x86, bts: provide in-kernel branch-trace interface

Impact: cleanup

Move the BTS bits from ptrace.c into ds.c.
Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

arch/x86/include/asm/ds.h

@@ -6,13 +6,13 @@
  * precise-event based sampling (PEBS).
  *
  * It manages:
- * - per-thread and per-cpu allocation of BTS and PEBS
+ * - DS and BTS hardware configuration
  * - buffer overflow handling (to be done)
  * - buffer access
  *
- * It assumes:
- * - get_task_struct on all traced tasks
- * - current is allowed to trace tasks
+ * It does not do:
+ * - security checking (is the caller allowed to trace the task)
+ * - buffer allocation (memory accounting)
  *
  *
  * Copyright (C) 2007-2008 Intel Corporation.
@@ -31,6 +31,7 @@
 #ifdef CONFIG_X86_DS
 
 struct task_struct;
+struct ds_context;
 struct ds_tracer;
 struct bts_tracer;
 struct pebs_tracer;
@@ -38,6 +39,38 @@ struct pebs_tracer;
 typedef void (*bts_ovfl_callback_t)(struct bts_tracer *);
 typedef void (*pebs_ovfl_callback_t)(struct pebs_tracer *);
 
+
+/*
+ * A list of features plus corresponding macros to talk about them in
+ * the ds_request function's flags parameter.
+ *
+ * We use the enum to index an array of corresponding control bits;
+ * we use the macro to index a flags bit-vector.
+ */
+enum ds_feature {
+	dsf_bts = 0,
+	dsf_bts_kernel,
+#define BTS_KERNEL (1 << dsf_bts_kernel)
+	/* trace kernel-mode branches */
+
+	dsf_bts_user,
+#define BTS_USER (1 << dsf_bts_user)
+	/* trace user-mode branches */
+
+	dsf_bts_overflow,
+	dsf_bts_max,
+	dsf_pebs = dsf_bts_max,
+
+	dsf_pebs_max,
+	dsf_ctl_max = dsf_pebs_max,
+	dsf_bts_timestamps = dsf_ctl_max,
+#define BTS_TIMESTAMPS (1 << dsf_bts_timestamps)
+	/* add timestamps into BTS trace */
+
+#define BTS_USER_FLAGS (BTS_KERNEL | BTS_USER | BTS_TIMESTAMPS)
+};
+
+
 /*
  * Request BTS or PEBS
  *
@@ -58,92 +91,135 @@ typedef void (*pebs_ovfl_callback_t)(struct pebs_tracer *);
  *       NULL if cyclic buffer requested
  * th: the interrupt threshold in records from the end of the buffer;
  *     -1 if no interrupt threshold is requested.
+ * flags: a bit-mask of the above flags
  */
 extern struct bts_tracer *ds_request_bts(struct task_struct *task,
 					 void *base, size_t size,
-					 bts_ovfl_callback_t ovfl, size_t th);
+					 bts_ovfl_callback_t ovfl,
+					 size_t th, unsigned int flags);
 extern struct pebs_tracer *ds_request_pebs(struct task_struct *task,
 					   void *base, size_t size,
 					   pebs_ovfl_callback_t ovfl,
-					   size_t th);
+					   size_t th, unsigned int flags);
 
 /*
  * Release BTS or PEBS resources
- *
- * Returns 0 on success; -Eerrno otherwise
+ * Suspend and resume BTS or PEBS tracing
  *
  * tracer: the tracer handle returned from ds_request_~()
  */
-extern int ds_release_bts(struct bts_tracer *tracer);
-extern int ds_release_pebs(struct pebs_tracer *tracer);
+extern void ds_release_bts(struct bts_tracer *tracer);
+extern void ds_suspend_bts(struct bts_tracer *tracer);
+extern void ds_resume_bts(struct bts_tracer *tracer);
+extern void ds_release_pebs(struct pebs_tracer *tracer);
+extern void ds_suspend_pebs(struct pebs_tracer *tracer);
+extern void ds_resume_pebs(struct pebs_tracer *tracer);
+
 
 /*
- * Get the (array) index of the write pointer.
- * (assuming an array of BTS/PEBS records)
- *
- * Returns 0 on success; -Eerrno on error
+ * The raw DS buffer state as it is used for BTS and PEBS recording.
  *
- * tracer: the tracer handle returned from ds_request_~()
- * pos (out): will hold the result
+ * This is the low-level, arch-dependent interface for working
+ * directly on the raw trace data.
  */
-extern int ds_get_bts_index(struct bts_tracer *tracer, size_t *pos);
-extern int ds_get_pebs_index(struct pebs_tracer *tracer, size_t *pos);
+struct ds_trace {
+	/* the number of bts/pebs records */
+	size_t n;
+	/* the size of a bts/pebs record in bytes */
+	size_t size;
+	/* pointers into the raw buffer:
+	   - to the first entry */
+	void *begin;
+	/* - one beyond the last entry */
+	void *end;
+	/* - one beyond the newest entry */
+	void *top;
+	/* - the interrupt threshold */
+	void *ith;
+	/* flags given on ds_request() */
+	unsigned int flags;
+};
 
 /*
- * Get the (array) index one record beyond the end of the array.
- * (assuming an array of BTS/PEBS records)
- *
- * Returns 0 on success; -Eerrno on error
- *
- * tracer: the tracer handle returned from ds_request_~()
- * pos (out): will hold the result
+ * An arch-independent view on branch trace data.
  */
-extern int ds_get_bts_end(struct bts_tracer *tracer, size_t *pos);
-extern int ds_get_pebs_end(struct pebs_tracer *tracer, size_t *pos);
+enum bts_qualifier {
+	bts_invalid,
+#define BTS_INVALID bts_invalid
+
+	bts_branch,
+#define BTS_BRANCH bts_branch
+
+	bts_task_arrives,
+#define BTS_TASK_ARRIVES bts_task_arrives
+
+	bts_task_departs,
+#define BTS_TASK_DEPARTS bts_task_departs
+
+	bts_qual_bit_size = 4,
+	bts_qual_max = (1 << bts_qual_bit_size),
+};
+
+struct bts_struct {
+	__u64 qualifier;
+	union {
+		/* BTS_BRANCH */
+		struct {
+			__u64 from;
+			__u64 to;
+		} lbr;
+		/* BTS_TASK_ARRIVES or BTS_TASK_DEPARTS */
+		struct {
+			__u64 jiffies;
+			pid_t pid;
+		} timestamp;
+	} variant;
+};
+
 
 /*
- * Provide a pointer to the BTS/PEBS record at parameter index.
- * (assuming an array of BTS/PEBS records)
- *
- * The pointer points directly into the buffer. The user is
- * responsible for copying the record.
- *
- * Returns the size of a single record on success; -Eerrno on error
+ * The BTS state.
  *
- * tracer: the tracer handle returned from ds_request_~()
- * index: the index of the requested record
- * record (out): pointer to the requested record
+ * This gives access to the raw DS state and adds functions to provide
+ * an arch-independent view of the BTS data.
  */
-extern int ds_access_bts(struct bts_tracer *tracer,
-			 size_t index, const void **record);
-extern int ds_access_pebs(struct pebs_tracer *tracer,
-			  size_t index, const void **record);
+struct bts_trace {
+	struct ds_trace ds;
+
+	int (*read)(struct bts_tracer *tracer, const void *at,
+		    struct bts_struct *out);
+	int (*write)(struct bts_tracer *tracer, const struct bts_struct *in);
+};
+
 
 /*
- * Write one or more BTS/PEBS records at the write pointer index and
- * advance the write pointer.
+ * The PEBS state.
  *
- * If size is not a multiple of the record size, trailing bytes are
- * zeroed out.
- *
- * May result in one or more overflow notifications.
- *
- * If called during overflow handling, that is, with index >=
- * interrupt threshold, the write will wrap around.
+ * This gives access to the raw DS state and the PEBS-specific counter
+ * reset value.
+ */
+struct pebs_trace {
+	struct ds_trace ds;
+
+	/* the PEBS reset value */
+	unsigned long long reset_value;
+};
+
+
+/*
+ * Read the BTS or PEBS trace.
  *
- * An overflow notification is given if and when the interrupt
- * threshold is reached during or after the write.
+ * Returns a view on the trace collected for the parameter tracer.
  *
- * Returns the number of bytes written or -Eerrno.
+ * The view remains valid as long as the traced task is not running or
+ * the tracer is suspended.
+ * Writes into the trace buffer are not reflected.
  *
  * tracer: the tracer handle returned from ds_request_~()
- * buffer: the buffer to write
- * size: the size of the buffer
  */
-extern int ds_write_bts(struct bts_tracer *tracer,
-			const void *buffer, size_t size);
-extern int ds_write_pebs(struct pebs_tracer *tracer,
-			 const void *buffer, size_t size);
+extern const struct bts_trace *ds_read_bts(struct bts_tracer *tracer);
+extern const struct pebs_trace *ds_read_pebs(struct pebs_tracer *tracer);
+
 
 /*
  * Reset the write pointer of the BTS/PEBS buffer.
@@ -155,27 +231,6 @@ extern int ds_write_pebs(struct pebs_tracer *tracer,
 extern int ds_reset_bts(struct bts_tracer *tracer);
 extern int ds_reset_pebs(struct pebs_tracer *tracer);
 
-/*
- * Clear the BTS/PEBS buffer and reset the write pointer.
- * The entire buffer will be zeroed out.
- *
- * Returns 0 on success; -Eerrno on error
- *
- * tracer: the tracer handle returned from ds_request_~()
- */
-extern int ds_clear_bts(struct bts_tracer *tracer);
-extern int ds_clear_pebs(struct pebs_tracer *tracer);
-
-/*
- * Provide the PEBS counter reset value.
- *
- * Returns 0 on success; -Eerrno on error
- *
- * tracer: the tracer handle returned from ds_request_pebs()
- * value (out): the counter reset value
- */
-extern int ds_get_pebs_reset(struct pebs_tracer *tracer, u64 *value);
-
 /*
  * Set the PEBS counter reset value.
  *
@@ -192,35 +247,17 @@ extern int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value);
 struct cpuinfo_x86;
 extern void __cpuinit ds_init_intel(struct cpuinfo_x86 *);
 
-
-
 /*
- * The DS context - part of struct thread_struct.
+ * Context switch work
  */
-#define MAX_SIZEOF_DS (12 * 8)
-
-struct ds_context {
-	/* pointer to the DS configuration; goes into MSR_IA32_DS_AREA */
-	unsigned char ds[MAX_SIZEOF_DS];
-	/* the owner of the BTS and PEBS configuration, respectively */
-	struct ds_tracer  *owner[2];
-	/* use count */
-	unsigned long count;
-	/* a pointer to the context location inside the thread_struct
-	 * or the per_cpu context array */
-	struct ds_context **this;
-	/* a pointer to the task owning this context, or NULL, if the
-	 * context is owned by a cpu */
-	struct task_struct *task;
-};
-
-/* called by exit_thread() to free leftover contexts */
-extern void ds_free(struct ds_context *context);
+extern void ds_switch_to(struct task_struct *prev, struct task_struct *next);
 
 #else /* CONFIG_X86_DS */
 
 struct cpuinfo_x86;
 static inline void __cpuinit ds_init_intel(struct cpuinfo_x86 *ignored) {}
+static inline void ds_switch_to(struct task_struct *prev,
+				struct task_struct *next) {}
 
 #endif /* CONFIG_X86_DS */
 #endif /* _ASM_X86_DS_H */

arch/x86/include/asm/processor.h

@@ -752,6 +752,19 @@ extern void switch_to_new_gdt(void);
 extern void cpu_init(void);
 extern void init_gdt(int cpu);
 
+static inline unsigned long get_debugctlmsr(void)
+{
+    unsigned long debugctlmsr = 0;
+
+#ifndef CONFIG_X86_DEBUGCTLMSR
+	if (boot_cpu_data.x86 < 6)
+		return 0;
+#endif
+	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
+
+    return debugctlmsr;
+}
+
 static inline void update_debugctlmsr(unsigned long debugctlmsr)
 {
 #ifndef CONFIG_X86_DEBUGCTLMSR

arch/x86/include/asm/ptrace.h

@@ -6,7 +6,6 @@
 #include <asm/processor-flags.h>
 
 #ifdef __KERNEL__
-#include <asm/ds.h>		/* the DS BTS struct is used for ptrace too */
 #include <asm/segment.h>
 #endif
 
@@ -128,34 +127,6 @@ struct pt_regs {
 #endif /* !__i386__ */
 
 
-#ifdef CONFIG_X86_PTRACE_BTS
-/* a branch trace record entry
- *
- * In order to unify the interface between various processor versions,
- * we use the below data structure for all processors.
- */
-enum bts_qualifier {
-	BTS_INVALID = 0,
-	BTS_BRANCH,
-	BTS_TASK_ARRIVES,
-	BTS_TASK_DEPARTS
-};
-
-struct bts_struct {
-	__u64 qualifier;
-	union {
-		/* BTS_BRANCH */
-		struct {
-			__u64 from_ip;
-			__u64 to_ip;
-		} lbr;
-		/* BTS_TASK_ARRIVES or
-		   BTS_TASK_DEPARTS */
-		__u64 jiffies;
-	} variant;
-};
-#endif /* CONFIG_X86_PTRACE_BTS */
-
 #ifdef __KERNEL__
 
 #include <linux/init.h>
@@ -163,13 +134,6 @@ struct bts_struct {
 struct cpuinfo_x86;
 struct task_struct;
 
-#ifdef CONFIG_X86_PTRACE_BTS
-extern void __cpuinit ptrace_bts_init_intel(struct cpuinfo_x86 *);
-extern void ptrace_bts_take_timestamp(struct task_struct *, enum bts_qualifier);
-#else
-#define ptrace_bts_init_intel(config) do {} while (0)
-#endif /* CONFIG_X86_PTRACE_BTS */
-
 extern unsigned long profile_pc(struct pt_regs *regs);
 
 extern unsigned long

arch/x86/include/asm/thread_info.h

@@ -93,7 +93,6 @@ struct thread_info {
 #define TIF_FORCED_TF		24	/* true if TF in eflags artificially */
 #define TIF_DEBUGCTLMSR		25	/* uses thread_struct.debugctlmsr */
 #define TIF_DS_AREA_MSR		26      /* uses thread_struct.ds_area_msr */
-#define TIF_BTS_TRACE_TS	27      /* record scheduling event timestamps */
 
 #define _TIF_SYSCALL_TRACE	(1 << TIF_SYSCALL_TRACE)
 #define _TIF_NOTIFY_RESUME	(1 << TIF_NOTIFY_RESUME)
@@ -115,7 +114,6 @@ struct thread_info {
 #define _TIF_FORCED_TF		(1 << TIF_FORCED_TF)
 #define _TIF_DEBUGCTLMSR	(1 << TIF_DEBUGCTLMSR)
 #define _TIF_DS_AREA_MSR	(1 << TIF_DS_AREA_MSR)
-#define _TIF_BTS_TRACE_TS	(1 << TIF_BTS_TRACE_TS)
 
 /* work to do in syscall_trace_enter() */
 #define _TIF_WORK_SYSCALL_ENTRY	\
@@ -141,8 +139,7 @@ struct thread_info {
 
 /* flags to check in __switch_to() */
 #define _TIF_WORK_CTXSW							\
-	(_TIF_IO_BITMAP|_TIF_DEBUGCTLMSR|_TIF_DS_AREA_MSR|_TIF_BTS_TRACE_TS| \
-								_TIF_NOTSC)
+	(_TIF_IO_BITMAP|_TIF_DEBUGCTLMSR|_TIF_DS_AREA_MSR|_TIF_NOTSC)
 
 #define _TIF_WORK_CTXSW_PREV _TIF_WORK_CTXSW
 #define _TIF_WORK_CTXSW_NEXT (_TIF_WORK_CTXSW|_TIF_DEBUG)

arch/x86/kernel/cpu/intel.c

@@ -11,7 +11,6 @@
 #include <asm/pgtable.h>
 #include <asm/msr.h>
 #include <asm/uaccess.h>
-#include <asm/ptrace.h>
 #include <asm/ds.h>
 #include <asm/bugs.h>
 
@@ -309,9 +308,6 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c)
 		set_cpu_cap(c, X86_FEATURE_P3);
 #endif
 
-	if (cpu_has_bts)
-		ptrace_bts_init_intel(c);
-
 	detect_extended_topology(c);
 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
 		/*

arch/x86/kernel/process_32.c

@@ -252,11 +252,14 @@ void exit_thread(void)
 		put_cpu();
 	}
 #ifdef CONFIG_X86_DS
-	/* Free any DS contexts that have not been properly released. */
-	if (unlikely(current->thread.ds_ctx)) {
-		/* we clear debugctl to make sure DS is not used. */
-		update_debugctlmsr(0);
-		ds_free(current->thread.ds_ctx);
+	/* Free any BTS tracers that have not been properly released. */
+	if (unlikely(current->bts)) {
+		ds_release_bts(current->bts);
+		current->bts = NULL;
+
+		kfree(current->bts_buffer);
+		current->bts_buffer = NULL;
+		current->bts_size = 0;
 	}
 #endif /* CONFIG_X86_DS */
 }
@@ -420,48 +423,19 @@ int set_tsc_mode(unsigned int val)
 	return 0;
 }
 
-#ifdef CONFIG_X86_DS
-static int update_debugctl(struct thread_struct *prev,
-			struct thread_struct *next, unsigned long debugctl)
-{
-	unsigned long ds_prev = 0;
-	unsigned long ds_next = 0;
-
-	if (prev->ds_ctx)
-		ds_prev = (unsigned long)prev->ds_ctx->ds;
-	if (next->ds_ctx)
-		ds_next = (unsigned long)next->ds_ctx->ds;
-
-	if (ds_next != ds_prev) {
-		/* we clear debugctl to make sure DS
-		 * is not in use when we change it */
-		debugctl = 0;
-		update_debugctlmsr(0);
-		wrmsr(MSR_IA32_DS_AREA, ds_next, 0);
-	}
-	return debugctl;
-}
-#else
-static int update_debugctl(struct thread_struct *prev,
-			struct thread_struct *next, unsigned long debugctl)
-{
-	return debugctl;
-}
-#endif /* CONFIG_X86_DS */
-
 static noinline void
 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 		 struct tss_struct *tss)
 {
 	struct thread_struct *prev, *next;
-	unsigned long debugctl;
 
 	prev = &prev_p->thread;
 	next = &next_p->thread;
 
-	debugctl = update_debugctl(prev, next, prev->debugctlmsr);
-
-	if (next->debugctlmsr != debugctl)
+	if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
+	    test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
+		ds_switch_to(prev_p, next_p);
+	else if (next->debugctlmsr != prev->debugctlmsr)
 		update_debugctlmsr(next->debugctlmsr);
 
 	if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
@@ -483,15 +457,6 @@ __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 			hard_enable_TSC();
 	}
 
-#ifdef CONFIG_X86_PTRACE_BTS
-	if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
-		ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
-
-	if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
-		ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
-#endif /* CONFIG_X86_PTRACE_BTS */
-
-
 	if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 		/*
 		 * Disable the bitmap via an invalid offset. We still cache

arch/x86/kernel/process_64.c

@@ -237,11 +237,14 @@ void exit_thread(void)
 		put_cpu();
 	}
 #ifdef CONFIG_X86_DS
-	/* Free any DS contexts that have not been properly released. */
-	if (unlikely(t->ds_ctx)) {
-		/* we clear debugctl to make sure DS is not used. */
-		update_debugctlmsr(0);
-		ds_free(t->ds_ctx);
+	/* Free any BTS tracers that have not been properly released. */
+	if (unlikely(current->bts)) {
+		ds_release_bts(current->bts);
+		current->bts = NULL;
+
+		kfree(current->bts_buffer);
+		current->bts_buffer = NULL;
+		current->bts_size = 0;
 	}
 #endif /* CONFIG_X86_DS */
 }
@@ -471,35 +474,14 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
 				    struct tss_struct *tss)
 {
 	struct thread_struct *prev, *next;
-	unsigned long debugctl;
 
 	prev = &prev_p->thread,
 	next = &next_p->thread;
 
-	debugctl = prev->debugctlmsr;
-
-#ifdef CONFIG_X86_DS
-	{
-		unsigned long ds_prev = 0, ds_next = 0;
-
-		if (prev->ds_ctx)
-			ds_prev = (unsigned long)prev->ds_ctx->ds;
-		if (next->ds_ctx)
-			ds_next = (unsigned long)next->ds_ctx->ds;
-
-		if (ds_next != ds_prev) {
-			/*
-			 * We clear debugctl to make sure DS
-			 * is not in use when we change it:
-			 */
-			debugctl = 0;
-			update_debugctlmsr(0);
-			wrmsrl(MSR_IA32_DS_AREA, ds_next);
-		}
-	}
-#endif /* CONFIG_X86_DS */
-
-	if (next->debugctlmsr != debugctl)
+	if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
+	    test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
+		ds_switch_to(prev_p, next_p);
+	else if (next->debugctlmsr != prev->debugctlmsr)
 		update_debugctlmsr(next->debugctlmsr);
 
 	if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
@@ -534,14 +516,6 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
 		 */
 		memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 	}
-
-#ifdef CONFIG_X86_PTRACE_BTS
-	if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
-		ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
-
-	if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
-		ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
-#endif /* CONFIG_X86_PTRACE_BTS */
 }
 
 /*

include/linux/sched.h

@@ -1176,6 +1176,7 @@ struct task_struct {
 	 * The buffer to hold the BTS data.
 	 */
 	void *bts_buffer;
+	size_t bts_size;
 #endif /* CONFIG_X86_PTRACE_BTS */
 
 	/* PID/PID hash table linkage. */