RE: [PATCH v2 2/9] NTB: Add indexed ports NTB API

RE: [PATCH v2 2/9] NTB: Add indexed ports NTB API

[Allen Hubbe]

From: Serge Semin 
> There are some NTB hardware, which can combine more than just two domains
> over NTB. For instance, some IDT PCIe-switches can have NTB-functions
> activated on more than two-ports. The different domains are distinguished
> by ports they are connected to. So the new port-related methods are added to
> the NTB API:
>  ntb_port_number() - return local port
>  ntb_peer_port_count() - return number of peers local port can connect to
>  ntb_peer_port_number(pdix) - return port number by it index
>  ntb_peer_port_idx(port) - return port index by it number
> 
> Current test-drivers aren't changed much. They still support two-ports devices
> for the time being while multi-ports hardware drivers aren't added.
> 

The port methods are the same for PRI/SEC drivers.  Rather than duplicating the code, multiport could be made optional in the api, and default implementations provided by ntb common code.

Some comments below.

> Signed-off-by: Serge Semin <fancer.lancer@gmail.com>
> 
> ---
>  drivers/ntb/hw/amd/ntb_hw_amd.c     | 47 ++++++++++++++++++++++++
>  drivers/ntb/hw/amd/ntb_hw_amd.h     |  9 +++++
>  drivers/ntb/hw/intel/ntb_hw_intel.c | 52 ++++++++++++++++++++++++++-
>  drivers/ntb/hw/intel/ntb_hw_intel.h |  9 +++++
>  drivers/ntb/ntb_transport.c         |  6 ++++
>  drivers/ntb/test/ntb_perf.c         |  4 +++
>  drivers/ntb/test/ntb_pingpong.c     |  6 ++++
>  drivers/ntb/test/ntb_tool.c         |  5 +++
>  include/linux/ntb.h                 | 71 +++++++++++++++++++++++++++++++++++++
>  9 files changed, 208 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/ntb/hw/amd/ntb_hw_amd.c b/drivers/ntb/hw/amd/ntb_hw_amd.c
> index 6704327..0b767ef 100644
> --- a/drivers/ntb/hw/amd/ntb_hw_amd.c
> +++ b/drivers/ntb/hw/amd/ntb_hw_amd.c
> @@ -71,6 +71,49 @@ MODULE_AUTHOR("AMD Inc.");
>  static const struct file_operations amd_ntb_debugfs_info;
>  static struct dentry *debugfs_dir;
> 
> +static int amd_ntb_port_number(struct ntb_dev *ntb)
> +{
> +	switch (ntb->topo) {
> +	case NTB_TOPO_PRI:
> +	case NTB_TOPO_B2B_USD:
> +		return NTB_PORT_PRI_USD;
> +	case NTB_TOPO_SEC:
> +	case NTB_TOPO_B2B_DSD:
> +		return NTB_PORT_SEC_DSD;
> +	default:
> +		break;
> +	}
> +
> +	return -EINVAL;
> +}
> +
> +static int amd_ntb_peer_port_count(struct ntb_dev *ntb)
> +{
> +	return NTB_PEER_CNT;
> +}
> +
> +static int amd_ntb_peer_port_number(struct ntb_dev *ntb, int pidx)
> +{
> +	int local_port = amd_ntb_port_number(ntb);
> +
> +	if (pidx > NTB_PIDX_MAX)
> +		return -EINVAL;

pidx may be negative.

> +
> +	return (local_port == NTB_PORT_PRI_USD ?
> +		NTB_PORT_SEC_DSD : NTB_PORT_PRI_USD);

local_port may be -EINVAL.

It may be simpler to have the same switch statement here as in amd_ntb_port_number(), with the return statements swapped.

> +}
> +
> +static int amd_ntb_peer_port_idx(struct ntb_dev *ntb, int port)
> +{
> +	int local_port = amd_ntb_port_number(ntb);
> +
> +	if ((local_port == NTB_PORT_PRI_USD && port != NTB_PORT_SEC_DSD) ||
> +	    (local_port == NTB_PORT_SEC_DSD && port != NTB_PORT_PRI_USD))
> +		return -EINVAL;
> +

How about:

peer_port = amd_ntb_peer_port_number(ntb, 0);

if (peer_port == -EINVAL || port != peer_port)
	return -EINVAL;

return 0;

> +	return 0;
> +}
> +
>  static int amd_link_is_up(struct amd_ntb_dev *ndev)
>  {
>  	if (!ndev->peer_sta)
> @@ -431,6 +474,10 @@ static int amd_ntb_peer_spad_write(struct ntb_dev *ntb,
>  }
> 
>  static const struct ntb_dev_ops amd_ntb_ops = {
> +	.port_number		= amd_ntb_port_number,
> +	.peer_port_count	= amd_ntb_peer_port_count,
> +	.peer_port_number	= amd_ntb_peer_port_number,
> +	.peer_port_idx		= amd_ntb_peer_port_idx,
>  	.link_is_up		= amd_ntb_link_is_up,
>  	.link_enable		= amd_ntb_link_enable,
>  	.link_disable		= amd_ntb_link_disable,
> diff --git a/drivers/ntb/hw/amd/ntb_hw_amd.h b/drivers/ntb/hw/amd/ntb_hw_amd.h
> index 2eac3cd..1aeb08f 100644
> --- a/drivers/ntb/hw/amd/ntb_hw_amd.h
> +++ b/drivers/ntb/hw/amd/ntb_hw_amd.h
> @@ -62,6 +62,10 @@
>  #define NTB_LNK_STA_SPEED(x)	(((x) & NTB_LNK_STA_SPEED_MASK) >> 16)
>  #define NTB_LNK_STA_WIDTH(x)	(((x) & NTB_LNK_STA_WIDTH_MASK) >> 20)
> 
> +/* port related constants */
> +#define NTB_PEER_CNT			(1)
> +#define NTB_PIDX_MAX			(0)

Just NTB_PEER_CNT is sufficient.  Anything that checks if (pidx > NTB_PIDX_MAX) could check if (pidx >= NTB_PEER_CNT).

> +
>  #ifndef read64
>  #ifdef readq
>  #define read64 readq
> @@ -91,6 +95,11 @@ static inline void _write64(u64 val, void __iomem *mmio)
>  #endif
>  #endif
> 
> +enum amd_ntb_port {
> +	NTB_PORT_PRI_USD,
> +	NTB_PORT_SEC_DSD
> +};

This could be part of ntb.h, since it will likely be the same for any of the PRI/SEC variety of NTB devices.  Making it a part of the api will encourage other PRI/SEC drivers to assign port numbers the same way using these values.  Or, drivers may be made consistent by sharing common code.

> +
>  enum {
>  	/* AMD NTB Capability */
>  	AMD_MW_CNT		= 3,
> diff --git a/drivers/ntb/hw/intel/ntb_hw_intel.c b/drivers/ntb/hw/intel/ntb_hw_intel.c
> index 68d9908..7e44dc3 100644
> --- a/drivers/ntb/hw/intel/ntb_hw_intel.c
> +++ b/drivers/ntb/hw/intel/ntb_hw_intel.c
> @@ -1035,6 +1035,49 @@ static void ndev_deinit_debugfs(struct intel_ntb_dev *ndev)
>  	debugfs_remove_recursive(ndev->debugfs_dir);
>  }

Apply same comments above to Intel driver.

> 
> +static int intel_ntb_port_number(struct ntb_dev *ntb)
> +{
> +	switch (ntb->topo) {
> +	case NTB_TOPO_PRI:
> +	case NTB_TOPO_B2B_USD:
> +		return NTB_PORT_PRI_USD;
> +	case NTB_TOPO_SEC:
> +	case NTB_TOPO_B2B_DSD:
> +		return NTB_PORT_SEC_DSD;
> +	default:
> +		break;
> +	}
> +
> +	return -EINVAL;
> +}
> +
> +static int intel_ntb_peer_port_count(struct ntb_dev *ntb)
> +{
> +	return NTB_PEER_CNT;
> +}
> +
> +static int intel_ntb_peer_port_number(struct ntb_dev *ntb, int pidx)
> +{
> +	int local_port = intel_ntb_port_number(ntb);
> +
> +	if (pidx > NTB_PIDX_MAX)
> +		return -EINVAL;
> +
> +	return (local_port == NTB_PORT_PRI_USD ?
> +		NTB_PORT_SEC_DSD : NTB_PORT_PRI_USD);
> +}
> +
> +static int intel_ntb_peer_port_idx(struct ntb_dev *ntb, int port)
> +{
> +	int local_port = intel_ntb_port_number(ntb);
> +
> +	if ((local_port == NTB_PORT_PRI_USD && port != NTB_PORT_SEC_DSD) ||
> +	    (local_port == NTB_PORT_SEC_DSD && port != NTB_PORT_PRI_USD))
> +		return -EINVAL;
> +
> +	return 0;
> +}
> +
>  static int intel_ntb_link_is_up(struct ntb_dev *ntb,
>  				enum ntb_speed *speed,
>  				enum ntb_width *width)
> @@ -1775,7 +1818,6 @@ static int skx_init_ntb(struct intel_ntb_dev *ndev)
>  {
>  	int rc;
> 
> -
>  	ndev->mw_count = XEON_MW_COUNT;
>  	ndev->spad_count = SKX_SPAD_COUNT;
>  	ndev->db_count = SKX_DB_COUNT;
> @@ -2883,6 +2925,10 @@ static const struct intel_ntb_xlat_reg skx_sec_xlat = {
> 
>  /* operations for primary side of local ntb */
>  static const struct ntb_dev_ops intel_ntb_ops = {
> +	.port_number		= intel_ntb_port_number,
> +	.peer_port_count	= intel_ntb_peer_port_count,
> +	.peer_port_number	= intel_ntb_peer_port_number,
> +	.peer_port_idx		= intel_ntb_peer_port_idx,
>  	.link_is_up		= intel_ntb_link_is_up,
>  	.link_enable		= intel_ntb_link_enable,
>  	.link_disable		= intel_ntb_link_disable,
> @@ -2909,6 +2955,10 @@ static const struct ntb_dev_ops intel_ntb_ops = {
>  };
> 
>  static const struct ntb_dev_ops intel_ntb3_ops = {
> +	.port_number		= intel_ntb_port_number,
> +	.peer_port_count	= intel_ntb_peer_port_count,
> +	.peer_port_number	= intel_ntb_peer_port_number,
> +	.peer_port_idx		= intel_ntb_peer_port_idx,
>  	.link_is_up		= intel_ntb_link_is_up,
>  	.link_enable		= intel_ntb3_link_enable,
>  	.link_disable		= intel_ntb_link_disable,
> diff --git a/drivers/ntb/hw/intel/ntb_hw_intel.h b/drivers/ntb/hw/intel/ntb_hw_intel.h
> index 6e8c182..f12c960 100644
> --- a/drivers/ntb/hw/intel/ntb_hw_intel.h
> +++ b/drivers/ntb/hw/intel/ntb_hw_intel.h
> @@ -295,6 +295,15 @@
>  #define NTB_BAR_MASK_64			~(0xfull)
>  #define NTB_BAR_MASK_32			~(0xfu)
> 
> +/* port related constants */
> +#define NTB_PEER_CNT			(1)
> +#define NTB_PIDX_MAX			(0)
> +
> +enum intel_ntb_port {
> +	NTB_PORT_PRI_USD,
> +	NTB_PORT_SEC_DSD
> +};
> +
>  struct intel_ntb_dev;
> 
>  struct intel_ntb_reg {
> diff --git a/drivers/ntb/ntb_transport.c b/drivers/ntb/ntb_transport.c
> index 4eb8adb..37d428d 100644
> --- a/drivers/ntb/ntb_transport.c
> +++ b/drivers/ntb/ntb_transport.c
> @@ -94,6 +94,9 @@ MODULE_PARM_DESC(use_dma, "Use DMA engine to perform large data copy");
> 
>  static struct dentry *nt_debugfs_dir;
> 
> +/* Only two-ports NTB devices are supported */
> +#define PIDX		0
> +
>  struct ntb_queue_entry {
>  	/* ntb_queue list reference */
>  	struct list_head entry;
> @@ -1083,6 +1086,9 @@ static int ntb_transport_probe(struct ntb_client *self, struct
> ntb_dev *ndev)
>  		dev_dbg(&ndev->dev,
>  			"scratchpad is unsafe, proceed anyway...\n");
> 
> +	if (ntb_peer_port_count(ndev) != 1)
> +		dev_warn(&ndev->dev, "Multi-port NTB devices unsupported\n");
> +
>  	node = dev_to_node(&ndev->dev);
> 
>  	nt = kzalloc_node(sizeof(*nt), GFP_KERNEL, node);
> diff --git a/drivers/ntb/test/ntb_perf.c b/drivers/ntb/test/ntb_perf.c
> index e75d4fd..481827a 100644
> --- a/drivers/ntb/test/ntb_perf.c
> +++ b/drivers/ntb/test/ntb_perf.c
> @@ -76,6 +76,7 @@
>  #define DMA_RETRIES		20
>  #define SZ_4G			(1ULL << 32)
>  #define MAX_SEG_ORDER		20 /* no larger than 1M for kmalloc buffer */
> +#define PIDX			0
> 
>  MODULE_LICENSE(DRIVER_LICENSE);
>  MODULE_VERSION(DRIVER_VERSION);
> @@ -764,6 +765,9 @@ static int perf_probe(struct ntb_client *client, struct ntb_dev *ntb)
>  		return -EIO;
>  	}
> 
> +	if (ntb_peer_port_count(ntb) != 1)
> +		dev_warn(&ntb->dev, "Multi-port NTB devices unsupported\n");
> +
>  	node = dev_to_node(&pdev->dev);
> 
>  	perf = kzalloc_node(sizeof(*perf), GFP_KERNEL, node);
> diff --git a/drivers/ntb/test/ntb_pingpong.c b/drivers/ntb/test/ntb_pingpong.c
> index 4358611..6dd7582 100644
> --- a/drivers/ntb/test/ntb_pingpong.c
> +++ b/drivers/ntb/test/ntb_pingpong.c
> @@ -90,6 +90,9 @@ static unsigned long db_init = 0x7;
>  module_param(db_init, ulong, 0644);
>  MODULE_PARM_DESC(db_init, "Initial doorbell bits to ring on the peer");
> 
> +/* Only two-ports NTB devices are supported */
> +#define PIDX		0
> +
>  struct pp_ctx {
>  	struct ntb_dev			*ntb;
>  	u64				db_bits;
> @@ -230,6 +233,9 @@ static int pp_probe(struct ntb_client *client,
>  		}
>  	}
> 
> +	if (ntb_peer_port_count(ntb) != 1)
> +		dev_warn(&ntb->dev, "multi-port NTB is unsupported\n");
> +
>  	pp = kmalloc(sizeof(*pp), GFP_KERNEL);
>  	if (!pp) {
>  		rc = -ENOMEM;
> diff --git a/drivers/ntb/test/ntb_tool.c b/drivers/ntb/test/ntb_tool.c
> index 61bf2ef..85b6417 100644
> --- a/drivers/ntb/test/ntb_tool.c
> +++ b/drivers/ntb/test/ntb_tool.c
> @@ -120,6 +120,8 @@ MODULE_AUTHOR(DRIVER_AUTHOR);
>  MODULE_DESCRIPTION(DRIVER_DESCRIPTION);
> 
>  #define MAX_MWS 16
> +/* Only two-ports devices are supported */
> +#define PIDX	0
> 
>  static struct dentry *tool_dbgfs;
> 
> @@ -919,6 +921,9 @@ static int tool_probe(struct ntb_client *self, struct ntb_dev *ntb)
>  	if (ntb_spad_is_unsafe(ntb))
>  		dev_dbg(&ntb->dev, "scratchpad is unsafe\n");
> 
> +	if (ntb_peer_port_count(ntb) != 1)
> +		dev_warn(&ntb->dev, "multi-port NTB is unsupported\n");
> +
>  	tc = kzalloc(sizeof(*tc), GFP_KERNEL);
>  	if (!tc) {
>  		rc = -ENOMEM;
> diff --git a/include/linux/ntb.h b/include/linux/ntb.h
> index 5d1f260..3216689 100644
> --- a/include/linux/ntb.h
> +++ b/include/linux/ntb.h
> @@ -179,6 +179,10 @@ static inline int ntb_ctx_ops_is_valid(const struct ntb_ctx_ops *ops)
> 
>  /**
>   * struct ntb_ctx_ops - ntb device operations
> + * @port_number:	See ntb_port_number().
> + * @peer_port_count:	See ntb_peer_port_count().
> + * @peer_port_number:	See ntb_peer_port_number().
> + * @peer_port_idx:	See ntb_peer_port_idx().
>   * @link_is_up:		See ntb_link_is_up().
>   * @link_enable:	See ntb_link_enable().
>   * @link_disable:	See ntb_link_disable().
> @@ -212,6 +216,11 @@ static inline int ntb_ctx_ops_is_valid(const struct ntb_ctx_ops *ops)
>   * @peer_spad_write:	See ntb_peer_spad_write().
>   */
>  struct ntb_dev_ops {
> +	int (*port_number)(struct ntb_dev *ntb);
> +	int (*peer_port_count)(struct ntb_dev *ntb);
> +	int (*peer_port_number)(struct ntb_dev *ntb, int pidx);
> +	int (*peer_port_idx)(struct ntb_dev *ntb, int port);
> +
>  	int (*link_is_up)(struct ntb_dev *ntb,
>  			  enum ntb_speed *speed, enum ntb_width *width);
>  	int (*link_enable)(struct ntb_dev *ntb,
> @@ -265,6 +274,10 @@ static inline int ntb_dev_ops_is_valid(const struct ntb_dev_ops *ops)
>  {
>  	/* commented callbacks are not required: */
>  	return
> +		ops->port_number			&&
> +		ops->peer_port_count			&&
> +		ops->peer_port_number			&&
> +		ops->peer_port_idx			&&

If these are made optional, then:

!ops->peer_port_count == !ops->port_number &&
!ops->peer_port_number == !ops->port_number &&
!ops->peer_port_idx == !ops->port_number &&

>  		ops->link_is_up				&&
>  		ops->link_enable			&&
>  		ops->link_disable			&&
> @@ -442,6 +455,64 @@ void ntb_link_event(struct ntb_dev *ntb);
>  void ntb_db_event(struct ntb_dev *ntb, int vector);
> 
>  /**
> + * ntb_port_number() - get the local port number
> + * @ntb:	NTB device context.
> + *
> + * Hardware must support at least simple two-ports ntb connection
> + *
> + * Return: the local port number
> + */
> +static inline int ntb_port_number(struct ntb_dev *ntb)
> +{

If these are made optional, then:

if (!ntb->ops->port_number)
	return ntb_default_port_number(ntb);

> +	return ntb->ops->port_number(ntb);
> +}
> +
> +/**
> + * ntb_peer_port_count() - get the number of peer device ports
> + * @ntb:	NTB device context.
> + *
> + * Hardware may support an access to memory of several remote domains
> + * over multi-port NTB devices. This method returns the number of peers,
> + * local device can have shared memory with.
> + *
> + * Return: the number of peer ports
> + */
> +static inline int ntb_peer_port_count(struct ntb_dev *ntb)
> +{
> +	return ntb->ops->peer_port_count(ntb);
> +}
> +
> +/**
> + * ntb_peer_port_number() - get the peer port by given index
> + * @ntb:	NTB device context.
> + * @pidx:	Peer port index.
> + *
> + * Peer ports are continuously enumerated by NTB API logic, so this methods
> + * lets to retrieve port real number by its index.
> + *
> + * Return: the peer device port or negative value indicating an error
> + */
> +static inline int ntb_peer_port_number(struct ntb_dev *ntb, int pidx)
> +{
> +	return ntb->ops->peer_port_number(ntb, pidx);
> +}
> +
> +/**
> + * ntb_peer_port_idx() - get the peer device port index by given port number
> + * @ntb:	NTB device context.
> + * @port:	Peer port number.
> + *
> + * Inverse operation of ntb_peer_port_number(), so one can get port index
> + * by specified port number.
> + *
> + * Return: the peer port index or negative value indicating an error
> + */
> +static inline int ntb_peer_port_idx(struct ntb_dev *ntb, int port)
> +{
> +	return ntb->ops->peer_port_idx(ntb, port);
> +}
> +
> +/**
>   * ntb_link_is_up() - get the current ntb link state
>   * @ntb:	NTB device context.
>   * @speed:	OUT - The link speed expressed as PCIe generation number.
> --
> 2.6.6

[PATCH 00/22] NTB: Alter kernel API to support multi-port devices

[Serge Semin]

There are devices, like IDT PCIe-switches, which have more than just two ports.
Particularly one device can have up to eight ports with NTB-function activated.
In order to support such devices, NTB kernel API should be altered since
currently it's optimized to work with two-ports devices only.

Here are the changes I made to conform the design we discussed a few months ago:

1) Port-index-related methods are added to KAPI
 ntb_port_number();
 ntb_peer_port_count();
 ntb_peer_port_number(pdix);
 ntb_peer_port_idx(port);

2) Link state method returns bitfield of link states for each reachable port
 u64 ntb_link_is_up();
 
3) Link enable/disable methods work with local link side of NTB
 ntb_link_enable()/ntb_link_disable();
 
4) NTB memory window related interface does the following things
 ntb_mw_count(pidx); - number of inbound memory windows, which can be allocated
for shared buffer with specified peer device.
 ntb_mw_get_align(pidx, widx); - get alignment and size restrition parameters
to properly allocate inbound memory region.
 ntb_peer_mw_count(); - get number of outbound memory windows.
 ntb_peer_mw_get_addr(widx); - get mapping address of an outbound memory window 

Inbound MW based hardware:
 ntb_mw_set_trans(pidx, widx); - set translation address of allocated inbound
memory window so a peer device could access it.
 ntb_mw_clear_trans(pidx, widx); - clear the translation address of an inbound
memory window.

Outbound MW based hardware:
 ntb_peer_mw_set_trans(pidx, widx); - set translation address of a memory
window retrieved from a peer device
 ntb_peer_mw_clear_trans(pidx, widx); - clear the translation address of an
outbound memory window

5) Scratchpad interface needs to support multi-port devices as well
 ntb_spad_count() - return number of Scratchpad per each port
 ntb_peer_spad_addr(pidx, sidx) - address of Scratchpad register of the
peer device with pidx-index
 ntb_peer_spad_read(pidx, sidx) - read specified Scratchpad register of the
peer with pidx-index
 ntb_peer_spad_write(pidx, sidx) - write data to Scratchpad register of the
peer with pidx-index

6) Introduce new messaging interface of NTB KAPI
 ntb_msg_count(); - get number of message registers
 ntb_msg_inbits(); - get bitfield of inbound message registers status
 ntb_msg_outbits(); - get bitfield of outbound message registers status
 ntb_msg_read_sts(); - read the inbound and outbound message registers status
 ntb_msg_clear_sts(); - clear status bits of message registers
 ntb_msg_set_mask(); - mask interrupts raised by status bits of message
registers.
 ntb_msg_clear_mask(); - clear interrupts mask bits of message registers
 ntb_msg_recv(midx, *pidx); - read message register with specified index,
additionally getting peer port index which data received from
 ntb_msg_send(midx, pidx); - write data to the specified message register
sending it to the passed peer device connected over a pidx port
 ntb_msg_event(); - notify driver context of a new message event
 
7) Topology reduced to be either P2P (port-to-port) or B2B (bridge-to-bridge).
Since there is port number introduced to be part of ntb_dev structure, real
port number can be used to determine Primary and Secondary sides. Intel and AMD
driver are altered to support this novation.

8) Standard test drivers: PingPong, Debugging tool and Raw Perf as well as NTB
Transport drivers don't support multi-port devices at the moment.
Since we haven't got any real multi-port hadrware driver, it's dangerous to
make any serious alterations in the debugging tools. So I have made those
drivers to work the way they wokred before, but using the new NTB API.
The situation will change when I finish porting my current IDT NTB driver to
support new API. Then I'll be able to refactor the tools and test them using
real multi-port hardware.

The changes are made on top of the NTB-fork of the kernel:
https://github.com/jonmason/ntb
of "ntb"-branch with last commit:
9c763584b7c8911106bb77af7e648bef09af9d80 Linux 4.9-rc6

Signed-off-by: Serge Semin <fancer.lancer@gmail.com>

Serge Semin (22):
  NTB: Move link state API being first in sources
  NTB: Add peer indexed ports NTB API
  NTB: Alter NTB API to support both inbound and outbound MW based
    interfaces
  NTB: Add messaging NTB API
  NTB: Alter Scratchpads NTB API to support multi-ports interface
  NTB: Slightly alter link state NTB API
  NTB: Fix a few ntb.h issues
  NTB: Add T-Platforms copyrights to NTB API
  NTB Intel: Move link-related methods being first in the driver
  NTB Intel: Add port-related NTB API callback methods
  NTB Intel: Alter MW interface to fit new NTB API
  NTB Intel: Alter Scratchpads interface to fit new NTB API
  NTB Intel: Add T-Platforms copyrights to Intel NTB driver
  NTB AMD: Move link-related methods being first in the driver
  NTB AMD: Add port-related NTB API callback methods
  NTB AMD: Alter MW interface to fit new NTB API
  NTB AMD: Alter Scratchpads interface to fit new NTB API
  NTB AMD: Add T-Platforms copyrights to AMD NTB driver
  NTB PingPong: Alter driver to work with two-ports NTB API
  NTB Tool: Alter driver to work with two-ports NTB API
  NTB Perf: Alter driver to work with two-ports NTB API
  NTB Transport: Alter driver to work with two-ports NTB API

 drivers/ntb/hw/amd/ntb_hw_amd.c     | 310 +++++++++------
 drivers/ntb/hw/amd/ntb_hw_amd.h     |  12 +
 drivers/ntb/hw/intel/ntb_hw_intel.c | 420 ++++++++++++--------
 drivers/ntb/hw/intel/ntb_hw_intel.h |  12 +
 drivers/ntb/ntb.c                   |  15 +
 drivers/ntb/ntb_transport.c         |  43 +-
 drivers/ntb/test/ntb_perf.c         |  27 +-
 drivers/ntb/test/ntb_pingpong.c     |  11 +-
 drivers/ntb/test/ntb_tool.c         |  87 +++--
 include/linux/ntb.h                 | 753 +++++++++++++++++++++++++++++-------
 10 files changed, 1221 insertions(+), 469 deletions(-)

-- 
2.6.6

[PATCH v2 0/9] NTB: Alter kernel API to support multi-port devices

[Serge Semin]

There are devices, like IDT PCIe-switches, which have more than just two ports.
Particularly one device can have up to eight ports with NTB-function activated.
In order to support such devices, NTB kernel API should be altered since
currently it's optimized to work with two-ports devices only.

Changelog v2:
 - Move comments from cover letter to individual patches
 - Combine patches to make code buildable
 - Alter patchset to support Intel SKX driver
 - Make sure all the API uses the same midx/widx/pidx/sidx arguments notation
 - Move new MW API usage description into Documention
 - Alter Spad/Msg API checking valid function to make spad and msg interfaces optional
 - Alter comments in ntb.h
 - Split: add NTB_SPEED_GEN4 and ntb.h comments into separate patches
 - Put copyrights into some of the existing patches
 - Get rid of TOPO updates

Signed-off-by: Serge Semin <fancer.lancer@gmail.com>

Serge Semin (9):
  NTB: Make link-state API being declared first
  NTB: Add indexed ports NTB API
  NTB: Alter link-state API to support multi-port devices
  NTB: Alter MW API to support multi-ports devices
  NTB: Alter Scratchpads API to support multi-ports devices
  NTB: Add Messaging NTB API
  NTB: Add new Memory Windows API documentation
  NTB: Add PCIe Gen4 link speed
  NTB: Add ntb.h comments

 Documentation/ntb.txt               |  99 +++++-
 drivers/ntb/hw/amd/ntb_hw_amd.c     | 291 ++++++++++------
 drivers/ntb/hw/amd/ntb_hw_amd.h     |  11 +
 drivers/ntb/hw/intel/ntb_hw_intel.c | 296 +++++++++++------
 drivers/ntb/hw/intel/ntb_hw_intel.h |  11 +
 drivers/ntb/ntb.c                   |  15 +
 drivers/ntb/ntb_transport.c         |  44 ++-
 drivers/ntb/test/ntb_perf.c         |  27 +-
 drivers/ntb/test/ntb_pingpong.c     |  14 +-
 drivers/ntb/test/ntb_tool.c         |  93 ++++--
 include/linux/ntb.h                 | 639 ++++++++++++++++++++++++++++++------
 11 files changed, 1173 insertions(+), 367 deletions(-)

-- 
2.6.6

[PATCH v2 2/9] NTB: Add indexed ports NTB API

[Serge Semin]

There are some NTB hardware, which can combine more than just two domains
over NTB. For instance, some IDT PCIe-switches can have NTB-functions
activated on more than two-ports. The different domains are distinguished
by ports they are connected to. So the new port-related methods are added to
the NTB API:
 ntb_port_number() - return local port
 ntb_peer_port_count() - return number of peers local port can connect to
 ntb_peer_port_number(pdix) - return port number by it index
 ntb_peer_port_idx(port) - return port index by it number

Current test-drivers aren't changed much. They still support two-ports devices
for the time being while multi-ports hardware drivers aren't added.

Signed-off-by: Serge Semin <fancer.lancer@gmail.com>

---
 drivers/ntb/hw/amd/ntb_hw_amd.c     | 47 ++++++++++++++++++++++++
 drivers/ntb/hw/amd/ntb_hw_amd.h     |  9 +++++
 drivers/ntb/hw/intel/ntb_hw_intel.c | 52 ++++++++++++++++++++++++++-
 drivers/ntb/hw/intel/ntb_hw_intel.h |  9 +++++
 drivers/ntb/ntb_transport.c         |  6 ++++
 drivers/ntb/test/ntb_perf.c         |  4 +++
 drivers/ntb/test/ntb_pingpong.c     |  6 ++++
 drivers/ntb/test/ntb_tool.c         |  5 +++
 include/linux/ntb.h                 | 71 +++++++++++++++++++++++++++++++++++++
 9 files changed, 208 insertions(+), 1 deletion(-)

diff --git a/drivers/ntb/hw/amd/ntb_hw_amd.c b/drivers/ntb/hw/amd/ntb_hw_amd.c
index 6704327..0b767ef 100644
--- a/drivers/ntb/hw/amd/ntb_hw_amd.c
+++ b/drivers/ntb/hw/amd/ntb_hw_amd.c
@@ -71,6 +71,49 @@ MODULE_AUTHOR("AMD Inc.");
 static const struct file_operations amd_ntb_debugfs_info;
 static struct dentry *debugfs_dir;
 
+static int amd_ntb_port_number(struct ntb_dev *ntb)
+{
+	switch (ntb->topo) {
+	case NTB_TOPO_PRI:
+	case NTB_TOPO_B2B_USD:
+		return NTB_PORT_PRI_USD;
+	case NTB_TOPO_SEC:
+	case NTB_TOPO_B2B_DSD:
+		return NTB_PORT_SEC_DSD;
+	default:
+		break;
+	}
+
+	return -EINVAL;
+}
+
+static int amd_ntb_peer_port_count(struct ntb_dev *ntb)
+{
+	return NTB_PEER_CNT;
+}
+
+static int amd_ntb_peer_port_number(struct ntb_dev *ntb, int pidx)
+{
+	int local_port = amd_ntb_port_number(ntb);
+
+	if (pidx > NTB_PIDX_MAX)
+		return -EINVAL;
+
+	return (local_port == NTB_PORT_PRI_USD ?
+		NTB_PORT_SEC_DSD : NTB_PORT_PRI_USD);
+}
+
+static int amd_ntb_peer_port_idx(struct ntb_dev *ntb, int port)
+{
+	int local_port = amd_ntb_port_number(ntb);
+
+	if ((local_port == NTB_PORT_PRI_USD && port != NTB_PORT_SEC_DSD) ||
+	    (local_port == NTB_PORT_SEC_DSD && port != NTB_PORT_PRI_USD))
+		return -EINVAL;
+
+	return 0;
+}
+
 static int amd_link_is_up(struct amd_ntb_dev *ndev)
 {
 	if (!ndev->peer_sta)
@@ -431,6 +474,10 @@ static int amd_ntb_peer_spad_write(struct ntb_dev *ntb,
 }
 
 static const struct ntb_dev_ops amd_ntb_ops = {
+	.port_number		= amd_ntb_port_number,
+	.peer_port_count	= amd_ntb_peer_port_count,
+	.peer_port_number	= amd_ntb_peer_port_number,
+	.peer_port_idx		= amd_ntb_peer_port_idx,
 	.link_is_up		= amd_ntb_link_is_up,
 	.link_enable		= amd_ntb_link_enable,
 	.link_disable		= amd_ntb_link_disable,
diff --git a/drivers/ntb/hw/amd/ntb_hw_amd.h b/drivers/ntb/hw/amd/ntb_hw_amd.h
index 2eac3cd..1aeb08f 100644
--- a/drivers/ntb/hw/amd/ntb_hw_amd.h
+++ b/drivers/ntb/hw/amd/ntb_hw_amd.h
@@ -62,6 +62,10 @@
 #define NTB_LNK_STA_SPEED(x)	(((x) & NTB_LNK_STA_SPEED_MASK) >> 16)
 #define NTB_LNK_STA_WIDTH(x)	(((x) & NTB_LNK_STA_WIDTH_MASK) >> 20)
 
+/* port related constants */
+#define NTB_PEER_CNT			(1)
+#define NTB_PIDX_MAX			(0)
+
 #ifndef read64
 #ifdef readq
 #define read64 readq
@@ -91,6 +95,11 @@ static inline void _write64(u64 val, void __iomem *mmio)
 #endif
 #endif
 
+enum amd_ntb_port {
+	NTB_PORT_PRI_USD,
+	NTB_PORT_SEC_DSD
+};
+
 enum {
 	/* AMD NTB Capability */
 	AMD_MW_CNT		= 3,
diff --git a/drivers/ntb/hw/intel/ntb_hw_intel.c b/drivers/ntb/hw/intel/ntb_hw_intel.c
index 68d9908..7e44dc3 100644
--- a/drivers/ntb/hw/intel/ntb_hw_intel.c
+++ b/drivers/ntb/hw/intel/ntb_hw_intel.c
@@ -1035,6 +1035,49 @@ static void ndev_deinit_debugfs(struct intel_ntb_dev *ndev)
 	debugfs_remove_recursive(ndev->debugfs_dir);
 }
 
+static int intel_ntb_port_number(struct ntb_dev *ntb)
+{
+	switch (ntb->topo) {
+	case NTB_TOPO_PRI:
+	case NTB_TOPO_B2B_USD:
+		return NTB_PORT_PRI_USD;
+	case NTB_TOPO_SEC:
+	case NTB_TOPO_B2B_DSD:
+		return NTB_PORT_SEC_DSD;
+	default:
+		break;
+	}
+
+	return -EINVAL;
+}
+
+static int intel_ntb_peer_port_count(struct ntb_dev *ntb)
+{
+	return NTB_PEER_CNT;
+}
+
+static int intel_ntb_peer_port_number(struct ntb_dev *ntb, int pidx)
+{
+	int local_port = intel_ntb_port_number(ntb);
+
+	if (pidx > NTB_PIDX_MAX)
+		return -EINVAL;
+
+	return (local_port == NTB_PORT_PRI_USD ?
+		NTB_PORT_SEC_DSD : NTB_PORT_PRI_USD);
+}
+
+static int intel_ntb_peer_port_idx(struct ntb_dev *ntb, int port)
+{
+	int local_port = intel_ntb_port_number(ntb);
+
+	if ((local_port == NTB_PORT_PRI_USD && port != NTB_PORT_SEC_DSD) ||
+	    (local_port == NTB_PORT_SEC_DSD && port != NTB_PORT_PRI_USD))
+		return -EINVAL;
+
+	return 0;
+}
+
 static int intel_ntb_link_is_up(struct ntb_dev *ntb,
 				enum ntb_speed *speed,
 				enum ntb_width *width)
@@ -1775,7 +1818,6 @@ static int skx_init_ntb(struct intel_ntb_dev *ndev)
 {
 	int rc;
 
-
 	ndev->mw_count = XEON_MW_COUNT;
 	ndev->spad_count = SKX_SPAD_COUNT;
 	ndev->db_count = SKX_DB_COUNT;
@@ -2883,6 +2925,10 @@ static const struct intel_ntb_xlat_reg skx_sec_xlat = {
 
 /* operations for primary side of local ntb */
 static const struct ntb_dev_ops intel_ntb_ops = {
+	.port_number		= intel_ntb_port_number,
+	.peer_port_count	= intel_ntb_peer_port_count,
+	.peer_port_number	= intel_ntb_peer_port_number,
+	.peer_port_idx		= intel_ntb_peer_port_idx,
 	.link_is_up		= intel_ntb_link_is_up,
 	.link_enable		= intel_ntb_link_enable,
 	.link_disable		= intel_ntb_link_disable,
@@ -2909,6 +2955,10 @@ static const struct ntb_dev_ops intel_ntb_ops = {
 };
 
 static const struct ntb_dev_ops intel_ntb3_ops = {
+	.port_number		= intel_ntb_port_number,
+	.peer_port_count	= intel_ntb_peer_port_count,
+	.peer_port_number	= intel_ntb_peer_port_number,
+	.peer_port_idx		= intel_ntb_peer_port_idx,
 	.link_is_up		= intel_ntb_link_is_up,
 	.link_enable		= intel_ntb3_link_enable,
 	.link_disable		= intel_ntb_link_disable,
diff --git a/drivers/ntb/hw/intel/ntb_hw_intel.h b/drivers/ntb/hw/intel/ntb_hw_intel.h
index 6e8c182..f12c960 100644
--- a/drivers/ntb/hw/intel/ntb_hw_intel.h
+++ b/drivers/ntb/hw/intel/ntb_hw_intel.h
@@ -295,6 +295,15 @@
 #define NTB_BAR_MASK_64			~(0xfull)
 #define NTB_BAR_MASK_32			~(0xfu)
 
+/* port related constants */
+#define NTB_PEER_CNT			(1)
+#define NTB_PIDX_MAX			(0)
+
+enum intel_ntb_port {
+	NTB_PORT_PRI_USD,
+	NTB_PORT_SEC_DSD
+};
+
 struct intel_ntb_dev;
 
 struct intel_ntb_reg {
diff --git a/drivers/ntb/ntb_transport.c b/drivers/ntb/ntb_transport.c
index 4eb8adb..37d428d 100644
--- a/drivers/ntb/ntb_transport.c
+++ b/drivers/ntb/ntb_transport.c
@@ -94,6 +94,9 @@ MODULE_PARM_DESC(use_dma, "Use DMA engine to perform large data copy");
 
 static struct dentry *nt_debugfs_dir;
 
+/* Only two-ports NTB devices are supported */
+#define PIDX		0
+
 struct ntb_queue_entry {
 	/* ntb_queue list reference */
 	struct list_head entry;
@@ -1083,6 +1086,9 @@ static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev)
 		dev_dbg(&ndev->dev,
 			"scratchpad is unsafe, proceed anyway...\n");
 
+	if (ntb_peer_port_count(ndev) != 1)
+		dev_warn(&ndev->dev, "Multi-port NTB devices unsupported\n");
+
 	node = dev_to_node(&ndev->dev);
 
 	nt = kzalloc_node(sizeof(*nt), GFP_KERNEL, node);
diff --git a/drivers/ntb/test/ntb_perf.c b/drivers/ntb/test/ntb_perf.c
index e75d4fd..481827a 100644
--- a/drivers/ntb/test/ntb_perf.c
+++ b/drivers/ntb/test/ntb_perf.c
@@ -76,6 +76,7 @@
 #define DMA_RETRIES		20
 #define SZ_4G			(1ULL << 32)
 #define MAX_SEG_ORDER		20 /* no larger than 1M for kmalloc buffer */
+#define PIDX			0
 
 MODULE_LICENSE(DRIVER_LICENSE);
 MODULE_VERSION(DRIVER_VERSION);
@@ -764,6 +765,9 @@ static int perf_probe(struct ntb_client *client, struct ntb_dev *ntb)
 		return -EIO;
 	}
 
+	if (ntb_peer_port_count(ntb) != 1)
+		dev_warn(&ntb->dev, "Multi-port NTB devices unsupported\n");
+
 	node = dev_to_node(&pdev->dev);
 
 	perf = kzalloc_node(sizeof(*perf), GFP_KERNEL, node);
diff --git a/drivers/ntb/test/ntb_pingpong.c b/drivers/ntb/test/ntb_pingpong.c
index 4358611..6dd7582 100644
--- a/drivers/ntb/test/ntb_pingpong.c
+++ b/drivers/ntb/test/ntb_pingpong.c
@@ -90,6 +90,9 @@ static unsigned long db_init = 0x7;
 module_param(db_init, ulong, 0644);
 MODULE_PARM_DESC(db_init, "Initial doorbell bits to ring on the peer");
 
+/* Only two-ports NTB devices are supported */
+#define PIDX		0
+
 struct pp_ctx {
 	struct ntb_dev			*ntb;
 	u64				db_bits;
@@ -230,6 +233,9 @@ static int pp_probe(struct ntb_client *client,
 		}
 	}
 
+	if (ntb_peer_port_count(ntb) != 1)
+		dev_warn(&ntb->dev, "multi-port NTB is unsupported\n");
+
 	pp = kmalloc(sizeof(*pp), GFP_KERNEL);
 	if (!pp) {
 		rc = -ENOMEM;
diff --git a/drivers/ntb/test/ntb_tool.c b/drivers/ntb/test/ntb_tool.c
index 61bf2ef..85b6417 100644
--- a/drivers/ntb/test/ntb_tool.c
+++ b/drivers/ntb/test/ntb_tool.c
@@ -120,6 +120,8 @@ MODULE_AUTHOR(DRIVER_AUTHOR);
 MODULE_DESCRIPTION(DRIVER_DESCRIPTION);
 
 #define MAX_MWS 16
+/* Only two-ports devices are supported */
+#define PIDX	0
 
 static struct dentry *tool_dbgfs;
 
@@ -919,6 +921,9 @@ static int tool_probe(struct ntb_client *self, struct ntb_dev *ntb)
 	if (ntb_spad_is_unsafe(ntb))
 		dev_dbg(&ntb->dev, "scratchpad is unsafe\n");
 
+	if (ntb_peer_port_count(ntb) != 1)
+		dev_warn(&ntb->dev, "multi-port NTB is unsupported\n");
+
 	tc = kzalloc(sizeof(*tc), GFP_KERNEL);
 	if (!tc) {
 		rc = -ENOMEM;
diff --git a/include/linux/ntb.h b/include/linux/ntb.h
index 5d1f260..3216689 100644
--- a/include/linux/ntb.h
+++ b/include/linux/ntb.h
@@ -179,6 +179,10 @@ static inline int ntb_ctx_ops_is_valid(const struct ntb_ctx_ops *ops)
 
 /**
  * struct ntb_ctx_ops - ntb device operations
+ * @port_number:	See ntb_port_number().
+ * @peer_port_count:	See ntb_peer_port_count().
+ * @peer_port_number:	See ntb_peer_port_number().
+ * @peer_port_idx:	See ntb_peer_port_idx().
  * @link_is_up:		See ntb_link_is_up().
  * @link_enable:	See ntb_link_enable().
  * @link_disable:	See ntb_link_disable().
@@ -212,6 +216,11 @@ static inline int ntb_ctx_ops_is_valid(const struct ntb_ctx_ops *ops)
  * @peer_spad_write:	See ntb_peer_spad_write().
  */
 struct ntb_dev_ops {
+	int (*port_number)(struct ntb_dev *ntb);
+	int (*peer_port_count)(struct ntb_dev *ntb);
+	int (*peer_port_number)(struct ntb_dev *ntb, int pidx);
+	int (*peer_port_idx)(struct ntb_dev *ntb, int port);
+
 	int (*link_is_up)(struct ntb_dev *ntb,
 			  enum ntb_speed *speed, enum ntb_width *width);
 	int (*link_enable)(struct ntb_dev *ntb,
@@ -265,6 +274,10 @@ static inline int ntb_dev_ops_is_valid(const struct ntb_dev_ops *ops)
 {
 	/* commented callbacks are not required: */
 	return
+		ops->port_number			&&
+		ops->peer_port_count			&&
+		ops->peer_port_number			&&
+		ops->peer_port_idx			&&
 		ops->link_is_up				&&
 		ops->link_enable			&&
 		ops->link_disable			&&
@@ -442,6 +455,64 @@ void ntb_link_event(struct ntb_dev *ntb);
 void ntb_db_event(struct ntb_dev *ntb, int vector);
 
 /**
+ * ntb_port_number() - get the local port number
+ * @ntb:	NTB device context.
+ *
+ * Hardware must support at least simple two-ports ntb connection
+ *
+ * Return: the local port number
+ */
+static inline int ntb_port_number(struct ntb_dev *ntb)
+{
+	return ntb->ops->port_number(ntb);
+}
+
+/**
+ * ntb_peer_port_count() - get the number of peer device ports
+ * @ntb:	NTB device context.
+ *
+ * Hardware may support an access to memory of several remote domains
+ * over multi-port NTB devices. This method returns the number of peers,
+ * local device can have shared memory with.
+ *
+ * Return: the number of peer ports
+ */
+static inline int ntb_peer_port_count(struct ntb_dev *ntb)
+{
+	return ntb->ops->peer_port_count(ntb);
+}
+
+/**
+ * ntb_peer_port_number() - get the peer port by given index
+ * @ntb:	NTB device context.
+ * @pidx:	Peer port index.
+ *
+ * Peer ports are continuously enumerated by NTB API logic, so this methods
+ * lets to retrieve port real number by its index.
+ *
+ * Return: the peer device port or negative value indicating an error
+ */
+static inline int ntb_peer_port_number(struct ntb_dev *ntb, int pidx)
+{
+	return ntb->ops->peer_port_number(ntb, pidx);
+}
+
+/**
+ * ntb_peer_port_idx() - get the peer device port index by given port number
+ * @ntb:	NTB device context.
+ * @port:	Peer port number.
+ *
+ * Inverse operation of ntb_peer_port_number(), so one can get port index
+ * by specified port number.
+ *
+ * Return: the peer port index or negative value indicating an error
+ */
+static inline int ntb_peer_port_idx(struct ntb_dev *ntb, int port)
+{
+	return ntb->ops->peer_port_idx(ntb, port);
+}
+
+/**
  * ntb_link_is_up() - get the current ntb link state
  * @ntb:	NTB device context.
  * @speed:	OUT - The link speed expressed as PCIe generation number.
-- 
2.6.6