acenic_np.c

该软件根据网络数据生成NetFlow记录。NetFlow可用于网络规划、负载均衡、安全监控等

/******************************************************************************
*                                                                             *
*   Copyright 2005 University of Cambridge Computer Laboratory.               *
*                                                                             *
*   This file is part of Nprobe.                                              *
*                                                                             *
*   Nprobe is free software; you can redistribute it and/or modify            *
*   it under the terms of the GNU General Public License as published by      *
*   the Free Software Foundation; either version 2 of the License, or         *
*   (at your option) any later version.                                       *
*                                                                             *
*   Nprobe is distributed in the hope that it will be useful,                 *
*   but WITHOUT ANY WARRANTY; without even the implied warranty of            *
*   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the             *
*   GNU General Public License for more details.                              *
*                                                                             *
*   You should have received a copy of the GNU General Public License         *
*   along with Nprobe; if not, write to the Free Software                     *
*   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA *
*                                                                             *
******************************************************************************/


/*
 * acenic.c: Linux driver for the Alteon AceNIC Gigabit Ethernet card
 *           and other Tigon based cards.
 *
 * Copyright 1998-2000 by Jes Sorensen, <[e-mail]>.
 *
 * Thanks to Alteon and 3Com for providing hardware and documentation
 * enabling me to write this driver.
 *
 * A mailing list for discussing the use of this driver has been
 * setup, please subscribe to the lists if you have any questions
 * about the driver. Send mail to [e-mail] to
 * see how to subscribe.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * Additional credits:
 *   Pete Wyckoff <[e-mail]>: Initial Linux/Alpha and trace
 *       dump support. The trace dump support has not been
 *       integrated yet however.
 *   Troy Benjegerdes: Big Endian (PPC) patches.
 *   Nate Stahl: Better out of memory handling and stats support.
 *   Aman Singla: Nasty race between interrupt handler and tx code dealing
 *                with 'testing the tx_ret_csm and setting tx_full'
 *   David S. Miller <[e-mail]>: conversion to new PCI dma mapping
 *                                       infrastructure and Sparc support
 *   Pierrick Pinasseau (CERN): For lending me an Ultra 5 to test the
 *                              driver under Linux/Sparc64
 *   Matt Domsch <[e-mail]>: Detect Alteon 1000baseT cards
 *   Chip Salzenberg <[e-mail]>: Fix race condition between tx
 *                                       handler and close() cleanup.
 *   Ken Aaker <[e-mail]>: Correct check for whether
 *                                       memory mapped IO is enabled to
 *                                       make the driver work on RS/6000.
 *   Takayoshi Kouchi <[e-mail]>: Identifying problem
 *                                       where the driver would disable
 *                                       bus master mode if it had to disable
 *                                       write and invalidate.
 *   Stephen Hack <[e-mail]>: Fixed ace_set_mac_addr for little
 *                                       endian systems.
 *   Val Henson <[e-mail]>:    Reset Jumbo skb producer and
 *                                       rx producer index when
 *                                       flushing the Jumbo ring.
 *   Hans Grobler <[e-mail]>:     Memory leak fixes in the
 *                                       driver init path.
 */

#include <linux/config.h>
#include <linux/module.h>
#include <linux/version.h>
#include <linux/types.h>
#include <linux/errno.h>
#include <linux/ioport.h>
#include <linux/pci.h>
#include <linux/kernel.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/skbuff.h>
#include <linux/init.h>
#include <linux/delay.h>
#include <linux/mm.h>
#include <linux/sockios.h>

#ifdef SIOCETHTOOL
#include <linux/ethtool.h>
#endif

#include <net/sock.h>
#include <net/ip.h>

#include <asm/system.h>
#include <asm/io.h>
#include <asm/irq.h>
#include <asm/byteorder.h>
#include <asm/uaccess.h>

#include "probe.h"
//#define NP_dev_kfree_skb(skb) dev_kfree_skb(skb)
//#define NP_dev_alloc_skb(sz) alloc_skb(sz, GFP_ATOMIC)
//#define NP_netif_rx(skb) netif_rx(skb)



#undef INDEX_DEBUG

#ifdef CONFIG_ACENIC_OMIT_TIGON_I
#define ACE_IS_TIGON_I(ap)	0
#else
#define ACE_IS_TIGON_I(ap)	(ap->version == 1)
#endif

#ifndef PCI_VENDOR_ID_ALTEON
#define PCI_VENDOR_ID_ALTEON		0x12ae	
#endif
#ifndef PCI_DEVICE_ID_ALTEON_ACENIC_FIBRE
#define PCI_DEVICE_ID_ALTEON_ACENIC_FIBRE  0x0001
#define PCI_DEVICE_ID_ALTEON_ACENIC_COPPER 0x0002
#endif
#ifndef PCI_DEVICE_ID_3COM_3C985
#define PCI_DEVICE_ID_3COM_3C985	0x0001
#endif
#ifndef PCI_VENDOR_ID_NETGEAR
#define PCI_VENDOR_ID_NETGEAR		0x1385
#define PCI_DEVICE_ID_NETGEAR_GA620	0x620a
#endif
#ifndef PCI_DEVICE_ID_NETGEAR_GA620T
#define PCI_DEVICE_ID_NETGEAR_GA620T	0x630a
#endif


/*
 * Farallon used the DEC vendor ID by mistake and they seem not
 * to care - stinky!
 */
#ifndef PCI_DEVICE_ID_FARALLON_PN9000SX
#define PCI_DEVICE_ID_FARALLON_PN9000SX	0x1a
#endif
#ifndef PCI_VENDOR_ID_SGI
#define PCI_VENDOR_ID_SGI		0x10a9
#endif
#ifndef PCI_DEVICE_ID_SGI_ACENIC
#define PCI_DEVICE_ID_SGI_ACENIC	0x0009
#endif

#if LINUX_VERSION_CODE >= 0x20400
static struct pci_device_id acenic_pci_tbl[] __initdata = {
	{ PCI_VENDOR_ID_ALTEON, PCI_DEVICE_ID_ALTEON_ACENIC_FIBRE,
	  PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, },
	{ PCI_VENDOR_ID_ALTEON, PCI_DEVICE_ID_ALTEON_ACENIC_COPPER,
	  PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, },
	{ PCI_VENDOR_ID_3COM, PCI_DEVICE_ID_3COM_3C985,
	  PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, },
	{ PCI_VENDOR_ID_NETGEAR, PCI_DEVICE_ID_NETGEAR_GA620,
	  PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, },
	{ PCI_VENDOR_ID_NETGEAR, PCI_DEVICE_ID_NETGEAR_GA620T,
	  PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, },
	/*
	 * Farallon used the DEC vendor ID on their cards incorrectly.
	 */
	{ PCI_VENDOR_ID_DEC, PCI_DEVICE_ID_FARALLON_PN9000SX,
	  PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, },
	{ PCI_VENDOR_ID_SGI, PCI_DEVICE_ID_SGI_ACENIC,
	  PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, },
	{ }
};
MODULE_DEVICE_TABLE(pci, acenic_pci_tbl);
#endif


#ifndef wmb
#define wmb()	mb()
#endif

#ifndef __exit
#define __exit
#endif

#ifndef __devinit
#define __devinit	__init
#endif


#ifndef SMP_CACHE_BYTES
#define SMP_CACHE_BYTES	L1_CACHE_BYTES
#endif

#if (BITS_PER_LONG == 64)
#define ACE_64BIT_PTR	1
#endif

#ifndef SET_MODULE_OWNER
#define SET_MODULE_OWNER(dev)		{do{} while(0);}
#define ACE_MOD_INC_USE_COUNT		MOD_INC_USE_COUNT
#define ACE_MOD_DEC_USE_COUNT		MOD_DEC_USE_COUNT
#else
#define ACE_MOD_INC_USE_COUNT		{do{} while(0);}
#define ACE_MOD_DEC_USE_COUNT		{do{} while(0);}
#endif


#if (LINUX_VERSION_CODE < 0x02030d)
#define pci_resource_start(dev, bar)	dev->base_address[bar]
#elif (LINUX_VERSION_CODE < 0x02032c)
#define pci_resource_start(dev, bar)	dev->resource[bar].start
#endif

#if (LINUX_VERSION_CODE < 0x02030e)
#define net_device device
#endif

#if (LINUX_VERSION_CODE < 0x02032a)
typedef u32 dma_addr_t;

static inline void *pci_alloc_consistent(struct pci_dev *hwdev, size_t size,
					 dma_addr_t *dma_handle)
{
	void *virt_ptr;

	virt_ptr = kmalloc(size, GFP_KERNEL);
	if (!virt_ptr)
		return NULL;
	*dma_handle = virt_to_bus(virt_ptr);
	return virt_ptr;
}
#define pci_free_consistent(cookie, size, ptr, dma_ptr)	kfree(ptr)
#define pci_map_single(cookie, address, size, dir)	virt_to_bus(address)
#define pci_unmap_single(cookie, address, size, dir)
#endif

#if (LINUX_VERSION_CODE < 0x02032b)
/*
 * SoftNet
 *
 * For pre-softnet kernels we need to tell the upper layer not to
 * re-enter start_xmit() while we are in there. However softnet
 * guarantees not to enter while we are in there so there is no need
 * to do the netif_stop_queue() dance unless the transmit queue really
 * gets stuck. This should also improve performance according to tests
 * done by Aman Singla.
 */
#define dev_kfree_skb_irq(a)			dev_kfree_skb(a)
#define netif_wake_queue(dev)			clear_bit(0, &dev->tbusy)
#define netif_stop_queue(dev)			set_bit(0, &dev->tbusy)
#define late_stop_netif_stop_queue(dev)		{do{} while(0);}
#define early_stop_netif_stop_queue(dev)	test_and_set_bit(0,&dev->tbusy)
#define early_stop_netif_wake_queue(dev)	netif_wake_queue(dev)

static inline void netif_start_queue(struct net_device *dev)
{
	dev->tbusy = 0;
	dev->interrupt = 0;
	dev->start = 1;
}

#define ace_mark_net_bh()			mark_bh(NET_BH)
#define netif_queue_stopped(dev)		dev->tbusy
#define netif_running(dev)			dev->start
#define ace_if_down(dev)			{do{dev->start = 0;} while(0);}

#define tasklet_struct				tq_struct
static inline void tasklet_schedule(struct tasklet_struct *tasklet)
{
	queue_task(tasklet, &tq_immediate);
	mark_bh(IMMEDIATE_BH);
}

static inline void tasklet_init(struct tasklet_struct *tasklet,
				void (*func)(unsigned long),
				unsigned long data)
{
	tasklet->next = NULL;
	tasklet->sync = 0;
	tasklet->routine = (void (*)(void *))func;
	tasklet->data = (void *)data;
}
#define tasklet_kill(tasklet)			{do{} while(0);}
#else
#define late_stop_netif_stop_queue(dev)		netif_stop_queue(dev)
#define early_stop_netif_stop_queue(dev)	0
#define early_stop_netif_wake_queue(dev)	{do{} while(0);}
#define ace_mark_net_bh()			{do{} while(0);}
#define ace_if_down(dev)			{do{} while(0);}
#endif

#if (LINUX_VERSION_CODE >= 0x02031b)
#define NEW_NETINIT
#define ACE_PROBE_ARG				void
#else
#define ACE_PROBE_ARG				struct net_device *dev
#endif

#define ACE_MAX_MOD_PARMS	8
#define BOARD_IDX_STATIC	0
#define BOARD_IDX_OVERFLOW	-1


#include "acenic_np.h"

/*
 * These must be defined before the firmware is included.
 */
#define MAX_TEXT_LEN	96*1024
#define MAX_RODATA_LEN	8*1024
#define MAX_DATA_LEN	2*1024

#include "acenic_firmware.h"

/*
 * This driver currently supports Tigon I and Tigon II based cards
 * including the Alteon AceNIC, the 3Com 3C985[B] and NetGear
 * GA620. The driver should also work on the SGI, DEC and Farallon
 * versions of the card, however I have not been able to test that
 * myself.
 *
 * This card is really neat, it supports receive hardware checksumming
 * and jumbo frames (up to 9000 bytes) and does a lot of work in the
 * firmware. Also the programming interface is quite neat, except for
 * the parts dealing with the i2c eeprom on the card ;-)
 *
 * Using jumbo frames:
 *
 * To enable jumbo frames, simply specify an mtu between 1500 and 9000
 * bytes to ifconfig. Jumbo frames can be enabled or disabled at any time
 * by running `ifconfig eth<X> mtu <MTU>' with <X> being the Ethernet
 * interface number and <MTU> being the MTU value.
 *
 * Module parameters:
 *
 * When compiled as a loadable module, the driver allows for a number
 * of module parameters to be specified. The driver supports the
 * following module parameters:
 *
 *  trace=<val> - Firmware trace level. This requires special traced
 *                firmware to replace the firmware supplied with
 *                the driver - for debugging purposes only.
 *
 *  link=<val>  - Link state. Normally you want to use the default link
 *                parameters set by the driver. This can be used to
 *                override these in case your switch doesn't negotiate
 *                the link properly. Valid values are:
 *         0x0001 - Force half duplex link.
 *         0x0002 - Do not negotiate line speed with the other end.
 *         0x0010 - 10Mbit/sec link.
 *         0x0020 - 100Mbit/sec link.
 *         0x0040 - 1000Mbit/sec link.
 *         0x0100 - Do not negotiate flow control.
 *         0x0200 - Enable RX flow control Y
 *         0x0400 - Enable TX flow control Y (Tigon II NICs only).
 *                Default value is 0x0270, ie. enable link+flow
 *                control negotiation. Negotiating the highest
 *                possible link speed with RX flow control enabled.
 *
 *                When disabling link speed negotiation, only one link
 *                speed is allowed to be specified!
 *
 *  tx_coal_tick=<val> - number of coalescing clock ticks (us) allowed
 *                to wait for more packets to arive before
 *                interrupting the host, from the time the first
 *                packet arrives.
 *
 *  rx_coal_tick=<val> - number of coalescing clock ticks (us) allowed
 *                to wait for more packets to arive in the transmit ring,
 *                before interrupting the host, after transmitting the
 *                first packet in the ring.
 *
 *  max_tx_desc=<val> - maximum number of transmit descriptors
 *                (packets) transmitted before interrupting the host.
 *
 *  max_rx_desc=<val> - maximum number of receive descriptors
 *                (packets) received before interrupting the host.
 *
 *  tx_ratio=<val> - 7 bit value (0 - 63) specifying the split in 64th
 *                increments of the NIC's on board memory to be used for
 *                transmit and receive buffers. For the 1MB NIC app. 800KB
 *                is available, on the 1/2MB NIC app. 300KB is available.
 *                68KB will always be available as a minimum for both
 *                directions. The default value is a 50/50 split.
 *  dis_pci_mem_inval=<val> - disable PCI memory write and invalidate
 *                operations, default (1) is to always disable this as
 *                that is what Alteon does on NT. I have not been able
 *                to measure any real performance differences with
 *                this on my systems. Set <val>=0 if you want to
 *                enable these operations.
 *
 * If you use more than one NIC, specify the parameters for the
 * individual NICs with a comma, ie. trace=0,0x00001fff,0 you want to
 * run tracing on NIC #2 but not on NIC #1 and #3.
 *
 * TODO:
 *
 * - Proper multicast support.
 * - NIC dump support.
 * - More tuning parameters.
 *
 * The mini ring is not used under Linux and I am not sure it makes sense
 * to actually use it.
 *
 * New interrupt handler strategy:
 *
 * The old interrupt handler worked using the traditional method of
 * replacing an skbuff with a new one when a packet arrives. However
 * the rx rings do not need to contain a static number of buffer
 * descriptors, thus it makes sense to move the memory allocation out
 * of the main interrupt handler and do it in a bottom half handler
 * and only allocate new buffers when the number of buffers in the
 * ring is below a certain threshold. In order to avoid starving the
 * NIC under heavy load it is however necessary to force allocation
 * when hitting a minimum threshold. The strategy for alloction is as
 * follows:
 *
 *     RX_LOW_BUF_THRES    - allocate buffers in the bottom half
 *     RX_PANIC_LOW_THRES  - we are very low on buffers, allocate
 *                           the buffers in the interrupt handler
 *     RX_RING_THRES       - maximum number of buffers in the rx ring
 *     RX_MINI_THRES       - maximum number of buffers in the mini ring
 *     RX_JUMBO_THRES      - maximum number of buffers in the jumbo ring
 *
 * One advantagous side effect of this allocation approach is that the
 * entire rx processing can be done without holding any spin lock
 * since the rx rings and registers are totally independant of the tx
 * ring and its registers.  This of course includes the kmalloc's of
 * new skb's. Thus start_xmit can run in parallel with rx processing
 * and the memory allocation on SMP systems.
 *
 * Note that running the skb reallocation in a bottom half opens up
 * another can of races which needs to be handled properly. In
 * particular it can happen that the interrupt handler tries to run
 * the reallocation while the bottom half is either running on another
 * CPU or was interrupted on the same CPU. To get around this the
 * driver uses bitops to prevent the reallocation routines from being
 * reentered.
 *
 * TX handling can also be done without holding any spin lock, wheee
 * this is fun! since tx_ret_csm is only written to by the interrupt
 * handler. The case to be aware of is when shutting down the device
 * and cleaning up where it is necessary to make sure that
 * start_xmit() is not running while this is happening. Well DaveM
 * informs me that this case is already protected against ... bye bye
 * Mr. Spin Lock, it was nice to know you.
 *
 * TX interrupts are now partly disabled so the NIC will only generate
 * TX interrupts for the number of coal ticks, not for the number of
 * TX packets in the queue. This should reduce the number of TX only,
 * ie. when no RX processing is done, interrupts seen.
 */

/*
 * Threshold values for RX buffer allocation - the low water marks for
 * when to start refilling the rings are set to 75% of the ring
 * sizes. It seems to make sense to refill the rings entirely from the
 * intrrupt handler once it gets below the panic threshold, that way
 * we don't risk that the refilling is moved to another CPU when the
 * one running the interrupt handler just got the slab code hot in its
 * cache.
 */
#define RX_RING_SIZE		128
#define RX_MINI_SIZE		128
#define RX_JUMBO_SIZE		48

#define RX_PANIC_STD_THRES	16
#define RX_PANIC_STD_REFILL	(3*RX_PANIC_STD_THRES)/2
#define RX_LOW_STD_THRES	(3*RX_RING_SIZE)/4
#define RX_PANIC_MINI_THRES	12