This is the mail archive of the systemtap@sourceware.org mailing list for the systemtap project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

System call instrumentation


Hi Ingo,

I looked at the system call instrumentation present in LTTng lately. I
tried different solutions, e.g. hooking a kernel-wide syscall trace in
do_syscall_trace, but it appears that I ended up re-doing another
syscall table, which consists of specialized functions which extracts
the string and data structure parameters from user-space. Since code
duplication is not exactly wanted, I think that the original approach
taken in my patchset, which is to instrument the kernel code at the
sys_* level (e.g. sys_open), which is the earliest level where the
parameter information is made available to the kernel, is still the best
way to go.

I would still identify the execution mode changes in the same way I do
currently, which is by instrumenting do_syscall_trace, just to know as
soon as possible when the mode has changed from user-space to
kernel-space so we can do time accounting more accurately. I already
have the patchset which adds the KERNEL_TRACE thread flag to every
architectures. It's tested in assembly in the same way SYSCALL_TRACE is
tested, but is activated globally by iterating on all the threads.

So, the currently proposed scheme for a system call would be (for the
open() example)

shown as : 
kernel stack
  trace: event name (parameters)


do_syscall_trace()
  trace: kernel_arch_syscall_entry (syscall id, instruction pointer)

do_sys_open()
  trace: fs_open (fd, filename)

do_syscall_trace()
  kernel_arch_syscall_exit (return value)

If we take this open() example, filename is ready only in do_sys_open,
which is called by sys_open and sys_openat. So the logical
instrumentation site for this would really be do_sys_open(). The
information about which system call has been done is made available in
the kernel_arch_syscall_entry event. It is not present anymore at the
do_sys_open level because this execution path can be called from more
than one syscall.

What do you think of this approach ?


Mathieu

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]