oprofile: arch-independent code for stack trace sampling
PATCH 2/11 oprofile: arch-independent code for stack trace sampling
Signature · 3 lines
-- Greg Banks, R&D Software Engineer, SGI Australian Software Group. I don't speak for SGI.
oprofile arch-independent updates, including some internal API changes and support for stack trace sampling.
---
drivers/oprofile/buffer_sync.c | 57 ++++++++++++--- drivers/oprofile/cpu_buffer.c | 141 ++++++++++++++++++++++++++++++-------- drivers/oprofile/cpu_buffer.h | 6 + drivers/oprofile/event_buffer.h | 2 drivers/oprofile/oprof.c | 74 +++++++++++-------- drivers/oprofile/oprof.h | 5 + drivers/oprofile/oprofile_files.c | 42 +++++++++-- drivers/oprofile/oprofile_stats.c | 4 + drivers/oprofile/oprofile_stats.h | 1 drivers/oprofile/timer_int.c | 19 +---- include/linux/oprofile.h | 20 ++++- 11 files changed, 279 insertions(+), 92 deletions(-)
drivers/oprofile/oprofile_files.c +37 -5
@@ -18,6 +18,37 @@ unsigned long fs_buffer_size = 131072;unsigned long fs_cpu_buffer_size =8192 ;unsigned long fs_buffer_watershed =32768 ;/* FIXME: tune */ +static ssize_t depth_read (struct file * file,char * buf,size_t count,loff_t * offset) +{ +return oprofilefs_ulong_to_user (backtrace_depth, buf, count, offset); +} + + +static ssize_t depth_write (struct file * file,char const * buf,size_t count,loff_t * offset) +{ +unsigned long val; +int retval; + +if (*offset) +return -EINVAL ; + + retval =oprofilefs_ulong_from_user (&val, buf, count); +if (retval) +return retval; + + retval =oprofile_set_backtrace (val); + +if (retval) +return retval; +return count; +} + + +static struct file_operations depth_fops = { + .read = depth_read, + .write = depth_write +}; +static ssize_t pointer_size_read (struct file * file,char __user * buf,size_t count,loff_t * offset) {
@@ -32,7 +63,7 @@ static struct file_operations pointer_sistatic ssize_t cpu_type_read (struct file * file,char __user * buf,size_t count,loff_t * offset) { -return oprofilefs_str_to_user (oprofile_ops->cpu_type, buf, count, offset); +return oprofilefs_str_to_user (oprofile_ops.cpu_type, buf, count, offset); }
@@ -47,7 +78,7 @@ static ssize_t enable_read(struct file * } -static ssize_t enable_write (struct file *file,char const __user * buf,size_t count,loff_t * offset) +static ssize_t enable_write (struct file * file,char const __user * buf,size_t count,loff_t * offset) {unsigned long val;int retval;
@@ -76,7 +107,7 @@ static struct file_operations enable_fop }; -static ssize_t dump_write (struct file *file,char const __user * buf,size_t count,loff_t * offset) +static ssize_t dump_write (struct file * file,char const __user * buf,size_t count,loff_t * offset) {wake_up_buffer_waiter ();return count;
@@ -96,8 +127,9 @@ void oprofile_create_files(struct super_oprofilefs_create_ulong (sb, root,"buffer_watershed" , &fs_buffer_watershed);oprofilefs_create_ulong (sb, root,"cpu_buffer_size" , &fs_cpu_buffer_size);oprofilefs_create_file (sb, root,"cpu_type" , &cpu_type_fops); +oprofilefs_create_file (sb, root,"backtrace_depth" , &depth_fops);oprofilefs_create_file (sb, root,"pointer_size" , &pointer_size_fops);oprofile_create_stats_files (sb, root); -if (oprofile_ops->create_files) - oprofile_ops->create_files (sb, root); +if (oprofile_ops.create_files) + oprofile_ops.create_files (sb, root); }
drivers/oprofile/oprofile_stats.h +1 -0
@@ -15,6 +15,7 @@struct oprofile_stat_struct {atomic_t sample_lost_no_mm;atomic_t sample_lost_no_mapping; +atomic_t bt_lost_no_mapping;atomic_t event_lost_overflow; };
include/linux/oprofile.h +15 -5
@@ -20,6 +20,7 @@struct super_block ;struct dentry ;struct file_operations ; +struct pt_regs ;/* Operations structure to be filled in */ struct oprofile_operations {
@@ -34,6 +35,8 @@ struct oprofile_operations {int (*start)(void );/* Stop delivering interrupts. */ void (*stop)(void ); +/* Initiate a stack backtrace. Optional. */ +void (*backtrace)(struct pt_regs *const regs,unsigned int depth);/* CPU identification string. */ char * cpu_type; };
@@ -41,11 +44,11 @@ struct oprofile_operations {/** * One-time initialisation. *ops must be set to a filled-in * operations structure. This is called even in timer interrupt -* mode. +* mode so an arch can set a backtrace callback. * -* Return 0 on success. +* If an error occurs, the fields should be left untouched. */ -int oprofile_arch_init (struct oprofile_operations ** ops); +void oprofile_arch_init (struct oprofile_operations * ops);/** * One-time exit/cleanup for the arch.
@@ -56,8 +59,15 @@ void oprofile_arch_exit(void);* Add a sample. This may be called from any context. Pass * smp_processor_id() as cpu. */ -extern void oprofile_add_sample (unsigned long eip,unsigned int is_kernel, -unsigned long event,int cpu); +void oprofile_add_sample (struct pt_regs *const regs,unsigned long event); + +/* Use this instead when the PC value is not from the regs. Doesn't +* backtrace. */ +void oprofile_add_pc (unsigned long pc,int is_kernel,unsigned long event); + +/* add a backtrace entry, to be called from the ->backtrace callback */ +void oprofile_add_trace (unsigned long eip); +/** * Create a file of the given name as a child of the given root, with
drivers/oprofile/cpu_buffer.c +115 -26
210 lines not on this page · raw message
drivers/oprofile/oprof.c +44 -30
127 lines not on this page · raw message
drivers/oprofile/cpu_buffer.h +6 -0
@@ -37,9 +37,11 @@ struct oprofile_cpu_buffer {unsigned long buffer_size;struct task_struct * last_task;int last_is_kernel; +int tracing;struct op_sample * buffer;unsigned long sample_received;unsigned long sample_lost_overflow; +unsigned long backtrace_aborted;int cpu;struct work_struct work; }____cacheline_aligned ;
@@ -48,4 +50,8 @@ extern struct oprofile_cpu_buffer cpu_buvoid cpu_buffer_reset (struct oprofile_cpu_buffer * cpu_buf); +/* transient events for the CPU buffer -> event buffer */ +#define CPU_IS_KERNEL 1 +#define CPU_TRACE_BEGIN 2 +#endif /* OPROFILE_CPU_BUFFER_H */
drivers/oprofile/buffer_sync.c +46 -11
136 lines not on this page · raw message
drivers/oprofile/timer_int.c +5 -14
@@ -17,10 +17,7 @@static int timer_notify (struct pt_regs *regs) { -int cpu =smp_processor_id (); -unsigned long eip =profile_pc (regs); - -oprofile_add_sample (eip, !user_mode (regs),0 , cpu); +oprofile_add_sample (regs,0 );return 0 ; }
@@ -36,15 +33,9 @@ static void timer_stop(void) } -static struct oprofile_operations timer_ops = { - .start = timer_start, - .stop = timer_stop, - .cpu_type ="timer" -}; - - -void __init timer_init (struct oprofile_operations ** ops) +void __init timer_init (struct oprofile_operations * ops) { - *ops = &timer_ops; -printk (KERN_INFO "oprofile: using timer interrupt.\n" ); + ops->start = timer_start; + ops->stop = timer_stop; + ops->cpu_type ="timer" ; }
drivers/oprofile/oprofile_stats.c +4 -0
@@ -59,6 +59,8 @@ void oprofile_create_stats_files(struct &cpu_buf->sample_received);oprofilefs_create_ro_ulong (sb, cpudir,"sample_lost_overflow" , &cpu_buf->sample_lost_overflow); +oprofilefs_create_ro_ulong (sb, cpudir,"backtrace_aborted" , + &cpu_buf->backtrace_aborted); } oprofilefs_create_ro_atomic(sb, dir, "sample_lost_no_mm",
@@ -67,4 +69,6 @@ void oprofile_create_stats_files(struct &oprofile_stats.sample_lost_no_mapping);oprofilefs_create_ro_atomic (sb, dir,"event_lost_overflow" , &oprofile_stats.event_lost_overflow); +oprofilefs_create_ro_atomic (sb, dir,"bt_lost_no_mapping" , + &oprofile_stats.bt_lost_no_mapping); }
drivers/oprofile/event_buffer.h +2 -0
@@ -32,6 +32,8 @@ void wake_up_buffer_waiter(void);#define KERNEL_EXIT_SWITCH_CODE 5#define MODULE_LOADED_CODE 6#define CTX_TGID_CODE 7 +#define TRACE_BEGIN_CODE 8 +#define TRACE_END_CODE 9/* add data to the event buffer */ void add_event_entry (unsigned long data);
drivers/oprofile/oprof.h +4 -1
@@ -24,12 +24,15 @@ struct oprofile_operations;extern unsigned long fs_buffer_size;extern unsigned long fs_cpu_buffer_size;extern unsigned long fs_buffer_watershed; -extern struct oprofile_operations * oprofile_ops; +extern struct oprofile_operations oprofile_ops;extern unsigned long oprofile_started; +extern unsigned long backtrace_depth;struct super_block ;struct dentry ;void oprofile_create_files (struct super_block * sb,struct dentry * root); + +int oprofile_set_backtrace (unsigned long depth);#endif /* OPROF_H */
Review
Quoted 2 lines · Greg Banks
(quoted diff) +1
-0+ struct oprofile_cpu_buffer * cpu_buf = &cpu_buffer[smp_processor_id()];
oprofile is currently doing suspicious things with smp_processor_id() in premptible reasons. Is this patch compounding things?
Quoted 6 lines · Andrew Morton
Review
Quoted 2 lines · Greg Banks
(quoted diff) +1
-0+ struct oprofile_cpu_buffer * cpu_buf = &cpu_buffer[smp_processor_id()];oprofile is currently doing suspicious things with
smp_processor_id()in premptible reasons. Is this patch compounding things?
It's not changing the contexts where smp_processor_id() is called, just pushing it down one level from a bunch of interrupt handlers to the 2 oprofile sampling functions they call. If it was busted before it's no more nor less busted now.
I presume the perceived problem is that with CONFIG_PREEMPT=y the thread can be pre-empted onto another CPU? If it makes everyone happier I can sprinkle a few preempt_disable()s around, but I'd prefer to do it in a subsequent patch rather than respin this.
Greg.
Signature · 2 lines
Greg Banks, R&D Software Engineer, SGI Australian Software Group. I don't speak for SGI.
Quoted 8 lines · Greg Banks
Quoted 2 lines
oprofile is currently doing suspicious things with
smp_processor_id()in premptible reasons. Is this patch compounding things?It's not changing the contexts where
smp_processor_id()is called, just pushing it down one level from a bunch of interrupt handlers to the 2 oprofile sampling functions they call. If it was busted before it's no more nor less busted now.
Quoted 4 lines
I presume the perceived problem is that with
CONFIG_PREEMPT=ythe thread can be pre-empted onto another CPU? If it makes everyone happier I can sprinkle a few preempt_disable()s around, but I'd prefer to do it in a subsequent patch rather than respin this.
Andrew: basically the warning is false, there is no bug in this code.
we don't want to use preempt_disable(). Instead we want some way to get a CPU ID and then carry on in pre-emptible fashion. It's only used to index into an array, and if we get pre-empted onto another CPU it's not a major deal.
(Yes, this breaks with CPU hotplug, but so does the rest of OProfile and I've yet to see a sensible API for handling this, that is a ctor/dtor style API)
regards
john