[PATCH v4 15/16] block-dma: properly take MMIO path

[Jason Gunthorpe]

On Thu, Aug 28, 2025 at 11:15:20AM -0600, Keith Busch wrote:
> On Thu, Aug 28, 2025 at 07:54:27PM +0300, Leon Romanovsky wrote:
> > On Thu, Aug 28, 2025 at 09:19:20AM -0600, Keith Busch wrote:
> > > On Tue, Aug 19, 2025 at 08:36:59PM +0300, Leon Romanovsky wrote:
> > > > diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
> > > > index 09b99d52fd36..283058bcb5b1 100644
> > > > --- a/include/linux/blk_types.h
> > > > +++ b/include/linux/blk_types.h
> > > > @@ -387,6 +387,7 @@ enum req_flag_bits {
> > > >  	__REQ_FS_PRIVATE,	/* for file system (submitter) use */
> > > >  	__REQ_ATOMIC,		/* for atomic write operations */
> > > >  	__REQ_P2PDMA,		/* contains P2P DMA pages */
> > > > +	__REQ_MMIO,		/* contains MMIO memory */
> > > >  	/*
> > > >  	 * Command specific flags, keep last:
> > > >  	 */
> > > > @@ -420,6 +421,7 @@ enum req_flag_bits {
> > > >  #define REQ_FS_PRIVATE	(__force blk_opf_t)(1ULL << __REQ_FS_PRIVATE)
> > > >  #define REQ_ATOMIC	(__force blk_opf_t)(1ULL << __REQ_ATOMIC)
> > > >  #define REQ_P2PDMA	(__force blk_opf_t)(1ULL << __REQ_P2PDMA)
> > > > +#define REQ_MMIO	(__force blk_opf_t)(1ULL << __REQ_MMIO)
> > > 
> > > Now that my integrity metadata DMA series is staged, I don't think we
> > > can use REQ flags like this because data and metadata may have different
> > > mapping types. I think we should add a flags field to the dma_iova_state
> > > instead.
> > 
> > Before integrity metadata code was merged, the assumption was that request is
> > only one type or p2p or host. Is it still holding now?
> 
> I don't think that was ever the case. Metadata is allocated
> independently of the data payload, usually by the kernel in
> bio_integrity_prep() just before dispatching the request. The bio may
> have a p2p data payload, but the integrity metadata is just a kmalloc
> buf in that path.

Then you should do two dma mapping operations today, that is how the
API was built. You shouldn't mix P2P and non P2P within a single
operation right now..

Jason