[PATCH 1/3] tracing/kprobes: Use global event perf buffers in kprobe tracer

From: Frederic Weisbecker
Date: Fri Oct 02 2009 - 22:32:13 EST


From: Masami Hiramatsu <mhiramat@xxxxxxxxxx>

Use new percpu global event buffer instead of stack in kprobe
tracer while tracing through perf.

Signed-off-by: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
Acked-by: Steven Rostedt <rostedt@xxxxxxxxxxx>
Acked-by: Ingo Molnar <mingo@xxxxxxx>
Cc: Jim Keniston <jkenisto@xxxxxxxxxx>
Cc: Ananth N Mavinakayanahalli <ananth@xxxxxxxxxx>
Cc: Andi Kleen <ak@xxxxxxxxxxxxxxx>
Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Cc: Frank Ch. Eigler <fche@xxxxxxxxxx>
Cc: H. Peter Anvin <hpa@xxxxxxxxx>
Cc: Jason Baron <jbaron@xxxxxxxxxx>
Cc: K.Prasad <prasad@xxxxxxxxxxxxxxxxxx>
Cc: Lai Jiangshan <laijs@xxxxxxxxxxxxxx>
Cc: Li Zefan <lizf@xxxxxxxxxxxxxx>
Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Cc: Srikar Dronamraju <srikar@xxxxxxxxxxxxxxxxxx>
Cc: Tom Zanussi <tzanussi@xxxxxxxxx>
LKML-Reference: <20090925182011.10157.60140.stgit@omoto>
Signed-off-by: Frederic Weisbecker <fweisbec@xxxxxxxxx>
---
kernel/trace/trace_kprobe.c | 115 +++++++++++++++++++++++++++----------------
1 files changed, 73 insertions(+), 42 deletions(-)

diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
index 09cba27..97309d4 100644
--- a/kernel/trace/trace_kprobe.c
+++ b/kernel/trace/trace_kprobe.c
@@ -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;
}

--
1.6.2.3

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/