]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
eth: remove the driver for acenic / tigon1&2
authorJakub Kicinski <kuba@kernel.org>
Fri, 3 Apr 2026 22:05:01 +0000 (15:05 -0700)
committerJakub Kicinski <kuba@kernel.org>
Tue, 7 Apr 2026 01:52:27 +0000 (18:52 -0700)
The entire git history for this driver looks like tree-wide
and automated cleanups. There's even more coming now with
AI, so let's try to delete it instead.

Acked-by: Jes Sorensen <jes@trained-monkey.org>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Link: https://patch.msgid.link/20260403220501.2263835-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
29 files changed:
MAINTAINERS
arch/loongarch/configs/loongson32_defconfig
arch/loongarch/configs/loongson64_defconfig
arch/mips/configs/cavium_octeon_defconfig
arch/mips/configs/loongson2k_defconfig
arch/mips/configs/loongson3_defconfig
arch/mips/configs/malta_qemu_32r6_defconfig
arch/mips/configs/maltaaprp_defconfig
arch/mips/configs/maltasmvp_defconfig
arch/mips/configs/maltasmvp_eva_defconfig
arch/mips/configs/maltaup_defconfig
arch/mips/configs/mtx1_defconfig
arch/parisc/configs/generic-32bit_defconfig
arch/parisc/configs/generic-64bit_defconfig
arch/powerpc/configs/44x/akebono_defconfig
arch/powerpc/configs/g5_defconfig
arch/powerpc/configs/powernv_defconfig
arch/powerpc/configs/ppc64_defconfig
arch/powerpc/configs/ppc64e_defconfig
arch/powerpc/configs/ppc6xx_defconfig
arch/powerpc/configs/skiroot_defconfig
arch/s390/configs/debug_defconfig
arch/s390/configs/defconfig
drivers/net/ethernet/Kconfig
drivers/net/ethernet/Makefile
drivers/net/ethernet/alteon/Kconfig [deleted file]
drivers/net/ethernet/alteon/Makefile [deleted file]
drivers/net/ethernet/alteon/acenic.c [deleted file]
drivers/net/ethernet/alteon/acenic.h [deleted file]

index c583c5478ef6fc5fc9a4ac4dd845940dde78ef57..cb48c9ad9ddc7b46cdaf0bfc265bb836508b1972 100644 (file)
@@ -261,12 +261,6 @@ L: linux-gpio@vger.kernel.org
 S:     Maintained
 F:     drivers/gpio/gpio-pcie-idio-24.c
 
-ACENIC DRIVER
-M:     Jes Sorensen <jes@trained-monkey.org>
-L:     linux-acenic@sunsite.dk
-S:     Maintained
-F:     drivers/net/ethernet/alteon/acenic*
-
 ACER ASPIRE ONE TEMPERATURE AND FAN DRIVER
 M:     Peter Kaestle <peter@piie.net>
 L:     platform-driver-x86@vger.kernel.org
index 276b1577e0be95f76d9a54154667892ffbe5e56a..37e5eada5a675947703d94d952edae36483d3485 100644 (file)
@@ -571,7 +571,6 @@ CONFIG_VIRTIO_NET=m
 # CONFIG_NET_VENDOR_ADAPTEC is not set
 # CONFIG_NET_VENDOR_AGERE is not set
 # CONFIG_NET_VENDOR_ALACRITECH is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 # CONFIG_NET_VENDOR_AMAZON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_AQUANTIA is not set
index a14db1a95e7e4a299fa784d0b354a33cd9775bc0..7648f6c73929245d41a5b5760d63ee2f4822d52a 100644 (file)
@@ -587,7 +587,6 @@ CONFIG_VIRTIO_NET=m
 # CONFIG_NET_VENDOR_ADAPTEC is not set
 # CONFIG_NET_VENDOR_AGERE is not set
 # CONFIG_NET_VENDOR_ALACRITECH is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 # CONFIG_NET_VENDOR_AMAZON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_AQUANTIA is not set
index 68c363366bceb8cb36bf53f5beb7c681fb0c9fd1..64da32b570aa5637f29d5c7c19e80469091c0d27 100644 (file)
@@ -60,7 +60,6 @@ CONFIG_PATA_OCTEON_CF=y
 CONFIG_NETDEVICES=y
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_ATHEROS is not set
 # CONFIG_NET_VENDOR_BROADCOM is not set
index a5c50b63d478ef1055ae869fbc33bf89e2743bf3..ca534a6b66deb77bf4ce9f152cf8c081d1650464 100644 (file)
@@ -137,7 +137,6 @@ CONFIG_NETDEVICES=y
 CONFIG_TUN=m
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_ARC is not set
 # CONFIG_NET_VENDOR_ATHEROS is not set
index 575aaf2423610d818b52ab7981ba77acf43df951..12cb1a6a13602c3508a2bfd71de7319ae43f5033 100644 (file)
@@ -184,7 +184,6 @@ CONFIG_VETH=m
 CONFIG_VIRTIO_NET=m
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_ARC is not set
 # CONFIG_NET_VENDOR_ATHEROS is not set
index accb471a1d93f19f0f9a985164031cf3481f0a7c..dafc9a716e856527100f029631fa8a8f8ad29494 100644 (file)
@@ -85,7 +85,6 @@ CONFIG_ATA_GENERIC=y
 CONFIG_NETDEVICES=y
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 CONFIG_PCNET32=y
 # CONFIG_NET_VENDOR_ATHEROS is not set
 # CONFIG_NET_VENDOR_BROADCOM is not set
index 6bda67c5f68f8732f99f2087b0e1f4bbcc93b992..7361bca6a4059c42ff178d5251e32f58f998394f 100644 (file)
@@ -87,7 +87,6 @@ CONFIG_ATA_GENERIC=y
 CONFIG_NETDEVICES=y
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 CONFIG_PCNET32=y
 # CONFIG_NET_VENDOR_ATHEROS is not set
 # CONFIG_NET_VENDOR_BROADCOM is not set
index e4082537f80fb1cd7eb70137cd419aea80566603..c848a1bfca5cf74a111ce781617aa86a52efc86d 100644 (file)
@@ -86,7 +86,6 @@ CONFIG_ATA_PIIX=y
 CONFIG_NETDEVICES=y
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 CONFIG_PCNET32=y
 # CONFIG_NET_VENDOR_ATHEROS is not set
 # CONFIG_NET_VENDOR_BROADCOM is not set
index 58f5af45fa983540d9d296b577830ee6e8dff0c5..905248e01b954cef2a6bf8dc8b42ace4da65a145 100644 (file)
@@ -89,7 +89,6 @@ CONFIG_ATA_GENERIC=y
 CONFIG_NETDEVICES=y
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 CONFIG_PCNET32=y
 # CONFIG_NET_VENDOR_ATHEROS is not set
 # CONFIG_NET_VENDOR_BROADCOM is not set
index 9bfef7de0d1cf060fa49cd999cc61284bf7af138..b9bbe02f3595b35a55407ff7aa9b290242a92100 100644 (file)
@@ -86,7 +86,6 @@ CONFIG_ATA_GENERIC=y
 CONFIG_NETDEVICES=y
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 CONFIG_PCNET32=y
 # CONFIG_NET_VENDOR_ATHEROS is not set
 # CONFIG_NET_VENDOR_BROADCOM is not set
index 77050ae3945fccf34114e1a2ee6a3a909ebdbe15..2428a6a72747e4fd9006f331eac76396cc683d33 100644 (file)
@@ -257,7 +257,6 @@ CONFIG_PCMCIA_3C589=m
 CONFIG_VORTEX=m
 CONFIG_TYPHOON=m
 CONFIG_ADAPTEC_STARFIRE=m
-CONFIG_ACENIC=m
 CONFIG_AMD8111_ETH=m
 CONFIG_PCNET32=m
 CONFIG_PCMCIA_NMCLAN=m
index 5444ce6405f3aaf1866bc46d7e1708403a427a79..ad2fb69184f3e765efa34141629949447c5aef79 100644 (file)
@@ -78,7 +78,6 @@ CONFIG_DUMMY=m
 CONFIG_TUN=m
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_ATHEROS is not set
 # CONFIG_NET_VENDOR_BROADCOM is not set
index ce91f9d1fdbfbba0b07390bae33ff5af622816ce..b21287a9250cabfdcdff05c738d1d4ed2e8b77e1 100644 (file)
@@ -97,7 +97,6 @@ CONFIG_NETCONSOLE_DYNAMIC=y
 CONFIG_TUN=y
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_ATHEROS is not set
 # CONFIG_NET_VENDOR_BROADCOM is not set
index 02e88648a2e631c17d3366e25990203f226b48bf..11ad5ed3cc903d6cb3923daa55f9d0ff46a93087 100644 (file)
@@ -47,7 +47,6 @@ CONFIG_SATA_AHCI_PLATFORM=y
 # CONFIG_ATA_SFF is not set
 # CONFIG_NET_VENDOR_3COM is not set
 # CONFIG_NET_VENDOR_ADAPTEC is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_ARC is not set
 # CONFIG_NET_VENDOR_ATHEROS is not set
index 428f17b455132105dbe648d4beaefb99f8e6dd9f..4247d9a30eba8fe2cf3ee72184397917c943224b 100644 (file)
@@ -92,8 +92,6 @@ CONFIG_NETDEVICES=y
 CONFIG_BONDING=m
 CONFIG_DUMMY=m
 CONFIG_TUN=m
-CONFIG_ACENIC=m
-CONFIG_ACENIC_OMIT_TIGON_I=y
 CONFIG_TIGON3=y
 CONFIG_E1000=y
 CONFIG_SUNGEM=y
index 9ac746cfb4be316cad421dd13ecc4b32c7c9cb16..ce72c6d22ca263c901133b2c71c316889c4509f2 100644 (file)
@@ -158,8 +158,6 @@ CONFIG_NETCONSOLE=m
 CONFIG_TUN=m
 CONFIG_VETH=m
 CONFIG_VORTEX=m
-CONFIG_ACENIC=m
-CONFIG_ACENIC_OMIT_TIGON_I=y
 CONFIG_PCNET32=m
 CONFIG_TIGON3=y
 CONFIG_BNX2X=m
index 2b0720f2753bc043c4e77e8e47e12ff8381c7b40..7603af31938560cdc9f3f6c8d4739712097baf09 100644 (file)
@@ -204,8 +204,6 @@ CONFIG_NETCONSOLE=y
 CONFIG_TUN=m
 CONFIG_VIRTIO_NET=m
 CONFIG_VORTEX=m
-CONFIG_ACENIC=m
-CONFIG_ACENIC_OMIT_TIGON_I=y
 CONFIG_PCNET32=m
 CONFIG_TIGON3=y
 CONFIG_BNX2X=m
index 90247b2a0ab0cf18d67b778b8627444c0ea11391..e44c65693b5158ab90955e60fcae71f82c910191 100644 (file)
@@ -96,8 +96,6 @@ CONFIG_DUMMY=m
 CONFIG_NETCONSOLE=y
 CONFIG_TUN=m
 CONFIG_VORTEX=y
-CONFIG_ACENIC=y
-CONFIG_ACENIC_OMIT_TIGON_I=y
 CONFIG_PCNET32=y
 CONFIG_TIGON3=y
 CONFIG_E100=y
index 3c08f46f3d41f9714bb6444564e4e1f91718579e..b70307b70af0943e7a32e74a71159697652ea4db 100644 (file)
@@ -410,7 +410,6 @@ CONFIG_PCMCIA_3C589=m
 CONFIG_VORTEX=m
 CONFIG_TYPHOON=m
 CONFIG_ADAPTEC_STARFIRE=m
-CONFIG_ACENIC=m
 CONFIG_AMD8111_ETH=m
 CONFIG_PCNET32=m
 CONFIG_PCMCIA_NMCLAN=m
index 86c74146824aa08c9ce7cfb0072a37420dcef4f0..ff1bed4b6d2cb16b351c60dfcb59bef409a692ff 100644 (file)
@@ -125,8 +125,6 @@ CONFIG_DM_MULTIPATH=m
 # CONFIG_NET_VENDOR_ADAPTEC is not set
 # CONFIG_NET_VENDOR_AGERE is not set
 # CONFIG_NET_VENDOR_ALACRITECH is not set
-CONFIG_ACENIC=m
-CONFIG_ACENIC_OMIT_TIGON_I=y
 # CONFIG_NET_VENDOR_AMAZON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_AQUANTIA is not set
index 98fd0a2f51c6af2dba075156eccf9ed618634dca..2a94abcabd7288e40f0ff631896ec7d449e7fbe3 100644 (file)
@@ -524,7 +524,6 @@ CONFIG_NLMON=m
 # CONFIG_NET_VENDOR_ADAPTEC is not set
 # CONFIG_NET_VENDOR_AGERE is not set
 # CONFIG_NET_VENDOR_ALACRITECH is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 # CONFIG_NET_VENDOR_AMAZON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_AQUANTIA is not set
index 0f4cedcab3cef4005fef1db2b4a10ea5205b7d36..763fc7db101f5087e91a5d31c4795b39609eedc7 100644 (file)
@@ -514,7 +514,6 @@ CONFIG_NLMON=m
 # CONFIG_NET_VENDOR_ADAPTEC is not set
 # CONFIG_NET_VENDOR_AGERE is not set
 # CONFIG_NET_VENDOR_ALACRITECH is not set
-# CONFIG_NET_VENDOR_ALTEON is not set
 # CONFIG_NET_VENDOR_AMAZON is not set
 # CONFIG_NET_VENDOR_AMD is not set
 # CONFIG_NET_VENDOR_AQUANTIA is not set
index aa7103e7f47f6e83dacaabbe00a052ff1af0599b..bdc29d1431606871ab4b46f7e369d6ac28a31549 100644 (file)
@@ -23,7 +23,6 @@ source "drivers/net/ethernet/agere/Kconfig"
 source "drivers/net/ethernet/airoha/Kconfig"
 source "drivers/net/ethernet/alacritech/Kconfig"
 source "drivers/net/ethernet/allwinner/Kconfig"
-source "drivers/net/ethernet/alteon/Kconfig"
 source "drivers/net/ethernet/altera/Kconfig"
 source "drivers/net/ethernet/amazon/Kconfig"
 source "drivers/net/ethernet/amd/Kconfig"
index 6615a67a63d561a26f24ac45a0dac45bba80be6a..6bffb60ba6440ce3c0244c2d5d18a2ae1b380734 100644 (file)
@@ -13,7 +13,6 @@ obj-$(CONFIG_NET_VENDOR_AGERE) += agere/
 obj-$(CONFIG_NET_VENDOR_AIROHA) += airoha/
 obj-$(CONFIG_NET_VENDOR_ALACRITECH) += alacritech/
 obj-$(CONFIG_NET_VENDOR_ALLWINNER) += allwinner/
-obj-$(CONFIG_NET_VENDOR_ALTEON) += alteon/
 obj-$(CONFIG_ALTERA_TSE) += altera/
 obj-$(CONFIG_NET_VENDOR_AMAZON) += amazon/
 obj-$(CONFIG_NET_VENDOR_AMD) += amd/
diff --git a/drivers/net/ethernet/alteon/Kconfig b/drivers/net/ethernet/alteon/Kconfig
deleted file mode 100644 (file)
index cfe1f31..0000000
+++ /dev/null
@@ -1,47 +0,0 @@
-# SPDX-License-Identifier: GPL-2.0-only
-#
-# Alteon network device configuration
-#
-
-config NET_VENDOR_ALTEON
-       bool "Alteon devices"
-       default y
-       depends on PCI
-       help
-         If you have a network (Ethernet) card belonging to this class, say Y.
-
-         Note that the answer to this question doesn't directly affect the
-         kernel: saying N will just cause the configurator to skip all
-         the questions about Alteon cards. If you say Y, you will be asked for
-         your specific card in the following questions.
-
-if NET_VENDOR_ALTEON
-
-config ACENIC
-       tristate "Alteon AceNIC/3Com 3C985/NetGear GA620 Gigabit support"
-       depends on PCI
-       help
-         Say Y here if you have an Alteon AceNIC, 3Com 3C985(B), NetGear
-         GA620, SGI Gigabit or Farallon PN9000-SX PCI Gigabit Ethernet
-         adapter. The driver allows for using the Jumbo Frame option (9000
-         bytes/frame) however it requires that your switches can handle this
-         as well. To enable Jumbo Frames, add `mtu 9000' to your ifconfig
-         line.
-
-         To compile this driver as a module, choose M here: the
-         module will be called acenic.
-
-config ACENIC_OMIT_TIGON_I
-       bool "Omit support for old Tigon I based AceNICs"
-       depends on ACENIC
-       help
-         Say Y here if you only have Tigon II based AceNICs and want to leave
-         out support for the older Tigon I based cards which are no longer
-         being sold (ie. the original Alteon AceNIC and 3Com 3C985 (non B
-         version)).  This will reduce the size of the driver object by
-         app. 100KB.  If you are not sure whether your card is a Tigon I or a
-         Tigon II, say N here.
-
-         The safe and default value for this is N.
-
-endif # NET_VENDOR_ALTEON
diff --git a/drivers/net/ethernet/alteon/Makefile b/drivers/net/ethernet/alteon/Makefile
deleted file mode 100644 (file)
index be52255..0000000
+++ /dev/null
@@ -1,6 +0,0 @@
-# SPDX-License-Identifier: GPL-2.0-only
-#
-# Makefile for the Alteon network device drivers.
-#
-
-obj-$(CONFIG_ACENIC) += acenic.o
diff --git a/drivers/net/ethernet/alteon/acenic.c b/drivers/net/ethernet/alteon/acenic.c
deleted file mode 100644 (file)
index 455ee89..0000000
+++ /dev/null
@@ -1,3178 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0-or-later
-/*
- * acenic.c: Linux driver for the Alteon AceNIC Gigabit Ethernet card
- *           and other Tigon based cards.
- *
- * Copyright 1998-2002 by Jes Sorensen, <jes@trained-monkey.org>.
- *
- * 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 linux-acenic-help@sunsite.auc.dk to
- * see how to subscribe.
- *
- * Additional credits:
- *   Pete Wyckoff <wyckoff@ca.sandia.gov>: 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 <davem@redhat.com>: 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 <Matt_Domsch@dell.com>: Detect Alteon 1000baseT cards
- *                                       ETHTOOL_GDRVINFO support
- *   Chip Salzenberg <chip@valinux.com>: Fix race condition between tx
- *                                       handler and close() cleanup.
- *   Ken Aaker <kdaaker@rchland.vnet.ibm.com>: Correct check for whether
- *                                       memory mapped IO is enabled to
- *                                       make the driver work on RS/6000.
- *   Takayoshi Kouchi <kouchi@hpc.bs1.fc.nec.co.jp>: Identifying problem
- *                                       where the driver would disable
- *                                       bus master mode if it had to disable
- *                                       write and invalidate.
- *   Stephen Hack <stephen_hack@hp.com>: Fixed ace_set_mac_addr for little
- *                                       endian systems.
- *   Val Henson <vhenson@esscom.com>:    Reset Jumbo skb producer and
- *                                       rx producer index when
- *                                       flushing the Jumbo ring.
- *   Hans Grobler <grobh@sun.ac.za>:     Memory leak fixes in the
- *                                       driver init path.
- *   Grant Grundler <grundler@cup.hp.com>: PCI write posting fixes.
- */
-
-#include <linux/module.h>
-#include <linux/moduleparam.h>
-#include <linux/types.h>
-#include <linux/errno.h>
-#include <linux/ioport.h>
-#include <linux/pci.h>
-#include <linux/dma-mapping.h>
-#include <linux/kernel.h>
-#include <linux/netdevice.h>
-#include <linux/etherdevice.h>
-#include <linux/skbuff.h>
-#include <linux/delay.h>
-#include <linux/mm.h>
-#include <linux/highmem.h>
-#include <linux/sockios.h>
-#include <linux/firmware.h>
-#include <linux/slab.h>
-#include <linux/prefetch.h>
-#include <linux/if_vlan.h>
-
-#ifdef SIOCETHTOOL
-#include <linux/ethtool.h>
-#endif
-
-#include <net/sock.h>
-#include <net/ip.h>
-
-#include <asm/io.h>
-#include <asm/irq.h>
-#include <asm/byteorder.h>
-#include <linux/uaccess.h>
-
-
-#define DRV_NAME "acenic"
-
-#undef INDEX_DEBUG
-
-#ifdef CONFIG_ACENIC_OMIT_TIGON_I
-#define ACE_IS_TIGON_I(ap)     0
-#define ACE_TX_RING_ENTRIES(ap)        MAX_TX_RING_ENTRIES
-#else
-#define ACE_IS_TIGON_I(ap)     (ap->version == 1)
-#define ACE_TX_RING_ENTRIES(ap)        ap->tx_ring_entries
-#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_DEVICE_ID_FARALLON_PN9100T
-#define PCI_DEVICE_ID_FARALLON_PN9100T  0xfa
-#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
-
-static const struct pci_device_id acenic_pci_tbl[] = {
-       { 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,
-        * then later Alteon's ID.
-        */
-       { PCI_VENDOR_ID_DEC, PCI_DEVICE_ID_FARALLON_PN9000SX,
-         PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_NETWORK_ETHERNET << 8, 0xffff00, },
-       { PCI_VENDOR_ID_ALTEON, PCI_DEVICE_ID_FARALLON_PN9100T,
-         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);
-
-#define ace_sync_irq(irq)      synchronize_irq(irq)
-
-#ifndef offset_in_page
-#define offset_in_page(ptr)    ((unsigned long)(ptr) & ~PAGE_MASK)
-#endif
-
-#define ACE_MAX_MOD_PARMS      8
-#define BOARD_IDX_STATIC       0
-#define BOARD_IDX_OVERFLOW     -1
-
-#include "acenic.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
-
-#ifndef tigon2FwReleaseLocal
-#define tigon2FwReleaseLocal 0
-#endif
-
-/*
- * 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 independent 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           72
-#define RX_MINI_SIZE           64
-#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
-#define RX_PANIC_MINI_REFILL   (3*RX_PANIC_MINI_THRES)/2
-#define RX_LOW_MINI_THRES      (3*RX_MINI_SIZE)/4
-#define RX_PANIC_JUMBO_THRES   6
-#define RX_PANIC_JUMBO_REFILL  (3*RX_PANIC_JUMBO_THRES)/2
-#define RX_LOW_JUMBO_THRES     (3*RX_JUMBO_SIZE)/4
-
-
-/*
- * Size of the mini ring entries, basically these just should be big
- * enough to take TCP ACKs
- */
-#define ACE_MINI_SIZE          100
-
-#define ACE_MINI_BUFSIZE       ACE_MINI_SIZE
-#define ACE_STD_BUFSIZE                (ACE_STD_MTU + ETH_HLEN + 4)
-#define ACE_JUMBO_BUFSIZE      (ACE_JUMBO_MTU + ETH_HLEN + 4)
-
-/*
- * There seems to be a magic difference in the effect between 995 and 996
- * but little difference between 900 and 995 ... no idea why.
- *
- * There is now a default set of tuning parameters which is set, depending
- * on whether or not the user enables Jumbo frames. It's assumed that if
- * Jumbo frames are enabled, the user wants optimal tuning for that case.
- */
-#define DEF_TX_COAL            400 /* 996 */
-#define DEF_TX_MAX_DESC                60  /* was 40 */
-#define DEF_RX_COAL            120 /* 1000 */
-#define DEF_RX_MAX_DESC                25
-#define DEF_TX_RATIO           21 /* 24 */
-
-#define DEF_JUMBO_TX_COAL      20
-#define DEF_JUMBO_TX_MAX_DESC  60
-#define DEF_JUMBO_RX_COAL      30
-#define DEF_JUMBO_RX_MAX_DESC  6
-#define DEF_JUMBO_TX_RATIO     21
-
-#if tigon2FwReleaseLocal < 20001118
-/*
- * Standard firmware and early modifications duplicate
- * IRQ load without this flag (coal timer is never reset).
- * Note that with this flag tx_coal should be less than
- * time to xmit full tx ring.
- * 400usec is not so bad for tx ring size of 128.
- */
-#define TX_COAL_INTS_ONLY      1       /* worth it */
-#else
-/*
- * With modified firmware, this is not necessary, but still useful.
- */
-#define TX_COAL_INTS_ONLY      1
-#endif
-
-#define DEF_TRACE              0
-#define DEF_STAT               (2 * TICKS_PER_SEC)
-
-
-static int link_state[ACE_MAX_MOD_PARMS];
-static int trace[ACE_MAX_MOD_PARMS];
-static int tx_coal_tick[ACE_MAX_MOD_PARMS];
-static int rx_coal_tick[ACE_MAX_MOD_PARMS];
-static int max_tx_desc[ACE_MAX_MOD_PARMS];
-static int max_rx_desc[ACE_MAX_MOD_PARMS];
-static int tx_ratio[ACE_MAX_MOD_PARMS];
-static int dis_pci_mem_inval[ACE_MAX_MOD_PARMS] = {1, 1, 1, 1, 1, 1, 1, 1};
-
-MODULE_AUTHOR("Jes Sorensen <jes@trained-monkey.org>");
-MODULE_LICENSE("GPL");
-MODULE_DESCRIPTION("AceNIC/3C985/GA620 Gigabit Ethernet driver");
-#ifndef CONFIG_ACENIC_OMIT_TIGON_I
-MODULE_FIRMWARE("acenic/tg1.bin");
-#endif
-MODULE_FIRMWARE("acenic/tg2.bin");
-
-module_param_array_named(link, link_state, int, NULL, 0);
-module_param_array(trace, int, NULL, 0);
-module_param_array(tx_coal_tick, int, NULL, 0);
-module_param_array(max_tx_desc, int, NULL, 0);
-module_param_array(rx_coal_tick, int, NULL, 0);
-module_param_array(max_rx_desc, int, NULL, 0);
-module_param_array(tx_ratio, int, NULL, 0);
-MODULE_PARM_DESC(link, "AceNIC/3C985/NetGear link state");
-MODULE_PARM_DESC(trace, "AceNIC/3C985/NetGear firmware trace level");
-MODULE_PARM_DESC(tx_coal_tick, "AceNIC/3C985/GA620 max clock ticks to wait from first tx descriptor arrives");
-MODULE_PARM_DESC(max_tx_desc, "AceNIC/3C985/GA620 max number of transmit descriptors to wait");
-MODULE_PARM_DESC(rx_coal_tick, "AceNIC/3C985/GA620 max clock ticks to wait from first rx descriptor arrives");
-MODULE_PARM_DESC(max_rx_desc, "AceNIC/3C985/GA620 max number of receive descriptors to wait");
-MODULE_PARM_DESC(tx_ratio, "AceNIC/3C985/GA620 ratio of NIC memory used for TX/RX descriptors (range 0-63)");
-
-
-static const char version[] =
-  "acenic.c: v0.92 08/05/2002  Jes Sorensen, linux-acenic@SunSITE.dk\n"
-  "                            http://home.cern.ch/~jes/gige/acenic.html\n";
-
-static int ace_get_link_ksettings(struct net_device *,
-                                 struct ethtool_link_ksettings *);
-static int ace_set_link_ksettings(struct net_device *,
-                                 const struct ethtool_link_ksettings *);
-static void ace_get_drvinfo(struct net_device *, struct ethtool_drvinfo *);
-
-static const struct ethtool_ops ace_ethtool_ops = {
-       .get_drvinfo = ace_get_drvinfo,
-       .get_link_ksettings = ace_get_link_ksettings,
-       .set_link_ksettings = ace_set_link_ksettings,
-};
-
-static void ace_watchdog(struct net_device *dev, unsigned int txqueue);
-
-static const struct net_device_ops ace_netdev_ops = {
-       .ndo_open               = ace_open,
-       .ndo_stop               = ace_close,
-       .ndo_tx_timeout         = ace_watchdog,
-       .ndo_get_stats          = ace_get_stats,
-       .ndo_start_xmit         = ace_start_xmit,
-       .ndo_set_rx_mode        = ace_set_multicast_list,
-       .ndo_validate_addr      = eth_validate_addr,
-       .ndo_set_mac_address    = ace_set_mac_addr,
-       .ndo_change_mtu         = ace_change_mtu,
-};
-
-static int acenic_probe_one(struct pci_dev *pdev,
-                           const struct pci_device_id *id)
-{
-       struct net_device *dev;
-       struct ace_private *ap;
-       static int boards_found;
-
-       dev = alloc_etherdev(sizeof(struct ace_private));
-       if (dev == NULL)
-               return -ENOMEM;
-
-       SET_NETDEV_DEV(dev, &pdev->dev);
-
-       ap = netdev_priv(dev);
-       ap->ndev = dev;
-       ap->pdev = pdev;
-       ap->name = pci_name(pdev);
-
-       dev->features |= NETIF_F_SG | NETIF_F_IP_CSUM;
-       dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
-
-       dev->watchdog_timeo = 5*HZ;
-       dev->min_mtu = 0;
-       dev->max_mtu = ACE_JUMBO_MTU;
-
-       dev->netdev_ops = &ace_netdev_ops;
-       dev->ethtool_ops = &ace_ethtool_ops;
-
-       /* we only display this string ONCE */
-       if (!boards_found)
-               printk(version);
-
-       if (pci_enable_device(pdev))
-               goto fail_free_netdev;
-
-       /*
-        * Enable master mode before we start playing with the
-        * pci_command word since pci_set_master() will modify
-        * it.
-        */
-       pci_set_master(pdev);
-
-       pci_read_config_word(pdev, PCI_COMMAND, &ap->pci_command);
-
-       /* OpenFirmware on Mac's does not set this - DOH.. */
-       if (!(ap->pci_command & PCI_COMMAND_MEMORY)) {
-               printk(KERN_INFO "%s: Enabling PCI Memory Mapped "
-                      "access - was not enabled by BIOS/Firmware\n",
-                      ap->name);
-               ap->pci_command = ap->pci_command | PCI_COMMAND_MEMORY;
-               pci_write_config_word(ap->pdev, PCI_COMMAND,
-                                     ap->pci_command);
-               wmb();
-       }
-
-       pci_read_config_byte(pdev, PCI_LATENCY_TIMER, &ap->pci_latency);
-       if (ap->pci_latency <= 0x40) {
-               ap->pci_latency = 0x40;
-               pci_write_config_byte(pdev, PCI_LATENCY_TIMER, ap->pci_latency);
-       }
-
-       /*
-        * Remap the regs into kernel space - this is abuse of
-        * dev->base_addr since it was means for I/O port
-        * addresses but who gives a damn.
-        */
-       dev->base_addr = pci_resource_start(pdev, 0);
-       ap->regs = ioremap(dev->base_addr, 0x4000);
-       if (!ap->regs) {
-               printk(KERN_ERR "%s:  Unable to map I/O register, "
-                      "AceNIC %i will be disabled.\n",
-                      ap->name, boards_found);
-               goto fail_free_netdev;
-       }
-
-       switch(pdev->vendor) {
-       case PCI_VENDOR_ID_ALTEON:
-               if (pdev->device == PCI_DEVICE_ID_FARALLON_PN9100T) {
-                       printk(KERN_INFO "%s: Farallon PN9100-T ",
-                              ap->name);
-               } else {
-                       printk(KERN_INFO "%s: Alteon AceNIC ",
-                              ap->name);
-               }
-               break;
-       case PCI_VENDOR_ID_3COM:
-               printk(KERN_INFO "%s: 3Com 3C985 ", ap->name);
-               break;
-       case PCI_VENDOR_ID_NETGEAR:
-               printk(KERN_INFO "%s: NetGear GA620 ", ap->name);
-               break;
-       case PCI_VENDOR_ID_DEC:
-               if (pdev->device == PCI_DEVICE_ID_FARALLON_PN9000SX) {
-                       printk(KERN_INFO "%s: Farallon PN9000-SX ",
-                              ap->name);
-                       break;
-               }
-               fallthrough;
-       case PCI_VENDOR_ID_SGI:
-               printk(KERN_INFO "%s: SGI AceNIC ", ap->name);
-               break;
-       default:
-               printk(KERN_INFO "%s: Unknown AceNIC ", ap->name);
-               break;
-       }
-
-       printk("Gigabit Ethernet at 0x%08lx, ", dev->base_addr);
-       printk("irq %d\n", pdev->irq);
-
-#ifdef CONFIG_ACENIC_OMIT_TIGON_I
-       if ((readl(&ap->regs->HostCtrl) >> 28) == 4) {
-               printk(KERN_ERR "%s: Driver compiled without Tigon I"
-                      " support - NIC disabled\n", dev->name);
-               goto fail_uninit;
-       }
-#endif
-
-       if (ace_allocate_descriptors(dev))
-               goto fail_free_netdev;
-
-#ifdef MODULE
-       if (boards_found >= ACE_MAX_MOD_PARMS)
-               ap->board_idx = BOARD_IDX_OVERFLOW;
-       else
-               ap->board_idx = boards_found;
-#else
-       ap->board_idx = BOARD_IDX_STATIC;
-#endif
-
-       if (ace_init(dev))
-               goto fail_free_netdev;
-
-       if (register_netdev(dev)) {
-               printk(KERN_ERR "acenic: device registration failed\n");
-               goto fail_uninit;
-       }
-       ap->name = dev->name;
-
-       dev->features |= NETIF_F_HIGHDMA;
-
-       pci_set_drvdata(pdev, dev);
-
-       boards_found++;
-       return 0;
-
- fail_uninit:
-       ace_init_cleanup(dev);
- fail_free_netdev:
-       free_netdev(dev);
-       return -ENODEV;
-}
-
-static void acenic_remove_one(struct pci_dev *pdev)
-{
-       struct net_device *dev = pci_get_drvdata(pdev);
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       short i;
-
-       unregister_netdev(dev);
-
-       writel(readl(&regs->CpuCtrl) | CPU_HALT, &regs->CpuCtrl);
-       if (ap->version >= 2)
-               writel(readl(&regs->CpuBCtrl) | CPU_HALT, &regs->CpuBCtrl);
-
-       /*
-        * This clears any pending interrupts
-        */
-       writel(1, &regs->Mb0Lo);
-       readl(&regs->CpuCtrl);  /* flush */
-
-       /*
-        * Make sure no other CPUs are processing interrupts
-        * on the card before the buffers are being released.
-        * Otherwise one might experience some `interesting'
-        * effects.
-        *
-        * Then release the RX buffers - jumbo buffers were
-        * already released in ace_close().
-        */
-       ace_sync_irq(dev->irq);
-
-       for (i = 0; i < RX_STD_RING_ENTRIES; i++) {
-               struct sk_buff *skb = ap->skb->rx_std_skbuff[i].skb;
-
-               if (skb) {
-                       struct ring_info *ringp;
-                       dma_addr_t mapping;
-
-                       ringp = &ap->skb->rx_std_skbuff[i];
-                       mapping = dma_unmap_addr(ringp, mapping);
-                       dma_unmap_page(&ap->pdev->dev, mapping,
-                                      ACE_STD_BUFSIZE, DMA_FROM_DEVICE);
-
-                       ap->rx_std_ring[i].size = 0;
-                       ap->skb->rx_std_skbuff[i].skb = NULL;
-                       dev_kfree_skb(skb);
-               }
-       }
-
-       if (ap->version >= 2) {
-               for (i = 0; i < RX_MINI_RING_ENTRIES; i++) {
-                       struct sk_buff *skb = ap->skb->rx_mini_skbuff[i].skb;
-
-                       if (skb) {
-                               struct ring_info *ringp;
-                               dma_addr_t mapping;
-
-                               ringp = &ap->skb->rx_mini_skbuff[i];
-                               mapping = dma_unmap_addr(ringp,mapping);
-                               dma_unmap_page(&ap->pdev->dev, mapping,
-                                              ACE_MINI_BUFSIZE,
-                                              DMA_FROM_DEVICE);
-
-                               ap->rx_mini_ring[i].size = 0;
-                               ap->skb->rx_mini_skbuff[i].skb = NULL;
-                               dev_kfree_skb(skb);
-                       }
-               }
-       }
-
-       for (i = 0; i < RX_JUMBO_RING_ENTRIES; i++) {
-               struct sk_buff *skb = ap->skb->rx_jumbo_skbuff[i].skb;
-               if (skb) {
-                       struct ring_info *ringp;
-                       dma_addr_t mapping;
-
-                       ringp = &ap->skb->rx_jumbo_skbuff[i];
-                       mapping = dma_unmap_addr(ringp, mapping);
-                       dma_unmap_page(&ap->pdev->dev, mapping,
-                                      ACE_JUMBO_BUFSIZE, DMA_FROM_DEVICE);
-
-                       ap->rx_jumbo_ring[i].size = 0;
-                       ap->skb->rx_jumbo_skbuff[i].skb = NULL;
-                       dev_kfree_skb(skb);
-               }
-       }
-
-       ace_init_cleanup(dev);
-       free_netdev(dev);
-}
-
-static struct pci_driver acenic_pci_driver = {
-       .name           = "acenic",
-       .id_table       = acenic_pci_tbl,
-       .probe          = acenic_probe_one,
-       .remove         = acenic_remove_one,
-};
-
-static void ace_free_descriptors(struct net_device *dev)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       int size;
-
-       if (ap->rx_std_ring != NULL) {
-               size = (sizeof(struct rx_desc) *
-                       (RX_STD_RING_ENTRIES +
-                        RX_JUMBO_RING_ENTRIES +
-                        RX_MINI_RING_ENTRIES +
-                        RX_RETURN_RING_ENTRIES));
-               dma_free_coherent(&ap->pdev->dev, size, ap->rx_std_ring,
-                                 ap->rx_ring_base_dma);
-               ap->rx_std_ring = NULL;
-               ap->rx_jumbo_ring = NULL;
-               ap->rx_mini_ring = NULL;
-               ap->rx_return_ring = NULL;
-       }
-       if (ap->evt_ring != NULL) {
-               size = (sizeof(struct event) * EVT_RING_ENTRIES);
-               dma_free_coherent(&ap->pdev->dev, size, ap->evt_ring,
-                                 ap->evt_ring_dma);
-               ap->evt_ring = NULL;
-       }
-       if (ap->tx_ring != NULL && !ACE_IS_TIGON_I(ap)) {
-               size = (sizeof(struct tx_desc) * MAX_TX_RING_ENTRIES);
-               dma_free_coherent(&ap->pdev->dev, size, ap->tx_ring,
-                                 ap->tx_ring_dma);
-       }
-       ap->tx_ring = NULL;
-
-       if (ap->evt_prd != NULL) {
-               dma_free_coherent(&ap->pdev->dev, sizeof(u32),
-                                 (void *)ap->evt_prd, ap->evt_prd_dma);
-               ap->evt_prd = NULL;
-       }
-       if (ap->rx_ret_prd != NULL) {
-               dma_free_coherent(&ap->pdev->dev, sizeof(u32),
-                                 (void *)ap->rx_ret_prd, ap->rx_ret_prd_dma);
-               ap->rx_ret_prd = NULL;
-       }
-       if (ap->tx_csm != NULL) {
-               dma_free_coherent(&ap->pdev->dev, sizeof(u32),
-                                 (void *)ap->tx_csm, ap->tx_csm_dma);
-               ap->tx_csm = NULL;
-       }
-}
-
-
-static int ace_allocate_descriptors(struct net_device *dev)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       int size;
-
-       size = (sizeof(struct rx_desc) *
-               (RX_STD_RING_ENTRIES +
-                RX_JUMBO_RING_ENTRIES +
-                RX_MINI_RING_ENTRIES +
-                RX_RETURN_RING_ENTRIES));
-
-       ap->rx_std_ring = dma_alloc_coherent(&ap->pdev->dev, size,
-                                            &ap->rx_ring_base_dma, GFP_KERNEL);
-       if (ap->rx_std_ring == NULL)
-               goto fail;
-
-       ap->rx_jumbo_ring = ap->rx_std_ring + RX_STD_RING_ENTRIES;
-       ap->rx_mini_ring = ap->rx_jumbo_ring + RX_JUMBO_RING_ENTRIES;
-       ap->rx_return_ring = ap->rx_mini_ring + RX_MINI_RING_ENTRIES;
-
-       size = (sizeof(struct event) * EVT_RING_ENTRIES);
-
-       ap->evt_ring = dma_alloc_coherent(&ap->pdev->dev, size,
-                                         &ap->evt_ring_dma, GFP_KERNEL);
-
-       if (ap->evt_ring == NULL)
-               goto fail;
-
-       /*
-        * Only allocate a host TX ring for the Tigon II, the Tigon I
-        * has to use PCI registers for this ;-(
-        */
-       if (!ACE_IS_TIGON_I(ap)) {
-               size = (sizeof(struct tx_desc) * MAX_TX_RING_ENTRIES);
-
-               ap->tx_ring = dma_alloc_coherent(&ap->pdev->dev, size,
-                                                &ap->tx_ring_dma, GFP_KERNEL);
-
-               if (ap->tx_ring == NULL)
-                       goto fail;
-       }
-
-       ap->evt_prd = dma_alloc_coherent(&ap->pdev->dev, sizeof(u32),
-                                        &ap->evt_prd_dma, GFP_KERNEL);
-       if (ap->evt_prd == NULL)
-               goto fail;
-
-       ap->rx_ret_prd = dma_alloc_coherent(&ap->pdev->dev, sizeof(u32),
-                                           &ap->rx_ret_prd_dma, GFP_KERNEL);
-       if (ap->rx_ret_prd == NULL)
-               goto fail;
-
-       ap->tx_csm = dma_alloc_coherent(&ap->pdev->dev, sizeof(u32),
-                                       &ap->tx_csm_dma, GFP_KERNEL);
-       if (ap->tx_csm == NULL)
-               goto fail;
-
-       return 0;
-
-fail:
-       /* Clean up. */
-       ace_init_cleanup(dev);
-       return 1;
-}
-
-
-/*
- * Generic cleanup handling data allocated during init. Used when the
- * module is unloaded or if an error occurs during initialization
- */
-static void ace_init_cleanup(struct net_device *dev)
-{
-       struct ace_private *ap;
-
-       ap = netdev_priv(dev);
-
-       ace_free_descriptors(dev);
-
-       if (ap->info)
-               dma_free_coherent(&ap->pdev->dev, sizeof(struct ace_info),
-                                 ap->info, ap->info_dma);
-       kfree(ap->skb);
-       kfree(ap->trace_buf);
-
-       if (dev->irq)
-               free_irq(dev->irq, dev);
-
-       iounmap(ap->regs);
-}
-
-
-/*
- * Commands are considered to be slow.
- */
-static inline void ace_issue_cmd(struct ace_regs __iomem *regs, struct cmd *cmd)
-{
-       u32 idx;
-
-       idx = readl(&regs->CmdPrd);
-
-       writel(*(u32 *)(cmd), &regs->CmdRng[idx]);
-       idx = (idx + 1) % CMD_RING_ENTRIES;
-
-       writel(idx, &regs->CmdPrd);
-}
-
-
-static int ace_init(struct net_device *dev)
-{
-       struct ace_private *ap;
-       struct ace_regs __iomem *regs;
-       struct ace_info *info = NULL;
-       struct pci_dev *pdev;
-       unsigned long myjif;
-       u64 tmp_ptr;
-       u32 tig_ver, mac1, mac2, tmp, pci_state;
-       int board_idx, ecode = 0;
-       short i;
-       unsigned char cache_size;
-       u8 addr[ETH_ALEN];
-
-       ap = netdev_priv(dev);
-       regs = ap->regs;
-
-       board_idx = ap->board_idx;
-
-       /*
-        * aman@sgi.com - its useful to do a NIC reset here to
-        * address the `Firmware not running' problem subsequent
-        * to any crashes involving the NIC
-        */
-       writel(HW_RESET | (HW_RESET << 24), &regs->HostCtrl);
-       readl(&regs->HostCtrl);         /* PCI write posting */
-       udelay(5);
-
-       /*
-        * Don't access any other registers before this point!
-        */
-#ifdef __BIG_ENDIAN
-       /*
-        * This will most likely need BYTE_SWAP once we switch
-        * to using __raw_writel()
-        */
-       writel((WORD_SWAP | CLR_INT | ((WORD_SWAP | CLR_INT) << 24)),
-              &regs->HostCtrl);
-#else
-       writel((CLR_INT | WORD_SWAP | ((CLR_INT | WORD_SWAP) << 24)),
-              &regs->HostCtrl);
-#endif
-       readl(&regs->HostCtrl);         /* PCI write posting */
-
-       /*
-        * Stop the NIC CPU and clear pending interrupts
-        */
-       writel(readl(&regs->CpuCtrl) | CPU_HALT, &regs->CpuCtrl);
-       readl(&regs->CpuCtrl);          /* PCI write posting */
-       writel(0, &regs->Mb0Lo);
-
-       tig_ver = readl(&regs->HostCtrl) >> 28;
-
-       switch(tig_ver){
-#ifndef CONFIG_ACENIC_OMIT_TIGON_I
-       case 4:
-       case 5:
-               printk(KERN_INFO "  Tigon I  (Rev. %i), Firmware: %i.%i.%i, ",
-                      tig_ver, ap->firmware_major, ap->firmware_minor,
-                      ap->firmware_fix);
-               writel(0, &regs->LocalCtrl);
-               ap->version = 1;
-               ap->tx_ring_entries = TIGON_I_TX_RING_ENTRIES;
-               break;
-#endif
-       case 6:
-               printk(KERN_INFO "  Tigon II (Rev. %i), Firmware: %i.%i.%i, ",
-                      tig_ver, ap->firmware_major, ap->firmware_minor,
-                      ap->firmware_fix);
-               writel(readl(&regs->CpuBCtrl) | CPU_HALT, &regs->CpuBCtrl);
-               readl(&regs->CpuBCtrl);         /* PCI write posting */
-               /*
-                * The SRAM bank size does _not_ indicate the amount
-                * of memory on the card, it controls the _bank_ size!
-                * Ie. a 1MB AceNIC will have two banks of 512KB.
-                */
-               writel(SRAM_BANK_512K, &regs->LocalCtrl);
-               writel(SYNC_SRAM_TIMING, &regs->MiscCfg);
-               ap->version = 2;
-               ap->tx_ring_entries = MAX_TX_RING_ENTRIES;
-               break;
-       default:
-               printk(KERN_WARNING "  Unsupported Tigon version detected "
-                      "(%i)\n", tig_ver);
-               ecode = -ENODEV;
-               goto init_error;
-       }
-
-       /*
-        * ModeStat _must_ be set after the SRAM settings as this change
-        * seems to corrupt the ModeStat and possible other registers.
-        * The SRAM settings survive resets and setting it to the same
-        * value a second time works as well. This is what caused the
-        * `Firmware not running' problem on the Tigon II.
-        */
-#ifdef __BIG_ENDIAN
-       writel(ACE_BYTE_SWAP_DMA | ACE_WARN | ACE_FATAL | ACE_BYTE_SWAP_BD |
-              ACE_WORD_SWAP_BD | ACE_NO_JUMBO_FRAG, &regs->ModeStat);
-#else
-       writel(ACE_BYTE_SWAP_DMA | ACE_WARN | ACE_FATAL |
-              ACE_WORD_SWAP_BD | ACE_NO_JUMBO_FRAG, &regs->ModeStat);
-#endif
-       readl(&regs->ModeStat);         /* PCI write posting */
-
-       mac1 = 0;
-       for(i = 0; i < 4; i++) {
-               int t;
-
-               mac1 = mac1 << 8;
-               t = read_eeprom_byte(dev, 0x8c+i);
-               if (t < 0) {
-                       ecode = -EIO;
-                       goto init_error;
-               } else
-                       mac1 |= (t & 0xff);
-       }
-       mac2 = 0;
-       for(i = 4; i < 8; i++) {
-               int t;
-
-               mac2 = mac2 << 8;
-               t = read_eeprom_byte(dev, 0x8c+i);
-               if (t < 0) {
-                       ecode = -EIO;
-                       goto init_error;
-               } else
-                       mac2 |= (t & 0xff);
-       }
-
-       writel(mac1, &regs->MacAddrHi);
-       writel(mac2, &regs->MacAddrLo);
-
-       addr[0] = (mac1 >> 8) & 0xff;
-       addr[1] = mac1 & 0xff;
-       addr[2] = (mac2 >> 24) & 0xff;
-       addr[3] = (mac2 >> 16) & 0xff;
-       addr[4] = (mac2 >> 8) & 0xff;
-       addr[5] = mac2 & 0xff;
-       eth_hw_addr_set(dev, addr);
-
-       printk("MAC: %pM\n", dev->dev_addr);
-
-       /*
-        * Looks like this is necessary to deal with on all architectures,
-        * even this %$#%$# N440BX Intel based thing doesn't get it right.
-        * Ie. having two NICs in the machine, one will have the cache
-        * line set at boot time, the other will not.
-        */
-       pdev = ap->pdev;
-       pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE, &cache_size);
-       cache_size <<= 2;
-       if (cache_size != SMP_CACHE_BYTES) {
-               printk(KERN_INFO "  PCI cache line size set incorrectly "
-                      "(%i bytes) by BIOS/FW, ", cache_size);
-               if (cache_size > SMP_CACHE_BYTES)
-                       printk("expecting %i\n", SMP_CACHE_BYTES);
-               else {
-                       printk("correcting to %i\n", SMP_CACHE_BYTES);
-                       pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE,
-                                             SMP_CACHE_BYTES >> 2);
-               }
-       }
-
-       pci_state = readl(&regs->PciState);
-       printk(KERN_INFO "  PCI bus width: %i bits, speed: %iMHz, "
-              "latency: %i clks\n",
-               (pci_state & PCI_32BIT) ? 32 : 64,
-               (pci_state & PCI_66MHZ) ? 66 : 33,
-               ap->pci_latency);
-
-       /*
-        * Set the max DMA transfer size. Seems that for most systems
-        * the performance is better when no MAX parameter is
-        * set. However for systems enabling PCI write and invalidate,
-        * DMA writes must be set to the L1 cache line size to get
-        * optimal performance.
-        *
-        * The default is now to turn the PCI write and invalidate off
-        * - that is what Alteon does for NT.
-        */
-       tmp = READ_CMD_MEM | WRITE_CMD_MEM;
-       if (ap->version >= 2) {
-               tmp |= (MEM_READ_MULTIPLE | (pci_state & PCI_66MHZ));
-               /*
-                * Tuning parameters only supported for 8 cards
-                */
-               if (board_idx == BOARD_IDX_OVERFLOW ||
-                   dis_pci_mem_inval[board_idx]) {
-                       if (ap->pci_command & PCI_COMMAND_INVALIDATE) {
-                               ap->pci_command &= ~PCI_COMMAND_INVALIDATE;
-                               pci_write_config_word(pdev, PCI_COMMAND,
-                                                     ap->pci_command);
-                               printk(KERN_INFO "  Disabling PCI memory "
-                                      "write and invalidate\n");
-                       }
-               } else if (ap->pci_command & PCI_COMMAND_INVALIDATE) {
-                       printk(KERN_INFO "  PCI memory write & invalidate "
-                              "enabled by BIOS, enabling counter measures\n");
-
-                       switch(SMP_CACHE_BYTES) {
-                       case 16:
-                               tmp |= DMA_WRITE_MAX_16;
-                               break;
-                       case 32:
-                               tmp |= DMA_WRITE_MAX_32;
-                               break;
-                       case 64:
-                               tmp |= DMA_WRITE_MAX_64;
-                               break;
-                       case 128:
-                               tmp |= DMA_WRITE_MAX_128;
-                               break;
-                       default:
-                               printk(KERN_INFO "  Cache line size %i not "
-                                      "supported, PCI write and invalidate "
-                                      "disabled\n", SMP_CACHE_BYTES);
-                               ap->pci_command &= ~PCI_COMMAND_INVALIDATE;
-                               pci_write_config_word(pdev, PCI_COMMAND,
-                                                     ap->pci_command);
-                       }
-               }
-       }
-
-#ifdef __sparc__
-       /*
-        * On this platform, we know what the best dma settings
-        * are.  We use 64-byte maximum bursts, because if we
-        * burst larger than the cache line size (or even cross
-        * a 64byte boundary in a single burst) the UltraSparc
-        * PCI controller will disconnect at 64-byte multiples.
-        *
-        * Read-multiple will be properly enabled above, and when
-        * set will give the PCI controller proper hints about
-        * prefetching.
-        */
-       tmp &= ~DMA_READ_WRITE_MASK;
-       tmp |= DMA_READ_MAX_64;
-       tmp |= DMA_WRITE_MAX_64;
-#endif
-#ifdef __alpha__
-       tmp &= ~DMA_READ_WRITE_MASK;
-       tmp |= DMA_READ_MAX_128;
-       /*
-        * All the docs say MUST NOT. Well, I did.
-        * Nothing terrible happens, if we load wrong size.
-        * Bit w&i still works better!
-        */
-       tmp |= DMA_WRITE_MAX_128;
-#endif
-       writel(tmp, &regs->PciState);
-
-#if 0
-       /*
-        * The Host PCI bus controller driver has to set FBB.
-        * If all devices on that PCI bus support FBB, then the controller
-        * can enable FBB support in the Host PCI Bus controller (or on
-        * the PCI-PCI bridge if that applies).
-        * -ggg
-        */
-       /*
-        * I have received reports from people having problems when this
-        * bit is enabled.
-        */
-       if (!(ap->pci_command & PCI_COMMAND_FAST_BACK)) {
-               printk(KERN_INFO "  Enabling PCI Fast Back to Back\n");
-               ap->pci_command |= PCI_COMMAND_FAST_BACK;
-               pci_write_config_word(pdev, PCI_COMMAND, ap->pci_command);
-       }
-#endif
-
-       /*
-        * Configure DMA attributes.
-        */
-       if (dma_set_mask(&pdev->dev, DMA_BIT_MASK(64))) {
-               ecode = -ENODEV;
-               goto init_error;
-       }
-
-       /*
-        * Initialize the generic info block and the command+event rings
-        * and the control blocks for the transmit and receive rings
-        * as they need to be setup once and for all.
-        */
-       if (!(info = dma_alloc_coherent(&ap->pdev->dev, sizeof(struct ace_info),
-                                       &ap->info_dma, GFP_KERNEL))) {
-               ecode = -EAGAIN;
-               goto init_error;
-       }
-       ap->info = info;
-
-       /*
-        * Get the memory for the skb rings.
-        */
-       if (!(ap->skb = kzalloc_obj(struct ace_skb))) {
-               ecode = -EAGAIN;
-               goto init_error;
-       }
-
-       ecode = request_irq(pdev->irq, ace_interrupt, IRQF_SHARED,
-                           DRV_NAME, dev);
-       if (ecode) {
-               printk(KERN_WARNING "%s: Requested IRQ %d is busy\n",
-                      DRV_NAME, pdev->irq);
-               goto init_error;
-       } else
-               dev->irq = pdev->irq;
-
-#ifdef INDEX_DEBUG
-       spin_lock_init(&ap->debug_lock);
-       ap->last_tx = ACE_TX_RING_ENTRIES(ap) - 1;
-       ap->last_std_rx = 0;
-       ap->last_mini_rx = 0;
-#endif
-
-       ecode = ace_load_firmware(dev);
-       if (ecode)
-               goto init_error;
-
-       ap->fw_running = 0;
-
-       tmp_ptr = ap->info_dma;
-       writel(tmp_ptr >> 32, &regs->InfoPtrHi);
-       writel(tmp_ptr & 0xffffffff, &regs->InfoPtrLo);
-
-       memset(ap->evt_ring, 0, EVT_RING_ENTRIES * sizeof(struct event));
-
-       set_aceaddr(&info->evt_ctrl.rngptr, ap->evt_ring_dma);
-       info->evt_ctrl.flags = 0;
-
-       *(ap->evt_prd) = 0;
-       wmb();
-       set_aceaddr(&info->evt_prd_ptr, ap->evt_prd_dma);
-       writel(0, &regs->EvtCsm);
-
-       set_aceaddr(&info->cmd_ctrl.rngptr, 0x100);
-       info->cmd_ctrl.flags = 0;
-       info->cmd_ctrl.max_len = 0;
-
-       for (i = 0; i < CMD_RING_ENTRIES; i++)
-               writel(0, &regs->CmdRng[i]);
-
-       writel(0, &regs->CmdPrd);
-       writel(0, &regs->CmdCsm);
-
-       tmp_ptr = ap->info_dma;
-       tmp_ptr += (unsigned long) &(((struct ace_info *)0)->s.stats);
-       set_aceaddr(&info->stats2_ptr, (dma_addr_t) tmp_ptr);
-
-       set_aceaddr(&info->rx_std_ctrl.rngptr, ap->rx_ring_base_dma);
-       info->rx_std_ctrl.max_len = ACE_STD_BUFSIZE;
-       info->rx_std_ctrl.flags =
-         RCB_FLG_TCP_UDP_SUM | RCB_FLG_NO_PSEUDO_HDR | RCB_FLG_VLAN_ASSIST;
-
-       memset(ap->rx_std_ring, 0,
-              RX_STD_RING_ENTRIES * sizeof(struct rx_desc));
-
-       for (i = 0; i < RX_STD_RING_ENTRIES; i++)
-               ap->rx_std_ring[i].flags = BD_FLG_TCP_UDP_SUM;
-
-       ap->rx_std_skbprd = 0;
-       atomic_set(&ap->cur_rx_bufs, 0);
-
-       set_aceaddr(&info->rx_jumbo_ctrl.rngptr,
-                   (ap->rx_ring_base_dma +
-                    (sizeof(struct rx_desc) * RX_STD_RING_ENTRIES)));
-       info->rx_jumbo_ctrl.max_len = 0;
-       info->rx_jumbo_ctrl.flags =
-         RCB_FLG_TCP_UDP_SUM | RCB_FLG_NO_PSEUDO_HDR | RCB_FLG_VLAN_ASSIST;
-
-       memset(ap->rx_jumbo_ring, 0,
-              RX_JUMBO_RING_ENTRIES * sizeof(struct rx_desc));
-
-       for (i = 0; i < RX_JUMBO_RING_ENTRIES; i++)
-               ap->rx_jumbo_ring[i].flags = BD_FLG_TCP_UDP_SUM | BD_FLG_JUMBO;
-
-       ap->rx_jumbo_skbprd = 0;
-       atomic_set(&ap->cur_jumbo_bufs, 0);
-
-       memset(ap->rx_mini_ring, 0,
-              RX_MINI_RING_ENTRIES * sizeof(struct rx_desc));
-
-       if (ap->version >= 2) {
-               set_aceaddr(&info->rx_mini_ctrl.rngptr,
-                           (ap->rx_ring_base_dma +
-                            (sizeof(struct rx_desc) *
-                             (RX_STD_RING_ENTRIES +
-                              RX_JUMBO_RING_ENTRIES))));
-               info->rx_mini_ctrl.max_len = ACE_MINI_SIZE;
-               info->rx_mini_ctrl.flags =
-                 RCB_FLG_TCP_UDP_SUM|RCB_FLG_NO_PSEUDO_HDR|RCB_FLG_VLAN_ASSIST;
-
-               for (i = 0; i < RX_MINI_RING_ENTRIES; i++)
-                       ap->rx_mini_ring[i].flags =
-                               BD_FLG_TCP_UDP_SUM | BD_FLG_MINI;
-       } else {
-               set_aceaddr(&info->rx_mini_ctrl.rngptr, 0);
-               info->rx_mini_ctrl.flags = RCB_FLG_RNG_DISABLE;
-               info->rx_mini_ctrl.max_len = 0;
-       }
-
-       ap->rx_mini_skbprd = 0;
-       atomic_set(&ap->cur_mini_bufs, 0);
-
-       set_aceaddr(&info->rx_return_ctrl.rngptr,
-                   (ap->rx_ring_base_dma +
-                    (sizeof(struct rx_desc) *
-                     (RX_STD_RING_ENTRIES +
-                      RX_JUMBO_RING_ENTRIES +
-                      RX_MINI_RING_ENTRIES))));
-       info->rx_return_ctrl.flags = 0;
-       info->rx_return_ctrl.max_len = RX_RETURN_RING_ENTRIES;
-
-       memset(ap->rx_return_ring, 0,
-              RX_RETURN_RING_ENTRIES * sizeof(struct rx_desc));
-
-       set_aceaddr(&info->rx_ret_prd_ptr, ap->rx_ret_prd_dma);
-       *(ap->rx_ret_prd) = 0;
-
-       writel(TX_RING_BASE, &regs->WinBase);
-
-       if (ACE_IS_TIGON_I(ap)) {
-               ap->tx_ring = (__force struct tx_desc *) regs->Window;
-               for (i = 0; i < (TIGON_I_TX_RING_ENTRIES
-                                * sizeof(struct tx_desc)) / sizeof(u32); i++)
-                       writel(0, (__force void __iomem *)ap->tx_ring  + i * 4);
-
-               set_aceaddr(&info->tx_ctrl.rngptr, TX_RING_BASE);
-       } else {
-               memset(ap->tx_ring, 0,
-                      MAX_TX_RING_ENTRIES * sizeof(struct tx_desc));
-
-               set_aceaddr(&info->tx_ctrl.rngptr, ap->tx_ring_dma);
-       }
-
-       info->tx_ctrl.max_len = ACE_TX_RING_ENTRIES(ap);
-       tmp = RCB_FLG_TCP_UDP_SUM | RCB_FLG_NO_PSEUDO_HDR | RCB_FLG_VLAN_ASSIST;
-
-       /*
-        * The Tigon I does not like having the TX ring in host memory ;-(
-        */
-       if (!ACE_IS_TIGON_I(ap))
-               tmp |= RCB_FLG_TX_HOST_RING;
-#if TX_COAL_INTS_ONLY
-       tmp |= RCB_FLG_COAL_INT_ONLY;
-#endif
-       info->tx_ctrl.flags = tmp;
-
-       set_aceaddr(&info->tx_csm_ptr, ap->tx_csm_dma);
-
-       /*
-        * Potential item for tuning parameter
-        */
-#if 0 /* NO */
-       writel(DMA_THRESH_16W, &regs->DmaReadCfg);
-       writel(DMA_THRESH_16W, &regs->DmaWriteCfg);
-#else
-       writel(DMA_THRESH_8W, &regs->DmaReadCfg);
-       writel(DMA_THRESH_8W, &regs->DmaWriteCfg);
-#endif
-
-       writel(0, &regs->MaskInt);
-       writel(1, &regs->IfIdx);
-#if 0
-       /*
-        * McKinley boxes do not like us fiddling with AssistState
-        * this early
-        */
-       writel(1, &regs->AssistState);
-#endif
-
-       writel(DEF_STAT, &regs->TuneStatTicks);
-       writel(DEF_TRACE, &regs->TuneTrace);
-
-       ace_set_rxtx_parms(dev, 0);
-
-       if (board_idx == BOARD_IDX_OVERFLOW) {
-               printk(KERN_WARNING "%s: more than %i NICs detected, "
-                      "ignoring module parameters!\n",
-                      ap->name, ACE_MAX_MOD_PARMS);
-       } else if (board_idx >= 0) {
-               if (tx_coal_tick[board_idx])
-                       writel(tx_coal_tick[board_idx],
-                              &regs->TuneTxCoalTicks);
-               if (max_tx_desc[board_idx])
-                       writel(max_tx_desc[board_idx], &regs->TuneMaxTxDesc);
-
-               if (rx_coal_tick[board_idx])
-                       writel(rx_coal_tick[board_idx],
-                              &regs->TuneRxCoalTicks);
-               if (max_rx_desc[board_idx])
-                       writel(max_rx_desc[board_idx], &regs->TuneMaxRxDesc);
-
-               if (trace[board_idx])
-                       writel(trace[board_idx], &regs->TuneTrace);
-
-               if ((tx_ratio[board_idx] > 0) && (tx_ratio[board_idx] < 64))
-                       writel(tx_ratio[board_idx], &regs->TxBufRat);
-       }
-
-       /*
-        * Default link parameters
-        */
-       tmp = LNK_ENABLE | LNK_FULL_DUPLEX | LNK_1000MB | LNK_100MB |
-               LNK_10MB | LNK_RX_FLOW_CTL_Y | LNK_NEG_FCTL | LNK_NEGOTIATE;
-       if(ap->version >= 2)
-               tmp |= LNK_TX_FLOW_CTL_Y;
-
-       /*
-        * Override link default parameters
-        */
-       if ((board_idx >= 0) && link_state[board_idx]) {
-               int option = link_state[board_idx];
-
-               tmp = LNK_ENABLE;
-
-               if (option & 0x01) {
-                       printk(KERN_INFO "%s: Setting half duplex link\n",
-                              ap->name);
-                       tmp &= ~LNK_FULL_DUPLEX;
-               }
-               if (option & 0x02)
-                       tmp &= ~LNK_NEGOTIATE;
-               if (option & 0x10)
-                       tmp |= LNK_10MB;
-               if (option & 0x20)
-                       tmp |= LNK_100MB;
-               if (option & 0x40)
-                       tmp |= LNK_1000MB;
-               if ((option & 0x70) == 0) {
-                       printk(KERN_WARNING "%s: No media speed specified, "
-                              "forcing auto negotiation\n", ap->name);
-                       tmp |= LNK_NEGOTIATE | LNK_1000MB |
-                               LNK_100MB | LNK_10MB;
-               }
-               if ((option & 0x100) == 0)
-                       tmp |= LNK_NEG_FCTL;
-               else
-                       printk(KERN_INFO "%s: Disabling flow control "
-                              "negotiation\n", ap->name);
-               if (option & 0x200)
-                       tmp |= LNK_RX_FLOW_CTL_Y;
-               if ((option & 0x400) && (ap->version >= 2)) {
-                       printk(KERN_INFO "%s: Enabling TX flow control\n",
-                              ap->name);
-                       tmp |= LNK_TX_FLOW_CTL_Y;
-               }
-       }
-
-       ap->link = tmp;
-       writel(tmp, &regs->TuneLink);
-       if (ap->version >= 2)
-               writel(tmp, &regs->TuneFastLink);
-
-       writel(ap->firmware_start, &regs->Pc);
-
-       writel(0, &regs->Mb0Lo);
-
-       /*
-        * Set tx_csm before we start receiving interrupts, otherwise
-        * the interrupt handler might think it is supposed to process
-        * tx ints before we are up and running, which may cause a null
-        * pointer access in the int handler.
-        */
-       ap->cur_rx = 0;
-       ap->tx_prd = *(ap->tx_csm) = ap->tx_ret_csm = 0;
-
-       wmb();
-       ace_set_txprd(regs, ap, 0);
-       writel(0, &regs->RxRetCsm);
-
-       /*
-        * Enable DMA engine now.
-        * If we do this sooner, Mckinley box pukes.
-        * I assume it's because Tigon II DMA engine wants to check
-        * *something* even before the CPU is started.
-        */
-       writel(1, &regs->AssistState);  /* enable DMA */
-
-       /*
-        * Start the NIC CPU
-        */
-       writel(readl(&regs->CpuCtrl) & ~(CPU_HALT|CPU_TRACE), &regs->CpuCtrl);
-       readl(&regs->CpuCtrl);
-
-       /*
-        * Wait for the firmware to spin up - max 3 seconds.
-        */
-       myjif = jiffies + 3 * HZ;
-       while (time_before(jiffies, myjif) && !ap->fw_running)
-               cpu_relax();
-
-       if (!ap->fw_running) {
-               printk(KERN_ERR "%s: Firmware NOT running!\n", ap->name);
-
-               ace_dump_trace(ap);
-               writel(readl(&regs->CpuCtrl) | CPU_HALT, &regs->CpuCtrl);
-               readl(&regs->CpuCtrl);
-
-               /* aman@sgi.com - account for badly behaving firmware/NIC:
-                * - have observed that the NIC may continue to generate
-                *   interrupts for some reason; attempt to stop it - halt
-                *   second CPU for Tigon II cards, and also clear Mb0
-                * - if we're a module, we'll fail to load if this was
-                *   the only GbE card in the system => if the kernel does
-                *   see an interrupt from the NIC, code to handle it is
-                *   gone and OOps! - so free_irq also
-                */
-               if (ap->version >= 2)
-                       writel(readl(&regs->CpuBCtrl) | CPU_HALT,
-                              &regs->CpuBCtrl);
-               writel(0, &regs->Mb0Lo);
-               readl(&regs->Mb0Lo);
-
-               ecode = -EBUSY;
-               goto init_error;
-       }
-
-       /*
-        * We load the ring here as there seem to be no way to tell the
-        * firmware to wipe the ring without re-initializing it.
-        */
-       if (!test_and_set_bit(0, &ap->std_refill_busy))
-               ace_load_std_rx_ring(dev, RX_RING_SIZE);
-       else
-               printk(KERN_ERR "%s: Someone is busy refilling the RX ring\n",
-                      ap->name);
-       if (ap->version >= 2) {
-               if (!test_and_set_bit(0, &ap->mini_refill_busy))
-                       ace_load_mini_rx_ring(dev, RX_MINI_SIZE);
-               else
-                       printk(KERN_ERR "%s: Someone is busy refilling "
-                              "the RX mini ring\n", ap->name);
-       }
-       return 0;
-
- init_error:
-       ace_init_cleanup(dev);
-       return ecode;
-}
-
-
-static void ace_set_rxtx_parms(struct net_device *dev, int jumbo)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       int board_idx = ap->board_idx;
-
-       if (board_idx >= 0) {
-               if (!jumbo) {
-                       if (!tx_coal_tick[board_idx])
-                               writel(DEF_TX_COAL, &regs->TuneTxCoalTicks);
-                       if (!max_tx_desc[board_idx])
-                               writel(DEF_TX_MAX_DESC, &regs->TuneMaxTxDesc);
-                       if (!rx_coal_tick[board_idx])
-                               writel(DEF_RX_COAL, &regs->TuneRxCoalTicks);
-                       if (!max_rx_desc[board_idx])
-                               writel(DEF_RX_MAX_DESC, &regs->TuneMaxRxDesc);
-                       if (!tx_ratio[board_idx])
-                               writel(DEF_TX_RATIO, &regs->TxBufRat);
-               } else {
-                       if (!tx_coal_tick[board_idx])
-                               writel(DEF_JUMBO_TX_COAL,
-                                      &regs->TuneTxCoalTicks);
-                       if (!max_tx_desc[board_idx])
-                               writel(DEF_JUMBO_TX_MAX_DESC,
-                                      &regs->TuneMaxTxDesc);
-                       if (!rx_coal_tick[board_idx])
-                               writel(DEF_JUMBO_RX_COAL,
-                                      &regs->TuneRxCoalTicks);
-                       if (!max_rx_desc[board_idx])
-                               writel(DEF_JUMBO_RX_MAX_DESC,
-                                      &regs->TuneMaxRxDesc);
-                       if (!tx_ratio[board_idx])
-                               writel(DEF_JUMBO_TX_RATIO, &regs->TxBufRat);
-               }
-       }
-}
-
-
-static void ace_watchdog(struct net_device *data, unsigned int txqueue)
-{
-       struct net_device *dev = data;
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-
-       /*
-        * We haven't received a stats update event for more than 2.5
-        * seconds and there is data in the transmit queue, thus we
-        * assume the card is stuck.
-        */
-       if (*ap->tx_csm != ap->tx_ret_csm) {
-               printk(KERN_WARNING "%s: Transmitter is stuck, %08x\n",
-                      dev->name, (unsigned int)readl(&regs->HostCtrl));
-               /* This can happen due to ieee flow control. */
-       } else {
-               printk(KERN_DEBUG "%s: BUG... transmitter died. Kicking it.\n",
-                      dev->name);
-#if 0
-               netif_wake_queue(dev);
-#endif
-       }
-}
-
-
-static void ace_bh_work(struct work_struct *work)
-{
-       struct ace_private *ap = from_work(ap, work, ace_bh_work);
-       struct net_device *dev = ap->ndev;
-       int cur_size;
-
-       cur_size = atomic_read(&ap->cur_rx_bufs);
-       if ((cur_size < RX_LOW_STD_THRES) &&
-           !test_and_set_bit(0, &ap->std_refill_busy)) {
-#ifdef DEBUG
-               printk("refilling buffers (current %i)\n", cur_size);
-#endif
-               ace_load_std_rx_ring(dev, RX_RING_SIZE - cur_size);
-       }
-
-       if (ap->version >= 2) {
-               cur_size = atomic_read(&ap->cur_mini_bufs);
-               if ((cur_size < RX_LOW_MINI_THRES) &&
-                   !test_and_set_bit(0, &ap->mini_refill_busy)) {
-#ifdef DEBUG
-                       printk("refilling mini buffers (current %i)\n",
-                              cur_size);
-#endif
-                       ace_load_mini_rx_ring(dev, RX_MINI_SIZE - cur_size);
-               }
-       }
-
-       cur_size = atomic_read(&ap->cur_jumbo_bufs);
-       if (ap->jumbo && (cur_size < RX_LOW_JUMBO_THRES) &&
-           !test_and_set_bit(0, &ap->jumbo_refill_busy)) {
-#ifdef DEBUG
-               printk("refilling jumbo buffers (current %i)\n", cur_size);
-#endif
-               ace_load_jumbo_rx_ring(dev, RX_JUMBO_SIZE - cur_size);
-       }
-       ap->bh_work_pending = 0;
-}
-
-
-/*
- * Copy the contents of the NIC's trace buffer to kernel memory.
- */
-static void ace_dump_trace(struct ace_private *ap)
-{
-#if 0
-       if (!ap->trace_buf)
-               if (!(ap->trace_buf = kmalloc(ACE_TRACE_SIZE, GFP_KERNEL)))
-                   return;
-#endif
-}
-
-
-/*
- * Load the standard rx ring.
- *
- * Loading rings is safe without holding the spin lock since this is
- * done only before the device is enabled, thus no interrupts are
- * generated and by the interrupt handler/bh handler.
- */
-static void ace_load_std_rx_ring(struct net_device *dev, int nr_bufs)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       short i, idx;
-
-
-       prefetchw(&ap->cur_rx_bufs);
-
-       idx = ap->rx_std_skbprd;
-
-       for (i = 0; i < nr_bufs; i++) {
-               struct sk_buff *skb;
-               struct rx_desc *rd;
-               dma_addr_t mapping;
-
-               skb = netdev_alloc_skb_ip_align(dev, ACE_STD_BUFSIZE);
-               if (!skb)
-                       break;
-
-               mapping = dma_map_page(&ap->pdev->dev,
-                                      virt_to_page(skb->data),
-                                      offset_in_page(skb->data),
-                                      ACE_STD_BUFSIZE, DMA_FROM_DEVICE);
-               ap->skb->rx_std_skbuff[idx].skb = skb;
-               dma_unmap_addr_set(&ap->skb->rx_std_skbuff[idx],
-                                  mapping, mapping);
-
-               rd = &ap->rx_std_ring[idx];
-               set_aceaddr(&rd->addr, mapping);
-               rd->size = ACE_STD_BUFSIZE;
-               rd->idx = idx;
-               idx = (idx + 1) % RX_STD_RING_ENTRIES;
-       }
-
-       if (!i)
-               goto error_out;
-
-       atomic_add(i, &ap->cur_rx_bufs);
-       ap->rx_std_skbprd = idx;
-
-       if (ACE_IS_TIGON_I(ap)) {
-               struct cmd cmd;
-               cmd.evt = C_SET_RX_PRD_IDX;
-               cmd.code = 0;
-               cmd.idx = ap->rx_std_skbprd;
-               ace_issue_cmd(regs, &cmd);
-       } else {
-               writel(idx, &regs->RxStdPrd);
-               wmb();
-       }
-
- out:
-       clear_bit(0, &ap->std_refill_busy);
-       return;
-
- error_out:
-       printk(KERN_INFO "Out of memory when allocating "
-              "standard receive buffers\n");
-       goto out;
-}
-
-
-static void ace_load_mini_rx_ring(struct net_device *dev, int nr_bufs)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       short i, idx;
-
-       prefetchw(&ap->cur_mini_bufs);
-
-       idx = ap->rx_mini_skbprd;
-       for (i = 0; i < nr_bufs; i++) {
-               struct sk_buff *skb;
-               struct rx_desc *rd;
-               dma_addr_t mapping;
-
-               skb = netdev_alloc_skb_ip_align(dev, ACE_MINI_BUFSIZE);
-               if (!skb)
-                       break;
-
-               mapping = dma_map_page(&ap->pdev->dev,
-                                      virt_to_page(skb->data),
-                                      offset_in_page(skb->data),
-                                      ACE_MINI_BUFSIZE, DMA_FROM_DEVICE);
-               ap->skb->rx_mini_skbuff[idx].skb = skb;
-               dma_unmap_addr_set(&ap->skb->rx_mini_skbuff[idx],
-                                  mapping, mapping);
-
-               rd = &ap->rx_mini_ring[idx];
-               set_aceaddr(&rd->addr, mapping);
-               rd->size = ACE_MINI_BUFSIZE;
-               rd->idx = idx;
-               idx = (idx + 1) % RX_MINI_RING_ENTRIES;
-       }
-
-       if (!i)
-               goto error_out;
-
-       atomic_add(i, &ap->cur_mini_bufs);
-
-       ap->rx_mini_skbprd = idx;
-
-       writel(idx, &regs->RxMiniPrd);
-       wmb();
-
- out:
-       clear_bit(0, &ap->mini_refill_busy);
-       return;
- error_out:
-       printk(KERN_INFO "Out of memory when allocating "
-              "mini receive buffers\n");
-       goto out;
-}
-
-
-/*
- * Load the jumbo rx ring, this may happen at any time if the MTU
- * is changed to a value > 1500.
- */
-static void ace_load_jumbo_rx_ring(struct net_device *dev, int nr_bufs)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       short i, idx;
-
-       idx = ap->rx_jumbo_skbprd;
-
-       for (i = 0; i < nr_bufs; i++) {
-               struct sk_buff *skb;
-               struct rx_desc *rd;
-               dma_addr_t mapping;
-
-               skb = netdev_alloc_skb_ip_align(dev, ACE_JUMBO_BUFSIZE);
-               if (!skb)
-                       break;
-
-               mapping = dma_map_page(&ap->pdev->dev,
-                                      virt_to_page(skb->data),
-                                      offset_in_page(skb->data),
-                                      ACE_JUMBO_BUFSIZE, DMA_FROM_DEVICE);
-               ap->skb->rx_jumbo_skbuff[idx].skb = skb;
-               dma_unmap_addr_set(&ap->skb->rx_jumbo_skbuff[idx],
-                                  mapping, mapping);
-
-               rd = &ap->rx_jumbo_ring[idx];
-               set_aceaddr(&rd->addr, mapping);
-               rd->size = ACE_JUMBO_BUFSIZE;
-               rd->idx = idx;
-               idx = (idx + 1) % RX_JUMBO_RING_ENTRIES;
-       }
-
-       if (!i)
-               goto error_out;
-
-       atomic_add(i, &ap->cur_jumbo_bufs);
-       ap->rx_jumbo_skbprd = idx;
-
-       if (ACE_IS_TIGON_I(ap)) {
-               struct cmd cmd;
-               cmd.evt = C_SET_RX_JUMBO_PRD_IDX;
-               cmd.code = 0;
-               cmd.idx = ap->rx_jumbo_skbprd;
-               ace_issue_cmd(regs, &cmd);
-       } else {
-               writel(idx, &regs->RxJumboPrd);
-               wmb();
-       }
-
- out:
-       clear_bit(0, &ap->jumbo_refill_busy);
-       return;
- error_out:
-       if (net_ratelimit())
-               printk(KERN_INFO "Out of memory when allocating "
-                      "jumbo receive buffers\n");
-       goto out;
-}
-
-
-/*
- * All events are considered to be slow (RX/TX ints do not generate
- * events) and are handled here, outside the main interrupt handler,
- * to reduce the size of the handler.
- */
-static u32 ace_handle_event(struct net_device *dev, u32 evtcsm, u32 evtprd)
-{
-       struct ace_private *ap;
-
-       ap = netdev_priv(dev);
-
-       while (evtcsm != evtprd) {
-               switch (ap->evt_ring[evtcsm].evt) {
-               case E_FW_RUNNING:
-                       printk(KERN_INFO "%s: Firmware up and running\n",
-                              ap->name);
-                       ap->fw_running = 1;
-                       wmb();
-                       break;
-               case E_STATS_UPDATED:
-                       break;
-               case E_LNK_STATE:
-               {
-                       u16 code = ap->evt_ring[evtcsm].code;
-                       switch (code) {
-                       case E_C_LINK_UP:
-                       {
-                               u32 state = readl(&ap->regs->GigLnkState);
-                               printk(KERN_WARNING "%s: Optical link UP "
-                                      "(%s Duplex, Flow Control: %s%s)\n",
-                                      ap->name,
-                                      state & LNK_FULL_DUPLEX ? "Full":"Half",
-                                      state & LNK_TX_FLOW_CTL_Y ? "TX " : "",
-                                      state & LNK_RX_FLOW_CTL_Y ? "RX" : "");
-                               break;
-                       }
-                       case E_C_LINK_DOWN:
-                               printk(KERN_WARNING "%s: Optical link DOWN\n",
-                                      ap->name);
-                               break;
-                       case E_C_LINK_10_100:
-                               printk(KERN_WARNING "%s: 10/100BaseT link "
-                                      "UP\n", ap->name);
-                               break;
-                       default:
-                               printk(KERN_ERR "%s: Unknown optical link "
-                                      "state %02x\n", ap->name, code);
-                       }
-                       break;
-               }
-               case E_ERROR:
-                       switch(ap->evt_ring[evtcsm].code) {
-                       case E_C_ERR_INVAL_CMD:
-                               printk(KERN_ERR "%s: invalid command error\n",
-                                      ap->name);
-                               break;
-                       case E_C_ERR_UNIMP_CMD:
-                               printk(KERN_ERR "%s: unimplemented command "
-                                      "error\n", ap->name);
-                               break;
-                       case E_C_ERR_BAD_CFG:
-                               printk(KERN_ERR "%s: bad config error\n",
-                                      ap->name);
-                               break;
-                       default:
-                               printk(KERN_ERR "%s: unknown error %02x\n",
-                                      ap->name, ap->evt_ring[evtcsm].code);
-                       }
-                       break;
-               case E_RESET_JUMBO_RNG:
-               {
-                       int i;
-                       for (i = 0; i < RX_JUMBO_RING_ENTRIES; i++) {
-                               if (ap->skb->rx_jumbo_skbuff[i].skb) {
-                                       ap->rx_jumbo_ring[i].size = 0;
-                                       set_aceaddr(&ap->rx_jumbo_ring[i].addr, 0);
-                                       dev_kfree_skb(ap->skb->rx_jumbo_skbuff[i].skb);
-                                       ap->skb->rx_jumbo_skbuff[i].skb = NULL;
-                               }
-                       }
-
-                       if (ACE_IS_TIGON_I(ap)) {
-                               struct cmd cmd;
-                               cmd.evt = C_SET_RX_JUMBO_PRD_IDX;
-                               cmd.code = 0;
-                               cmd.idx = 0;
-                               ace_issue_cmd(ap->regs, &cmd);
-                       } else {
-                               writel(0, &((ap->regs)->RxJumboPrd));
-                               wmb();
-                       }
-
-                       ap->jumbo = 0;
-                       ap->rx_jumbo_skbprd = 0;
-                       printk(KERN_INFO "%s: Jumbo ring flushed\n",
-                              ap->name);
-                       clear_bit(0, &ap->jumbo_refill_busy);
-                       break;
-               }
-               default:
-                       printk(KERN_ERR "%s: Unhandled event 0x%02x\n",
-                              ap->name, ap->evt_ring[evtcsm].evt);
-               }
-               evtcsm = (evtcsm + 1) % EVT_RING_ENTRIES;
-       }
-
-       return evtcsm;
-}
-
-
-static void ace_rx_int(struct net_device *dev, u32 rxretprd, u32 rxretcsm)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       u32 idx;
-       int mini_count = 0, std_count = 0;
-
-       idx = rxretcsm;
-
-       prefetchw(&ap->cur_rx_bufs);
-       prefetchw(&ap->cur_mini_bufs);
-
-       while (idx != rxretprd) {
-               struct ring_info *rip;
-               struct sk_buff *skb;
-               struct rx_desc *retdesc;
-               u32 skbidx;
-               int bd_flags, desc_type, mapsize;
-               u16 csum;
-
-
-               /* make sure the rx descriptor isn't read before rxretprd */
-               if (idx == rxretcsm)
-                       rmb();
-
-               retdesc = &ap->rx_return_ring[idx];
-               skbidx = retdesc->idx;
-               bd_flags = retdesc->flags;
-               desc_type = bd_flags & (BD_FLG_JUMBO | BD_FLG_MINI);
-
-               switch(desc_type) {
-                       /*
-                        * Normal frames do not have any flags set
-                        *
-                        * Mini and normal frames arrive frequently,
-                        * so use a local counter to avoid doing
-                        * atomic operations for each packet arriving.
-                        */
-               case 0:
-                       rip = &ap->skb->rx_std_skbuff[skbidx];
-                       mapsize = ACE_STD_BUFSIZE;
-                       std_count++;
-                       break;
-               case BD_FLG_JUMBO:
-                       rip = &ap->skb->rx_jumbo_skbuff[skbidx];
-                       mapsize = ACE_JUMBO_BUFSIZE;
-                       atomic_dec(&ap->cur_jumbo_bufs);
-                       break;
-               case BD_FLG_MINI:
-                       rip = &ap->skb->rx_mini_skbuff[skbidx];
-                       mapsize = ACE_MINI_BUFSIZE;
-                       mini_count++;
-                       break;
-               default:
-                       printk(KERN_INFO "%s: unknown frame type (0x%02x) "
-                              "returned by NIC\n", dev->name,
-                              retdesc->flags);
-                       goto error;
-               }
-
-               skb = rip->skb;
-               rip->skb = NULL;
-               dma_unmap_page(&ap->pdev->dev, dma_unmap_addr(rip, mapping),
-                              mapsize, DMA_FROM_DEVICE);
-               skb_put(skb, retdesc->size);
-
-               /*
-                * Fly baby, fly!
-                */
-               csum = retdesc->tcp_udp_csum;
-
-               skb->protocol = eth_type_trans(skb, dev);
-
-               /*
-                * Instead of forcing the poor tigon mips cpu to calculate
-                * pseudo hdr checksum, we do this ourselves.
-                */
-               if (bd_flags & BD_FLG_TCP_UDP_SUM) {
-                       skb->csum = htons(csum);
-                       skb->ip_summed = CHECKSUM_COMPLETE;
-               } else {
-                       skb_checksum_none_assert(skb);
-               }
-
-               /* send it up */
-               if ((bd_flags & BD_FLG_VLAN_TAG))
-                       __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), retdesc->vlan);
-               netif_rx(skb);
-
-               dev->stats.rx_packets++;
-               dev->stats.rx_bytes += retdesc->size;
-
-               idx = (idx + 1) % RX_RETURN_RING_ENTRIES;
-       }
-
-       atomic_sub(std_count, &ap->cur_rx_bufs);
-       if (!ACE_IS_TIGON_I(ap))
-               atomic_sub(mini_count, &ap->cur_mini_bufs);
-
- out:
-       /*
-        * According to the documentation RxRetCsm is obsolete with
-        * the 12.3.x Firmware - my Tigon I NICs seem to disagree!
-        */
-       if (ACE_IS_TIGON_I(ap)) {
-               writel(idx, &ap->regs->RxRetCsm);
-       }
-       ap->cur_rx = idx;
-
-       return;
- error:
-       idx = rxretprd;
-       goto out;
-}
-
-
-static inline void ace_tx_int(struct net_device *dev,
-                             u32 txcsm, u32 idx)
-{
-       struct ace_private *ap = netdev_priv(dev);
-
-       do {
-               struct sk_buff *skb;
-               struct tx_ring_info *info;
-
-               info = ap->skb->tx_skbuff + idx;
-               skb = info->skb;
-
-               if (dma_unmap_len(info, maplen)) {
-                       dma_unmap_page(&ap->pdev->dev,
-                                      dma_unmap_addr(info, mapping),
-                                      dma_unmap_len(info, maplen),
-                                      DMA_TO_DEVICE);
-                       dma_unmap_len_set(info, maplen, 0);
-               }
-
-               if (skb) {
-                       dev->stats.tx_packets++;
-                       dev->stats.tx_bytes += skb->len;
-                       dev_consume_skb_irq(skb);
-                       info->skb = NULL;
-               }
-
-               idx = (idx + 1) % ACE_TX_RING_ENTRIES(ap);
-       } while (idx != txcsm);
-
-       if (netif_queue_stopped(dev))
-               netif_wake_queue(dev);
-
-       wmb();
-       ap->tx_ret_csm = txcsm;
-
-       /* So... tx_ret_csm is advanced _after_ check for device wakeup.
-        *
-        * We could try to make it before. In this case we would get
-        * the following race condition: hard_start_xmit on other cpu
-        * enters after we advanced tx_ret_csm and fills space,
-        * which we have just freed, so that we make illegal device wakeup.
-        * There is no good way to workaround this (at entry
-        * to ace_start_xmit detects this condition and prevents
-        * ring corruption, but it is not a good workaround.)
-        *
-        * When tx_ret_csm is advanced after, we wake up device _only_
-        * if we really have some space in ring (though the core doing
-        * hard_start_xmit can see full ring for some period and has to
-        * synchronize.) Superb.
-        * BUT! We get another subtle race condition. hard_start_xmit
-        * may think that ring is full between wakeup and advancing
-        * tx_ret_csm and will stop device instantly! It is not so bad.
-        * We are guaranteed that there is something in ring, so that
-        * the next irq will resume transmission. To speedup this we could
-        * mark descriptor, which closes ring with BD_FLG_COAL_NOW
-        * (see ace_start_xmit).
-        *
-        * Well, this dilemma exists in all lock-free devices.
-        * We, following scheme used in drivers by Donald Becker,
-        * select the least dangerous.
-        *                                                      --ANK
-        */
-}
-
-
-static irqreturn_t ace_interrupt(int irq, void *dev_id)
-{
-       struct net_device *dev = (struct net_device *)dev_id;
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       u32 idx;
-       u32 txcsm, rxretcsm, rxretprd;
-       u32 evtcsm, evtprd;
-
-       /*
-        * In case of PCI shared interrupts or spurious interrupts,
-        * we want to make sure it is actually our interrupt before
-        * spending any time in here.
-        */
-       if (!(readl(&regs->HostCtrl) & IN_INT))
-               return IRQ_NONE;
-
-       /*
-        * ACK intr now. Otherwise we will lose updates to rx_ret_prd,
-        * which happened _after_ rxretprd = *ap->rx_ret_prd; but before
-        * writel(0, &regs->Mb0Lo).
-        *
-        * "IRQ avoidance" recommended in docs applies to IRQs served
-        * threads and it is wrong even for that case.
-        */
-       writel(0, &regs->Mb0Lo);
-       readl(&regs->Mb0Lo);
-
-       /*
-        * There is no conflict between transmit handling in
-        * start_xmit and receive processing, thus there is no reason
-        * to take a spin lock for RX handling. Wait until we start
-        * working on the other stuff - hey we don't need a spin lock
-        * anymore.
-        */
-       rxretprd = *ap->rx_ret_prd;
-       rxretcsm = ap->cur_rx;
-
-       if (rxretprd != rxretcsm)
-               ace_rx_int(dev, rxretprd, rxretcsm);
-
-       txcsm = *ap->tx_csm;
-       idx = ap->tx_ret_csm;
-
-       if (txcsm != idx) {
-               /*
-                * If each skb takes only one descriptor this check degenerates
-                * to identity, because new space has just been opened.
-                * But if skbs are fragmented we must check that this index
-                * update releases enough of space, otherwise we just
-                * wait for device to make more work.
-                */
-               if (!tx_ring_full(ap, txcsm, ap->tx_prd))
-                       ace_tx_int(dev, txcsm, idx);
-       }
-
-       evtcsm = readl(&regs->EvtCsm);
-       evtprd = *ap->evt_prd;
-
-       if (evtcsm != evtprd) {
-               evtcsm = ace_handle_event(dev, evtcsm, evtprd);
-               writel(evtcsm, &regs->EvtCsm);
-       }
-
-       /*
-        * This has to go last in the interrupt handler and run with
-        * the spin lock released ... what lock?
-        */
-       if (netif_running(dev)) {
-               int cur_size;
-               int run_bh_work = 0;
-
-               cur_size = atomic_read(&ap->cur_rx_bufs);
-               if (cur_size < RX_LOW_STD_THRES) {
-                       if ((cur_size < RX_PANIC_STD_THRES) &&
-                           !test_and_set_bit(0, &ap->std_refill_busy)) {
-#ifdef DEBUG
-                               printk("low on std buffers %i\n", cur_size);
-#endif
-                               ace_load_std_rx_ring(dev,
-                                                    RX_RING_SIZE - cur_size);
-                       } else
-                               run_bh_work = 1;
-               }
-
-               if (!ACE_IS_TIGON_I(ap)) {
-                       cur_size = atomic_read(&ap->cur_mini_bufs);
-                       if (cur_size < RX_LOW_MINI_THRES) {
-                               if ((cur_size < RX_PANIC_MINI_THRES) &&
-                                   !test_and_set_bit(0,
-                                                     &ap->mini_refill_busy)) {
-#ifdef DEBUG
-                                       printk("low on mini buffers %i\n",
-                                              cur_size);
-#endif
-                                       ace_load_mini_rx_ring(dev,
-                                                             RX_MINI_SIZE - cur_size);
-                               } else
-                                       run_bh_work = 1;
-                       }
-               }
-
-               if (ap->jumbo) {
-                       cur_size = atomic_read(&ap->cur_jumbo_bufs);
-                       if (cur_size < RX_LOW_JUMBO_THRES) {
-                               if ((cur_size < RX_PANIC_JUMBO_THRES) &&
-                                   !test_and_set_bit(0,
-                                                     &ap->jumbo_refill_busy)){
-#ifdef DEBUG
-                                       printk("low on jumbo buffers %i\n",
-                                              cur_size);
-#endif
-                                       ace_load_jumbo_rx_ring(dev,
-                                                              RX_JUMBO_SIZE - cur_size);
-                               } else
-                                       run_bh_work = 1;
-                       }
-               }
-               if (run_bh_work && !ap->bh_work_pending) {
-                       ap->bh_work_pending = 1;
-                       queue_work(system_bh_wq, &ap->ace_bh_work);
-               }
-       }
-
-       return IRQ_HANDLED;
-}
-
-static int ace_open(struct net_device *dev)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       struct cmd cmd;
-
-       if (!(ap->fw_running)) {
-               printk(KERN_WARNING "%s: Firmware not running!\n", dev->name);
-               return -EBUSY;
-       }
-
-       writel(dev->mtu + ETH_HLEN + 4, &regs->IfMtu);
-
-       cmd.evt = C_CLEAR_STATS;
-       cmd.code = 0;
-       cmd.idx = 0;
-       ace_issue_cmd(regs, &cmd);
-
-       cmd.evt = C_HOST_STATE;
-       cmd.code = C_C_STACK_UP;
-       cmd.idx = 0;
-       ace_issue_cmd(regs, &cmd);
-
-       if (ap->jumbo &&
-           !test_and_set_bit(0, &ap->jumbo_refill_busy))
-               ace_load_jumbo_rx_ring(dev, RX_JUMBO_SIZE);
-
-       if (dev->flags & IFF_PROMISC) {
-               cmd.evt = C_SET_PROMISC_MODE;
-               cmd.code = C_C_PROMISC_ENABLE;
-               cmd.idx = 0;
-               ace_issue_cmd(regs, &cmd);
-
-               ap->promisc = 1;
-       }else
-               ap->promisc = 0;
-       ap->mcast_all = 0;
-
-#if 0
-       cmd.evt = C_LNK_NEGOTIATION;
-       cmd.code = 0;
-       cmd.idx = 0;
-       ace_issue_cmd(regs, &cmd);
-#endif
-
-       netif_start_queue(dev);
-
-       /*
-        * Setup the bottom half rx ring refill handler
-        */
-       INIT_WORK(&ap->ace_bh_work, ace_bh_work);
-       return 0;
-}
-
-
-static int ace_close(struct net_device *dev)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       struct cmd cmd;
-       unsigned long flags;
-       short i;
-
-       /*
-        * Without (or before) releasing irq and stopping hardware, this
-        * is an absolute non-sense, by the way. It will be reset instantly
-        * by the first irq.
-        */
-       netif_stop_queue(dev);
-
-
-       if (ap->promisc) {
-               cmd.evt = C_SET_PROMISC_MODE;
-               cmd.code = C_C_PROMISC_DISABLE;
-               cmd.idx = 0;
-               ace_issue_cmd(regs, &cmd);
-               ap->promisc = 0;
-       }
-
-       cmd.evt = C_HOST_STATE;
-       cmd.code = C_C_STACK_DOWN;
-       cmd.idx = 0;
-       ace_issue_cmd(regs, &cmd);
-
-       cancel_work_sync(&ap->ace_bh_work);
-
-       /*
-        * Make sure one CPU is not processing packets while
-        * buffers are being released by another.
-        */
-
-       local_irq_save(flags);
-       ace_mask_irq(dev);
-
-       for (i = 0; i < ACE_TX_RING_ENTRIES(ap); i++) {
-               struct sk_buff *skb;
-               struct tx_ring_info *info;
-
-               info = ap->skb->tx_skbuff + i;
-               skb = info->skb;
-
-               if (dma_unmap_len(info, maplen)) {
-                       if (ACE_IS_TIGON_I(ap)) {
-                               /* NB: TIGON_1 is special, tx_ring is in io space */
-                               struct tx_desc __iomem *tx;
-                               tx = (__force struct tx_desc __iomem *) &ap->tx_ring[i];
-                               writel(0, &tx->addr.addrhi);
-                               writel(0, &tx->addr.addrlo);
-                               writel(0, &tx->flagsize);
-                       } else
-                               memset(ap->tx_ring + i, 0,
-                                      sizeof(struct tx_desc));
-                       dma_unmap_page(&ap->pdev->dev,
-                                      dma_unmap_addr(info, mapping),
-                                      dma_unmap_len(info, maplen),
-                                      DMA_TO_DEVICE);
-                       dma_unmap_len_set(info, maplen, 0);
-               }
-               if (skb) {
-                       dev_kfree_skb(skb);
-                       info->skb = NULL;
-               }
-       }
-
-       if (ap->jumbo) {
-               cmd.evt = C_RESET_JUMBO_RNG;
-               cmd.code = 0;
-               cmd.idx = 0;
-               ace_issue_cmd(regs, &cmd);
-       }
-
-       ace_unmask_irq(dev);
-       local_irq_restore(flags);
-
-       return 0;
-}
-
-
-static inline dma_addr_t
-ace_map_tx_skb(struct ace_private *ap, struct sk_buff *skb,
-              struct sk_buff *tail, u32 idx)
-{
-       dma_addr_t mapping;
-       struct tx_ring_info *info;
-
-       mapping = dma_map_page(&ap->pdev->dev, virt_to_page(skb->data),
-                              offset_in_page(skb->data), skb->len,
-                              DMA_TO_DEVICE);
-
-       info = ap->skb->tx_skbuff + idx;
-       info->skb = tail;
-       dma_unmap_addr_set(info, mapping, mapping);
-       dma_unmap_len_set(info, maplen, skb->len);
-       return mapping;
-}
-
-
-static inline void
-ace_load_tx_bd(struct ace_private *ap, struct tx_desc *desc, u64 addr,
-              u32 flagsize, u32 vlan_tag)
-{
-#if !USE_TX_COAL_NOW
-       flagsize &= ~BD_FLG_COAL_NOW;
-#endif
-
-       if (ACE_IS_TIGON_I(ap)) {
-               struct tx_desc __iomem *io = (__force struct tx_desc __iomem *) desc;
-               writel(addr >> 32, &io->addr.addrhi);
-               writel(addr & 0xffffffff, &io->addr.addrlo);
-               writel(flagsize, &io->flagsize);
-               writel(vlan_tag, &io->vlanres);
-       } else {
-               desc->addr.addrhi = addr >> 32;
-               desc->addr.addrlo = addr;
-               desc->flagsize = flagsize;
-               desc->vlanres = vlan_tag;
-       }
-}
-
-
-static netdev_tx_t ace_start_xmit(struct sk_buff *skb,
-                                 struct net_device *dev)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       struct tx_desc *desc;
-       u32 idx, flagsize;
-       unsigned long maxjiff = jiffies + 3*HZ;
-
-restart:
-       idx = ap->tx_prd;
-
-       if (tx_ring_full(ap, ap->tx_ret_csm, idx))
-               goto overflow;
-
-       if (!skb_shinfo(skb)->nr_frags) {
-               dma_addr_t mapping;
-               u32 vlan_tag = 0;
-
-               mapping = ace_map_tx_skb(ap, skb, skb, idx);
-               flagsize = (skb->len << 16) | (BD_FLG_END);
-               if (skb->ip_summed == CHECKSUM_PARTIAL)
-                       flagsize |= BD_FLG_TCP_UDP_SUM;
-               if (skb_vlan_tag_present(skb)) {
-                       flagsize |= BD_FLG_VLAN_TAG;
-                       vlan_tag = skb_vlan_tag_get(skb);
-               }
-               desc = ap->tx_ring + idx;
-               idx = (idx + 1) % ACE_TX_RING_ENTRIES(ap);
-
-               /* Look at ace_tx_int for explanations. */
-               if (tx_ring_full(ap, ap->tx_ret_csm, idx))
-                       flagsize |= BD_FLG_COAL_NOW;
-
-               ace_load_tx_bd(ap, desc, mapping, flagsize, vlan_tag);
-       } else {
-               dma_addr_t mapping;
-               u32 vlan_tag = 0;
-               int i;
-
-               mapping = ace_map_tx_skb(ap, skb, NULL, idx);
-               flagsize = (skb_headlen(skb) << 16);
-               if (skb->ip_summed == CHECKSUM_PARTIAL)
-                       flagsize |= BD_FLG_TCP_UDP_SUM;
-               if (skb_vlan_tag_present(skb)) {
-                       flagsize |= BD_FLG_VLAN_TAG;
-                       vlan_tag = skb_vlan_tag_get(skb);
-               }
-
-               ace_load_tx_bd(ap, ap->tx_ring + idx, mapping, flagsize, vlan_tag);
-
-               idx = (idx + 1) % ACE_TX_RING_ENTRIES(ap);
-
-               for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
-                       const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
-                       struct tx_ring_info *info;
-
-                       info = ap->skb->tx_skbuff + idx;
-                       desc = ap->tx_ring + idx;
-
-                       mapping = skb_frag_dma_map(&ap->pdev->dev, frag, 0,
-                                                  skb_frag_size(frag),
-                                                  DMA_TO_DEVICE);
-
-                       flagsize = skb_frag_size(frag) << 16;
-                       if (skb->ip_summed == CHECKSUM_PARTIAL)
-                               flagsize |= BD_FLG_TCP_UDP_SUM;
-                       idx = (idx + 1) % ACE_TX_RING_ENTRIES(ap);
-
-                       if (i == skb_shinfo(skb)->nr_frags - 1) {
-                               flagsize |= BD_FLG_END;
-                               if (tx_ring_full(ap, ap->tx_ret_csm, idx))
-                                       flagsize |= BD_FLG_COAL_NOW;
-
-                               /*
-                                * Only the last fragment frees
-                                * the skb!
-                                */
-                               info->skb = skb;
-                       } else {
-                               info->skb = NULL;
-                       }
-                       dma_unmap_addr_set(info, mapping, mapping);
-                       dma_unmap_len_set(info, maplen, skb_frag_size(frag));
-                       ace_load_tx_bd(ap, desc, mapping, flagsize, vlan_tag);
-               }
-       }
-
-       wmb();
-       ap->tx_prd = idx;
-       ace_set_txprd(regs, ap, idx);
-
-       if (flagsize & BD_FLG_COAL_NOW) {
-               netif_stop_queue(dev);
-
-               /*
-                * A TX-descriptor producer (an IRQ) might have gotten
-                * between, making the ring free again. Since xmit is
-                * serialized, this is the only situation we have to
-                * re-test.
-                */
-               if (!tx_ring_full(ap, ap->tx_ret_csm, idx))
-                       netif_wake_queue(dev);
-       }
-
-       return NETDEV_TX_OK;
-
-overflow:
-       /*
-        * This race condition is unavoidable with lock-free drivers.
-        * We wake up the queue _before_ tx_prd is advanced, so that we can
-        * enter hard_start_xmit too early, while tx ring still looks closed.
-        * This happens ~1-4 times per 100000 packets, so that we can allow
-        * to loop syncing to other CPU. Probably, we need an additional
-        * wmb() in ace_tx_intr as well.
-        *
-        * Note that this race is relieved by reserving one more entry
-        * in tx ring than it is necessary (see original non-SG driver).
-        * However, with SG we need to reserve 2*MAX_SKB_FRAGS+1, which
-        * is already overkill.
-        *
-        * Alternative is to return with 1 not throttling queue. In this
-        * case loop becomes longer, no more useful effects.
-        */
-       if (time_before(jiffies, maxjiff)) {
-               barrier();
-               cpu_relax();
-               goto restart;
-       }
-
-       /* The ring is stuck full. */
-       printk(KERN_WARNING "%s: Transmit ring stuck full\n", dev->name);
-       return NETDEV_TX_BUSY;
-}
-
-
-static int ace_change_mtu(struct net_device *dev, int new_mtu)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-
-       writel(new_mtu + ETH_HLEN + 4, &regs->IfMtu);
-       WRITE_ONCE(dev->mtu, new_mtu);
-
-       if (new_mtu > ACE_STD_MTU) {
-               if (!(ap->jumbo)) {
-                       printk(KERN_INFO "%s: Enabling Jumbo frame "
-                              "support\n", dev->name);
-                       ap->jumbo = 1;
-                       if (!test_and_set_bit(0, &ap->jumbo_refill_busy))
-                               ace_load_jumbo_rx_ring(dev, RX_JUMBO_SIZE);
-                       ace_set_rxtx_parms(dev, 1);
-               }
-       } else {
-               while (test_and_set_bit(0, &ap->jumbo_refill_busy));
-               ace_sync_irq(dev->irq);
-               ace_set_rxtx_parms(dev, 0);
-               if (ap->jumbo) {
-                       struct cmd cmd;
-
-                       cmd.evt = C_RESET_JUMBO_RNG;
-                       cmd.code = 0;
-                       cmd.idx = 0;
-                       ace_issue_cmd(regs, &cmd);
-               }
-       }
-
-       return 0;
-}
-
-static int ace_get_link_ksettings(struct net_device *dev,
-                                 struct ethtool_link_ksettings *cmd)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       u32 link;
-       u32 supported;
-
-       memset(cmd, 0, sizeof(struct ethtool_link_ksettings));
-
-       supported = (SUPPORTED_10baseT_Half | SUPPORTED_10baseT_Full |
-                    SUPPORTED_100baseT_Half | SUPPORTED_100baseT_Full |
-                    SUPPORTED_1000baseT_Half | SUPPORTED_1000baseT_Full |
-                    SUPPORTED_Autoneg | SUPPORTED_FIBRE);
-
-       cmd->base.port = PORT_FIBRE;
-
-       link = readl(&regs->GigLnkState);
-       if (link & LNK_1000MB) {
-               cmd->base.speed = SPEED_1000;
-       } else {
-               link = readl(&regs->FastLnkState);
-               if (link & LNK_100MB)
-                       cmd->base.speed = SPEED_100;
-               else if (link & LNK_10MB)
-                       cmd->base.speed = SPEED_10;
-               else
-                       cmd->base.speed = 0;
-       }
-       if (link & LNK_FULL_DUPLEX)
-               cmd->base.duplex = DUPLEX_FULL;
-       else
-               cmd->base.duplex = DUPLEX_HALF;
-
-       if (link & LNK_NEGOTIATE)
-               cmd->base.autoneg = AUTONEG_ENABLE;
-       else
-               cmd->base.autoneg = AUTONEG_DISABLE;
-
-#if 0
-       /*
-        * Current struct ethtool_cmd is insufficient
-        */
-       ecmd->trace = readl(&regs->TuneTrace);
-
-       ecmd->txcoal = readl(&regs->TuneTxCoalTicks);
-       ecmd->rxcoal = readl(&regs->TuneRxCoalTicks);
-#endif
-
-       ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
-                                               supported);
-
-       return 0;
-}
-
-static int ace_set_link_ksettings(struct net_device *dev,
-                                 const struct ethtool_link_ksettings *cmd)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       u32 link, speed;
-
-       link = readl(&regs->GigLnkState);
-       if (link & LNK_1000MB)
-               speed = SPEED_1000;
-       else {
-               link = readl(&regs->FastLnkState);
-               if (link & LNK_100MB)
-                       speed = SPEED_100;
-               else if (link & LNK_10MB)
-                       speed = SPEED_10;
-               else
-                       speed = SPEED_100;
-       }
-
-       link = LNK_ENABLE | LNK_1000MB | LNK_100MB | LNK_10MB |
-               LNK_RX_FLOW_CTL_Y | LNK_NEG_FCTL;
-       if (!ACE_IS_TIGON_I(ap))
-               link |= LNK_TX_FLOW_CTL_Y;
-       if (cmd->base.autoneg == AUTONEG_ENABLE)
-               link |= LNK_NEGOTIATE;
-       if (cmd->base.speed != speed) {
-               link &= ~(LNK_1000MB | LNK_100MB | LNK_10MB);
-               switch (cmd->base.speed) {
-               case SPEED_1000:
-                       link |= LNK_1000MB;
-                       break;
-               case SPEED_100:
-                       link |= LNK_100MB;
-                       break;
-               case SPEED_10:
-                       link |= LNK_10MB;
-                       break;
-               }
-       }
-
-       if (cmd->base.duplex == DUPLEX_FULL)
-               link |= LNK_FULL_DUPLEX;
-
-       if (link != ap->link) {
-               struct cmd cmd;
-               printk(KERN_INFO "%s: Renegotiating link state\n",
-                      dev->name);
-
-               ap->link = link;
-               writel(link, &regs->TuneLink);
-               if (!ACE_IS_TIGON_I(ap))
-                       writel(link, &regs->TuneFastLink);
-               wmb();
-
-               cmd.evt = C_LNK_NEGOTIATION;
-               cmd.code = 0;
-               cmd.idx = 0;
-               ace_issue_cmd(regs, &cmd);
-       }
-       return 0;
-}
-
-static void ace_get_drvinfo(struct net_device *dev,
-                           struct ethtool_drvinfo *info)
-{
-       struct ace_private *ap = netdev_priv(dev);
-
-       strscpy(info->driver, "acenic", sizeof(info->driver));
-       snprintf(info->fw_version, sizeof(info->version), "%i.%i.%i",
-                ap->firmware_major, ap->firmware_minor, ap->firmware_fix);
-
-       if (ap->pdev)
-               strscpy(info->bus_info, pci_name(ap->pdev),
-                       sizeof(info->bus_info));
-
-}
-
-/*
- * Set the hardware MAC address.
- */
-static int ace_set_mac_addr(struct net_device *dev, void *p)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       struct sockaddr *addr=p;
-       const u8 *da;
-       struct cmd cmd;
-
-       if(netif_running(dev))
-               return -EBUSY;
-
-       eth_hw_addr_set(dev, addr->sa_data);
-
-       da = (const u8 *)dev->dev_addr;
-
-       writel(da[0] << 8 | da[1], &regs->MacAddrHi);
-       writel((da[2] << 24) | (da[3] << 16) | (da[4] << 8) | da[5],
-              &regs->MacAddrLo);
-
-       cmd.evt = C_SET_MAC_ADDR;
-       cmd.code = 0;
-       cmd.idx = 0;
-       ace_issue_cmd(regs, &cmd);
-
-       return 0;
-}
-
-
-static void ace_set_multicast_list(struct net_device *dev)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       struct cmd cmd;
-
-       if ((dev->flags & IFF_ALLMULTI) && !(ap->mcast_all)) {
-               cmd.evt = C_SET_MULTICAST_MODE;
-               cmd.code = C_C_MCAST_ENABLE;
-               cmd.idx = 0;
-               ace_issue_cmd(regs, &cmd);
-               ap->mcast_all = 1;
-       } else if (ap->mcast_all) {
-               cmd.evt = C_SET_MULTICAST_MODE;
-               cmd.code = C_C_MCAST_DISABLE;
-               cmd.idx = 0;
-               ace_issue_cmd(regs, &cmd);
-               ap->mcast_all = 0;
-       }
-
-       if ((dev->flags & IFF_PROMISC) && !(ap->promisc)) {
-               cmd.evt = C_SET_PROMISC_MODE;
-               cmd.code = C_C_PROMISC_ENABLE;
-               cmd.idx = 0;
-               ace_issue_cmd(regs, &cmd);
-               ap->promisc = 1;
-       }else if (!(dev->flags & IFF_PROMISC) && (ap->promisc)) {
-               cmd.evt = C_SET_PROMISC_MODE;
-               cmd.code = C_C_PROMISC_DISABLE;
-               cmd.idx = 0;
-               ace_issue_cmd(regs, &cmd);
-               ap->promisc = 0;
-       }
-
-       /*
-        * For the time being multicast relies on the upper layers
-        * filtering it properly. The Firmware does not allow one to
-        * set the entire multicast list at a time and keeping track of
-        * it here is going to be messy.
-        */
-       if (!netdev_mc_empty(dev) && !ap->mcast_all) {
-               cmd.evt = C_SET_MULTICAST_MODE;
-               cmd.code = C_C_MCAST_ENABLE;
-               cmd.idx = 0;
-               ace_issue_cmd(regs, &cmd);
-       }else if (!ap->mcast_all) {
-               cmd.evt = C_SET_MULTICAST_MODE;
-               cmd.code = C_C_MCAST_DISABLE;
-               cmd.idx = 0;
-               ace_issue_cmd(regs, &cmd);
-       }
-}
-
-
-static struct net_device_stats *ace_get_stats(struct net_device *dev)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_mac_stats __iomem *mac_stats =
-               (struct ace_mac_stats __iomem *)ap->regs->Stats;
-
-       dev->stats.rx_missed_errors = readl(&mac_stats->drop_space);
-       dev->stats.multicast = readl(&mac_stats->kept_mc);
-       dev->stats.collisions = readl(&mac_stats->coll);
-
-       return &dev->stats;
-}
-
-
-static void ace_copy(struct ace_regs __iomem *regs, const __be32 *src,
-                    u32 dest, int size)
-{
-       void __iomem *tdest;
-       short tsize, i;
-
-       if (size <= 0)
-               return;
-
-       while (size > 0) {
-               tsize = min_t(u32, ((~dest & (ACE_WINDOW_SIZE - 1)) + 1),
-                           min_t(u32, size, ACE_WINDOW_SIZE));
-               tdest = (void __iomem *) &regs->Window +
-                       (dest & (ACE_WINDOW_SIZE - 1));
-               writel(dest & ~(ACE_WINDOW_SIZE - 1), &regs->WinBase);
-               for (i = 0; i < (tsize / 4); i++) {
-                       /* Firmware is big-endian */
-                       writel(be32_to_cpup(src), tdest);
-                       src++;
-                       tdest += 4;
-                       dest += 4;
-                       size -= 4;
-               }
-       }
-}
-
-
-static void ace_clear(struct ace_regs __iomem *regs, u32 dest, int size)
-{
-       void __iomem *tdest;
-       short tsize = 0, i;
-
-       if (size <= 0)
-               return;
-
-       while (size > 0) {
-               tsize = min_t(u32, ((~dest & (ACE_WINDOW_SIZE - 1)) + 1),
-                               min_t(u32, size, ACE_WINDOW_SIZE));
-               tdest = (void __iomem *) &regs->Window +
-                       (dest & (ACE_WINDOW_SIZE - 1));
-               writel(dest & ~(ACE_WINDOW_SIZE - 1), &regs->WinBase);
-
-               for (i = 0; i < (tsize / 4); i++) {
-                       writel(0, tdest + i*4);
-               }
-
-               dest += tsize;
-               size -= tsize;
-       }
-}
-
-
-/*
- * Download the firmware into the SRAM on the NIC
- *
- * This operation requires the NIC to be halted and is performed with
- * interrupts disabled and with the spinlock hold.
- */
-static int ace_load_firmware(struct net_device *dev)
-{
-       const struct firmware *fw;
-       const char *fw_name = "acenic/tg2.bin";
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       const __be32 *fw_data;
-       u32 load_addr;
-       int ret;
-
-       if (!(readl(&regs->CpuCtrl) & CPU_HALTED)) {
-               printk(KERN_ERR "%s: trying to download firmware while the "
-                      "CPU is running!\n", ap->name);
-               return -EFAULT;
-       }
-
-       if (ACE_IS_TIGON_I(ap))
-               fw_name = "acenic/tg1.bin";
-
-       ret = request_firmware(&fw, fw_name, &ap->pdev->dev);
-       if (ret) {
-               printk(KERN_ERR "%s: Failed to load firmware \"%s\"\n",
-                      ap->name, fw_name);
-               return ret;
-       }
-
-       fw_data = (void *)fw->data;
-
-       /* Firmware blob starts with version numbers, followed by
-          load and start address. Remainder is the blob to be loaded
-          contiguously from load address. We don't bother to represent
-          the BSS/SBSS sections any more, since we were clearing the
-          whole thing anyway. */
-       ap->firmware_major = fw->data[0];
-       ap->firmware_minor = fw->data[1];
-       ap->firmware_fix = fw->data[2];
-
-       ap->firmware_start = be32_to_cpu(fw_data[1]);
-       if (ap->firmware_start < 0x4000 || ap->firmware_start >= 0x80000) {
-               printk(KERN_ERR "%s: bogus load address %08x in \"%s\"\n",
-                      ap->name, ap->firmware_start, fw_name);
-               ret = -EINVAL;
-               goto out;
-       }
-
-       load_addr = be32_to_cpu(fw_data[2]);
-       if (load_addr < 0x4000 || load_addr >= 0x80000) {
-               printk(KERN_ERR "%s: bogus load address %08x in \"%s\"\n",
-                      ap->name, load_addr, fw_name);
-               ret = -EINVAL;
-               goto out;
-       }
-
-       /*
-        * Do not try to clear more than 512KiB or we end up seeing
-        * funny things on NICs with only 512KiB SRAM
-        */
-       ace_clear(regs, 0x2000, 0x80000-0x2000);
-       ace_copy(regs, &fw_data[3], load_addr, fw->size-12);
- out:
-       release_firmware(fw);
-       return ret;
-}
-
-
-/*
- * The eeprom on the AceNIC is an Atmel i2c EEPROM.
- *
- * Accessing the EEPROM is `interesting' to say the least - don't read
- * this code right after dinner.
- *
- * This is all about black magic and bit-banging the device .... I
- * wonder in what hospital they have put the guy who designed the i2c
- * specs.
- *
- * Oh yes, this is only the beginning!
- *
- * Thanks to Stevarino Webinski for helping tracking down the bugs in the
- * code i2c readout code by beta testing all my hacks.
- */
-static void eeprom_start(struct ace_regs __iomem *regs)
-{
-       u32 local;
-
-       readl(&regs->LocalCtrl);
-       udelay(ACE_SHORT_DELAY);
-       local = readl(&regs->LocalCtrl);
-       local |= EEPROM_DATA_OUT | EEPROM_WRITE_ENABLE;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_SHORT_DELAY);
-       local |= EEPROM_CLK_OUT;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_SHORT_DELAY);
-       local &= ~EEPROM_DATA_OUT;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_SHORT_DELAY);
-       local &= ~EEPROM_CLK_OUT;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-}
-
-
-static void eeprom_prep(struct ace_regs __iomem *regs, u8 magic)
-{
-       short i;
-       u32 local;
-
-       udelay(ACE_SHORT_DELAY);
-       local = readl(&regs->LocalCtrl);
-       local &= ~EEPROM_DATA_OUT;
-       local |= EEPROM_WRITE_ENABLE;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-
-       for (i = 0; i < 8; i++, magic <<= 1) {
-               udelay(ACE_SHORT_DELAY);
-               if (magic & 0x80)
-                       local |= EEPROM_DATA_OUT;
-               else
-                       local &= ~EEPROM_DATA_OUT;
-               writel(local, &regs->LocalCtrl);
-               readl(&regs->LocalCtrl);
-               mb();
-
-               udelay(ACE_SHORT_DELAY);
-               local |= EEPROM_CLK_OUT;
-               writel(local, &regs->LocalCtrl);
-               readl(&regs->LocalCtrl);
-               mb();
-               udelay(ACE_SHORT_DELAY);
-               local &= ~(EEPROM_CLK_OUT | EEPROM_DATA_OUT);
-               writel(local, &regs->LocalCtrl);
-               readl(&regs->LocalCtrl);
-               mb();
-       }
-}
-
-
-static int eeprom_check_ack(struct ace_regs __iomem *regs)
-{
-       int state;
-       u32 local;
-
-       local = readl(&regs->LocalCtrl);
-       local &= ~EEPROM_WRITE_ENABLE;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_LONG_DELAY);
-       local |= EEPROM_CLK_OUT;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_SHORT_DELAY);
-       /* sample data in middle of high clk */
-       state = (readl(&regs->LocalCtrl) & EEPROM_DATA_IN) != 0;
-       udelay(ACE_SHORT_DELAY);
-       mb();
-       writel(readl(&regs->LocalCtrl) & ~EEPROM_CLK_OUT, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-
-       return state;
-}
-
-
-static void eeprom_stop(struct ace_regs __iomem *regs)
-{
-       u32 local;
-
-       udelay(ACE_SHORT_DELAY);
-       local = readl(&regs->LocalCtrl);
-       local |= EEPROM_WRITE_ENABLE;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_SHORT_DELAY);
-       local &= ~EEPROM_DATA_OUT;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_SHORT_DELAY);
-       local |= EEPROM_CLK_OUT;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_SHORT_DELAY);
-       local |= EEPROM_DATA_OUT;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_LONG_DELAY);
-       local &= ~EEPROM_CLK_OUT;
-       writel(local, &regs->LocalCtrl);
-       mb();
-}
-
-
-/*
- * Read a whole byte from the EEPROM.
- */
-static int read_eeprom_byte(struct net_device *dev, unsigned long offset)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-       unsigned long flags;
-       u32 local;
-       int result = 0;
-       short i;
-
-       /*
-        * Don't take interrupts on this CPU will bit banging
-        * the %#%#@$ I2C device
-        */
-       local_irq_save(flags);
-
-       eeprom_start(regs);
-
-       eeprom_prep(regs, EEPROM_WRITE_SELECT);
-       if (eeprom_check_ack(regs)) {
-               local_irq_restore(flags);
-               printk(KERN_ERR "%s: Unable to sync eeprom\n", ap->name);
-               result = -EIO;
-               goto eeprom_read_error;
-       }
-
-       eeprom_prep(regs, (offset >> 8) & 0xff);
-       if (eeprom_check_ack(regs)) {
-               local_irq_restore(flags);
-               printk(KERN_ERR "%s: Unable to set address byte 0\n",
-                      ap->name);
-               result = -EIO;
-               goto eeprom_read_error;
-       }
-
-       eeprom_prep(regs, offset & 0xff);
-       if (eeprom_check_ack(regs)) {
-               local_irq_restore(flags);
-               printk(KERN_ERR "%s: Unable to set address byte 1\n",
-                      ap->name);
-               result = -EIO;
-               goto eeprom_read_error;
-       }
-
-       eeprom_start(regs);
-       eeprom_prep(regs, EEPROM_READ_SELECT);
-       if (eeprom_check_ack(regs)) {
-               local_irq_restore(flags);
-               printk(KERN_ERR "%s: Unable to set READ_SELECT\n",
-                      ap->name);
-               result = -EIO;
-               goto eeprom_read_error;
-       }
-
-       for (i = 0; i < 8; i++) {
-               local = readl(&regs->LocalCtrl);
-               local &= ~EEPROM_WRITE_ENABLE;
-               writel(local, &regs->LocalCtrl);
-               readl(&regs->LocalCtrl);
-               udelay(ACE_LONG_DELAY);
-               mb();
-               local |= EEPROM_CLK_OUT;
-               writel(local, &regs->LocalCtrl);
-               readl(&regs->LocalCtrl);
-               mb();
-               udelay(ACE_SHORT_DELAY);
-               /* sample data mid high clk */
-               result = (result << 1) |
-                       ((readl(&regs->LocalCtrl) & EEPROM_DATA_IN) != 0);
-               udelay(ACE_SHORT_DELAY);
-               mb();
-               local = readl(&regs->LocalCtrl);
-               local &= ~EEPROM_CLK_OUT;
-               writel(local, &regs->LocalCtrl);
-               readl(&regs->LocalCtrl);
-               udelay(ACE_SHORT_DELAY);
-               mb();
-               if (i == 7) {
-                       local |= EEPROM_WRITE_ENABLE;
-                       writel(local, &regs->LocalCtrl);
-                       readl(&regs->LocalCtrl);
-                       mb();
-                       udelay(ACE_SHORT_DELAY);
-               }
-       }
-
-       local |= EEPROM_DATA_OUT;
-       writel(local, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_SHORT_DELAY);
-       writel(readl(&regs->LocalCtrl) | EEPROM_CLK_OUT, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       udelay(ACE_LONG_DELAY);
-       writel(readl(&regs->LocalCtrl) & ~EEPROM_CLK_OUT, &regs->LocalCtrl);
-       readl(&regs->LocalCtrl);
-       mb();
-       udelay(ACE_SHORT_DELAY);
-       eeprom_stop(regs);
-
-       local_irq_restore(flags);
- out:
-       return result;
-
- eeprom_read_error:
-       printk(KERN_ERR "%s: Unable to read eeprom byte 0x%02lx\n",
-              ap->name, offset);
-       goto out;
-}
-
-module_pci_driver(acenic_pci_driver);
diff --git a/drivers/net/ethernet/alteon/acenic.h b/drivers/net/ethernet/alteon/acenic.h
deleted file mode 100644 (file)
index 0e45a97..0000000
+++ /dev/null
@@ -1,791 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _ACENIC_H_
-#define _ACENIC_H_
-#include <linux/interrupt.h>
-#include <linux/workqueue.h>
-
-/*
- * Generate TX index update each time, when TX ring is closed.
- * Normally, this is not useful, because results in more dma (and irqs
- * without TX_COAL_INTS_ONLY).
- */
-#define USE_TX_COAL_NOW         0
-
-/*
- * Addressing:
- *
- * The Tigon uses 64-bit host addresses, regardless of their actual
- * length, and it expects a big-endian format. For 32 bit systems the
- * upper 32 bits of the address are simply ignored (zero), however for
- * little endian 64 bit systems (Alpha) this looks strange with the
- * two parts of the address word being swapped.
- *
- * The addresses are split in two 32 bit words for all architectures
- * as some of them are in PCI shared memory and it is necessary to use
- * readl/writel to access them.
- *
- * The addressing code is derived from Pete Wyckoff's work, but
- * modified to deal properly with readl/writel usage.
- */
-
-struct ace_regs {
-       u32     pad0[16];       /* PCI control registers */
-
-       u32     HostCtrl;       /* 0x40 */
-       u32     LocalCtrl;
-
-       u32     pad1[2];
-
-       u32     MiscCfg;        /* 0x50 */
-
-       u32     pad2[2];
-
-       u32     PciState;
-
-       u32     pad3[2];        /* 0x60 */
-
-       u32     WinBase;
-       u32     WinData;
-
-       u32     pad4[12];       /* 0x70 */
-
-       u32     DmaWriteState;  /* 0xa0 */
-       u32     pad5[3];
-       u32     DmaReadState;   /* 0xb0 */
-
-       u32     pad6[26];
-
-       u32     AssistState;
-
-       u32     pad7[8];        /* 0x120 */
-
-       u32     CpuCtrl;        /* 0x140 */
-       u32     Pc;
-
-       u32     pad8[3];
-
-       u32     SramAddr;       /* 0x154 */
-       u32     SramData;
-
-       u32     pad9[49];
-
-       u32     MacRxState;     /* 0x220 */
-
-       u32     pad10[7];
-
-       u32     CpuBCtrl;       /* 0x240 */
-       u32     PcB;
-
-       u32     pad11[3];
-
-       u32     SramBAddr;      /* 0x254 */
-       u32     SramBData;
-
-       u32     pad12[105];
-
-       u32     pad13[32];      /* 0x400 */
-       u32     Stats[32];
-
-       u32     Mb0Hi;          /* 0x500 */
-       u32     Mb0Lo;
-       u32     Mb1Hi;
-       u32     CmdPrd;
-       u32     Mb2Hi;
-       u32     TxPrd;
-       u32     Mb3Hi;
-       u32     RxStdPrd;
-       u32     Mb4Hi;
-       u32     RxJumboPrd;
-       u32     Mb5Hi;
-       u32     RxMiniPrd;
-       u32     Mb6Hi;
-       u32     Mb6Lo;
-       u32     Mb7Hi;
-       u32     Mb7Lo;
-       u32     Mb8Hi;
-       u32     Mb8Lo;
-       u32     Mb9Hi;
-       u32     Mb9Lo;
-       u32     MbAHi;
-       u32     MbALo;
-       u32     MbBHi;
-       u32     MbBLo;
-       u32     MbCHi;
-       u32     MbCLo;
-       u32     MbDHi;
-       u32     MbDLo;
-       u32     MbEHi;
-       u32     MbELo;
-       u32     MbFHi;
-       u32     MbFLo;
-
-       u32     pad14[32];
-
-       u32     MacAddrHi;      /* 0x600 */
-       u32     MacAddrLo;
-       u32     InfoPtrHi;
-       u32     InfoPtrLo;
-       u32     MultiCastHi;    /* 0x610 */
-       u32     MultiCastLo;
-       u32     ModeStat;
-       u32     DmaReadCfg;
-       u32     DmaWriteCfg;    /* 0x620 */
-       u32     TxBufRat;
-       u32     EvtCsm;
-       u32     CmdCsm;
-       u32     TuneRxCoalTicks;/* 0x630 */
-       u32     TuneTxCoalTicks;
-       u32     TuneStatTicks;
-       u32     TuneMaxTxDesc;
-       u32     TuneMaxRxDesc;  /* 0x640 */
-       u32     TuneTrace;
-       u32     TuneLink;
-       u32     TuneFastLink;
-       u32     TracePtr;       /* 0x650 */
-       u32     TraceStrt;
-       u32     TraceLen;
-       u32     IfIdx;
-       u32     IfMtu;          /* 0x660 */
-       u32     MaskInt;
-       u32     GigLnkState;
-       u32     FastLnkState;
-       u32     pad16[4];       /* 0x670 */
-       u32     RxRetCsm;       /* 0x680 */
-
-       u32     pad17[31];
-
-       u32     CmdRng[64];     /* 0x700 */
-       u32     Window[0x200];
-};
-
-
-typedef struct {
-       u32 addrhi;
-       u32 addrlo;
-} aceaddr;
-
-
-#define ACE_WINDOW_SIZE        0x800
-
-#define ACE_JUMBO_MTU 9000
-#define ACE_STD_MTU 1500
-
-#define ACE_TRACE_SIZE 0x8000
-
-/*
- * Host control register bits.
- */
-
-#define IN_INT         0x01
-#define CLR_INT                0x02
-#define HW_RESET       0x08
-#define BYTE_SWAP      0x10
-#define WORD_SWAP      0x20
-#define MASK_INTS      0x40
-
-/*
- * Local control register bits.
- */
-
-#define EEPROM_DATA_IN         0x800000
-#define EEPROM_DATA_OUT                0x400000
-#define EEPROM_WRITE_ENABLE    0x200000
-#define EEPROM_CLK_OUT         0x100000
-
-#define EEPROM_BASE            0xa0000000
-
-#define EEPROM_WRITE_SELECT    0xa0
-#define EEPROM_READ_SELECT     0xa1
-
-#define SRAM_BANK_512K         0x200
-
-
-/*
- * udelay() values for when clocking the eeprom
- */
-#define ACE_SHORT_DELAY                2
-#define ACE_LONG_DELAY         4
-
-
-/*
- * Misc Config bits
- */
-
-#define SYNC_SRAM_TIMING       0x100000
-
-
-/*
- * CPU state bits.
- */
-
-#define CPU_RESET              0x01
-#define CPU_TRACE              0x02
-#define CPU_PROM_FAILED                0x10
-#define CPU_HALT               0x00010000
-#define CPU_HALTED             0xffff0000
-
-
-/*
- * PCI State bits.
- */
-
-#define DMA_READ_MAX_4         0x04
-#define DMA_READ_MAX_16                0x08
-#define DMA_READ_MAX_32                0x0c
-#define DMA_READ_MAX_64                0x10
-#define DMA_READ_MAX_128       0x14
-#define DMA_READ_MAX_256       0x18
-#define DMA_READ_MAX_1K                0x1c
-#define DMA_WRITE_MAX_4                0x20
-#define DMA_WRITE_MAX_16       0x40
-#define DMA_WRITE_MAX_32       0x60
-#define DMA_WRITE_MAX_64       0x80
-#define DMA_WRITE_MAX_128      0xa0
-#define DMA_WRITE_MAX_256      0xc0
-#define DMA_WRITE_MAX_1K       0xe0
-#define DMA_READ_WRITE_MASK    0xfc
-#define MEM_READ_MULTIPLE      0x00020000
-#define PCI_66MHZ              0x00080000
-#define PCI_32BIT              0x00100000
-#define DMA_WRITE_ALL_ALIGN    0x00800000
-#define READ_CMD_MEM           0x06000000
-#define WRITE_CMD_MEM          0x70000000
-
-
-/*
- * Mode status
- */
-
-#define ACE_BYTE_SWAP_BD       0x02
-#define ACE_WORD_SWAP_BD       0x04            /* not actually used */
-#define ACE_WARN               0x08
-#define ACE_BYTE_SWAP_DMA      0x10
-#define ACE_NO_JUMBO_FRAG      0x200
-#define ACE_FATAL              0x40000000
-
-
-/*
- * DMA config
- */
-
-#define DMA_THRESH_1W          0x10
-#define DMA_THRESH_2W          0x20
-#define DMA_THRESH_4W          0x40
-#define DMA_THRESH_8W          0x80
-#define DMA_THRESH_16W         0x100
-#define DMA_THRESH_32W         0x0     /* not described in doc, but exists. */
-
-
-/*
- * Tuning parameters
- */
-
-#define TICKS_PER_SEC          1000000
-
-
-/*
- * Link bits
- */
-
-#define LNK_PREF               0x00008000
-#define LNK_10MB               0x00010000
-#define LNK_100MB              0x00020000
-#define LNK_1000MB             0x00040000
-#define LNK_FULL_DUPLEX                0x00080000
-#define LNK_HALF_DUPLEX                0x00100000
-#define LNK_TX_FLOW_CTL_Y      0x00200000
-#define LNK_NEG_ADVANCED       0x00400000
-#define LNK_RX_FLOW_CTL_Y      0x00800000
-#define LNK_NIC                        0x01000000
-#define LNK_JAM                        0x02000000
-#define LNK_JUMBO              0x04000000
-#define LNK_ALTEON             0x08000000
-#define LNK_NEG_FCTL           0x10000000
-#define LNK_NEGOTIATE          0x20000000
-#define LNK_ENABLE             0x40000000
-#define LNK_UP                 0x80000000
-
-
-/*
- * Event definitions
- */
-
-#define EVT_RING_ENTRIES       256
-#define EVT_RING_SIZE  (EVT_RING_ENTRIES * sizeof(struct event))
-
-struct event {
-#ifdef __LITTLE_ENDIAN_BITFIELD
-       u32     idx:12;
-       u32     code:12;
-       u32     evt:8;
-#else
-       u32     evt:8;
-       u32     code:12;
-       u32     idx:12;
-#endif
-       u32     pad;
-};
-
-
-/*
- * Events
- */
-
-#define E_FW_RUNNING           0x01
-#define E_STATS_UPDATED                0x04
-
-#define E_STATS_UPDATE         0x04
-
-#define E_LNK_STATE            0x06
-#define E_C_LINK_UP            0x01
-#define E_C_LINK_DOWN          0x02
-#define E_C_LINK_10_100                0x03
-
-#define E_ERROR                        0x07
-#define E_C_ERR_INVAL_CMD      0x01
-#define E_C_ERR_UNIMP_CMD      0x02
-#define E_C_ERR_BAD_CFG                0x03
-
-#define E_MCAST_LIST           0x08
-#define E_C_MCAST_ADDR_ADD     0x01
-#define E_C_MCAST_ADDR_DEL     0x02
-
-#define E_RESET_JUMBO_RNG      0x09
-
-
-/*
- * Commands
- */
-
-#define CMD_RING_ENTRIES       64
-
-struct cmd {
-#ifdef __LITTLE_ENDIAN_BITFIELD
-       u32     idx:12;
-       u32     code:12;
-       u32     evt:8;
-#else
-       u32     evt:8;
-       u32     code:12;
-       u32     idx:12;
-#endif
-};
-
-
-#define C_HOST_STATE           0x01
-#define C_C_STACK_UP           0x01
-#define C_C_STACK_DOWN         0x02
-
-#define C_FDR_FILTERING                0x02
-#define C_C_FDR_FILT_ENABLE    0x01
-#define C_C_FDR_FILT_DISABLE   0x02
-
-#define C_SET_RX_PRD_IDX       0x03
-#define C_UPDATE_STATS         0x04
-#define C_RESET_JUMBO_RNG      0x05
-#define C_ADD_MULTICAST_ADDR   0x08
-#define C_DEL_MULTICAST_ADDR   0x09
-
-#define C_SET_PROMISC_MODE     0x0a
-#define C_C_PROMISC_ENABLE     0x01
-#define C_C_PROMISC_DISABLE    0x02
-
-#define C_LNK_NEGOTIATION      0x0b
-#define C_C_NEGOTIATE_BOTH     0x00
-#define C_C_NEGOTIATE_GIG      0x01
-#define C_C_NEGOTIATE_10_100   0x02
-
-#define C_SET_MAC_ADDR         0x0c
-#define C_CLEAR_PROFILE                0x0d
-
-#define C_SET_MULTICAST_MODE   0x0e
-#define C_C_MCAST_ENABLE       0x01
-#define C_C_MCAST_DISABLE      0x02
-
-#define C_CLEAR_STATS          0x0f
-#define C_SET_RX_JUMBO_PRD_IDX 0x10
-#define C_REFRESH_STATS                0x11
-
-
-/*
- * Descriptor flags
- */
-#define BD_FLG_TCP_UDP_SUM     0x01
-#define BD_FLG_IP_SUM          0x02
-#define BD_FLG_END             0x04
-#define BD_FLG_MORE            0x08
-#define BD_FLG_JUMBO           0x10
-#define BD_FLG_UCAST           0x20
-#define BD_FLG_MCAST           0x40
-#define BD_FLG_BCAST           0x60
-#define BD_FLG_TYP_MASK                0x60
-#define BD_FLG_IP_FRAG         0x80
-#define BD_FLG_IP_FRAG_END     0x100
-#define BD_FLG_VLAN_TAG                0x200
-#define BD_FLG_FRAME_ERROR     0x400
-#define BD_FLG_COAL_NOW                0x800
-#define BD_FLG_MINI            0x1000
-
-
-/*
- * Ring Control block flags
- */
-#define RCB_FLG_TCP_UDP_SUM    0x01
-#define RCB_FLG_IP_SUM         0x02
-#define RCB_FLG_NO_PSEUDO_HDR  0x08
-#define RCB_FLG_VLAN_ASSIST    0x10
-#define RCB_FLG_COAL_INT_ONLY  0x20
-#define RCB_FLG_TX_HOST_RING   0x40
-#define RCB_FLG_IEEE_SNAP_SUM  0x80
-#define RCB_FLG_EXT_RX_BD      0x100
-#define RCB_FLG_RNG_DISABLE    0x200
-
-
-/*
- * TX ring - maximum TX ring entries for Tigon I's is 128
- */
-#define MAX_TX_RING_ENTRIES    256
-#define TIGON_I_TX_RING_ENTRIES        128
-#define TX_RING_SIZE           (MAX_TX_RING_ENTRIES * sizeof(struct tx_desc))
-#define TX_RING_BASE           0x3800
-
-struct tx_desc{
-        aceaddr        addr;
-       u32     flagsize;
-#if 0
-/*
- * This is in PCI shared mem and must be accessed with readl/writel
- * real layout is:
- */
-#if __LITTLE_ENDIAN
-       u16     flags;
-       u16     size;
-       u16     vlan;
-       u16     reserved;
-#else
-       u16     size;
-       u16     flags;
-       u16     reserved;
-       u16     vlan;
-#endif
-#endif
-       u32     vlanres;
-};
-
-
-#define RX_STD_RING_ENTRIES    512
-#define RX_STD_RING_SIZE       (RX_STD_RING_ENTRIES * sizeof(struct rx_desc))
-
-#define RX_JUMBO_RING_ENTRIES  256
-#define RX_JUMBO_RING_SIZE     (RX_JUMBO_RING_ENTRIES *sizeof(struct rx_desc))
-
-#define RX_MINI_RING_ENTRIES   1024
-#define RX_MINI_RING_SIZE      (RX_MINI_RING_ENTRIES *sizeof(struct rx_desc))
-
-#define RX_RETURN_RING_ENTRIES 2048
-#define RX_RETURN_RING_SIZE    (RX_MAX_RETURN_RING_ENTRIES * \
-                                sizeof(struct rx_desc))
-
-struct rx_desc{
-       aceaddr addr;
-#ifdef __LITTLE_ENDIAN
-       u16     size;
-       u16     idx;
-#else
-       u16     idx;
-       u16     size;
-#endif
-#ifdef __LITTLE_ENDIAN
-       u16     flags;
-       u16     type;
-#else
-       u16     type;
-       u16     flags;
-#endif
-#ifdef __LITTLE_ENDIAN
-       u16     tcp_udp_csum;
-       u16     ip_csum;
-#else
-       u16     ip_csum;
-       u16     tcp_udp_csum;
-#endif
-#ifdef __LITTLE_ENDIAN
-       u16     vlan;
-       u16     err_flags;
-#else
-       u16     err_flags;
-       u16     vlan;
-#endif
-       u32     reserved;
-       u32     opague;
-};
-
-
-/*
- * This struct is shared with the NIC firmware.
- */
-struct ring_ctrl {
-       aceaddr rngptr;
-#ifdef __LITTLE_ENDIAN
-       u16     flags;
-       u16     max_len;
-#else
-       u16     max_len;
-       u16     flags;
-#endif
-       u32     pad;
-};
-
-
-struct ace_mac_stats {
-       u32 excess_colls;
-       u32 coll_1;
-       u32 coll_2;
-       u32 coll_3;
-       u32 coll_4;
-       u32 coll_5;
-       u32 coll_6;
-       u32 coll_7;
-       u32 coll_8;
-       u32 coll_9;
-       u32 coll_10;
-       u32 coll_11;
-       u32 coll_12;
-       u32 coll_13;
-       u32 coll_14;
-       u32 coll_15;
-       u32 late_coll;
-       u32 defers;
-       u32 crc_err;
-       u32 underrun;
-       u32 crs_err;
-       u32 pad[3];
-       u32 drop_ula;
-       u32 drop_mc;
-       u32 drop_fc;
-       u32 drop_space;
-       u32 coll;
-       u32 kept_bc;
-       u32 kept_mc;
-       u32 kept_uc;
-};
-
-
-struct ace_info {
-       union {
-               u32 stats[256];
-       } s;
-       struct ring_ctrl        evt_ctrl;
-       struct ring_ctrl        cmd_ctrl;
-       struct ring_ctrl        tx_ctrl;
-       struct ring_ctrl        rx_std_ctrl;
-       struct ring_ctrl        rx_jumbo_ctrl;
-       struct ring_ctrl        rx_mini_ctrl;
-       struct ring_ctrl        rx_return_ctrl;
-       aceaddr evt_prd_ptr;
-       aceaddr rx_ret_prd_ptr;
-       aceaddr tx_csm_ptr;
-       aceaddr stats2_ptr;
-};
-
-
-struct ring_info {
-       struct sk_buff          *skb;
-       DEFINE_DMA_UNMAP_ADDR(mapping);
-};
-
-
-/*
- * Funny... As soon as we add maplen on alpha, it starts to work
- * much slower. Hmm... is it because struct does not fit to one cacheline?
- * So, split tx_ring_info.
- */
-struct tx_ring_info {
-       struct sk_buff          *skb;
-       DEFINE_DMA_UNMAP_ADDR(mapping);
-       DEFINE_DMA_UNMAP_LEN(maplen);
-};
-
-
-/*
- * struct ace_skb holding the rings of skb's. This is an awful lot of
- * pointers, but I don't see any other smart mode to do this in an
- * efficient manner ;-(
- */
-struct ace_skb
-{
-       struct tx_ring_info     tx_skbuff[MAX_TX_RING_ENTRIES];
-       struct ring_info        rx_std_skbuff[RX_STD_RING_ENTRIES];
-       struct ring_info        rx_mini_skbuff[RX_MINI_RING_ENTRIES];
-       struct ring_info        rx_jumbo_skbuff[RX_JUMBO_RING_ENTRIES];
-};
-
-
-/*
- * Struct private for the AceNIC.
- *
- * Elements are grouped so variables used by the tx handling goes
- * together, and will go into the same cache lines etc. in order to
- * avoid cache line contention between the rx and tx handling on SMP.
- *
- * Frequently accessed variables are put at the beginning of the
- * struct to help the compiler generate better/shorter code.
- */
-struct ace_private
-{
-       struct net_device       *ndev;          /* backpointer */
-       struct ace_info         *info;
-       struct ace_regs __iomem *regs;          /* register base */
-       struct ace_skb          *skb;
-       dma_addr_t              info_dma;       /* 32/64 bit */
-
-       int                     version, link;
-       int                     promisc, mcast_all;
-
-       /*
-        * TX elements
-        */
-       struct tx_desc          *tx_ring;
-       u32                     tx_prd;
-       volatile u32            tx_ret_csm;
-       int                     tx_ring_entries;
-
-       /*
-        * RX elements
-        */
-       unsigned long           std_refill_busy
-                               __attribute__ ((aligned (SMP_CACHE_BYTES)));
-       unsigned long           mini_refill_busy, jumbo_refill_busy;
-       atomic_t                cur_rx_bufs;
-       atomic_t                cur_mini_bufs;
-       atomic_t                cur_jumbo_bufs;
-       u32                     rx_std_skbprd, rx_mini_skbprd, rx_jumbo_skbprd;
-       u32                     cur_rx;
-
-       struct rx_desc          *rx_std_ring;
-       struct rx_desc          *rx_jumbo_ring;
-       struct rx_desc          *rx_mini_ring;
-       struct rx_desc          *rx_return_ring;
-
-       int                     bh_work_pending, jumbo;
-       struct work_struct      ace_bh_work;
-
-       struct event            *evt_ring;
-
-       volatile u32            *evt_prd, *rx_ret_prd, *tx_csm;
-
-       dma_addr_t              tx_ring_dma;    /* 32/64 bit */
-       dma_addr_t              rx_ring_base_dma;
-       dma_addr_t              evt_ring_dma;
-       dma_addr_t              evt_prd_dma, rx_ret_prd_dma, tx_csm_dma;
-
-       unsigned char           *trace_buf;
-       struct pci_dev          *pdev;
-       struct net_device       *next;
-       volatile int            fw_running;
-       int                     board_idx;
-       u16                     pci_command;
-       u8                      pci_latency;
-       const char              *name;
-#ifdef INDEX_DEBUG
-       spinlock_t              debug_lock
-                               __attribute__ ((aligned (SMP_CACHE_BYTES)));
-       u32                     last_tx, last_std_rx, last_mini_rx;
-#endif
-       u8                      firmware_major;
-       u8                      firmware_minor;
-       u8                      firmware_fix;
-       u32                     firmware_start;
-};
-
-
-#define TX_RESERVED    MAX_SKB_FRAGS
-
-static inline int tx_space (struct ace_private *ap, u32 csm, u32 prd)
-{
-       return (csm - prd - 1) & (ACE_TX_RING_ENTRIES(ap) - 1);
-}
-
-#define tx_free(ap)            tx_space((ap)->tx_ret_csm, (ap)->tx_prd, ap)
-#define tx_ring_full(ap, csm, prd)     (tx_space(ap, csm, prd) <= TX_RESERVED)
-
-static inline void set_aceaddr(aceaddr *aa, dma_addr_t addr)
-{
-       u64 baddr = (u64) addr;
-       aa->addrlo = baddr & 0xffffffff;
-       aa->addrhi = baddr >> 32;
-       wmb();
-}
-
-
-static inline void ace_set_txprd(struct ace_regs __iomem *regs,
-                                struct ace_private *ap, u32 value)
-{
-#ifdef INDEX_DEBUG
-       unsigned long flags;
-       spin_lock_irqsave(&ap->debug_lock, flags);
-       writel(value, &regs->TxPrd);
-       if (value == ap->last_tx)
-               printk(KERN_ERR "AceNIC RACE ALERT! writing identical value "
-                      "to tx producer (%i)\n", value);
-       ap->last_tx = value;
-       spin_unlock_irqrestore(&ap->debug_lock, flags);
-#else
-       writel(value, &regs->TxPrd);
-#endif
-       wmb();
-}
-
-
-static inline void ace_mask_irq(struct net_device *dev)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-
-       if (ACE_IS_TIGON_I(ap))
-               writel(1, &regs->MaskInt);
-       else
-               writel(readl(&regs->HostCtrl) | MASK_INTS, &regs->HostCtrl);
-
-       ace_sync_irq(dev->irq);
-}
-
-
-static inline void ace_unmask_irq(struct net_device *dev)
-{
-       struct ace_private *ap = netdev_priv(dev);
-       struct ace_regs __iomem *regs = ap->regs;
-
-       if (ACE_IS_TIGON_I(ap))
-               writel(0, &regs->MaskInt);
-       else
-               writel(readl(&regs->HostCtrl) & ~MASK_INTS, &regs->HostCtrl);
-}
-
-
-/*
- * Prototypes
- */
-static int ace_init(struct net_device *dev);
-static void ace_load_std_rx_ring(struct net_device *dev, int nr_bufs);
-static void ace_load_mini_rx_ring(struct net_device *dev, int nr_bufs);
-static void ace_load_jumbo_rx_ring(struct net_device *dev, int nr_bufs);
-static irqreturn_t ace_interrupt(int irq, void *dev_id);
-static int ace_load_firmware(struct net_device *dev);
-static int ace_open(struct net_device *dev);
-static netdev_tx_t ace_start_xmit(struct sk_buff *skb,
-                                 struct net_device *dev);
-static int ace_close(struct net_device *dev);
-static void ace_bh_work(struct work_struct *work);
-static void ace_dump_trace(struct ace_private *ap);
-static void ace_set_multicast_list(struct net_device *dev);
-static int ace_change_mtu(struct net_device *dev, int new_mtu);
-static int ace_set_mac_addr(struct net_device *dev, void *p);
-static void ace_set_rxtx_parms(struct net_device *dev, int jumbo);
-static int ace_allocate_descriptors(struct net_device *dev);
-static void ace_free_descriptors(struct net_device *dev);
-static void ace_init_cleanup(struct net_device *dev);
-static struct net_device_stats *ace_get_stats(struct net_device *dev);
-static int read_eeprom_byte(struct net_device *dev, unsigned long offset);
-
-#endif /* _ACENIC_H_ */