Re: [RFC 1/1] swiotlb: Reduce calls to swiotlb_find_pool()

[Petr Tesařík]

V Fri, 28 Jun 2024 08:01:29 +0200
"hch@xxxxxx" <hch@xxxxxx> napsáno:

> On Thu, Jun 27, 2024 at 04:02:59PM +0000, Michael Kelley wrote:
> > > > Conceptually, it's still being used as a boolean function based on
> > > > whether the return value is NULL.  Renaming it to swiotlb_get_pool()
> > > > more accurately describes the return value, but obscures the
> > > > intent of determining if it is a swiotlb buffer.  I'll think about it.
> > > > Suggestions are welcome.  
> > > 
> > > Just keep is_swiotlb_buffer as a trivial inline helper that returns
> > > bool.  
> > 
> > I don't understand what you are suggesting.  Could you elaborate a bit?
> > is_swiotlb_buffer() can't be trivial when CONFIG_SWIOTLB_DYNAMIC
> > is set.  
> 
> Call the main function that finds and retuns the pool swiotlb_find_pool,
> and then have a is_swiotlb_buffer wrapper that just returns bool.
> 

I see. That's not my point. After applying Michael's patch, the return
value is always used, except here:

bool dma_direct_need_sync(struct device *dev, dma_addr_t dma_addr)
{
	return !dev_is_dma_coherent(dev) ||
	       is_swiotlb_buffer(dev, dma_to_phys(dev, dma_addr));
}

I don't think this one occurrence in the entire source tree is worth a
separate inline function.

If nobody has a better idea, I'm not really offended by keeping the
original name, is_swiotlb_buffer(). It would just become the only
function which starts with "is_" and provides more information in the
return value than a simple yes/no, and I thought there must be an
unwritten convention about that.

Petr T