falcon.c

/****************************************************************************
 * Driver for Solarflare Solarstorm network controllers and boards
 * Copyright 2005-2006 Fen Systems Ltd.
 * Copyright 2006-2009 Solarflare Communications Inc.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 as published
 * by the Free Software Foundation, incorporated herein by reference.
 */

#include <linux/bitops.h>
#include <linux/delay.h>
#include <linux/pci.h>
#include <linux/module.h>
#include <linux/seq_file.h>
#include <linux/i2c.h>
#include <linux/mii.h>
#include <linux/slab.h>
#include "net_driver.h"
#include "bitfield.h"
#include "efx.h"
#include "mac.h"
#include "spi.h"
#include "nic.h"
#include "regs.h"
#include "io.h"
#include "mdio_10g.h"
#include "phy.h"
#include "workarounds.h"

/* Hardware control for SFC4000 (aka Falcon). */

static const unsigned int
/* "Large" EEPROM device: Atmel AT25640 or similar
 * 8 KB, 16-bit address, 32 B write block */
large_eeprom_type = ((13 << SPI_DEV_TYPE_SIZE_LBN)
		     | (2 << SPI_DEV_TYPE_ADDR_LEN_LBN)
		     | (5 << SPI_DEV_TYPE_BLOCK_SIZE_LBN)),
/* Default flash device: Atmel AT25F1024
 * 128 KB, 24-bit address, 32 KB erase block, 256 B write block */
default_flash_type = ((17 << SPI_DEV_TYPE_SIZE_LBN)
		      | (3 << SPI_DEV_TYPE_ADDR_LEN_LBN)
		      | (0x52 << SPI_DEV_TYPE_ERASE_CMD_LBN)
		      | (15 << SPI_DEV_TYPE_ERASE_SIZE_LBN)
		      | (8 << SPI_DEV_TYPE_BLOCK_SIZE_LBN));

/**************************************************************************
 *
 * I2C bus - this is a bit-bashing interface using GPIO pins
 * Note that it uses the output enables to tristate the outputs
 * SDA is the data pin and SCL is the clock
 *
 **************************************************************************
 */
static void falcon_setsda(void *data, int state)
{
	struct efx_nic *efx = (struct efx_nic *)data;
	efx_oword_t reg;

	efx_reado(efx, &reg, FR_AB_GPIO_CTL);
	EFX_SET_OWORD_FIELD(reg, FRF_AB_GPIO3_OEN, !state);
	efx_writeo(efx, &reg, FR_AB_GPIO_CTL);
}

static void falcon_setscl(void *data, int state)
{
	struct efx_nic *efx = (struct efx_nic *)data;
	efx_oword_t reg;

	efx_reado(efx, &reg, FR_AB_GPIO_CTL);
	EFX_SET_OWORD_FIELD(reg, FRF_AB_GPIO0_OEN, !state);
	efx_writeo(efx, &reg, FR_AB_GPIO_CTL);
}

static int falcon_getsda(void *data)
{
	struct efx_nic *efx = (struct efx_nic *)data;
	efx_oword_t reg;

	efx_reado(efx, &reg, FR_AB_GPIO_CTL);
	return EFX_OWORD_FIELD(reg, FRF_AB_GPIO3_IN);
}

static int falcon_getscl(void *data)
{
	struct efx_nic *efx = (struct efx_nic *)data;
	efx_oword_t reg;

	efx_reado(efx, &reg, FR_AB_GPIO_CTL);
	return EFX_OWORD_FIELD(reg, FRF_AB_GPIO0_IN);
}

static struct i2c_algo_bit_data falcon_i2c_bit_operations = {
	.setsda		= falcon_setsda,
	.setscl		= falcon_setscl,
	.getsda		= falcon_getsda,
	.getscl		= falcon_getscl,
	.udelay		= 5,
	/* Wait up to 50 ms for slave to let us pull SCL high */
	.timeout	= DIV_ROUND_UP(HZ, 20),
};

static void falcon_push_irq_moderation(struct efx_channel *channel)
{
	efx_dword_t timer_cmd;
	struct efx_nic *efx = channel->efx;

	/* Set timer register */
	if (channel->irq_moderation) {
		EFX_POPULATE_DWORD_2(timer_cmd,
				     FRF_AB_TC_TIMER_MODE,
				     FFE_BB_TIMER_MODE_INT_HLDOFF,
				     FRF_AB_TC_TIMER_VAL,
				     channel->irq_moderation - 1);
	} else {
		EFX_POPULATE_DWORD_2(timer_cmd,
				     FRF_AB_TC_TIMER_MODE,
				     FFE_BB_TIMER_MODE_DIS,
				     FRF_AB_TC_TIMER_VAL, 0);
	}
	BUILD_BUG_ON(FR_AA_TIMER_COMMAND_KER != FR_BZ_TIMER_COMMAND_P0);
	efx_writed_page_locked(efx, &timer_cmd, FR_BZ_TIMER_COMMAND_P0,
			       channel->channel);
}

static void falcon_deconfigure_mac_wrapper(struct efx_nic *efx);

static void falcon_prepare_flush(struct efx_nic *efx)
{
	falcon_deconfigure_mac_wrapper(efx);

	/* Wait for the tx and rx fifo's to get to the next packet boundary
	 * (~1ms without back-pressure), then to drain the remainder of the
	 * fifo's at data path speeds (negligible), with a healthy margin. */
	msleep(10);
}

/* Acknowledge a legacy interrupt from Falcon
 *
 * This acknowledges a legacy (not MSI) interrupt via INT_ACK_KER_REG.
 *
 * Due to SFC bug 3706 (silicon revision <=A1) reads can be duplicated in the
 * BIU. Interrupt acknowledge is read sensitive so must write instead
 * (then read to ensure the BIU collector is flushed)
 *
 * NB most hardware supports MSI interrupts
 */
inline void falcon_irq_ack_a1(struct efx_nic *efx)
{
	efx_dword_t reg;

	EFX_POPULATE_DWORD_1(reg, FRF_AA_INT_ACK_KER_FIELD, 0xb7eb7e);
	efx_writed(efx, &reg, FR_AA_INT_ACK_KER);
	efx_readd(efx, &reg, FR_AA_WORK_AROUND_BROKEN_PCI_READS);
}


irqreturn_t falcon_legacy_interrupt_a1(int irq, void *dev_id)
{
	struct efx_nic *efx = dev_id;
	efx_oword_t *int_ker = efx->irq_status.addr;
	struct efx_channel *channel;
	int syserr;
	int queues;

	/* Check to see if this is our interrupt.  If it isn't, we
	 * exit without having touched the hardware.
	 */
	if (unlikely(EFX_OWORD_IS_ZERO(*int_ker))) {
		EFX_TRACE(efx, "IRQ %d on CPU %d not for me\n", irq,
			  raw_smp_processor_id());
		return IRQ_NONE;
	}
	efx->last_irq_cpu = raw_smp_processor_id();
	EFX_TRACE(efx, "IRQ %d on CPU %d status " EFX_OWORD_FMT "\n",
		  irq, raw_smp_processor_id(), EFX_OWORD_VAL(*int_ker));

	/* Determine interrupting queues, clear interrupt status
	 * register and acknowledge the device interrupt.
	 */
	BUILD_BUG_ON(FSF_AZ_NET_IVEC_INT_Q_WIDTH > EFX_MAX_CHANNELS);
	queues = EFX_OWORD_FIELD(*int_ker, FSF_AZ_NET_IVEC_INT_Q);

	/* Check to see if we have a serious error condition */
	if (queues & (1U << efx->fatal_irq_level)) {
		syserr = EFX_OWORD_FIELD(*int_ker, FSF_AZ_NET_IVEC_FATAL_INT);
		if (unlikely(syserr))
			return efx_nic_fatal_interrupt(efx);
	}

	EFX_ZERO_OWORD(*int_ker);
	wmb(); /* Ensure the vector is cleared before interrupt ack */
	falcon_irq_ack_a1(efx);

	/* Schedule processing of any interrupting queues */
	channel = &efx->channel[0];
	while (queues) {
		if (queues & 0x01)
			efx_schedule_channel(channel);
		channel++;
		queues >>= 1;
	}

	return IRQ_HANDLED;
}
/**************************************************************************
 *
 * EEPROM/flash
 *
 **************************************************************************
 */

#define FALCON_SPI_MAX_LEN sizeof(efx_oword_t)

static int falcon_spi_poll(struct efx_nic *efx)
{
	efx_oword_t reg;
	efx_reado(efx, &reg, FR_AB_EE_SPI_HCMD);
	return EFX_OWORD_FIELD(reg, FRF_AB_EE_SPI_HCMD_CMD_EN) ? -EBUSY : 0;
}

/* Wait for SPI command completion */
static int falcon_spi_wait(struct efx_nic *efx)
{
	/* Most commands will finish quickly, so we start polling at
	 * very short intervals.  Sometimes the command may have to
	 * wait for VPD or expansion ROM access outside of our
	 * control, so we allow up to 100 ms. */
	unsigned long timeout = jiffies + 1 + DIV_ROUND_UP(HZ, 10);
	int i;

	for (i = 0; i < 10; i++) {
		if (!falcon_spi_poll(efx))
			return 0;
		udelay(10);
	}

	for (;;) {
		if (!falcon_spi_poll(efx))
			return 0;
		if (time_after_eq(jiffies, timeout)) {
			EFX_ERR(efx, "timed out waiting for SPI\n");
			return -ETIMEDOUT;
		}
		schedule_timeout_uninterruptible(1);
	}
}

int falcon_spi_cmd(struct efx_nic *efx, const struct efx_spi_device *spi,
		   unsigned int command, int address,
		   const void *in, void *out, size_t len)
{
	bool addressed = (address >= 0);
	bool reading = (out != NULL);
	efx_oword_t reg;
	int rc;

	/* Input validation */
	if (len > FALCON_SPI_MAX_LEN)
		return -EINVAL;
	BUG_ON(!mutex_is_locked(&efx->spi_lock));

	/* Check that previous command is not still running */
	rc = falcon_spi_poll(efx);
	if (rc)
		return rc;

	/* Program address register, if we have an address */
	if (addressed) {
		EFX_POPULATE_OWORD_1(reg, FRF_AB_EE_SPI_HADR_ADR, address);
		efx_writeo(efx, &reg, FR_AB_EE_SPI_HADR);
	}

	/* Program data register, if we have data */
	if (in != NULL) {
		memcpy(&reg, in, len);
		efx_writeo(efx, &reg, FR_AB_EE_SPI_HDATA);
	}

	/* Issue read/write command */
	EFX_POPULATE_OWORD_7(reg,
			     FRF_AB_EE_SPI_HCMD_CMD_EN, 1,
			     FRF_AB_EE_SPI_HCMD_SF_SEL, spi->device_id,
			     FRF_AB_EE_SPI_HCMD_DABCNT, len,
			     FRF_AB_EE_SPI_HCMD_READ, reading,
			     FRF_AB_EE_SPI_HCMD_DUBCNT, 0,
			     FRF_AB_EE_SPI_HCMD_ADBCNT,
			     (addressed ? spi->addr_len : 0),
			     FRF_AB_EE_SPI_HCMD_ENC, command);
	efx_writeo(efx, &reg, FR_AB_EE_SPI_HCMD);

	/* Wait for read/write to complete */
	rc = falcon_spi_wait(efx);
	if (rc)
		return rc;

	/* Read data */
	if (out != NULL) {
		efx_reado(efx, &reg, FR_AB_EE_SPI_HDATA);
		memcpy(out, &reg, len);
	}

	return 0;
}

static size_t
falcon_spi_write_limit(const struct efx_spi_device *spi, size_t start)
{
	return min(FALCON_SPI_MAX_LEN,
		   (spi->block_size - (start & (spi->block_size - 1))));
}

static inline u8
efx_spi_munge_command(const struct efx_spi_device *spi,
		      const u8 command, const unsigned int address)
{
	return command | (((address >> 8) & spi->munge_address) << 3);
}

/* Wait up to 10 ms for buffered write completion */
int
falcon_spi_wait_write(struct efx_nic *efx, const struct efx_spi_device *spi)
{
	unsigned long timeout = jiffies + 1 + DIV_ROUND_UP(HZ, 100);
	u8 status;
	int rc;

	for (;;) {
		rc = falcon_spi_cmd(efx, spi, SPI_RDSR, -1, NULL,
				    &status, sizeof(status));
		if (rc)
			return rc;
		if (!(status & SPI_STATUS_NRDY))
			return 0;
		if (time_after_eq(jiffies, timeout)) {
			EFX_ERR(efx, "SPI write timeout on device %d"
				" last status=0x%02x\n",
				spi->device_id, status);
			return -ETIMEDOUT;
		}
		schedule_timeout_uninterruptible(1);
	}
}

int falcon_spi_read(struct efx_nic *efx, const struct efx_spi_device *spi,
		    loff_t start, size_t len, size_t *retlen, u8 *buffer)
{
	size_t block_len, pos = 0;
	unsigned int command;
	int rc = 0;

	while (pos < len) {
		block_len = min(len - pos, FALCON_SPI_MAX_LEN);

		command = efx_spi_munge_command(spi, SPI_READ, start + pos);
		rc = falcon_spi_cmd(efx, spi, command, start + pos, NULL,
				    buffer + pos, block_len);
		if (rc)
			break;
		pos += block_len;

		/* Avoid locking up the system */
		cond_resched();
		if (signal_pending(current)) {
			rc = -EINTR;
			break;
		}
	}

	if (retlen)
		*retlen = pos;
	return rc;
}

int
falcon_spi_write(struct efx_nic *efx, const struct efx_spi_device *spi,
		 loff_t start, size_t len, size_t *retlen, const u8 *buffer)
{
	u8 verify_buffer[FALCON_SPI_MAX_LEN];
	size_t block_len, pos = 0;
	unsigned int command;
	int rc = 0;

	while (pos < len) {
		rc = falcon_spi_cmd(efx, spi, SPI_WREN, -1, NULL, NULL, 0);
		if (rc)
			break;

		block_len = min(len - pos,
				falcon_spi_write_limit(spi, start + pos));
		command = efx_spi_munge_command(spi, SPI_WRITE, start + pos);
		rc = falcon_spi_cmd(efx, spi, command, start + pos,
				    buffer + pos, NULL, block_len);
		if (rc)
			break;

		rc = falcon_spi_wait_write(efx, spi);
		if (rc)
			break;

		command = efx_spi_munge_command(spi, SPI_READ, start + pos);
		rc = falcon_spi_cmd(efx, spi, command, start + pos,
				    NULL, verify_buffer, block_len);
		if (memcmp(verify_buffer, buffer + pos, block_len)) {
			rc = -EIO;
			break;
		}

		pos += block_len;

		/* Avoid locking up the system */
		cond_resched();
		if (signal_pending(current)) {
			rc = -EINTR;
			break;
		}
	}

	if (retlen)
		*retlen = pos;
	return rc;
}

/**************************************************************************
 *
 * MAC wrapper
 *
 **************************************************************************
 */

static void falcon_push_multicast_hash(struct efx_nic *efx)
{
	union efx_multicast_hash *mc_hash = &efx->multicast_hash;

	WARN_ON(!mutex_is_locked(&efx->mac_lock));

	efx_writeo(efx, &mc_hash->oword[0], FR_AB_MAC_MC_HASH_REG0);
	efx_writeo(efx, &mc_hash->oword[1], FR_AB_MAC_MC_HASH_REG1);
}

static void falcon_reset_macs(struct efx_nic *efx)
{
	struct falcon_nic_data *nic_data = efx->nic_data;
	efx_oword_t reg, mac_ctrl;
	int count;

	if (efx_nic_rev(efx) < EFX_REV_FALCON_B0) {
		/* It's not safe to use GLB_CTL_REG to reset the
		 * macs, so instead use the internal MAC resets
		 */
		if (!EFX_IS10G(efx)) {
			EFX_POPULATE_OWORD_1(reg, FRF_AB_GM_SW_RST, 1);
			efx_writeo(efx, &reg, FR_AB_GM_CFG1);
			udelay(1000);

			EFX_POPULATE_OWORD_1(reg, FRF_AB_GM_SW_RST, 0);
			efx_writeo(efx, &reg, FR_AB_GM_CFG1);
			udelay(1000);
			return;
		} else {
			EFX_POPULATE_OWORD_1(reg, FRF_AB_XM_CORE_RST, 1);
			efx_writeo(efx, &reg, FR_AB_XM_GLB_CFG);

			for (count = 0; count < 10000; count++) {
				efx_reado(efx, &reg, FR_AB_XM_GLB_CFG);
				if (EFX_OWORD_FIELD(reg, FRF_AB_XM_CORE_RST) ==
				    0)
					return;
				udelay(10);
			}

			EFX_ERR(efx, "timed out waiting for XMAC core reset\n");
		}
	}

	/* Mac stats will fail whist the TX fifo is draining */
	WARN_ON(nic_data->stats_disable_count == 0);

	efx_reado(efx, &mac_ctrl, FR_AB_MAC_CTRL);
	EFX_SET_OWORD_FIELD(mac_ctrl, FRF_BB_TXFIFO_DRAIN_EN, 1);
	efx_writeo(efx, &mac_ctrl, FR_AB_MAC_CTRL);

	efx_reado(efx, &reg, FR_AB_GLB_CTL);
	EFX_SET_OWORD_FIELD(reg, FRF_AB_RST_XGTX, 1);
	EFX_SET_OWORD_FIELD(reg, FRF_AB_RST_XGRX, 1);
	EFX_SET_OWORD_FIELD(reg, FRF_AB_RST_EM, 1);
	efx_writeo(efx, &reg, FR_AB_GLB_CTL);

	count = 0;
	while (1) {
		efx_reado(efx, &reg, FR_AB_GLB_CTL);
		if (!EFX_OWORD_FIELD(reg, FRF_AB_RST_XGTX) &&
		    !EFX_OWORD_FIELD(reg, FRF_AB_RST_XGRX) &&
		    !EFX_OWORD_FIELD(reg, FRF_AB_RST_EM)) {
			EFX_LOG(efx, "Completed MAC reset after %d loops\n",
				count);
			break;
		}
		if (count > 20) {
			EFX_ERR(efx, "MAC reset failed\n");
			break;
		}
		count++;
		udelay(10);
	}

	/* Ensure the correct MAC is selected before statistics
	 * are re-enabled by the caller */
	efx_writeo(efx, &mac_ctrl, FR_AB_MAC_CTRL);

	/* This can run even when the GMAC is selected */
	falcon_setup_xaui(efx);
}

void falcon_drain_tx_fifo(struct efx_nic *efx)
{
	efx_oword_t reg;

	if ((efx_nic_rev(efx) < EFX_REV_FALCON_B0) ||
	    (efx->loopback_mode != LOOPBACK_NONE))
		return;

	efx_reado(efx, &reg, FR_AB_MAC_CTRL);
	/* There is no point in draining more than once */
	if (EFX_OWORD_FIELD(reg, FRF_BB_TXFIFO_DRAIN_EN))
		return;

	falcon_reset_macs(efx);
}

static void falcon_deconfigure_mac_wrapper(struct efx_nic *efx)
{
	efx_oword_t reg;

	if (efx_nic_rev(efx) < EFX_REV_FALCON_B0)
		return;

	/* Isolate the MAC -> RX */
	efx_reado(efx, &reg, FR_AZ_RX_CFG);
	EFX_SET_OWORD_FIELD(reg, FRF_BZ_RX_INGR_EN, 0);
	efx_writeo(efx, &reg, FR_AZ_RX_CFG);

	/* Isolate TX -> MAC */
	falcon_drain_tx_fifo(efx);
}

void falcon_reconfigure_mac_wrapper(struct efx_nic *efx)
{
	struct efx_link_state *link_state = &efx->link_state;
	efx_oword_t reg;
	int link_speed;

	switch (link_state->speed) {
	case 10000: link_speed = 3; break;
	case 1000:  link_speed = 2; break;
	case 100:   link_speed = 1; break;
	default:    link_speed = 0; break;
	}
	/* MAC_LINK_STATUS controls MAC backpressure but doesn't work
	 * as advertised.  Disable to ensure packets are not
	 * indefinitely held and TX queue can be flushed at any point
	 * while the link is down. */
	EFX_POPULATE_OWORD_5(reg,
			     FRF_AB_MAC_XOFF_VAL, 0xffff /* max pause time */,
			     FRF_AB_MAC_BCAD_ACPT, 1,
			     FRF_AB_MAC_UC_PROM, efx->promiscuous,
			     FRF_AB_MAC_LINK_STATUS, 1, /* always set */
			     FRF_AB_MAC_SPEED, link_speed);
	/* On B0, MAC backpressure can be disabled and packets get
	 * discarded. */
	if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
		EFX_SET_OWORD_FIELD(reg, FRF_BB_TXFIFO_DRAIN_EN,
				    !link_state->up);
	}

	efx_writeo(efx, &reg, FR_AB_MAC_CTRL);

	/* Restore the multicast hash registers. */
	falcon_push_multicast_hash(efx);

	efx_reado(efx, &reg, FR_AZ_RX_CFG);
	/* Enable XOFF signal from RX FIFO (we enabled it during NIC
	 * initialisation but it may read back as 0) */
	EFX_SET_OWORD_FIELD(reg, FRF_AZ_RX_XOFF_MAC_EN, 1);
	/* Unisolate the MAC -> RX */
	if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0)
		EFX_SET_OWORD_FIELD(reg, FRF_BZ_RX_INGR_EN, 1);
	efx_writeo(efx, &reg, FR_AZ_RX_CFG);
}

static void falcon_stats_request(struct efx_nic *efx)
{
	struct falcon_nic_data *nic_data = efx->nic_data;
	efx_oword_t reg;

	WARN_ON(nic_data->stats_pending);
	WARN_ON(nic_data->stats_disable_count);

	if (nic_data->stats_dma_done == NULL)
		return;	/* no mac selected */

	*nic_data->stats_dma_done = FALCON_STATS_NOT_DONE;
	nic_data->stats_pending = true;
	wmb(); /* ensure done flag is clear */

	/* Initiate DMA transfer of stats */
	EFX_POPULATE_OWORD_2(reg,
			     FRF_AB_MAC_STAT_DMA_CMD, 1,
			     FRF_AB_MAC_STAT_DMA_ADR,
			     efx->stats_buffer.dma_addr);
	efx_writeo(efx, &reg, FR_AB_MAC_STAT_DMA);

	mod_timer(&nic_data->stats_timer, round_jiffies_up(jiffies + HZ / 2));
}

static void falcon_stats_complete(struct efx_nic *efx)
{
	struct falcon_nic_data *nic_data = efx->nic_data;

	if (!nic_data->stats_pending)
		return;

	nic_data->stats_pending = 0;
	if (*nic_data->stats_dma_done == FALCON_STATS_DONE) {
		rmb(); /* read the done flag before the stats */
		efx->mac_op->update_stats(efx);
	} else {
		EFX_ERR(efx, "timed out waiting for statistics\n");
	}
}

static void falcon_stats_timer_func(unsigned long context)
{
	struct efx_nic *efx = (struct efx_nic *)context;
	struct falcon_nic_data *nic_data = efx->nic_data;

	spin_lock(&efx->stats_lock);

	falcon_stats_complete(efx);
	if (nic_data->stats_disable_count == 0)
		falcon_stats_request(efx);

	spin_unlock(&efx->stats_lock);
}

static void falcon_switch_mac(struct efx_nic *efx);

static bool falcon_loopback_link_poll(struct efx_nic *efx)
{
	struct efx_link_state old_state = efx->link_state;

	WARN_ON(!mutex_is_locked(&efx->mac_lock));
	WARN_ON(!LOOPBACK_INTERNAL(efx));

	efx->link_state.fd = true;
	efx->link_state.fc = efx->wanted_fc;
	efx->link_state.up = true;

	if (efx->loopback_mode == LOOPBACK_GMAC)
		efx->link_state.speed = 1000;
	else
		efx->link_state.speed = 10000;

	return !efx_link_state_equal(&efx->link_state, &old_state);
}

static int falcon_reconfigure_port(struct efx_nic *efx)
{
	int rc;

	WARN_ON(efx_nic_rev(efx) > EFX_REV_FALCON_B0);

	/* Poll the PHY link state *before* reconfiguring it. This means we
	 * will pick up the correct speed (in loopback) to select the correct
	 * MAC.
	 */
	if (LOOPBACK_INTERNAL(efx))
		falcon_loopback_link_poll(efx);
	else
		efx->phy_op->poll(efx);

	falcon_stop_nic_stats(efx);
	falcon_deconfigure_mac_wrapper(efx);

	falcon_switch_mac(efx);

	efx->phy_op->reconfigure(efx);
	rc = efx->mac_op->reconfigure(efx);
	BUG_ON(rc);

	falcon_start_nic_stats(efx);

	/* Synchronise efx->link_state with the kernel */
	efx_link_status_changed(efx);

	return 0;
}

/**************************************************************************
 *
 * PHY access via GMII
 *
 **************************************************************************
 */

/* Wait for GMII access to complete */
static int falcon_gmii_wait(struct efx_nic *efx)
{
	efx_oword_t md_stat;
	int count;

	/* wait upto 50ms - taken max from datasheet */
	for (count = 0; count < 5000; count++) {
		efx_reado(efx, &md_stat, FR_AB_MD_STAT);
		if (EFX_OWORD_FIELD(md_stat, FRF_AB_MD_BSY) == 0) {
			if (EFX_OWORD_FIELD(md_stat, FRF_AB_MD_LNFL) != 0 ||
			    EFX_OWORD_FIELD(md_stat, FRF_AB_MD_BSERR) != 0) {
				EFX_ERR(efx, "error from GMII access "
					EFX_OWORD_FMT"\n",
					EFX_OWORD_VAL(md_stat));
				return -EIO;
			}
			return 0;
		}
		udelay(10);
	}
	EFX_ERR(efx, "timed out waiting for GMII\n");
	return -ETIMEDOUT;
}

/* Write an MDIO register of a PHY connected to Falcon. */
static int falcon_mdio_write(struct net_device *net_dev,
			     int prtad, int devad, u16 addr, u16 value)
{
	struct efx_nic *efx = netdev_priv(net_dev);
	efx_oword_t reg;
	int rc;

	EFX_REGDUMP(efx, "writing MDIO %d register %d.%d with 0x%04x\n",
		    prtad, devad, addr, value);

	mutex_lock(&efx->mdio_lock);

	/* Check MDIO not currently being accessed */
	rc = falcon_gmii_wait(efx);
	if (rc)
		goto out;

	/* Write the address/ID register */
	EFX_POPULATE_OWORD_1(reg, FRF_AB_MD_PHY_ADR, addr);
	efx_writeo(efx, &reg, FR_AB_MD_PHY_ADR);

	EFX_POPULATE_OWORD_2(reg, FRF_AB_MD_PRT_ADR, prtad,
			     FRF_AB_MD_DEV_ADR, devad);
	efx_writeo(efx, &reg, FR_AB_MD_ID);

	/* Write data */
	EFX_POPULATE_OWORD_1(reg, FRF_AB_MD_TXD, value);
	efx_writeo(efx, &reg, FR_AB_MD_TXD);

	EFX_POPULATE_OWORD_2(reg,
			     FRF_AB_MD_WRC, 1,
			     FRF_AB_MD_GC, 0);
	efx_writeo(efx, &reg, FR_AB_MD_CS);

	/* Wait for data to be written */
	rc = falcon_gmii_wait(efx);
	if (rc) {
		/* Abort the write operation */
		EFX_POPULATE_OWORD_2(reg,
				     FRF_AB_MD_WRC, 0,
				     FRF_AB_MD_GC, 1);
		efx_writeo(efx, &reg, FR_AB_MD_CS);
		udelay(10);
	}

out:
	mutex_unlock(&efx->mdio_lock);
	return rc;
}

/* Read an MDIO register of a PHY connected to Falcon. */
static int falcon_mdio_read(struct net_device *net_dev,
			    int prtad, int devad, u16 addr)
{
	struct efx_nic *efx = netdev_priv(net_dev);
	efx_oword_t reg;
	int rc;

	mutex_lock(&efx->mdio_lock);

	/* Check MDIO not currently being accessed */
	rc = falcon_gmii_wait(efx);
	if (rc)
		goto out;

	EFX_POPULATE_OWORD_1(reg, FRF_AB_MD_PHY_ADR, addr);
	efx_writeo(efx, &reg, FR_AB_MD_PHY_ADR);

	EFX_POPULATE_OWORD_2(reg, FRF_AB_MD_PRT_ADR, prtad,
			     FRF_AB_MD_DEV_ADR, devad);
	efx_writeo(efx, &reg, FR_AB_MD_ID);

	/* Request data to be read */
	EFX_POPULATE_OWORD_2(reg, FRF_AB_MD_RDC, 1, FRF_AB_MD_GC, 0);
	efx_writeo(efx, &reg, FR_AB_MD_CS);

	/* Wait for data to become available */
	rc = falcon_gmii_wait(efx);
	if (rc == 0) {
		efx_reado(efx, &reg, FR_AB_MD_RXD);
		rc = EFX_OWORD_FIELD(reg, FRF_AB_MD_RXD);
		EFX_REGDUMP(efx, "read from MDIO %d register %d.%d, got %04x\n",
			    prtad, devad, addr, rc);
	} else {
		/* Abort the read operation */
		EFX_POPULATE_OWORD_2(reg,
				     FRF_AB_MD_RIC, 0,
				     FRF_AB_MD_GC, 1);
		efx_writeo(efx, &reg, FR_AB_MD_CS);

		EFX_LOG(efx, "read from MDIO %d register %d.%d, got error %d\n",
			prtad, devad, addr, rc);
	}

out:
	mutex_unlock(&efx->mdio_lock);
	return rc;
}

static void falcon_clock_mac(struct efx_nic *efx)
{
	unsigned strap_val;
	efx_oword_t nic_stat;

	/* Configure the NIC generated MAC clock correctly */
	efx_reado(efx, &nic_stat, FR_AB_NIC_STAT);
	strap_val = EFX_IS10G(efx) ? 5 : 3;
	if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
		EFX_SET_OWORD_FIELD(nic_stat, FRF_BB_EE_STRAP_EN, 1);
		EFX_SET_OWORD_FIELD(nic_stat, FRF_BB_EE_STRAP, strap_val);
		efx_writeo(efx, &nic_stat, FR_AB_NIC_STAT);
	} else {
		/* Falcon A1 does not support 1G/10G speed switching
		 * and must not be used with a PHY that does. */
		BUG_ON(EFX_OWORD_FIELD(nic_stat, FRF_AB_STRAP_PINS) !=
		       strap_val);
	}
}

static void falcon_switch_mac(struct efx_nic *efx)
{
	struct efx_mac_operations *old_mac_op = efx->mac_op;
	struct falcon_nic_data *nic_data = efx->nic_data;
	unsigned int stats_done_offset;

	WARN_ON(!mutex_is_locked(&efx->mac_lock));
	WARN_ON(nic_data->stats_disable_count == 0);

	efx->mac_op = (EFX_IS10G(efx) ?
		       &falcon_xmac_operations : &falcon_gmac_operations);

	if (EFX_IS10G(efx))
		stats_done_offset = XgDmaDone_offset;
	else
		stats_done_offset = GDmaDone_offset;
	nic_data->stats_dma_done = efx->stats_buffer.addr + stats_done_offset;

	if (old_mac_op == efx->mac_op)
		return;

	falcon_clock_mac(efx);

	EFX_LOG(efx, "selected %cMAC\n", EFX_IS10G(efx) ? 'X' : 'G');
	/* Not all macs support a mac-level link state */
	efx->xmac_poll_required = false;
	falcon_reset_macs(efx);
}

/* This call is responsible for hooking in the MAC and PHY operations */
static int falcon_probe_port(struct efx_nic *efx)
{
	int rc;

	switch (efx->phy_type) {
	case PHY_TYPE_SFX7101:
		efx->phy_op = &falcon_sfx7101_phy_ops;
		break;
	case PHY_TYPE_SFT9001A:
	case PHY_TYPE_SFT9001B:
		efx->phy_op = &falcon_sft9001_phy_ops;
		break;
	case PHY_TYPE_QT2022C2:
	case PHY_TYPE_QT2025C:
		efx->phy_op = &falcon_qt202x_phy_ops;
		break;
	default:
		EFX_ERR(efx, "Unknown PHY type %d\n",
			efx->phy_type);
		return -ENODEV;
	}

	/* Fill out MDIO structure and loopback modes */
	efx->mdio.mdio_read = falcon_mdio_read;
	efx->mdio.mdio_write = falcon_mdio_write;
	rc = efx->phy_op->probe(efx);
	if (rc != 0)
		return rc;

	/* Initial assumption */
	efx->link_state.speed = 10000;
	efx->link_state.fd = true;

	/* Hardware flow ctrl. FalconA RX FIFO too small for pause generation */
	if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0)
		efx->wanted_fc = EFX_FC_RX | EFX_FC_TX;
	else
		efx->wanted_fc = EFX_FC_RX;
	if (efx->mdio.mmds & MDIO_DEVS_AN)
		efx->wanted_fc |= EFX_FC_AUTO;

	/* Allocate buffer for stats */
	rc = efx_nic_alloc_buffer(efx, &efx->stats_buffer,
				  FALCON_MAC_STATS_SIZE);
	if (rc)
		return rc;
	EFX_LOG(efx, "stats buffer at %llx (virt %p phys %llx)\n",
		(u64)efx->stats_buffer.dma_addr,
		efx->stats_buffer.addr,
		(u64)virt_to_phys(efx->stats_buffer.addr));

	return 0;
}

static void falcon_remove_port(struct efx_nic *efx)
{
	efx->phy_op->remove(efx);
	efx_nic_free_buffer(efx, &efx->stats_buffer);
}

/**************************************************************************
 *
 * Falcon test code
 *
 **************************************************************************/

static int
falcon_read_nvram(struct efx_nic *efx, struct falcon_nvconfig *nvconfig_out)
{
	struct falcon_nvconfig *nvconfig;
	struct efx_spi_device *spi;
	void *region;
	int rc, magic_num, struct_ver;
	__le16 *word, *limit;
	u32 csum;

	spi = efx->spi_flash ? efx->spi_flash : efx->spi_eeprom;
	if (!spi)
		return -EINVAL;

	region = kmalloc(FALCON_NVCONFIG_END, GFP_KERNEL);
	if (!region)
		return -ENOMEM;
	nvconfig = region + FALCON_NVCONFIG_OFFSET;

	mutex_lock(&efx->spi_lock);
	rc = falcon_spi_read(efx, spi, 0, FALCON_NVCONFIG_END, NULL, region);
	mutex_unlock(&efx->spi_lock);
	if (rc) {
		EFX_ERR(efx, "Failed to read %s\n",
			efx->spi_flash ? "flash" : "EEPROM");
		rc = -EIO;
		goto out;
	}

	magic_num = le16_to_cpu(nvconfig->board_magic_num);
	struct_ver = le16_to_cpu(nvconfig->board_struct_ver);

	rc = -EINVAL;
	if (magic_num != FALCON_NVCONFIG_BOARD_MAGIC_NUM) {
		EFX_ERR(efx, "NVRAM bad magic 0x%x\n", magic_num);
		goto out;
	}
	if (struct_ver < 2) {
		EFX_ERR(efx, "NVRAM has ancient version 0x%x\n", struct_ver);
		goto out;
	} else if (struct_ver < 4) {
		word = &nvconfig->board_magic_num;
		limit = (__le16 *) (nvconfig + 1);
	} else {
		word = region;
		limit = region + FALCON_NVCONFIG_END;
	}
	for (csum = 0; word < limit; ++word)
		csum += le16_to_cpu(*word);

	if (~csum & 0xffff) {
		EFX_ERR(efx, "NVRAM has incorrect checksum\n");
		goto out;
	}

	rc = 0;
	if (nvconfig_out)
		memcpy(nvconfig_out, nvconfig, sizeof(*nvconfig));

 out:
	kfree(region);
	return rc;
}

static int falcon_test_nvram(struct efx_nic *efx)
{
	return falcon_read_nvram(efx, NULL);
}

static const struct efx_nic_register_test falcon_b0_register_tests[] = {
	{ FR_AZ_ADR_REGION,
	  EFX_OWORD32(0x0003FFFF, 0x0003FFFF, 0x0003FFFF, 0x0003FFFF) },
	{ FR_AZ_RX_CFG,
	  EFX_OWORD32(0xFFFFFFFE, 0x00017FFF, 0x00000000, 0x00000000) },
	{ FR_AZ_TX_CFG,
	  EFX_OWORD32(0x7FFF0037, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AZ_TX_RESERVED,
	  EFX_OWORD32(0xFFFEFE80, 0x1FFFFFFF, 0x020000FE, 0x007FFFFF) },
	{ FR_AB_MAC_CTRL,
	  EFX_OWORD32(0xFFFF0000, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AZ_SRM_TX_DC_CFG,
	  EFX_OWORD32(0x001FFFFF, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AZ_RX_DC_CFG,
	  EFX_OWORD32(0x0000000F, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AZ_RX_DC_PF_WM,
	  EFX_OWORD32(0x000003FF, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_BZ_DP_CTRL,
	  EFX_OWORD32(0x00000FFF, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AB_GM_CFG2,
	  EFX_OWORD32(0x00007337, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AB_GMF_CFG0,
	  EFX_OWORD32(0x00001F1F, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AB_XM_GLB_CFG,
	  EFX_OWORD32(0x00000C68, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AB_XM_TX_CFG,
	  EFX_OWORD32(0x00080164, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AB_XM_RX_CFG,
	  EFX_OWORD32(0x07100A0C, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AB_XM_RX_PARAM,
	  EFX_OWORD32(0x00001FF8, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AB_XM_FC,
	  EFX_OWORD32(0xFFFF0001, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AB_XM_ADR_LO,
	  EFX_OWORD32(0xFFFFFFFF, 0x00000000, 0x00000000, 0x00000000) },
	{ FR_AB_XX_SD_CTL,
	  EFX_OWORD32(0x0003FF0F, 0x00000000, 0x00000000, 0x00000000) },
};

static int falcon_b0_test_registers(struct efx_nic *efx)
{
	return efx_nic_test_registers(efx, falcon_b0_register_tests,
				      ARRAY_SIZE(falcon_b0_register_tests));
}

/**************************************************************************
 *
 * Device reset
 *
 **************************************************************************
 */

/* Resets NIC to known state.  This routine must be called in process
 * context and is allowed to sleep. */
static int falcon_reset_hw(struct efx_nic *efx, enum reset_type method)
{
	struct falcon_nic_data *nic_data = efx->nic_data;
	efx_oword_t glb_ctl_reg_ker;
	int rc;

	EFX_LOG(efx, "performing %s hardware reset\n", RESET_TYPE(method));

	/* Initiate device reset */
	if (method == RESET_TYPE_WORLD) {
		rc = pci_save_state(efx->pci_dev);
		if (rc) {
			EFX_ERR(efx, "failed to backup PCI state of primary "
				"function prior to hardware reset\n");
			goto fail1;
		}
		if (efx_nic_is_dual_func(efx)) {
			rc = pci_save_state(nic_data->pci_dev2);
			if (rc) {
				EFX_ERR(efx, "failed to backup PCI state of "
					"secondary function prior to "
					"hardware reset\n");
				goto fail2;
			}
		}

		EFX_POPULATE_OWORD_2(glb_ctl_reg_ker,
				     FRF_AB_EXT_PHY_RST_DUR,
				     FFE_AB_EXT_PHY_RST_DUR_10240US,
				     FRF_AB_SWRST, 1);
	} else {
		EFX_POPULATE_OWORD_7(glb_ctl_reg_ker,
				     /* exclude PHY from "invisible" reset */
				     FRF_AB_EXT_PHY_RST_CTL,
				     method == RESET_TYPE_INVISIBLE,
				     /* exclude EEPROM/flash and PCIe */
				     FRF_AB_PCIE_CORE_RST_CTL, 1,
				     FRF_AB_PCIE_NSTKY_RST_CTL, 1,
				     FRF_AB_PCIE_SD_RST_CTL, 1,
				     FRF_AB_EE_RST_CTL, 1,
				     FRF_AB_EXT_PHY_RST_DUR,
				     FFE_AB_EXT_PHY_RST_DUR_10240US,
				     FRF_AB_SWRST, 1);
	}
	efx_writeo(efx, &glb_ctl_reg_ker, FR_AB_GLB_CTL);

	EFX_LOG(efx, "waiting for hardware reset\n");
	schedule_timeout_uninterruptible(HZ / 20);

	/* Restore PCI configuration if needed */
	if (method == RESET_TYPE_WORLD) {
		if (efx_nic_is_dual_func(efx)) {
			rc = pci_restore_state(nic_data->pci_dev2);
			if (rc) {
				EFX_ERR(efx, "failed to restore PCI config for "
					"the secondary function\n");
				goto fail3;
			}
		}
		rc = pci_restore_state(efx->pci_dev);
		if (rc) {
			EFX_ERR(efx, "failed to restore PCI config for the "
				"primary function\n");
			goto fail4;
		}
		EFX_LOG(efx, "successfully restored PCI config\n");
	}

	/* Assert that reset complete */
	efx_reado(efx, &glb_ctl_reg_ker, FR_AB_GLB_CTL);
	if (EFX_OWORD_FIELD(glb_ctl_reg_ker, FRF_AB_SWRST) != 0) {
		rc = -ETIMEDOUT;
		EFX_ERR(efx, "timed out waiting for hardware reset\n");
		goto fail5;
	}
	EFX_LOG(efx, "hardware reset complete\n");

	return 0;

	/* pci_save_state() and pci_restore_state() MUST be called in pairs */
fail2:
fail3:
	pci_restore_state(efx->pci_dev);
fail1:
fail4:
fail5:
	return rc;
}

static void falcon_monitor(struct efx_nic *efx)
{
	bool link_changed;
	int rc;

	BUG_ON(!mutex_is_locked(&efx->mac_lock));

	rc = falcon_board(efx)->type->monitor(efx);
	if (rc) {
		EFX_ERR(efx, "Board sensor %s; shutting down PHY\n",
			(rc == -ERANGE) ? "reported fault" : "failed");
		efx->phy_mode |= PHY_MODE_LOW_POWER;
		rc = __efx_reconfigure_port(efx);
		WARN_ON(rc);
	}

	if (LOOPBACK_INTERNAL(efx))
		link_changed = falcon_loopback_link_poll(efx);
	else
		link_changed = efx->phy_op->poll(efx);

	if (link_changed) {
		falcon_stop_nic_stats(efx);
		falcon_deconfigure_mac_wrapper(efx);

		falcon_switch_mac(efx);
		rc = efx->mac_op->reconfigure(efx);
		BUG_ON(rc);

		falcon_start_nic_stats(efx);

		efx_link_status_changed(efx);
	}

	if (EFX_IS10G(efx))
		falcon_poll_xmac(efx);
}

/* Zeroes out the SRAM contents.  This routine must be called in
 * process context and is allowed to sleep.
 */
static int falcon_reset_sram(struct efx_nic *efx)
{
	efx_oword_t srm_cfg_reg_ker, gpio_cfg_reg_ker;
	int count;

	/* Set the SRAM wake/sleep GPIO appropriately. */
	efx_reado(efx, &gpio_cfg_reg_ker, FR_AB_GPIO_CTL);
	EFX_SET_OWORD_FIELD(gpio_cfg_reg_ker, FRF_AB_GPIO1_OEN, 1);
	EFX_SET_OWORD_FIELD(gpio_cfg_reg_ker, FRF_AB_GPIO1_OUT, 1);
	efx_writeo(efx, &gpio_cfg_reg_ker, FR_AB_GPIO_CTL);

	/* Initiate SRAM reset */
	EFX_POPULATE_OWORD_2(srm_cfg_reg_ker,
			     FRF_AZ_SRM_INIT_EN, 1,
			     FRF_AZ_SRM_NB_SZ, 0);
	efx_writeo(efx, &srm_cfg_reg_ker, FR_AZ_SRM_CFG);

	/* Wait for SRAM reset to complete */
	count = 0;
	do {
		EFX_LOG(efx, "waiting for SRAM reset (attempt %d)...\n", count);

		/* SRAM reset is slow; expect around 16ms */
		schedule_timeout_uninterruptible(HZ / 50);

		/* Check for reset complete */
		efx_reado(efx, &srm_cfg_reg_ker, FR_AZ_SRM_CFG);
		if (!EFX_OWORD_FIELD(srm_cfg_reg_ker, FRF_AZ_SRM_INIT_EN)) {
			EFX_LOG(efx, "SRAM reset complete\n");

			return 0;
		}
	} while (++count < 20);	/* wait upto 0.4 sec */

	EFX_ERR(efx, "timed out waiting for SRAM reset\n");
	return -ETIMEDOUT;
}

static int falcon_spi_device_init(struct efx_nic *efx,
				  struct efx_spi_device **spi_device_ret,
				  unsigned int device_id, u32 device_type)
{
	struct efx_spi_device *spi_device;

	if (device_type != 0) {
		spi_device = kzalloc(sizeof(*spi_device), GFP_KERNEL);
		if (!spi_device)
			return -ENOMEM;
		spi_device->device_id = device_id;
		spi_device->size =
			1 << SPI_DEV_TYPE_FIELD(device_type, SPI_DEV_TYPE_SIZE);
		spi_device->addr_len =
			SPI_DEV_TYPE_FIELD(device_type, SPI_DEV_TYPE_ADDR_LEN);
		spi_device->munge_address = (spi_device->size == 1 << 9 &&
					     spi_device->addr_len == 1);
		spi_device->erase_command =
			SPI_DEV_TYPE_FIELD(device_type, SPI_DEV_TYPE_ERASE_CMD);
		spi_device->erase_size =
			1 << SPI_DEV_TYPE_FIELD(device_type,
						SPI_DEV_TYPE_ERASE_SIZE);
		spi_device->block_size =
			1 << SPI_DEV_TYPE_FIELD(device_type,
						SPI_DEV_TYPE_BLOCK_SIZE);
	} else {
		spi_device = NULL;
	}

	kfree(*spi_device_ret);
	*spi_device_ret = spi_device;
	return 0;
}

static void falcon_remove_spi_devices(struct efx_nic *efx)
{
	kfree(efx->spi_eeprom);
	efx->spi_eeprom = NULL;
	kfree(efx->spi_flash);
	efx->spi_flash = NULL;
}

/* Extract non-volatile configuration */
static int falcon_probe_nvconfig(struct efx_nic *efx)
{
	struct falcon_nvconfig *nvconfig;
	int board_rev;
	int rc;

	nvconfig = kmalloc(sizeof(*nvconfig), GFP_KERNEL);
	if (!nvconfig)
		return -ENOMEM;

	rc = falcon_read_nvram(efx, nvconfig);
	if (rc == -EINVAL) {
		EFX_ERR(efx, "NVRAM is invalid therefore using defaults\n");
		efx->phy_type = PHY_TYPE_NONE;
		efx->mdio.prtad = MDIO_PRTAD_NONE;
		board_rev = 0;
		rc = 0;
	} else if (rc) {
		goto fail1;
	} else {
		struct falcon_nvconfig_board_v2 *v2 = &nvconfig->board_v2;
		struct falcon_nvconfig_board_v3 *v3 = &nvconfig->board_v3;

		efx->phy_type = v2->port0_phy_type;
		efx->mdio.prtad = v2->port0_phy_addr;
		board_rev = le16_to_cpu(v2->board_revision);

		if (le16_to_cpu(nvconfig->board_struct_ver) >= 3) {
			rc = falcon_spi_device_init(
				efx, &efx->spi_flash, FFE_AB_SPI_DEVICE_FLASH,
				le32_to_cpu(v3->spi_device_type
					    [FFE_AB_SPI_DEVICE_FLASH]));
			if (rc)
				goto fail2;
			rc = falcon_spi_device_init(
				efx, &efx->spi_eeprom, FFE_AB_SPI_DEVICE_EEPROM,
				le32_to_cpu(v3->spi_device_type
					    [FFE_AB_SPI_DEVICE_EEPROM]));
			if (rc)
				goto fail2;
		}
	}

	/* Read the MAC addresses */
	memcpy(efx->mac_address, nvconfig->mac_address[0], ETH_ALEN);

	EFX_LOG(efx, "PHY is %d phy_id %d\n", efx->phy_type, efx->mdio.prtad);

	rc = falcon_probe_board(efx, board_rev);
	if (rc)
		goto fail2;

	kfree(nvconfig);
	return 0;

 fail2:
	falcon_remove_spi_devices(efx);
 fail1:
	kfree(nvconfig);
	return rc;
}

/* Probe all SPI devices on the NIC */
static void falcon_probe_spi_devices(struct efx_nic *efx)
{
	efx_oword_t nic_stat, gpio_ctl, ee_vpd_cfg;
	int boot_dev;

	efx_reado(efx, &gpio_ctl, FR_AB_GPIO_CTL);
	efx_reado(efx, &nic_stat, FR_AB_NIC_STAT);
	efx_reado(efx, &ee_vpd_cfg, FR_AB_EE_VPD_CFG0);

	if (EFX_OWORD_FIELD(gpio_ctl, FRF_AB_GPIO3_PWRUP_VALUE)) {
		boot_dev = (EFX_OWORD_FIELD(nic_stat, FRF_AB_SF_PRST) ?
			    FFE_AB_SPI_DEVICE_FLASH : FFE_AB_SPI_DEVICE_EEPROM);
		EFX_LOG(efx, "Booted from %s\n",
			boot_dev == FFE_AB_SPI_DEVICE_FLASH ? "flash" : "EEPROM");
	} else {
		/* Disable VPD and set clock dividers to safe
		 * values for initial programming. */
		boot_dev = -1;
		EFX_LOG(efx, "Booted from internal ASIC settings;"
			" setting SPI config\n");
		EFX_POPULATE_OWORD_3(ee_vpd_cfg, FRF_AB_EE_VPD_EN, 0,
				     /* 125 MHz / 7 ~= 20 MHz */
				     FRF_AB_EE_SF_CLOCK_DIV, 7,
				     /* 125 MHz / 63 ~= 2 MHz */
				     FRF_AB_EE_EE_CLOCK_DIV, 63);
		efx_writeo(efx, &ee_vpd_cfg, FR_AB_EE_VPD_CFG0);
	}

	if (boot_dev == FFE_AB_SPI_DEVICE_FLASH)
		falcon_spi_device_init(efx, &efx->spi_flash,
				       FFE_AB_SPI_DEVICE_FLASH,
				       default_flash_type);
	if (boot_dev == FFE_AB_SPI_DEVICE_EEPROM)
		falcon_spi_device_init(efx, &efx->spi_eeprom,
				       FFE_AB_SPI_DEVICE_EEPROM,
				       large_eeprom_type);
}

static int falcon_probe_nic(struct efx_nic *efx)
{
	struct falcon_nic_data *nic_data;
	struct falcon_board *board;
	int rc;

	/* Allocate storage for hardware specific data */
	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
	if (!nic_data)
		return -ENOMEM;
	efx->nic_data = nic_data;

	rc = -ENODEV;

	if (efx_nic_fpga_ver(efx) != 0) {
		EFX_ERR(efx, "Falcon FPGA not supported\n");
		goto fail1;
	}

	if (efx_nic_rev(efx) <= EFX_REV_FALCON_A1) {
		efx_oword_t nic_stat;
		struct pci_dev *dev;
		u8 pci_rev = efx->pci_dev->revision;

		if ((pci_rev == 0xff) || (pci_rev == 0)) {
			EFX_ERR(efx, "Falcon rev A0 not supported\n");
			goto fail1;
		}
		efx_reado(efx, &nic_stat, FR_AB_NIC_STAT);
		if (EFX_OWORD_FIELD(nic_stat, FRF_AB_STRAP_10G) == 0) {
			EFX_ERR(efx, "Falcon rev A1 1G not supported\n");
			goto fail1;
		}
		if (EFX_OWORD_FIELD(nic_stat, FRF_AA_STRAP_PCIE) == 0) {
			EFX_ERR(efx, "Falcon rev A1 PCI-X not supported\n");
			goto fail1;
		}

		dev = pci_dev_get(efx->pci_dev);
		while ((dev = pci_get_device(EFX_VENDID_SFC, FALCON_A_S_DEVID,
					     dev))) {
			if (dev->bus == efx->pci_dev->bus &&
			    dev->devfn == efx->pci_dev->devfn + 1) {
				nic_data->pci_dev2 = dev;
				break;
			}
		}
		if (!nic_data->pci_dev2) {
			EFX_ERR(efx, "failed to find secondary function\n");
			rc = -ENODEV;
			goto fail2;
		}
	}

	/* Now we can reset the NIC */
	rc = falcon_reset_hw(efx, RESET_TYPE_ALL);
	if (rc) {
		EFX_ERR(efx, "failed to reset NIC\n");
		goto fail3;
	}

	/* Allocate memory for INT_KER */
	rc = efx_nic_alloc_buffer(efx, &efx->irq_status, sizeof(efx_oword_t));
	if (rc)
		goto fail4;
	BUG_ON(efx->irq_status.dma_addr & 0x0f);

	EFX_LOG(efx, "INT_KER at %llx (virt %p phys %llx)\n",
		(u64)efx->irq_status.dma_addr,
		efx->irq_status.addr, (u64)virt_to_phys(efx->irq_status.addr));

	falcon_probe_spi_devices(efx);

	/* Read in the non-volatile configuration */
	rc = falcon_probe_nvconfig(efx);
	if (rc)
		goto fail5;

	/* Initialise I2C adapter */
	board = falcon_board(efx);
	board->i2c_adap.owner = THIS_MODULE;
	board->i2c_data = falcon_i2c_bit_operations;
	board->i2c_data.data = efx;
	board->i2c_adap.algo_data = &board->i2c_data;
	board->i2c_adap.dev.parent = &efx->pci_dev->dev;
	strlcpy(board->i2c_adap.name, "SFC4000 GPIO",
		sizeof(board->i2c_adap.name));
	rc = i2c_bit_add_bus(&board->i2c_adap);
	if (rc)
		goto fail5;

	rc = falcon_board(efx)->type->init(efx);
	if (rc) {
		EFX_ERR(efx, "failed to initialise board\n");
		goto fail6;
	}

	nic_data->stats_disable_count = 1;
	setup_timer(&nic_data->stats_timer, &falcon_stats_timer_func,
		    (unsigned long)efx);

	return 0;

 fail6:
	BUG_ON(i2c_del_adapter(&board->i2c_adap));
	memset(&board->i2c_adap, 0, sizeof(board->i2c_adap));
 fail5:
	falcon_remove_spi_devices(efx);
	efx_nic_free_buffer(efx, &efx->irq_status);
 fail4:
 fail3:
	if (nic_data->pci_dev2) {
		pci_dev_put(nic_data->pci_dev2);
		nic_data->pci_dev2 = NULL;
	}
 fail2:
 fail1:
	kfree(efx->nic_data);
	return rc;
}

static void falcon_init_rx_cfg(struct efx_nic *efx)
{
	/* Prior to Siena the RX DMA engine will split each frame at
	 * intervals of RX_USR_BUF_SIZE (32-byte units). We set it to
	 * be so large that that never happens. */
	const unsigned huge_buf_size = (3 * 4096) >> 5;
	/* RX control FIFO thresholds (32 entries) */
	const unsigned ctrl_xon_thr = 20;
	const unsigned ctrl_xoff_thr = 25;
	/* RX data FIFO thresholds (256-byte units; size varies) */
	int data_xon_thr = efx_nic_rx_xon_thresh >> 8;
	int data_xoff_thr = efx_nic_rx_xoff_thresh >> 8;
	efx_oword_t reg;

	efx_reado(efx, &reg, FR_AZ_RX_CFG);
	if (efx_nic_rev(efx) <= EFX_REV_FALCON_A1) {
		/* Data FIFO size is 5.5K */
		if (data_xon_thr < 0)
			data_xon_thr = 512 >> 8;
		if (data_xoff_thr < 0)
			data_xoff_thr = 2048 >> 8;
		EFX_SET_OWORD_FIELD(reg, FRF_AA_RX_DESC_PUSH_EN, 0);
		EFX_SET_OWORD_FIELD(reg, FRF_AA_RX_USR_BUF_SIZE,
				    huge_buf_size);
		EFX_SET_OWORD_FIELD(reg, FRF_AA_RX_XON_MAC_TH, data_xon_thr);
		EFX_SET_OWORD_FIELD(reg, FRF_AA_RX_XOFF_MAC_TH, data_xoff_thr);
		EFX_SET_OWORD_FIELD(reg, FRF_AA_RX_XON_TX_TH, ctrl_xon_thr);
		EFX_SET_OWORD_FIELD(reg, FRF_AA_RX_XOFF_TX_TH, ctrl_xoff_thr);
	} else {
		/* Data FIFO size is 80K; register fields moved */
		if (data_xon_thr < 0)
			data_xon_thr = 27648 >> 8; /* ~3*max MTU */
		if (data_xoff_thr < 0)
			data_xoff_thr = 54272 >> 8; /* ~80Kb - 3*max MTU */
		EFX_SET_OWORD_FIELD(reg, FRF_BZ_RX_DESC_PUSH_EN, 0);
		EFX_SET_OWORD_FIELD(reg, FRF_BZ_RX_USR_BUF_SIZE,
				    huge_buf_size);
		EFX_SET_OWORD_FIELD(reg, FRF_BZ_RX_XON_MAC_TH, data_xon_thr);
		EFX_SET_OWORD_FIELD(reg, FRF_BZ_RX_XOFF_MAC_TH, data_xoff_thr);
		EFX_SET_OWORD_FIELD(reg, FRF_BZ_RX_XON_TX_TH, ctrl_xon_thr);
		EFX_SET_OWORD_FIELD(reg, FRF_BZ_RX_XOFF_TX_TH, ctrl_xoff_thr);
		EFX_SET_OWORD_FIELD(reg, FRF_BZ_RX_INGR_EN, 1);
	}
	/* Always enable XOFF signal from RX FIFO.  We enable
	 * or disable transmission of pause frames at the MAC. */
	EFX_SET_OWORD_FIELD(reg, FRF_AZ_RX_XOFF_MAC_EN, 1);
	efx_writeo(efx, &reg, FR_AZ_RX_CFG);
}

/* This call performs hardware-specific global initialisation, such as
 * defining the descriptor cache sizes and number of RSS channels.
 * It does not set up any buffers, descriptor rings or event queues.
 */
static int falcon_init_nic(struct efx_nic *efx)
{
	efx_oword_t temp;
	int rc;

	/* Use on-chip SRAM */
	efx_reado(efx, &temp, FR_AB_NIC_STAT);
	EFX_SET_OWORD_FIELD(temp, FRF_AB_ONCHIP_SRAM, 1);
	efx_writeo(efx, &temp, FR_AB_NIC_STAT);

	/* Set the source of the GMAC clock */
	if (efx_nic_rev(efx) == EFX_REV_FALCON_B0) {
		efx_reado(efx, &temp, FR_AB_GPIO_CTL);
		EFX_SET_OWORD_FIELD(temp, FRF_AB_USE_NIC_CLK, true);
		efx_writeo(efx, &temp, FR_AB_GPIO_CTL);
	}

	/* Select the correct MAC */
	falcon_clock_mac(efx);

	rc = falcon_reset_sram(efx);
	if (rc)
		return rc;

	/* Clear the parity enables on the TX data fifos as
	 * they produce false parity errors because of timing issues
	 */
	if (EFX_WORKAROUND_5129(efx)) {
		efx_reado(efx, &temp, FR_AZ_CSR_SPARE);
		EFX_SET_OWORD_FIELD(temp, FRF_AB_MEM_PERR_EN_TX_DATA, 0);
		efx_writeo(efx, &temp, FR_AZ_CSR_SPARE);
	}

	if (EFX_WORKAROUND_7244(efx)) {
		efx_reado(efx, &temp, FR_BZ_RX_FILTER_CTL);
		EFX_SET_OWORD_FIELD(temp, FRF_BZ_UDP_FULL_SRCH_LIMIT, 8);
		EFX_SET_OWORD_FIELD(temp, FRF_BZ_UDP_WILD_SRCH_LIMIT, 8);
		EFX_SET_OWORD_FIELD(temp, FRF_BZ_TCP_FULL_SRCH_LIMIT, 8);
		EFX_SET_OWORD_FIELD(temp, FRF_BZ_TCP_WILD_SRCH_LIMIT, 8);
		efx_writeo(efx, &temp, FR_BZ_RX_FILTER_CTL);
	}

	/* XXX This is documented only for Falcon A0/A1 */
	/* Setup RX.  Wait for descriptor is broken and must
	 * be disabled.  RXDP recovery shouldn't be needed, but is.
	 */
	efx_reado(efx, &temp, FR_AA_RX_SELF_RST);
	EFX_SET_OWORD_FIELD(temp, FRF_AA_RX_NODESC_WAIT_DIS, 1);
	EFX_SET_OWORD_FIELD(temp, FRF_AA_RX_SELF_RST_EN, 1);
	if (EFX_WORKAROUND_5583(efx))
		EFX_SET_OWORD_FIELD(temp, FRF_AA_RX_ISCSI_DIS, 1);
	efx_writeo(efx, &temp, FR_AA_RX_SELF_RST);

	/* Do not enable TX_NO_EOP_DISC_EN, since it limits packets to 16
	 * descriptors (which is bad).
	 */
	efx_reado(efx, &temp, FR_AZ_TX_CFG);
	EFX_SET_OWORD_FIELD(temp, FRF_AZ_TX_NO_EOP_DISC_EN, 0);
	efx_writeo(efx, &temp, FR_AZ_TX_CFG);

	falcon_init_rx_cfg(efx);

	/* Set destination of both TX and RX Flush events */
	if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) {
		EFX_POPULATE_OWORD_1(temp, FRF_BZ_FLS_EVQ_ID, 0);
		efx_writeo(efx, &temp, FR_BZ_DP_CTRL);
	}

	efx_nic_init_common(efx);

	return 0;
}

static void falcon_remove_nic(struct efx_nic *efx)
{
	struct falcon_nic_data *nic_data = efx->nic_data;
	struct falcon_board *board = falcon_board(efx);
	int rc;

	board->type->fini(efx);

	/* Remove I2C adapter and clear it in preparation for a retry */
	rc = i2c_del_adapter(&board->i2c_adap);
	BUG_ON(rc);
	memset(&board->i2c_adap, 0, sizeof(board->i2c_adap));

	falcon_remove_spi_devices(efx);
	efx_nic_free_buffer(efx, &efx->irq_status);

	falcon_reset_hw(efx, RESET_TYPE_ALL);

	/* Release the second function after the reset */
	if (nic_data->pci_dev2) {
		pci_dev_put(nic_data->pci_dev2);
		nic_data->pci_dev2 = NULL;
	}

	/* Tear down the private nic state */
	kfree(efx->nic_data);
	efx->nic_data = NULL;
}

static void falcon_update_nic_stats(struct efx_nic *efx)
{
	struct falcon_nic_data *nic_data = efx->nic_data;
	efx_oword_t cnt;

	if (nic_data->stats_disable_count)
		return;

	efx_reado(efx, &cnt, FR_AZ_RX_NODESC_DROP);
	efx->n_rx_nodesc_drop_cnt +=
		EFX_OWORD_FIELD(cnt, FRF_AB_RX_NODESC_DROP_CNT);

	if (nic_data->stats_pending &&
	    *nic_data->stats_dma_done == FALCON_STATS_DONE) {
		nic_data->stats_pending = false;
		rmb(); /* read the done flag before the stats */
		efx->mac_op->update_stats(efx);
	}
}

void falcon_start_nic_stats(struct efx_nic *efx)
{
	struct falcon_nic_data *nic_data = efx->nic_data;

	spin_lock_bh(&efx->stats_lock);
	if (--nic_data->stats_disable_count == 0)
		falcon_stats_request(efx);
	spin_unlock_bh(&efx->stats_lock);
}

void falcon_stop_nic_stats(struct efx_nic *efx)
{
	struct falcon_nic_data *nic_data = efx->nic_data;
	int i;

	might_sleep();

	spin_lock_bh(&efx->stats_lock);
	++nic_data->stats_disable_count;
	spin_unlock_bh(&efx->stats_lock);

	del_timer_sync(&nic_data->stats_timer);

	/* Wait enough time for the most recent transfer to
	 * complete. */
	for (i = 0; i < 4 && nic_data->stats_pending; i++) {
		if (*nic_data->stats_dma_done == FALCON_STATS_DONE)
			break;
		msleep(1);
	}

	spin_lock_bh(&efx->stats_lock);
	falcon_stats_complete(efx);
	spin_unlock_bh(&efx->stats_lock);
}

static void falcon_set_id_led(struct efx_nic *efx, enum efx_led_mode mode)
{
	falcon_board(efx)->type->set_id_led(efx, mode);
}

/**************************************************************************
 *
 * Wake on LAN
 *
 **************************************************************************
 */

static void falcon_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
{
	wol->supported = 0;
	wol->wolopts = 0;
	memset(&wol->sopass, 0, sizeof(wol->sopass));
}

static int falcon_set_wol(struct efx_nic *efx, u32 type)
{
	if (type != 0)
		return -EINVAL;
	return 0;
}

/**************************************************************************
 *
 * Revision-dependent attributes used by efx.c and nic.c
 *
 **************************************************************************
 */

struct efx_nic_type falcon_a1_nic_type = {
	.probe = falcon_probe_nic,
	.remove = falcon_remove_nic,
	.init = falcon_init_nic,
	.fini = efx_port_dummy_op_void,
	.monitor = falcon_monitor,
	.reset = falcon_reset_hw,
	.probe_port = falcon_probe_port,
	.remove_port = falcon_remove_port,
	.prepare_flush = falcon_prepare_flush,
	.update_stats = falcon_update_nic_stats,
	.start_stats = falcon_start_nic_stats,
	.stop_stats = falcon_stop_nic_stats,
	.set_id_led = falcon_set_id_led,
	.push_irq_moderation = falcon_push_irq_moderation,
	.push_multicast_hash = falcon_push_multicast_hash,
	.reconfigure_port = falcon_reconfigure_port,
	.get_wol = falcon_get_wol,
	.set_wol = falcon_set_wol,
	.resume_wol = efx_port_dummy_op_void,
	.test_nvram = falcon_test_nvram,
	.default_mac_ops = &falcon_xmac_operations,

	.revision = EFX_REV_FALCON_A1,
	.mem_map_size = 0x20000,
	.txd_ptr_tbl_base = FR_AA_TX_DESC_PTR_TBL_KER,
	.rxd_ptr_tbl_base = FR_AA_RX_DESC_PTR_TBL_KER,
	.buf_tbl_base = FR_AA_BUF_FULL_TBL_KER,
	.evq_ptr_tbl_base = FR_AA_EVQ_PTR_TBL_KER,
	.evq_rptr_tbl_base = FR_AA_EVQ_RPTR_KER,
	.max_dma_mask = DMA_BIT_MASK(FSF_AZ_TX_KER_BUF_ADDR_WIDTH),
	.rx_buffer_padding = 0x24,
	.max_interrupt_mode = EFX_INT_MODE_MSI,
	.phys_addr_channels = 4,
	.tx_dc_base = 0x130000,
	.rx_dc_base = 0x100000,
	.offload_features = NETIF_F_IP_CSUM,
	.reset_world_flags = ETH_RESET_IRQ,
};

struct efx_nic_type falcon_b0_nic_type = {
	.probe = falcon_probe_nic,
	.remove = falcon_remove_nic,
	.init = falcon_init_nic,
	.fini = efx_port_dummy_op_void,
	.monitor = falcon_monitor,
	.reset = falcon_reset_hw,
	.probe_port = falcon_probe_port,
	.remove_port = falcon_remove_port,
	.prepare_flush = falcon_prepare_flush,
	.update_stats = falcon_update_nic_stats,
	.start_stats = falcon_start_nic_stats,
	.stop_stats = falcon_stop_nic_stats,
	.set_id_led = falcon_set_id_led,
	.push_irq_moderation = falcon_push_irq_moderation,
	.push_multicast_hash = falcon_push_multicast_hash,
	.reconfigure_port = falcon_reconfigure_port,
	.get_wol = falcon_get_wol,
	.set_wol = falcon_set_wol,
	.resume_wol = efx_port_dummy_op_void,
	.test_registers = falcon_b0_test_registers,
	.test_nvram = falcon_test_nvram,
	.default_mac_ops = &falcon_xmac_operations,

	.revision = EFX_REV_FALCON_B0,
	/* Map everything up to and including the RSS indirection
	 * table.  Don't map MSI-X table, MSI-X PBA since Linux
	 * requires that they not be mapped.  */
	.mem_map_size = (FR_BZ_RX_INDIRECTION_TBL +
			 FR_BZ_RX_INDIRECTION_TBL_STEP *
			 FR_BZ_RX_INDIRECTION_TBL_ROWS),
	.txd_ptr_tbl_base = FR_BZ_TX_DESC_PTR_TBL,
	.rxd_ptr_tbl_base = FR_BZ_RX_DESC_PTR_TBL,
	.buf_tbl_base = FR_BZ_BUF_FULL_TBL,
	.evq_ptr_tbl_base = FR_BZ_EVQ_PTR_TBL,
	.evq_rptr_tbl_base = FR_BZ_EVQ_RPTR,
	.max_dma_mask = DMA_BIT_MASK(FSF_AZ_TX_KER_BUF_ADDR_WIDTH),
	.rx_buffer_padding = 0,
	.max_interrupt_mode = EFX_INT_MODE_MSIX,
	.phys_addr_channels = 32, /* Hardware limit is 64, but the legacy
				   * interrupt handler only supports 32
				   * channels */
	.tx_dc_base = 0x130000,
	.rx_dc_base = 0x100000,
	.offload_features = NETIF_F_IP_CSUM,
	.reset_world_flags = ETH_RESET_IRQ,
};