ntb_hw.c

/*
 * This file is provided under a dual BSD/GPLv2 license.  When using or
 *   redistributing this file, you may do so under either license.
 *
 *   GPL LICENSE SUMMARY
 *
 *   Copyright(c) 2012 Intel Corporation. All rights reserved.
 *
 *   This program is free software; you can redistribute it and/or modify
 *   it under the terms of version 2 of the GNU General Public License as
 *   published by the Free Software Foundation.
 *
 *   BSD LICENSE
 *
 *   Copyright(c) 2012 Intel Corporation. All rights reserved.
 *
 *   Redistribution and use in source and binary forms, with or without
 *   modification, are permitted provided that the following conditions
 *   are met:
 *
 *     * Redistributions of source code must retain the above copyright
 *       notice, this list of conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above copy
 *       notice, this list of conditions and the following disclaimer in
 *       the documentation and/or other materials provided with the
 *       distribution.
 *     * Neither the name of Intel Corporation nor the names of its
 *       contributors may be used to endorse or promote products derived
 *       from this software without specific prior written permission.
 *
 *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 * Intel PCIe NTB Linux driver
 *
 * Contact Information:
 * Jon Mason <jon.mason@intel.com>
 */
#include <linux/debugfs.h>
#include <linux/delay.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/module.h>
#include <linux/pci.h>
#include <linux/random.h>
#include <linux/slab.h>
#include "ntb_hw.h"
#include "ntb_regs.h"

#define NTB_NAME	"Intel(R) PCI-E Non-Transparent Bridge Driver"
#define NTB_VER		"0.25"

MODULE_DESCRIPTION(NTB_NAME);
MODULE_VERSION(NTB_VER);
MODULE_LICENSE("Dual BSD/GPL");
MODULE_AUTHOR("Intel Corporation");

static bool xeon_errata_workaround = true;
module_param(xeon_errata_workaround, bool, 0644);
MODULE_PARM_DESC(xeon_errata_workaround, "Workaround for the Xeon Errata");

enum {
	NTB_CONN_CLASSIC = 0,
	NTB_CONN_B2B,
	NTB_CONN_RP,
};

enum {
	NTB_DEV_USD = 0,
	NTB_DEV_DSD,
};

enum {
	SNB_HW = 0,
	BWD_HW,
};

static struct dentry *debugfs_dir;

#define BWD_LINK_RECOVERY_TIME	500

/* Translate memory window 0,1 to BAR 2,4 */
#define MW_TO_BAR(mw)	(mw * NTB_MAX_NUM_MW + 2)

static DEFINE_PCI_DEVICE_TABLE(ntb_pci_tbl) = {
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_BWD)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_JSF)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_SNB)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_IVT)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_HSX)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_JSF)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_SNB)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_IVT)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_HSX)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_JSF)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_SNB)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_IVT)},
	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_HSX)},
	{0}
};
MODULE_DEVICE_TABLE(pci, ntb_pci_tbl);

/**
 * ntb_register_event_callback() - register event callback
 * @ndev: pointer to ntb_device instance
 * @func: callback function to register
 *
 * This function registers a callback for any HW driver events such as link
 * up/down, power management notices and etc.
 *
 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
 */
int ntb_register_event_callback(struct ntb_device *ndev,
			    void (*func)(void *handle, enum ntb_hw_event event))
{
	if (ndev->event_cb)
		return -EINVAL;

	ndev->event_cb = func;

	return 0;
}

/**
 * ntb_unregister_event_callback() - unregisters the event callback
 * @ndev: pointer to ntb_device instance
 *
 * This function unregisters the existing callback from transport
 */
void ntb_unregister_event_callback(struct ntb_device *ndev)
{
	ndev->event_cb = NULL;
}

/**
 * ntb_register_db_callback() - register a callback for doorbell interrupt
 * @ndev: pointer to ntb_device instance
 * @idx: doorbell index to register callback, zero based
 * @func: callback function to register
 *
 * This function registers a callback function for the doorbell interrupt
 * on the primary side. The function will unmask the doorbell as well to
 * allow interrupt.
 *
 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
 */
int ntb_register_db_callback(struct ntb_device *ndev, unsigned int idx,
			     void *data, void (*func)(void *data, int db_num))
{
	unsigned long mask;

	if (idx >= ndev->max_cbs || ndev->db_cb[idx].callback) {
		dev_warn(&ndev->pdev->dev, "Invalid Index.\n");
		return -EINVAL;
	}

	ndev->db_cb[idx].callback = func;
	ndev->db_cb[idx].data = data;

	/* unmask interrupt */
	mask = readw(ndev->reg_ofs.pdb_mask);
	clear_bit(idx * ndev->bits_per_vector, &mask);
	writew(mask, ndev->reg_ofs.pdb_mask);

	return 0;
}

/**
 * ntb_unregister_db_callback() - unregister a callback for doorbell interrupt
 * @ndev: pointer to ntb_device instance
 * @idx: doorbell index to register callback, zero based
 *
 * This function unregisters a callback function for the doorbell interrupt
 * on the primary side. The function will also mask the said doorbell.
 */
void ntb_unregister_db_callback(struct ntb_device *ndev, unsigned int idx)
{
	unsigned long mask;

	if (idx >= ndev->max_cbs || !ndev->db_cb[idx].callback)
		return;

	mask = readw(ndev->reg_ofs.pdb_mask);
	set_bit(idx * ndev->bits_per_vector, &mask);
	writew(mask, ndev->reg_ofs.pdb_mask);

	ndev->db_cb[idx].callback = NULL;
}

/**
 * ntb_find_transport() - find the transport pointer
 * @transport: pointer to pci device
 *
 * Given the pci device pointer, return the transport pointer passed in when
 * the transport attached when it was inited.
 *
 * RETURNS: pointer to transport.
 */
void *ntb_find_transport(struct pci_dev *pdev)
{
	struct ntb_device *ndev = pci_get_drvdata(pdev);
	return ndev->ntb_transport;
}

/**
 * ntb_register_transport() - Register NTB transport with NTB HW driver
 * @transport: transport identifier
 *
 * This function allows a transport to reserve the hardware driver for
 * NTB usage.
 *
 * RETURNS: pointer to ntb_device, NULL on error.
 */
struct ntb_device *ntb_register_transport(struct pci_dev *pdev, void *transport)
{
	struct ntb_device *ndev = pci_get_drvdata(pdev);

	if (ndev->ntb_transport)
		return NULL;

	ndev->ntb_transport = transport;
	return ndev;
}

/**
 * ntb_unregister_transport() - Unregister the transport with the NTB HW driver
 * @ndev - ntb_device of the transport to be freed
 *
 * This function unregisters the transport from the HW driver and performs any
 * necessary cleanups.
 */
void ntb_unregister_transport(struct ntb_device *ndev)
{
	int i;

	if (!ndev->ntb_transport)
		return;

	for (i = 0; i < ndev->max_cbs; i++)
		ntb_unregister_db_callback(ndev, i);

	ntb_unregister_event_callback(ndev);
	ndev->ntb_transport = NULL;
}

/**
 * ntb_write_local_spad() - write to the secondary scratchpad register
 * @ndev: pointer to ntb_device instance
 * @idx: index to the scratchpad register, 0 based
 * @val: the data value to put into the register
 *
 * This function allows writing of a 32bit value to the indexed scratchpad
 * register. This writes over the data mirrored to the local scratchpad register
 * by the remote system.
 *
 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
 */
int ntb_write_local_spad(struct ntb_device *ndev, unsigned int idx, u32 val)
{
	if (idx >= ndev->limits.max_spads)
		return -EINVAL;

	dev_dbg(&ndev->pdev->dev, "Writing %x to local scratch pad index %d\n",
		val, idx);
	writel(val, ndev->reg_ofs.spad_read + idx * 4);

	return 0;
}

/**
 * ntb_read_local_spad() - read from the primary scratchpad register
 * @ndev: pointer to ntb_device instance
 * @idx: index to scratchpad register, 0 based
 * @val: pointer to 32bit integer for storing the register value
 *
 * This function allows reading of the 32bit scratchpad register on
 * the primary (internal) side.  This allows the local system to read data
 * written and mirrored to the scratchpad register by the remote system.
 *
 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
 */
int ntb_read_local_spad(struct ntb_device *ndev, unsigned int idx, u32 *val)
{
	if (idx >= ndev->limits.max_spads)
		return -EINVAL;

	*val = readl(ndev->reg_ofs.spad_write + idx * 4);
	dev_dbg(&ndev->pdev->dev,
		"Reading %x from local scratch pad index %d\n", *val, idx);

	return 0;
}

/**
 * ntb_write_remote_spad() - write to the secondary scratchpad register
 * @ndev: pointer to ntb_device instance
 * @idx: index to the scratchpad register, 0 based
 * @val: the data value to put into the register
 *
 * This function allows writing of a 32bit value to the indexed scratchpad
 * register. The register resides on the secondary (external) side.  This allows
 * the local system to write data to be mirrored to the remote systems
 * scratchpad register.
 *
 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
 */
int ntb_write_remote_spad(struct ntb_device *ndev, unsigned int idx, u32 val)
{
	if (idx >= ndev->limits.max_spads)
		return -EINVAL;

	dev_dbg(&ndev->pdev->dev, "Writing %x to remote scratch pad index %d\n",
		val, idx);
	writel(val, ndev->reg_ofs.spad_write + idx * 4);

	return 0;
}

/**
 * ntb_read_remote_spad() - read from the primary scratchpad register
 * @ndev: pointer to ntb_device instance
 * @idx: index to scratchpad register, 0 based
 * @val: pointer to 32bit integer for storing the register value
 *
 * This function allows reading of the 32bit scratchpad register on
 * the primary (internal) side.  This alloows the local system to read the data
 * it wrote to be mirrored on the remote system.
 *
 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
 */
int ntb_read_remote_spad(struct ntb_device *ndev, unsigned int idx, u32 *val)
{
	if (idx >= ndev->limits.max_spads)
		return -EINVAL;

	*val = readl(ndev->reg_ofs.spad_read + idx * 4);
	dev_dbg(&ndev->pdev->dev,
		"Reading %x from remote scratch pad index %d\n", *val, idx);

	return 0;
}

/**
 * ntb_get_mw_vbase() - get virtual addr for the NTB memory window
 * @ndev: pointer to ntb_device instance
 * @mw: memory window number
 *
 * This function provides the base virtual address of the memory window
 * specified.
 *
 * RETURNS: pointer to virtual address, or NULL on error.
 */
void __iomem *ntb_get_mw_vbase(struct ntb_device *ndev, unsigned int mw)
{
	if (mw >= ntb_max_mw(ndev))
		return NULL;

	return ndev->mw[mw].vbase;
}

/**
 * ntb_get_mw_size() - return size of NTB memory window
 * @ndev: pointer to ntb_device instance
 * @mw: memory window number
 *
 * This function provides the physical size of the memory window specified
 *
 * RETURNS: the size of the memory window or zero on error
 */
u64 ntb_get_mw_size(struct ntb_device *ndev, unsigned int mw)
{
	if (mw >= ntb_max_mw(ndev))
		return 0;

	return ndev->mw[mw].bar_sz;
}

/**
 * ntb_set_mw_addr - set the memory window address
 * @ndev: pointer to ntb_device instance
 * @mw: memory window number
 * @addr: base address for data
 *
 * This function sets the base physical address of the memory window.  This
 * memory address is where data from the remote system will be transfered into
 * or out of depending on how the transport is configured.
 */
void ntb_set_mw_addr(struct ntb_device *ndev, unsigned int mw, u64 addr)
{
	if (mw >= ntb_max_mw(ndev))
		return;

	dev_dbg(&ndev->pdev->dev, "Writing addr %Lx to BAR %d\n", addr,
		MW_TO_BAR(mw));

	ndev->mw[mw].phys_addr = addr;

	switch (MW_TO_BAR(mw)) {
	case NTB_BAR_23:
		writeq(addr, ndev->reg_ofs.sbar2_xlat);
		break;
	case NTB_BAR_45:
		writeq(addr, ndev->reg_ofs.sbar4_xlat);
		break;
	}
}

/**
 * ntb_ring_sdb() - Set the doorbell on the secondary/external side
 * @ndev: pointer to ntb_device instance
 * @db: doorbell to ring
 *
 * This function allows triggering of a doorbell on the secondary/external
 * side that will initiate an interrupt on the remote host
 *
 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
 */
void ntb_ring_sdb(struct ntb_device *ndev, unsigned int db)
{
	dev_dbg(&ndev->pdev->dev, "%s: ringing doorbell %d\n", __func__, db);

	if (ndev->hw_type == BWD_HW)
		writeq((u64) 1 << db, ndev->reg_ofs.sdb);
	else
		writew(((1 << ndev->bits_per_vector) - 1) <<
		       (db * ndev->bits_per_vector), ndev->reg_ofs.sdb);
}

static void bwd_recover_link(struct ntb_device *ndev)
{
	u32 status;

	/* Driver resets the NTB ModPhy lanes - magic! */
	writeb(0xe0, ndev->reg_base + BWD_MODPHY_PCSREG6);
	writeb(0x40, ndev->reg_base + BWD_MODPHY_PCSREG4);
	writeb(0x60, ndev->reg_base + BWD_MODPHY_PCSREG4);
	writeb(0x60, ndev->reg_base + BWD_MODPHY_PCSREG6);

	/* Driver waits 100ms to allow the NTB ModPhy to settle */
	msleep(100);

	/* Clear AER Errors, write to clear */
	status = readl(ndev->reg_base + BWD_ERRCORSTS_OFFSET);
	dev_dbg(&ndev->pdev->dev, "ERRCORSTS = %x\n", status);
	status &= PCI_ERR_COR_REP_ROLL;
	writel(status, ndev->reg_base + BWD_ERRCORSTS_OFFSET);

	/* Clear unexpected electrical idle event in LTSSM, write to clear */
	status = readl(ndev->reg_base + BWD_LTSSMERRSTS0_OFFSET);
	dev_dbg(&ndev->pdev->dev, "LTSSMERRSTS0 = %x\n", status);
	status |= BWD_LTSSMERRSTS0_UNEXPECTEDEI;
	writel(status, ndev->reg_base + BWD_LTSSMERRSTS0_OFFSET);

	/* Clear DeSkew Buffer error, write to clear */
	status = readl(ndev->reg_base + BWD_DESKEWSTS_OFFSET);
	dev_dbg(&ndev->pdev->dev, "DESKEWSTS = %x\n", status);
	status |= BWD_DESKEWSTS_DBERR;
	writel(status, ndev->reg_base + BWD_DESKEWSTS_OFFSET);

	status = readl(ndev->reg_base + BWD_IBSTERRRCRVSTS0_OFFSET);
	dev_dbg(&ndev->pdev->dev, "IBSTERRRCRVSTS0 = %x\n", status);
	status &= BWD_IBIST_ERR_OFLOW;
	writel(status, ndev->reg_base + BWD_IBSTERRRCRVSTS0_OFFSET);

	/* Releases the NTB state machine to allow the link to retrain */
	status = readl(ndev->reg_base + BWD_LTSSMSTATEJMP_OFFSET);
	dev_dbg(&ndev->pdev->dev, "LTSSMSTATEJMP = %x\n", status);
	status &= ~BWD_LTSSMSTATEJMP_FORCEDETECT;
	writel(status, ndev->reg_base + BWD_LTSSMSTATEJMP_OFFSET);
}

static void ntb_link_event(struct ntb_device *ndev, int link_state)
{
	unsigned int event;

	if (ndev->link_status == link_state)
		return;

	if (link_state == NTB_LINK_UP) {
		u16 status;

		dev_info(&ndev->pdev->dev, "Link Up\n");
		ndev->link_status = NTB_LINK_UP;
		event = NTB_EVENT_HW_LINK_UP;

		if (ndev->hw_type == BWD_HW)
			status = readw(ndev->reg_ofs.lnk_stat);
		else {
			int rc = pci_read_config_word(ndev->pdev,
						      SNB_LINK_STATUS_OFFSET,
						      &status);
			if (rc)
				return;
		}

		ndev->link_width = (status & NTB_LINK_WIDTH_MASK) >> 4;
		ndev->link_speed = (status & NTB_LINK_SPEED_MASK);
		dev_info(&ndev->pdev->dev, "Link Width %d, Link Speed %d\n",
			 ndev->link_width, ndev->link_speed);
	} else {
		dev_info(&ndev->pdev->dev, "Link Down\n");
		ndev->link_status = NTB_LINK_DOWN;
		event = NTB_EVENT_HW_LINK_DOWN;
		/* Don't modify link width/speed, we need it in link recovery */
	}

	/* notify the upper layer if we have an event change */
	if (ndev->event_cb)
		ndev->event_cb(ndev->ntb_transport, event);
}

static int ntb_link_status(struct ntb_device *ndev)
{
	int link_state;

	if (ndev->hw_type == BWD_HW) {
		u32 ntb_cntl;

		ntb_cntl = readl(ndev->reg_ofs.lnk_cntl);
		if (ntb_cntl & BWD_CNTL_LINK_DOWN)
			link_state = NTB_LINK_DOWN;
		else
			link_state = NTB_LINK_UP;
	} else {
		u16 status;
		int rc;

		rc = pci_read_config_word(ndev->pdev, SNB_LINK_STATUS_OFFSET,
					  &status);
		if (rc)
			return rc;

		if (status & NTB_LINK_STATUS_ACTIVE)
			link_state = NTB_LINK_UP;
		else
			link_state = NTB_LINK_DOWN;
	}

	ntb_link_event(ndev, link_state);

	return 0;
}

static void bwd_link_recovery(struct work_struct *work)
{
	struct ntb_device *ndev = container_of(work, struct ntb_device,
					       lr_timer.work);
	u32 status32;

	bwd_recover_link(ndev);
	/* There is a potential race between the 2 NTB devices recovering at the
	 * same time.  If the times are the same, the link will not recover and
	 * the driver will be stuck in this loop forever.  Add a random interval
	 * to the recovery time to prevent this race.
	 */
	msleep(BWD_LINK_RECOVERY_TIME + prandom_u32() % BWD_LINK_RECOVERY_TIME);

	status32 = readl(ndev->reg_base + BWD_LTSSMSTATEJMP_OFFSET);
	if (status32 & BWD_LTSSMSTATEJMP_FORCEDETECT)
		goto retry;

	status32 = readl(ndev->reg_base + BWD_IBSTERRRCRVSTS0_OFFSET);
	if (status32 & BWD_IBIST_ERR_OFLOW)
		goto retry;

	status32 = readl(ndev->reg_ofs.lnk_cntl);
	if (!(status32 & BWD_CNTL_LINK_DOWN)) {
		unsigned char speed, width;
		u16 status16;

		status16 = readw(ndev->reg_ofs.lnk_stat);
		width = (status16 & NTB_LINK_WIDTH_MASK) >> 4;
		speed = (status16 & NTB_LINK_SPEED_MASK);
		if (ndev->link_width != width || ndev->link_speed != speed)
			goto retry;
	}

	schedule_delayed_work(&ndev->hb_timer, NTB_HB_TIMEOUT);
	return;

retry:
	schedule_delayed_work(&ndev->lr_timer, NTB_HB_TIMEOUT);
}

/* BWD doesn't have link status interrupt, poll on that platform */
static void bwd_link_poll(struct work_struct *work)
{
	struct ntb_device *ndev = container_of(work, struct ntb_device,
					       hb_timer.work);
	unsigned long ts = jiffies;

	/* If we haven't gotten an interrupt in a while, check the BWD link
	 * status bit
	 */
	if (ts > ndev->last_ts + NTB_HB_TIMEOUT) {
		int rc = ntb_link_status(ndev);
		if (rc)
			dev_err(&ndev->pdev->dev,
				"Error determining link status\n");

		/* Check to see if a link error is the cause of the link down */
		if (ndev->link_status == NTB_LINK_DOWN) {
			u32 status32 = readl(ndev->reg_base +
					     BWD_LTSSMSTATEJMP_OFFSET);
			if (status32 & BWD_LTSSMSTATEJMP_FORCEDETECT) {
				schedule_delayed_work(&ndev->lr_timer, 0);
				return;
			}
		}
	}

	schedule_delayed_work(&ndev->hb_timer, NTB_HB_TIMEOUT);
}

static int ntb_xeon_setup(struct ntb_device *ndev)
{
	int rc;
	u8 val;

	ndev->hw_type = SNB_HW;

	rc = pci_read_config_byte(ndev->pdev, NTB_PPD_OFFSET, &val);
	if (rc)
		return rc;

	switch (val & SNB_PPD_CONN_TYPE) {
	case NTB_CONN_B2B:
		ndev->conn_type = NTB_CONN_B2B;
		break;
	case NTB_CONN_CLASSIC:
	case NTB_CONN_RP:
	default:
		dev_err(&ndev->pdev->dev, "Only B2B supported at this time\n");
		return -EINVAL;
	}

	if (val & SNB_PPD_DEV_TYPE)
		ndev->dev_type = NTB_DEV_USD;
	else
		ndev->dev_type = NTB_DEV_DSD;

	ndev->reg_ofs.pdb = ndev->reg_base + SNB_PDOORBELL_OFFSET;
	ndev->reg_ofs.pdb_mask = ndev->reg_base + SNB_PDBMSK_OFFSET;
	ndev->reg_ofs.sbar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET;
	ndev->reg_ofs.sbar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET;
	ndev->reg_ofs.lnk_cntl = ndev->reg_base + SNB_NTBCNTL_OFFSET;
	ndev->reg_ofs.lnk_stat = ndev->reg_base + SNB_LINK_STATUS_OFFSET;
	ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET;
	ndev->reg_ofs.spci_cmd = ndev->reg_base + SNB_PCICMD_OFFSET;

	/* There is a Xeon hardware errata related to writes to
	 * SDOORBELL or B2BDOORBELL in conjunction with inbound access
	 * to NTB MMIO Space, which may hang the system.  To workaround
	 * this use the second memory window to access the interrupt and
	 * scratch pad registers on the remote system.
	 */
	if (xeon_errata_workaround) {
		if (!ndev->mw[1].bar_sz)
			return -EINVAL;

		ndev->limits.max_mw = SNB_ERRATA_MAX_MW;
		ndev->reg_ofs.spad_write = ndev->mw[1].vbase +
					   SNB_SPAD_OFFSET;
		ndev->reg_ofs.sdb = ndev->mw[1].vbase +
				    SNB_PDOORBELL_OFFSET;

		/* Set the Limit register to 4k, the minimum size, to
		 * prevent an illegal access
		 */
		writeq(ndev->mw[1].bar_sz + 0x1000, ndev->reg_base +
		       SNB_PBAR4LMT_OFFSET);
	} else {
		ndev->limits.max_mw = SNB_MAX_MW;
		ndev->reg_ofs.spad_write = ndev->reg_base +
					   SNB_B2B_SPAD_OFFSET;
		ndev->reg_ofs.sdb = ndev->reg_base +
				    SNB_B2B_DOORBELL_OFFSET;

		/* Disable the Limit register, just incase it is set to
		 * something silly
		 */
		writeq(0, ndev->reg_base + SNB_PBAR4LMT_OFFSET);
	}

	/* The Xeon errata workaround requires setting SBAR Base
	 * addresses to known values, so that the PBAR XLAT can be
	 * pointed at SBAR0 of the remote system.
	 */
	if (ndev->dev_type == NTB_DEV_USD) {
		writeq(SNB_MBAR23_DSD_ADDR, ndev->reg_base +
		       SNB_PBAR2XLAT_OFFSET);
		if (xeon_errata_workaround)
			writeq(SNB_MBAR01_DSD_ADDR, ndev->reg_base +
			       SNB_PBAR4XLAT_OFFSET);
		else {
			writeq(SNB_MBAR45_DSD_ADDR, ndev->reg_base +
			       SNB_PBAR4XLAT_OFFSET);
			/* B2B_XLAT_OFFSET is a 64bit register, but can
			 * only take 32bit writes
			 */
			writel(SNB_MBAR01_USD_ADDR & 0xffffffff,
			       ndev->reg_base + SNB_B2B_XLAT_OFFSETL);
			writel(SNB_MBAR01_DSD_ADDR >> 32,
			       ndev->reg_base + SNB_B2B_XLAT_OFFSETU);
		}

		writeq(SNB_MBAR01_USD_ADDR, ndev->reg_base +
		       SNB_SBAR0BASE_OFFSET);
		writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base +
		       SNB_SBAR2BASE_OFFSET);
		writeq(SNB_MBAR45_USD_ADDR, ndev->reg_base +
		       SNB_SBAR4BASE_OFFSET);
	} else {
		writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base +
		       SNB_PBAR2XLAT_OFFSET);
		if (xeon_errata_workaround)
			writeq(SNB_MBAR01_USD_ADDR, ndev->reg_base +
			       SNB_PBAR4XLAT_OFFSET);
		else {
			writeq(SNB_MBAR45_USD_ADDR, ndev->reg_base +
			       SNB_PBAR4XLAT_OFFSET);
			/* B2B_XLAT_OFFSET is a 64bit register, but can
			 * only take 32bit writes
			 */
			writel(SNB_MBAR01_USD_ADDR & 0xffffffff,
			       ndev->reg_base + SNB_B2B_XLAT_OFFSETL);
			writel(SNB_MBAR01_USD_ADDR >> 32,
			       ndev->reg_base + SNB_B2B_XLAT_OFFSETU);
		}
		writeq(SNB_MBAR01_DSD_ADDR, ndev->reg_base +
		       SNB_SBAR0BASE_OFFSET);
		writeq(SNB_MBAR23_DSD_ADDR, ndev->reg_base +
		       SNB_SBAR2BASE_OFFSET);
		writeq(SNB_MBAR45_DSD_ADDR, ndev->reg_base +
		       SNB_SBAR4BASE_OFFSET);
	}

	ndev->limits.max_spads = SNB_MAX_B2B_SPADS;
	ndev->limits.max_db_bits = SNB_MAX_DB_BITS;
	ndev->limits.msix_cnt = SNB_MSIX_CNT;
	ndev->bits_per_vector = SNB_DB_BITS_PER_VEC;

	return 0;
}

static int ntb_bwd_setup(struct ntb_device *ndev)
{
	int rc;
	u32 val;

	ndev->hw_type = BWD_HW;

	rc = pci_read_config_dword(ndev->pdev, NTB_PPD_OFFSET, &val);
	if (rc)
		return rc;

	switch ((val & BWD_PPD_CONN_TYPE) >> 8) {
	case NTB_CONN_B2B:
		ndev->conn_type = NTB_CONN_B2B;
		break;
	case NTB_CONN_RP:
	default:
		dev_err(&ndev->pdev->dev, "Only B2B supported at this time\n");
		return -EINVAL;
	}

	if (val & BWD_PPD_DEV_TYPE)
		ndev->dev_type = NTB_DEV_DSD;
	else
		ndev->dev_type = NTB_DEV_USD;

	/* Initiate PCI-E link training */
	rc = pci_write_config_dword(ndev->pdev, NTB_PPD_OFFSET,
				    val | BWD_PPD_INIT_LINK);
	if (rc)
		return rc;

	ndev->reg_ofs.pdb = ndev->reg_base + BWD_PDOORBELL_OFFSET;
	ndev->reg_ofs.pdb_mask = ndev->reg_base + BWD_PDBMSK_OFFSET;
	ndev->reg_ofs.sbar2_xlat = ndev->reg_base + BWD_SBAR2XLAT_OFFSET;
	ndev->reg_ofs.sbar4_xlat = ndev->reg_base + BWD_SBAR4XLAT_OFFSET;
	ndev->reg_ofs.lnk_cntl = ndev->reg_base + BWD_NTBCNTL_OFFSET;
	ndev->reg_ofs.lnk_stat = ndev->reg_base + BWD_LINK_STATUS_OFFSET;
	ndev->reg_ofs.spad_read = ndev->reg_base + BWD_SPAD_OFFSET;
	ndev->reg_ofs.spci_cmd = ndev->reg_base + BWD_PCICMD_OFFSET;

	if (ndev->conn_type == NTB_CONN_B2B) {
		ndev->reg_ofs.sdb = ndev->reg_base + BWD_B2B_DOORBELL_OFFSET;
		ndev->reg_ofs.spad_write = ndev->reg_base + BWD_B2B_SPAD_OFFSET;
		ndev->limits.max_spads = BWD_MAX_SPADS;
	} else {
		ndev->reg_ofs.sdb = ndev->reg_base + BWD_PDOORBELL_OFFSET;
		ndev->reg_ofs.spad_write = ndev->reg_base + BWD_SPAD_OFFSET;
		ndev->limits.max_spads = BWD_MAX_COMPAT_SPADS;
	}

	ndev->limits.max_mw = BWD_MAX_MW;
	ndev->limits.max_db_bits = BWD_MAX_DB_BITS;
	ndev->limits.msix_cnt = BWD_MSIX_CNT;
	ndev->bits_per_vector = BWD_DB_BITS_PER_VEC;

	/* Since bwd doesn't have a link interrupt, setup a poll timer */
	INIT_DELAYED_WORK(&ndev->hb_timer, bwd_link_poll);
	INIT_DELAYED_WORK(&ndev->lr_timer, bwd_link_recovery);
	schedule_delayed_work(&ndev->hb_timer, NTB_HB_TIMEOUT);

	return 0;
}

static int ntb_device_setup(struct ntb_device *ndev)
{
	int rc;

	switch (ndev->pdev->device) {
	case PCI_DEVICE_ID_INTEL_NTB_SS_JSF:
	case PCI_DEVICE_ID_INTEL_NTB_SS_SNB:
	case PCI_DEVICE_ID_INTEL_NTB_SS_IVT:
	case PCI_DEVICE_ID_INTEL_NTB_SS_HSX:
	case PCI_DEVICE_ID_INTEL_NTB_PS_JSF:
	case PCI_DEVICE_ID_INTEL_NTB_PS_SNB:
	case PCI_DEVICE_ID_INTEL_NTB_PS_IVT:
	case PCI_DEVICE_ID_INTEL_NTB_PS_HSX:
	case PCI_DEVICE_ID_INTEL_NTB_B2B_JSF:
	case PCI_DEVICE_ID_INTEL_NTB_B2B_SNB:
	case PCI_DEVICE_ID_INTEL_NTB_B2B_IVT:
	case PCI_DEVICE_ID_INTEL_NTB_B2B_HSX:
		rc = ntb_xeon_setup(ndev);
		break;
	case PCI_DEVICE_ID_INTEL_NTB_B2B_BWD:
		rc = ntb_bwd_setup(ndev);
		break;
	default:
		rc = -ENODEV;
	}

	if (rc)
		return rc;

	dev_info(&ndev->pdev->dev, "Device Type = %s\n",
		 ndev->dev_type == NTB_DEV_USD ? "USD/DSP" : "DSD/USP");

	/* Enable Bus Master and Memory Space on the secondary side */
	writew(PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER, ndev->reg_ofs.spci_cmd);

	return 0;
}

static void ntb_device_free(struct ntb_device *ndev)
{
	if (ndev->hw_type == BWD_HW) {
		cancel_delayed_work_sync(&ndev->hb_timer);
		cancel_delayed_work_sync(&ndev->lr_timer);
	}
}

static irqreturn_t bwd_callback_msix_irq(int irq, void *data)
{
	struct ntb_db_cb *db_cb = data;
	struct ntb_device *ndev = db_cb->ndev;

	dev_dbg(&ndev->pdev->dev, "MSI-X irq %d received for DB %d\n", irq,
		db_cb->db_num);

	if (db_cb->callback)
		db_cb->callback(db_cb->data, db_cb->db_num);

	/* No need to check for the specific HB irq, any interrupt means
	 * we're connected.
	 */
	ndev->last_ts = jiffies;

	writeq((u64) 1 << db_cb->db_num, ndev->reg_ofs.pdb);

	return IRQ_HANDLED;
}

static irqreturn_t xeon_callback_msix_irq(int irq, void *data)
{
	struct ntb_db_cb *db_cb = data;
	struct ntb_device *ndev = db_cb->ndev;

	dev_dbg(&ndev->pdev->dev, "MSI-X irq %d received for DB %d\n", irq,
		db_cb->db_num);

	if (db_cb->callback)
		db_cb->callback(db_cb->data, db_cb->db_num);

	/* On Sandybridge, there are 16 bits in the interrupt register
	 * but only 4 vectors.  So, 5 bits are assigned to the first 3
	 * vectors, with the 4th having a single bit for link
	 * interrupts.
	 */
	writew(((1 << ndev->bits_per_vector) - 1) <<
	       (db_cb->db_num * ndev->bits_per_vector), ndev->reg_ofs.pdb);

	return IRQ_HANDLED;
}

/* Since we do not have a HW doorbell in BWD, this is only used in JF/JT */
static irqreturn_t xeon_event_msix_irq(int irq, void *dev)
{
	struct ntb_device *ndev = dev;
	int rc;

	dev_dbg(&ndev->pdev->dev, "MSI-X irq %d received for Events\n", irq);

	rc = ntb_link_status(ndev);
	if (rc)
		dev_err(&ndev->pdev->dev, "Error determining link status\n");

	/* bit 15 is always the link bit */
	writew(1 << ndev->limits.max_db_bits, ndev->reg_ofs.pdb);

	return IRQ_HANDLED;
}

static irqreturn_t ntb_interrupt(int irq, void *dev)
{
	struct ntb_device *ndev = dev;
	unsigned int i = 0;

	if (ndev->hw_type == BWD_HW) {
		u64 pdb = readq(ndev->reg_ofs.pdb);

		dev_dbg(&ndev->pdev->dev, "irq %d - pdb = %Lx\n", irq, pdb);

		while (pdb) {
			i = __ffs(pdb);
			pdb &= pdb - 1;
			bwd_callback_msix_irq(irq, &ndev->db_cb[i]);
		}
	} else {
		u16 pdb = readw(ndev->reg_ofs.pdb);

		dev_dbg(&ndev->pdev->dev, "irq %d - pdb = %x sdb %x\n", irq,
			pdb, readw(ndev->reg_ofs.sdb));

		if (pdb & SNB_DB_HW_LINK) {
			xeon_event_msix_irq(irq, dev);
			pdb &= ~SNB_DB_HW_LINK;
		}

		while (pdb) {
			i = __ffs(pdb);
			pdb &= pdb - 1;
			xeon_callback_msix_irq(irq, &ndev->db_cb[i]);
		}
	}

	return IRQ_HANDLED;
}

static int ntb_setup_msix(struct ntb_device *ndev)
{
	struct pci_dev *pdev = ndev->pdev;
	struct msix_entry *msix;
	int msix_entries;
	int rc, i, pos;
	u16 val;

	pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
	if (!pos) {
		rc = -EIO;
		goto err;
	}

	rc = pci_read_config_word(pdev, pos + PCI_MSIX_FLAGS, &val);
	if (rc)
		goto err;

	msix_entries = msix_table_size(val);
	if (msix_entries > ndev->limits.msix_cnt) {
		rc = -EINVAL;
		goto err;
	}

	ndev->msix_entries = kmalloc(sizeof(struct msix_entry) * msix_entries,
				     GFP_KERNEL);
	if (!ndev->msix_entries) {
		rc = -ENOMEM;
		goto err;
	}

	for (i = 0; i < msix_entries; i++)
		ndev->msix_entries[i].entry = i;

	rc = pci_enable_msix(pdev, ndev->msix_entries, msix_entries);
	if (rc < 0)
		goto err1;
	if (rc > 0) {
		/* On SNB, the link interrupt is always tied to 4th vector.  If