linux/kernel/trace/trace_uprobe.c, branch v3.14-rc7

tracing/probes: Fix build break on !CONFIG_KPROBE_EVENT

2014-01-03T20:27:18Z

When kprobe-based dynamic event tracer is not enabled, it caused following build error: kernel/built-in.o: In function `traceprobe_update_arg': (.text+0x10c8dd): undefined reference to `fetch_symbol_u8' kernel/built-in.o: In function `traceprobe_update_arg': (.text+0x10c8e9): undefined reference to `fetch_symbol_u16' kernel/built-in.o: In function `traceprobe_update_arg': (.text+0x10c8f5): undefined reference to `fetch_symbol_u32' kernel/built-in.o: In function `traceprobe_update_arg': (.text+0x10c901): undefined reference to `fetch_symbol_u64' kernel/built-in.o: In function `traceprobe_update_arg': (.text+0x10c909): undefined reference to `fetch_symbol_string' kernel/built-in.o: In function `traceprobe_update_arg': (.text+0x10c913): undefined reference to `fetch_symbol_string_size' ... It was due to the fetch methods are referred from CHECK_FETCH_FUNCS macro and since it was only defined in trace_kprobe.c. Move NULL definition of such fetch functions to the header file. Note, it also requires CONFIG_BRANCH_PROFILING enabled to trigger this failure as well. This is because the "fetch_symbol_*" variables are referenced in a "else if" statement that will only call update_symbol_cache(), which is a static inline stub function when CONFIG_KPROBE_EVENT is not enabled. gcc is smart enough to optimize this "else if" out and that also removes the code that references the undefined variables. But when BRANCH_PROFILING is enabled, it fools gcc into keeping the if statement around and thus references the undefined symbols and fails to build. Reported-by: kbuild test robot Signed-off-by: Namhyung Kim Signed-off-by: Steven Rostedt

tracing/uprobes: Add @+file_offset fetch method

2014-01-03T01:57:05Z

Enable to fetch data from a file offset. Currently it only supports fetching from same binary uprobe set. It'll translate the file offset to a proper virtual address in the process. The syntax is "@+OFFSET" as it does similar to normal memory fetching (@ADDR) which does no address translation. Suggested-by: Oleg Nesterov Acked-by: Masami Hiramatsu Acked-by: Oleg Nesterov Cc: Srikar Dronamraju Cc: zhangwei(Jovi) Cc: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim

tracing/uprobes: Fetch args before reserving a ring buffer

2014-01-02T21:17:44Z

Fetching from user space should be done in a non-atomic context. So use a per-cpu buffer and copy its content to the ring buffer atomically. Note that we can migrate during accessing user memory thus use a per-cpu mutex to protect concurrent accesses. This is needed since we'll be able to fetch args from an user memory which can be swapped out. Before that uprobes could fetch args from registers only which saved in a kernel space. While at it, use __get_data_size() and store_trace_args() to reduce code duplication. And add struct uprobe_cpu_buffer and its helpers as suggested by Oleg. Reviewed-by: Masami Hiramatsu Acked-by: Oleg Nesterov Cc: Srikar Dronamraju Cc: zhangwei(Jovi) Cc: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim

tracing/uprobes: Pass 'is_return' to traceprobe_parse_probe_arg()

2014-01-02T21:17:43Z

Currently uprobes don't pass is_return to the argument parser so that it cannot make use of "$retval" fetch method since it only works for return probes. Reviewed-by: Masami Hiramatsu Acked-by: Oleg Nesterov Cc: Srikar Dronamraju Cc: zhangwei(Jovi) Cc: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim

tracing/probes: Implement 'memory' fetch method for uprobes

2014-01-02T21:17:43Z

Use separate method to fetch from memory. Move existing functions to trace_kprobe.c and make them static. Also add new memory fetch implementation for uprobes. Acked-by: Masami Hiramatsu Acked-by: Oleg Nesterov Cc: Srikar Dronamraju Cc: zhangwei(Jovi) Cc: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim

tracing/probes: Move 'symbol' fetch method to kprobes

2014-01-02T21:17:41Z

Move existing functions to trace_kprobe.c and add NULL entries to the uprobes fetch type table. I don't make them static since some generic routines like update/free_XXX_fetch_param() require pointers to the functions. Acked-by: Oleg Nesterov Cc: Masami Hiramatsu Cc: Srikar Dronamraju Cc: zhangwei(Jovi) Cc: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim

tracing/probes: Implement 'stack' fetch method for uprobes

2014-01-02T21:17:40Z

Use separate method to fetch from stack. Move existing functions to trace_kprobe.c and make them static. Also add new stack fetch implementation for uprobes. Acked-by: Oleg Nesterov Cc: Masami Hiramatsu Cc: Srikar Dronamraju Cc: zhangwei(Jovi) Cc: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim

tracing/probes: Split [ku]probes_fetch_type_table

2014-01-02T21:17:39Z

Use separate fetch_type_table for kprobes and uprobes. It currently shares all fetch methods but some of them will be implemented differently later. This is not to break build if [ku]probes is configured alone (like !CONFIG_KPROBE_EVENT and CONFIG_UPROBE_EVENT). So I added '__weak' to the table declaration so that it can be safely omitted when it configured out. Acked-by: Oleg Nesterov Acked-by: Masami Hiramatsu Cc: Srikar Dronamraju Cc: zhangwei(Jovi) Cc: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim

tracing/probes: Integrate duplicate set_print_fmt()

2014-01-02T21:17:38Z

The set_print_fmt() functions are implemented almost same for [ku]probes. Move it to a common place and get rid of the duplication. Acked-by: Masami Hiramatsu Acked-by: Oleg Nesterov Cc: Srikar Dronamraju Cc: zhangwei(Jovi) Cc: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim

tracing/uprobes: Convert to struct trace_probe

2014-01-02T21:17:36Z

Convert struct trace_uprobe to make use of the common trace_probe structure. Reviewed-by: Masami Hiramatsu Acked-by: Srikar Dronamraju Acked-by: Oleg Nesterov Cc: zhangwei(Jovi) Cc: Arnaldo Carvalho de Melo Signed-off-by: Namhyung Kim