PCI Peer-to-Peer DMA Support

The PCI bus has pretty decent support for performing DMA transfers between two devices on the bus. This type of transaction is henceforth called Peer-to-Peer (or P2P). However, there are a number of issues that make P2P transactions tricky to do in a perfectly safe way.

One of the biggest issues is that PCI doesn’t require forwarding transactions between hierarchy domains, and in PCIe, each Root Port defines a separate hierarchy domain. To make things worse, there is no simple way to determine if a given Root Complex supports this or not. (See PCIe r4.0, sec 1.3.1). Therefore, as of this writing, the kernel only supports doing P2P when the endpoints involved are all behind the same PCI bridge, as such devices are all in the same PCI hierarchy domain, and the spec guarantees that all transactions within the hierarchy will be routable, but it does not require routing between hierarchies.

The second issue is that to make use of existing interfaces in Linux, memory that is used for P2P transactions needs to be backed by struct pages. However, PCI BARs are not typically cache coherent so there are a few corner case gotchas with these pages so developers need to be careful about what they do with them.

Driver Writer’s Guide

In a given P2P implementation there may be three or more different types of kernel drivers in play:

• Provider - A driver which provides or publishes P2P resources like memory or doorbell registers to other drivers.
• Client - A driver which makes use of a resource by setting up a DMA transaction to or from it.
• Orchestrator - A driver which orchestrates the flow of data between clients and providers.

In many cases there could be overlap between these three types (i.e., it may be typical for a driver to be both a provider and a client).

For example, in the NVMe Target Copy Offload implementation:

• The NVMe PCI driver is both a client, provider and orchestrator in that it exposes any CMB (Controller Memory Buffer) as a P2P memory resource (provider), it accepts P2P memory pages as buffers in requests to be used directly (client) and it can also make use of the CMB as submission queue entries (orchestrator).
• The RDMA driver is a client in this arrangement so that an RNIC can DMA directly to the memory exposed by the NVMe device.
• The NVMe Target driver (nvmet) can orchestrate the data from the RNIC to the P2P memory (CMB) and then to the NVMe device (and vice versa).

This is currently the only arrangement supported by the kernel but one could imagine slight tweaks to this that would allow for the same functionality. For example, if a specific RNIC added a BAR with some memory behind it, its driver could add support as a P2P provider and then the NVMe Target could use the RNIC’s memory instead of the CMB in cases where the NVMe cards in use do not have CMB support.

Provider Drivers

A provider simply needs to register a BAR (or a portion of a BAR) as a P2P DMA resource using pci_p2pdma_add_resource(). This will register struct pages for all the specified memory.

After that it may optionally publish all of its resources as P2P memory using pci_p2pmem_publish(). This will allow any orchestrator drivers to find and use the memory. When marked in this way, the resource must be regular memory with no side effects.

For the time being this is fairly rudimentary in that all resources are typically going to be P2P memory. Future work will likely expand this to include other types of resources like doorbells.

Client Drivers

A client driver typically only has to conditionally change its DMA map routine to use the mapping function pci_p2pdma_map_sg() instead of the usual dma_map_sg() function. Memory mapped in this way does not need to be unmapped.

The client may also, optionally, make use of is_pci_p2pdma_page() to determine when to use the P2P mapping functions and when to use the regular mapping functions. In some situations, it may be more appropriate to use a flag to indicate a given request is P2P memory and map appropriately. It is important to ensure that struct pages that back P2P memory stay out of code that does not have support for them as other code may treat the pages as regular memory which may not be appropriate.

Orchestrator Drivers

The first task an orchestrator driver must do is compile a list of all client devices that will be involved in a given transaction. For example, the NVMe Target driver creates a list including the namespace block device and the RNIC in use. If the orchestrator has access to a specific P2P provider to use it may check compatibility using pci_p2pdma_distance() otherwise it may find a memory provider that’s compatible with all clients using pci_p2pmem_find(). If more than one provider is supported, the one nearest to all the clients will be chosen first. If more than one provider is an equal distance away, the one returned will be chosen at random (it is not an arbitrary but truly random). This function returns the PCI device to use for the provider with a reference taken and therefore when it’s no longer needed it should be returned with pci_dev_put().

Once a provider is selected, the orchestrator can then use pci_alloc_p2pmem() and pci_free_p2pmem() to allocate P2P memory from the provider. pci_p2pmem_alloc_sgl() and pci_p2pmem_free_sgl() are convenience functions for allocating scatter-gather lists with P2P memory.

Struct Page Caveats

Driver writers should be very careful about not passing these special struct pages to code that isn’t prepared for it. At this time, the kernel interfaces do not have any checks for ensuring this. This obviously precludes passing these pages to userspace.

P2P memory is also technically IO memory but should never have any side effects behind it. Thus, the order of loads and stores should not be important and ioreadX(), iowriteX() and friends should not be necessary.

P2P DMA Support Library

int pci_p2pdma_add_resource(struct pci_dev * pdev, int bar, size_t size, u64 offset)

add memory for use as p2p memory

Parameters

struct pci_dev * pdev

the device to add the memory to

int bar

PCI BAR to add

size_t size

size of the memory to add, may be zero to use the whole BAR

u64 offset

offset into the PCI BAR

Description

The memory will be given ZONE_DEVICE struct pages so that it may be used with any DMA request.

int pci_p2pdma_distance_many(struct pci_dev * provider, struct device ** clients, int num_clients, bool verbose)

Determine the cumulative distance between a p2pdma provider and the clients in use.

Parameters

struct pci_dev * provider

p2pdma provider to check against the client list

struct device ** clients

array of devices to check (NULL-terminated)

int num_clients

number of clients in the array

bool verbose

if true, print warnings for devices when we return -1

Description

Returns -1 if any of the clients are not compatible (behind the same root port as the provider), otherwise returns a positive number where a lower number is the preferable choice. (If there’s one client that’s the same as the provider it will return 0, which is best choice).

For now, “compatible” means the provider and the clients are all behind the same PCI root port. This cuts out cases that may work but is safest for the user. Future work can expand this to white-list root complexes that can safely forward between each ports.

bool pci_has_p2pmem(struct pci_dev * pdev)

check if a given PCI device has published any p2pmem

Parameters

struct pci_dev * pdev

PCI device to check

struct pci_dev * pci_p2pmem_find_many(struct device ** clients, int num_clients)

find a peer-to-peer DMA memory device compatible with the specified list of clients and shortest distance (as determined by pci_p2pmem_dma())

Parameters

struct device ** clients

array of devices to check (NULL-terminated)

int num_clients

number of client devices in the list

Description

If multiple devices are behind the same switch, the one “closest” to the client devices in use will be chosen first. (So if one of the providers is the same as one of the clients, that provider will be used ahead of any other providers that are unrelated). If multiple providers are an equal distance away, one will be chosen at random.

Returns a pointer to the PCI device with a reference taken (use pci_dev_put to return the reference) or NULL if no compatible device is found. The found provider will also be assigned to the client list.

void * pci_alloc_p2pmem(struct pci_dev * pdev, size_t size)

allocate peer-to-peer DMA memory

Parameters

struct pci_dev * pdev

the device to allocate memory from

size_t size

number of bytes to allocate

Description

Returns the allocated memory or NULL on error.

void pci_free_p2pmem(struct pci_dev * pdev, void * addr, size_t size)

free peer-to-peer DMA memory

Parameters

struct pci_dev * pdev

the device the memory was allocated from

void * addr

address of the memory that was allocated

size_t size

number of bytes that were allocated

pci_bus_addr_t pci_p2pmem_virt_to_bus(struct pci_dev * pdev, void * addr)

return the PCI bus address for a given virtual address obtained with pci_alloc_p2pmem()

Parameters

struct pci_dev * pdev

the device the memory was allocated from

void * addr

address of the memory that was allocated

struct scatterlist * pci_p2pmem_alloc_sgl(struct pci_dev * pdev, unsigned int * nents, u32 length)

allocate peer-to-peer DMA memory in a scatterlist

Parameters

struct pci_dev * pdev

the device to allocate memory from

unsigned int * nents

the number of SG entries in the list

u32 length

number of bytes to allocate

Return

NULL on error or struct scatterlist pointer and nents on success

void pci_p2pmem_free_sgl(struct pci_dev * pdev, struct scatterlist * sgl)

free a scatterlist allocated by pci_p2pmem_alloc_sgl()

Parameters

struct pci_dev * pdev

the device to allocate memory from

struct scatterlist * sgl

the allocated scatterlist

void pci_p2pmem_publish(struct pci_dev * pdev, bool publish)

publish the peer-to-peer DMA memory for use by other devices with pci_p2pmem_find()

Parameters

struct pci_dev * pdev

the device with peer-to-peer DMA memory to publish

bool publish

set to true to publish the memory, false to unpublish it

Description

Published memory can be used by other PCI device drivers for peer-2-peer DMA operations. Non-published memory is reserved for exclusive use of the device driver that registers the peer-to-peer memory.

int pci_p2pdma_map_sg(struct device * dev, struct scatterlist * sg, int nents, enum dma_data_direction dir)

map a PCI peer-to-peer scatterlist for DMA

Parameters

struct device * dev

device doing the DMA request

struct scatterlist * sg

scatter list to map

int nents

elements in the scatterlist

enum dma_data_direction dir

DMA direction

Description

Scatterlists mapped with this function should not be unmapped in any way.

Returns the number of SG entries mapped or 0 on error.

int pci_p2pdma_enable_store(const char * page, struct pci_dev ** p2p_dev, bool * use_p2pdma)

parse a configfs/sysfs attribute store to enable p2pdma

Parameters

const char * page

contents of the value to be stored

struct pci_dev ** p2p_dev

returns the PCI device that was selected to be used (if one was specified in the stored value)

bool * use_p2pdma

returns whether to enable p2pdma or not

Description

Parses an attribute value to decide whether to enable p2pdma. The value can select a PCI device (using its full BDF device name) or a boolean (in any format strtobool() accepts). A false value disables p2pdma, a true value expects the caller to automatically find a compatible device and specifying a PCI device expects the caller to use the specific provider.

pci_p2pdma_enable_show() should be used as the show operation for the attribute.

Returns 0 on success

ssize_t pci_p2pdma_enable_show(char * page, struct pci_dev * p2p_dev, bool use_p2pdma)

show a configfs/sysfs attribute indicating whether p2pdma is enabled

Parameters

char * page

contents of the stored value

struct pci_dev * p2p_dev

the selected p2p device (NULL if no device is selected)

bool use_p2pdma

whether p2pdma has been enabled

Description

Attributes that use pci_p2pdma_enable_store() should use this function to show the value of the attribute.

Returns 0 on success