Re: [PATCH v8 01/23] dma-direct: return struct page from dma_direct_alloc_from_pool()

On Thu, Jul 23, 2026 at 10:57:04AM +0300, Leon Romanovsky wrote:
> On Wed, Jul 22, 2026 at 04:59:12PM -0300, Jason Gunthorpe wrote:
> > On Tue, Jul 21, 2026 at 06:33:21PM +0300, Leon Romanovsky wrote:
> > 
> > > Something like, but probably void* needs to be phys_addr_t:
> > > 
> > > static inline struct page *dma_phys_to_page(void *addr)
> > > {
> > > #ifdef CONFIG_DMA_DIRECT_REMAP
> > > 	return vmalloc_to_page(addr);
> > > #else
> > > 	return virt_to_page(addr);
> > > #endif
> > > }
> > 
> > I don't like this, we don't know for sure the addr will be in a vmap
> > and this will fail if it isn't.
> 
> Of course we know. The existing "#ifdef CONFIG_DMA_DIRECT_REMAP" is
> relevant for addresses acquired from pool.

Yeah, but I still don't like it :) It is hard to follow if you make
those kinds of leaps, someone will call this new helper on something
they shouldn't

> dma_phys_to_page() is a bad name for some low-level conversion function.
> It needs to be internal to DMA logic, in the level when we convert from
> phys to page.

I think we should not convert from phys to page, that's also easy to
do wrong

Jason