Re: Warning from ring buffer code (Was: Re: linux-next: tip treebuild warning)

[Frederic Weisbecker]

On Mon, Sep 14, 2009 at 02:17:18PM -0400, Steven Rostedt wrote:
> On Mon, 2009-09-14 at 10:09 -0700, Christopher Li wrote:
> > On Mon, Sep 14, 2009 at 8:16 AM, Steven Rostedt <srostedt@xxxxxxxxxx> wrote:
> > > static void func(int size_me) {
> > >        char array[size_me];
> > >
> > >        memcpy(array, "hello", size);
> > > };
> > >
> > > and sparse failed on it as well. Note, you need to have something call
> > > func, or sparse will ignore it.
> > 
> > Gcc allows variable size. Sparse expects the size of an array is constant.
> > For the kernel using variable array size is consider bad. Because the kernel
> > has very limited stack size. (8K if I remember correctly). Using dynamic array
> > is very easy to overflow the stack without realizing it.
> > 
> > It deserves a warning. I agree the warning message can use a better description
> > though.
> 
> Good point!
> 
> I've added Frederic to the Cc list, since he wrote the code.
> 
> Frederic, how big can one of those events get. The ring buffer (and
> TRACE_EVENT) allow up to almost a page size, which is very hefty for the
> stack. This code needs to either be rewritten or we need to set a limit
> to the size of a profile entry.
> 
> We could add:
> 
> 	if (__entry_size > 256)
> 		return;
> 
> Thoughts?
> 


Well it can be big, especially once we play with array fields or
__string().

I can manage the __string() that said, by only copying their
pointer and later delay the copy.

Well actually I would like to rewrite all that entirely to avoid
any stack allocation, especially for arrays and string.

Lemme think about a CPP magic way to directly interact with perf
buffer. I think it's possible.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/