}
+
+int
+pfm_install_alt_pmu_interrupt(pfm_intr_handler_desc_t *hdl)
+{
+	int ret, i;
+	int reserve_cpu;
+
+	/* some sanity checks */
+	if (hdl == NULL || hdl->handler == NULL) return -EINVAL;
+
+	/* do the easy test first */
+	if (pfm_alt_intr_handler) return -EBUSY;
+
+	/* one at a time in the install or remove, just fail the others */
+	if (!spin_trylock(&pfm_alt_install_check)) {
+		return -EBUSY;
+	}
+
+	/* reserve our session */
+	for_each_online_cpu(reserve_cpu) {
+		ret = pfm_reserve_session(NULL, 1, reserve_cpu);
+		if (ret) goto cleanup_reserve;
+	}
+
+	/* save the current system wide pmu states */
+	ret = on_each_cpu(pfm_alt_save_pmu_state, NULL, 0, 1);
+	if (ret) {
+		DPRINT(("on_each_cpu() failed: %d\n", ret));
+		goto cleanup_reserve;
+	}
+
+	/* officially change to the alternate interrupt handler */
+	pfm_alt_intr_handler = hdl;
+
+	spin_unlock(&pfm_alt_install_check);
+
+	return 0;
+
+cleanup_reserve:
+	for_each_online_cpu(i) {
+		/* don't unreserve more than we reserved */
+		if (i >= reserve_cpu) break;
+
+		pfm_unreserve_session(NULL, 1, i);
+	}
+
+	spin_unlock(&pfm_alt_install_check);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(pfm_install_alt_pmu_interrupt);
+
+int
+pfm_remove_alt_pmu_interrupt(pfm_intr_handler_desc_t *hdl)
+{
+	int i;
+	int ret;
+
+	if (hdl == NULL) return -EINVAL;
+
+	/* cannot remove someone else's handler! */
+	if (pfm_alt_intr_handler != hdl) return -EINVAL;
+
+	/* one at a time in the install or remove, just fail the others */
+	if (!spin_trylock(&pfm_alt_install_check)) {
+		return -EBUSY;
+	}
+
+	pfm_alt_intr_handler = NULL;
+
+	ret = on_each_cpu(pfm_alt_restore_pmu_state, NULL, 0, 1);
+	if (ret) {
+		DPRINT(("on_each_cpu() failed: %d\n", ret));
+	}
+
+	for_each_online_cpu(i) {
+		pfm_unreserve_session(NULL, 1, i);
+	}
+
+	spin_unlock(&pfm_alt_install_check);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(pfm_remove_alt_pmu_interrupt);
+
 /*
  * perfmon initialization routine, called from the initcall() table
  */
diff --git a/include/asm-ia64/perfmon.h b/include/asm-ia64/perfmon.h
index ed5416c5b1ac..7f3333dd00e4 100644
--- a/include/asm-ia64/perfmon.h
+++ b/include/asm-ia64/perfmon.h
@@ -177,6 +177,10 @@ typedef union {
 
 extern long perfmonctl(int fd, int cmd, void *arg, int narg);
 
+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 *);
 
@@ -187,6 +191,10 @@ extern void pfm_syst_wide_update_task(struct task_struct *, unsigned long info,
 extern void pfm_inherit(struct task_struct *task, struct pt_regs *regs);
 extern void pfm_init_percpu(void);
 extern void pfm_handle_work(void);
+extern int  pfm_install_alt_pmu_interrupt(pfm_intr_handler_desc_t *h);
+extern int  pfm_remove_alt_pmu_interrupt(pfm_intr_handler_desc_t *h);
+
+
 
 /*
  * Reset PMD register flags