All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
To: "Leon Romanovsky" <leon@kernel.org>,
	"Bjorn Helgaas" <bhelgaas@google.com>,
	"Logan Gunthorpe" <logang@deltatee.com>,
	"Jason Gunthorpe" <jgg@ziepe.ca>,
	"Joerg Roedel (AMD)" <joro@8bytes.org>,
	"Will Deacon" <will@kernel.org>,
	"Robin Murphy" <robin.murphy@arm.com>,
	"Christian König" <christian.koenig@amd.com>
Cc: linux-pci@vger.kernel.org, linux-kernel@vger.kernel.org,
	 linux-doc@vger.kernel.org, iommu@lists.linux.dev,
	Tushar Dave <tdave@nvidia.com>,
		linux-media@vger.kernel.org, dri-devel@lists.freedesktop.org,
		linaro-mm-sig@lists.linaro.org, linux-rdma@vger.kernel.org,
	kvm@vger.kernel.org,  Chaitanya Kulkarni	 <kch@nvidia.com>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Jens Axboe	 <axboe@kernel.dk>, Alex Williamson <alex@shazbot.org>,
	Ankit Agrawal <ankita@nvidia.com>,
	Jonathan Corbet <corbet@lwn.net>,
	Shuah Khan	 <skhan@linuxfoundation.org>,
	Randy Dunlap <rdunlap@infradead.org>,
	Sumit Semwal	 <sumit.semwal@linaro.org>
Subject: Re: [PATCH v8 22/23] PCI/P2PDMA: Evaluate the ATS path for clients with ATS enabled
Date: Mon, 28 Sep 2026 14:43:06 +0200	[thread overview]
Message-ID: <e61fcc1552447e628b7ec938ac7e4e36555ebec2.camel@linux.intel.com> (raw)
In-Reply-To: <20260928-fix-p2p-acs-v4-0-v8-22-404453b9c435@nvidia.com>

On Mon, 2026-09-28 at 14:19 +0300, Leon Romanovsky wrote:
> From: Leon Romanovsky <leonro@nvidia.com>
> 
> P2PDMA assumes every client issues Untranslated Requests. A device
> with
> ATS enabled may translate any address it is handed, including a bus
> address, and Translation Blocking can reject its Translated Requests
> on
> a route the Untranslated answer called usable.
> 
> Unless the client declared per-mapping ATS, take the Address Type
> from
> its ATS Enable bit. Never hand such a client bus addresses: only an
> IOVA
> survives translation, and whatever it issues untranslated goes
> through
> the host bridge, whose route must therefore be usable. Apply this
> after
> the cache, which keeps holding answers that depend only on the
> topology.
> 
> Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

> ---
>  Documentation/driver-api/pci/p2pdma.rst |  8 ++++++
>  drivers/pci/p2pdma.c                    | 50
> +++++++++++++++++++++++++++++++--
>  include/linux/pci-p2pdma.h              |  3 +-
>  3 files changed, 58 insertions(+), 3 deletions(-)
> 
> diff --git a/Documentation/driver-api/pci/p2pdma.rst
> b/Documentation/driver-api/pci/p2pdma.rst
> index 7a7f29f455b0..bf9f96c07b58 100644
> --- a/Documentation/driver-api/pci/p2pdma.rst
> +++ b/Documentation/driver-api/pci/p2pdma.rst
> @@ -40,6 +40,14 @@ A client therefore describes its traffic with
> ``enum pci_p2pdma_tlp_flags``
>  and asks ``pci_p2pdma_map_type_tlp()``. ``pci_p2pdma_map_type()``
> answers for
>  the default: strictly ordered Requests carrying an Untranslated
> address.
>  
> +The PCIe ATS Enable bit covers the whole device, and most devices
> translate
> +any address they are handed once it is set. Unless a driver has
> declared with
> +``pcim_p2pdma_set_ats_per_mapping()`` that its device chooses ATS
> per mapping,
> +that bit rather than the flags decides the Address Type, and a
> client with ATS
> +enabled is never handed a bus address, which it would translate as
> though it
> +were an IOVA. Its Translated Requests may still route directly, but
> anything
> +it issues untranslated reaches the host bridge, so that route has to
> work too.
> +
>  The two directions are decided independently. Translation Blocking
> (TB),
>  Direct Translated P2P (DT), Request Redirect (RR) and Egress Control
> (EC) on
>  the client-side port decide the Request:
> diff --git a/drivers/pci/p2pdma.c b/drivers/pci/p2pdma.c
> index ae676509f9f3..ca79af2e64c7 100644
> --- a/drivers/pci/p2pdma.c
> +++ b/drivers/pci/p2pdma.c
> @@ -1595,6 +1595,41 @@ void pcim_p2pdma_set_ats_per_mapping(struct
> pci_dev *pdev)
>  }
>  EXPORT_SYMBOL_GPL(pcim_p2pdma_set_ats_per_mapping);
>  
> +static unsigned int pci_p2pdma_client_tlp_flags(struct pci_dev
> *client,
> +						unsigned int
> tlp_flags)
> +{
> +	if (client->ats_enabled)
> +		return tlp_flags | PCI_P2PDMA_TLP_TRANSLATED;
> +
> +	return tlp_flags & ~PCI_P2PDMA_TLP_TRANSLATED;
> +}
> +
> +/*
> + * A client that translates every address it is handed cannot be
> handed a bus
> + * address, which it would translate as though it were an IOVA. Its
> Translated
> + * Requests may still route directly, but a Request it issues
> without a
> + * translation carries the IOVA to the host bridge, so that route
> has to work
> + * as well.
> + */
> +static enum pci_p2pdma_map_type
> +pci_p2pdma_client_map_type(struct pci_dev *provider, struct pci_dev
> *client,
> +			   bool per_mapping, enum
> pci_p2pdma_map_type type)
> +{
> +	if (per_mapping || !client->ats_enabled ||
> +	    type != PCI_P2PDMA_MAP_BUS_ADDR)
> +		return type;
> +
> +	pci_dbg(client,
> +		"P2PDMA ACS: provider=%s bus address withheld; ATS
> is enabled for the whole client\n",
> +		 pci_name(provider));
> +
> +	if (cpu_supports_p2pdma() ||
> +	    host_bridge_whitelist(provider, client, false))
> +		return PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
> +
> +	return PCI_P2PDMA_MAP_NOT_SUPPORTED;
> +}
> +
>  /**
>   * pci_p2pdma_map_type_tlp - Determine the mapping type for P2PDMA
> transfers
>   * @provider: P2PDMA provider structure
> @@ -1609,6 +1644,11 @@
> EXPORT_SYMBOL_GPL(pcim_p2pdma_set_ats_per_mapping);
>   * ACS routes a peer-to-peer transaction by the attributes its TLPs
> carry, so
>   * the answer depends on @tlp_flags. A caller that passes flags its
> traffic
>   * does not match gets a mapping the fabric will not deliver.
> + *
> + * Only a client whose driver called
> pcim_p2pdma_set_ats_per_mapping() takes
> + * the Address Type from @tlp_flags. For any other client its ATS
> Enable bit
> + * decides, and a client that translates every address it is handed
> never gets
> + * %PCI_P2PDMA_MAP_BUS_ADDR.
>   */
>  enum pci_p2pdma_map_type
>  pci_p2pdma_map_type_tlp(struct p2pdma_provider *provider, struct
> device *dev,
> @@ -1620,6 +1660,7 @@ pci_p2pdma_map_type_tlp(struct p2pdma_provider
> *provider, struct device *dev,
>  	struct pci_p2pdma *p2pdma;
>  	struct pci_dev *client;
>  	bool provider_state;
> +	bool per_mapping;
>  	int dist;
>  
>  	if (!pdev->p2pdma) {
> @@ -1639,6 +1680,9 @@ pci_p2pdma_map_type_tlp(struct p2pdma_provider
> *provider, struct device *dev,
>  	cache_index = map_types_idx(client);
>  
>  	rcu_read_lock();
> +	/* The declaration belongs to the client, the cache to the
> provider. */
> +	p2pdma = rcu_dereference(client->p2pdma);
> +	per_mapping = p2pdma && p2pdma->ats_per_mapping;
>  	p2pdma = rcu_dereference(pdev->p2pdma);
>  
>  	if (p2pdma)
> @@ -1646,6 +1690,8 @@ pci_p2pdma_map_type_tlp(struct p2pdma_provider
> *provider, struct device *dev,
>  					     cache_index));
>  	provider_state = !!p2pdma;
>  	rcu_read_unlock();
> +	if (!per_mapping)
> +		tlp_flags = pci_p2pdma_client_tlp_flags(client,
> tlp_flags);
>  	type = pci_p2pdma_map_types_unpack(cached, tlp_flags);
>  	pci_dbg(client,
>  		 "P2PDMA ACS: map lookup provider=%s index=%#lx tlp-
> flags=%#x cached=%s(%d) provider-state=%u\n",
> @@ -1653,10 +1699,10 @@ pci_p2pdma_map_type_tlp(struct
> p2pdma_provider *provider, struct device *dev,
>  		 pci_p2pdma_map_type_name(type), type,
> provider_state);
>  
>  	if (type == PCI_P2PDMA_MAP_UNKNOWN)
> -		return calc_map_type_and_dist(pdev, client, &dist,
> tlp_flags,
> +		type = calc_map_type_and_dist(pdev, client, &dist,
> tlp_flags,
>  					      true);
>  
> -	return type;
> +	return pci_p2pdma_client_map_type(pdev, client, per_mapping,
> type);
>  }
>  
>  void __pci_p2pdma_update_state(struct pci_p2pdma_map_state *state,
> diff --git a/include/linux/pci-p2pdma.h b/include/linux/pci-p2pdma.h
> index 543a09426a11..8d96659df9df 100644
> --- a/include/linux/pci-p2pdma.h
> +++ b/include/linux/pci-p2pdma.h
> @@ -207,7 +207,8 @@ static inline struct pci_dev
> *pci_p2pmem_find(struct device *client)
>   * @dev: Client device that initiates the transfer
>   *
>   * Same as pci_p2pdma_map_type_tlp() for a client issuing strictly
> ordered
> - * Requests that carry an Untranslated address.
> + * Requests. Their Address Type is Untranslated unless the client
> enables ATS
> + * for the whole device.
>   */
>  static inline enum pci_p2pdma_map_type
>  pci_p2pdma_map_type(struct p2pdma_provider *provider, struct device
> *dev)

  reply	other threads:[~2026-09-28 12:43 UTC|newest]

Thread overview: 48+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-09-28 11:19 [PATCH v8 00/23] PCI/P2PDMA: Route peer-to-peer DMA by TLP class Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 01/23] PCI/P2PDMA: Document the TLP attribute assumptions Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 02/23] PCI/P2PDMA: Derive routing from directional ACS controls Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 03/23] PCI: Reject unreadable ACS controls in isolation checks Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 04/23] PCI/P2PDMA: Evaluate ACS controls at the path divergence Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 05/23] PCI/P2PDMA: Document directional ACS routing Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 06/23] PCI/P2PDMA: Collect the path's ACS controls before deciding Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 07/23] PCI/P2PDMA: Answer routing per TLP class Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 08/23] PCI/P2PDMA: Route Relaxed Ordering Completions directly Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 09/23] PCI/P2PDMA: Reject Translated Requests blocked by Translation Blocking Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 10/23] PCI/P2PDMA: Route Translated Requests under Direct Translated P2P Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 11/23] PCI/P2PDMA: Log detailed ACS routing diagnostics Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 12/23] PCI/P2PDMA: Add KUnit tests for the ACS routing decisions Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 13/23] PCI/P2PDMA: Test the ACS P2P routing walk Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 14/23] PCI: Add KUnit coverage for ACS isolation checks Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 15/23] PCI/P2PDMA: Document TLP-class routing Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 16/23] dma-buf: Let importers ask how peer-to-peer traffic is routed Leon Romanovsky
2026-09-29  9:34   ` Christian König
2026-09-29 13:23     ` Leon Romanovsky
2026-09-29 13:39       ` Christian König
2026-09-29 17:57         ` Leon Romanovsky
2026-09-30  7:00           ` Christian König
2026-09-30  8:18             ` Leon Romanovsky
2026-09-30  8:34               ` Christian König
2026-09-30 11:43                 ` Leon Romanovsky
2026-09-30 11:52                   ` Christian König
2026-09-30 14:32                     ` Leon Romanovsky
2026-09-30 14:52                       ` Christian König
2026-10-01  6:17                         ` Leon Romanovsky
2026-10-01  7:06                           ` Christian König
2026-09-30 14:57                       ` Thomas Hellström
2026-09-30 15:03                         ` Christian König
2026-09-30 16:52                         ` Jason Gunthorpe
2026-10-02  8:19                           ` Alistair Popple
2026-10-01  6:37                         ` Leon Romanovsky
2026-10-01  7:49                           ` Christian König
2026-10-01  8:27                             ` Leon Romanovsky
2026-10-01  9:07                               ` Christian König
2026-10-01  8:11                           ` Thomas Hellström
2026-09-28 11:19 ` [PATCH v8 17/23] vfio/pci: Hand out the P2PDMA provider behind a dma-buf Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 18/23] RDMA/uverbs: " Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 19/23] RDMA/mlx5: Ask P2PDMA whether ATS takes a direct peer-to-peer route Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 20/23] PCI/P2PDMA: Let a client declare that it selects ATS per mapping Leon Romanovsky
2026-09-28 12:44   ` Thomas Hellström
2026-09-28 11:19 ` [PATCH v8 21/23] RDMA/mlx5: Declare to P2PDMA that ATS is selected per memory key Leon Romanovsky
2026-09-28 11:19 ` [PATCH v8 22/23] PCI/P2PDMA: Evaluate the ATS path for clients with ATS enabled Leon Romanovsky
2026-09-28 12:43   ` Thomas Hellström [this message]
2026-09-28 11:19 ` [PATCH v8 23/23] PCI/P2PDMA: Test the routing of " Leon Romanovsky

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=e61fcc1552447e628b7ec938ac7e4e36555ebec2.camel@linux.intel.com \
    --to=thomas.hellstrom@linux.intel.com \
    --cc=alex@shazbot.org \
    --cc=ankita@nvidia.com \
    --cc=axboe@kernel.dk \
    --cc=bhelgaas@google.com \
    --cc=christian.koenig@amd.com \
    --cc=corbet@lwn.net \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=gregkh@linuxfoundation.org \
    --cc=iommu@lists.linux.dev \
    --cc=jgg@ziepe.ca \
    --cc=joro@8bytes.org \
    --cc=kch@nvidia.com \
    --cc=kvm@vger.kernel.org \
    --cc=leon@kernel.org \
    --cc=linaro-mm-sig@lists.linaro.org \
    --cc=linux-doc@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-media@vger.kernel.org \
    --cc=linux-pci@vger.kernel.org \
    --cc=linux-rdma@vger.kernel.org \
    --cc=logang@deltatee.com \
    --cc=rdunlap@infradead.org \
    --cc=robin.murphy@arm.com \
    --cc=skhan@linuxfoundation.org \
    --cc=sumit.semwal@linaro.org \
    --cc=tdave@nvidia.com \
    --cc=will@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.