2354544c3cdc172ed440ef0c74fc6f924844245c
[dpdk.git] / drivers / net / e1000 / em_ethdev.c
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33
34 #include <sys/queue.h>
35 #include <stdio.h>
36 #include <errno.h>
37 #include <stdint.h>
38 #include <stdarg.h>
39
40 #include <rte_common.h>
41 #include <rte_interrupts.h>
42 #include <rte_byteorder.h>
43 #include <rte_log.h>
44 #include <rte_debug.h>
45 #include <rte_pci.h>
46 #include <rte_ether.h>
47 #include <rte_ethdev.h>
48 #include <rte_memory.h>
49 #include <rte_memzone.h>
50 #include <rte_eal.h>
51 #include <rte_atomic.h>
52 #include <rte_malloc.h>
53 #include <rte_dev.h>
54
55 #include "e1000_logs.h"
56 #include "base/e1000_api.h"
57 #include "e1000_ethdev.h"
58
59 #define EM_EIAC                 0x000DC
60
61 #define PMD_ROUNDUP(x,y)        (((x) + (y) - 1)/(y) * (y))
62
63
64 static int eth_em_configure(struct rte_eth_dev *dev);
65 static int eth_em_start(struct rte_eth_dev *dev);
66 static void eth_em_stop(struct rte_eth_dev *dev);
67 static void eth_em_close(struct rte_eth_dev *dev);
68 static void eth_em_promiscuous_enable(struct rte_eth_dev *dev);
69 static void eth_em_promiscuous_disable(struct rte_eth_dev *dev);
70 static void eth_em_allmulticast_enable(struct rte_eth_dev *dev);
71 static void eth_em_allmulticast_disable(struct rte_eth_dev *dev);
72 static int eth_em_link_update(struct rte_eth_dev *dev,
73                                 int wait_to_complete);
74 static void eth_em_stats_get(struct rte_eth_dev *dev,
75                                 struct rte_eth_stats *rte_stats);
76 static void eth_em_stats_reset(struct rte_eth_dev *dev);
77 static void eth_em_infos_get(struct rte_eth_dev *dev,
78                                 struct rte_eth_dev_info *dev_info);
79 static int eth_em_flow_ctrl_get(struct rte_eth_dev *dev,
80                                 struct rte_eth_fc_conf *fc_conf);
81 static int eth_em_flow_ctrl_set(struct rte_eth_dev *dev,
82                                 struct rte_eth_fc_conf *fc_conf);
83 static int eth_em_interrupt_setup(struct rte_eth_dev *dev);
84 static int eth_em_rxq_interrupt_setup(struct rte_eth_dev *dev);
85 static int eth_em_interrupt_get_status(struct rte_eth_dev *dev);
86 static int eth_em_interrupt_action(struct rte_eth_dev *dev);
87 static void eth_em_interrupt_handler(struct rte_intr_handle *handle,
88                                                         void *param);
89
90 static int em_hw_init(struct e1000_hw *hw);
91 static int em_hardware_init(struct e1000_hw *hw);
92 static void em_hw_control_acquire(struct e1000_hw *hw);
93 static void em_hw_control_release(struct e1000_hw *hw);
94 static void em_init_manageability(struct e1000_hw *hw);
95 static void em_release_manageability(struct e1000_hw *hw);
96
97 static int eth_em_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
98
99 static int eth_em_vlan_filter_set(struct rte_eth_dev *dev,
100                 uint16_t vlan_id, int on);
101 static void eth_em_vlan_offload_set(struct rte_eth_dev *dev, int mask);
102 static void em_vlan_hw_filter_enable(struct rte_eth_dev *dev);
103 static void em_vlan_hw_filter_disable(struct rte_eth_dev *dev);
104 static void em_vlan_hw_strip_enable(struct rte_eth_dev *dev);
105 static void em_vlan_hw_strip_disable(struct rte_eth_dev *dev);
106
107 /*
108 static void eth_em_vlan_filter_set(struct rte_eth_dev *dev,
109                                         uint16_t vlan_id, int on);
110 */
111
112 static int eth_em_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id);
113 static int eth_em_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id);
114 static void em_lsc_intr_disable(struct e1000_hw *hw);
115 static void em_rxq_intr_enable(struct e1000_hw *hw);
116 static void em_rxq_intr_disable(struct e1000_hw *hw);
117
118 static int eth_em_led_on(struct rte_eth_dev *dev);
119 static int eth_em_led_off(struct rte_eth_dev *dev);
120
121 static int em_get_rx_buffer_size(struct e1000_hw *hw);
122 static void eth_em_rar_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
123                 uint32_t index, uint32_t pool);
124 static void eth_em_rar_clear(struct rte_eth_dev *dev, uint32_t index);
125
126 static int eth_em_set_mc_addr_list(struct rte_eth_dev *dev,
127                                    struct ether_addr *mc_addr_set,
128                                    uint32_t nb_mc_addr);
129
130 #define EM_FC_PAUSE_TIME 0x0680
131 #define EM_LINK_UPDATE_CHECK_TIMEOUT  90  /* 9s */
132 #define EM_LINK_UPDATE_CHECK_INTERVAL 100 /* ms */
133
134 static enum e1000_fc_mode em_fc_setting = e1000_fc_full;
135
136 /*
137  * The set of PCI devices this driver supports
138  */
139 static const struct rte_pci_id pci_id_em_map[] = {
140
141 #define RTE_PCI_DEV_ID_DECL_EM(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
142 #include "rte_pci_dev_ids.h"
143
144 {0},
145 };
146
147 static const struct eth_dev_ops eth_em_ops = {
148         .dev_configure        = eth_em_configure,
149         .dev_start            = eth_em_start,
150         .dev_stop             = eth_em_stop,
151         .dev_close            = eth_em_close,
152         .promiscuous_enable   = eth_em_promiscuous_enable,
153         .promiscuous_disable  = eth_em_promiscuous_disable,
154         .allmulticast_enable  = eth_em_allmulticast_enable,
155         .allmulticast_disable = eth_em_allmulticast_disable,
156         .link_update          = eth_em_link_update,
157         .stats_get            = eth_em_stats_get,
158         .stats_reset          = eth_em_stats_reset,
159         .dev_infos_get        = eth_em_infos_get,
160         .mtu_set              = eth_em_mtu_set,
161         .vlan_filter_set      = eth_em_vlan_filter_set,
162         .vlan_offload_set     = eth_em_vlan_offload_set,
163         .rx_queue_setup       = eth_em_rx_queue_setup,
164         .rx_queue_release     = eth_em_rx_queue_release,
165         .rx_queue_count       = eth_em_rx_queue_count,
166         .rx_descriptor_done   = eth_em_rx_descriptor_done,
167         .tx_queue_setup       = eth_em_tx_queue_setup,
168         .tx_queue_release     = eth_em_tx_queue_release,
169         .rx_queue_intr_enable = eth_em_rx_queue_intr_enable,
170         .rx_queue_intr_disable = eth_em_rx_queue_intr_disable,
171         .dev_led_on           = eth_em_led_on,
172         .dev_led_off          = eth_em_led_off,
173         .flow_ctrl_get        = eth_em_flow_ctrl_get,
174         .flow_ctrl_set        = eth_em_flow_ctrl_set,
175         .mac_addr_add         = eth_em_rar_set,
176         .mac_addr_remove      = eth_em_rar_clear,
177         .set_mc_addr_list     = eth_em_set_mc_addr_list,
178 };
179
180 /**
181  * Atomically reads the link status information from global
182  * structure rte_eth_dev.
183  *
184  * @param dev
185  *   - Pointer to the structure rte_eth_dev to read from.
186  *   - Pointer to the buffer to be saved with the link status.
187  *
188  * @return
189  *   - On success, zero.
190  *   - On failure, negative value.
191  */
192 static inline int
193 rte_em_dev_atomic_read_link_status(struct rte_eth_dev *dev,
194                                 struct rte_eth_link *link)
195 {
196         struct rte_eth_link *dst = link;
197         struct rte_eth_link *src = &(dev->data->dev_link);
198
199         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
200                                         *(uint64_t *)src) == 0)
201                 return -1;
202
203         return 0;
204 }
205
206 /**
207  * Atomically writes the link status information into global
208  * structure rte_eth_dev.
209  *
210  * @param dev
211  *   - Pointer to the structure rte_eth_dev to read from.
212  *   - Pointer to the buffer to be saved with the link status.
213  *
214  * @return
215  *   - On success, zero.
216  *   - On failure, negative value.
217  */
218 static inline int
219 rte_em_dev_atomic_write_link_status(struct rte_eth_dev *dev,
220                                 struct rte_eth_link *link)
221 {
222         struct rte_eth_link *dst = &(dev->data->dev_link);
223         struct rte_eth_link *src = link;
224
225         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
226                                         *(uint64_t *)src) == 0)
227                 return -1;
228
229         return 0;
230 }
231
232 static int
233 eth_em_dev_init(struct rte_eth_dev *eth_dev)
234 {
235         struct rte_pci_device *pci_dev;
236         struct e1000_adapter *adapter =
237                 E1000_DEV_PRIVATE(eth_dev->data->dev_private);
238         struct e1000_hw *hw =
239                 E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
240         struct e1000_vfta * shadow_vfta =
241                 E1000_DEV_PRIVATE_TO_VFTA(eth_dev->data->dev_private);
242
243         pci_dev = eth_dev->pci_dev;
244         eth_dev->dev_ops = &eth_em_ops;
245         eth_dev->rx_pkt_burst = (eth_rx_burst_t)&eth_em_recv_pkts;
246         eth_dev->tx_pkt_burst = (eth_tx_burst_t)&eth_em_xmit_pkts;
247
248         /* for secondary processes, we don't initialise any further as primary
249          * has already done this work. Only check we don't need a different
250          * RX function */
251         if (rte_eal_process_type() != RTE_PROC_PRIMARY){
252                 if (eth_dev->data->scattered_rx)
253                         eth_dev->rx_pkt_burst =
254                                 (eth_rx_burst_t)&eth_em_recv_scattered_pkts;
255                 return 0;
256         }
257
258         hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
259         hw->device_id = pci_dev->id.device_id;
260         adapter->stopped = 0;
261
262         /* For ICH8 support we'll need to map the flash memory BAR */
263
264         if (e1000_setup_init_funcs(hw, TRUE) != E1000_SUCCESS ||
265                         em_hw_init(hw) != 0) {
266                 PMD_INIT_LOG(ERR, "port_id %d vendorID=0x%x deviceID=0x%x: "
267                         "failed to init HW",
268                         eth_dev->data->port_id, pci_dev->id.vendor_id,
269                         pci_dev->id.device_id);
270                 return -(ENODEV);
271         }
272
273         /* Allocate memory for storing MAC addresses */
274         eth_dev->data->mac_addrs = rte_zmalloc("e1000", ETHER_ADDR_LEN *
275                         hw->mac.rar_entry_count, 0);
276         if (eth_dev->data->mac_addrs == NULL) {
277                 PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to "
278                         "store MAC addresses",
279                         ETHER_ADDR_LEN * hw->mac.rar_entry_count);
280                 return -(ENOMEM);
281         }
282
283         /* Copy the permanent MAC address */
284         ether_addr_copy((struct ether_addr *) hw->mac.addr,
285                 eth_dev->data->mac_addrs);
286
287         /* initialize the vfta */
288         memset(shadow_vfta, 0, sizeof(*shadow_vfta));
289
290         PMD_INIT_LOG(DEBUG, "port_id %d vendorID=0x%x deviceID=0x%x",
291                      eth_dev->data->port_id, pci_dev->id.vendor_id,
292                      pci_dev->id.device_id);
293
294         rte_intr_callback_register(&(pci_dev->intr_handle),
295                 eth_em_interrupt_handler, (void *)eth_dev);
296
297         return (0);
298 }
299
300 static int
301 eth_em_dev_uninit(struct rte_eth_dev *eth_dev)
302 {
303         struct rte_pci_device *pci_dev;
304         struct e1000_adapter *adapter =
305                 E1000_DEV_PRIVATE(eth_dev->data->dev_private);
306
307         PMD_INIT_FUNC_TRACE();
308
309         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
310                 return -EPERM;
311
312         pci_dev = eth_dev->pci_dev;
313
314         if (adapter->stopped == 0)
315                 eth_em_close(eth_dev);
316
317         eth_dev->dev_ops = NULL;
318         eth_dev->rx_pkt_burst = NULL;
319         eth_dev->tx_pkt_burst = NULL;
320
321         rte_free(eth_dev->data->mac_addrs);
322         eth_dev->data->mac_addrs = NULL;
323
324         /* disable uio intr before callback unregister */
325         rte_intr_disable(&(pci_dev->intr_handle));
326         rte_intr_callback_unregister(&(pci_dev->intr_handle),
327                 eth_em_interrupt_handler, (void *)eth_dev);
328
329         return 0;
330 }
331
332 static struct eth_driver rte_em_pmd = {
333         .pci_drv = {
334                 .name = "rte_em_pmd",
335                 .id_table = pci_id_em_map,
336                 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
337                         RTE_PCI_DRV_DETACHABLE,
338         },
339         .eth_dev_init = eth_em_dev_init,
340         .eth_dev_uninit = eth_em_dev_uninit,
341         .dev_private_size = sizeof(struct e1000_adapter),
342 };
343
344 static int
345 rte_em_pmd_init(const char *name __rte_unused, const char *params __rte_unused)
346 {
347         rte_eth_driver_register(&rte_em_pmd);
348         return 0;
349 }
350
351 static int
352 em_hw_init(struct e1000_hw *hw)
353 {
354         int diag;
355
356         diag = hw->mac.ops.init_params(hw);
357         if (diag != 0) {
358                 PMD_INIT_LOG(ERR, "MAC Initialization Error");
359                 return diag;
360         }
361         diag = hw->nvm.ops.init_params(hw);
362         if (diag != 0) {
363                 PMD_INIT_LOG(ERR, "NVM Initialization Error");
364                 return diag;
365         }
366         diag = hw->phy.ops.init_params(hw);
367         if (diag != 0) {
368                 PMD_INIT_LOG(ERR, "PHY Initialization Error");
369                 return diag;
370         }
371         (void) e1000_get_bus_info(hw);
372
373         hw->mac.autoneg = 1;
374         hw->phy.autoneg_wait_to_complete = 0;
375         hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
376
377         e1000_init_script_state_82541(hw, TRUE);
378         e1000_set_tbi_compatibility_82543(hw, TRUE);
379
380         /* Copper options */
381         if (hw->phy.media_type == e1000_media_type_copper) {
382                 hw->phy.mdix = 0; /* AUTO_ALL_MODES */
383                 hw->phy.disable_polarity_correction = 0;
384                 hw->phy.ms_type = e1000_ms_hw_default;
385         }
386
387         /*
388          * Start from a known state, this is important in reading the nvm
389          * and mac from that.
390          */
391         e1000_reset_hw(hw);
392
393         /* Make sure we have a good EEPROM before we read from it */
394         if (e1000_validate_nvm_checksum(hw) < 0) {
395                 /*
396                  * Some PCI-E parts fail the first check due to
397                  * the link being in sleep state, call it again,
398                  * if it fails a second time its a real issue.
399                  */
400                 diag = e1000_validate_nvm_checksum(hw);
401                 if (diag < 0) {
402                         PMD_INIT_LOG(ERR, "EEPROM checksum invalid");
403                         goto error;
404                 }
405         }
406
407         /* Read the permanent MAC address out of the EEPROM */
408         diag = e1000_read_mac_addr(hw);
409         if (diag != 0) {
410                 PMD_INIT_LOG(ERR, "EEPROM error while reading MAC address");
411                 goto error;
412         }
413
414         /* Now initialize the hardware */
415         diag = em_hardware_init(hw);
416         if (diag != 0) {
417                 PMD_INIT_LOG(ERR, "Hardware initialization failed");
418                 goto error;
419         }
420
421         hw->mac.get_link_status = 1;
422
423         /* Indicate SOL/IDER usage */
424         diag = e1000_check_reset_block(hw);
425         if (diag < 0) {
426                 PMD_INIT_LOG(ERR, "PHY reset is blocked due to "
427                         "SOL/IDER session");
428         }
429         return (0);
430
431 error:
432         em_hw_control_release(hw);
433         return (diag);
434 }
435
436 static int
437 eth_em_configure(struct rte_eth_dev *dev)
438 {
439         struct e1000_interrupt *intr =
440                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
441
442         PMD_INIT_FUNC_TRACE();
443         intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
444         PMD_INIT_FUNC_TRACE();
445
446         return (0);
447 }
448
449 static void
450 em_set_pba(struct e1000_hw *hw)
451 {
452         uint32_t pba;
453
454         /*
455          * Packet Buffer Allocation (PBA)
456          * Writing PBA sets the receive portion of the buffer
457          * the remainder is used for the transmit buffer.
458          * Devices before the 82547 had a Packet Buffer of 64K.
459          * After the 82547 the buffer was reduced to 40K.
460          */
461         switch (hw->mac.type) {
462                 case e1000_82547:
463                 case e1000_82547_rev_2:
464                 /* 82547: Total Packet Buffer is 40K */
465                         pba = E1000_PBA_22K; /* 22K for Rx, 18K for Tx */
466                         break;
467                 case e1000_82571:
468                 case e1000_82572:
469                 case e1000_80003es2lan:
470                         pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
471                         break;
472                 case e1000_82573: /* 82573: Total Packet Buffer is 32K */
473                         pba = E1000_PBA_12K; /* 12K for Rx, 20K for Tx */
474                         break;
475                 case e1000_82574:
476                 case e1000_82583:
477                         pba = E1000_PBA_20K; /* 20K for Rx, 20K for Tx */
478                         break;
479                 case e1000_ich8lan:
480                         pba = E1000_PBA_8K;
481                         break;
482                 case e1000_ich9lan:
483                 case e1000_ich10lan:
484                         pba = E1000_PBA_10K;
485                         break;
486                 case e1000_pchlan:
487                 case e1000_pch2lan:
488                         pba = E1000_PBA_26K;
489                         break;
490                 default:
491                         pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */
492         }
493
494         E1000_WRITE_REG(hw, E1000_PBA, pba);
495 }
496
497 static int
498 eth_em_start(struct rte_eth_dev *dev)
499 {
500         struct e1000_adapter *adapter =
501                 E1000_DEV_PRIVATE(dev->data->dev_private);
502         struct e1000_hw *hw =
503                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
504         struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
505         int ret, mask;
506         uint32_t intr_vector = 0;
507
508         PMD_INIT_FUNC_TRACE();
509
510         eth_em_stop(dev);
511
512         e1000_power_up_phy(hw);
513
514         /* Set default PBA value */
515         em_set_pba(hw);
516
517         /* Put the address into the Receive Address Array */
518         e1000_rar_set(hw, hw->mac.addr, 0);
519
520         /*
521          * With the 82571 adapter, RAR[0] may be overwritten
522          * when the other port is reset, we make a duplicate
523          * in RAR[14] for that eventuality, this assures
524          * the interface continues to function.
525          */
526         if (hw->mac.type == e1000_82571) {
527                 e1000_set_laa_state_82571(hw, TRUE);
528                 e1000_rar_set(hw, hw->mac.addr, E1000_RAR_ENTRIES - 1);
529         }
530
531         /* Initialize the hardware */
532         if (em_hardware_init(hw)) {
533                 PMD_INIT_LOG(ERR, "Unable to initialize the hardware");
534                 return (-EIO);
535         }
536
537         E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN);
538
539         /* Configure for OS presence */
540         em_init_manageability(hw);
541
542         if (dev->data->dev_conf.intr_conf.rxq != 0) {
543                 intr_vector = dev->data->nb_rx_queues;
544                 if (rte_intr_efd_enable(intr_handle, intr_vector))
545                         return -1;
546         }
547
548         if (rte_intr_dp_is_en(intr_handle)) {
549                 intr_handle->intr_vec =
550                         rte_zmalloc("intr_vec",
551                                         dev->data->nb_rx_queues * sizeof(int), 0);
552                 if (intr_handle->intr_vec == NULL) {
553                         PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
554                                                 " intr_vec\n", dev->data->nb_rx_queues);
555                         return -ENOMEM;
556                 }
557
558                 /* enable rx interrupt */
559                 em_rxq_intr_enable(hw);
560         }
561
562         eth_em_tx_init(dev);
563
564         ret = eth_em_rx_init(dev);
565         if (ret) {
566                 PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
567                 em_dev_clear_queues(dev);
568                 return ret;
569         }
570
571         e1000_clear_hw_cntrs_base_generic(hw);
572
573         mask = ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK | \
574                         ETH_VLAN_EXTEND_MASK;
575         eth_em_vlan_offload_set(dev, mask);
576
577         /* Set Interrupt Throttling Rate to maximum allowed value. */
578         E1000_WRITE_REG(hw, E1000_ITR, UINT16_MAX);
579
580         /* Setup link speed and duplex */
581         switch (dev->data->dev_conf.link_speed) {
582         case ETH_LINK_SPEED_AUTONEG:
583                 if (dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX)
584                         hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
585                 else if (dev->data->dev_conf.link_duplex ==
586                                         ETH_LINK_HALF_DUPLEX)
587                         hw->phy.autoneg_advertised = E1000_ALL_HALF_DUPLEX;
588                 else if (dev->data->dev_conf.link_duplex ==
589                                         ETH_LINK_FULL_DUPLEX)
590                         hw->phy.autoneg_advertised = E1000_ALL_FULL_DUPLEX;
591                 else
592                         goto error_invalid_config;
593                 break;
594         case ETH_LINK_SPEED_10:
595                 if (dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX)
596                         hw->phy.autoneg_advertised = E1000_ALL_10_SPEED;
597                 else if (dev->data->dev_conf.link_duplex ==
598                                         ETH_LINK_HALF_DUPLEX)
599                         hw->phy.autoneg_advertised = ADVERTISE_10_HALF;
600                 else if (dev->data->dev_conf.link_duplex ==
601                                         ETH_LINK_FULL_DUPLEX)
602                         hw->phy.autoneg_advertised = ADVERTISE_10_FULL;
603                 else
604                         goto error_invalid_config;
605                 break;
606         case ETH_LINK_SPEED_100:
607                 if (dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX)
608                         hw->phy.autoneg_advertised = E1000_ALL_100_SPEED;
609                 else if (dev->data->dev_conf.link_duplex ==
610                                         ETH_LINK_HALF_DUPLEX)
611                         hw->phy.autoneg_advertised = ADVERTISE_100_HALF;
612                 else if (dev->data->dev_conf.link_duplex ==
613                                         ETH_LINK_FULL_DUPLEX)
614                         hw->phy.autoneg_advertised = ADVERTISE_100_FULL;
615                 else
616                         goto error_invalid_config;
617                 break;
618         case ETH_LINK_SPEED_1000:
619                 if ((dev->data->dev_conf.link_duplex ==
620                                 ETH_LINK_AUTONEG_DUPLEX) ||
621                         (dev->data->dev_conf.link_duplex ==
622                                         ETH_LINK_FULL_DUPLEX))
623                         hw->phy.autoneg_advertised = ADVERTISE_1000_FULL;
624                 else
625                         goto error_invalid_config;
626                 break;
627         case ETH_LINK_SPEED_10000:
628         default:
629                 goto error_invalid_config;
630         }
631         e1000_setup_link(hw);
632
633         if (rte_intr_allow_others(intr_handle)) {
634                 /* check if lsc interrupt is enabled */
635                 if (dev->data->dev_conf.intr_conf.lsc != 0)
636                         ret = eth_em_interrupt_setup(dev);
637                         if (ret) {
638                                 PMD_INIT_LOG(ERR, "Unable to setup interrupts");
639                                 em_dev_clear_queues(dev);
640                                 return ret;
641                         }
642         } else {
643                 rte_intr_callback_unregister(intr_handle,
644                                                 eth_em_interrupt_handler,
645                                                 (void *)dev);
646                 if (dev->data->dev_conf.intr_conf.lsc != 0)
647                         PMD_INIT_LOG(INFO, "lsc won't enable because of"
648                                      " no intr multiplex\n");
649         }
650         /* check if rxq interrupt is enabled */
651         if (dev->data->dev_conf.intr_conf.rxq != 0)
652                 eth_em_rxq_interrupt_setup(dev);
653
654         rte_intr_enable(intr_handle);
655
656         adapter->stopped = 0;
657
658         PMD_INIT_LOG(DEBUG, "<<");
659
660         return (0);
661
662 error_invalid_config:
663         PMD_INIT_LOG(ERR, "Invalid link_speed/link_duplex (%u/%u) for port %u",
664                      dev->data->dev_conf.link_speed,
665                      dev->data->dev_conf.link_duplex, dev->data->port_id);
666         em_dev_clear_queues(dev);
667         return (-EINVAL);
668 }
669
670 /*********************************************************************
671  *
672  *  This routine disables all traffic on the adapter by issuing a
673  *  global reset on the MAC.
674  *
675  **********************************************************************/
676 static void
677 eth_em_stop(struct rte_eth_dev *dev)
678 {
679         struct rte_eth_link link;
680         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
681         struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
682
683         em_rxq_intr_disable(hw);
684         em_lsc_intr_disable(hw);
685
686         e1000_reset_hw(hw);
687         if (hw->mac.type >= e1000_82544)
688                 E1000_WRITE_REG(hw, E1000_WUC, 0);
689
690         /* Power down the phy. Needed to make the link go down */
691         e1000_power_down_phy(hw);
692
693         em_dev_clear_queues(dev);
694
695         /* clear the recorded link status */
696         memset(&link, 0, sizeof(link));
697         rte_em_dev_atomic_write_link_status(dev, &link);
698
699         if (!rte_intr_allow_others(intr_handle))
700                 /* resume to the default handler */
701                 rte_intr_callback_register(intr_handle,
702                                            eth_em_interrupt_handler,
703                                            (void *)dev);
704
705         /* Clean datapath event and queue/vec mapping */
706         rte_intr_efd_disable(intr_handle);
707         if (intr_handle->intr_vec != NULL) {
708                 rte_free(intr_handle->intr_vec);
709                 intr_handle->intr_vec = NULL;
710         }
711 }
712
713 static void
714 eth_em_close(struct rte_eth_dev *dev)
715 {
716         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
717         struct e1000_adapter *adapter =
718                 E1000_DEV_PRIVATE(dev->data->dev_private);
719
720         eth_em_stop(dev);
721         adapter->stopped = 1;
722         em_dev_free_queues(dev);
723         e1000_phy_hw_reset(hw);
724         em_release_manageability(hw);
725         em_hw_control_release(hw);
726 }
727
728 static int
729 em_get_rx_buffer_size(struct e1000_hw *hw)
730 {
731         uint32_t rx_buf_size;
732
733         rx_buf_size = ((E1000_READ_REG(hw, E1000_PBA) & UINT16_MAX) << 10);
734         return rx_buf_size;
735 }
736
737 /*********************************************************************
738  *
739  *  Initialize the hardware
740  *
741  **********************************************************************/
742 static int
743 em_hardware_init(struct e1000_hw *hw)
744 {
745         uint32_t rx_buf_size;
746         int diag;
747
748         /* Issue a global reset */
749         e1000_reset_hw(hw);
750
751         /* Let the firmware know the OS is in control */
752         em_hw_control_acquire(hw);
753
754         /*
755          * These parameters control the automatic generation (Tx) and
756          * response (Rx) to Ethernet PAUSE frames.
757          * - High water mark should allow for at least two standard size (1518)
758          *   frames to be received after sending an XOFF.
759          * - Low water mark works best when it is very near the high water mark.
760          *   This allows the receiver to restart by sending XON when it has
761          *   drained a bit. Here we use an arbitrary value of 1500 which will
762          *   restart after one full frame is pulled from the buffer. There
763          *   could be several smaller frames in the buffer and if so they will
764          *   not trigger the XON until their total number reduces the buffer
765          *   by 1500.
766          * - The pause time is fairly large at 1000 x 512ns = 512 usec.
767          */
768         rx_buf_size = em_get_rx_buffer_size(hw);
769
770         hw->fc.high_water = rx_buf_size - PMD_ROUNDUP(ETHER_MAX_LEN * 2, 1024);
771         hw->fc.low_water = hw->fc.high_water - 1500;
772
773         if (hw->mac.type == e1000_80003es2lan)
774                 hw->fc.pause_time = UINT16_MAX;
775         else
776                 hw->fc.pause_time = EM_FC_PAUSE_TIME;
777
778         hw->fc.send_xon = 1;
779
780         /* Set Flow control, use the tunable location if sane */
781         if (em_fc_setting <= e1000_fc_full)
782                 hw->fc.requested_mode = em_fc_setting;
783         else
784                 hw->fc.requested_mode = e1000_fc_none;
785
786         /* Workaround: no TX flow ctrl for PCH */
787         if (hw->mac.type == e1000_pchlan)
788                 hw->fc.requested_mode = e1000_fc_rx_pause;
789
790         /* Override - settings for PCH2LAN, ya its magic :) */
791         if (hw->mac.type == e1000_pch2lan) {
792                 hw->fc.high_water = 0x5C20;
793                 hw->fc.low_water = 0x5048;
794                 hw->fc.pause_time = 0x0650;
795                 hw->fc.refresh_time = 0x0400;
796         }
797
798         diag = e1000_init_hw(hw);
799         if (diag < 0)
800                 return (diag);
801         e1000_check_for_link(hw);
802         return (0);
803 }
804
805 /* This function is based on em_update_stats_counters() in e1000/if_em.c */
806 static void
807 eth_em_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
808 {
809         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
810         struct e1000_hw_stats *stats =
811                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
812         int pause_frames;
813
814         if(hw->phy.media_type == e1000_media_type_copper ||
815                         (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
816                 stats->symerrs += E1000_READ_REG(hw,E1000_SYMERRS);
817                 stats->sec += E1000_READ_REG(hw, E1000_SEC);
818         }
819
820         stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
821         stats->mpc += E1000_READ_REG(hw, E1000_MPC);
822         stats->scc += E1000_READ_REG(hw, E1000_SCC);
823         stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
824
825         stats->mcc += E1000_READ_REG(hw, E1000_MCC);
826         stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
827         stats->colc += E1000_READ_REG(hw, E1000_COLC);
828         stats->dc += E1000_READ_REG(hw, E1000_DC);
829         stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
830         stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
831         stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
832
833         /*
834          * For watchdog management we need to know if we have been
835          * paused during the last interval, so capture that here.
836          */
837         pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
838         stats->xoffrxc += pause_frames;
839         stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
840         stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
841         stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
842         stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
843         stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
844         stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
845         stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
846         stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
847         stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
848         stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
849         stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
850         stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
851
852         /*
853          * For the 64-bit byte counters the low dword must be read first.
854          * Both registers clear on the read of the high dword.
855          */
856
857         stats->gorc += E1000_READ_REG(hw, E1000_GORCL);
858         stats->gorc += ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
859         stats->gotc += E1000_READ_REG(hw, E1000_GOTCL);
860         stats->gotc += ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
861
862         stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
863         stats->ruc += E1000_READ_REG(hw, E1000_RUC);
864         stats->rfc += E1000_READ_REG(hw, E1000_RFC);
865         stats->roc += E1000_READ_REG(hw, E1000_ROC);
866         stats->rjc += E1000_READ_REG(hw, E1000_RJC);
867
868         stats->tor += E1000_READ_REG(hw, E1000_TORH);
869         stats->tot += E1000_READ_REG(hw, E1000_TOTH);
870
871         stats->tpr += E1000_READ_REG(hw, E1000_TPR);
872         stats->tpt += E1000_READ_REG(hw, E1000_TPT);
873         stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
874         stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
875         stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
876         stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
877         stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
878         stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
879         stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
880         stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
881
882         /* Interrupt Counts */
883
884         if (hw->mac.type >= e1000_82571) {
885                 stats->iac += E1000_READ_REG(hw, E1000_IAC);
886                 stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
887                 stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
888                 stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
889                 stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
890                 stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
891                 stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
892                 stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
893                 stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
894         }
895
896         if (hw->mac.type >= e1000_82543) {
897                 stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
898                 stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
899                 stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
900                 stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
901                 stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
902                 stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
903         }
904
905         if (rte_stats == NULL)
906                 return;
907
908         /* Rx Errors */
909         rte_stats->ibadcrc = stats->crcerrs;
910         rte_stats->ibadlen = stats->rlec + stats->ruc + stats->roc;
911         rte_stats->imissed = stats->mpc;
912         rte_stats->ierrors = rte_stats->ibadcrc +
913                              rte_stats->ibadlen +
914                              rte_stats->imissed +
915                              stats->rxerrc + stats->algnerrc + stats->cexterr;
916
917         /* Tx Errors */
918         rte_stats->oerrors = stats->ecol + stats->latecol;
919
920         rte_stats->ipackets = stats->gprc;
921         rte_stats->opackets = stats->gptc;
922         rte_stats->ibytes   = stats->gorc;
923         rte_stats->obytes   = stats->gotc;
924
925         /* XON/XOFF pause frames stats registers */
926         rte_stats->tx_pause_xon  = stats->xontxc;
927         rte_stats->rx_pause_xon  = stats->xonrxc;
928         rte_stats->tx_pause_xoff = stats->xofftxc;
929         rte_stats->rx_pause_xoff = stats->xoffrxc;
930 }
931
932 static void
933 eth_em_stats_reset(struct rte_eth_dev *dev)
934 {
935         struct e1000_hw_stats *hw_stats =
936                         E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
937
938         /* HW registers are cleared on read */
939         eth_em_stats_get(dev, NULL);
940
941         /* Reset software totals */
942         memset(hw_stats, 0, sizeof(*hw_stats));
943 }
944
945 static int
946 eth_em_rx_queue_intr_enable(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id)
947 {
948         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
949
950         em_rxq_intr_enable(hw);
951         rte_intr_enable(&dev->pci_dev->intr_handle);
952
953         return 0;
954 }
955
956 static int
957 eth_em_rx_queue_intr_disable(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id)
958 {
959         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
960
961         em_rxq_intr_disable(hw);
962
963         return 0;
964 }
965
966 static uint32_t
967 em_get_max_pktlen(const struct e1000_hw *hw)
968 {
969         switch (hw->mac.type) {
970         case e1000_82571:
971         case e1000_82572:
972         case e1000_ich9lan:
973         case e1000_ich10lan:
974         case e1000_pch2lan:
975         case e1000_82574:
976         case e1000_80003es2lan: /* 9K Jumbo Frame size */
977         case e1000_82583:
978                 return (0x2412);
979         case e1000_pchlan:
980                 return (0x1000);
981         /* Adapters that do not support jumbo frames */
982         case e1000_ich8lan:
983                 return (ETHER_MAX_LEN);
984         default:
985                 return (MAX_JUMBO_FRAME_SIZE);
986         }
987 }
988
989 static void
990 eth_em_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
991 {
992         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
993
994         dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
995         dev_info->max_rx_pktlen = em_get_max_pktlen(hw);
996         dev_info->max_mac_addrs = hw->mac.rar_entry_count;
997
998         /*
999          * Starting with 631xESB hw supports 2 TX/RX queues per port.
1000          * Unfortunatelly, all these nics have just one TX context.
1001          * So we have few choises for TX:
1002          * - Use just one TX queue.
1003          * - Allow cksum offload only for one TX queue.
1004          * - Don't allow TX cksum offload at all.
1005          * For now, option #1 was chosen.
1006          * To use second RX queue we have to use extended RX descriptor
1007          * (Multiple Receive Queues are mutually exclusive with UDP
1008          * fragmentation and are not supported when a legacy receive
1009          * descriptor format is used).
1010          * Which means separate RX routinies - as legacy nics (82540, 82545)
1011          * don't support extended RXD.
1012          * To avoid it we support just one RX queue for now (no RSS).
1013          */
1014
1015         dev_info->max_rx_queues = 1;
1016         dev_info->max_tx_queues = 1;
1017 }
1018
1019 /* return 0 means link status changed, -1 means not changed */
1020 static int
1021 eth_em_link_update(struct rte_eth_dev *dev, int wait_to_complete)
1022 {
1023         struct e1000_hw *hw =
1024                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1025         struct rte_eth_link link, old;
1026         int link_check, count;
1027
1028         link_check = 0;
1029         hw->mac.get_link_status = 1;
1030
1031         /* possible wait-to-complete in up to 9 seconds */
1032         for (count = 0; count < EM_LINK_UPDATE_CHECK_TIMEOUT; count ++) {
1033                 /* Read the real link status */
1034                 switch (hw->phy.media_type) {
1035                 case e1000_media_type_copper:
1036                         /* Do the work to read phy */
1037                         e1000_check_for_link(hw);
1038                         link_check = !hw->mac.get_link_status;
1039                         break;
1040
1041                 case e1000_media_type_fiber:
1042                         e1000_check_for_link(hw);
1043                         link_check = (E1000_READ_REG(hw, E1000_STATUS) &
1044                                         E1000_STATUS_LU);
1045                         break;
1046
1047                 case e1000_media_type_internal_serdes:
1048                         e1000_check_for_link(hw);
1049                         link_check = hw->mac.serdes_has_link;
1050                         break;
1051
1052                 default:
1053                         break;
1054                 }
1055                 if (link_check || wait_to_complete == 0)
1056                         break;
1057                 rte_delay_ms(EM_LINK_UPDATE_CHECK_INTERVAL);
1058         }
1059         memset(&link, 0, sizeof(link));
1060         rte_em_dev_atomic_read_link_status(dev, &link);
1061         old = link;
1062
1063         /* Now we check if a transition has happened */
1064         if (link_check && (link.link_status == 0)) {
1065                 hw->mac.ops.get_link_up_info(hw, &link.link_speed,
1066                         &link.link_duplex);
1067                 link.link_status = 1;
1068         } else if (!link_check && (link.link_status == 1)) {
1069                 link.link_speed = 0;
1070                 link.link_duplex = 0;
1071                 link.link_status = 0;
1072         }
1073         rte_em_dev_atomic_write_link_status(dev, &link);
1074
1075         /* not changed */
1076         if (old.link_status == link.link_status)
1077                 return -1;
1078
1079         /* changed */
1080         return 0;
1081 }
1082
1083 /*
1084  * em_hw_control_acquire sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
1085  * For ASF and Pass Through versions of f/w this means
1086  * that the driver is loaded. For AMT version type f/w
1087  * this means that the network i/f is open.
1088  */
1089 static void
1090 em_hw_control_acquire(struct e1000_hw *hw)
1091 {
1092         uint32_t ctrl_ext, swsm;
1093
1094         /* Let firmware know the driver has taken over */
1095         if (hw->mac.type == e1000_82573) {
1096                 swsm = E1000_READ_REG(hw, E1000_SWSM);
1097                 E1000_WRITE_REG(hw, E1000_SWSM, swsm | E1000_SWSM_DRV_LOAD);
1098
1099         } else {
1100                 ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
1101                 E1000_WRITE_REG(hw, E1000_CTRL_EXT,
1102                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1103         }
1104 }
1105
1106 /*
1107  * em_hw_control_release resets {CTRL_EXTT|FWSM}:DRV_LOAD bit.
1108  * For ASF and Pass Through versions of f/w this means that the
1109  * driver is no longer loaded. For AMT versions of the
1110  * f/w this means that the network i/f is closed.
1111  */
1112 static void
1113 em_hw_control_release(struct e1000_hw *hw)
1114 {
1115         uint32_t ctrl_ext, swsm;
1116
1117         /* Let firmware taken over control of h/w */
1118         if (hw->mac.type == e1000_82573) {
1119                 swsm = E1000_READ_REG(hw, E1000_SWSM);
1120                 E1000_WRITE_REG(hw, E1000_SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
1121         } else {
1122                 ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
1123                 E1000_WRITE_REG(hw, E1000_CTRL_EXT,
1124                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1125         }
1126 }
1127
1128 /*
1129  * Bit of a misnomer, what this really means is
1130  * to enable OS management of the system... aka
1131  * to disable special hardware management features.
1132  */
1133 static void
1134 em_init_manageability(struct e1000_hw *hw)
1135 {
1136         if (e1000_enable_mng_pass_thru(hw)) {
1137                 uint32_t manc2h = E1000_READ_REG(hw, E1000_MANC2H);
1138                 uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
1139
1140                 /* disable hardware interception of ARP */
1141                 manc &= ~(E1000_MANC_ARP_EN);
1142
1143                 /* enable receiving management packets to the host */
1144                 manc |= E1000_MANC_EN_MNG2HOST;
1145                 manc2h |= 1 << 5;  /* Mng Port 623 */
1146                 manc2h |= 1 << 6;  /* Mng Port 664 */
1147                 E1000_WRITE_REG(hw, E1000_MANC2H, manc2h);
1148                 E1000_WRITE_REG(hw, E1000_MANC, manc);
1149         }
1150 }
1151
1152 /*
1153  * Give control back to hardware management
1154  * controller if there is one.
1155  */
1156 static void
1157 em_release_manageability(struct e1000_hw *hw)
1158 {
1159         uint32_t manc;
1160
1161         if (e1000_enable_mng_pass_thru(hw)) {
1162                 manc = E1000_READ_REG(hw, E1000_MANC);
1163
1164                 /* re-enable hardware interception of ARP */
1165                 manc |= E1000_MANC_ARP_EN;
1166                 manc &= ~E1000_MANC_EN_MNG2HOST;
1167
1168                 E1000_WRITE_REG(hw, E1000_MANC, manc);
1169         }
1170 }
1171
1172 static void
1173 eth_em_promiscuous_enable(struct rte_eth_dev *dev)
1174 {
1175         struct e1000_hw *hw =
1176                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1177         uint32_t rctl;
1178
1179         rctl = E1000_READ_REG(hw, E1000_RCTL);
1180         rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1181         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1182 }
1183
1184 static void
1185 eth_em_promiscuous_disable(struct rte_eth_dev *dev)
1186 {
1187         struct e1000_hw *hw =
1188                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1189         uint32_t rctl;
1190
1191         rctl = E1000_READ_REG(hw, E1000_RCTL);
1192         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_SBP);
1193         if (dev->data->all_multicast == 1)
1194                 rctl |= E1000_RCTL_MPE;
1195         else
1196                 rctl &= (~E1000_RCTL_MPE);
1197         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1198 }
1199
1200 static void
1201 eth_em_allmulticast_enable(struct rte_eth_dev *dev)
1202 {
1203         struct e1000_hw *hw =
1204                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1205         uint32_t rctl;
1206
1207         rctl = E1000_READ_REG(hw, E1000_RCTL);
1208         rctl |= E1000_RCTL_MPE;
1209         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1210 }
1211
1212 static void
1213 eth_em_allmulticast_disable(struct rte_eth_dev *dev)
1214 {
1215         struct e1000_hw *hw =
1216                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1217         uint32_t rctl;
1218
1219         if (dev->data->promiscuous == 1)
1220                 return; /* must remain in all_multicast mode */
1221         rctl = E1000_READ_REG(hw, E1000_RCTL);
1222         rctl &= (~E1000_RCTL_MPE);
1223         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1224 }
1225
1226 static int
1227 eth_em_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1228 {
1229         struct e1000_hw *hw =
1230                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1231         struct e1000_vfta * shadow_vfta =
1232                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
1233         uint32_t vfta;
1234         uint32_t vid_idx;
1235         uint32_t vid_bit;
1236
1237         vid_idx = (uint32_t) ((vlan_id >> E1000_VFTA_ENTRY_SHIFT) &
1238                               E1000_VFTA_ENTRY_MASK);
1239         vid_bit = (uint32_t) (1 << (vlan_id & E1000_VFTA_ENTRY_BIT_SHIFT_MASK));
1240         vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, vid_idx);
1241         if (on)
1242                 vfta |= vid_bit;
1243         else
1244                 vfta &= ~vid_bit;
1245         E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, vid_idx, vfta);
1246
1247         /* update local VFTA copy */
1248         shadow_vfta->vfta[vid_idx] = vfta;
1249
1250         return 0;
1251 }
1252
1253 static void
1254 em_vlan_hw_filter_disable(struct rte_eth_dev *dev)
1255 {
1256         struct e1000_hw *hw =
1257                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1258         uint32_t reg;
1259
1260         /* Filter Table Disable */
1261         reg = E1000_READ_REG(hw, E1000_RCTL);
1262         reg &= ~E1000_RCTL_CFIEN;
1263         reg &= ~E1000_RCTL_VFE;
1264         E1000_WRITE_REG(hw, E1000_RCTL, reg);
1265 }
1266
1267 static void
1268 em_vlan_hw_filter_enable(struct rte_eth_dev *dev)
1269 {
1270         struct e1000_hw *hw =
1271                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1272         struct e1000_vfta * shadow_vfta =
1273                 E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
1274         uint32_t reg;
1275         int i;
1276
1277         /* Filter Table Enable, CFI not used for packet acceptance */
1278         reg = E1000_READ_REG(hw, E1000_RCTL);
1279         reg &= ~E1000_RCTL_CFIEN;
1280         reg |= E1000_RCTL_VFE;
1281         E1000_WRITE_REG(hw, E1000_RCTL, reg);
1282
1283         /* restore vfta from local copy */
1284         for (i = 0; i < IGB_VFTA_SIZE; i++)
1285                 E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, shadow_vfta->vfta[i]);
1286 }
1287
1288 static void
1289 em_vlan_hw_strip_disable(struct rte_eth_dev *dev)
1290 {
1291         struct e1000_hw *hw =
1292                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1293         uint32_t reg;
1294
1295         /* VLAN Mode Disable */
1296         reg = E1000_READ_REG(hw, E1000_CTRL);
1297         reg &= ~E1000_CTRL_VME;
1298         E1000_WRITE_REG(hw, E1000_CTRL, reg);
1299
1300 }
1301
1302 static void
1303 em_vlan_hw_strip_enable(struct rte_eth_dev *dev)
1304 {
1305         struct e1000_hw *hw =
1306                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1307         uint32_t reg;
1308
1309         /* VLAN Mode Enable */
1310         reg = E1000_READ_REG(hw, E1000_CTRL);
1311         reg |= E1000_CTRL_VME;
1312         E1000_WRITE_REG(hw, E1000_CTRL, reg);
1313 }
1314
1315 static void
1316 eth_em_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1317 {
1318         if(mask & ETH_VLAN_STRIP_MASK){
1319                 if (dev->data->dev_conf.rxmode.hw_vlan_strip)
1320                         em_vlan_hw_strip_enable(dev);
1321                 else
1322                         em_vlan_hw_strip_disable(dev);
1323         }
1324
1325         if(mask & ETH_VLAN_FILTER_MASK){
1326                 if (dev->data->dev_conf.rxmode.hw_vlan_filter)
1327                         em_vlan_hw_filter_enable(dev);
1328                 else
1329                         em_vlan_hw_filter_disable(dev);
1330         }
1331 }
1332
1333 /*
1334  * It enables the interrupt mask and then enable the interrupt.
1335  *
1336  * @param dev
1337  *  Pointer to struct rte_eth_dev.
1338  *
1339  * @return
1340  *  - On success, zero.
1341  *  - On failure, a negative value.
1342  */
1343 static int
1344 eth_em_interrupt_setup(struct rte_eth_dev *dev)
1345 {
1346         uint32_t regval;
1347         struct e1000_hw *hw =
1348                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1349
1350         /* clear interrupt */
1351         E1000_READ_REG(hw, E1000_ICR);
1352         regval = E1000_READ_REG(hw, E1000_IMS);
1353         E1000_WRITE_REG(hw, E1000_IMS, regval | E1000_ICR_LSC);
1354         return (0);
1355 }
1356
1357 /*
1358  * It clears the interrupt causes and enables the interrupt.
1359  * It will be called once only during nic initialized.
1360  *
1361  * @param dev
1362  *  Pointer to struct rte_eth_dev.
1363  *
1364  * @return
1365  *  - On success, zero.
1366  *  - On failure, a negative value.
1367  */
1368 static int
1369 eth_em_rxq_interrupt_setup(struct rte_eth_dev *dev)
1370 {
1371         struct e1000_hw *hw =
1372         E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1373
1374         E1000_READ_REG(hw, E1000_ICR);
1375         em_rxq_intr_enable(hw);
1376         return 0;
1377 }
1378
1379 /*
1380  * It enable receive packet interrupt.
1381  * @param hw
1382  * Pointer to struct e1000_hw
1383  *
1384  * @return
1385  */
1386 static void
1387 em_rxq_intr_enable(struct e1000_hw *hw)
1388 {
1389         E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_RXT0);
1390         E1000_WRITE_FLUSH(hw);
1391 }
1392
1393 /*
1394  * It disabled lsc interrupt.
1395  * @param hw
1396  * Pointer to struct e1000_hw
1397  *
1398  * @return
1399  */
1400 static void
1401 em_lsc_intr_disable(struct e1000_hw *hw)
1402 {
1403         E1000_WRITE_REG(hw, E1000_IMC, E1000_IMS_LSC);
1404         E1000_WRITE_FLUSH(hw);
1405 }
1406
1407 /*
1408  * It disabled receive packet interrupt.
1409  * @param hw
1410  * Pointer to struct e1000_hw
1411  *
1412  * @return
1413  */
1414 static void
1415 em_rxq_intr_disable(struct e1000_hw *hw)
1416 {
1417         E1000_READ_REG(hw, E1000_ICR);
1418         E1000_WRITE_REG(hw, E1000_IMC, E1000_IMS_RXT0);
1419         E1000_WRITE_FLUSH(hw);
1420 }
1421
1422 /*
1423  * It reads ICR and gets interrupt causes, check it and set a bit flag
1424  * to update link status.
1425  *
1426  * @param dev
1427  *  Pointer to struct rte_eth_dev.
1428  *
1429  * @return
1430  *  - On success, zero.
1431  *  - On failure, a negative value.
1432  */
1433 static int
1434 eth_em_interrupt_get_status(struct rte_eth_dev *dev)
1435 {
1436         uint32_t icr;
1437         struct e1000_hw *hw =
1438                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1439         struct e1000_interrupt *intr =
1440                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
1441
1442         /* read-on-clear nic registers here */
1443         icr = E1000_READ_REG(hw, E1000_ICR);
1444         if (icr & E1000_ICR_LSC) {
1445                 intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
1446         }
1447
1448         return 0;
1449 }
1450
1451 /*
1452  * It executes link_update after knowing an interrupt is prsent.
1453  *
1454  * @param dev
1455  *  Pointer to struct rte_eth_dev.
1456  *
1457  * @return
1458  *  - On success, zero.
1459  *  - On failure, a negative value.
1460  */
1461 static int
1462 eth_em_interrupt_action(struct rte_eth_dev *dev)
1463 {
1464         struct e1000_hw *hw =
1465                 E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1466         struct e1000_interrupt *intr =
1467                 E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
1468         uint32_t tctl, rctl;
1469         struct rte_eth_link link;
1470         int ret;
1471
1472         if (!(intr->flags & E1000_FLAG_NEED_LINK_UPDATE))
1473                 return -1;
1474
1475         intr->flags &= ~E1000_FLAG_NEED_LINK_UPDATE;
1476         rte_intr_enable(&(dev->pci_dev->intr_handle));
1477
1478         /* set get_link_status to check register later */
1479         hw->mac.get_link_status = 1;
1480         ret = eth_em_link_update(dev, 0);
1481
1482         /* check if link has changed */
1483         if (ret < 0)
1484                 return 0;
1485
1486         memset(&link, 0, sizeof(link));
1487         rte_em_dev_atomic_read_link_status(dev, &link);
1488         if (link.link_status) {
1489                 PMD_INIT_LOG(INFO, " Port %d: Link Up - speed %u Mbps - %s",
1490                              dev->data->port_id, (unsigned)link.link_speed,
1491                              link.link_duplex == ETH_LINK_FULL_DUPLEX ?
1492                              "full-duplex" : "half-duplex");
1493         } else {
1494                 PMD_INIT_LOG(INFO, " Port %d: Link Down", dev->data->port_id);
1495         }
1496         PMD_INIT_LOG(DEBUG, "PCI Address: %04d:%02d:%02d:%d",
1497                      dev->pci_dev->addr.domain, dev->pci_dev->addr.bus,
1498                      dev->pci_dev->addr.devid, dev->pci_dev->addr.function);
1499
1500         tctl = E1000_READ_REG(hw, E1000_TCTL);
1501         rctl = E1000_READ_REG(hw, E1000_RCTL);
1502         if (link.link_status) {
1503                 /* enable Tx/Rx */
1504                 tctl |= E1000_TCTL_EN;
1505                 rctl |= E1000_RCTL_EN;
1506         } else {
1507                 /* disable Tx/Rx */
1508                 tctl &= ~E1000_TCTL_EN;
1509                 rctl &= ~E1000_RCTL_EN;
1510         }
1511         E1000_WRITE_REG(hw, E1000_TCTL, tctl);
1512         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1513         E1000_WRITE_FLUSH(hw);
1514
1515         return 0;
1516 }
1517
1518 /**
1519  * Interrupt handler which shall be registered at first.
1520  *
1521  * @param handle
1522  *  Pointer to interrupt handle.
1523  * @param param
1524  *  The address of parameter (struct rte_eth_dev *) regsitered before.
1525  *
1526  * @return
1527  *  void
1528  */
1529 static void
1530 eth_em_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1531                                                         void *param)
1532 {
1533         struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
1534
1535         eth_em_interrupt_get_status(dev);
1536         eth_em_interrupt_action(dev);
1537         _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC);
1538 }
1539
1540 static int
1541 eth_em_led_on(struct rte_eth_dev *dev)
1542 {
1543         struct e1000_hw *hw;
1544
1545         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1546         return (e1000_led_on(hw) == E1000_SUCCESS ? 0 : -ENOTSUP);
1547 }
1548
1549 static int
1550 eth_em_led_off(struct rte_eth_dev *dev)
1551 {
1552         struct e1000_hw *hw;
1553
1554         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1555         return (e1000_led_off(hw) == E1000_SUCCESS ? 0 : -ENOTSUP);
1556 }
1557
1558 static int
1559 eth_em_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1560 {
1561         struct e1000_hw *hw;
1562         uint32_t ctrl;
1563         int tx_pause;
1564         int rx_pause;
1565
1566         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1567         fc_conf->pause_time = hw->fc.pause_time;
1568         fc_conf->high_water = hw->fc.high_water;
1569         fc_conf->low_water = hw->fc.low_water;
1570         fc_conf->send_xon = hw->fc.send_xon;
1571         fc_conf->autoneg = hw->mac.autoneg;
1572
1573         /*
1574          * Return rx_pause and tx_pause status according to actual setting of
1575          * the TFCE and RFCE bits in the CTRL register.
1576          */
1577         ctrl = E1000_READ_REG(hw, E1000_CTRL);
1578         if (ctrl & E1000_CTRL_TFCE)
1579                 tx_pause = 1;
1580         else
1581                 tx_pause = 0;
1582
1583         if (ctrl & E1000_CTRL_RFCE)
1584                 rx_pause = 1;
1585         else
1586                 rx_pause = 0;
1587
1588         if (rx_pause && tx_pause)
1589                 fc_conf->mode = RTE_FC_FULL;
1590         else if (rx_pause)
1591                 fc_conf->mode = RTE_FC_RX_PAUSE;
1592         else if (tx_pause)
1593                 fc_conf->mode = RTE_FC_TX_PAUSE;
1594         else
1595                 fc_conf->mode = RTE_FC_NONE;
1596
1597         return 0;
1598 }
1599
1600 static int
1601 eth_em_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1602 {
1603         struct e1000_hw *hw;
1604         int err;
1605         enum e1000_fc_mode rte_fcmode_2_e1000_fcmode[] = {
1606                 e1000_fc_none,
1607                 e1000_fc_rx_pause,
1608                 e1000_fc_tx_pause,
1609                 e1000_fc_full
1610         };
1611         uint32_t rx_buf_size;
1612         uint32_t max_high_water;
1613         uint32_t rctl;
1614
1615         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1616         if (fc_conf->autoneg != hw->mac.autoneg)
1617                 return -ENOTSUP;
1618         rx_buf_size = em_get_rx_buffer_size(hw);
1619         PMD_INIT_LOG(DEBUG, "Rx packet buffer size = 0x%x", rx_buf_size);
1620
1621         /* At least reserve one Ethernet frame for watermark */
1622         max_high_water = rx_buf_size - ETHER_MAX_LEN;
1623         if ((fc_conf->high_water > max_high_water) ||
1624             (fc_conf->high_water < fc_conf->low_water)) {
1625                 PMD_INIT_LOG(ERR, "e1000 incorrect high/low water value");
1626                 PMD_INIT_LOG(ERR, "high water must <= 0x%x", max_high_water);
1627                 return (-EINVAL);
1628         }
1629
1630         hw->fc.requested_mode = rte_fcmode_2_e1000_fcmode[fc_conf->mode];
1631         hw->fc.pause_time     = fc_conf->pause_time;
1632         hw->fc.high_water     = fc_conf->high_water;
1633         hw->fc.low_water      = fc_conf->low_water;
1634         hw->fc.send_xon       = fc_conf->send_xon;
1635
1636         err = e1000_setup_link_generic(hw);
1637         if (err == E1000_SUCCESS) {
1638
1639                 /* check if we want to forward MAC frames - driver doesn't have native
1640                  * capability to do that, so we'll write the registers ourselves */
1641
1642                 rctl = E1000_READ_REG(hw, E1000_RCTL);
1643
1644                 /* set or clear MFLCN.PMCF bit depending on configuration */
1645                 if (fc_conf->mac_ctrl_frame_fwd != 0)
1646                         rctl |= E1000_RCTL_PMCF;
1647                 else
1648                         rctl &= ~E1000_RCTL_PMCF;
1649
1650                 E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1651                 E1000_WRITE_FLUSH(hw);
1652
1653                 return 0;
1654         }
1655
1656         PMD_INIT_LOG(ERR, "e1000_setup_link_generic = 0x%x", err);
1657         return (-EIO);
1658 }
1659
1660 static void
1661 eth_em_rar_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1662                 uint32_t index, __rte_unused uint32_t pool)
1663 {
1664         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1665
1666         e1000_rar_set(hw, mac_addr->addr_bytes, index);
1667 }
1668
1669 static void
1670 eth_em_rar_clear(struct rte_eth_dev *dev, uint32_t index)
1671 {
1672         uint8_t addr[ETHER_ADDR_LEN];
1673         struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1674
1675         memset(addr, 0, sizeof(addr));
1676
1677         e1000_rar_set(hw, addr, index);
1678 }
1679
1680 static int
1681 eth_em_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1682 {
1683         struct rte_eth_dev_info dev_info;
1684         struct e1000_hw *hw;
1685         uint32_t frame_size;
1686         uint32_t rctl;
1687
1688         eth_em_infos_get(dev, &dev_info);
1689         frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + VLAN_TAG_SIZE;
1690
1691         /* check that mtu is within the allowed range */
1692         if ((mtu < ETHER_MIN_MTU) || (frame_size > dev_info.max_rx_pktlen))
1693                 return -EINVAL;
1694
1695         /* refuse mtu that requires the support of scattered packets when this
1696          * feature has not been enabled before. */
1697         if (!dev->data->scattered_rx &&
1698             frame_size > dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM)
1699                 return -EINVAL;
1700
1701         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1702         rctl = E1000_READ_REG(hw, E1000_RCTL);
1703
1704         /* switch to jumbo mode if needed */
1705         if (frame_size > ETHER_MAX_LEN) {
1706                 dev->data->dev_conf.rxmode.jumbo_frame = 1;
1707                 rctl |= E1000_RCTL_LPE;
1708         } else {
1709                 dev->data->dev_conf.rxmode.jumbo_frame = 0;
1710                 rctl &= ~E1000_RCTL_LPE;
1711         }
1712         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
1713
1714         /* update max frame size */
1715         dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1716         return 0;
1717 }
1718
1719 static int
1720 eth_em_set_mc_addr_list(struct rte_eth_dev *dev,
1721                         struct ether_addr *mc_addr_set,
1722                         uint32_t nb_mc_addr)
1723 {
1724         struct e1000_hw *hw;
1725
1726         hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1727         e1000_update_mc_addr_list(hw, (u8 *)mc_addr_set, nb_mc_addr);
1728         return 0;
1729 }
1730
1731 struct rte_driver em_pmd_drv = {
1732         .type = PMD_PDEV,
1733         .init = rte_em_pmd_init,
1734 };
1735
1736 PMD_REGISTER_DRIVER(em_pmd_drv);