Introduce callstack legacy implementation header
authorMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Tue, 21 May 2019 21:51:23 +0000 (17:51 -0400)
committerMathieu Desnoyers <mathieu.desnoyers@efficios.com>
Wed, 22 May 2019 14:48:30 +0000 (10:48 -0400)
Split the callstack code: keep boilerplate code within the
C implementation file, and move the parts which depend on the
"legacy" (pre-stackwalk) stacktrace kernel API to a separate
implementation header.

This is a preparation step to introduce a new implementation
header for the stackwalk API, added in Linux 5.2 and gradually
integrated within each architecture.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
lttng-context-callstack-legacy-impl.h [new file with mode: 0644]
lttng-context-callstack.c

diff --git a/lttng-context-callstack-legacy-impl.h b/lttng-context-callstack-legacy-impl.h
new file mode 100644 (file)
index 0000000..a68e8a1
--- /dev/null
@@ -0,0 +1,190 @@
+/* SPDX-License-Identifier: (GPL-2.0 or LGPL-2.1)
+ *
+ * lttng-context-callstack-legacy-impl.h
+ *
+ * LTTng callstack event context, legacy implementation. Targets
+ * kernels and architectures not yet using the stacktrace common
+ * infrastructure introduced in the upstream Linux kernel by commit
+ * 214d8ca6ee "stacktrace: Provide common infrastructure" (merged in
+ * Linux 5.2, then gradually introduced within architectures).
+ *
+ * Copyright (C) 2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ * Copyright (C) 2014 Francis Giraldeau <francis.giraldeau@gmail.com>
+ */
+
+#define MAX_ENTRIES 128
+
+enum lttng_cs_ctx_modes {
+       CALLSTACK_KERNEL = 0,
+       CALLSTACK_USER = 1,
+       NR_CALLSTACK_MODES,
+};
+
+struct lttng_cs_dispatch {
+       struct stack_trace stack_trace;
+       unsigned long entries[MAX_ENTRIES];
+};
+
+struct lttng_cs {
+       struct lttng_cs_dispatch dispatch[RING_BUFFER_MAX_NESTING];
+};
+
+struct field_data {
+       struct lttng_cs __percpu *cs_percpu;
+       enum lttng_cs_ctx_modes mode;
+};
+
+struct lttng_cs_type {
+       const char *name;
+       const char *save_func_name;
+       void (*save_func)(struct stack_trace *trace);
+};
+
+static struct lttng_cs_type cs_types[] = {
+       {
+               .name           = "callstack_kernel",
+               .save_func_name = "save_stack_trace",
+               .save_func      = NULL,
+       },
+       {
+               .name           = "callstack_user",
+               .save_func_name = "save_stack_trace_user",
+               .save_func      = NULL,
+       },
+};
+
+static
+int init_type(enum lttng_cs_ctx_modes mode)
+{
+       unsigned long func;
+
+       if (cs_types[mode].save_func)
+               return 0;
+       func = kallsyms_lookup_funcptr(cs_types[mode].save_func_name);
+       if (!func) {
+               printk(KERN_WARNING "LTTng: symbol lookup failed: %s\n",
+                               cs_types[mode].save_func_name);
+               return -EINVAL;
+       }
+       cs_types[mode].save_func = (void *) func;
+       return 0;
+}
+
+/* Keep track of nesting inside userspace callstack context code */
+DEFINE_PER_CPU(int, callstack_user_nesting);
+
+static
+struct stack_trace *stack_trace_context(struct lttng_ctx_field *field,
+                                       struct lib_ring_buffer_ctx *ctx)
+{
+       int buffer_nesting, cs_user_nesting;
+       struct lttng_cs *cs;
+       struct field_data *fdata = field->priv;
+
+       /*
+        * Do not gather the userspace callstack context when the event was
+        * triggered by the userspace callstack context saving mechanism.
+        */
+       cs_user_nesting = per_cpu(callstack_user_nesting, ctx->cpu);
+
+       if (fdata->mode == CALLSTACK_USER && cs_user_nesting >= 1)
+               return NULL;
+
+       /*
+        * get_cpu() is not required, preemption is already
+        * disabled while event is written.
+        *
+        * max nesting is checked in lib_ring_buffer_get_cpu().
+        * Check it again as a safety net.
+        */
+       cs = per_cpu_ptr(fdata->cs_percpu, ctx->cpu);
+       buffer_nesting = per_cpu(lib_ring_buffer_nesting, ctx->cpu) - 1;
+       if (buffer_nesting >= RING_BUFFER_MAX_NESTING)
+               return NULL;
+
+       return &cs->dispatch[buffer_nesting].stack_trace;
+}
+
+/*
+ * In order to reserve the correct size, the callstack is computed. The
+ * resulting callstack is saved to be accessed in the record step.
+ */
+static
+size_t lttng_callstack_get_size(size_t offset, struct lttng_ctx_field *field,
+                               struct lib_ring_buffer_ctx *ctx,
+                               struct lttng_channel *chan)
+{
+       struct stack_trace *trace;
+       struct field_data *fdata = field->priv;
+       size_t orig_offset = offset;
+
+       /* do not write data if no space is available */
+       trace = stack_trace_context(field, ctx);
+       if (unlikely(!trace)) {
+               offset += lib_ring_buffer_align(offset, lttng_alignof(unsigned int));
+               offset += sizeof(unsigned int);
+               offset += lib_ring_buffer_align(offset, lttng_alignof(unsigned long));
+               return offset - orig_offset;
+       }
+
+       /* reset stack trace, no need to clear memory */
+       trace->nr_entries = 0;
+
+       if (fdata->mode == CALLSTACK_USER)
+               ++per_cpu(callstack_user_nesting, ctx->cpu);
+
+       /* do the real work and reserve space */
+       cs_types[fdata->mode].save_func(trace);
+
+       if (fdata->mode == CALLSTACK_USER)
+               per_cpu(callstack_user_nesting, ctx->cpu)--;
+
+       /*
+        * Remove final ULONG_MAX delimiter. If we cannot find it, add
+        * our own marker to show that the stack is incomplete. This is
+        * more compact for a trace.
+        */
+       if (trace->nr_entries > 0
+                       && trace->entries[trace->nr_entries - 1] == ULONG_MAX) {
+               trace->nr_entries--;
+       }
+       offset += lib_ring_buffer_align(offset, lttng_alignof(unsigned int));
+       offset += sizeof(unsigned int);
+       offset += lib_ring_buffer_align(offset, lttng_alignof(unsigned long));
+       offset += sizeof(unsigned long) * trace->nr_entries;
+       /* Add our own ULONG_MAX delimiter to show incomplete stack. */
+       if (trace->nr_entries == trace->max_entries)
+               offset += sizeof(unsigned long);
+       return offset - orig_offset;
+}
+
+static
+void lttng_callstack_record(struct lttng_ctx_field *field,
+                       struct lib_ring_buffer_ctx *ctx,
+                       struct lttng_channel *chan)
+{
+       struct stack_trace *trace = stack_trace_context(field, ctx);
+       unsigned int nr_seq_entries;
+
+       if (unlikely(!trace)) {
+               nr_seq_entries = 0;
+               lib_ring_buffer_align_ctx(ctx, lttng_alignof(unsigned int));
+               chan->ops->event_write(ctx, &nr_seq_entries, sizeof(unsigned int));
+               lib_ring_buffer_align_ctx(ctx, lttng_alignof(unsigned long));
+               return;
+       }
+       lib_ring_buffer_align_ctx(ctx, lttng_alignof(unsigned int));
+       nr_seq_entries = trace->nr_entries;
+       if (trace->nr_entries == trace->max_entries)
+               nr_seq_entries++;
+       chan->ops->event_write(ctx, &nr_seq_entries, sizeof(unsigned int));
+       lib_ring_buffer_align_ctx(ctx, lttng_alignof(unsigned long));
+       chan->ops->event_write(ctx, trace->entries,
+                       sizeof(unsigned long) * trace->nr_entries);
+       /* Add our own ULONG_MAX delimiter to show incomplete stack. */
+       if (trace->nr_entries == trace->max_entries) {
+               unsigned long delim = ULONG_MAX;
+
+               chan->ops->event_write(ctx, &delim, sizeof(unsigned long));
+       }
+}
index 0876b12269ecc26a90c73b948d8c0dc9bb954f6e..3fccfcf270bf34e48053a5ab30b761d1dc5bffd6 100644 (file)
 #include "wrapper/vmalloc.h"
 #include "lttng-tracer.h"
 
-#define MAX_ENTRIES 128
-
-enum lttng_cs_ctx_modes {
-       CALLSTACK_KERNEL = 0,
-       CALLSTACK_USER = 1,
-       NR_CALLSTACK_MODES,
-};
-
-struct lttng_cs_dispatch {
-       struct stack_trace stack_trace;
-       unsigned long entries[MAX_ENTRIES];
-};
-
-struct lttng_cs {
-       struct lttng_cs_dispatch dispatch[RING_BUFFER_MAX_NESTING];
-};
-
-struct field_data {
-       struct lttng_cs __percpu *cs_percpu;
-       enum lttng_cs_ctx_modes mode;
-};
-
-struct lttng_cs_type {
-       const char *name;
-       const char *save_func_name;
-       void (*save_func)(struct stack_trace *trace);
-};
-
-static struct lttng_cs_type cs_types[] = {
-       {
-               .name           = "callstack_kernel",
-               .save_func_name = "save_stack_trace",
-               .save_func      = NULL,
-       },
-       {
-               .name           = "callstack_user",
-               .save_func_name = "save_stack_trace_user",
-               .save_func      = NULL,
-       },
-};
-
-static
-int init_type(enum lttng_cs_ctx_modes mode)
-{
-       unsigned long func;
-
-       if (cs_types[mode].save_func)
-               return 0;
-       func = kallsyms_lookup_funcptr(cs_types[mode].save_func_name);
-       if (!func) {
-               printk(KERN_WARNING "LTTng: symbol lookup failed: %s\n",
-                               cs_types[mode].save_func_name);
-               return -EINVAL;
-       }
-       cs_types[mode].save_func = (void *) func;
-       return 0;
-}
-
-/* Keep track of nesting inside userspace callstack context code */
-DEFINE_PER_CPU(int, callstack_user_nesting);
-
-static
-struct stack_trace *stack_trace_context(struct lttng_ctx_field *field,
-                                       struct lib_ring_buffer_ctx *ctx)
-{
-       int buffer_nesting, cs_user_nesting;
-       struct lttng_cs *cs;
-       struct field_data *fdata = field->priv;
-
-       /*
-        * Do not gather the userspace callstack context when the event was
-        * triggered by the userspace callstack context saving mechanism.
-        */
-       cs_user_nesting = per_cpu(callstack_user_nesting, ctx->cpu);
-
-       if (fdata->mode == CALLSTACK_USER && cs_user_nesting >= 1)
-               return NULL;
-
-       /*
-        * get_cpu() is not required, preemption is already
-        * disabled while event is written.
-        *
-        * max nesting is checked in lib_ring_buffer_get_cpu().
-        * Check it again as a safety net.
-        */
-       cs = per_cpu_ptr(fdata->cs_percpu, ctx->cpu);
-       buffer_nesting = per_cpu(lib_ring_buffer_nesting, ctx->cpu) - 1;
-       if (buffer_nesting >= RING_BUFFER_MAX_NESTING)
-               return NULL;
-
-       return &cs->dispatch[buffer_nesting].stack_trace;
-}
-
-/*
- * In order to reserve the correct size, the callstack is computed. The
- * resulting callstack is saved to be accessed in the record step.
- */
-static
-size_t lttng_callstack_get_size(size_t offset, struct lttng_ctx_field *field,
-                               struct lib_ring_buffer_ctx *ctx,
-                               struct lttng_channel *chan)
-{
-       struct stack_trace *trace;
-       struct field_data *fdata = field->priv;
-       size_t orig_offset = offset;
-
-       /* do not write data if no space is available */
-       trace = stack_trace_context(field, ctx);
-       if (unlikely(!trace)) {
-               offset += lib_ring_buffer_align(offset, lttng_alignof(unsigned int));
-               offset += sizeof(unsigned int);
-               offset += lib_ring_buffer_align(offset, lttng_alignof(unsigned long));
-               return offset - orig_offset;
-       }
-
-       /* reset stack trace, no need to clear memory */
-       trace->nr_entries = 0;
-
-       if (fdata->mode == CALLSTACK_USER)
-               ++per_cpu(callstack_user_nesting, ctx->cpu);
-
-       /* do the real work and reserve space */
-       cs_types[fdata->mode].save_func(trace);
-
-       if (fdata->mode == CALLSTACK_USER)
-               per_cpu(callstack_user_nesting, ctx->cpu)--;
-
-       /*
-        * Remove final ULONG_MAX delimiter. If we cannot find it, add
-        * our own marker to show that the stack is incomplete. This is
-        * more compact for a trace.
-        */
-       if (trace->nr_entries > 0
-                       && trace->entries[trace->nr_entries - 1] == ULONG_MAX) {
-               trace->nr_entries--;
-       }
-       offset += lib_ring_buffer_align(offset, lttng_alignof(unsigned int));
-       offset += sizeof(unsigned int);
-       offset += lib_ring_buffer_align(offset, lttng_alignof(unsigned long));
-       offset += sizeof(unsigned long) * trace->nr_entries;
-       /* Add our own ULONG_MAX delimiter to show incomplete stack. */
-       if (trace->nr_entries == trace->max_entries)
-               offset += sizeof(unsigned long);
-       return offset - orig_offset;
-}
-
-static
-void lttng_callstack_record(struct lttng_ctx_field *field,
-                       struct lib_ring_buffer_ctx *ctx,
-                       struct lttng_channel *chan)
-{
-       struct stack_trace *trace = stack_trace_context(field, ctx);
-       unsigned int nr_seq_entries;
-
-       if (unlikely(!trace)) {
-               nr_seq_entries = 0;
-               lib_ring_buffer_align_ctx(ctx, lttng_alignof(unsigned int));
-               chan->ops->event_write(ctx, &nr_seq_entries, sizeof(unsigned int));
-               lib_ring_buffer_align_ctx(ctx, lttng_alignof(unsigned long));
-               return;
-       }
-       lib_ring_buffer_align_ctx(ctx, lttng_alignof(unsigned int));
-       nr_seq_entries = trace->nr_entries;
-       if (trace->nr_entries == trace->max_entries)
-               nr_seq_entries++;
-       chan->ops->event_write(ctx, &nr_seq_entries, sizeof(unsigned int));
-       lib_ring_buffer_align_ctx(ctx, lttng_alignof(unsigned long));
-       chan->ops->event_write(ctx, trace->entries,
-                       sizeof(unsigned long) * trace->nr_entries);
-       /* Add our own ULONG_MAX delimiter to show incomplete stack. */
-       if (trace->nr_entries == trace->max_entries) {
-               unsigned long delim = ULONG_MAX;
-
-               chan->ops->event_write(ctx, &delim, sizeof(unsigned long));
-       }
-}
+#include "lttng-context-callstack-legacy-impl.h"
 
 static
 void field_data_free(struct field_data *fdata)
This page took 0.029517 seconds and 4 git commands to generate.