tracing/kprobes: kprobe-tracer updates
cover tracing/kprobes: kprobe-tracer updates
Hi Frederic,
Here is a few updates for kprobe-tracer which fix reported bugs and is for catching up the latest ftrace.
- Use dynamic percpu profiling buffer patch is for catching up the latest ftrace change.
- Checking the result of create_event_dir() patch improves ftrace code pointed by Li Zefan (thanks!).
- Add VIA instructions to x86 insn decoder patch fixes decoding bug reported by Ingo Molnar (thanks!).
Soon after this series, I'll send perf-kprobe patch series.
Thank you,
---
Masami Hiramatsu (3):
tracing/ftrace: Fix to check create_event_dir() when adding new events
x86: Add VIA processor instructions
tracing/kprobes: Use global event profile buffer in kprobe tracer
arch/x86/lib/x86-opcode-map.txt | 8 ++- kernel/trace/trace_events.c | 29 +++++++--- kernel/trace/trace_kprobe.c | 115 +++++++++++++++++++++++++-------------- 3 files changed, 101 insertions(+), 51 deletions(-)
Signature · 7 lines
Masami Hiramatsu Software Engineer Hitachi Computer Products (America), Inc. Software Solutions Division e-mail: [email protected]
PATCH 1/3 tracing/kprobes: Use global event profile buffer in kprobe tracer
Use new percpu global event profile buffer instead of stack in kprobe tracer.
---
kernel/trace/trace_kprobe.c | 115 +++++++++++++++++++++++++++---------------- 1 files changed, 73 insertions(+), 42 deletions(-)
kernel/trace/trace_kprobe.c +73 -42
@@ -1149,35 +1149,49 @@ static __kprobes int kprobe_profile_func(struct kprobe *kp, struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);struct ftrace_event_call *call = &tp->call;struct kprobe_trace_entry *entry; -int size, __size, i, pc; +struct trace_entry *ent; +int size, __size, i, pc, __cpu;unsigned long irq_flags; +char *raw_data; -local_save_flags (irq_flags); pc =preempt_count (); - __size =SIZEOF_KPROBE_TRACE_ENTRY (tp->nr_args); size =ALIGN (__size +sizeof (u32),sizeof (u64)); size -=sizeof (u32); +if (WARN_ONCE (size >FTRACE_MAX_PROFILE_SIZE , +"profile buffer not large enough" )) +return 0 ; -do { -char raw_data[size]; -struct trace_entry *ent; -/* -* Zero dead bytes from alignment to avoid stack leak -* to userspace -*/ - *(u64 *)(&raw_data[size -sizeof (u64)]) =0ULL ; - entry = (struct kprobe_trace_entry *)raw_data; - ent = &entry->ent; - -tracing_generic_entry_update (ent, irq_flags, pc); - ent->type = call->id; - entry->nargs = tp->nr_args; - entry->ip = (unsigned long )kp->addr; -for (i =0 ; i < tp->nr_args; i++) - entry->args[i] =call_fetch (&tp->args[i].fetch, regs); -perf_tp_event (call->id, entry->ip,1 , entry, size); - }while (0 ); +/* +* Protect the non nmi buffer +* This also protects the rcu read side +*/ +local_irq_save (irq_flags); + __cpu =smp_processor_id (); + +if (in_nmi ()) + raw_data =rcu_dereference (trace_profile_buf_nmi); +else + raw_data =rcu_dereference (trace_profile_buf); + +if (!raw_data) + goto end; + + raw_data =per_cpu_ptr (raw_data, __cpu); +/* Zero dead bytes from alignment to avoid buffer leak to userspace */ + *(u64 *)(&raw_data[size -sizeof (u64)]) =0ULL ; + entry = (struct kprobe_trace_entry *)raw_data; + ent = &entry->ent; + +tracing_generic_entry_update (ent, irq_flags, pc); + ent->type = call->id; + entry->nargs = tp->nr_args; + entry->ip = (unsigned long )kp->addr; +for (i =0 ; i < tp->nr_args; i++) + entry->args[i] =call_fetch (&tp->args[i].fetch, regs); +perf_tp_event (call->id, entry->ip,1 , entry, size); +end: +local_irq_restore (irq_flags);return 0 ; }
@@ -1188,33 +1202,50 @@ static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri, struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);struct ftrace_event_call *call = &tp->call;struct kretprobe_trace_entry *entry; -int size, __size, i, pc; +struct trace_entry *ent; +int size, __size, i, pc, __cpu;unsigned long irq_flags; +char *raw_data; -local_save_flags (irq_flags); pc =preempt_count (); - __size =SIZEOF_KRETPROBE_TRACE_ENTRY (tp->nr_args); size =ALIGN (__size +sizeof (u32),sizeof (u64)); size -=sizeof (u32); +if (WARN_ONCE (size >FTRACE_MAX_PROFILE_SIZE , +"profile buffer not large enough" )) +return 0 ; + +/* +* Protect the non nmi buffer +* This also protects the rcu read side +*/ +local_irq_save (irq_flags); + __cpu =smp_processor_id (); + +if (in_nmi ()) + raw_data =rcu_dereference (trace_profile_buf_nmi); +else + raw_data =rcu_dereference (trace_profile_buf); + +if (!raw_data) + goto end; + + raw_data =per_cpu_ptr (raw_data, __cpu); +/* Zero dead bytes from alignment to avoid buffer leak to userspace */ + *(u64 *)(&raw_data[size -sizeof (u64)]) =0ULL ; + entry = (struct kretprobe_trace_entry *)raw_data; + ent = &entry->ent; -do { -char raw_data[size]; -struct trace_entry *ent; - - *(u64 *)(&raw_data[size -sizeof (u64)]) =0ULL ; - entry = (struct kretprobe_trace_entry *)raw_data; - ent = &entry->ent; - -tracing_generic_entry_update (ent, irq_flags, pc); - ent->type = call->id; - entry->nargs = tp->nr_args; - entry->func = (unsigned long )tp->rp.kp.addr; - entry->ret_ip = (unsigned long )ri->ret_addr; -for (i =0 ; i < tp->nr_args; i++) - entry->args[i] =call_fetch (&tp->args[i].fetch, regs); -perf_tp_event (call->id, entry->ret_ip,1 , entry, size); - }while (0 ); +tracing_generic_entry_update (ent, irq_flags, pc); + ent->type = call->id; + entry->nargs = tp->nr_args; + entry->func = (unsigned long )tp->rp.kp.addr; + entry->ret_ip = (unsigned long )ri->ret_addr; +for (i =0 ; i < tp->nr_args; i++) + entry->args[i] =call_fetch (&tp->args[i].fetch, regs); +perf_tp_event (call->id, entry->ret_ip,1 , entry, size); +end: +local_irq_restore (irq_flags);return 0 ; }
Signature · 7 lines
Masami Hiramatsu Software Engineer Hitachi Computer Products (America), Inc. Software Solutions Division e-mail: [email protected]
PATCH 2/3 x86: Add VIA processor instructions
In reply to Masami HiramatsuAdd VIA processor's Padlock instructions(MONTMUL, XSHA1, XSHA256).
---
arch/x86/lib/x86-opcode-map.txt | 8 +++++++- 1 files changed, 7 insertions(+), 1 deletions(-)
arch/x86/lib/x86-opcode-map.txt +7 -1
@@ -469,7 +469,7 @@ a2: CPUID a3: BT Ev,Gv a4: SHLD Ev,Gv,Ib a5: SHLD Ev,Gv,CL -a6: +a6: GrpPDLK a7: GrpRNG a8: PUSH GS (d64) a9: POP GS (d64)
@@ -803,6 +803,12 @@ GrpTable: Grp16 3: prefetch T2 EndTable +GrpTable: GrpPDLK +0: MONTMUL +1: XSHA1 +2: XSHA2 +EndTable + GrpTable: GrpRNG 0: xstore-rng 1: xcrypt-ecb
Signature · 7 lines
Masami Hiramatsu Software Engineer Hitachi Computer Products (America), Inc. Software Solutions Division e-mail: [email protected]
PATCH 3/3 tracing/ftrace: Fix to check create_event_dir() when adding new events
In reply to Masami HiramatsuCheck result of create_event_dir() and add ftrace_event_call to ftrace_events list only if it is succeeded. Thanks Li for pointing it out.
---
kernel/trace/trace_events.c | 29 +++++++++++++++++++++-------- 1 files changed, 21 insertions(+), 8 deletions(-)
kernel/trace/trace_events.c +21 -8
@@ -957,12 +957,14 @@ static int __trace_add_event_call(struct ftrace_event_call *call)if (!d_events)return -ENOENT ; -list_add (&call->list, &ftrace_events); ret =event_create_dir (call, d_events, &ftrace_event_id_fops, &ftrace_enable_fops, &ftrace_event_filter_fops, &ftrace_event_format_fops);if (ret <0 ) -list_del (&call->list); +pr_warning ("Could not create directory of trace events/%s\n" , + call->name); +else +list_add (&call->list, &ftrace_events);return ret; }
@@ -1124,10 +1126,15 @@ static void trace_module_add_events(struct module *mod)return ; } call->mod = mod; + ret =event_create_dir (call, d_events, + &file_ops->id, &file_ops->enable, + &file_ops->filter, &file_ops->format); +if (ret <0 ) { +pr_warning ("Could not create directory of trace " +"point events/%s\n" , call->name); +continue ; + }list_add (&call->list, &ftrace_events); -event_create_dir (call, d_events, - &file_ops->id, &file_ops->enable, - &file_ops->filter, &file_ops->format); } }
@@ -1267,10 +1274,16 @@ static __init int event_trace_init(void)continue ; } } + ret =event_create_dir (call, d_events, &ftrace_event_id_fops, + &ftrace_enable_fops, + &ftrace_event_filter_fops, + &ftrace_event_format_fops); +if (ret <0 ) { +pr_warning ("Could not create directory of trace " +"point events/%s\n" , call->name); +continue ; + }list_add (&call->list, &ftrace_events); -event_create_dir (call, d_events, &ftrace_event_id_fops, - &ftrace_enable_fops, &ftrace_event_filter_fops, - &ftrace_event_format_fops); }while (true) {
Signature · 7 lines
Masami Hiramatsu Software Engineer Hitachi Computer Products (America), Inc. Software Solutions Division e-mail: [email protected]
Quoted 5 lines · Masami Hiramatsu
Hi Frederic,
Here is a few updates for kprobe-tracer which fix reported bugs and is for catching up the latest ftrace.
Quoted 8 lines
- Use dynamic percpu profiling buffer patch is for catching up the latest ftrace change.
- Checking the result ofcreate_event_dir()patch improves ftrace code pointed by Li Zefan (thanks!).
- Add VIA instructions to x86 insn decoder patch fixes decoding bug reported by Ingo Molnar (thanks!).Soon after this series, I'll send perf-kprobe patch series.
-- Steve
Re: [PATCH tracing/kprobes 3/3] tracing/ftrace: Fix to check create_event_dir() when adding new events
In reply to Masami HiramatsuQuoted 31 lines · Masami Hiramatsu
Check result of
create_event_dir()and addftrace_event_calltoftrace_eventslist only if it is succeeded. Thanks Li for pointing it out.Signed-off-by Masami Hiramatsu <[email protected]>Cc Steven Rostedt <[email protected]>Cc Jim Keniston <[email protected]>Cc Ananth N Mavinakayanahalli <[email protected]>Cc Andi Kleen <[email protected]>Cc Christoph Hellwig <[email protected]>Cc Frank Ch. Eigler <[email protected]>Cc Frederic Weisbecker <[email protected]>Cc H. Peter Anvin <[email protected]>Cc Ingo Molnar <[email protected]>Cc Jason Baron <[email protected]>Cc K.Prasad <[email protected]>Cc Lai Jiangshan <[email protected]>Cc Li Zefan <[email protected]>Cc Peter Zijlstra <[email protected]>Cc Srikar Dronamraju <[email protected]>Cc Tom Zanussi <[email protected]>---
kernel/trace/trace_events.c | 29 +++++++++++++++++++++-------- 1 files changed, 21 insertions(+), 8 deletions(-)kernel/trace/trace_events.c +0
-0@@ -957,12 +957,14 @@ static int __trace_add_event_call(struct ftrace_event_call *call)if (!d_events)
kernel/trace/trace_events.c static int __trace_add_event_call(struct ftrace_event_call *call) · line 965
Quoted 10 lines
kernel/trace/trace_events.c +2
-2return -ENOENT ; -list_add (&call->list, &ftrace_events); ret =event_create_dir (call, d_events, &ftrace_event_id_fops, &ftrace_enable_fops, &ftrace_event_filter_fops, &ftrace_event_format_fops);if (ret <0 ) -list_del (&call->list); +pr_warning ("Could not create directory of trace events/%s\n" , + call->name);
We already have such warnings in event_create_dir, with even more granularity against the failure reason. The patch is fine, but I'll just remove the warning introduced inside while applying it.
Thanks.
PS: Also I guess we could add the event in the list right from event_create_dir()...
Frederic Weisbecker
AppliedQuoted 12 lines · Masami Hiramatsu
Hi Frederic,
Here is a few updates for kprobe-tracer which fix reported bugs and is for catching up the latest ftrace.
- Use dynamic percpu profiling buffer patch is for catching up the latest ftrace change.
- Checking the result ofcreate_event_dir()patch improves ftrace code pointed by Li Zefan (thanks!).
- Add VIA instructions to x86 insn decoder patch fixes decoding bug reported by Ingo Molnar (thanks!).
Quoted 3 lines
Soon after this series, I'll send perf-kprobe patch series.
Thank you,
Applied, thanks Masami!
[tip:perf/probes] tracing/kprobes: Use global event perf buffers in kprobe tracer
In reply to Masami HiramatsuCommit-ID: a1a138d05fa060ac4238c19a1e890aacc25ed3ba Gitweb: http://git.kernel.org/tip/a1a138d05fa060ac4238c19a1e890aacc25ed3ba Author: Masami Hiramatsu <[email protected]> AuthorDate: Fri, 25 Sep 2009 11:20:12 -0700 Committer: Frederic Weisbecker <[email protected]> CommitDate: Sat, 3 Oct 2009 02:21:39 +0200
tracing/kprobes: Use global event perf buffers in kprobe tracer
Use new percpu global event buffer instead of stack in kprobe tracer while tracing through perf.
LKML-Reference: <20090925182011.10157.60140.stgit@omoto>
--- kernel/trace/trace_kprobe.c | 115 +++++++++++++++++++++++++++---------------- 1 files changed, 73 insertions(+), 42 deletions(-)
kernel/trace/trace_kprobe.c +73 -42
@@ -1149,35 +1149,49 @@ static __kprobes int kprobe_profile_func(struct kprobe *kp, struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);struct ftrace_event_call *call = &tp->call;struct kprobe_trace_entry *entry; -int size, __size, i, pc; +struct trace_entry *ent; +int size, __size, i, pc, __cpu;unsigned long irq_flags; +char *raw_data; -local_save_flags (irq_flags); pc =preempt_count (); - __size =SIZEOF_KPROBE_TRACE_ENTRY (tp->nr_args); size =ALIGN (__size +sizeof (u32),sizeof (u64)); size -=sizeof (u32); +if (WARN_ONCE (size >FTRACE_MAX_PROFILE_SIZE , +"profile buffer not large enough" )) +return 0 ; -do { -char raw_data[size]; -struct trace_entry *ent; -/* -* Zero dead bytes from alignment to avoid stack leak -* to userspace -*/ - *(u64 *)(&raw_data[size -sizeof (u64)]) =0ULL ; - entry = (struct kprobe_trace_entry *)raw_data; - ent = &entry->ent; - -tracing_generic_entry_update (ent, irq_flags, pc); - ent->type = call->id; - entry->nargs = tp->nr_args; - entry->ip = (unsigned long )kp->addr; -for (i =0 ; i < tp->nr_args; i++) - entry->args[i] =call_fetch (&tp->args[i].fetch, regs); -perf_tp_event (call->id, entry->ip,1 , entry, size); - }while (0 ); +/* +* Protect the non nmi buffer +* This also protects the rcu read side +*/ +local_irq_save (irq_flags); + __cpu =smp_processor_id (); + +if (in_nmi ()) + raw_data =rcu_dereference (trace_profile_buf_nmi); +else + raw_data =rcu_dereference (trace_profile_buf); + +if (!raw_data) + goto end; + + raw_data =per_cpu_ptr (raw_data, __cpu); +/* Zero dead bytes from alignment to avoid buffer leak to userspace */ + *(u64 *)(&raw_data[size -sizeof (u64)]) =0ULL ; + entry = (struct kprobe_trace_entry *)raw_data; + ent = &entry->ent; + +tracing_generic_entry_update (ent, irq_flags, pc); + ent->type = call->id; + entry->nargs = tp->nr_args; + entry->ip = (unsigned long )kp->addr; +for (i =0 ; i < tp->nr_args; i++) + entry->args[i] =call_fetch (&tp->args[i].fetch, regs); +perf_tp_event (call->id, entry->ip,1 , entry, size); +end: +local_irq_restore (irq_flags);return 0 ; }
@@ -1188,33 +1202,50 @@ static __kprobes int kretprobe_profile_func(struct kretprobe_instance *ri, struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);struct ftrace_event_call *call = &tp->call;struct kretprobe_trace_entry *entry; -int size, __size, i, pc; +struct trace_entry *ent; +int size, __size, i, pc, __cpu;unsigned long irq_flags; +char *raw_data; -local_save_flags (irq_flags); pc =preempt_count (); - __size =SIZEOF_KRETPROBE_TRACE_ENTRY (tp->nr_args); size =ALIGN (__size +sizeof (u32),sizeof (u64)); size -=sizeof (u32); +if (WARN_ONCE (size >FTRACE_MAX_PROFILE_SIZE , +"profile buffer not large enough" )) +return 0 ; + +/* +* Protect the non nmi buffer +* This also protects the rcu read side +*/ +local_irq_save (irq_flags); + __cpu =smp_processor_id (); + +if (in_nmi ()) + raw_data =rcu_dereference (trace_profile_buf_nmi); +else + raw_data =rcu_dereference (trace_profile_buf); + +if (!raw_data) + goto end; + + raw_data =per_cpu_ptr (raw_data, __cpu); +/* Zero dead bytes from alignment to avoid buffer leak to userspace */ + *(u64 *)(&raw_data[size -sizeof (u64)]) =0ULL ; + entry = (struct kretprobe_trace_entry *)raw_data; + ent = &entry->ent; -do { -char raw_data[size]; -struct trace_entry *ent; - - *(u64 *)(&raw_data[size -sizeof (u64)]) =0ULL ; - entry = (struct kretprobe_trace_entry *)raw_data; - ent = &entry->ent; - -tracing_generic_entry_update (ent, irq_flags, pc); - ent->type = call->id; - entry->nargs = tp->nr_args; - entry->func = (unsigned long )tp->rp.kp.addr; - entry->ret_ip = (unsigned long )ri->ret_addr; -for (i =0 ; i < tp->nr_args; i++) - entry->args[i] =call_fetch (&tp->args[i].fetch, regs); -perf_tp_event (call->id, entry->ret_ip,1 , entry, size); - }while (0 ); +tracing_generic_entry_update (ent, irq_flags, pc); + ent->type = call->id; + entry->nargs = tp->nr_args; + entry->func = (unsigned long )tp->rp.kp.addr; + entry->ret_ip = (unsigned long )ri->ret_addr; +for (i =0 ; i < tp->nr_args; i++) + entry->args[i] =call_fetch (&tp->args[i].fetch, regs); +perf_tp_event (call->id, entry->ret_ip,1 , entry, size); +end: +local_irq_restore (irq_flags);return 0 ; }
[tip:perf/probes] x86: Add VIA processor instructions in opcodes decoder
In reply to Masami HiramatsuCommit-ID: c0b11d3af164947c71e2491912c5b8418900dafb Gitweb: http://git.kernel.org/tip/c0b11d3af164947c71e2491912c5b8418900dafb Author: Masami Hiramatsu <[email protected]> AuthorDate: Fri, 25 Sep 2009 11:20:38 -0700 Committer: Frederic Weisbecker <[email protected]> CommitDate: Sat, 3 Oct 2009 02:43:00 +0200
x86: Add VIA processor instructions in opcodes decoder
Add VIA processor's Padlock instructions(MONTMUL, XSHA1, XSHA256) as parts of the kernel may use them.
This fixes the following crash in opcodes decoder selftests:
make[2]: `scripts/unifdef' is up to date.
TEST posttest
Error: c145cf71: f3 0f a6 d0 repz xsha256
Error: objdump says 4 bytes, but insn_get_length() says 3 (attr:0) make[1]: *** [posttest] Error 2
make: *** [bzImage] Error 2
LKML-Reference: <20090925182037.10157.3180.stgit@omoto>
--- arch/x86/lib/x86-opcode-map.txt | 8 +++++++- 1 files changed, 7 insertions(+), 1 deletions(-)
arch/x86/lib/x86-opcode-map.txt +7 -1
@@ -469,7 +469,7 @@ a2: CPUID a3: BT Ev,Gv a4: SHLD Ev,Gv,Ib a5: SHLD Ev,Gv,CL -a6: +a6: GrpPDLK a7: GrpRNG a8: PUSH GS (d64) a9: POP GS (d64)
@@ -803,6 +803,12 @@ GrpTable: Grp16 3: prefetch T2 EndTable +GrpTable: GrpPDLK +0: MONTMUL +1: XSHA1 +2: XSHA2 +EndTable + GrpTable: GrpRNG 0: xstore-rng 1: xcrypt-ecb
[tip:perf/probes] tracing/ftrace: Fix to check create_event_dir() when adding new events
In reply to Masami HiramatsuCommit-ID: 88f70d7590538e427c8405a2e02ac2624847386c Gitweb: http://git.kernel.org/tip/88f70d7590538e427c8405a2e02ac2624847386c Author: Masami Hiramatsu <[email protected]> AuthorDate: Fri, 25 Sep 2009 11:20:54 -0700 Committer: Frederic Weisbecker <[email protected]> CommitDate: Sat, 3 Oct 2009 03:04:58 +0200
tracing/ftrace: Fix to check create_event_dir() when adding new events
Check result of event_create_dir() and add ftrace_event_call to ftrace_events list only if it is succeeded. Thanks to Li for pointing it out.
LKML-Reference: <20090925182054.10157.55219.stgit@omoto>
--- kernel/trace/trace_events.c | 25 ++++++++++++++----------- 1 files changed, 14 insertions(+), 11 deletions(-)
kernel/trace/trace_events.c +14 -11
@@ -957,12 +957,12 @@ static int __trace_add_event_call(struct ftrace_event_call *call)if (!d_events)return -ENOENT ; -list_add (&call->list, &ftrace_events); ret =event_create_dir (call, d_events, &ftrace_event_id_fops, &ftrace_enable_fops, &ftrace_event_filter_fops, &ftrace_event_format_fops); -if (ret <0 ) -list_del (&call->list); +if (!ret) +list_add (&call->list, &ftrace_events); +return ret; }
@@ -1124,10 +1124,11 @@ static void trace_module_add_events(struct module *mod)return ; } call->mod = mod; -list_add (&call->list, &ftrace_events); -event_create_dir (call, d_events, - &file_ops->id, &file_ops->enable, - &file_ops->filter, &file_ops->format); + ret =event_create_dir (call, d_events, + &file_ops->id, &file_ops->enable, + &file_ops->filter, &file_ops->format); +if (!ret) +list_add (&call->list, &ftrace_events); } }
@@ -1267,10 +1268,12 @@ static __init int event_trace_init(void)continue ; } } -list_add (&call->list, &ftrace_events); -event_create_dir (call, d_events, &ftrace_event_id_fops, - &ftrace_enable_fops, &ftrace_event_filter_fops, - &ftrace_event_format_fops); + ret =event_create_dir (call, d_events, &ftrace_event_id_fops, + &ftrace_enable_fops, + &ftrace_event_filter_fops, + &ftrace_event_format_fops); +if (!ret) +list_add (&call->list, &ftrace_events); }while (true) {