Re: [PATCH v7 0/4] trace: Introduce objtrace trigger to trace the kernel object

[Masami Hiramatsu]

Hi Jeff,

On Wed, 19 Jan 2022 10:32:49 +0800
Jeff Xie <xiehuan09@xxxxxxxxx> wrote:

> > > Note: when change to use the ftrace_test_recursion_trylock, all the functions
> > > will call the copy_from_kernel_nofault, I don't know where this is the problem now,
> > > maybe should fall back to the usage in v6.
> > >
> > > for example:
> > >
> > > cat-118     [000] ...1.     1.458998: __bio_add_page <-bio_add_page object:0xffff88811a12e9e8 value:0x0
> > > cat-118     [000] ...2.     1.458998: copy_from_kernel_nofault <-trace_object_events_call object:0xffff88811a12e9e8 value:0x1000
> > > cat-118     [000] ...2.     1.458998: copy_from_kernel_nofault_allowed <-copy_from_kernel_nofault object:0xffff88811a12e9e8 value:0x1000
> > > cat-118     [000] ...1.     1.458998: __rcu_read_lock <-xa_load object:0xffff88811a12e9e8 value:0x1000
> > > cat-118     [000] ...2.     1.458998: copy_from_kernel_nofault <-trace_object_events_call object:0xffff88811a12e9e8 value:0x1000
> > > cat-118     [000] ...2.     1.458998: copy_from_kernel_nofault_allowed <-copy_from_kernel_nofault object:0xffff88811a12e9e8 value:0x1000
> > > cat-118     [000] ...1.     1.458998: __rcu_read_unlock <-xa_load object:0xffff88811a12e9e8 value:0x1000
> > > cat-118     [000] ...3.     1.458998: copy_from_kernel_nofault <-trace_object_events_call object:0xffff88811a12e9e8 value:0x1000
> > > cat-118     [000] ...3.     1.458998: copy_from_kernel_nofault_allowed <-copy_from_kernel_nofault object:0xffff88811a12e9e8 value:0x1000
> > > ....
> >
> > Hmm, this is strange, but I got it is the expected behavior, since the
> > ftrace_test_recursion_trylock() accepts one stage recursion for the
> > first event in the interrupt as transition event.

I think you should revert that change and back to your own per-cpu recursion
flag instead of using ftrace_test_recursion_trylock().

Thank you,

-- 
Masami Hiramatsu <mhiramat@xxxxxxxxxx>