Re: [PATCH 1/3] i3c: mipi-i3c-hci: Make bounce buffer code generic to all DMA transfers

[Frank Li <Frank.li@nxp.com>]

On Thu, Jun 19, 2025 at 04:37:30PM +0300, Jarkko Nikula wrote:
> Hi
>
> On 6/17/25 6:09 PM, Frank Li wrote:
> > > Thanks, I finially understand the problem you faced. I think other master
> > > controller face similar issue if they use dma. Let me think more. If alloc
> > > buffer size is not align to cache line, swtlib will bounce again.
> > >
> > > rough idea i3c core layer provide an i3c_(un)map_api to do that.
> >
> > struct i3c_dma {
> > 	void *bounce;
> > 	dma_addr_t addr;
> > 	enum dma_data_direction;
> > 	void *data;
> > 	size_t data_size;
> > 	size_t bounce_size;
> > }
> >
> > struct i3c_dma *i3c_dma_map_single(struct device *dev, void *data, size_t sz, enum dma_data_direction dir)
> > {
> > 	struct i3c_dma *p
> >
> > 	p = kmalloc(sizeof(*p));
> > 	if (!p)
> > 		return p;
> >
> > 	if (is_vmalloc_addr(data)) {
> > 		p->bounce = kmalloc(align_up(sz, CACHE_LINE_SIZE)); //avoid swtlib bounce again.
>
> I'll try to look at your idea next week. Do you have pointers to the above
> allocation size rounding is it required?
>
> I've understood from the Documentation/core-api/dma-api-howto.rst
> "Architectures must ensure that kmalloc'ed buffer is DMA-safe" and
> "ARCH_DMA_MINALIGN must be set so that the memory allocator makes sure that
> kmalloc'ed buffer doesn't share a cache line with the others.".

But sz in dma_map_single() will be checked by swtlib, if size is not align
with CACHE_LINE, it will bounce it.

Frank

>
> --
> linux-i3c mailing list
> linux-i3c@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-i3c

-- 
linux-i3c mailing list
linux-i3c@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-i3c