Re: sound/pci/ctxfi/ctdaio.c: duplicate function removal question

[David Laight]

On Sun, 29 Dec 2024 13:03:56 +0100
Takashi Iwai <tiwai@xxxxxxx> wrote:

> On Sun, 29 Dec 2024 11:57:22 +0100,
> David Laight wrote:
> > 
> > On Sun, 29 Dec 2024 10:10:09 +0100
> > Takashi Iwai <tiwai@xxxxxxx> wrote:
> >   
> > > On Sun, 29 Dec 2024 10:02:28 +0100,
> > > David Laight wrote:  
> > > > 
> > > > On Tue, 24 Dec 2024 23:16:17 +0000
> > > > Ethan Carter Edwards <ethan@xxxxxxxxxxxxxxxxx> wrote:
> > > >     
> > > > > Hello all,
> > > > > 
> > > > > First of all, happy holidays. 
> > > > > 
> > > > > I was browsing the ctdaio.c code and I noticed a lot of 
> > > > > duplicate code and functions, specifically:
> > > > > 
> > > > > dao_set_{right,left}_input and
> > > > > dao_clear_{right,left}_input functions.
> > > > > 
> > > > > The functions are pretty much identical. They only 
> > > > > differ in the side (left, right). What was the original
> > > > > idea in doing this? Wouldn't it make more since to just
> > > > > have an ENUM (left, right) as an argument that would 
> > > > > determine the side and just reduce the function to 
> > > > > dao_set_input and dao_clear_input.    
> > > > 
> > > > Hmmm... you'd have a lot of conditionals inside the function.
> > > > 
> > > > They also look like a memory leak just waiting to happen.
> > > > I guess that an earlier implementation used a separate kmalloc()
> > > > for each imappers[].
> > > > 
> > > > Why is imappers[] an array of pointers not an array of the items?
> > > > Each is just 8 bytes plus a 'list_head' (2 pointers?).    
> > > 
> > > AFAIUC, it's a setup of a chained element, so no leak there as of
> > > now.  
> > 
> > Unless someone calls the functions in the wrong order.
> > And that seems to be outside the control of this code.  
> 
> Could you elaborate?  The chain element gets released also by the
> destructor, dao_rsc_uninit().  And this is no exported function to be
> used by other drivers, but an internal one only for ctxfi driver.

The code is just too horrid :-)
I missed that when dao_set_right_input() calls dao->ops->clear_right_input()
it is just calling the static function just below.

I think the two 'clear' functions could be combined as:

static int dao_clear_input(struct dao *dao, unsigned int start, unsigned int end)
{
	struct imapper *to_free = dao->imappers[start];
	unsigned int i;

	if (!to_free)
		return 0;
	for (i = start; i < end; i++) {
		dao->mgr->imap_delete(dao->mgr, dao->imappers[i]);
		dao->imappers[i] = NULL;
	}

	kfree(to_free);
	return 0;
}

static int dao_clear_left_input(struct dao *dao)
{
	return dao_clear_input(dao, 0, dao->rscl.msr);
}

static int dao_clear_right_input(struct dao *dao)
{
	return dao_clear_input(dao, dao->rscl.msr, dao->rscl.msr + dao->rscr.msr);
}

Although I'm missing any other code that actually indexes that imappers[] array.
So I suspect only [0] and [dao->rscl.msr] are ever used??

Stanger things happen in the set functions.

	David

> 
> 
> thanks,
> 
> Takashi