Re: [PATCH 2/5] ftrace: use code patching for ftrace graph tracer

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 26 Nov 2008 11:46:58 -0500 (EST) Steven Rostedt <rostedt@goodmis.org> wrote:

> 
> 
> On Tue, 25 Nov 2008, Andrew Morton wrote:
> 
> > On Wed, 26 Nov 2008 00:16:24 -0500 Steven Rostedt <rostedt@goodmis.org> wrote:
> > 
> > > From: Steven Rostedt <rostedt@goodmis.org>
> > > 
> > > Impact: more efficient code for ftrace graph tracer
> > > 
> > > This patch uses the dynamic patching, when available, to patch
> > > the function graph code into the kernel.
> > > 
> > > This patch will ease the way for letting both function tracing
> > > and function graph tracing run together.
> > > 
> > > ...
> > >
> > > +static int ftrace_mod_jmp(unsigned long ip,
> > > +			  int old_offset, int new_offset)
> > > +{
> > > +	unsigned char code[MCOUNT_INSN_SIZE];
> > > +
> > > +	if (probe_kernel_read(code, (void *)ip, MCOUNT_INSN_SIZE))
> > > +		return -EFAULT;
> > > +
> > > +	if (code[0] != 0xe9 || old_offset != *(int *)(&code[1]))
> > 
> > erk.  I suspect that there's a nicer way of doing this amongst our
> > forest of get_unaligned_foo() interfaces.  Harvey will know.
> 
> Hmm, I may be able to make a struct out of code.
> 
>   struct {
> 	unsigned char op;
> 	unsigned int  offset;
>   } code __attribute__((packed));
> 
> Would that look better?

nah, let's do something more generic for this.

> > 
> > > +		return -EINVAL;
> > > +
> > > +	*(int *)(&code[1]) = new_offset;
> > 
> > Might be able to use put_unaligned_foo() here.
> > 
> > The problem is that these functions use sizeof(*ptr) to work out what
> > to do, so a cast is still needed.  A get_unaligned32(ptr) would be
> > nice.  One which takes a void* and assumes CPU ordering.
> 
> Is there a correctness concern here? This is arch specific code, so I'm
> not worried about other archs.

No, the code is OK as-is.

It's just that "read a word from an [maybe-]unaligned address" is such
a common operation that there should be a nice clean simple function to
do it, rather than doing open-coded (and different) weird C tricks at each
and\ every site.