Re: [PATCH v8 16/23] dma-buf: Let importers ask how peer-to-peer traffic is routed

["Christian König" <christian.koenig@amd.com>]

On 9/28/26 13:19, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@nvidia.com>
> 
> Exporters keep the &struct p2pdma_provider backing a buffer in their own
> private data. An importer cannot reach it, so it has no way to learn how
> its own peer-to-peer traffic would be routed before it programs its
> hardware.
> 
> Add an optional @p2pdma_provider callback for an exporter to hand that
> provider out, and dma_buf_p2pdma_map_type() for an importer to ask by TLP
> class. Exporters keep the provider where it already lives, so this adds an
> operation rather than changing any existing signature or structure.
> 
> Tested-by: Tushar Dave <tdave@nvidia.com>
> Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> ---
>  drivers/dma-buf/dma-buf-mapping.c | 39 +++++++++++++++++++++++++++++++++++++++
>  include/linux/dma-buf-mapping.h   |  3 +++
>  include/linux/dma-buf.h           | 19 +++++++++++++++++++
>  3 files changed, 61 insertions(+)
> 
> diff --git a/drivers/dma-buf/dma-buf-mapping.c b/drivers/dma-buf/dma-buf-mapping.c
> index 794acff2546a..41702045f803 100644
> --- a/drivers/dma-buf/dma-buf-mapping.c
> +++ b/drivers/dma-buf/dma-buf-mapping.c
> @@ -6,6 +6,45 @@
>  #include <linux/dma-buf-mapping.h>
>  #include <linux/dma-resv.h>
>  
> +/**
> + * dma_buf_p2pdma_map_type - How peer-to-peer traffic to a buffer is routed
> + * @attach:	attachment of the importer that will issue the traffic
> + * @tlp_flags:	&enum pci_p2pdma_tlp_flags describing the TLPs it will issue
> + *
> + * Reports how the PCIe fabric routes @tlp_flags traffic between the buffer
> + * behind @attach and the importer attached to it, so that an importer can
> + * choose the TLP attributes that earn it a direct route before it programs
> + * its hardware.
> + *
> + * The caller must hold the reservation lock. The provider is owned by the
> + * exporter's driver binding, and exporters revoke under that lock before
> + * letting an unbind proceed, so dropping it any earlier would leave this
> + * racing a teardown that frees the provider.
> + *
> + * Return: the mapping type for @tlp_flags traffic, or PCI_P2PDMA_MAP_NONE
> + * when the exporter names no &struct p2pdma_provider and nothing is known
> + * about the route.
> + */
> +enum pci_p2pdma_map_type
> +dma_buf_p2pdma_map_type(struct dma_buf_attachment *attach,
> +			unsigned int tlp_flags)
> +{
> +	struct dma_buf *dmabuf = attach->dmabuf;
> +	struct p2pdma_provider *provider;
> +
> +	dma_resv_assert_held(dmabuf->resv);
> +
> +	if (!dmabuf->ops->p2pdma_provider)
> +		return PCI_P2PDMA_MAP_NONE;
> +
> +	provider = dmabuf->ops->p2pdma_provider(dmabuf);
> +	if (!provider)
> +		return PCI_P2PDMA_MAP_NONE;
> +


> +	return pci_p2pdma_map_type_tlp(provider, attach->dev, tlp_flags);

This function call here *must* be in the exporter and not the DMA-buf framework.

So clear NAK to putting that here.

Regards,
Christian.

> +}
> +EXPORT_SYMBOL_NS_GPL(dma_buf_p2pdma_map_type, "DMA_BUF");
> +
>  static struct scatterlist *fill_sg_entry(struct scatterlist *sgl, size_t length,
>  					 dma_addr_t addr)
>  {
> diff --git a/include/linux/dma-buf-mapping.h b/include/linux/dma-buf-mapping.h
> index 09bde3f748e4..37b3da10b17a 100644
> --- a/include/linux/dma-buf-mapping.h
> +++ b/include/linux/dma-buf-mapping.h
> @@ -7,6 +7,9 @@
>  #define __DMA_BUF_MAPPING_H__
>  #include <linux/dma-buf.h>
>  
> +enum pci_p2pdma_map_type
> +dma_buf_p2pdma_map_type(struct dma_buf_attachment *attach,
> +			unsigned int tlp_flags);
>  struct sg_table *dma_buf_phys_vec_to_sgt(struct dma_buf_attachment *attach,
>  					 struct p2pdma_provider *provider,
>  					 struct phys_vec *phys_vec,
> diff --git a/include/linux/dma-buf.h b/include/linux/dma-buf.h
> index d1203da56fc5..d56a48d41215 100644
> --- a/include/linux/dma-buf.h
> +++ b/include/linux/dma-buf.h
> @@ -186,6 +186,25 @@ struct dma_buf_ops {
>  	 * if the call would block.
>  	 */
>  
> +	/**
> +	 * @p2pdma_provider:
> +	 *
> +	 * Returns the &struct p2pdma_provider backing this buffer, so that an
> +	 * importer can ask how its peer-to-peer traffic would be routed before
> +	 * it programs its hardware. Importers reach this through
> +	 * dma_buf_p2pdma_map_type() rather than calling it directly.
> +	 *
> +	 * Exporters of MMIO memory that is reachable peer-to-peer should
> +	 * implement this. Called with the &dma_buf.resv reservation lock held,
> +	 * which is what lets an exporter refuse once it has revoked the buffer
> +	 * and started tearing the provider down. This callback is optional.
> +	 *
> +	 * Returns:
> +	 *
> +	 * The provider backing the buffer.
> +	 */
> +	struct p2pdma_provider *(*p2pdma_provider)(struct dma_buf *dmabuf);
> +
>  	/**
>  	 * @release:
>  	 *
>