Gitiles
Code Review Sign In
code.illumos.org / illumos-gate / bbb9d5d65bf8372aae4b8821c80e218b8b832846 / . / usr / src / uts / common / io / e1000g / e1000g_main.c
blob: de4ebc29718ff99f18db8d398c4f430afe18d531 [file] [log] [blame]
/*
* This file is provided under a CDDLv1 license. When using or
* redistributing this file, you may do so under this license.
* In redistributing this file this license must be included
* and no other modification of this header file is permitted.
*
* CDDL LICENSE SUMMARY
*
* Copyright(c) 1999 - 2009 Intel Corporation. All rights reserved.
*
* The contents of this file are subject to the terms of Version
* 1.0 of the Common Development and Distribution License (the "License").
*
* You should have received a copy of the License with this software.
* You can obtain a copy of the License at
* http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*/
/*
* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
*/
/*
* Copyright 2012 DEY Storage Systems, Inc. All rights reserved.
* Copyright 2013 Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2017, Joyent, Inc.
*/
/*
* **********************************************************************
* *
* Module Name: *
* e1000g_main.c *
* *
* Abstract: *
* This file contains the interface routines for the solaris OS. *
* It has all DDI entry point routines and GLD entry point routines. *
* *
* This file also contains routines that take care of initialization *
* uninit routine and interrupt routine. *
* *
* **********************************************************************
*/
#include <sys/dlpi.h>
#include <sys/mac.h>
#include "e1000g_sw.h"
#include "e1000g_debug.h"
static char ident[] = "Intel PRO/1000 Ethernet";
/* LINTED E_STATIC_UNUSED */
static char e1000g_version[] = "Driver Ver. 5.3.24";
/*
* Proto types for DDI entry points
*/
static int e1000g_attach(dev_info_t *, ddi_attach_cmd_t);
static int e1000g_detach(dev_info_t *, ddi_detach_cmd_t);
static int e1000g_quiesce(dev_info_t *);
/*
* init and intr routines prototype
*/
static int e1000g_resume(dev_info_t *);
static int e1000g_suspend(dev_info_t *);
static uint_t e1000g_intr_pciexpress(caddr_t);
static uint_t e1000g_intr(caddr_t);
static void e1000g_intr_work(struct e1000g *, uint32_t);
#pragma inline(e1000g_intr_work)
static int e1000g_init(struct e1000g *);
static int e1000g_start(struct e1000g *, boolean_t);
static void e1000g_stop(struct e1000g *, boolean_t);
static int e1000g_m_start(void *);
static void e1000g_m_stop(void *);
static int e1000g_m_promisc(void *, boolean_t);
static boolean_t e1000g_m_getcapab(void *, mac_capab_t, void *);
static int e1000g_m_multicst(void *, boolean_t, const uint8_t *);
static void e1000g_m_ioctl(void *, queue_t *, mblk_t *);
static int e1000g_m_setprop(void *, const char *, mac_prop_id_t,
uint_t, const void *);
static int e1000g_m_getprop(void *, const char *, mac_prop_id_t,
uint_t, void *);
static void e1000g_m_propinfo(void *, const char *, mac_prop_id_t,
mac_prop_info_handle_t);
static int e1000g_set_priv_prop(struct e1000g *, const char *, uint_t,
const void *);
static int e1000g_get_priv_prop(struct e1000g *, const char *, uint_t, void *);
static void e1000g_init_locks(struct e1000g *);
static void e1000g_destroy_locks(struct e1000g *);
static int e1000g_identify_hardware(struct e1000g *);
static int e1000g_regs_map(struct e1000g *);
static int e1000g_set_driver_params(struct e1000g *);
static void e1000g_set_bufsize(struct e1000g *);
static int e1000g_register_mac(struct e1000g *);
static boolean_t e1000g_rx_drain(struct e1000g *);
static boolean_t e1000g_tx_drain(struct e1000g *);
static void e1000g_init_unicst(struct e1000g *);
static int e1000g_unicst_set(struct e1000g *, const uint8_t *, int);
static int e1000g_alloc_rx_data(struct e1000g *);
static void e1000g_release_multicast(struct e1000g *);
static void e1000g_pch_limits(struct e1000g *);
static uint32_t e1000g_mtu2maxframe(uint32_t);
/*
* Local routines
*/
static boolean_t e1000g_reset_adapter(struct e1000g *);
static void e1000g_tx_clean(struct e1000g *);
static void e1000g_rx_clean(struct e1000g *);
static void e1000g_link_timer(void *);
static void e1000g_local_timer(void *);
static boolean_t e1000g_link_check(struct e1000g *);
static boolean_t e1000g_stall_check(struct e1000g *);
static void e1000g_smartspeed(struct e1000g *);
static void e1000g_get_conf(struct e1000g *);
static boolean_t e1000g_get_prop(struct e1000g *, char *, int, int, int,
int *);
static void enable_watchdog_timer(struct e1000g *);
static void disable_watchdog_timer(struct e1000g *);
static void start_watchdog_timer(struct e1000g *);
static void restart_watchdog_timer(struct e1000g *);
static void stop_watchdog_timer(struct e1000g *);
static void stop_link_timer(struct e1000g *);
static void stop_82547_timer(e1000g_tx_ring_t *);
static void e1000g_force_speed_duplex(struct e1000g *);
static void e1000g_setup_max_mtu(struct e1000g *);
static void e1000g_get_max_frame_size(struct e1000g *);
static boolean_t is_valid_mac_addr(uint8_t *);
static void e1000g_unattach(dev_info_t *, struct e1000g *);
static int e1000g_get_bar_info(dev_info_t *, int, bar_info_t *);
#ifdef E1000G_DEBUG
static void e1000g_ioc_peek_reg(struct e1000g *, e1000g_peekpoke_t *);
static void e1000g_ioc_poke_reg(struct e1000g *, e1000g_peekpoke_t *);
static void e1000g_ioc_peek_mem(struct e1000g *, e1000g_peekpoke_t *);
static void e1000g_ioc_poke_mem(struct e1000g *, e1000g_peekpoke_t *);
static enum ioc_reply e1000g_pp_ioctl(struct e1000g *,
struct iocblk *, mblk_t *);
#endif
static enum ioc_reply e1000g_loopback_ioctl(struct e1000g *,
struct iocblk *, mblk_t *);
static boolean_t e1000g_check_loopback_support(struct e1000_hw *);
static boolean_t e1000g_set_loopback_mode(struct e1000g *, uint32_t);
static void e1000g_set_internal_loopback(struct e1000g *);
static void e1000g_set_external_loopback_1000(struct e1000g *);
static void e1000g_set_external_loopback_100(struct e1000g *);
static void e1000g_set_external_loopback_10(struct e1000g *);
static int e1000g_add_intrs(struct e1000g *);
static int e1000g_intr_add(struct e1000g *, int);
static int e1000g_rem_intrs(struct e1000g *);
static int e1000g_enable_intrs(struct e1000g *);
static int e1000g_disable_intrs(struct e1000g *);
static boolean_t e1000g_link_up(struct e1000g *);
#ifdef __sparc
static boolean_t e1000g_find_mac_address(struct e1000g *);
#endif
static void e1000g_get_phy_state(struct e1000g *);
static int e1000g_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err,
const void *impl_data);
static void e1000g_fm_init(struct e1000g *Adapter);
static void e1000g_fm_fini(struct e1000g *Adapter);
static void e1000g_param_sync(struct e1000g *);
static void e1000g_get_driver_control(struct e1000_hw *);
static void e1000g_release_driver_control(struct e1000_hw *);
static void e1000g_restore_promisc(struct e1000g *Adapter);
char *e1000g_priv_props[] = {
"_tx_bcopy_threshold",
"_tx_interrupt_enable",
"_tx_intr_delay",
"_tx_intr_abs_delay",
"_rx_bcopy_threshold",
"_max_num_rcv_packets",
"_rx_intr_delay",
"_rx_intr_abs_delay",
"_intr_throttling_rate",
"_intr_adaptive",
"_adv_pause_cap",
"_adv_asym_pause_cap",
NULL
};
static struct cb_ops cb_ws_ops = {
nulldev, /* cb_open */
nulldev, /* cb_close */
nodev, /* cb_strategy */
nodev, /* cb_print */
nodev, /* cb_dump */
nodev, /* cb_read */
nodev, /* cb_write */
nodev, /* cb_ioctl */
nodev, /* cb_devmap */
nodev, /* cb_mmap */
nodev, /* cb_segmap */
nochpoll, /* cb_chpoll */
ddi_prop_op, /* cb_prop_op */
NULL, /* cb_stream */
D_MP | D_HOTPLUG, /* cb_flag */
CB_REV, /* cb_rev */
nodev, /* cb_aread */
nodev /* cb_awrite */
};
static struct dev_ops ws_ops = {
DEVO_REV, /* devo_rev */
0, /* devo_refcnt */
NULL, /* devo_getinfo */
nulldev, /* devo_identify */
nulldev, /* devo_probe */
e1000g_attach, /* devo_attach */
e1000g_detach, /* devo_detach */
nodev, /* devo_reset */
&cb_ws_ops, /* devo_cb_ops */
NULL, /* devo_bus_ops */
ddi_power, /* devo_power */
e1000g_quiesce /* devo_quiesce */
};
static struct modldrv modldrv = {
&mod_driverops, /* Type of module. This one is a driver */
ident, /* Discription string */
&ws_ops, /* driver ops */
};
static struct modlinkage modlinkage = {
MODREV_1, &modldrv, NULL
};
/* Access attributes for register mapping */
static ddi_device_acc_attr_t e1000g_regs_acc_attr = {
DDI_DEVICE_ATTR_V1,
DDI_STRUCTURE_LE_ACC,
DDI_STRICTORDER_ACC,
DDI_FLAGERR_ACC
};
#define E1000G_M_CALLBACK_FLAGS \
(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP | MC_PROPINFO)
static mac_callbacks_t e1000g_m_callbacks = {
E1000G_M_CALLBACK_FLAGS,
e1000g_m_stat,
e1000g_m_start,
e1000g_m_stop,
e1000g_m_promisc,
e1000g_m_multicst,
NULL,
e1000g_m_tx,
NULL,
e1000g_m_ioctl,
e1000g_m_getcapab,
NULL,
NULL,
e1000g_m_setprop,
e1000g_m_getprop,
e1000g_m_propinfo
};
/*
* Global variables
*/
uint32_t e1000g_jumbo_mtu = MAXIMUM_MTU_9K;
uint32_t e1000g_mblks_pending = 0;
/*
* Workaround for Dynamic Reconfiguration support, for x86 platform only.
* Here we maintain a private dev_info list if e1000g_force_detach is
* enabled. If we force the driver to detach while there are still some
* rx buffers retained in the upper layer, we have to keep a copy of the
* dev_info. In some cases (Dynamic Reconfiguration), the dev_info data
* structure will be freed after the driver is detached. However when we
* finally free those rx buffers released by the upper layer, we need to
* refer to the dev_info to free the dma buffers. So we save a copy of
* the dev_info for this purpose. On x86 platform, we assume this copy
* of dev_info is always valid, but on SPARC platform, it could be invalid
* after the system board level DR operation. For this reason, the global
* variable e1000g_force_detach must be B_FALSE on SPARC platform.
*/
#ifdef __sparc
boolean_t e1000g_force_detach = B_FALSE;
#else
boolean_t e1000g_force_detach = B_TRUE;
#endif
private_devi_list_t *e1000g_private_devi_list = NULL;
/*
* The mutex e1000g_rx_detach_lock is defined to protect the processing of
* the private dev_info list, and to serialize the processing of rx buffer
* freeing and rx buffer recycling.
*/
kmutex_t e1000g_rx_detach_lock;
/*
* The rwlock e1000g_dma_type_lock is defined to protect the global flag
* e1000g_dma_type. For SPARC, the initial value of the flag is "USE_DVMA".
* If there are many e1000g instances, the system may run out of DVMA
* resources during the initialization of the instances, then the flag will
* be changed to "USE_DMA". Because different e1000g instances are initialized
* in parallel, we need to use this lock to protect the flag.
*/
krwlock_t e1000g_dma_type_lock;
/*
* The 82546 chipset is a dual-port device, both the ports share one eeprom.
* Based on the information from Intel, the 82546 chipset has some hardware
* problem. When one port is being reset and the other port is trying to
* access the eeprom, it could cause system hang or panic. To workaround this
* hardware problem, we use a global mutex to prevent such operations from
* happening simultaneously on different instances. This workaround is applied
* to all the devices supported by this driver.
*/
kmutex_t e1000g_nvm_lock;
/*
* Loadable module configuration entry points for the driver
*/
/*
* _init - module initialization
*/
int
_init(void)
{
int status;
mac_init_ops(&ws_ops, WSNAME);
status = mod_install(&modlinkage);
if (status != DDI_SUCCESS)
mac_fini_ops(&ws_ops);
else {
mutex_init(&e1000g_rx_detach_lock, NULL, MUTEX_DRIVER, NULL);
rw_init(&e1000g_dma_type_lock, NULL, RW_DRIVER, NULL);
mutex_init(&e1000g_nvm_lock, NULL, MUTEX_DRIVER, NULL);
}
return (status);
}
/*
* _fini - module finalization
*/
int
_fini(void)
{
int status;
if (e1000g_mblks_pending != 0)
return (EBUSY);
status = mod_remove(&modlinkage);
if (status == DDI_SUCCESS) {
mac_fini_ops(&ws_ops);
if (e1000g_force_detach) {
private_devi_list_t *devi_node;
mutex_enter(&e1000g_rx_detach_lock);
while (e1000g_private_devi_list != NULL) {
devi_node = e1000g_private_devi_list;
e1000g_private_devi_list =
e1000g_private_devi_list->next;
kmem_free(devi_node->priv_dip,
sizeof (struct dev_info));
kmem_free(devi_node,
sizeof (private_devi_list_t));
}
mutex_exit(&e1000g_rx_detach_lock);
}
mutex_destroy(&e1000g_rx_detach_lock);
rw_destroy(&e1000g_dma_type_lock);
mutex_destroy(&e1000g_nvm_lock);
}
return (status);
}
/*
* _info - module information
*/
int
_info(struct modinfo *modinfop)
{
return (mod_info(&modlinkage, modinfop));
}
/*
* e1000g_attach - driver attach
*
* This function is the device-specific initialization entry
* point. This entry point is required and must be written.
* The DDI_ATTACH command must be provided in the attach entry
* point. When attach() is called with cmd set to DDI_ATTACH,
* all normal kernel services (such as kmem_alloc(9F)) are
* available for use by the driver.
*
* The attach() function will be called once for each instance
* of the device on the system with cmd set to DDI_ATTACH.
* Until attach() succeeds, the only driver entry points which
* may be called are open(9E) and getinfo(9E).
*/
static int
e1000g_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
{
struct e1000g *Adapter;
struct e1000_hw *hw;
struct e1000g_osdep *osdep;
int instance;
switch (cmd) {
default:
e1000g_log(NULL, CE_WARN,
"Unsupported command send to e1000g_attach... ");
return (DDI_FAILURE);
case DDI_RESUME:
return (e1000g_resume(devinfo));
case DDI_ATTACH:
break;
}
/*
* get device instance number
*/
instance = ddi_get_instance(devinfo);
/*
* Allocate soft data structure
*/
Adapter =
(struct e1000g *)kmem_zalloc(sizeof (*Adapter), KM_SLEEP);
Adapter->dip = devinfo;
Adapter->instance = instance;
Adapter->tx_ring->adapter = Adapter;
Adapter->rx_ring->adapter = Adapter;
hw = &Adapter->shared;
osdep = &Adapter->osdep;
hw->back = osdep;
osdep->adapter = Adapter;
ddi_set_driver_private(devinfo, (caddr_t)Adapter);
/*
* Initialize for fma support
*/
(void) e1000g_get_prop(Adapter, "fm-capable",
0, 0x0f,
DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE,
&Adapter->fm_capabilities);
e1000g_fm_init(Adapter);
Adapter->attach_progress |= ATTACH_PROGRESS_FMINIT;
/*
* PCI Configure
*/
if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
e1000g_log(Adapter, CE_WARN, "PCI configuration failed");
goto attach_fail;
}
Adapter->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
/*
* Setup hardware
*/
if (e1000g_identify_hardware(Adapter) != DDI_SUCCESS) {
e1000g_log(Adapter, CE_WARN, "Identify hardware failed");
goto attach_fail;
}
/*
* Map in the device registers.
*/
if (e1000g_regs_map(Adapter) != DDI_SUCCESS) {
e1000g_log(Adapter, CE_WARN, "Mapping registers failed");
goto attach_fail;
}
Adapter->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
/*
* Initialize driver parameters
*/
if (e1000g_set_driver_params(Adapter) != DDI_SUCCESS) {
goto attach_fail;
}
Adapter->attach_progress |= ATTACH_PROGRESS_SETUP;
if (e1000g_check_acc_handle(Adapter->osdep.cfg_handle) != DDI_FM_OK) {
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_LOST);
goto attach_fail;
}
/*
* Disable ULP support
*/
(void) e1000_disable_ulp_lpt_lp(hw, TRUE);
/*
* Initialize interrupts
*/
if (e1000g_add_intrs(Adapter) != DDI_SUCCESS) {
e1000g_log(Adapter, CE_WARN, "Add interrupts failed");
goto attach_fail;
}
Adapter->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
/*
* Initialize mutex's for this device.
* Do this before enabling the interrupt handler and
* register the softint to avoid the condition where
* interrupt handler can try using uninitialized mutex
*/
e1000g_init_locks(Adapter);
Adapter->attach_progress |= ATTACH_PROGRESS_LOCKS;
/*
* Initialize Driver Counters
*/
if (e1000g_init_stats(Adapter) != DDI_SUCCESS) {
e1000g_log(Adapter, CE_WARN, "Init stats failed");
goto attach_fail;
}
Adapter->attach_progress |= ATTACH_PROGRESS_KSTATS;
/*
* Initialize chip hardware and software structures
*/
rw_enter(&Adapter->chip_lock, RW_WRITER);
if (e1000g_init(Adapter) != DDI_SUCCESS) {
rw_exit(&Adapter->chip_lock);
e1000g_log(Adapter, CE_WARN, "Adapter initialization failed");
goto attach_fail;
}
rw_exit(&Adapter->chip_lock);
Adapter->attach_progress |= ATTACH_PROGRESS_INIT;
/*
* Register the driver to the MAC
*/
if (e1000g_register_mac(Adapter) != DDI_SUCCESS) {
e1000g_log(Adapter, CE_WARN, "Register MAC failed");
goto attach_fail;
}
Adapter->attach_progress |= ATTACH_PROGRESS_MAC;
/*
* Now that mutex locks are initialized, and the chip is also
* initialized, enable interrupts.
*/
if (e1000g_enable_intrs(Adapter) != DDI_SUCCESS) {
e1000g_log(Adapter, CE_WARN, "Enable DDI interrupts failed");
goto attach_fail;
}
Adapter->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
/*
* If e1000g_force_detach is enabled, in global private dip list,
* we will create a new entry, which maintains the priv_dip for DR
* supports after driver detached.
*/
if (e1000g_force_detach) {
private_devi_list_t *devi_node;
Adapter->priv_dip =
kmem_zalloc(sizeof (struct dev_info), KM_SLEEP);
bcopy(DEVI(devinfo), DEVI(Adapter->priv_dip),
sizeof (struct dev_info));
devi_node =
kmem_zalloc(sizeof (private_devi_list_t), KM_SLEEP);
mutex_enter(&e1000g_rx_detach_lock);
devi_node->priv_dip = Adapter->priv_dip;
devi_node->flag = E1000G_PRIV_DEVI_ATTACH;
devi_node->pending_rx_count = 0;
Adapter->priv_devi_node = devi_node;
if (e1000g_private_devi_list == NULL) {
devi_node->prev = NULL;
devi_node->next = NULL;
e1000g_private_devi_list = devi_node;
} else {
devi_node->prev = NULL;
devi_node->next = e1000g_private_devi_list;
e1000g_private_devi_list->prev = devi_node;
e1000g_private_devi_list = devi_node;
}
mutex_exit(&e1000g_rx_detach_lock);
}
Adapter->e1000g_state = E1000G_INITIALIZED;
return (DDI_SUCCESS);
attach_fail:
e1000g_unattach(devinfo, Adapter);
return (DDI_FAILURE);
}
static int
e1000g_register_mac(struct e1000g *Adapter)
{
struct e1000_hw *hw = &Adapter->shared;
mac_register_t *mac;
int err;
if ((mac = mac_alloc(MAC_VERSION)) == NULL)
return (DDI_FAILURE);
mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
mac->m_driver = Adapter;
mac->m_dip = Adapter->dip;
mac->m_src_addr = hw->mac.addr;
mac->m_callbacks = &e1000g_m_callbacks;
mac->m_min_sdu = 0;
mac->m_max_sdu = Adapter->default_mtu;
mac->m_margin = VLAN_TAGSZ;
mac->m_priv_props = e1000g_priv_props;
mac->m_v12n = MAC_VIRT_LEVEL1;
err = mac_register(mac, &Adapter->mh);
mac_free(mac);
return (err == 0 ? DDI_SUCCESS : DDI_FAILURE);
}
static int
e1000g_identify_hardware(struct e1000g *Adapter)
{
struct e1000_hw *hw = &Adapter->shared;
struct e1000g_osdep *osdep = &Adapter->osdep;
/* Get the device id */
hw->vendor_id =
pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
hw->device_id =
pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
hw->revision_id =
pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
hw->subsystem_device_id =
pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
hw->subsystem_vendor_id =
pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
E1000G_DEBUGLOG_0(Adapter, E1000G_INFO_LEVEL,
"MAC type could not be set properly.");
return (DDI_FAILURE);
}
return (DDI_SUCCESS);
}
static int
e1000g_regs_map(struct e1000g *Adapter)
{
dev_info_t *devinfo = Adapter->dip;
struct e1000_hw *hw = &Adapter->shared;
struct e1000g_osdep *osdep = &Adapter->osdep;
off_t mem_size;
bar_info_t bar_info;
int offset, rnumber;
rnumber = ADAPTER_REG_SET;
/* Get size of adapter register memory */
if (ddi_dev_regsize(devinfo, rnumber, &mem_size) !=
DDI_SUCCESS) {
E1000G_DEBUGLOG_0(Adapter, CE_WARN,
"ddi_dev_regsize for registers failed");
return (DDI_FAILURE);
}
/* Map adapter register memory */
if ((ddi_regs_map_setup(devinfo, rnumber,
(caddr_t *)&hw->hw_addr, 0, mem_size, &e1000g_regs_acc_attr,
&osdep->reg_handle)) != DDI_SUCCESS) {
E1000G_DEBUGLOG_0(Adapter, CE_WARN,
"ddi_regs_map_setup for registers failed");
goto regs_map_fail;
}
/* ICH needs to map flash memory */
switch (hw->mac.type) {
case e1000_ich8lan:
case e1000_ich9lan:
case e1000_ich10lan:
case e1000_pchlan:
case e1000_pch2lan:
case e1000_pch_lpt:
rnumber = ICH_FLASH_REG_SET;
/* get flash size */
if (ddi_dev_regsize(devinfo, rnumber,
&mem_size) != DDI_SUCCESS) {
E1000G_DEBUGLOG_0(Adapter, CE_WARN,
"ddi_dev_regsize for ICH flash failed");
goto regs_map_fail;
}
/* map flash in */
if (ddi_regs_map_setup(devinfo, rnumber,
(caddr_t *)&hw->flash_address, 0,
mem_size, &e1000g_regs_acc_attr,
&osdep->ich_flash_handle) != DDI_SUCCESS) {
E1000G_DEBUGLOG_0(Adapter, CE_WARN,
"ddi_regs_map_setup for ICH flash failed");
goto regs_map_fail;
}
break;
case e1000_pch_spt:
/*
* On the SPT, the device flash is actually in BAR0, not a
* separate BAR. Therefore we end up setting the
* ich_flash_handle to be the same as the register handle.
* We mark the same to reduce the confusion in the other
* functions and macros. Though this does make the set up and
* tear-down path slightly more complicated.
*/
osdep->ich_flash_handle = osdep->reg_handle;
hw->flash_address = hw->hw_addr;
default:
break;
}
/* map io space */
switch (hw->mac.type) {
case e1000_82544:
case e1000_82540:
case e1000_82545:
case e1000_82546:
case e1000_82541:
case e1000_82541_rev_2:
/* find the IO bar */
rnumber = -1;
for (offset = PCI_CONF_BASE1;
offset <= PCI_CONF_BASE5; offset += 4) {
if (e1000g_get_bar_info(devinfo, offset, &bar_info)
!= DDI_SUCCESS)
continue;
if (bar_info.type == E1000G_BAR_IO) {
rnumber = bar_info.rnumber;
break;
}
}
if (rnumber < 0) {
E1000G_DEBUGLOG_0(Adapter, CE_WARN,
"No io space is found");
goto regs_map_fail;
}
/* get io space size */
if (ddi_dev_regsize(devinfo, rnumber,
&mem_size) != DDI_SUCCESS) {
E1000G_DEBUGLOG_0(Adapter, CE_WARN,
"ddi_dev_regsize for io space failed");
goto regs_map_fail;
}
/* map io space */
if ((ddi_regs_map_setup(devinfo, rnumber,
(caddr_t *)&hw->io_base, 0, mem_size,
&e1000g_regs_acc_attr,
&osdep->io_reg_handle)) != DDI_SUCCESS) {
E1000G_DEBUGLOG_0(Adapter, CE_WARN,
"ddi_regs_map_setup for io space failed");
goto regs_map_fail;
}
break;
default:
hw->io_base = 0;
break;
}
return (DDI_SUCCESS);
regs_map_fail:
if (osdep->reg_handle != NULL)
ddi_regs_map_free(&osdep->reg_handle);
if (osdep->ich_flash_handle != NULL && hw->mac.type != e1000_pch_spt)
ddi_regs_map_free(&osdep->ich_flash_handle);
return (DDI_FAILURE);
}
static int
e1000g_set_driver_params(struct e1000g *Adapter)
{
struct e1000_hw *hw;
hw = &Adapter->shared;
/* Set MAC type and initialize hardware functions */
if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
E1000G_DEBUGLOG_0(Adapter, CE_WARN,
"Could not setup hardware functions");
return (DDI_FAILURE);
}
/* Get bus information */
if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
E1000G_DEBUGLOG_0(Adapter, CE_WARN,
"Could not get bus information");
return (DDI_FAILURE);
}
e1000_read_pci_cfg(hw, PCI_COMMAND_REGISTER, &hw->bus.pci_cmd_word);
hw->mac.autoneg_failed = B_TRUE;
/* Set the autoneg_wait_to_complete flag to B_FALSE */
hw->phy.autoneg_wait_to_complete = B_FALSE;
/* Adaptive IFS related changes */
hw->mac.adaptive_ifs = B_TRUE;
/* Enable phy init script for IGP phy of 82541/82547 */
if ((hw->mac.type == e1000_82547) ||
(hw->mac.type == e1000_82541) ||
(hw->mac.type == e1000_82547_rev_2) ||
(hw->mac.type == e1000_82541_rev_2))
e1000_init_script_state_82541(hw, B_TRUE);
/* Enable the TTL workaround for 82541/82547 */
e1000_set_ttl_workaround_state_82541(hw, B_TRUE);
#ifdef __sparc
Adapter->strip_crc = B_TRUE;
#else
Adapter->strip_crc = B_FALSE;
#endif
/* setup the maximum MTU size of the chip */
e1000g_setup_max_mtu(Adapter);
/* Get speed/duplex settings in conf file */
hw->mac.forced_speed_duplex = ADVERTISE_100_FULL;
hw->phy.autoneg_advertised = AUTONEG_ADVERTISE_SPEED_DEFAULT;
e1000g_force_speed_duplex(Adapter);
/* Get Jumbo Frames settings in conf file */
e1000g_get_max_frame_size(Adapter);
/* Get conf file properties */
e1000g_get_conf(Adapter);
/* enforce PCH limits */
e1000g_pch_limits(Adapter);
/* Set Rx/Tx buffer size */
e1000g_set_bufsize(Adapter);
/* Master Latency Timer */
Adapter->master_latency_timer = DEFAULT_MASTER_LATENCY_TIMER;
/* copper options */
if (hw->phy.media_type == e1000_media_type_copper) {
hw->phy.mdix = 0; /* AUTO_ALL_MODES */
hw->phy.disable_polarity_correction = B_FALSE;
hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
}
/* The initial link state should be "unknown" */
Adapter->link_state = LINK_STATE_UNKNOWN;
/* Initialize rx parameters */
Adapter->rx_intr_delay = DEFAULT_RX_INTR_DELAY;
Adapter->rx_intr_abs_delay = DEFAULT_RX_INTR_ABS_DELAY;
/* Initialize tx parameters */
Adapter->tx_intr_enable = DEFAULT_TX_INTR_ENABLE;
Adapter->tx_bcopy_thresh = DEFAULT_TX_BCOPY_THRESHOLD;
Adapter->tx_intr_delay = DEFAULT_TX_INTR_DELAY;
Adapter->tx_intr_abs_delay = DEFAULT_TX_INTR_ABS_DELAY;
/* Initialize rx parameters */
Adapter->rx_bcopy_thresh = DEFAULT_RX_BCOPY_THRESHOLD;
return (DDI_SUCCESS);
}
static void
e1000g_setup_max_mtu(struct e1000g *Adapter)
{
struct e1000_mac_info *mac = &Adapter->shared.mac;
struct e1000_phy_info *phy = &Adapter->shared.phy;
switch (mac->type) {
/* types that do not support jumbo frames */
case e1000_ich8lan:
case e1000_82573:
case e1000_82583:
Adapter->max_mtu = ETHERMTU;
break;
/* ich9 supports jumbo frames except on one phy type */
case e1000_ich9lan:
if (phy->type == e1000_phy_ife)
Adapter->max_mtu = ETHERMTU;
else
Adapter->max_mtu = MAXIMUM_MTU_9K;
break;
/* pch can do jumbo frames up to 4K */
case e1000_pchlan:
Adapter->max_mtu = MAXIMUM_MTU_4K;
break;
/* pch2 can do jumbo frames up to 9K */
case e1000_pch2lan:
case e1000_pch_lpt:
case e1000_pch_spt:
Adapter->max_mtu = MAXIMUM_MTU_9K;
break;
/* types with a special limit */
case e1000_82571:
case e1000_82572:
case e1000_82574:
case e1000_80003es2lan:
case e1000_ich10lan:
if (e1000g_jumbo_mtu >= ETHERMTU &&
e1000g_jumbo_mtu <= MAXIMUM_MTU_9K) {
Adapter->max_mtu = e1000g_jumbo_mtu;
} else {
Adapter->max_mtu = MAXIMUM_MTU_9K;
}
break;
/* default limit is 16K */
default:
Adapter->max_mtu = FRAME_SIZE_UPTO_16K -
sizeof (struct ether_vlan_header) - ETHERFCSL;
break;
}
}
static void
e1000g_set_bufsize(struct e1000g *Adapter)
{
struct e1000_mac_info *mac = &Adapter->shared.mac;
uint64_t rx_size;
uint64_t tx_size;
dev_info_t *devinfo = Adapter->dip;
#ifdef __sparc
ulong_t iommu_pagesize;
#endif
/* Get the system page size */
Adapter->sys_page_sz = ddi_ptob(devinfo, (ulong_t)1);
#ifdef __sparc
iommu_pagesize = dvma_pagesize(devinfo);
if (iommu_pagesize != 0) {
if (Adapter->sys_page_sz == iommu_pagesize) {
if (iommu_pagesize > 0x4000)
Adapter->sys_page_sz = 0x4000;
} else {
if (Adapter->sys_page_sz > iommu_pagesize)
Adapter->sys_page_sz = iommu_pagesize;
}
}
if (Adapter->lso_enable) {
Adapter->dvma_page_num = E1000_LSO_MAXLEN /
Adapter->sys_page_sz + E1000G_DEFAULT_DVMA_PAGE_NUM;
} else {
Adapter->dvma_page_num = Adapter->max_frame_size /
Adapter->sys_page_sz + E1000G_DEFAULT_DVMA_PAGE_NUM;
}
ASSERT(Adapter->dvma_page_num >= E1000G_DEFAULT_DVMA_PAGE_NUM);
#endif
Adapter->min_frame_size = ETHERMIN + ETHERFCSL;
if (Adapter->mem_workaround_82546 &&
((mac->type == e1000_82545) ||
(mac->type == e1000_82546) ||
(mac->type == e1000_82546_rev_3))) {
Adapter->rx_buffer_size = E1000_RX_BUFFER_SIZE_2K;
} else {
rx_size = Adapter->max_frame_size;
if ((rx_size > FRAME_SIZE_UPTO_2K) &&
(rx_size <= FRAME_SIZE_UPTO_4K))
Adapter->rx_buffer_size = E1000_RX_BUFFER_SIZE_4K;
else if ((rx_size > FRAME_SIZE_UPTO_4K) &&
(rx_size <= FRAME_SIZE_UPTO_8K))
Adapter->rx_buffer_size = E1000_RX_BUFFER_SIZE_8K;
else if ((rx_size > FRAME_SIZE_UPTO_8K) &&
(rx_size <= FRAME_SIZE_UPTO_16K))
Adapter->rx_buffer_size = E1000_RX_BUFFER_SIZE_16K;
else
Adapter->rx_buffer_size = E1000_RX_BUFFER_SIZE_2K;
}
Adapter->rx_buffer_size += E1000G_IPALIGNROOM;
tx_size = Adapter->max_frame_size;
if ((tx_size > FRAME_SIZE_UPTO_2K) && (tx_size <= FRAME_SIZE_UPTO_4K))
Adapter->tx_buffer_size = E1000_TX_BUFFER_SIZE_4K;
else if ((tx_size > FRAME_SIZE_UPTO_4K) &&
(tx_size <= FRAME_SIZE_UPTO_8K))
Adapter->tx_buffer_size = E1000_TX_BUFFER_SIZE_8K;
else if ((tx_size > FRAME_SIZE_UPTO_8K) &&
(tx_size <= FRAME_SIZE_UPTO_16K))
Adapter->tx_buffer_size = E1000_TX_BUFFER_SIZE_16K;
else
Adapter->tx_buffer_size = E1000_TX_BUFFER_SIZE_2K;
/*
* For Wiseman adapters we have an requirement of having receive
* buffers aligned at 256 byte boundary. Since Livengood does not
* require this and forcing it for all hardwares will have
* performance implications, I am making it applicable only for
* Wiseman and for Jumbo frames enabled mode as rest of the time,
* it is okay to have normal frames...but it does involve a
* potential risk where we may loose data if buffer is not
* aligned...so all wiseman boards to have 256 byte aligned
* buffers
*/
if (mac->type < e1000_82543)
Adapter->rx_buf_align = RECEIVE_BUFFER_ALIGN_SIZE;
else
Adapter->rx_buf_align = 1;
}
/*
* e1000g_detach - driver detach
*
* The detach() function is the complement of the attach routine.
* If cmd is set to DDI_DETACH, detach() is used to remove the
* state associated with a given instance of a device node
* prior to the removal of that instance from the system.
*
* The detach() function will be called once for each instance
* of the device for which there has been a successful attach()
* once there are no longer any opens on the device.
*
* Interrupts routine are disabled, All memory allocated by this
* driver are freed.
*/
static int
e1000g_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
{
struct e1000g *Adapter;
boolean_t rx_drain;
switch (cmd) {
default:
return (DDI_FAILURE);
case DDI_SUSPEND:
return (e1000g_suspend(devinfo));
case DDI_DETACH:
break;
}
Adapter = (struct e1000g *)ddi_get_driver_private(devinfo);
if (Adapter == NULL)
return (DDI_FAILURE);
rx_drain = e1000g_rx_drain(Adapter);
if (!rx_drain && !e1000g_force_detach)
return (DDI_FAILURE);
if (mac_unregister(Adapter->mh) != 0) {
e1000g_log(Adapter, CE_WARN, "Unregister MAC failed");
return (DDI_FAILURE);
}
Adapter->attach_progress &= ~ATTACH_PROGRESS_MAC;
ASSERT(!(Adapter->e1000g_state & E1000G_STARTED));
if (!e1000g_force_detach && !rx_drain)
return (DDI_FAILURE);
e1000g_unattach(devinfo, Adapter);
return (DDI_SUCCESS);
}
/*
* e1000g_free_priv_devi_node - free a priv_dip entry for driver instance
*/
void
e1000g_free_priv_devi_node(private_devi_list_t *devi_node)
{
ASSERT(e1000g_private_devi_list != NULL);
ASSERT(devi_node != NULL);
if (devi_node->prev != NULL)
devi_node->prev->next = devi_node->next;
if (devi_node->next != NULL)
devi_node->next->prev = devi_node->prev;
if (devi_node == e1000g_private_devi_list)
e1000g_private_devi_list = devi_node->next;
kmem_free(devi_node->priv_dip,
sizeof (struct dev_info));
kmem_free(devi_node,
sizeof (private_devi_list_t));
}
static void
e1000g_unattach(dev_info_t *devinfo, struct e1000g *Adapter)
{
private_devi_list_t *devi_node;
int result;
if (Adapter->e1000g_blink != NULL) {
ddi_periodic_delete(Adapter->e1000g_blink);
Adapter->e1000g_blink = NULL;
}
if (Adapter->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
(void) e1000g_disable_intrs(Adapter);
}
if (Adapter->attach_progress & ATTACH_PROGRESS_MAC) {
(void) mac_unregister(Adapter->mh);
}
if (Adapter->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
(void) e1000g_rem_intrs(Adapter);
}
if (Adapter->attach_progress & ATTACH_PROGRESS_SETUP) {
(void) ddi_prop_remove_all(devinfo);
}
if (Adapter->attach_progress & ATTACH_PROGRESS_KSTATS) {
kstat_delete((kstat_t *)Adapter->e1000g_ksp);
}
if (Adapter->attach_progress & ATTACH_PROGRESS_INIT) {
stop_link_timer(Adapter);
mutex_enter(&e1000g_nvm_lock);
result = e1000_reset_hw(&Adapter->shared);
mutex_exit(&e1000g_nvm_lock);
if (result != E1000_SUCCESS) {
e1000g_fm_ereport(Adapter, DDI_FM_DEVICE_INVAL_STATE);
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_LOST);
}
}
e1000g_release_multicast(Adapter);
if (Adapter->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
if (Adapter->osdep.reg_handle != NULL)
ddi_regs_map_free(&Adapter->osdep.reg_handle);
if (Adapter->osdep.ich_flash_handle != NULL &&
Adapter->shared.mac.type != e1000_pch_spt)
ddi_regs_map_free(&Adapter->osdep.ich_flash_handle);
if (Adapter->osdep.io_reg_handle != NULL)
ddi_regs_map_free(&Adapter->osdep.io_reg_handle);
}
if (Adapter->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
if (Adapter->osdep.cfg_handle != NULL)
pci_config_teardown(&Adapter->osdep.cfg_handle);
}
if (Adapter->attach_progress & ATTACH_PROGRESS_LOCKS) {
e1000g_destroy_locks(Adapter);
}
if (Adapter->attach_progress & ATTACH_PROGRESS_FMINIT) {
e1000g_fm_fini(Adapter);
}
mutex_enter(&e1000g_rx_detach_lock);
if (e1000g_force_detach && (Adapter->priv_devi_node != NULL)) {
devi_node = Adapter->priv_devi_node;
devi_node->flag |= E1000G_PRIV_DEVI_DETACH;
if (devi_node->pending_rx_count == 0) {
e1000g_free_priv_devi_node(devi_node);
}
}
mutex_exit(&e1000g_rx_detach_lock);
kmem_free((caddr_t)Adapter, sizeof (struct e1000g));
/*
* Another hotplug spec requirement,
* run ddi_set_driver_private(devinfo, null);
*/
ddi_set_driver_private(devinfo, NULL);
}
/*
* Get the BAR type and rnumber for a given PCI BAR offset
*/
static int
e1000g_get_bar_info(dev_info_t *dip, int bar_offset, bar_info_t *bar_info)
{
pci_regspec_t *regs;
uint_t regs_length;
int type, rnumber, rcount;
ASSERT((bar_offset >= PCI_CONF_BASE0) &&
(bar_offset <= PCI_CONF_BASE5));
/*
* Get the DDI "reg" property
*/
if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dip,
DDI_PROP_DONTPASS, "reg", (int **)&regs,
&regs_length) != DDI_PROP_SUCCESS) {
return (DDI_FAILURE);
}
rcount = regs_length * sizeof (int) / sizeof (pci_regspec_t);
/*
* Check the BAR offset
*/
for (rnumber = 0; rnumber < rcount; ++rnumber) {
if (PCI_REG_REG_G(regs[rnumber].pci_phys_hi) == bar_offset) {
type = regs[rnumber].pci_phys_hi & PCI_ADDR_MASK;
break;
}
}
ddi_prop_free(regs);
if (rnumber >= rcount)
return (DDI_FAILURE);
switch (type) {
case PCI_ADDR_CONFIG:
bar_info->type = E1000G_BAR_CONFIG;
break;
case PCI_ADDR_IO:
bar_info->type = E1000G_BAR_IO;
break;
case PCI_ADDR_MEM32:
bar_info->type = E1000G_BAR_MEM32;
break;
case PCI_ADDR_MEM64:
bar_info->type = E1000G_BAR_MEM64;
break;
default:
return (DDI_FAILURE);
}
bar_info->rnumber = rnumber;
return (DDI_SUCCESS);
}
static void
e1000g_init_locks(struct e1000g *Adapter)
{
e1000g_tx_ring_t *tx_ring;
e1000g_rx_ring_t *rx_ring;
rw_init(&Adapter->chip_lock, NULL,
RW_DRIVER, DDI_INTR_PRI(Adapter->intr_pri));
mutex_init(&Adapter->link_lock, NULL,
MUTEX_DRIVER, DDI_INTR_PRI(Adapter->intr_pri));
mutex_init(&Adapter->watchdog_lock, NULL,
MUTEX_DRIVER, DDI_INTR_PRI(Adapter->intr_pri));
tx_ring = Adapter->tx_ring;
mutex_init(&tx_ring->tx_lock, NULL,
MUTEX_DRIVER, DDI_INTR_PRI(Adapter->intr_pri));
mutex_init(&tx_ring->usedlist_lock, NULL,
MUTEX_DRIVER, DDI_INTR_PRI(Adapter->intr_pri));
mutex_init(&tx_ring->freelist_lock, NULL,
MUTEX_DRIVER, DDI_INTR_PRI(Adapter->intr_pri));
rx_ring = Adapter->rx_ring;
mutex_init(&rx_ring->rx_lock, NULL,
MUTEX_DRIVER, DDI_INTR_PRI(Adapter->intr_pri));
mutex_init(&Adapter->e1000g_led_lock, NULL,
MUTEX_DRIVER, DDI_INTR_PRI(Adapter->intr_pri));
}
static void
e1000g_destroy_locks(struct e1000g *Adapter)
{
e1000g_tx_ring_t *tx_ring;
e1000g_rx_ring_t *rx_ring;
mutex_destroy(&Adapter->e1000g_led_lock);
tx_ring = Adapter->tx_ring;
mutex_destroy(&tx_ring->tx_lock);
mutex_destroy(&tx_ring->usedlist_lock);
mutex_destroy(&tx_ring->freelist_lock);
rx_ring = Adapter->rx_ring;
mutex_destroy(&rx_ring->rx_lock);
mutex_destroy(&Adapter->link_lock);
mutex_destroy(&Adapter->watchdog_lock);
rw_destroy(&Adapter->chip_lock);
/* destory mutex initialized in shared code */
e1000_destroy_hw_mutex(&Adapter->shared);
}
static int
e1000g_resume(dev_info_t *devinfo)
{
struct e1000g *Adapter;
Adapter = (struct e1000g *)ddi_get_driver_private(devinfo);
if (Adapter == NULL)
e1000g_log(Adapter, CE_PANIC,
"Instance pointer is null\n");
if (Adapter->dip != devinfo)
e1000g_log(Adapter, CE_PANIC,
"Devinfo is not the same as saved devinfo\n");
rw_enter(&Adapter->chip_lock, RW_WRITER);
if (Adapter->e1000g_state & E1000G_STARTED) {
if (e1000g_start(Adapter, B_FALSE) != DDI_SUCCESS) {
rw_exit(&Adapter->chip_lock);
/*
* We note the failure, but return success, as the
* system is still usable without this controller.
*/
e1000g_log(Adapter, CE_WARN,
"e1000g_resume: failed to restart controller\n");
return (DDI_SUCCESS);
}
/* Enable and start the watchdog timer */
enable_watchdog_timer(Adapter);
}
Adapter->e1000g_state &= ~E1000G_SUSPENDED;
rw_exit(&Adapter->chip_lock);
return (DDI_SUCCESS);
}
static int
e1000g_suspend(dev_info_t *devinfo)
{
struct e1000g *Adapter;
Adapter = (struct e1000g *)ddi_get_driver_private(devinfo);
if (Adapter == NULL)
return (DDI_FAILURE);
rw_enter(&Adapter->chip_lock, RW_WRITER);
Adapter->e1000g_state |= E1000G_SUSPENDED;
/* if the port isn't plumbed, we can simply return */
if (!(Adapter->e1000g_state & E1000G_STARTED)) {
rw_exit(&Adapter->chip_lock);
return (DDI_SUCCESS);
}
e1000g_stop(Adapter, B_FALSE);
rw_exit(&Adapter->chip_lock);
/* Disable and stop all the timers */
disable_watchdog_timer(Adapter);
stop_link_timer(Adapter);
stop_82547_timer(Adapter->tx_ring);
return (DDI_SUCCESS);
}
static int
e1000g_init(struct e1000g *Adapter)
{
uint32_t pba;
uint32_t high_water;
struct e1000_hw *hw;
clock_t link_timeout;
int result;
hw = &Adapter->shared;
/*
* reset to put the hardware in a known state
* before we try to do anything with the eeprom
*/
mutex_enter(&e1000g_nvm_lock);
result = e1000_reset_hw(hw);
mutex_exit(&e1000g_nvm_lock);
if (result != E1000_SUCCESS) {
e1000g_fm_ereport(Adapter, DDI_FM_DEVICE_INVAL_STATE);
goto init_fail;
}
mutex_enter(&e1000g_nvm_lock);
result = e1000_validate_nvm_checksum(hw);
if (result < E1000_SUCCESS) {
/*
* Some PCI-E parts fail the first check due to
* the link being in sleep state. Call it again,
* if it fails a second time its a real issue.
*/
result = e1000_validate_nvm_checksum(hw);
}
mutex_exit(&e1000g_nvm_lock);
if (result < E1000_SUCCESS) {
e1000g_log(Adapter, CE_WARN,
"Invalid NVM checksum. Please contact "
"the vendor to update the NVM.");
e1000g_fm_ereport(Adapter, DDI_FM_DEVICE_INVAL_STATE);
goto init_fail;
}
result = 0;
#ifdef __sparc
/*
* First, we try to get the local ethernet address from OBP. If
* failed, then we get it from the EEPROM of NIC card.
*/
result = e1000g_find_mac_address(Adapter);
#endif
/* Get the local ethernet address. */
if (!result) {
mutex_enter(&e1000g_nvm_lock);
result = e1000_read_mac_addr(hw);
mutex_exit(&e1000g_nvm_lock);
}
if (result < E1000_SUCCESS) {
e1000g_log(Adapter, CE_WARN, "Read mac addr failed");
e1000g_fm_ereport(Adapter, DDI_FM_DEVICE_INVAL_STATE);
goto init_fail;
}
/* check for valid mac address */
if (!is_valid_mac_addr(hw->mac.addr)) {
e1000g_log(Adapter, CE_WARN, "Invalid mac addr");
e1000g_fm_ereport(Adapter, DDI_FM_DEVICE_INVAL_STATE);
goto init_fail;
}
/* Set LAA state for 82571 chipset */
e1000_set_laa_state_82571(hw, B_TRUE);
/* Master Latency Timer implementation */
if (Adapter->master_latency_timer) {
pci_config_put8(Adapter->osdep.cfg_handle,
PCI_CONF_LATENCY_TIMER, Adapter->master_latency_timer);
}
if (hw->mac.type < e1000_82547) {
/*
* Total FIFO is 64K
*/
if (Adapter->max_frame_size > FRAME_SIZE_UPTO_8K)
pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */
else
pba = E1000_PBA_48K; /* 48K for Rx, 16K for Tx */
} else if ((hw->mac.type == e1000_82571) ||
(hw->mac.type == e1000_82572) ||
(hw->mac.type == e1000_80003es2lan)) {
/*
* Total FIFO is 48K
*/
if (Adapter->max_frame_size > FRAME_SIZE_UPTO_8K)
pba = E1000_PBA_30K; /* 30K for Rx, 18K for Tx */
else
pba = E1000_PBA_38K; /* 38K for Rx, 10K for Tx */
} else if (hw->mac.type == e1000_82573) {
pba = E1000_PBA_20K; /* 20K for Rx, 12K for Tx */
} else if (hw->mac.type == e1000_82574) {
/* Keep adapter default: 20K for Rx, 20K for Tx */
pba = E1000_READ_REG(hw, E1000_PBA);
} else if (hw->mac.type == e1000_ich8lan) {
pba = E1000_PBA_8K; /* 8K for Rx, 12K for Tx */
} else if (hw->mac.type == e1000_ich9lan) {
pba = E1000_PBA_10K;
} else if (hw->mac.type == e1000_ich10lan) {
pba = E1000_PBA_10K;
} else if (hw->mac.type == e1000_pchlan) {
pba = E1000_PBA_26K;
} else if (hw->mac.type == e1000_pch2lan) {
pba = E1000_PBA_26K;
} else if (hw->mac.type == e1000_pch_lpt) {
pba = E1000_PBA_26K;
} else if (hw->mac.type == e1000_pch_spt) {
pba = E1000_PBA_26K;
} else {
/*
* Total FIFO is 40K
*/
if (Adapter->max_frame_size > FRAME_SIZE_UPTO_8K)
pba = E1000_PBA_22K; /* 22K for Rx, 18K for Tx */
else
pba = E1000_PBA_30K; /* 30K for Rx, 10K for Tx */
}
E1000_WRITE_REG(hw, E1000_PBA, pba);
/*
* These parameters set thresholds for the adapter's generation(Tx)
* and response(Rx) to Ethernet PAUSE frames. These are just threshold
* settings. Flow control is enabled or disabled in the configuration
* file.
* High-water mark is set down from the top of the rx fifo (not
* sensitive to max_frame_size) and low-water is set just below
* high-water mark.
* The high water mark must be low enough to fit one full frame above
* it in the rx FIFO. Should be the lower of:
* 90% of the Rx FIFO size and the full Rx FIFO size minus the early
* receive size (assuming ERT set to E1000_ERT_2048), or the full
* Rx FIFO size minus one full frame.
*/
high_water = min(((pba << 10) * 9 / 10),
((hw->mac.type == e1000_82573 || hw->mac.type == e1000_82574 ||
hw->mac.type == e1000_ich9lan || hw->mac.type == e1000_ich10lan) ?
((pba << 10) - (E1000_ERT_2048 << 3)) :
((pba << 10) - Adapter->max_frame_size)));
hw->fc.high_water = high_water & 0xFFF8;
hw->fc.low_water = hw->fc.high_water - 8;
if (hw->mac.type == e1000_80003es2lan)
hw->fc.pause_time = 0xFFFF;
else
hw->fc.pause_time = E1000_FC_PAUSE_TIME;
hw->fc.send_xon = B_TRUE;
/*
* Reset the adapter hardware the second time.
*/
mutex_enter(&e1000g_nvm_lock);
result = e1000_reset_hw(hw);
mutex_exit(&e1000g_nvm_lock);
if (result != E1000_SUCCESS) {
e1000g_fm_ereport(Adapter, DDI_FM_DEVICE_INVAL_STATE);
goto init_fail;
}
/* disable wakeup control by default */
if (hw->mac.type >= e1000_82544)
E1000_WRITE_REG(hw, E1000_WUC, 0);
/*
* MWI should be disabled on 82546.
*/
if (hw->mac.type == e1000_82546)
e1000_pci_clear_mwi(hw);
else
e1000_pci_set_mwi(hw);
/*
* Configure/Initialize hardware
*/
mutex_enter(&e1000g_nvm_lock);
result = e1000_init_hw(hw);
mutex_exit(&e1000g_nvm_lock);
if (result < E1000_SUCCESS) {
e1000g_log(Adapter, CE_WARN, "Initialize hw failed");
e1000g_fm_ereport(Adapter, DDI_FM_DEVICE_INVAL_STATE);
goto init_fail;
}
/*
* Restore LED settings to the default from EEPROM
* to meet the standard for Sun platforms.
*/
(void) e1000_cleanup_led(hw);
/* Disable Smart Power Down */
phy_spd_state(hw, B_FALSE);
/* Make sure driver has control */
e1000g_get_driver_control(hw);
/*
* Initialize unicast addresses.
*/
e1000g_init_unicst(Adapter);
/*
* Setup and initialize the mctable structures. After this routine
* completes Multicast table will be set
*/
e1000_update_mc_addr_list(hw,
(uint8_t *)Adapter->mcast_table, Adapter->mcast_count);
msec_delay(5);
/*
* Implement Adaptive IFS
*/
e1000_reset_adaptive(hw);
/* Setup Interrupt Throttling Register */
if (hw->mac.type >= e1000_82540) {
E1000_WRITE_REG(hw, E1000_ITR, Adapter->intr_throttling_rate);
} else
Adapter->intr_adaptive = B_FALSE;
/* Start the timer for link setup */
if (hw->mac.autoneg)
link_timeout = PHY_AUTO_NEG_LIMIT * drv_usectohz(100000);
else
link_timeout = PHY_FORCE_LIMIT * drv_usectohz(100000);
mutex_enter(&Adapter->link_lock);
if (hw->phy.autoneg_wait_to_complete) {
Adapter->link_complete = B_TRUE;
} else {
Adapter->link_complete = B_FALSE;
Adapter->link_tid = timeout(e1000g_link_timer,
(void *)Adapter, link_timeout);
}
mutex_exit(&Adapter->link_lock);
/* Save the state of the phy */
e1000g_get_phy_state(Adapter);
e1000g_param_sync(Adapter);
Adapter->init_count++;
if (e1000g_check_acc_handle(Adapter->osdep.cfg_handle) != DDI_FM_OK) {
goto init_fail;
}
if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) {
goto init_fail;
}
Adapter->poll_mode = e1000g_poll_mode;
return (DDI_SUCCESS);
init_fail:
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_LOST);
return (DDI_FAILURE);
}
static int
e1000g_alloc_rx_data(struct e1000g *Adapter)
{
e1000g_rx_ring_t *rx_ring;
e1000g_rx_data_t *rx_data;
rx_ring = Adapter->rx_ring;
rx_data = kmem_zalloc(sizeof (e1000g_rx_data_t), KM_NOSLEEP);
if (rx_data == NULL)
return (DDI_FAILURE);
rx_data->priv_devi_node = Adapter->priv_devi_node;
rx_data->rx_ring = rx_ring;
mutex_init(&rx_data->freelist_lock, NULL,
MUTEX_DRIVER, DDI_INTR_PRI(Adapter->intr_pri));
mutex_init(&rx_data->recycle_lock, NULL,
MUTEX_DRIVER, DDI_INTR_PRI(Adapter->intr_pri));
rx_ring->rx_data = rx_data;
return (DDI_SUCCESS);
}
void
e1000g_free_rx_pending_buffers(e1000g_rx_data_t *rx_data)
{
rx_sw_packet_t *packet, *next_packet;
if (rx_data == NULL)
return;
packet = rx_data->packet_area;
while (packet != NULL) {
next_packet = packet->next;
e1000g_free_rx_sw_packet(packet, B_TRUE);
packet = next_packet;
}
rx_data->packet_area = NULL;
}
void
e1000g_free_rx_data(e1000g_rx_data_t *rx_data)
{
if (rx_data == NULL)
return;
mutex_destroy(&rx_data->freelist_lock);
mutex_destroy(&rx_data->recycle_lock);
kmem_free(rx_data, sizeof (e1000g_rx_data_t));
}
/*
* Check if the link is up
*/
static boolean_t
e1000g_link_up(struct e1000g *Adapter)
{
struct e1000_hw *hw = &Adapter->shared;
boolean_t link_up = B_FALSE;
/*
* get_link_status is set in the interrupt handler on link-status-change
* or rx sequence error interrupt. get_link_status will stay
* false until the e1000_check_for_link establishes link only
* for copper adapters.
*/
switch (hw->phy.media_type) {
case e1000_media_type_copper:
if (hw->mac.get_link_status) {
/*
* SPT devices need a bit of extra time before we ask
* them.
*/
if (hw->mac.type == e1000_pch_spt)
msec_delay(50);
(void) e1000_check_for_link(hw);
if ((E1000_READ_REG(hw, E1000_STATUS) &
E1000_STATUS_LU)) {
link_up = B_TRUE;
} else {
link_up = !hw->mac.get_link_status;
}
} else {
link_up = B_TRUE;
}
break;
case e1000_media_type_fiber:
(void) e1000_check_for_link(hw);
link_up = (E1000_READ_REG(hw, E1000_STATUS) &
E1000_STATUS_LU);
break;
case e1000_media_type_internal_serdes:
(void) e1000_check_for_link(hw);
link_up = hw->mac.serdes_has_link;
break;
}
return (link_up);
}
static void
e1000g_m_ioctl(void *arg, queue_t *q, mblk_t *mp)
{
struct iocblk *iocp;
struct e1000g *e1000gp;
enum ioc_reply status;
iocp = (struct iocblk *)(uintptr_t)mp->b_rptr;
iocp->ioc_error = 0;
e1000gp = (struct e1000g *)arg;
ASSERT(e1000gp);
if (e1000gp == NULL) {
miocnak(q, mp, 0, EINVAL);
return;
}
rw_enter(&e1000gp->chip_lock, RW_READER);
if (e1000gp->e1000g_state & E1000G_SUSPENDED) {
rw_exit(&e1000gp->chip_lock);
miocnak(q, mp, 0, EINVAL);
return;
}
rw_exit(&e1000gp->chip_lock);
switch (iocp->ioc_cmd) {
case LB_GET_INFO_SIZE:
case LB_GET_INFO:
case LB_GET_MODE:
case LB_SET_MODE:
status = e1000g_loopback_ioctl(e1000gp, iocp, mp);
break;
#ifdef E1000G_DEBUG
case E1000G_IOC_REG_PEEK:
case E1000G_IOC_REG_POKE:
status = e1000g_pp_ioctl(e1000gp, iocp, mp);
break;
case E1000G_IOC_CHIP_RESET:
e1000gp->reset_count++;
if (e1000g_reset_adapter(e1000gp))
status = IOC_ACK;
else
status = IOC_INVAL;
break;
#endif
default:
status = IOC_INVAL;
break;
}
/*
* Decide how to reply
*/
switch (status) {
default:
case IOC_INVAL:
/*
* Error, reply with a NAK and EINVAL or the specified error
*/
miocnak(q, mp, 0, iocp->ioc_error == 0 ?
EINVAL : iocp->ioc_error);
break;
case IOC_DONE:
/*
* OK, reply already sent
*/
break;
case IOC_ACK:
/*
* OK, reply with an ACK
*/
miocack(q, mp, 0, 0);
break;
case IOC_REPLY:
/*
* OK, send prepared reply as ACK or NAK
*/
mp->b_datap->db_type = iocp->ioc_error == 0 ?
M_IOCACK : M_IOCNAK;
qreply(q, mp);
break;
}
}
/*
* The default value of e1000g_poll_mode == 0 assumes that the NIC is
* capable of supporting only one interrupt and we shouldn't disable
* the physical interrupt. In this case we let the interrupt come and
* we queue the packets in the rx ring itself in case we are in polling
* mode (better latency but slightly lower performance and a very
* high intrrupt count in mpstat which is harmless).
*
* e1000g_poll_mode == 1 assumes that we have per Rx ring interrupt
* which can be disabled in poll mode. This gives better overall
* throughput (compared to the mode above), shows very low interrupt
* count but has slightly higher latency since we pick the packets when
* the poll thread does polling.
*
* Currently, this flag should be enabled only while doing performance
* measurement or when it can be guaranteed that entire NIC going
* in poll mode will not harm any traffic like cluster heartbeat etc.
*/
int e1000g_poll_mode = 0;
/*
* Called from the upper layers when driver is in polling mode to
* pick up any queued packets. Care should be taken to not block
* this thread.
*/
static mblk_t *e1000g_poll_ring(void *arg, int bytes_to_pickup)
{
e1000g_rx_ring_t *rx_ring = (e1000g_rx_ring_t *)arg;
mblk_t *mp = NULL;
mblk_t *tail;
struct e1000g *adapter;
adapter = rx_ring->adapter;
rw_enter(&adapter->chip_lock, RW_READER);
if (adapter->e1000g_state & E1000G_SUSPENDED) {
rw_exit(&adapter->chip_lock);
return (NULL);
}
mutex_enter(&rx_ring->rx_lock);
mp = e1000g_receive(rx_ring, &tail, bytes_to_pickup);
mutex_exit(&rx_ring->rx_lock);
rw_exit(&adapter->chip_lock);
return (mp);
}
static int
e1000g_m_start(void *arg)
{
struct e1000g *Adapter = (struct e1000g *)arg;
rw_enter(&Adapter->chip_lock, RW_WRITER);
if (Adapter->e1000g_state & E1000G_SUSPENDED) {
rw_exit(&Adapter->chip_lock);
return (ECANCELED);
}
if (e1000g_start(Adapter, B_TRUE) != DDI_SUCCESS) {
rw_exit(&Adapter->chip_lock);
return (ENOTACTIVE);
}
Adapter->e1000g_state |= E1000G_STARTED;
rw_exit(&Adapter->chip_lock);
/* Enable and start the watchdog timer */
enable_watchdog_timer(Adapter);
return (0);
}
static int
e1000g_start(struct e1000g *Adapter, boolean_t global)
{
e1000g_rx_data_t *rx_data;
if (global) {
if (e1000g_alloc_rx_data(Adapter) != DDI_SUCCESS) {
e1000g_log(Adapter, CE_WARN, "Allocate rx data failed");
goto start_fail;
}
/* Allocate dma resources for descriptors and buffers */
if (e1000g_alloc_dma_resources(Adapter) != DDI_SUCCESS) {
e1000g_log(Adapter, CE_WARN,
"Alloc DMA resources failed");
goto start_fail;
}
Adapter->rx_buffer_setup = B_FALSE;
}
if (!(Adapter->attach_progress & ATTACH_PROGRESS_INIT)) {
if (e1000g_init(Adapter) != DDI_SUCCESS) {
e1000g_log(Adapter, CE_WARN,
"Adapter initialization failed");
goto start_fail;
}
}
/* Setup and initialize the transmit structures */
e1000g_tx_setup(Adapter);
msec_delay(5);
/* Setup and initialize the receive structures */
e1000g_rx_setup(Adapter);
msec_delay(5);
/* Restore the e1000g promiscuous mode */
e1000g_restore_promisc(Adapter);
e1000g_mask_interrupt(Adapter);
Adapter->attach_progress |= ATTACH_PROGRESS_INIT;
if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) {
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_LOST);
goto start_fail;
}
return (DDI_SUCCESS);
start_fail:
rx_data = Adapter->rx_ring->rx_data;
if (global) {
e1000g_release_dma_resources(Adapter);
e1000g_free_rx_pending_buffers(rx_data);
e1000g_free_rx_data(rx_data);
}
mutex_enter(&e1000g_nvm_lock);
(void) e1000_reset_hw(&Adapter->shared);
mutex_exit(&e1000g_nvm_lock);
return (DDI_FAILURE);
}
/*
* The I219 has the curious property that if the descriptor rings are not
* emptied before resetting the hardware or before changing the device state
* based on runtime power management, it'll cause the card to hang. This can
* then only be fixed by a PCI reset. As such, for the I219 and it alone, we
* have to flush the rings if we're in this state.
*/
static void
e1000g_flush_desc_rings(struct e1000g *Adapter)
{
struct e1000_hw *hw = &Adapter->shared;
u16 hang_state;
u32 fext_nvm11, tdlen;
/* First, disable MULR fix in FEXTNVM11 */
fext_nvm11 = E1000_READ_REG(hw, E1000_FEXTNVM11);
fext_nvm11 |= E1000_FEXTNVM11_DISABLE_MULR_FIX;
E1000_WRITE_REG(hw, E1000_FEXTNVM11, fext_nvm11);
/* do nothing if we're not in faulty state, or if the queue is empty */
tdlen = E1000_READ_REG(hw, E1000_TDLEN(0));
hang_state = pci_config_get16(Adapter->osdep.cfg_handle,
PCICFG_DESC_RING_STATUS);
if (!(hang_state & FLUSH_DESC_REQUIRED) || !tdlen)
return;
e1000g_flush_tx_ring(Adapter);
/* recheck, maybe the fault is caused by the rx ring */
hang_state = pci_config_get16(Adapter->osdep.cfg_handle,
PCICFG_DESC_RING_STATUS);
if (hang_state & FLUSH_DESC_REQUIRED)
e1000g_flush_rx_ring(Adapter);
}
static void
e1000g_m_stop(void *arg)
{
struct e1000g *Adapter = (struct e1000g *)arg;
/* Drain tx sessions */
(void) e1000g_tx_drain(Adapter);
rw_enter(&Adapter->chip_lock, RW_WRITER);
if (Adapter->e1000g_state & E1000G_SUSPENDED) {
rw_exit(&Adapter->chip_lock);
return;
}
Adapter->e1000g_state &= ~E1000G_STARTED;
e1000g_stop(Adapter, B_TRUE);
rw_exit(&Adapter->chip_lock);
/* Disable and stop all the timers */
disable_watchdog_timer(Adapter);
stop_link_timer(Adapter);
stop_82547_timer(Adapter->tx_ring);
}
static void
e1000g_stop(struct e1000g *Adapter, boolean_t global)
{
private_devi_list_t *devi_node;
e1000g_rx_data_t *rx_data;
int result;
Adapter->attach_progress &= ~ATTACH_PROGRESS_INIT;
/* Stop the chip and release pending resources */
/* Tell firmware driver is no longer in control */
e1000g_release_driver_control(&Adapter->shared);
e1000g_clear_all_interrupts(Adapter);
mutex_enter(&e1000g_nvm_lock);
result = e1000_reset_hw(&Adapter->shared);
mutex_exit(&e1000g_nvm_lock);
if (result != E1000_SUCCESS) {
e1000g_fm_ereport(Adapter, DDI_FM_DEVICE_INVAL_STATE);
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_LOST);
}
mutex_enter(&Adapter->link_lock);
Adapter->link_complete = B_FALSE;
mutex_exit(&Adapter->link_lock);
/* Release resources still held by the TX descriptors */
e1000g_tx_clean(Adapter);
if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK)
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_LOST);
/* Clean the pending rx jumbo packet fragment */
e1000g_rx_clean(Adapter);
/*
* The I219, eg. the pch_spt, has bugs such that we must ensure that
* rings are flushed before we do anything else. This must be done
* before we release DMA resources.
*/
if (Adapter->shared.mac.type == e1000_pch_spt)
e1000g_flush_desc_rings(Adapter);
if (global) {
e1000g_release_dma_resources(Adapter);
mutex_enter(&e1000g_rx_detach_lock);
rx_data = Adapter->rx_ring->rx_data;
rx_data->flag |= E1000G_RX_STOPPED;
if (rx_data->pending_count == 0) {
e1000g_free_rx_pending_buffers(rx_data);
e1000g_free_rx_data(rx_data);
} else {
devi_node = rx_data->priv_devi_node;
if (devi_node != NULL)
atomic_inc_32(&devi_node->pending_rx_count);
else
atomic_inc_32(&Adapter->pending_rx_count);
}
mutex_exit(&e1000g_rx_detach_lock);
}
if (Adapter->link_state != LINK_STATE_UNKNOWN) {
Adapter->link_state = LINK_STATE_UNKNOWN;
if (!Adapter->reset_flag)
mac_link_update(Adapter->mh, Adapter->link_state);
}
}
static void
e1000g_rx_clean(struct e1000g *Adapter)
{
e1000g_rx_data_t *rx_data = Adapter->rx_ring->rx_data;
if (rx_data == NULL)
return;
if (rx_data->rx_mblk != NULL) {
freemsg(rx_data->rx_mblk);
rx_data->rx_mblk = NULL;
rx_data->rx_mblk_tail = NULL;
rx_data->rx_mblk_len = 0;
}
}
static void
e1000g_tx_clean(struct e1000g *Adapter)
{
e1000g_tx_ring_t *tx_ring;
p_tx_sw_packet_t packet;
mblk_t *mp;
mblk_t *nmp;
uint32_t packet_count;
tx_ring = Adapter->tx_ring;
/*
* Here we don't need to protect the lists using
* the usedlist_lock and freelist_lock, for they
* have been protected by the chip_lock.
*/
mp = NULL;
nmp = NULL;
packet_count = 0;
packet = (p_tx_sw_packet_t)QUEUE_GET_HEAD(&tx_ring->used_list);
while (packet != NULL) {
if (packet->mp != NULL) {
/* Assemble the message chain */
if (mp == NULL) {
mp = packet->mp;
nmp = packet->mp;
} else {
nmp->b_next = packet->mp;
nmp = packet->mp;
}
/* Disconnect the message from the sw packet */
packet->mp = NULL;
}
e1000g_free_tx_swpkt(packet);
packet_count++;
packet = (p_tx_sw_packet_t)
QUEUE_GET_NEXT(&tx_ring->used_list, &packet->Link);
}
if (mp != NULL)
freemsgchain(mp);
if (packet_count > 0) {
QUEUE_APPEND(&tx_ring->free_list, &tx_ring->used_list);
QUEUE_INIT_LIST(&tx_ring->used_list);
/* Setup TX descriptor pointers */
tx_ring->tbd_next = tx_ring->tbd_first;
tx_ring->tbd_oldest = tx_ring->tbd_first;
/* Setup our HW Tx Head & Tail descriptor pointers */
E1000_WRITE_REG(&Adapter->shared, E1000_TDH(0), 0);
E1000_WRITE_REG(&Adapter->shared, E1000_TDT(0), 0);
}
}
static boolean_t
e1000g_tx_drain(struct e1000g *Adapter)
{
int i;
boolean_t done;
e1000g_tx_ring_t *tx_ring;
tx_ring = Adapter->tx_ring;
/* Allow up to 'wsdraintime' for pending xmit's to complete. */
for (i = 0; i < TX_DRAIN_TIME; i++) {
mutex_enter(&tx_ring->usedlist_lock);
done = IS_QUEUE_EMPTY(&tx_ring->used_list);
mutex_exit(&tx_ring->usedlist_lock);
if (done)
break;
msec_delay(1);
}
return (done);
}
static boolean_t
e1000g_rx_drain(struct e1000g *Adapter)
{
int i;
boolean_t done;
/*
* Allow up to RX_DRAIN_TIME for pending received packets to complete.
*/
for (i = 0; i < RX_DRAIN_TIME; i++) {
done = (Adapter->pending_rx_count == 0);
if (done)
break;
msec_delay(1);
}
return (done);
}
static boolean_t
e1000g_reset_adapter(struct e1000g *Adapter)
{
/* Disable and stop all the timers */
disable_watchdog_timer(Adapter);
stop_link_timer(Adapter);
stop_82547_timer(Adapter->tx_ring);
rw_enter(&Adapter->chip_lock, RW_WRITER);
if (Adapter->stall_flag) {
Adapter->stall_flag = B_FALSE;
Adapter->reset_flag = B_TRUE;
}
if (!(Adapter->e1000g_state & E1000G_STARTED)) {
rw_exit(&Adapter->chip_lock);
return (B_TRUE);
}
e1000g_stop(Adapter, B_FALSE);
if (e1000g_start(Adapter, B_FALSE) != DDI_SUCCESS) {
rw_exit(&Adapter->chip_lock);
e1000g_log(Adapter, CE_WARN, "Reset failed");
return (B_FALSE);
}
rw_exit(&Adapter->chip_lock);
/* Enable and start the watchdog timer */
enable_watchdog_timer(Adapter);
return (B_TRUE);
}
boolean_t
e1000g_global_reset(struct e1000g *Adapter)
{
/* Disable and stop all the timers */
disable_watchdog_timer(Adapter);
stop_link_timer(Adapter);
stop_82547_timer(Adapter->tx_ring);
rw_enter(&Adapter->chip_lock, RW_WRITER);
e1000g_stop(Adapter, B_TRUE);
Adapter->init_count = 0;
if (e1000g_start(Adapter, B_TRUE) != DDI_SUCCESS) {
rw_exit(&Adapter->chip_lock);
e1000g_log(Adapter, CE_WARN, "Reset failed");
return (B_FALSE);
}
rw_exit(&Adapter->chip_lock);
/* Enable and start the watchdog timer */
enable_watchdog_timer(Adapter);
return (B_TRUE);
}
/*
* e1000g_intr_pciexpress - ISR for PCI Express chipsets
*
* This interrupt service routine is for PCI-Express adapters.
* The ICR contents is valid only when the E1000_ICR_INT_ASSERTED
* bit is set.
*/
static uint_t
e1000g_intr_pciexpress(caddr_t arg)
{
struct e1000g *Adapter;
uint32_t icr;
Adapter = (struct e1000g *)(uintptr_t)arg;
icr = E1000_READ_REG(&Adapter->shared, E1000_ICR);
if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) {
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED);
return (DDI_INTR_CLAIMED);
}
if (icr & E1000_ICR_INT_ASSERTED) {
/*
* E1000_ICR_INT_ASSERTED bit was set:
* Read(Clear) the ICR, claim this interrupt,
* look for work to do.
*/
e1000g_intr_work(Adapter, icr);
return (DDI_INTR_CLAIMED);
} else {
/*
* E1000_ICR_INT_ASSERTED bit was not set:
* Don't claim this interrupt, return immediately.
*/
return (DDI_INTR_UNCLAIMED);
}
}
/*
* e1000g_intr - ISR for PCI/PCI-X chipsets
*
* This interrupt service routine is for PCI/PCI-X adapters.
* We check the ICR contents no matter the E1000_ICR_INT_ASSERTED
* bit is set or not.
*/
static uint_t
e1000g_intr(caddr_t arg)
{
struct e1000g *Adapter;
uint32_t icr;
Adapter = (struct e1000g *)(uintptr_t)arg;
icr = E1000_READ_REG(&Adapter->shared, E1000_ICR);
if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) {
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED);
return (DDI_INTR_CLAIMED);
}
if (icr) {
/*
* Any bit was set in ICR:
* Read(Clear) the ICR, claim this interrupt,
* look for work to do.
*/
e1000g_intr_work(Adapter, icr);
return (DDI_INTR_CLAIMED);
} else {
/*
* No bit was set in ICR:
* Don't claim this interrupt, return immediately.
*/
return (DDI_INTR_UNCLAIMED);
}
}
/*
* e1000g_intr_work - actual processing of ISR
*
* Read(clear) the ICR contents and call appropriate interrupt
* processing routines.
*/
static void
e1000g_intr_work(struct e1000g *Adapter, uint32_t icr)
{
struct e1000_hw *hw;
hw = &Adapter->shared;
e1000g_tx_ring_t *tx_ring = Adapter->tx_ring;
Adapter->rx_pkt_cnt = 0;
Adapter->tx_pkt_cnt = 0;
rw_enter(&Adapter->chip_lock, RW_READER);
if (Adapter->e1000g_state & E1000G_SUSPENDED) {
rw_exit(&Adapter->chip_lock);
return;
}
/*
* Here we need to check the "e1000g_state" flag within the chip_lock to
* ensure the receive routine will not execute when the adapter is
* being reset.
*/
if (!(Adapter->e1000g_state & E1000G_STARTED)) {
rw_exit(&Adapter->chip_lock);
return;
}
if (icr & E1000_ICR_RXT0) {
mblk_t *mp = NULL;
mblk_t *tail = NULL;
e1000g_rx_ring_t *rx_ring;
rx_ring = Adapter->rx_ring;
mutex_enter(&rx_ring->rx_lock);
/*
* Sometimes with legacy interrupts, it possible that
* there is a single interrupt for Rx/Tx. In which
* case, if poll flag is set, we shouldn't really
* be doing Rx processing.
*/
if (!rx_ring->poll_flag)
mp = e1000g_receive(rx_ring, &tail,
E1000G_CHAIN_NO_LIMIT);
mutex_exit(&rx_ring->rx_lock);
rw_exit(&Adapter->chip_lock);
if (mp != NULL)
mac_rx_ring(Adapter->mh, rx_ring->mrh,
mp, rx_ring->ring_gen_num);
} else
rw_exit(&Adapter->chip_lock);
if (icr & E1000_ICR_TXDW) {
if (!Adapter->tx_intr_enable)
e1000g_clear_tx_interrupt(Adapter);
/* Recycle the tx descriptors */
rw_enter(&Adapter->chip_lock, RW_READER);
(void) e1000g_recycle(tx_ring);
E1000G_DEBUG_STAT(tx_ring->stat_recycle_intr);
rw_exit(&Adapter->chip_lock);
if (tx_ring->resched_needed &&
(tx_ring->tbd_avail > DEFAULT_TX_UPDATE_THRESHOLD)) {
tx_ring->resched_needed = B_FALSE;
mac_tx_update(Adapter->mh);
E1000G_STAT(tx_ring->stat_reschedule);
}
}
/*
* The Receive Sequence errors RXSEQ and the link status change LSC
* are checked to detect that the cable has been pulled out. For
* the Wiseman 2.0 silicon, the receive sequence errors interrupt
* are an indication that cable is not connected.
*/
if ((icr & E1000_ICR_RXSEQ) ||
(icr & E1000_ICR_LSC) ||
(icr & E1000_ICR_GPI_EN1)) {
boolean_t link_changed;
timeout_id_t tid = 0;
stop_watchdog_timer(Adapter);
rw_enter(&Adapter->chip_lock, RW_WRITER);
/*
* Because we got a link-status-change interrupt, force
* e1000_check_for_link() to look at phy
*/
Adapter->shared.mac.get_link_status = B_TRUE;
/* e1000g_link_check takes care of link status change */
link_changed = e1000g_link_check(Adapter);
/* Get new phy state */
e1000g_get_phy_state(Adapter);
/*
* If the link timer has not timed out, we'll not notify
* the upper layer with any link state until the link is up.
*/
if (link_changed && !Adapter->link_complete) {
if (Adapter->link_state == LINK_STATE_UP) {
mutex_enter(&Adapter->link_lock);
Adapter->link_complete = B_TRUE;
tid = Adapter->link_tid;
Adapter->link_tid = 0;
mutex_exit(&Adapter->link_lock);
} else {
link_changed = B_FALSE;
}
}
rw_exit(&Adapter->chip_lock);
if (link_changed) {
if (tid != 0)
(void) untimeout(tid);
/*
* Workaround for esb2. Data stuck in fifo on a link
* down event. Stop receiver here and reset in watchdog.
*/
if ((Adapter->link_state == LINK_STATE_DOWN) &&
(Adapter->shared.mac.type == e1000_80003es2lan)) {
uint32_t rctl = E1000_READ_REG(hw, E1000_RCTL);
E1000_WRITE_REG(hw, E1000_RCTL,
rctl & ~E1000_RCTL_EN);
e1000g_log(Adapter, CE_WARN,
"ESB2 receiver disabled");
Adapter->esb2_workaround = B_TRUE;
}
if (!Adapter->reset_flag)
mac_link_update(Adapter->mh,
Adapter->link_state);
if (Adapter->link_state == LINK_STATE_UP)
Adapter->reset_flag = B_FALSE;
}
start_watchdog_timer(Adapter);
}
}
static void
e1000g_init_unicst(struct e1000g *Adapter)
{
struct e1000_hw *hw;
int slot;
hw = &Adapter->shared;
if (Adapter->init_count == 0) {
/* Initialize the multiple unicast addresses */
Adapter->unicst_total = min(hw->mac.rar_entry_count,
MAX_NUM_UNICAST_ADDRESSES);
/*
* The common code does not correctly calculate the number of
* rar's that could be reserved by firmware for the pch_lpt and
* pch_spt macs. The interface has one primary rar, and 11
* additional ones. Those 11 additional ones are not always
* available. According to the datasheet, we need to check a
* few of the bits set in the FWSM register. If the value is
* zero, everything is available. If the value is 1, none of the
* additional registers are available. If the value is 2-7, only
* that number are available.
*/
if (hw->mac.type == e1000_pch_lpt ||
hw->mac.type == e1000_pch_spt) {
uint32_t locked, rar;
locked = E1000_READ_REG(hw, E1000_FWSM) &
E1000_FWSM_WLOCK_MAC_MASK;
locked >>= E1000_FWSM_WLOCK_MAC_SHIFT;
rar = 1;
if (locked == 0)
rar += 11;
else if (locked == 1)
rar += 0;
else
rar += locked;
Adapter->unicst_total = min(rar,
MAX_NUM_UNICAST_ADDRESSES);
}
/* Workaround for an erratum of 82571 chipst */
if ((hw->mac.type == e1000_82571) &&
(e1000_get_laa_state_82571(hw) == B_TRUE))
Adapter->unicst_total--;
/* VMware doesn't support multiple mac addresses properly */
if (hw->subsystem_vendor_id == 0x15ad)
Adapter->unicst_total = 1;
Adapter->unicst_avail = Adapter->unicst_total;
for (slot = 0; slot < Adapter->unicst_total; slot++) {
/* Clear both the flag and MAC address */
Adapter->unicst_addr[slot].reg.high = 0;
Adapter->unicst_addr[slot].reg.low = 0;
}
} else {
/* Workaround for an erratum of 82571 chipst */
if ((hw->mac.type == e1000_82571) &&
(e1000_get_laa_state_82571(hw) == B_TRUE))
(void) e1000_rar_set(hw, hw->mac.addr, LAST_RAR_ENTRY);
/* Re-configure the RAR registers */
for (slot = 0; slot < Adapter->unicst_total; slot++)
if (Adapter->unicst_addr[slot].mac.set == 1)
(void) e1000_rar_set(hw,
Adapter->unicst_addr[slot].mac.addr, slot);
}
if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK)
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED);
}
static int
e1000g_unicst_set(struct e1000g *Adapter, const uint8_t *mac_addr,
int slot)
{
struct e1000_hw *hw;
hw = &Adapter->shared;
/*
* The first revision of Wiseman silicon (rev 2.0) has an errata
* that requires the receiver to be in reset when any of the
* receive address registers (RAR regs) are accessed. The first
* rev of Wiseman silicon also requires MWI to be disabled when
* a global reset or a receive reset is issued. So before we
* initialize the RARs, we check the rev of the Wiseman controller
* and work around any necessary HW errata.
*/
if ((hw->mac.type == e1000_82542) &&
(hw->revision_id == E1000_REVISION_2)) {
e1000_pci_clear_mwi(hw);
E1000_WRITE_REG(hw, E1000_RCTL, E1000_RCTL_RST);
msec_delay(5);
}
if (mac_addr == NULL) {
E1000_WRITE_REG_ARRAY(hw, E1000_RA, slot << 1, 0);
E1000_WRITE_FLUSH(hw);
E1000_WRITE_REG_ARRAY(hw, E1000_RA, (slot << 1) + 1, 0);
E1000_WRITE_FLUSH(hw);
/* Clear both the flag and MAC address */
Adapter->unicst_addr[slot].reg.high = 0;
Adapter->unicst_addr[slot].reg.low = 0;
} else {
bcopy(mac_addr, Adapter->unicst_addr[slot].mac.addr,
ETHERADDRL);
(void) e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
Adapter->unicst_addr[slot].mac.set = 1;
}
/* Workaround for an erratum of 82571 chipst */
if (slot == 0) {
if ((hw->mac.type == e1000_82571) &&
(e1000_get_laa_state_82571(hw) == B_TRUE))
if (mac_addr == NULL) {
E1000_WRITE_REG_ARRAY(hw, E1000_RA,
slot << 1, 0);
E1000_WRITE_FLUSH(hw);
E1000_WRITE_REG_ARRAY(hw, E1000_RA,
(slot << 1) + 1, 0);
E1000_WRITE_FLUSH(hw);
} else {
(void) e1000_rar_set(hw, (uint8_t *)mac_addr,
LAST_RAR_ENTRY);
}
}
/*
* If we are using Wiseman rev 2.0 silicon, we will have previously
* put the receive in reset, and disabled MWI, to work around some
* HW errata. Now we should take the receiver out of reset, and
* re-enabled if MWI if it was previously enabled by the PCI BIOS.
*/
if ((hw->mac.type == e1000_82542) &&
(hw->revision_id == E1000_REVISION_2)) {
E1000_WRITE_REG(hw, E1000_RCTL, 0);
msec_delay(1);
if (hw->bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
e1000_pci_set_mwi(hw);
e1000g_rx_setup(Adapter);
}
if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) {
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED);
return (EIO);
}
return (0);
}
static int
multicst_add(struct e1000g *Adapter, const uint8_t *multiaddr)
{
struct e1000_hw *hw = &Adapter->shared;
struct ether_addr *newtable;
size_t new_len;
size_t old_len;
int res = 0;
if ((multiaddr[0] & 01) == 0) {
res = EINVAL;
e1000g_log(Adapter, CE_WARN, "Illegal multicast address");
goto done;
}
if (Adapter->mcast_count >= Adapter->mcast_max_num) {
res = ENOENT;
e1000g_log(Adapter, CE_WARN,
"Adapter requested more than %d mcast addresses",
Adapter->mcast_max_num);
goto done;
}
if (Adapter->mcast_count == Adapter->mcast_alloc_count) {
old_len = Adapter->mcast_alloc_count *
sizeof (struct ether_addr);
new_len = (Adapter->mcast_alloc_count + MCAST_ALLOC_SIZE) *
sizeof (struct ether_addr);
newtable = kmem_alloc(new_len, KM_NOSLEEP);
if (newtable == NULL) {
res = ENOMEM;
e1000g_log(Adapter, CE_WARN,
"Not enough memory to alloc mcast table");
goto done;
}
if (Adapter->mcast_table != NULL) {
bcopy(Adapter->mcast_table, newtable, old_len);
kmem_free(Adapter->mcast_table, old_len);
}
Adapter->mcast_alloc_count += MCAST_ALLOC_SIZE;
Adapter->mcast_table = newtable;
}
bcopy(multiaddr,
&Adapter->mcast_table[Adapter->mcast_count], ETHERADDRL);
Adapter->mcast_count++;
/*
* Update the MC table in the hardware
*/
e1000g_clear_interrupt(Adapter);
e1000_update_mc_addr_list(hw,
(uint8_t *)Adapter->mcast_table, Adapter->mcast_count);
e1000g_mask_interrupt(Adapter);
if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) {
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED);
res = EIO;
}
done:
return (res);
}
static int
multicst_remove(struct e1000g *Adapter, const uint8_t *multiaddr)
{
struct e1000_hw *hw = &Adapter->shared;
struct ether_addr *newtable;
size_t new_len;
size_t old_len;
unsigned i;
for (i = 0; i < Adapter->mcast_count; i++) {
if (bcmp(multiaddr, &Adapter->mcast_table[i],
ETHERADDRL) == 0) {
for (i++; i < Adapter->mcast_count; i++) {
Adapter->mcast_table[i - 1] =
Adapter->mcast_table[i];
}
Adapter->mcast_count--;
break;
}
}
if ((Adapter->mcast_alloc_count - Adapter->mcast_count) >
MCAST_ALLOC_SIZE) {
old_len = Adapter->mcast_alloc_count *
sizeof (struct ether_addr);
new_len = (Adapter->mcast_alloc_count - MCAST_ALLOC_SIZE) *
sizeof (struct ether_addr);
newtable = kmem_alloc(new_len, KM_NOSLEEP);
if (newtable != NULL) {
bcopy(Adapter->mcast_table, newtable, new_len);
kmem_free(Adapter->mcast_table, old_len);
Adapter->mcast_alloc_count -= MCAST_ALLOC_SIZE;
Adapter->mcast_table = newtable;
}
}
/*
* Update the MC table in the hardware
*/
e1000g_clear_interrupt(Adapter);
e1000_update_mc_addr_list(hw,
(uint8_t *)Adapter->mcast_table, Adapter->mcast_count);
e1000g_mask_interrupt(Adapter);
if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) {
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED);
return (EIO);
}
return (0);
}
static void
e1000g_release_multicast(struct e1000g *Adapter)
{
if (Adapter->mcast_table != NULL) {
kmem_free(Adapter->mcast_table,
Adapter->mcast_alloc_count * sizeof (struct ether_addr));
Adapter->mcast_table = NULL;
}
}
int
e1000g_m_multicst(void *arg, boolean_t add, const uint8_t *addr)
{
struct e1000g *Adapter = (struct e1000g *)arg;
int result;
rw_enter(&Adapter->chip_lock, RW_WRITER);
if (Adapter->e1000g_state & E1000G_SUSPENDED) {
result = ECANCELED;
goto done;
}
result = (add) ? multicst_add(Adapter, addr)
: multicst_remove(Adapter, addr);
done:
rw_exit(&Adapter->chip_lock);
return (result);
}
int
e1000g_m_promisc(void *arg, boolean_t on)
{
struct e1000g *Adapter = (struct e1000g *)arg;
uint32_t rctl;
rw_enter(&Adapter->chip_lock, RW_WRITER);
if (Adapter->e1000g_state & E1000G_SUSPENDED) {
rw_exit(&Adapter->chip_lock);
return (ECANCELED);
}
rctl = E1000_READ_REG(&Adapter->shared, E1000_RCTL);
if (on)
rctl |=
(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_BAM);
else
rctl &= (~(E1000_RCTL_UPE | E1000_RCTL_MPE));
E1000_WRITE_REG(&Adapter->shared, E1000_RCTL, rctl);
Adapter->e1000g_promisc = on;
rw_exit(&Adapter->chip_lock);
if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) {
ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED);
return (EIO);