Merge branch 'linus' into tracing/hw-branch-tracing
[linux-2.6] / drivers / net / sfc / falcon.c
1 /****************************************************************************
2  * Driver for Solarflare Solarstorm network controllers and boards
3  * Copyright 2005-2006 Fen Systems Ltd.
4  * Copyright 2006-2008 Solarflare Communications Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License version 2 as published
8  * by the Free Software Foundation, incorporated herein by reference.
9  */
10
11 #include <linux/bitops.h>
12 #include <linux/delay.h>
13 #include <linux/pci.h>
14 #include <linux/module.h>
15 #include <linux/seq_file.h>
16 #include <linux/i2c.h>
17 #include <linux/i2c-algo-bit.h>
18 #include <linux/mii.h>
19 #include "net_driver.h"
20 #include "bitfield.h"
21 #include "efx.h"
22 #include "mac.h"
23 #include "spi.h"
24 #include "falcon.h"
25 #include "falcon_hwdefs.h"
26 #include "falcon_io.h"
27 #include "mdio_10g.h"
28 #include "phy.h"
29 #include "boards.h"
30 #include "workarounds.h"
31
32 /* Falcon hardware control.
33  * Falcon is the internal codename for the SFC4000 controller that is
34  * present in SFE400X evaluation boards
35  */
36
37 /**
38  * struct falcon_nic_data - Falcon NIC state
39  * @next_buffer_table: First available buffer table id
40  * @pci_dev2: The secondary PCI device if present
41  * @i2c_data: Operations and state for I2C bit-bashing algorithm
42  * @int_error_count: Number of internal errors seen recently
43  * @int_error_expire: Time at which error count will be expired
44  */
45 struct falcon_nic_data {
46         unsigned next_buffer_table;
47         struct pci_dev *pci_dev2;
48         struct i2c_algo_bit_data i2c_data;
49
50         unsigned int_error_count;
51         unsigned long int_error_expire;
52 };
53
54 /**************************************************************************
55  *
56  * Configurable values
57  *
58  **************************************************************************
59  */
60
61 static int disable_dma_stats;
62
63 /* This is set to 16 for a good reason.  In summary, if larger than
64  * 16, the descriptor cache holds more than a default socket
65  * buffer's worth of packets (for UDP we can only have at most one
66  * socket buffer's worth outstanding).  This combined with the fact
67  * that we only get 1 TX event per descriptor cache means the NIC
68  * goes idle.
69  */
70 #define TX_DC_ENTRIES 16
71 #define TX_DC_ENTRIES_ORDER 0
72 #define TX_DC_BASE 0x130000
73
74 #define RX_DC_ENTRIES 64
75 #define RX_DC_ENTRIES_ORDER 2
76 #define RX_DC_BASE 0x100000
77
78 static const unsigned int
79 /* "Large" EEPROM device: Atmel AT25640 or similar
80  * 8 KB, 16-bit address, 32 B write block */
81 large_eeprom_type = ((13 << SPI_DEV_TYPE_SIZE_LBN)
82                      | (2 << SPI_DEV_TYPE_ADDR_LEN_LBN)
83                      | (5 << SPI_DEV_TYPE_BLOCK_SIZE_LBN)),
84 /* Default flash device: Atmel AT25F1024
85  * 128 KB, 24-bit address, 32 KB erase block, 256 B write block */
86 default_flash_type = ((17 << SPI_DEV_TYPE_SIZE_LBN)
87                       | (3 << SPI_DEV_TYPE_ADDR_LEN_LBN)
88                       | (0x52 << SPI_DEV_TYPE_ERASE_CMD_LBN)
89                       | (15 << SPI_DEV_TYPE_ERASE_SIZE_LBN)
90                       | (8 << SPI_DEV_TYPE_BLOCK_SIZE_LBN));
91
92 /* RX FIFO XOFF watermark
93  *
94  * When the amount of the RX FIFO increases used increases past this
95  * watermark send XOFF. Only used if RX flow control is enabled (ethtool -A)
96  * This also has an effect on RX/TX arbitration
97  */
98 static int rx_xoff_thresh_bytes = -1;
99 module_param(rx_xoff_thresh_bytes, int, 0644);
100 MODULE_PARM_DESC(rx_xoff_thresh_bytes, "RX fifo XOFF threshold");
101
102 /* RX FIFO XON watermark
103  *
104  * When the amount of the RX FIFO used decreases below this
105  * watermark send XON. Only used if TX flow control is enabled (ethtool -A)
106  * This also has an effect on RX/TX arbitration
107  */
108 static int rx_xon_thresh_bytes = -1;
109 module_param(rx_xon_thresh_bytes, int, 0644);
110 MODULE_PARM_DESC(rx_xon_thresh_bytes, "RX fifo XON threshold");
111
112 /* TX descriptor ring size - min 512 max 4k */
113 #define FALCON_TXD_RING_ORDER TX_DESCQ_SIZE_1K
114 #define FALCON_TXD_RING_SIZE 1024
115 #define FALCON_TXD_RING_MASK (FALCON_TXD_RING_SIZE - 1)
116
117 /* RX descriptor ring size - min 512 max 4k */
118 #define FALCON_RXD_RING_ORDER RX_DESCQ_SIZE_1K
119 #define FALCON_RXD_RING_SIZE 1024
120 #define FALCON_RXD_RING_MASK (FALCON_RXD_RING_SIZE - 1)
121
122 /* Event queue size - max 32k */
123 #define FALCON_EVQ_ORDER EVQ_SIZE_4K
124 #define FALCON_EVQ_SIZE 4096
125 #define FALCON_EVQ_MASK (FALCON_EVQ_SIZE - 1)
126
127 /* If FALCON_MAX_INT_ERRORS internal errors occur within
128  * FALCON_INT_ERROR_EXPIRE seconds, we consider the NIC broken and
129  * disable it.
130  */
131 #define FALCON_INT_ERROR_EXPIRE 3600
132 #define FALCON_MAX_INT_ERRORS 5
133
134 /* We poll for events every FLUSH_INTERVAL ms, and check FLUSH_POLL_COUNT times
135  */
136 #define FALCON_FLUSH_INTERVAL 10
137 #define FALCON_FLUSH_POLL_COUNT 100
138
139 /**************************************************************************
140  *
141  * Falcon constants
142  *
143  **************************************************************************
144  */
145
146 /* DMA address mask */
147 #define FALCON_DMA_MASK DMA_BIT_MASK(46)
148
149 /* TX DMA length mask (13-bit) */
150 #define FALCON_TX_DMA_MASK (4096 - 1)
151
152 /* Size and alignment of special buffers (4KB) */
153 #define FALCON_BUF_SIZE 4096
154
155 /* Dummy SRAM size code */
156 #define SRM_NB_BSZ_ONCHIP_ONLY (-1)
157
158 #define FALCON_IS_DUAL_FUNC(efx)                \
159         (falcon_rev(efx) < FALCON_REV_B0)
160
161 /**************************************************************************
162  *
163  * Falcon hardware access
164  *
165  **************************************************************************/
166
167 /* Read the current event from the event queue */
168 static inline efx_qword_t *falcon_event(struct efx_channel *channel,
169                                         unsigned int index)
170 {
171         return (((efx_qword_t *) (channel->eventq.addr)) + index);
172 }
173
174 /* See if an event is present
175  *
176  * We check both the high and low dword of the event for all ones.  We
177  * wrote all ones when we cleared the event, and no valid event can
178  * have all ones in either its high or low dwords.  This approach is
179  * robust against reordering.
180  *
181  * Note that using a single 64-bit comparison is incorrect; even
182  * though the CPU read will be atomic, the DMA write may not be.
183  */
184 static inline int falcon_event_present(efx_qword_t *event)
185 {
186         return (!(EFX_DWORD_IS_ALL_ONES(event->dword[0]) |
187                   EFX_DWORD_IS_ALL_ONES(event->dword[1])));
188 }
189
190 /**************************************************************************
191  *
192  * I2C bus - this is a bit-bashing interface using GPIO pins
193  * Note that it uses the output enables to tristate the outputs
194  * SDA is the data pin and SCL is the clock
195  *
196  **************************************************************************
197  */
198 static void falcon_setsda(void *data, int state)
199 {
200         struct efx_nic *efx = (struct efx_nic *)data;
201         efx_oword_t reg;
202
203         falcon_read(efx, &reg, GPIO_CTL_REG_KER);
204         EFX_SET_OWORD_FIELD(reg, GPIO3_OEN, !state);
205         falcon_write(efx, &reg, GPIO_CTL_REG_KER);
206 }
207
208 static void falcon_setscl(void *data, int state)
209 {
210         struct efx_nic *efx = (struct efx_nic *)data;
211         efx_oword_t reg;
212
213         falcon_read(efx, &reg, GPIO_CTL_REG_KER);
214         EFX_SET_OWORD_FIELD(reg, GPIO0_OEN, !state);
215         falcon_write(efx, &reg, GPIO_CTL_REG_KER);
216 }
217
218 static int falcon_getsda(void *data)
219 {
220         struct efx_nic *efx = (struct efx_nic *)data;
221         efx_oword_t reg;
222
223         falcon_read(efx, &reg, GPIO_CTL_REG_KER);
224         return EFX_OWORD_FIELD(reg, GPIO3_IN);
225 }
226
227 static int falcon_getscl(void *data)
228 {
229         struct efx_nic *efx = (struct efx_nic *)data;
230         efx_oword_t reg;
231
232         falcon_read(efx, &reg, GPIO_CTL_REG_KER);
233         return EFX_OWORD_FIELD(reg, GPIO0_IN);
234 }
235
236 static struct i2c_algo_bit_data falcon_i2c_bit_operations = {
237         .setsda         = falcon_setsda,
238         .setscl         = falcon_setscl,
239         .getsda         = falcon_getsda,
240         .getscl         = falcon_getscl,
241         .udelay         = 5,
242         /* Wait up to 50 ms for slave to let us pull SCL high */
243         .timeout        = DIV_ROUND_UP(HZ, 20),
244 };
245
246 /**************************************************************************
247  *
248  * Falcon special buffer handling
249  * Special buffers are used for event queues and the TX and RX
250  * descriptor rings.
251  *
252  *************************************************************************/
253
254 /*
255  * Initialise a Falcon special buffer
256  *
257  * This will define a buffer (previously allocated via
258  * falcon_alloc_special_buffer()) in Falcon's buffer table, allowing
259  * it to be used for event queues, descriptor rings etc.
260  */
261 static void
262 falcon_init_special_buffer(struct efx_nic *efx,
263                            struct efx_special_buffer *buffer)
264 {
265         efx_qword_t buf_desc;
266         int index;
267         dma_addr_t dma_addr;
268         int i;
269
270         EFX_BUG_ON_PARANOID(!buffer->addr);
271
272         /* Write buffer descriptors to NIC */
273         for (i = 0; i < buffer->entries; i++) {
274                 index = buffer->index + i;
275                 dma_addr = buffer->dma_addr + (i * 4096);
276                 EFX_LOG(efx, "mapping special buffer %d at %llx\n",
277                         index, (unsigned long long)dma_addr);
278                 EFX_POPULATE_QWORD_4(buf_desc,
279                                      IP_DAT_BUF_SIZE, IP_DAT_BUF_SIZE_4K,
280                                      BUF_ADR_REGION, 0,
281                                      BUF_ADR_FBUF, (dma_addr >> 12),
282                                      BUF_OWNER_ID_FBUF, 0);
283                 falcon_write_sram(efx, &buf_desc, index);
284         }
285 }
286
287 /* Unmaps a buffer from Falcon and clears the buffer table entries */
288 static void
289 falcon_fini_special_buffer(struct efx_nic *efx,
290                            struct efx_special_buffer *buffer)
291 {
292         efx_oword_t buf_tbl_upd;
293         unsigned int start = buffer->index;
294         unsigned int end = (buffer->index + buffer->entries - 1);
295
296         if (!buffer->entries)
297                 return;
298
299         EFX_LOG(efx, "unmapping special buffers %d-%d\n",
300                 buffer->index, buffer->index + buffer->entries - 1);
301
302         EFX_POPULATE_OWORD_4(buf_tbl_upd,
303                              BUF_UPD_CMD, 0,
304                              BUF_CLR_CMD, 1,
305                              BUF_CLR_END_ID, end,
306                              BUF_CLR_START_ID, start);
307         falcon_write(efx, &buf_tbl_upd, BUF_TBL_UPD_REG_KER);
308 }
309
310 /*
311  * Allocate a new Falcon special buffer
312  *
313  * This allocates memory for a new buffer, clears it and allocates a
314  * new buffer ID range.  It does not write into Falcon's buffer table.
315  *
316  * This call will allocate 4KB buffers, since Falcon can't use 8KB
317  * buffers for event queues and descriptor rings.
318  */
319 static int falcon_alloc_special_buffer(struct efx_nic *efx,
320                                        struct efx_special_buffer *buffer,
321                                        unsigned int len)
322 {
323         struct falcon_nic_data *nic_data = efx->nic_data;
324
325         len = ALIGN(len, FALCON_BUF_SIZE);
326
327         buffer->addr = pci_alloc_consistent(efx->pci_dev, len,
328                                             &buffer->dma_addr);
329         if (!buffer->addr)
330                 return -ENOMEM;
331         buffer->len = len;
332         buffer->entries = len / FALCON_BUF_SIZE;
333         BUG_ON(buffer->dma_addr & (FALCON_BUF_SIZE - 1));
334
335         /* All zeros is a potentially valid event so memset to 0xff */
336         memset(buffer->addr, 0xff, len);
337
338         /* Select new buffer ID */
339         buffer->index = nic_data->next_buffer_table;
340         nic_data->next_buffer_table += buffer->entries;
341
342         EFX_LOG(efx, "allocating special buffers %d-%d at %llx+%x "
343                 "(virt %p phys %llx)\n", buffer->index,
344                 buffer->index + buffer->entries - 1,
345                 (u64)buffer->dma_addr, len,
346                 buffer->addr, (u64)virt_to_phys(buffer->addr));
347
348         return 0;
349 }
350
351 static void falcon_free_special_buffer(struct efx_nic *efx,
352                                        struct efx_special_buffer *buffer)
353 {
354         if (!buffer->addr)
355                 return;
356
357         EFX_LOG(efx, "deallocating special buffers %d-%d at %llx+%x "
358                 "(virt %p phys %llx)\n", buffer->index,
359                 buffer->index + buffer->entries - 1,
360                 (u64)buffer->dma_addr, buffer->len,
361                 buffer->addr, (u64)virt_to_phys(buffer->addr));
362
363         pci_free_consistent(efx->pci_dev, buffer->len, buffer->addr,
364                             buffer->dma_addr);
365         buffer->addr = NULL;
366         buffer->entries = 0;
367 }
368
369 /**************************************************************************
370  *
371  * Falcon generic buffer handling
372  * These buffers are used for interrupt status and MAC stats
373  *
374  **************************************************************************/
375
376 static int falcon_alloc_buffer(struct efx_nic *efx,
377                                struct efx_buffer *buffer, unsigned int len)
378 {
379         buffer->addr = pci_alloc_consistent(efx->pci_dev, len,
380                                             &buffer->dma_addr);
381         if (!buffer->addr)
382                 return -ENOMEM;
383         buffer->len = len;
384         memset(buffer->addr, 0, len);
385         return 0;
386 }
387
388 static void falcon_free_buffer(struct efx_nic *efx, struct efx_buffer *buffer)
389 {
390         if (buffer->addr) {
391                 pci_free_consistent(efx->pci_dev, buffer->len,
392                                     buffer->addr, buffer->dma_addr);
393                 buffer->addr = NULL;
394         }
395 }
396
397 /**************************************************************************
398  *
399  * Falcon TX path
400  *
401  **************************************************************************/
402
403 /* Returns a pointer to the specified transmit descriptor in the TX
404  * descriptor queue belonging to the specified channel.
405  */
406 static inline efx_qword_t *falcon_tx_desc(struct efx_tx_queue *tx_queue,
407                                                unsigned int index)
408 {
409         return (((efx_qword_t *) (tx_queue->txd.addr)) + index);
410 }
411
412 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
413 static inline void falcon_notify_tx_desc(struct efx_tx_queue *tx_queue)
414 {
415         unsigned write_ptr;
416         efx_dword_t reg;
417
418         write_ptr = tx_queue->write_count & FALCON_TXD_RING_MASK;
419         EFX_POPULATE_DWORD_1(reg, TX_DESC_WPTR_DWORD, write_ptr);
420         falcon_writel_page(tx_queue->efx, &reg,
421                            TX_DESC_UPD_REG_KER_DWORD, tx_queue->queue);
422 }
423
424
425 /* For each entry inserted into the software descriptor ring, create a
426  * descriptor in the hardware TX descriptor ring (in host memory), and
427  * write a doorbell.
428  */
429 void falcon_push_buffers(struct efx_tx_queue *tx_queue)
430 {
431
432         struct efx_tx_buffer *buffer;
433         efx_qword_t *txd;
434         unsigned write_ptr;
435
436         BUG_ON(tx_queue->write_count == tx_queue->insert_count);
437
438         do {
439                 write_ptr = tx_queue->write_count & FALCON_TXD_RING_MASK;
440                 buffer = &tx_queue->buffer[write_ptr];
441                 txd = falcon_tx_desc(tx_queue, write_ptr);
442                 ++tx_queue->write_count;
443
444                 /* Create TX descriptor ring entry */
445                 EFX_POPULATE_QWORD_5(*txd,
446                                      TX_KER_PORT, 0,
447                                      TX_KER_CONT, buffer->continuation,
448                                      TX_KER_BYTE_CNT, buffer->len,
449                                      TX_KER_BUF_REGION, 0,
450                                      TX_KER_BUF_ADR, buffer->dma_addr);
451         } while (tx_queue->write_count != tx_queue->insert_count);
452
453         wmb(); /* Ensure descriptors are written before they are fetched */
454         falcon_notify_tx_desc(tx_queue);
455 }
456
457 /* Allocate hardware resources for a TX queue */
458 int falcon_probe_tx(struct efx_tx_queue *tx_queue)
459 {
460         struct efx_nic *efx = tx_queue->efx;
461         return falcon_alloc_special_buffer(efx, &tx_queue->txd,
462                                            FALCON_TXD_RING_SIZE *
463                                            sizeof(efx_qword_t));
464 }
465
466 void falcon_init_tx(struct efx_tx_queue *tx_queue)
467 {
468         efx_oword_t tx_desc_ptr;
469         struct efx_nic *efx = tx_queue->efx;
470
471         tx_queue->flushed = false;
472
473         /* Pin TX descriptor ring */
474         falcon_init_special_buffer(efx, &tx_queue->txd);
475
476         /* Push TX descriptor ring to card */
477         EFX_POPULATE_OWORD_10(tx_desc_ptr,
478                               TX_DESCQ_EN, 1,
479                               TX_ISCSI_DDIG_EN, 0,
480                               TX_ISCSI_HDIG_EN, 0,
481                               TX_DESCQ_BUF_BASE_ID, tx_queue->txd.index,
482                               TX_DESCQ_EVQ_ID, tx_queue->channel->channel,
483                               TX_DESCQ_OWNER_ID, 0,
484                               TX_DESCQ_LABEL, tx_queue->queue,
485                               TX_DESCQ_SIZE, FALCON_TXD_RING_ORDER,
486                               TX_DESCQ_TYPE, 0,
487                               TX_NON_IP_DROP_DIS_B0, 1);
488
489         if (falcon_rev(efx) >= FALCON_REV_B0) {
490                 int csum = tx_queue->queue == EFX_TX_QUEUE_OFFLOAD_CSUM;
491                 EFX_SET_OWORD_FIELD(tx_desc_ptr, TX_IP_CHKSM_DIS_B0, !csum);
492                 EFX_SET_OWORD_FIELD(tx_desc_ptr, TX_TCP_CHKSM_DIS_B0, !csum);
493         }
494
495         falcon_write_table(efx, &tx_desc_ptr, efx->type->txd_ptr_tbl_base,
496                            tx_queue->queue);
497
498         if (falcon_rev(efx) < FALCON_REV_B0) {
499                 efx_oword_t reg;
500
501                 /* Only 128 bits in this register */
502                 BUILD_BUG_ON(EFX_TX_QUEUE_COUNT >= 128);
503
504                 falcon_read(efx, &reg, TX_CHKSM_CFG_REG_KER_A1);
505                 if (tx_queue->queue == EFX_TX_QUEUE_OFFLOAD_CSUM)
506                         clear_bit_le(tx_queue->queue, (void *)&reg);
507                 else
508                         set_bit_le(tx_queue->queue, (void *)&reg);
509                 falcon_write(efx, &reg, TX_CHKSM_CFG_REG_KER_A1);
510         }
511 }
512
513 static void falcon_flush_tx_queue(struct efx_tx_queue *tx_queue)
514 {
515         struct efx_nic *efx = tx_queue->efx;
516         efx_oword_t tx_flush_descq;
517
518         /* Post a flush command */
519         EFX_POPULATE_OWORD_2(tx_flush_descq,
520                              TX_FLUSH_DESCQ_CMD, 1,
521                              TX_FLUSH_DESCQ, tx_queue->queue);
522         falcon_write(efx, &tx_flush_descq, TX_FLUSH_DESCQ_REG_KER);
523 }
524
525 void falcon_fini_tx(struct efx_tx_queue *tx_queue)
526 {
527         struct efx_nic *efx = tx_queue->efx;
528         efx_oword_t tx_desc_ptr;
529
530         /* The queue should have been flushed */
531         WARN_ON(!tx_queue->flushed);
532
533         /* Remove TX descriptor ring from card */
534         EFX_ZERO_OWORD(tx_desc_ptr);
535         falcon_write_table(efx, &tx_desc_ptr, efx->type->txd_ptr_tbl_base,
536                            tx_queue->queue);
537
538         /* Unpin TX descriptor ring */
539         falcon_fini_special_buffer(efx, &tx_queue->txd);
540 }
541
542 /* Free buffers backing TX queue */
543 void falcon_remove_tx(struct efx_tx_queue *tx_queue)
544 {
545         falcon_free_special_buffer(tx_queue->efx, &tx_queue->txd);
546 }
547
548 /**************************************************************************
549  *
550  * Falcon RX path
551  *
552  **************************************************************************/
553
554 /* Returns a pointer to the specified descriptor in the RX descriptor queue */
555 static inline efx_qword_t *falcon_rx_desc(struct efx_rx_queue *rx_queue,
556                                                unsigned int index)
557 {
558         return (((efx_qword_t *) (rx_queue->rxd.addr)) + index);
559 }
560
561 /* This creates an entry in the RX descriptor queue */
562 static inline void falcon_build_rx_desc(struct efx_rx_queue *rx_queue,
563                                         unsigned index)
564 {
565         struct efx_rx_buffer *rx_buf;
566         efx_qword_t *rxd;
567
568         rxd = falcon_rx_desc(rx_queue, index);
569         rx_buf = efx_rx_buffer(rx_queue, index);
570         EFX_POPULATE_QWORD_3(*rxd,
571                              RX_KER_BUF_SIZE,
572                              rx_buf->len -
573                              rx_queue->efx->type->rx_buffer_padding,
574                              RX_KER_BUF_REGION, 0,
575                              RX_KER_BUF_ADR, rx_buf->dma_addr);
576 }
577
578 /* This writes to the RX_DESC_WPTR register for the specified receive
579  * descriptor ring.
580  */
581 void falcon_notify_rx_desc(struct efx_rx_queue *rx_queue)
582 {
583         efx_dword_t reg;
584         unsigned write_ptr;
585
586         while (rx_queue->notified_count != rx_queue->added_count) {
587                 falcon_build_rx_desc(rx_queue,
588                                      rx_queue->notified_count &
589                                      FALCON_RXD_RING_MASK);
590                 ++rx_queue->notified_count;
591         }
592
593         wmb();
594         write_ptr = rx_queue->added_count & FALCON_RXD_RING_MASK;
595         EFX_POPULATE_DWORD_1(reg, RX_DESC_WPTR_DWORD, write_ptr);
596         falcon_writel_page(rx_queue->efx, &reg,
597                            RX_DESC_UPD_REG_KER_DWORD, rx_queue->queue);
598 }
599
600 int falcon_probe_rx(struct efx_rx_queue *rx_queue)
601 {
602         struct efx_nic *efx = rx_queue->efx;
603         return falcon_alloc_special_buffer(efx, &rx_queue->rxd,
604                                            FALCON_RXD_RING_SIZE *
605                                            sizeof(efx_qword_t));
606 }
607
608 void falcon_init_rx(struct efx_rx_queue *rx_queue)
609 {
610         efx_oword_t rx_desc_ptr;
611         struct efx_nic *efx = rx_queue->efx;
612         bool is_b0 = falcon_rev(efx) >= FALCON_REV_B0;
613         bool iscsi_digest_en = is_b0;
614
615         EFX_LOG(efx, "RX queue %d ring in special buffers %d-%d\n",
616                 rx_queue->queue, rx_queue->rxd.index,
617                 rx_queue->rxd.index + rx_queue->rxd.entries - 1);
618
619         rx_queue->flushed = false;
620
621         /* Pin RX descriptor ring */
622         falcon_init_special_buffer(efx, &rx_queue->rxd);
623
624         /* Push RX descriptor ring to card */
625         EFX_POPULATE_OWORD_10(rx_desc_ptr,
626                               RX_ISCSI_DDIG_EN, iscsi_digest_en,
627                               RX_ISCSI_HDIG_EN, iscsi_digest_en,
628                               RX_DESCQ_BUF_BASE_ID, rx_queue->rxd.index,
629                               RX_DESCQ_EVQ_ID, rx_queue->channel->channel,
630                               RX_DESCQ_OWNER_ID, 0,
631                               RX_DESCQ_LABEL, rx_queue->queue,
632                               RX_DESCQ_SIZE, FALCON_RXD_RING_ORDER,
633                               RX_DESCQ_TYPE, 0 /* kernel queue */ ,
634                               /* For >=B0 this is scatter so disable */
635                               RX_DESCQ_JUMBO, !is_b0,
636                               RX_DESCQ_EN, 1);
637         falcon_write_table(efx, &rx_desc_ptr, efx->type->rxd_ptr_tbl_base,
638                            rx_queue->queue);
639 }
640
641 static void falcon_flush_rx_queue(struct efx_rx_queue *rx_queue)
642 {
643         struct efx_nic *efx = rx_queue->efx;
644         efx_oword_t rx_flush_descq;
645
646         /* Post a flush command */
647         EFX_POPULATE_OWORD_2(rx_flush_descq,
648                              RX_FLUSH_DESCQ_CMD, 1,
649                              RX_FLUSH_DESCQ, rx_queue->queue);
650         falcon_write(efx, &rx_flush_descq, RX_FLUSH_DESCQ_REG_KER);
651 }
652
653 void falcon_fini_rx(struct efx_rx_queue *rx_queue)
654 {
655         efx_oword_t rx_desc_ptr;
656         struct efx_nic *efx = rx_queue->efx;
657
658         /* The queue should already have been flushed */
659         WARN_ON(!rx_queue->flushed);
660
661         /* Remove RX descriptor ring from card */
662         EFX_ZERO_OWORD(rx_desc_ptr);
663         falcon_write_table(efx, &rx_desc_ptr, efx->type->rxd_ptr_tbl_base,
664                            rx_queue->queue);
665
666         /* Unpin RX descriptor ring */
667         falcon_fini_special_buffer(efx, &rx_queue->rxd);
668 }
669
670 /* Free buffers backing RX queue */
671 void falcon_remove_rx(struct efx_rx_queue *rx_queue)
672 {
673         falcon_free_special_buffer(rx_queue->efx, &rx_queue->rxd);
674 }
675
676 /**************************************************************************
677  *
678  * Falcon event queue processing
679  * Event queues are processed by per-channel tasklets.
680  *
681  **************************************************************************/
682
683 /* Update a channel's event queue's read pointer (RPTR) register
684  *
685  * This writes the EVQ_RPTR_REG register for the specified channel's
686  * event queue.
687  *
688  * Note that EVQ_RPTR_REG contains the index of the "last read" event,
689  * whereas channel->eventq_read_ptr contains the index of the "next to
690  * read" event.
691  */
692 void falcon_eventq_read_ack(struct efx_channel *channel)
693 {
694         efx_dword_t reg;
695         struct efx_nic *efx = channel->efx;
696
697         EFX_POPULATE_DWORD_1(reg, EVQ_RPTR_DWORD, channel->eventq_read_ptr);
698         falcon_writel_table(efx, &reg, efx->type->evq_rptr_tbl_base,
699                             channel->channel);
700 }
701
702 /* Use HW to insert a SW defined event */
703 void falcon_generate_event(struct efx_channel *channel, efx_qword_t *event)
704 {
705         efx_oword_t drv_ev_reg;
706
707         EFX_POPULATE_OWORD_2(drv_ev_reg,
708                              DRV_EV_QID, channel->channel,
709                              DRV_EV_DATA,
710                              EFX_QWORD_FIELD64(*event, WHOLE_EVENT));
711         falcon_write(channel->efx, &drv_ev_reg, DRV_EV_REG_KER);
712 }
713
714 /* Handle a transmit completion event
715  *
716  * Falcon batches TX completion events; the message we receive is of
717  * the form "complete all TX events up to this index".
718  */
719 static void falcon_handle_tx_event(struct efx_channel *channel,
720                                    efx_qword_t *event)
721 {
722         unsigned int tx_ev_desc_ptr;
723         unsigned int tx_ev_q_label;
724         struct efx_tx_queue *tx_queue;
725         struct efx_nic *efx = channel->efx;
726
727         if (likely(EFX_QWORD_FIELD(*event, TX_EV_COMP))) {
728                 /* Transmit completion */
729                 tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, TX_EV_DESC_PTR);
730                 tx_ev_q_label = EFX_QWORD_FIELD(*event, TX_EV_Q_LABEL);
731                 tx_queue = &efx->tx_queue[tx_ev_q_label];
732                 channel->irq_mod_score +=
733                         (tx_ev_desc_ptr - tx_queue->read_count) &
734                         efx->type->txd_ring_mask;
735                 efx_xmit_done(tx_queue, tx_ev_desc_ptr);
736         } else if (EFX_QWORD_FIELD(*event, TX_EV_WQ_FF_FULL)) {
737                 /* Rewrite the FIFO write pointer */
738                 tx_ev_q_label = EFX_QWORD_FIELD(*event, TX_EV_Q_LABEL);
739                 tx_queue = &efx->tx_queue[tx_ev_q_label];
740
741                 if (efx_dev_registered(efx))
742                         netif_tx_lock(efx->net_dev);
743                 falcon_notify_tx_desc(tx_queue);
744                 if (efx_dev_registered(efx))
745                         netif_tx_unlock(efx->net_dev);
746         } else if (EFX_QWORD_FIELD(*event, TX_EV_PKT_ERR) &&
747                    EFX_WORKAROUND_10727(efx)) {
748                 efx_schedule_reset(efx, RESET_TYPE_TX_DESC_FETCH);
749         } else {
750                 EFX_ERR(efx, "channel %d unexpected TX event "
751                         EFX_QWORD_FMT"\n", channel->channel,
752                         EFX_QWORD_VAL(*event));
753         }
754 }
755
756 /* Detect errors included in the rx_evt_pkt_ok bit. */
757 static void falcon_handle_rx_not_ok(struct efx_rx_queue *rx_queue,
758                                     const efx_qword_t *event,
759                                     bool *rx_ev_pkt_ok,
760                                     bool *discard)
761 {
762         struct efx_nic *efx = rx_queue->efx;
763         bool rx_ev_buf_owner_id_err, rx_ev_ip_hdr_chksum_err;
764         bool rx_ev_tcp_udp_chksum_err, rx_ev_eth_crc_err;
765         bool rx_ev_frm_trunc, rx_ev_drib_nib, rx_ev_tobe_disc;
766         bool rx_ev_other_err, rx_ev_pause_frm;
767         bool rx_ev_ip_frag_err, rx_ev_hdr_type, rx_ev_mcast_pkt;
768         unsigned rx_ev_pkt_type;
769
770         rx_ev_hdr_type = EFX_QWORD_FIELD(*event, RX_EV_HDR_TYPE);
771         rx_ev_mcast_pkt = EFX_QWORD_FIELD(*event, RX_EV_MCAST_PKT);
772         rx_ev_tobe_disc = EFX_QWORD_FIELD(*event, RX_EV_TOBE_DISC);
773         rx_ev_pkt_type = EFX_QWORD_FIELD(*event, RX_EV_PKT_TYPE);
774         rx_ev_buf_owner_id_err = EFX_QWORD_FIELD(*event,
775                                                  RX_EV_BUF_OWNER_ID_ERR);
776         rx_ev_ip_frag_err = EFX_QWORD_FIELD(*event, RX_EV_IF_FRAG_ERR);
777         rx_ev_ip_hdr_chksum_err = EFX_QWORD_FIELD(*event,
778                                                   RX_EV_IP_HDR_CHKSUM_ERR);
779         rx_ev_tcp_udp_chksum_err = EFX_QWORD_FIELD(*event,
780                                                    RX_EV_TCP_UDP_CHKSUM_ERR);
781         rx_ev_eth_crc_err = EFX_QWORD_FIELD(*event, RX_EV_ETH_CRC_ERR);
782         rx_ev_frm_trunc = EFX_QWORD_FIELD(*event, RX_EV_FRM_TRUNC);
783         rx_ev_drib_nib = ((falcon_rev(efx) >= FALCON_REV_B0) ?
784                           0 : EFX_QWORD_FIELD(*event, RX_EV_DRIB_NIB));
785         rx_ev_pause_frm = EFX_QWORD_FIELD(*event, RX_EV_PAUSE_FRM_ERR);
786
787         /* Every error apart from tobe_disc and pause_frm */
788         rx_ev_other_err = (rx_ev_drib_nib | rx_ev_tcp_udp_chksum_err |
789                            rx_ev_buf_owner_id_err | rx_ev_eth_crc_err |
790                            rx_ev_frm_trunc | rx_ev_ip_hdr_chksum_err);
791
792         /* Count errors that are not in MAC stats.  Ignore expected
793          * checksum errors during self-test. */
794         if (rx_ev_frm_trunc)
795                 ++rx_queue->channel->n_rx_frm_trunc;
796         else if (rx_ev_tobe_disc)
797                 ++rx_queue->channel->n_rx_tobe_disc;
798         else if (!efx->loopback_selftest) {
799                 if (rx_ev_ip_hdr_chksum_err)
800                         ++rx_queue->channel->n_rx_ip_hdr_chksum_err;
801                 else if (rx_ev_tcp_udp_chksum_err)
802                         ++rx_queue->channel->n_rx_tcp_udp_chksum_err;
803         }
804         if (rx_ev_ip_frag_err)
805                 ++rx_queue->channel->n_rx_ip_frag_err;
806
807         /* The frame must be discarded if any of these are true. */
808         *discard = (rx_ev_eth_crc_err | rx_ev_frm_trunc | rx_ev_drib_nib |
809                     rx_ev_tobe_disc | rx_ev_pause_frm);
810
811         /* TOBE_DISC is expected on unicast mismatches; don't print out an
812          * error message.  FRM_TRUNC indicates RXDP dropped the packet due
813          * to a FIFO overflow.
814          */
815 #ifdef EFX_ENABLE_DEBUG
816         if (rx_ev_other_err) {
817                 EFX_INFO_RL(efx, " RX queue %d unexpected RX event "
818                             EFX_QWORD_FMT "%s%s%s%s%s%s%s%s\n",
819                             rx_queue->queue, EFX_QWORD_VAL(*event),
820                             rx_ev_buf_owner_id_err ? " [OWNER_ID_ERR]" : "",
821                             rx_ev_ip_hdr_chksum_err ?
822                             " [IP_HDR_CHKSUM_ERR]" : "",
823                             rx_ev_tcp_udp_chksum_err ?
824                             " [TCP_UDP_CHKSUM_ERR]" : "",
825                             rx_ev_eth_crc_err ? " [ETH_CRC_ERR]" : "",
826                             rx_ev_frm_trunc ? " [FRM_TRUNC]" : "",
827                             rx_ev_drib_nib ? " [DRIB_NIB]" : "",
828                             rx_ev_tobe_disc ? " [TOBE_DISC]" : "",
829                             rx_ev_pause_frm ? " [PAUSE]" : "");
830         }
831 #endif
832 }
833
834 /* Handle receive events that are not in-order. */
835 static void falcon_handle_rx_bad_index(struct efx_rx_queue *rx_queue,
836                                        unsigned index)
837 {
838         struct efx_nic *efx = rx_queue->efx;
839         unsigned expected, dropped;
840
841         expected = rx_queue->removed_count & FALCON_RXD_RING_MASK;
842         dropped = ((index + FALCON_RXD_RING_SIZE - expected) &
843                    FALCON_RXD_RING_MASK);
844         EFX_INFO(efx, "dropped %d events (index=%d expected=%d)\n",
845                 dropped, index, expected);
846
847         efx_schedule_reset(efx, EFX_WORKAROUND_5676(efx) ?
848                            RESET_TYPE_RX_RECOVERY : RESET_TYPE_DISABLE);
849 }
850
851 /* Handle a packet received event
852  *
853  * Falcon silicon gives a "discard" flag if it's a unicast packet with the
854  * wrong destination address
855  * Also "is multicast" and "matches multicast filter" flags can be used to
856  * discard non-matching multicast packets.
857  */
858 static void falcon_handle_rx_event(struct efx_channel *channel,
859                                    const efx_qword_t *event)
860 {
861         unsigned int rx_ev_desc_ptr, rx_ev_byte_cnt;
862         unsigned int rx_ev_hdr_type, rx_ev_mcast_pkt;
863         unsigned expected_ptr;
864         bool rx_ev_pkt_ok, discard = false, checksummed;
865         struct efx_rx_queue *rx_queue;
866         struct efx_nic *efx = channel->efx;
867
868         /* Basic packet information */
869         rx_ev_byte_cnt = EFX_QWORD_FIELD(*event, RX_EV_BYTE_CNT);
870         rx_ev_pkt_ok = EFX_QWORD_FIELD(*event, RX_EV_PKT_OK);
871         rx_ev_hdr_type = EFX_QWORD_FIELD(*event, RX_EV_HDR_TYPE);
872         WARN_ON(EFX_QWORD_FIELD(*event, RX_EV_JUMBO_CONT));
873         WARN_ON(EFX_QWORD_FIELD(*event, RX_EV_SOP) != 1);
874         WARN_ON(EFX_QWORD_FIELD(*event, RX_EV_Q_LABEL) != channel->channel);
875
876         rx_queue = &efx->rx_queue[channel->channel];
877
878         rx_ev_desc_ptr = EFX_QWORD_FIELD(*event, RX_EV_DESC_PTR);
879         expected_ptr = rx_queue->removed_count & FALCON_RXD_RING_MASK;
880         if (unlikely(rx_ev_desc_ptr != expected_ptr))
881                 falcon_handle_rx_bad_index(rx_queue, rx_ev_desc_ptr);
882
883         if (likely(rx_ev_pkt_ok)) {
884                 /* If packet is marked as OK and packet type is TCP/IPv4 or
885                  * UDP/IPv4, then we can rely on the hardware checksum.
886                  */
887                 checksummed = RX_EV_HDR_TYPE_HAS_CHECKSUMS(rx_ev_hdr_type);
888         } else {
889                 falcon_handle_rx_not_ok(rx_queue, event, &rx_ev_pkt_ok,
890                                         &discard);
891                 checksummed = false;
892         }
893
894         /* Detect multicast packets that didn't match the filter */
895         rx_ev_mcast_pkt = EFX_QWORD_FIELD(*event, RX_EV_MCAST_PKT);
896         if (rx_ev_mcast_pkt) {
897                 unsigned int rx_ev_mcast_hash_match =
898                         EFX_QWORD_FIELD(*event, RX_EV_MCAST_HASH_MATCH);
899
900                 if (unlikely(!rx_ev_mcast_hash_match))
901                         discard = true;
902         }
903
904         channel->irq_mod_score += 2;
905
906         /* Handle received packet */
907         efx_rx_packet(rx_queue, rx_ev_desc_ptr, rx_ev_byte_cnt,
908                       checksummed, discard);
909 }
910
911 /* Global events are basically PHY events */
912 static void falcon_handle_global_event(struct efx_channel *channel,
913                                        efx_qword_t *event)
914 {
915         struct efx_nic *efx = channel->efx;
916         bool handled = false;
917
918         if (EFX_QWORD_FIELD(*event, G_PHY0_INTR) ||
919             EFX_QWORD_FIELD(*event, G_PHY1_INTR) ||
920             EFX_QWORD_FIELD(*event, XG_PHY_INTR) ||
921             EFX_QWORD_FIELD(*event, XFP_PHY_INTR)) {
922                 efx->phy_op->clear_interrupt(efx);
923                 queue_work(efx->workqueue, &efx->phy_work);
924                 handled = true;
925         }
926
927         if ((falcon_rev(efx) >= FALCON_REV_B0) &&
928             EFX_QWORD_FIELD(*event, XG_MNT_INTR_B0)) {
929                 queue_work(efx->workqueue, &efx->mac_work);
930                 handled = true;
931         }
932
933         if (EFX_QWORD_FIELD_VER(efx, *event, RX_RECOVERY)) {
934                 EFX_ERR(efx, "channel %d seen global RX_RESET "
935                         "event. Resetting.\n", channel->channel);
936
937                 atomic_inc(&efx->rx_reset);
938                 efx_schedule_reset(efx, EFX_WORKAROUND_6555(efx) ?
939                                    RESET_TYPE_RX_RECOVERY : RESET_TYPE_DISABLE);
940                 handled = true;
941         }
942
943         if (!handled)
944                 EFX_ERR(efx, "channel %d unknown global event "
945                         EFX_QWORD_FMT "\n", channel->channel,
946                         EFX_QWORD_VAL(*event));
947 }
948
949 static void falcon_handle_driver_event(struct efx_channel *channel,
950                                        efx_qword_t *event)
951 {
952         struct efx_nic *efx = channel->efx;
953         unsigned int ev_sub_code;
954         unsigned int ev_sub_data;
955
956         ev_sub_code = EFX_QWORD_FIELD(*event, DRIVER_EV_SUB_CODE);
957         ev_sub_data = EFX_QWORD_FIELD(*event, DRIVER_EV_SUB_DATA);
958
959         switch (ev_sub_code) {
960         case TX_DESCQ_FLS_DONE_EV_DECODE:
961                 EFX_TRACE(efx, "channel %d TXQ %d flushed\n",
962                           channel->channel, ev_sub_data);
963                 break;
964         case RX_DESCQ_FLS_DONE_EV_DECODE:
965                 EFX_TRACE(efx, "channel %d RXQ %d flushed\n",
966                           channel->channel, ev_sub_data);
967                 break;
968         case EVQ_INIT_DONE_EV_DECODE:
969                 EFX_LOG(efx, "channel %d EVQ %d initialised\n",
970                         channel->channel, ev_sub_data);
971                 break;
972         case SRM_UPD_DONE_EV_DECODE:
973                 EFX_TRACE(efx, "channel %d SRAM update done\n",
974                           channel->channel);
975                 break;
976         case WAKE_UP_EV_DECODE:
977                 EFX_TRACE(efx, "channel %d RXQ %d wakeup event\n",
978                           channel->channel, ev_sub_data);
979                 break;
980         case TIMER_EV_DECODE:
981                 EFX_TRACE(efx, "channel %d RX queue %d timer expired\n",
982                           channel->channel, ev_sub_data);
983                 break;
984         case RX_RECOVERY_EV_DECODE:
985                 EFX_ERR(efx, "channel %d seen DRIVER RX_RESET event. "
986                         "Resetting.\n", channel->channel);
987                 atomic_inc(&efx->rx_reset);
988                 efx_schedule_reset(efx,
989                                    EFX_WORKAROUND_6555(efx) ?
990                                    RESET_TYPE_RX_RECOVERY :
991                                    RESET_TYPE_DISABLE);
992                 break;
993         case RX_DSC_ERROR_EV_DECODE:
994                 EFX_ERR(efx, "RX DMA Q %d reports descriptor fetch error."
995                         " RX Q %d is disabled.\n", ev_sub_data, ev_sub_data);
996                 efx_schedule_reset(efx, RESET_TYPE_RX_DESC_FETCH);
997                 break;
998         case TX_DSC_ERROR_EV_DECODE:
999                 EFX_ERR(efx, "TX DMA Q %d reports descriptor fetch error."
1000                         " TX Q %d is disabled.\n", ev_sub_data, ev_sub_data);
1001                 efx_schedule_reset(efx, RESET_TYPE_TX_DESC_FETCH);
1002                 break;
1003         default:
1004                 EFX_TRACE(efx, "channel %d unknown driver event code %d "
1005                           "data %04x\n", channel->channel, ev_sub_code,
1006                           ev_sub_data);
1007                 break;
1008         }
1009 }
1010
1011 int falcon_process_eventq(struct efx_channel *channel, int rx_quota)
1012 {
1013         unsigned int read_ptr;
1014         efx_qword_t event, *p_event;
1015         int ev_code;
1016         int rx_packets = 0;
1017
1018         read_ptr = channel->eventq_read_ptr;
1019
1020         do {
1021                 p_event = falcon_event(channel, read_ptr);
1022                 event = *p_event;
1023
1024                 if (!falcon_event_present(&event))
1025                         /* End of events */
1026                         break;
1027
1028                 EFX_TRACE(channel->efx, "channel %d event is "EFX_QWORD_FMT"\n",
1029                           channel->channel, EFX_QWORD_VAL(event));
1030
1031                 /* Clear this event by marking it all ones */
1032                 EFX_SET_QWORD(*p_event);
1033
1034                 ev_code = EFX_QWORD_FIELD(event, EV_CODE);
1035
1036                 switch (ev_code) {
1037                 case RX_IP_EV_DECODE:
1038                         falcon_handle_rx_event(channel, &event);
1039                         ++rx_packets;
1040                         break;
1041                 case TX_IP_EV_DECODE:
1042                         falcon_handle_tx_event(channel, &event);
1043                         break;
1044                 case DRV_GEN_EV_DECODE:
1045                         channel->eventq_magic
1046                                 = EFX_QWORD_FIELD(event, EVQ_MAGIC);
1047                         EFX_LOG(channel->efx, "channel %d received generated "
1048                                 "event "EFX_QWORD_FMT"\n", channel->channel,
1049                                 EFX_QWORD_VAL(event));
1050                         break;
1051                 case GLOBAL_EV_DECODE:
1052                         falcon_handle_global_event(channel, &event);
1053                         break;
1054                 case DRIVER_EV_DECODE:
1055                         falcon_handle_driver_event(channel, &event);
1056                         break;
1057                 default:
1058                         EFX_ERR(channel->efx, "channel %d unknown event type %d"
1059                                 " (data " EFX_QWORD_FMT ")\n", channel->channel,
1060                                 ev_code, EFX_QWORD_VAL(event));
1061                 }
1062
1063                 /* Increment read pointer */
1064                 read_ptr = (read_ptr + 1) & FALCON_EVQ_MASK;
1065
1066         } while (rx_packets < rx_quota);
1067
1068         channel->eventq_read_ptr = read_ptr;
1069         return rx_packets;
1070 }
1071
1072 void falcon_set_int_moderation(struct efx_channel *channel)
1073 {
1074         efx_dword_t timer_cmd;
1075         struct efx_nic *efx = channel->efx;
1076
1077         /* Set timer register */
1078         if (channel->irq_moderation) {
1079                 /* Round to resolution supported by hardware.  The value we
1080                  * program is based at 0.  So actual interrupt moderation
1081                  * achieved is ((x + 1) * res).
1082                  */
1083                 channel->irq_moderation -= (channel->irq_moderation %
1084                                             FALCON_IRQ_MOD_RESOLUTION);
1085                 if (channel->irq_moderation < FALCON_IRQ_MOD_RESOLUTION)
1086                         channel->irq_moderation = FALCON_IRQ_MOD_RESOLUTION;
1087                 EFX_POPULATE_DWORD_2(timer_cmd,
1088                                      TIMER_MODE, TIMER_MODE_INT_HLDOFF,
1089                                      TIMER_VAL,
1090                                      channel->irq_moderation /
1091                                      FALCON_IRQ_MOD_RESOLUTION - 1);
1092         } else {
1093                 EFX_POPULATE_DWORD_2(timer_cmd,
1094                                      TIMER_MODE, TIMER_MODE_DIS,
1095                                      TIMER_VAL, 0);
1096         }
1097         falcon_writel_page_locked(efx, &timer_cmd, TIMER_CMD_REG_KER,
1098                                   channel->channel);
1099
1100 }
1101
1102 /* Allocate buffer table entries for event queue */
1103 int falcon_probe_eventq(struct efx_channel *channel)
1104 {
1105         struct efx_nic *efx = channel->efx;
1106         unsigned int evq_size;
1107
1108         evq_size = FALCON_EVQ_SIZE * sizeof(efx_qword_t);
1109         return falcon_alloc_special_buffer(efx, &channel->eventq, evq_size);
1110 }
1111
1112 void falcon_init_eventq(struct efx_channel *channel)
1113 {
1114         efx_oword_t evq_ptr;
1115         struct efx_nic *efx = channel->efx;
1116
1117         EFX_LOG(efx, "channel %d event queue in special buffers %d-%d\n",
1118                 channel->channel, channel->eventq.index,
1119                 channel->eventq.index + channel->eventq.entries - 1);
1120
1121         /* Pin event queue buffer */
1122         falcon_init_special_buffer(efx, &channel->eventq);
1123
1124         /* Fill event queue with all ones (i.e. empty events) */
1125         memset(channel->eventq.addr, 0xff, channel->eventq.len);
1126
1127         /* Push event queue to card */
1128         EFX_POPULATE_OWORD_3(evq_ptr,
1129                              EVQ_EN, 1,
1130                              EVQ_SIZE, FALCON_EVQ_ORDER,
1131                              EVQ_BUF_BASE_ID, channel->eventq.index);
1132         falcon_write_table(efx, &evq_ptr, efx->type->evq_ptr_tbl_base,
1133                            channel->channel);
1134
1135         falcon_set_int_moderation(channel);
1136 }
1137
1138 void falcon_fini_eventq(struct efx_channel *channel)
1139 {
1140         efx_oword_t eventq_ptr;
1141         struct efx_nic *efx = channel->efx;
1142
1143         /* Remove event queue from card */
1144         EFX_ZERO_OWORD(eventq_ptr);
1145         falcon_write_table(efx, &eventq_ptr, efx->type->evq_ptr_tbl_base,
1146                            channel->channel);
1147
1148         /* Unpin event queue */
1149         falcon_fini_special_buffer(efx, &channel->eventq);
1150 }
1151
1152 /* Free buffers backing event queue */
1153 void falcon_remove_eventq(struct efx_channel *channel)
1154 {
1155         falcon_free_special_buffer(channel->efx, &channel->eventq);
1156 }
1157
1158
1159 /* Generates a test event on the event queue.  A subsequent call to
1160  * process_eventq() should pick up the event and place the value of
1161  * "magic" into channel->eventq_magic;
1162  */
1163 void falcon_generate_test_event(struct efx_channel *channel, unsigned int magic)
1164 {
1165         efx_qword_t test_event;
1166
1167         EFX_POPULATE_QWORD_2(test_event,
1168                              EV_CODE, DRV_GEN_EV_DECODE,
1169                              EVQ_MAGIC, magic);
1170         falcon_generate_event(channel, &test_event);
1171 }
1172
1173 void falcon_sim_phy_event(struct efx_nic *efx)
1174 {
1175         efx_qword_t phy_event;
1176
1177         EFX_POPULATE_QWORD_1(phy_event, EV_CODE, GLOBAL_EV_DECODE);
1178         if (EFX_IS10G(efx))
1179                 EFX_SET_OWORD_FIELD(phy_event, XG_PHY_INTR, 1);
1180         else
1181                 EFX_SET_OWORD_FIELD(phy_event, G_PHY0_INTR, 1);
1182
1183         falcon_generate_event(&efx->channel[0], &phy_event);
1184 }
1185
1186 /**************************************************************************
1187  *
1188  * Flush handling
1189  *
1190  **************************************************************************/
1191
1192
1193 static void falcon_poll_flush_events(struct efx_nic *efx)
1194 {
1195         struct efx_channel *channel = &efx->channel[0];
1196         struct efx_tx_queue *tx_queue;
1197         struct efx_rx_queue *rx_queue;
1198         unsigned int read_ptr = channel->eventq_read_ptr;
1199         unsigned int end_ptr = (read_ptr - 1) & FALCON_EVQ_MASK;
1200
1201         do {
1202                 efx_qword_t *event = falcon_event(channel, read_ptr);
1203                 int ev_code, ev_sub_code, ev_queue;
1204                 bool ev_failed;
1205
1206                 if (!falcon_event_present(event))
1207                         break;
1208
1209                 ev_code = EFX_QWORD_FIELD(*event, EV_CODE);
1210                 ev_sub_code = EFX_QWORD_FIELD(*event, DRIVER_EV_SUB_CODE);
1211                 if (ev_code == DRIVER_EV_DECODE &&
1212                     ev_sub_code == TX_DESCQ_FLS_DONE_EV_DECODE) {
1213                         ev_queue = EFX_QWORD_FIELD(*event,
1214                                                    DRIVER_EV_TX_DESCQ_ID);
1215                         if (ev_queue < EFX_TX_QUEUE_COUNT) {
1216                                 tx_queue = efx->tx_queue + ev_queue;
1217                                 tx_queue->flushed = true;
1218                         }
1219                 } else if (ev_code == DRIVER_EV_DECODE &&
1220                            ev_sub_code == RX_DESCQ_FLS_DONE_EV_DECODE) {
1221                         ev_queue = EFX_QWORD_FIELD(*event,
1222                                                    DRIVER_EV_RX_DESCQ_ID);
1223                         ev_failed = EFX_QWORD_FIELD(*event,
1224                                                     DRIVER_EV_RX_FLUSH_FAIL);
1225                         if (ev_queue < efx->n_rx_queues) {
1226                                 rx_queue = efx->rx_queue + ev_queue;
1227
1228                                 /* retry the rx flush */
1229                                 if (ev_failed)
1230                                         falcon_flush_rx_queue(rx_queue);
1231                                 else
1232                                         rx_queue->flushed = true;
1233                         }
1234                 }
1235
1236                 read_ptr = (read_ptr + 1) & FALCON_EVQ_MASK;
1237         } while (read_ptr != end_ptr);
1238 }
1239
1240 /* Handle tx and rx flushes at the same time, since they run in
1241  * parallel in the hardware and there's no reason for us to
1242  * serialise them */
1243 int falcon_flush_queues(struct efx_nic *efx)
1244 {
1245         struct efx_rx_queue *rx_queue;
1246         struct efx_tx_queue *tx_queue;
1247         int i;
1248         bool outstanding;
1249
1250         /* Issue flush requests */
1251         efx_for_each_tx_queue(tx_queue, efx) {
1252                 tx_queue->flushed = false;
1253                 falcon_flush_tx_queue(tx_queue);
1254         }
1255         efx_for_each_rx_queue(rx_queue, efx) {
1256                 rx_queue->flushed = false;
1257                 falcon_flush_rx_queue(rx_queue);
1258         }
1259
1260         /* Poll the evq looking for flush completions. Since we're not pushing
1261          * any more rx or tx descriptors at this point, we're in no danger of
1262          * overflowing the evq whilst we wait */
1263         for (i = 0; i < FALCON_FLUSH_POLL_COUNT; ++i) {
1264                 msleep(FALCON_FLUSH_INTERVAL);
1265                 falcon_poll_flush_events(efx);
1266
1267                 /* Check if every queue has been succesfully flushed */
1268                 outstanding = false;
1269                 efx_for_each_tx_queue(tx_queue, efx)
1270                         outstanding |= !tx_queue->flushed;
1271                 efx_for_each_rx_queue(rx_queue, efx)
1272                         outstanding |= !rx_queue->flushed;
1273                 if (!outstanding)
1274                         return 0;
1275         }
1276
1277         /* Mark the queues as all flushed. We're going to return failure
1278          * leading to a reset, or fake up success anyway. "flushed" now
1279          * indicates that we tried to flush. */
1280         efx_for_each_tx_queue(tx_queue, efx) {
1281                 if (!tx_queue->flushed)
1282                         EFX_ERR(efx, "tx queue %d flush command timed out\n",
1283                                 tx_queue->queue);
1284                 tx_queue->flushed = true;
1285         }
1286         efx_for_each_rx_queue(rx_queue, efx) {
1287                 if (!rx_queue->flushed)
1288                         EFX_ERR(efx, "rx queue %d flush command timed out\n",
1289                                 rx_queue->queue);
1290                 rx_queue->flushed = true;
1291         }
1292
1293         if (EFX_WORKAROUND_7803(efx))
1294                 return 0;
1295
1296         return -ETIMEDOUT;
1297 }
1298
1299 /**************************************************************************
1300  *
1301  * Falcon hardware interrupts
1302  * The hardware interrupt handler does very little work; all the event
1303  * queue processing is carried out by per-channel tasklets.
1304  *
1305  **************************************************************************/
1306
1307 /* Enable/disable/generate Falcon interrupts */
1308 static inline void falcon_interrupts(struct efx_nic *efx, int enabled,
1309                                      int force)
1310 {
1311         efx_oword_t int_en_reg_ker;
1312
1313         EFX_POPULATE_OWORD_2(int_en_reg_ker,
1314                              KER_INT_KER, force,
1315                              DRV_INT_EN_KER, enabled);
1316         falcon_write(efx, &int_en_reg_ker, INT_EN_REG_KER);
1317 }
1318
1319 void falcon_enable_interrupts(struct efx_nic *efx)
1320 {
1321         efx_oword_t int_adr_reg_ker;
1322         struct efx_channel *channel;
1323
1324         EFX_ZERO_OWORD(*((efx_oword_t *) efx->irq_status.addr));
1325         wmb(); /* Ensure interrupt vector is clear before interrupts enabled */
1326
1327         /* Program address */
1328         EFX_POPULATE_OWORD_2(int_adr_reg_ker,
1329                              NORM_INT_VEC_DIS_KER, EFX_INT_MODE_USE_MSI(efx),
1330                              INT_ADR_KER, efx->irq_status.dma_addr);
1331         falcon_write(efx, &int_adr_reg_ker, INT_ADR_REG_KER);
1332
1333         /* Enable interrupts */
1334         falcon_interrupts(efx, 1, 0);
1335
1336         /* Force processing of all the channels to get the EVQ RPTRs up to
1337            date */
1338         efx_for_each_channel(channel, efx)
1339                 efx_schedule_channel(channel);
1340 }
1341
1342 void falcon_disable_interrupts(struct efx_nic *efx)
1343 {
1344         /* Disable interrupts */
1345         falcon_interrupts(efx, 0, 0);
1346 }
1347
1348 /* Generate a Falcon test interrupt
1349  * Interrupt must already have been enabled, otherwise nasty things
1350  * may happen.
1351  */
1352 void falcon_generate_interrupt(struct efx_nic *efx)
1353 {
1354         falcon_interrupts(efx, 1, 1);
1355 }
1356
1357 /* Acknowledge a legacy interrupt from Falcon
1358  *
1359  * This acknowledges a legacy (not MSI) interrupt via INT_ACK_KER_REG.
1360  *
1361  * Due to SFC bug 3706 (silicon revision <=A1) reads can be duplicated in the
1362  * BIU. Interrupt acknowledge is read sensitive so must write instead
1363  * (then read to ensure the BIU collector is flushed)
1364  *
1365  * NB most hardware supports MSI interrupts
1366  */
1367 static inline void falcon_irq_ack_a1(struct efx_nic *efx)
1368 {
1369         efx_dword_t reg;
1370
1371         EFX_POPULATE_DWORD_1(reg, INT_ACK_DUMMY_DATA, 0xb7eb7e);
1372         falcon_writel(efx, &reg, INT_ACK_REG_KER_A1);
1373         falcon_readl(efx, &reg, WORK_AROUND_BROKEN_PCI_READS_REG_KER_A1);
1374 }
1375
1376 /* Process a fatal interrupt
1377  * Disable bus mastering ASAP and schedule a reset
1378  */
1379 static irqreturn_t falcon_fatal_interrupt(struct efx_nic *efx)
1380 {
1381         struct falcon_nic_data *nic_data = efx->nic_data;
1382         efx_oword_t *int_ker = efx->irq_status.addr;
1383         efx_oword_t fatal_intr;
1384         int error, mem_perr;
1385
1386         falcon_read(efx, &fatal_intr, FATAL_INTR_REG_KER);
1387         error = EFX_OWORD_FIELD(fatal_intr, INT_KER_ERROR);
1388
1389         EFX_ERR(efx, "SYSTEM ERROR " EFX_OWORD_FMT " status "
1390                 EFX_OWORD_FMT ": %s\n", EFX_OWORD_VAL(*int_ker),
1391                 EFX_OWORD_VAL(fatal_intr),
1392                 error ? "disabling bus mastering" : "no recognised error");
1393         if (error == 0)
1394                 goto out;
1395
1396         /* If this is a memory parity error dump which blocks are offending */
1397         mem_perr = EFX_OWORD_FIELD(fatal_intr, MEM_PERR_INT_KER);
1398         if (mem_perr) {
1399                 efx_oword_t reg;
1400                 falcon_read(efx, &reg, MEM_STAT_REG_KER);
1401                 EFX_ERR(efx, "SYSTEM ERROR: memory parity error "
1402                         EFX_OWORD_FMT "\n", EFX_OWORD_VAL(reg));
1403         }
1404
1405         /* Disable both devices */
1406         pci_clear_master(efx->pci_dev);
1407         if (FALCON_IS_DUAL_FUNC(efx))
1408                 pci_clear_master(nic_data->pci_dev2);
1409         falcon_disable_interrupts(efx);
1410
1411         /* Count errors and reset or disable the NIC accordingly */
1412         if (nic_data->int_error_count == 0 ||
1413             time_after(jiffies, nic_data->int_error_expire)) {
1414                 nic_data->int_error_count = 0;
1415                 nic_data->int_error_expire =
1416                         jiffies + FALCON_INT_ERROR_EXPIRE * HZ;
1417         }
1418         if (++nic_data->int_error_count < FALCON_MAX_INT_ERRORS) {
1419                 EFX_ERR(efx, "SYSTEM ERROR - reset scheduled\n");
1420                 efx_schedule_reset(efx, RESET_TYPE_INT_ERROR);
1421         } else {
1422                 EFX_ERR(efx, "SYSTEM ERROR - max number of errors seen."
1423                         "NIC will be disabled\n");
1424                 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1425         }
1426 out:
1427         return IRQ_HANDLED;
1428 }
1429
1430 /* Handle a legacy interrupt from Falcon
1431  * Acknowledges the interrupt and schedule event queue processing.
1432  */
1433 static irqreturn_t falcon_legacy_interrupt_b0(int irq, void *dev_id)
1434 {
1435         struct efx_nic *efx = dev_id;
1436         efx_oword_t *int_ker = efx->irq_status.addr;
1437         irqreturn_t result = IRQ_NONE;
1438         struct efx_channel *channel;
1439         efx_dword_t reg;
1440         u32 queues;
1441         int syserr;
1442
1443         /* Read the ISR which also ACKs the interrupts */
1444         falcon_readl(efx, &reg, INT_ISR0_B0);
1445         queues = EFX_EXTRACT_DWORD(reg, 0, 31);
1446
1447         /* Check to see if we have a serious error condition */
1448         syserr = EFX_OWORD_FIELD(*int_ker, FATAL_INT);
1449         if (unlikely(syserr))
1450                 return falcon_fatal_interrupt(efx);
1451
1452         /* Schedule processing of any interrupting queues */
1453         efx_for_each_channel(channel, efx) {
1454                 if ((queues & 1) ||
1455                     falcon_event_present(
1456                             falcon_event(channel, channel->eventq_read_ptr))) {
1457                         efx_schedule_channel(channel);
1458                         result = IRQ_HANDLED;
1459                 }
1460                 queues >>= 1;
1461         }
1462
1463         if (result == IRQ_HANDLED) {
1464                 efx->last_irq_cpu = raw_smp_processor_id();
1465                 EFX_TRACE(efx, "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1466                           irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1467         }
1468
1469         return result;
1470 }
1471
1472
1473 static irqreturn_t falcon_legacy_interrupt_a1(int irq, void *dev_id)
1474 {
1475         struct efx_nic *efx = dev_id;
1476         efx_oword_t *int_ker = efx->irq_status.addr;
1477         struct efx_channel *channel;
1478         int syserr;
1479         int queues;
1480
1481         /* Check to see if this is our interrupt.  If it isn't, we
1482          * exit without having touched the hardware.
1483          */
1484         if (unlikely(EFX_OWORD_IS_ZERO(*int_ker))) {
1485                 EFX_TRACE(efx, "IRQ %d on CPU %d not for me\n", irq,
1486                           raw_smp_processor_id());
1487                 return IRQ_NONE;
1488         }
1489         efx->last_irq_cpu = raw_smp_processor_id();
1490         EFX_TRACE(efx, "IRQ %d on CPU %d status " EFX_OWORD_FMT "\n",
1491                   irq, raw_smp_processor_id(), EFX_OWORD_VAL(*int_ker));
1492
1493         /* Check to see if we have a serious error condition */
1494         syserr = EFX_OWORD_FIELD(*int_ker, FATAL_INT);
1495         if (unlikely(syserr))
1496                 return falcon_fatal_interrupt(efx);
1497
1498         /* Determine interrupting queues, clear interrupt status
1499          * register and acknowledge the device interrupt.
1500          */
1501         BUILD_BUG_ON(INT_EVQS_WIDTH > EFX_MAX_CHANNELS);
1502         queues = EFX_OWORD_FIELD(*int_ker, INT_EVQS);
1503         EFX_ZERO_OWORD(*int_ker);
1504         wmb(); /* Ensure the vector is cleared before interrupt ack */
1505         falcon_irq_ack_a1(efx);
1506
1507         /* Schedule processing of any interrupting queues */
1508         channel = &efx->channel[0];
1509         while (queues) {
1510                 if (queues & 0x01)
1511                         efx_schedule_channel(channel);
1512                 channel++;
1513                 queues >>= 1;
1514         }
1515
1516         return IRQ_HANDLED;
1517 }
1518
1519 /* Handle an MSI interrupt from Falcon
1520  *
1521  * Handle an MSI hardware interrupt.  This routine schedules event
1522  * queue processing.  No interrupt acknowledgement cycle is necessary.
1523  * Also, we never need to check that the interrupt is for us, since
1524  * MSI interrupts cannot be shared.
1525  */
1526 static irqreturn_t falcon_msi_interrupt(int irq, void *dev_id)
1527 {
1528         struct efx_channel *channel = dev_id;
1529         struct efx_nic *efx = channel->efx;
1530         efx_oword_t *int_ker = efx->irq_status.addr;
1531         int syserr;
1532
1533         efx->last_irq_cpu = raw_smp_processor_id();
1534         EFX_TRACE(efx, "IRQ %d on CPU %d status " EFX_OWORD_FMT "\n",
1535                   irq, raw_smp_processor_id(), EFX_OWORD_VAL(*int_ker));
1536
1537         /* Check to see if we have a serious error condition */
1538         syserr = EFX_OWORD_FIELD(*int_ker, FATAL_INT);
1539         if (unlikely(syserr))
1540                 return falcon_fatal_interrupt(efx);
1541
1542         /* Schedule processing of the channel */
1543         efx_schedule_channel(channel);
1544
1545         return IRQ_HANDLED;
1546 }
1547
1548
1549 /* Setup RSS indirection table.
1550  * This maps from the hash value of the packet to RXQ
1551  */
1552 static void falcon_setup_rss_indir_table(struct efx_nic *efx)
1553 {
1554         int i = 0;
1555         unsigned long offset;
1556         efx_dword_t dword;
1557
1558         if (falcon_rev(efx) < FALCON_REV_B0)
1559                 return;
1560
1561         for (offset = RX_RSS_INDIR_TBL_B0;
1562              offset < RX_RSS_INDIR_TBL_B0 + 0x800;
1563              offset += 0x10) {
1564                 EFX_POPULATE_DWORD_1(dword, RX_RSS_INDIR_ENT_B0,
1565                                      i % efx->n_rx_queues);
1566                 falcon_writel(efx, &dword, offset);
1567                 i++;
1568         }
1569 }
1570
1571 /* Hook interrupt handler(s)
1572  * Try MSI and then legacy interrupts.
1573  */
1574 int falcon_init_interrupt(struct efx_nic *efx)
1575 {
1576         struct efx_channel *channel;
1577         int rc;
1578
1579         if (!EFX_INT_MODE_USE_MSI(efx)) {
1580                 irq_handler_t handler;
1581                 if (falcon_rev(efx) >= FALCON_REV_B0)
1582                         handler = falcon_legacy_interrupt_b0;
1583                 else
1584                         handler = falcon_legacy_interrupt_a1;
1585
1586                 rc = request_irq(efx->legacy_irq, handler, IRQF_SHARED,
1587                                  efx->name, efx);
1588                 if (rc) {
1589                         EFX_ERR(efx, "failed to hook legacy IRQ %d\n",
1590                                 efx->pci_dev->irq);
1591                         goto fail1;
1592                 }
1593                 return 0;
1594         }
1595
1596         /* Hook MSI or MSI-X interrupt */
1597         efx_for_each_channel(channel, efx) {
1598                 rc = request_irq(channel->irq, falcon_msi_interrupt,
1599                                  IRQF_PROBE_SHARED, /* Not shared */
1600                                  channel->name, channel);
1601                 if (rc) {
1602                         EFX_ERR(efx, "failed to hook IRQ %d\n", channel->irq);
1603                         goto fail2;
1604                 }
1605         }
1606
1607         return 0;
1608
1609  fail2:
1610         efx_for_each_channel(channel, efx)
1611                 free_irq(channel->irq, channel);
1612  fail1:
1613         return rc;
1614 }
1615
1616 void falcon_fini_interrupt(struct efx_nic *efx)
1617 {
1618         struct efx_channel *channel;
1619         efx_oword_t reg;
1620
1621         /* Disable MSI/MSI-X interrupts */
1622         efx_for_each_channel(channel, efx) {
1623                 if (channel->irq)
1624                         free_irq(channel->irq, channel);
1625         }
1626
1627         /* ACK legacy interrupt */
1628         if (falcon_rev(efx) >= FALCON_REV_B0)
1629                 falcon_read(efx, &reg, INT_ISR0_B0);
1630         else
1631                 falcon_irq_ack_a1(efx);
1632
1633         /* Disable legacy interrupt */
1634         if (efx->legacy_irq)
1635                 free_irq(efx->legacy_irq, efx);
1636 }
1637
1638 /**************************************************************************
1639  *
1640  * EEPROM/flash
1641  *
1642  **************************************************************************
1643  */
1644
1645 #define FALCON_SPI_MAX_LEN sizeof(efx_oword_t)
1646
1647 static int falcon_spi_poll(struct efx_nic *efx)
1648 {
1649         efx_oword_t reg;
1650         falcon_read(efx, &reg, EE_SPI_HCMD_REG_KER);
1651         return EFX_OWORD_FIELD(reg, EE_SPI_HCMD_CMD_EN) ? -EBUSY : 0;
1652 }
1653
1654 /* Wait for SPI command completion */
1655 static int falcon_spi_wait(struct efx_nic *efx)
1656 {
1657         /* Most commands will finish quickly, so we start polling at
1658          * very short intervals.  Sometimes the command may have to
1659          * wait for VPD or expansion ROM access outside of our
1660          * control, so we allow up to 100 ms. */
1661         unsigned long timeout = jiffies + 1 + DIV_ROUND_UP(HZ, 10);
1662         int i;
1663
1664         for (i = 0; i < 10; i++) {
1665                 if (!falcon_spi_poll(efx))
1666                         return 0;
1667                 udelay(10);
1668         }
1669
1670         for (;;) {
1671                 if (!falcon_spi_poll(efx))
1672                         return 0;
1673                 if (time_after_eq(jiffies, timeout)) {
1674                         EFX_ERR(efx, "timed out waiting for SPI\n");
1675                         return -ETIMEDOUT;
1676                 }
1677                 schedule_timeout_uninterruptible(1);
1678         }
1679 }
1680
1681 int falcon_spi_cmd(const struct efx_spi_device *spi,
1682                    unsigned int command, int address,
1683                    const void *in, void *out, size_t len)
1684 {
1685         struct efx_nic *efx = spi->efx;
1686         bool addressed = (address >= 0);
1687         bool reading = (out != NULL);
1688         efx_oword_t reg;
1689         int rc;
1690
1691         /* Input validation */
1692         if (len > FALCON_SPI_MAX_LEN)
1693                 return -EINVAL;
1694         BUG_ON(!mutex_is_locked(&efx->spi_lock));
1695
1696         /* Check that previous command is not still running */
1697         rc = falcon_spi_poll(efx);
1698         if (rc)
1699                 return rc;
1700
1701         /* Program address register, if we have an address */
1702         if (addressed) {
1703                 EFX_POPULATE_OWORD_1(reg, EE_SPI_HADR_ADR, address);
1704                 falcon_write(efx, &reg, EE_SPI_HADR_REG_KER);
1705         }
1706
1707         /* Program data register, if we have data */
1708         if (in != NULL) {
1709                 memcpy(&reg, in, len);
1710                 falcon_write(efx, &reg, EE_SPI_HDATA_REG_KER);
1711         }
1712
1713         /* Issue read/write command */
1714         EFX_POPULATE_OWORD_7(reg,
1715                              EE_SPI_HCMD_CMD_EN, 1,
1716                              EE_SPI_HCMD_SF_SEL, spi->device_id,
1717                              EE_SPI_HCMD_DABCNT, len,
1718                              EE_SPI_HCMD_READ, reading,
1719                              EE_SPI_HCMD_DUBCNT, 0,
1720                              EE_SPI_HCMD_ADBCNT,
1721                              (addressed ? spi->addr_len : 0),
1722                              EE_SPI_HCMD_ENC, command);
1723         falcon_write(efx, &reg, EE_SPI_HCMD_REG_KER);
1724
1725         /* Wait for read/write to complete */
1726         rc = falcon_spi_wait(efx);
1727         if (rc)
1728                 return rc;
1729
1730         /* Read data */
1731         if (out != NULL) {
1732                 falcon_read(efx, &reg, EE_SPI_HDATA_REG_KER);
1733                 memcpy(out, &reg, len);
1734         }
1735
1736         return 0;
1737 }
1738
1739 static size_t
1740 falcon_spi_write_limit(const struct efx_spi_device *spi, size_t start)
1741 {
1742         return min(FALCON_SPI_MAX_LEN,
1743                    (spi->block_size - (start & (spi->block_size - 1))));
1744 }
1745
1746 static inline u8
1747 efx_spi_munge_command(const struct efx_spi_device *spi,
1748                       const u8 command, const unsigned int address)
1749 {
1750         return command | (((address >> 8) & spi->munge_address) << 3);
1751 }
1752
1753 /* Wait up to 10 ms for buffered write completion */
1754 int falcon_spi_wait_write(const struct efx_spi_device *spi)
1755 {
1756         struct efx_nic *efx = spi->efx;
1757         unsigned long timeout = jiffies + 1 + DIV_ROUND_UP(HZ, 100);
1758         u8 status;
1759         int rc;
1760
1761         for (;;) {
1762                 rc = falcon_spi_cmd(spi, SPI_RDSR, -1, NULL,
1763                                     &status, sizeof(status));
1764                 if (rc)
1765                         return rc;
1766                 if (!(status & SPI_STATUS_NRDY))
1767                         return 0;
1768                 if (time_after_eq(jiffies, timeout)) {
1769                         EFX_ERR(efx, "SPI write timeout on device %d"
1770                                 " last status=0x%02x\n",
1771                                 spi->device_id, status);
1772                         return -ETIMEDOUT;
1773                 }
1774                 schedule_timeout_uninterruptible(1);
1775         }
1776 }
1777
1778 int falcon_spi_read(const struct efx_spi_device *spi, loff_t start,
1779                     size_t len, size_t *retlen, u8 *buffer)
1780 {
1781         size_t block_len, pos = 0;
1782         unsigned int command;
1783         int rc = 0;
1784
1785         while (pos < len) {
1786                 block_len = min(len - pos, FALCON_SPI_MAX_LEN);
1787
1788                 command = efx_spi_munge_command(spi, SPI_READ, start + pos);
1789                 rc = falcon_spi_cmd(spi, command, start + pos, NULL,
1790                                     buffer + pos, block_len);
1791                 if (rc)
1792                         break;
1793                 pos += block_len;
1794
1795                 /* Avoid locking up the system */
1796                 cond_resched();
1797                 if (signal_pending(current)) {
1798                         rc = -EINTR;
1799                         break;
1800                 }
1801         }
1802
1803         if (retlen)
1804                 *retlen = pos;
1805         return rc;
1806 }
1807
1808 int falcon_spi_write(const struct efx_spi_device *spi, loff_t start,
1809                      size_t len, size_t *retlen, const u8 *buffer)
1810 {
1811         u8 verify_buffer[FALCON_SPI_MAX_LEN];
1812         size_t block_len, pos = 0;
1813         unsigned int command;
1814         int rc = 0;
1815
1816         while (pos < len) {
1817                 rc = falcon_spi_cmd(spi, SPI_WREN, -1, NULL, NULL, 0);
1818                 if (rc)
1819                         break;
1820
1821                 block_len = min(len - pos,
1822                                 falcon_spi_write_limit(spi, start + pos));
1823                 command = efx_spi_munge_command(spi, SPI_WRITE, start + pos);
1824                 rc = falcon_spi_cmd(spi, command, start + pos,
1825                                     buffer + pos, NULL, block_len);
1826                 if (rc)
1827                         break;
1828
1829                 rc = falcon_spi_wait_write(spi);
1830                 if (rc)
1831                         break;
1832
1833                 command = efx_spi_munge_command(spi, SPI_READ, start + pos);
1834                 rc = falcon_spi_cmd(spi, command, start + pos,
1835                                     NULL, verify_buffer, block_len);
1836                 if (memcmp(verify_buffer, buffer + pos, block_len)) {
1837                         rc = -EIO;
1838                         break;
1839                 }
1840
1841                 pos += block_len;
1842
1843                 /* Avoid locking up the system */
1844                 cond_resched();
1845                 if (signal_pending(current)) {
1846                         rc = -EINTR;
1847                         break;
1848                 }
1849         }
1850
1851         if (retlen)
1852                 *retlen = pos;
1853         return rc;
1854 }
1855
1856 /**************************************************************************
1857  *
1858  * MAC wrapper
1859  *
1860  **************************************************************************
1861  */
1862
1863 static int falcon_reset_macs(struct efx_nic *efx)
1864 {
1865         efx_oword_t reg;
1866         int count;
1867
1868         if (falcon_rev(efx) < FALCON_REV_B0) {
1869                 /* It's not safe to use GLB_CTL_REG to reset the
1870                  * macs, so instead use the internal MAC resets
1871                  */
1872                 if (!EFX_IS10G(efx)) {
1873                         EFX_POPULATE_OWORD_1(reg, GM_SW_RST, 1);
1874                         falcon_write(efx, &reg, GM_CFG1_REG);
1875                         udelay(1000);
1876
1877                         EFX_POPULATE_OWORD_1(reg, GM_SW_RST, 0);
1878                         falcon_write(efx, &reg, GM_CFG1_REG);
1879                         udelay(1000);
1880                         return 0;
1881                 } else {
1882                         EFX_POPULATE_OWORD_1(reg, XM_CORE_RST, 1);
1883                         falcon_write(efx, &reg, XM_GLB_CFG_REG);
1884
1885                         for (count = 0; count < 10000; count++) {
1886                                 falcon_read(efx, &reg, XM_GLB_CFG_REG);
1887                                 if (EFX_OWORD_FIELD(reg, XM_CORE_RST) == 0)
1888                                         return 0;
1889                                 udelay(10);
1890                         }
1891
1892                         EFX_ERR(efx, "timed out waiting for XMAC core reset\n");
1893                         return -ETIMEDOUT;
1894                 }
1895         }
1896
1897         /* MAC stats will fail whilst the TX fifo is draining. Serialise
1898          * the drain sequence with the statistics fetch */
1899         efx_stats_disable(efx);
1900
1901         falcon_read(efx, &reg, MAC0_CTRL_REG_KER);
1902         EFX_SET_OWORD_FIELD(reg, TXFIFO_DRAIN_EN_B0, 1);
1903         falcon_write(efx, &reg, MAC0_CTRL_REG_KER);
1904
1905         falcon_read(efx, &reg, GLB_CTL_REG_KER);
1906         EFX_SET_OWORD_FIELD(reg, RST_XGTX, 1);
1907         EFX_SET_OWORD_FIELD(reg, RST_XGRX, 1);
1908         EFX_SET_OWORD_FIELD(reg, RST_EM, 1);
1909         falcon_write(efx, &reg, GLB_CTL_REG_KER);
1910
1911         count = 0;
1912         while (1) {
1913                 falcon_read(efx, &reg, GLB_CTL_REG_KER);
1914                 if (!EFX_OWORD_FIELD(reg, RST_XGTX) &&
1915                     !EFX_OWORD_FIELD(reg, RST_XGRX) &&
1916                     !EFX_OWORD_FIELD(reg, RST_EM)) {
1917                         EFX_LOG(efx, "Completed MAC reset after %d loops\n",
1918                                 count);
1919                         break;
1920                 }
1921                 if (count > 20) {
1922                         EFX_ERR(efx, "MAC reset failed\n");
1923                         break;
1924                 }
1925                 count++;
1926                 udelay(10);
1927         }
1928
1929         efx_stats_enable(efx);
1930
1931         /* If we've reset the EM block and the link is up, then
1932          * we'll have to kick the XAUI link so the PHY can recover */
1933         if (efx->link_up && EFX_IS10G(efx) && EFX_WORKAROUND_5147(efx))
1934                 falcon_reset_xaui(efx);
1935
1936         return 0;
1937 }
1938
1939 void falcon_drain_tx_fifo(struct efx_nic *efx)
1940 {
1941         efx_oword_t reg;
1942
1943         if ((falcon_rev(efx) < FALCON_REV_B0) ||
1944             (efx->loopback_mode != LOOPBACK_NONE))
1945                 return;
1946
1947         falcon_read(efx, &reg, MAC0_CTRL_REG_KER);
1948         /* There is no point in draining more than once */
1949         if (EFX_OWORD_FIELD(reg, TXFIFO_DRAIN_EN_B0))
1950                 return;
1951
1952         falcon_reset_macs(efx);
1953 }
1954
1955 void falcon_deconfigure_mac_wrapper(struct efx_nic *efx)
1956 {
1957         efx_oword_t reg;
1958
1959         if (falcon_rev(efx) < FALCON_REV_B0)
1960                 return;
1961
1962         /* Isolate the MAC -> RX */
1963         falcon_read(efx, &reg, RX_CFG_REG_KER);
1964         EFX_SET_OWORD_FIELD(reg, RX_INGR_EN_B0, 0);
1965         falcon_write(efx, &reg, RX_CFG_REG_KER);
1966
1967         if (!efx->link_up)
1968                 falcon_drain_tx_fifo(efx);
1969 }
1970
1971 void falcon_reconfigure_mac_wrapper(struct efx_nic *efx)
1972 {
1973         efx_oword_t reg;
1974         int link_speed;
1975         bool tx_fc;
1976
1977         switch (efx->link_speed) {
1978         case 10000: link_speed = 3; break;
1979         case 1000:  link_speed = 2; break;
1980         case 100:   link_speed = 1; break;
1981         default:    link_speed = 0; break;
1982         }
1983         /* MAC_LINK_STATUS controls MAC backpressure but doesn't work
1984          * as advertised.  Disable to ensure packets are not
1985          * indefinitely held and TX queue can be flushed at any point
1986          * while the link is down. */
1987         EFX_POPULATE_OWORD_5(reg,
1988                              MAC_XOFF_VAL, 0xffff /* max pause time */,
1989                              MAC_BCAD_ACPT, 1,
1990                              MAC_UC_PROM, efx->promiscuous,
1991                              MAC_LINK_STATUS, 1, /* always set */
1992                              MAC_SPEED, link_speed);
1993         /* On B0, MAC backpressure can be disabled and packets get
1994          * discarded. */
1995         if (falcon_rev(efx) >= FALCON_REV_B0) {
1996                 EFX_SET_OWORD_FIELD(reg, TXFIFO_DRAIN_EN_B0,
1997                                     !efx->link_up);
1998         }
1999
2000         falcon_write(efx, &reg, MAC0_CTRL_REG_KER);
2001
2002         /* Restore the multicast hash registers. */
2003         falcon_set_multicast_hash(efx);
2004
2005         /* Transmission of pause frames when RX crosses the threshold is
2006          * covered by RX_XOFF_MAC_EN and XM_TX_CFG_REG:XM_FCNTL.
2007          * Action on receipt of pause frames is controller by XM_DIS_FCNTL */
2008         tx_fc = !!(efx->link_fc & EFX_FC_TX);
2009         falcon_read(efx, &reg, RX_CFG_REG_KER);
2010         EFX_SET_OWORD_FIELD_VER(efx, reg, RX_XOFF_MAC_EN, tx_fc);
2011
2012         /* Unisolate the MAC -> RX */
2013         if (falcon_rev(efx) >= FALCON_REV_B0)
2014                 EFX_SET_OWORD_FIELD(reg, RX_INGR_EN_B0, 1);
2015         falcon_write(efx, &reg, RX_CFG_REG_KER);
2016 }
2017
2018 int falcon_dma_stats(struct efx_nic *efx, unsigned int done_offset)
2019 {
2020         efx_oword_t reg;
2021         u32 *dma_done;
2022         int i;
2023
2024         if (disable_dma_stats)
2025                 return 0;
2026
2027         /* Statistics fetch will fail if the MAC is in TX drain */
2028         if (falcon_rev(efx) >= FALCON_REV_B0) {
2029                 efx_oword_t temp;
2030                 falcon_read(efx, &temp, MAC0_CTRL_REG_KER);
2031                 if (EFX_OWORD_FIELD(temp, TXFIFO_DRAIN_EN_B0))
2032                         return 0;
2033         }
2034
2035         dma_done = (efx->stats_buffer.addr + done_offset);
2036         *dma_done = FALCON_STATS_NOT_DONE;
2037         wmb(); /* ensure done flag is clear */
2038
2039         /* Initiate DMA transfer of stats */
2040         EFX_POPULATE_OWORD_2(reg,
2041                              MAC_STAT_DMA_CMD, 1,
2042                              MAC_STAT_DMA_ADR,
2043                              efx->stats_buffer.dma_addr);
2044         falcon_write(efx, &reg, MAC0_STAT_DMA_REG_KER);
2045
2046         /* Wait for transfer to complete */
2047         for (i = 0; i < 400; i++) {
2048                 if (*(volatile u32 *)dma_done == FALCON_STATS_DONE) {
2049                         rmb(); /* Ensure the stats are valid. */
2050                         return 0;
2051                 }
2052                 udelay(10);
2053         }
2054
2055         EFX_ERR(efx, "timed out waiting for statistics\n");
2056         return -ETIMEDOUT;
2057 }
2058
2059 /**************************************************************************
2060  *
2061  * PHY access via GMII
2062  *
2063  **************************************************************************
2064  */
2065
2066 /* Use the top bit of the MII PHY id to indicate the PHY type
2067  * (1G/10G), with the remaining bits as the actual PHY id.
2068  *
2069  * This allows us to avoid leaking information from the mii_if_info
2070  * structure into other data structures.
2071  */
2072 #define FALCON_PHY_ID_ID_WIDTH  EFX_WIDTH(MD_PRT_DEV_ADR)
2073 #define FALCON_PHY_ID_ID_MASK   ((1 << FALCON_PHY_ID_ID_WIDTH) - 1)
2074 #define FALCON_PHY_ID_WIDTH     (FALCON_PHY_ID_ID_WIDTH + 1)
2075 #define FALCON_PHY_ID_MASK      ((1 << FALCON_PHY_ID_WIDTH) - 1)
2076 #define FALCON_PHY_ID_10G       (1 << (FALCON_PHY_ID_WIDTH - 1))
2077
2078
2079 /* Packing the clause 45 port and device fields into a single value */
2080 #define MD_PRT_ADR_COMP_LBN   (MD_PRT_ADR_LBN - MD_DEV_ADR_LBN)
2081 #define MD_PRT_ADR_COMP_WIDTH  MD_PRT_ADR_WIDTH
2082 #define MD_DEV_ADR_COMP_LBN    0
2083 #define MD_DEV_ADR_COMP_WIDTH  MD_DEV_ADR_WIDTH
2084
2085
2086 /* Wait for GMII access to complete */
2087 static int falcon_gmii_wait(struct efx_nic *efx)
2088 {
2089         efx_dword_t md_stat;
2090         int count;
2091
2092         /* wait upto 50ms - taken max from datasheet */
2093         for (count = 0; count < 5000; count++) {
2094                 falcon_readl(efx, &md_stat, MD_STAT_REG_KER);
2095                 if (EFX_DWORD_FIELD(md_stat, MD_BSY) == 0) {
2096                         if (EFX_DWORD_FIELD(md_stat, MD_LNFL) != 0 ||
2097                             EFX_DWORD_FIELD(md_stat, MD_BSERR) != 0) {
2098                                 EFX_ERR(efx, "error from GMII access "
2099                                         EFX_DWORD_FMT"\n",
2100                                         EFX_DWORD_VAL(md_stat));
2101                                 return -EIO;
2102                         }
2103                         return 0;
2104                 }
2105                 udelay(10);
2106         }
2107         EFX_ERR(efx, "timed out waiting for GMII\n");
2108         return -ETIMEDOUT;
2109 }
2110
2111 /* Writes a GMII register of a PHY connected to Falcon using MDIO. */
2112 static void falcon_mdio_write(struct net_device *net_dev, int phy_id,
2113                               int addr, int value)
2114 {
2115         struct efx_nic *efx = netdev_priv(net_dev);
2116         unsigned int phy_id2 = phy_id & FALCON_PHY_ID_ID_MASK;
2117         efx_oword_t reg;
2118
2119         /* The 'generic' prt/dev packing in mdio_10g.h is conveniently
2120          * chosen so that the only current user, Falcon, can take the
2121          * packed value and use them directly.
2122          * Fail to build if this assumption is broken.
2123          */
2124         BUILD_BUG_ON(FALCON_PHY_ID_10G != MDIO45_XPRT_ID_IS10G);
2125         BUILD_BUG_ON(FALCON_PHY_ID_ID_WIDTH != MDIO45_PRT_DEV_WIDTH);
2126         BUILD_BUG_ON(MD_PRT_ADR_COMP_LBN != MDIO45_PRT_ID_COMP_LBN);
2127         BUILD_BUG_ON(MD_DEV_ADR_COMP_LBN != MDIO45_DEV_ID_COMP_LBN);
2128
2129         if (phy_id2 == PHY_ADDR_INVALID)
2130                 return;
2131
2132         /* See falcon_mdio_read for an explanation. */
2133         if (!(phy_id & FALCON_PHY_ID_10G)) {
2134                 int mmd = ffs(efx->phy_op->mmds) - 1;
2135                 EFX_TRACE(efx, "Fixing erroneous clause22 write\n");
2136                 phy_id2 = mdio_clause45_pack(phy_id2, mmd)
2137                         & FALCON_PHY_ID_ID_MASK;
2138         }
2139
2140         EFX_REGDUMP(efx, "writing GMII %d register %02x with %04x\n", phy_id,
2141                     addr, value);
2142
2143         spin_lock_bh(&efx->phy_lock);
2144
2145         /* Check MII not currently being accessed */
2146         if (falcon_gmii_wait(efx) != 0)
2147                 goto out;
2148
2149         /* Write the address/ID register */
2150         EFX_POPULATE_OWORD_1(reg, MD_PHY_ADR, addr);
2151         falcon_write(efx, &reg, MD_PHY_ADR_REG_KER);
2152
2153         EFX_POPULATE_OWORD_1(reg, MD_PRT_DEV_ADR, phy_id2);
2154         falcon_write(efx, &reg, MD_ID_REG_KER);
2155
2156         /* Write data */
2157         EFX_POPULATE_OWORD_1(reg, MD_TXD, value);
2158         falcon_write(efx, &reg, MD_TXD_REG_KER);
2159
2160         EFX_POPULATE_OWORD_2(reg,
2161                              MD_WRC, 1,
2162                              MD_GC, 0);
2163         falcon_write(efx, &reg, MD_CS_REG_KER);
2164
2165         /* Wait for data to be written */
2166         if (falcon_gmii_wait(efx) != 0) {
2167                 /* Abort the write operation */
2168                 EFX_POPULATE_OWORD_2(reg,
2169                                      MD_WRC, 0,
2170                                      MD_GC, 1);
2171                 falcon_write(efx, &reg, MD_CS_REG_KER);
2172                 udelay(10);
2173         }
2174
2175  out:
2176         spin_unlock_bh(&efx->phy_lock);
2177 }
2178
2179 /* Reads a GMII register from a PHY connected to Falcon.  If no value
2180  * could be read, -1 will be returned. */
2181 static int falcon_mdio_read(struct net_device *net_dev, int phy_id, int addr)
2182 {
2183         struct efx_nic *efx = netdev_priv(net_dev);
2184         unsigned int phy_addr = phy_id & FALCON_PHY_ID_ID_MASK;
2185         efx_oword_t reg;
2186         int value = -1;
2187
2188         if (phy_addr == PHY_ADDR_INVALID)
2189                 return -1;
2190
2191         /* Our PHY code knows whether it needs to talk clause 22(1G) or 45(10G)
2192          * but the generic Linux code does not make any distinction or have
2193          * any state for this.
2194          * We spot the case where someone tried to talk 22 to a 45 PHY and
2195          * redirect the request to the lowest numbered MMD as a clause45
2196          * request. This is enough to allow simple queries like id and link
2197          * state to succeed. TODO: We may need to do more in future.
2198          */
2199         if (!(phy_id & FALCON_PHY_ID_10G)) {
2200                 int mmd = ffs(efx->phy_op->mmds) - 1;
2201                 EFX_TRACE(efx, "Fixing erroneous clause22 read\n");
2202                 phy_addr = mdio_clause45_pack(phy_addr, mmd)
2203                         & FALCON_PHY_ID_ID_MASK;
2204         }
2205
2206         spin_lock_bh(&efx->phy_lock);
2207
2208         /* Check MII not currently being accessed */
2209         if (falcon_gmii_wait(efx) != 0)
2210                 goto out;
2211
2212         EFX_POPULATE_OWORD_1(reg, MD_PHY_ADR, addr);
2213         falcon_write(efx, &reg, MD_PHY_ADR_REG_KER);
2214
2215         EFX_POPULATE_OWORD_1(reg, MD_PRT_DEV_ADR, phy_addr);
2216         falcon_write(efx, &reg, MD_ID_REG_KER);
2217
2218         /* Request data to be read */
2219         EFX_POPULATE_OWORD_2(reg, MD_RDC, 1, MD_GC, 0);
2220         falcon_write(efx, &reg, MD_CS_REG_KER);
2221
2222         /* Wait for data to become available */
2223         value = falcon_gmii_wait(efx);
2224         if (value == 0) {
2225                 falcon_read(efx, &reg, MD_RXD_REG_KER);
2226                 value = EFX_OWORD_FIELD(reg, MD_RXD);
2227                 EFX_REGDUMP(efx, "read from GMII %d register %02x, got %04x\n",
2228                             phy_id, addr, value);
2229         } else {
2230                 /* Abort the read operation */
2231                 EFX_POPULATE_OWORD_2(reg,
2232                                      MD_RIC, 0,
2233                                      MD_GC, 1);
2234                 falcon_write(efx, &reg, MD_CS_REG_KER);
2235
2236                 EFX_LOG(efx, "read from GMII 0x%x register %02x, got "
2237                         "error %d\n", phy_id, addr, value);
2238         }
2239
2240  out:
2241         spin_unlock_bh(&efx->phy_lock);
2242
2243         return value;
2244 }
2245
2246 static void falcon_init_mdio(struct mii_if_info *gmii)
2247 {
2248         gmii->mdio_read = falcon_mdio_read;
2249         gmii->mdio_write = falcon_mdio_write;
2250         gmii->phy_id_mask = FALCON_PHY_ID_MASK;
2251         gmii->reg_num_mask = ((1 << EFX_WIDTH(MD_PHY_ADR)) - 1);
2252 }
2253
2254 static int falcon_probe_phy(struct efx_nic *efx)
2255 {
2256         switch (efx->phy_type) {
2257         case PHY_TYPE_SFX7101:
2258                 efx->phy_op = &falcon_sfx7101_phy_ops;
2259                 break;
2260         case PHY_TYPE_SFT9001A:
2261         case PHY_TYPE_SFT9001B:
2262                 efx->phy_op = &falcon_sft9001_phy_ops;
2263                 break;
2264         case PHY_TYPE_QT2022C2:
2265         case PHY_TYPE_QT2025C:
2266                 efx->phy_op = &falcon_xfp_phy_ops;
2267                 break;
2268         default:
2269                 EFX_ERR(efx, "Unknown PHY type %d\n",
2270                         efx->phy_type);
2271                 return -1;
2272         }
2273
2274         if (efx->phy_op->macs & EFX_XMAC)
2275                 efx->loopback_modes |= ((1 << LOOPBACK_XGMII) |
2276                                         (1 << LOOPBACK_XGXS) |
2277                                         (1 << LOOPBACK_XAUI));
2278         if (efx->phy_op->macs & EFX_GMAC)
2279                 efx->loopback_modes |= (1 << LOOPBACK_GMAC);
2280         efx->loopback_modes |= efx->phy_op->loopbacks;
2281
2282         return 0;
2283 }
2284
2285 int falcon_switch_mac(struct efx_nic *efx)
2286 {
2287         struct efx_mac_operations *old_mac_op = efx->mac_op;
2288         efx_oword_t nic_stat;
2289         unsigned strap_val;
2290         int rc = 0;
2291
2292         /* Don't try to fetch MAC stats while we're switching MACs */
2293         efx_stats_disable(efx);
2294
2295         /* Internal loopbacks override the phy speed setting */
2296         if (efx->loopback_mode == LOOPBACK_GMAC) {
2297                 efx->link_speed = 1000;
2298                 efx->link_fd = true;
2299         } else if (LOOPBACK_INTERNAL(efx)) {
2300                 efx->link_speed = 10000;
2301                 efx->link_fd = true;
2302         }
2303
2304         WARN_ON(!mutex_is_locked(&efx->mac_lock));
2305         efx->mac_op = (EFX_IS10G(efx) ?
2306                        &falcon_xmac_operations : &falcon_gmac_operations);
2307
2308         /* Always push the NIC_STAT_REG setting even if the mac hasn't
2309          * changed, because this function is run post online reset */
2310         falcon_read(efx, &nic_stat, NIC_STAT_REG);
2311         strap_val = EFX_IS10G(efx) ? 5 : 3;
2312         if (falcon_rev(efx) >= FALCON_REV_B0) {
2313                 EFX_SET_OWORD_FIELD(nic_stat, EE_STRAP_EN, 1);
2314                 EFX_SET_OWORD_FIELD(nic_stat, EE_STRAP_OVR, strap_val);
2315                 falcon_write(efx, &nic_stat, NIC_STAT_REG);
2316         } else {
2317                 /* Falcon A1 does not support 1G/10G speed switching
2318                  * and must not be used with a PHY that does. */
2319                 BUG_ON(EFX_OWORD_FIELD(nic_stat, STRAP_PINS) != strap_val);
2320         }
2321
2322         if (old_mac_op == efx->mac_op)
2323                 goto out;
2324
2325         EFX_LOG(efx, "selected %cMAC\n", EFX_IS10G(efx) ? 'X' : 'G');
2326         /* Not all macs support a mac-level link state */
2327         efx->mac_up = true;
2328
2329         rc = falcon_reset_macs(efx);
2330 out:
2331         efx_stats_enable(efx);
2332         return rc;
2333 }
2334
2335 /* This call is responsible for hooking in the MAC and PHY operations */
2336 int falcon_probe_port(struct efx_nic *efx)
2337 {
2338         int rc;
2339
2340         /* Hook in PHY operations table */
2341         rc = falcon_probe_phy(efx);
2342         if (rc)
2343                 return rc;
2344
2345         /* Set up GMII structure for PHY */
2346         efx->mii.supports_gmii = true;
2347         falcon_init_mdio(&efx->mii);
2348
2349         /* Hardware flow ctrl. FalconA RX FIFO too small for pause generation */
2350         if (falcon_rev(efx) >= FALCON_REV_B0)
2351                 efx->wanted_fc = EFX_FC_RX | EFX_FC_TX;
2352         else
2353                 efx->wanted_fc = EFX_FC_RX;
2354
2355         /* Allocate buffer for stats */
2356         rc = falcon_alloc_buffer(efx, &efx->stats_buffer,
2357                                  FALCON_MAC_STATS_SIZE);
2358         if (rc)
2359                 return rc;
2360         EFX_LOG(efx, "stats buffer at %llx (virt %p phys %llx)\n",
2361                 (u64)efx->stats_buffer.dma_addr,
2362                 efx->stats_buffer.addr,
2363                 (u64)virt_to_phys(efx->stats_buffer.addr));
2364
2365         return 0;
2366 }
2367
2368 void falcon_remove_port(struct efx_nic *efx)
2369 {
2370         falcon_free_buffer(efx, &efx->stats_buffer);
2371 }
2372
2373 /**************************************************************************
2374  *
2375  * Multicast filtering
2376  *
2377  **************************************************************************
2378  */
2379
2380 void falcon_set_multicast_hash(struct efx_nic *efx)
2381 {
2382         union efx_multicast_hash *mc_hash = &efx->multicast_hash;
2383
2384         /* Broadcast packets go through the multicast hash filter.
2385          * ether_crc_le() of the broadcast address is 0xbe2612ff
2386          * so we always add bit 0xff to the mask.
2387          */
2388         set_bit_le(0xff, mc_hash->byte);
2389
2390         falcon_write(efx, &mc_hash->oword[0], MAC_MCAST_HASH_REG0_KER);
2391         falcon_write(efx, &mc_hash->oword[1], MAC_MCAST_HASH_REG1_KER);
2392 }
2393
2394
2395 /**************************************************************************
2396  *
2397  * Falcon test code
2398  *
2399  **************************************************************************/
2400
2401 int falcon_read_nvram(struct efx_nic *efx, struct falcon_nvconfig *nvconfig_out)
2402 {
2403         struct falcon_nvconfig *nvconfig;
2404         struct efx_spi_device *spi;
2405         void *region;
2406         int rc, magic_num, struct_ver;
2407         __le16 *word, *limit;
2408         u32 csum;
2409
2410         spi = efx->spi_flash ? efx->spi_flash : efx->spi_eeprom;
2411         if (!spi)
2412                 return -EINVAL;
2413
2414         region = kmalloc(FALCON_NVCONFIG_END, GFP_KERNEL);
2415         if (!region)
2416                 return -ENOMEM;
2417         nvconfig = region + NVCONFIG_OFFSET;
2418
2419         mutex_lock(&efx->spi_lock);
2420         rc = falcon_spi_read(spi, 0, FALCON_NVCONFIG_END, NULL, region);
2421         mutex_unlock(&efx->spi_lock);
2422         if (rc) {
2423                 EFX_ERR(efx, "Failed to read %s\n",
2424                         efx->spi_flash ? "flash" : "EEPROM");
2425                 rc = -EIO;
2426                 goto out;
2427         }
2428
2429         magic_num = le16_to_cpu(nvconfig->board_magic_num);
2430         struct_ver = le16_to_cpu(nvconfig->board_struct_ver);
2431
2432         rc = -EINVAL;
2433         if (magic_num != NVCONFIG_BOARD_MAGIC_NUM) {
2434                 EFX_ERR(efx, "NVRAM bad magic 0x%x\n", magic_num);
2435                 goto out;
2436         }
2437         if (struct_ver < 2) {
2438                 EFX_ERR(efx, "NVRAM has ancient version 0x%x\n", struct_ver);
2439                 goto out;
2440         } else if (struct_ver < 4) {
2441                 word = &nvconfig->board_magic_num;
2442                 limit = (__le16 *) (nvconfig + 1);
2443         } else {
2444                 word = region;
2445                 limit = region + FALCON_NVCONFIG_END;
2446         }
2447         for (csum = 0; word < limit; ++word)
2448                 csum += le16_to_cpu(*word);
2449
2450         if (~csum & 0xffff) {
2451                 EFX_ERR(efx, "NVRAM has incorrect checksum\n");
2452                 goto out;
2453         }
2454
2455         rc = 0;
2456         if (nvconfig_out)
2457                 memcpy(nvconfig_out, nvconfig, sizeof(*nvconfig));
2458
2459  out:
2460         kfree(region);
2461         return rc;
2462 }
2463
2464 /* Registers tested in the falcon register test */
2465 static struct {
2466         unsigned address;
2467         efx_oword_t mask;
2468 } efx_test_registers[] = {
2469         { ADR_REGION_REG_KER,
2470           EFX_OWORD32(0x0001FFFF, 0x0001FFFF, 0x0001FFFF, 0x0001FFFF) },
2471         { RX_CFG_REG_KER,
2472           EFX_OWORD32(0xFFFFFFFE, 0x00017FFF, 0x00000000, 0x00000000) },
2473         { TX_CFG_REG_KER,
2474           EFX_OWORD32(0x7FFF0037, 0x00000000, 0x00000000, 0x00000000) },
2475         { TX_CFG2_REG_KER,
2476           EFX_OWORD32(0xFFFEFE80, 0x1FFFFFFF, 0x020000FE, 0x007FFFFF) },
2477         { MAC0_CTRL_REG_KER,
2478           EFX_OWORD32(0xFFFF0000, 0x00000000, 0x00000000, 0x00000000) },
2479         { SRM_TX_DC_CFG_REG_KER,
2480           EFX_OWORD32(0x001FFFFF, 0x00000000, 0x00000000, 0x00000000) },
2481         { RX_DC_CFG_REG_KER,
2482           EFX_OWORD32(0x0000000F, 0x00000000, 0x00000000, 0x00000000) },
2483         { RX_DC_PF_WM_REG_KER,
2484           EFX_OWORD32(0x000003FF, 0x00000000, 0x00000000, 0x00000000) },
2485         { DP_CTRL_REG,
2486           EFX_OWORD32(0x00000FFF, 0x00000000, 0x00000000, 0x00000000) },
2487         { GM_CFG2_REG,
2488           EFX_OWORD32(0x00007337, 0x00000000, 0x00000000, 0x00000000) },
2489         { GMF_CFG0_REG,
2490           EFX_OWORD32(0x00001F1F, 0x00000000, 0x00000000, 0x00000000) },
2491         { XM_GLB_CFG_REG,
2492           EFX_OWORD32(0x00000C68, 0x00000000, 0x00000000, 0x00000000) },
2493         { XM_TX_CFG_REG,
2494           EFX_OWORD32(0x00080164, 0x00000000, 0x00000000, 0x00000000) },
2495         { XM_RX_CFG_REG,
2496           EFX_OWORD32(0x07100A0C, 0x00000000, 0x00000000, 0x00000000) },
2497         { XM_RX_PARAM_REG,
2498           EFX_OWORD32(0x00001FF8, 0x00000000, 0x00000000, 0x00000000) },
2499         { XM_FC_REG,
2500           EFX_OWORD32(0xFFFF0001, 0x00000000, 0x00000000, 0x00000000) },
2501         { XM_ADR_LO_REG,
2502           EFX_OWORD32(0xFFFFFFFF, 0x00000000, 0x00000000, 0x00000000) },
2503         { XX_SD_CTL_REG,
2504           EFX_OWORD32(0x0003FF0F, 0x00000000, 0x00000000, 0x00000000) },
2505 };
2506
2507 static bool efx_masked_compare_oword(const efx_oword_t *a, const efx_oword_t *b,
2508                                      const efx_oword_t *mask)
2509 {
2510         return ((a->u64[0] ^ b->u64[0]) & mask->u64[0]) ||
2511                 ((a->u64[1] ^ b->u64[1]) & mask->u64[1]);
2512 }
2513
2514 int falcon_test_registers(struct efx_nic *efx)
2515 {
2516         unsigned address = 0, i, j;
2517         efx_oword_t mask, imask, original, reg, buf;
2518
2519         /* Falcon should be in loopback to isolate the XMAC from the PHY */
2520         WARN_ON(!LOOPBACK_INTERNAL(efx));
2521
2522         for (i = 0; i < ARRAY_SIZE(efx_test_registers); ++i) {
2523                 address = efx_test_registers[i].address;
2524                 mask = imask = efx_test_registers[i].mask;
2525                 EFX_INVERT_OWORD(imask);
2526
2527                 falcon_read(efx, &original, address);
2528
2529                 /* bit sweep on and off */
2530                 for (j = 0; j < 128; j++) {
2531                         if (!EFX_EXTRACT_OWORD32(mask, j, j))
2532                                 continue;
2533
2534                         /* Test this testable bit can be set in isolation */
2535                         EFX_AND_OWORD(reg, original, mask);
2536                         EFX_SET_OWORD32(reg, j, j, 1);
2537
2538                         falcon_write(efx, &reg, address);
2539                         falcon_read(efx, &buf, address);
2540
2541                         if (efx_masked_compare_oword(&reg, &buf, &mask))
2542                                 goto fail;
2543
2544                         /* Test this testable bit can be cleared in isolation */
2545                         EFX_OR_OWORD(reg, original, mask);
2546                         EFX_SET_OWORD32(reg, j, j, 0);
2547
2548                         falcon_write(efx, &reg, address);
2549                         falcon_read(efx, &buf, address);
2550
2551                         if (efx_masked_compare_oword(&reg, &buf, &mask))
2552                                 goto fail;
2553                 }
2554
2555                 falcon_write(efx, &original, address);
2556         }
2557
2558         return 0;
2559
2560 fail:
2561         EFX_ERR(efx, "wrote "EFX_OWORD_FMT" read "EFX_OWORD_FMT
2562                 " at address 0x%x mask "EFX_OWORD_FMT"\n", EFX_OWORD_VAL(reg),
2563                 EFX_OWORD_VAL(buf), address, EFX_OWORD_VAL(mask));
2564         return -EIO;
2565 }
2566
2567 /**************************************************************************
2568  *
2569  * Device reset
2570  *
2571  **************************************************************************
2572  */
2573
2574 /* Resets NIC to known state.  This routine must be called in process
2575  * context and is allowed to sleep. */
2576 int falcon_reset_hw(struct efx_nic *efx, enum reset_type method)
2577 {
2578         struct falcon_nic_data *nic_data = efx->nic_data;
2579         efx_oword_t glb_ctl_reg_ker;
2580         int rc;
2581
2582         EFX_LOG(efx, "performing hardware reset (%d)\n", method);
2583
2584         /* Initiate device reset */
2585         if (method == RESET_TYPE_WORLD) {
2586                 rc = pci_save_state(efx->pci_dev);
2587                 if (rc) {
2588                         EFX_ERR(efx, "failed to backup PCI state of primary "
2589                                 "function prior to hardware reset\n");
2590                         goto fail1;
2591                 }
2592                 if (FALCON_IS_DUAL_FUNC(efx)) {
2593                         rc = pci_save_state(nic_data->pci_dev2);
2594                         if (rc) {
2595                                 EFX_ERR(efx, "failed to backup PCI state of "
2596                                         "secondary function prior to "
2597                                         "hardware reset\n");
2598                                 goto fail2;
2599                         }
2600                 }
2601
2602                 EFX_POPULATE_OWORD_2(glb_ctl_reg_ker,
2603                                      EXT_PHY_RST_DUR, 0x7,
2604                                      SWRST, 1);
2605         } else {
2606                 int reset_phy = (method == RESET_TYPE_INVISIBLE ?
2607                                  EXCLUDE_FROM_RESET : 0);
2608
2609                 EFX_POPULATE_OWORD_7(glb_ctl_reg_ker,
2610                                      EXT_PHY_RST_CTL, reset_phy,
2611                                      PCIE_CORE_RST_CTL, EXCLUDE_FROM_RESET,
2612                                      PCIE_NSTCK_RST_CTL, EXCLUDE_FROM_RESET,
2613                                      PCIE_SD_RST_CTL, EXCLUDE_FROM_RESET,
2614                                      EE_RST_CTL, EXCLUDE_FROM_RESET,
2615                                      EXT_PHY_RST_DUR, 0x7 /* 10ms */,
2616                                      SWRST, 1);
2617         }
2618         falcon_write(efx, &glb_ctl_reg_ker, GLB_CTL_REG_KER);
2619
2620         EFX_LOG(efx, "waiting for hardware reset\n");
2621         schedule_timeout_uninterruptible(HZ / 20);
2622
2623         /* Restore PCI configuration if needed */
2624         if (method == RESET_TYPE_WORLD) {
2625                 if (FALCON_IS_DUAL_FUNC(efx)) {
2626                         rc = pci_restore_state(nic_data->pci_dev2);
2627                         if (rc) {
2628                                 EFX_ERR(efx, "failed to restore PCI config for "
2629                                         "the secondary function\n");
2630                                 goto fail3;
2631                         }
2632                 }
2633                 rc = pci_restore_state(efx->pci_dev);
2634                 if (rc) {
2635                         EFX_ERR(efx, "failed to restore PCI config for the "
2636                                 "primary function\n");
2637                         goto fail4;
2638                 }
2639                 EFX_LOG(efx, "successfully restored PCI config\n");
2640         }
2641
2642         /* Assert that reset complete */
2643         falcon_read(efx, &glb_ctl_reg_ker, GLB_CTL_REG_KER);
2644         if (EFX_OWORD_FIELD(glb_ctl_reg_ker, SWRST) != 0) {
2645                 rc = -ETIMEDOUT;
2646                 EFX_ERR(efx, "timed out waiting for hardware reset\n");
2647                 goto fail5;
2648         }
2649         EFX_LOG(efx, "hardware reset complete\n");
2650
2651         return 0;
2652
2653         /* pci_save_state() and pci_restore_state() MUST be called in pairs */
2654 fail2:
2655 fail3:
2656         pci_restore_state(efx->pci_dev);
2657 fail1:
2658 fail4:
2659 fail5:
2660         return rc;
2661 }
2662
2663 /* Zeroes out the SRAM contents.  This routine must be called in
2664  * process context and is allowed to sleep.
2665  */
2666 static int falcon_reset_sram(struct efx_nic *efx)
2667 {
2668         efx_oword_t srm_cfg_reg_ker, gpio_cfg_reg_ker;
2669         int count;
2670
2671         /* Set the SRAM wake/sleep GPIO appropriately. */
2672         falcon_read(efx, &gpio_cfg_reg_ker, GPIO_CTL_REG_KER);
2673         EFX_SET_OWORD_FIELD(gpio_cfg_reg_ker, GPIO1_OEN, 1);
2674         EFX_SET_OWORD_FIELD(gpio_cfg_reg_ker, GPIO1_OUT, 1);
2675         falcon_write(efx, &gpio_cfg_reg_ker, GPIO_CTL_REG_KER);
2676
2677         /* Initiate SRAM reset */
2678         EFX_POPULATE_OWORD_2(srm_cfg_reg_ker,
2679                              SRAM_OOB_BT_INIT_EN, 1,
2680                              SRM_NUM_BANKS_AND_BANK_SIZE, 0);
2681         falcon_write(efx, &srm_cfg_reg_ker, SRM_CFG_REG_KER);
2682
2683         /* Wait for SRAM reset to complete */
2684         count = 0;
2685         do {
2686                 EFX_LOG(efx, "waiting for SRAM reset (attempt %d)...\n", count);
2687
2688                 /* SRAM reset is slow; expect around 16ms */
2689                 schedule_timeout_uninterruptible(HZ / 50);
2690
2691                 /* Check for reset complete */
2692                 falcon_read(efx, &srm_cfg_reg_ker, SRM_CFG_REG_KER);
2693                 if (!EFX_OWORD_FIELD(srm_cfg_reg_ker, SRAM_OOB_BT_INIT_EN)) {
2694                         EFX_LOG(efx, "SRAM reset complete\n");
2695
2696                         return 0;
2697                 }
2698         } while (++count < 20); /* wait upto 0.4 sec */
2699
2700         EFX_ERR(efx, "timed out waiting for SRAM reset\n");
2701         return -ETIMEDOUT;
2702 }
2703
2704 static int falcon_spi_device_init(struct efx_nic *efx,
2705                                   struct efx_spi_device **spi_device_ret,
2706                                   unsigned int device_id, u32 device_type)
2707 {
2708         struct efx_spi_device *spi_device;
2709
2710         if (device_type != 0) {
2711                 spi_device = kzalloc(sizeof(*spi_device), GFP_KERNEL);
2712                 if (!spi_device)
2713                         return -ENOMEM;
2714                 spi_device->device_id = device_id;
2715                 spi_device->size =
2716                         1 << SPI_DEV_TYPE_FIELD(device_type, SPI_DEV_TYPE_SIZE);
2717                 spi_device->addr_len =
2718                         SPI_DEV_TYPE_FIELD(device_type, SPI_DEV_TYPE_ADDR_LEN);
2719                 spi_device->munge_address = (spi_device->size == 1 << 9 &&
2720                                              spi_device->addr_len == 1);
2721                 spi_device->erase_command =
2722                         SPI_DEV_TYPE_FIELD(device_type, SPI_DEV_TYPE_ERASE_CMD);
2723                 spi_device->erase_size =
2724                         1 << SPI_DEV_TYPE_FIELD(device_type,
2725                                                 SPI_DEV_TYPE_ERASE_SIZE);
2726                 spi_device->block_size =
2727                         1 << SPI_DEV_TYPE_FIELD(device_type,
2728                                                 SPI_DEV_TYPE_BLOCK_SIZE);
2729
2730                 spi_device->efx = efx;
2731         } else {
2732                 spi_device = NULL;
2733         }
2734
2735         kfree(*spi_device_ret);
2736         *spi_device_ret = spi_device;
2737         return 0;
2738 }
2739
2740
2741 static void falcon_remove_spi_devices(struct efx_nic *efx)
2742 {
2743         kfree(efx->spi_eeprom);
2744         efx->spi_eeprom = NULL;
2745         kfree(efx->spi_flash);
2746         efx->spi_flash = NULL;
2747 }
2748
2749 /* Extract non-volatile configuration */
2750 static int falcon_probe_nvconfig(struct efx_nic *efx)
2751 {
2752         struct falcon_nvconfig *nvconfig;
2753         int board_rev;
2754         int rc;
2755
2756         nvconfig = kmalloc(sizeof(*nvconfig), GFP_KERNEL);
2757         if (!nvconfig)
2758                 return -ENOMEM;
2759
2760         rc = falcon_read_nvram(efx, nvconfig);
2761         if (rc == -EINVAL) {
2762                 EFX_ERR(efx, "NVRAM is invalid therefore using defaults\n");
2763                 efx->phy_type = PHY_TYPE_NONE;
2764                 efx->mii.phy_id = PHY_ADDR_INVALID;
2765                 board_rev = 0;
2766                 rc = 0;
2767         } else if (rc) {
2768                 goto fail1;
2769         } else {
2770                 struct falcon_nvconfig_board_v2 *v2 = &nvconfig->board_v2;
2771                 struct falcon_nvconfig_board_v3 *v3 = &nvconfig->board_v3;
2772
2773                 efx->phy_type = v2->port0_phy_type;
2774                 efx->mii.phy_id = v2->port0_phy_addr;
2775                 board_rev = le16_to_cpu(v2->board_revision);
2776
2777                 if (le16_to_cpu(nvconfig->board_struct_ver) >= 3) {
2778                         __le32 fl = v3->spi_device_type[EE_SPI_FLASH];
2779                         __le32 ee = v3->spi_device_type[EE_SPI_EEPROM];
2780                         rc = falcon_spi_device_init(efx, &efx->spi_flash,
2781                                                     EE_SPI_FLASH,
2782                                                     le32_to_cpu(fl));
2783                         if (rc)
2784                                 goto fail2;
2785                         rc = falcon_spi_device_init(efx, &efx->spi_eeprom,
2786                                                     EE_SPI_EEPROM,
2787                                                     le32_to_cpu(ee));
2788                         if (rc)
2789                                 goto fail2;
2790                 }
2791         }
2792
2793         /* Read the MAC addresses */
2794         memcpy(efx->mac_address, nvconfig->mac_address[0], ETH_ALEN);
2795
2796         EFX_LOG(efx, "PHY is %d phy_id %d\n", efx->phy_type, efx->mii.phy_id);
2797
2798         efx_set_board_info(efx, board_rev);
2799
2800         kfree(nvconfig);
2801         return 0;
2802
2803  fail2:
2804         falcon_remove_spi_devices(efx);
2805  fail1:
2806         kfree(nvconfig);
2807         return rc;
2808 }
2809
2810 /* Probe the NIC variant (revision, ASIC vs FPGA, function count, port
2811  * count, port speed).  Set workaround and feature flags accordingly.
2812  */
2813 static int falcon_probe_nic_variant(struct efx_nic *efx)
2814 {
2815         efx_oword_t altera_build;
2816         efx_oword_t nic_stat;
2817
2818         falcon_read(efx, &altera_build, ALTERA_BUILD_REG_KER);
2819         if (EFX_OWORD_FIELD(altera_build, VER_ALL)) {
2820                 EFX_ERR(efx, "Falcon FPGA not supported\n");
2821                 return -ENODEV;
2822         }
2823
2824         falcon_read(efx, &nic_stat, NIC_STAT_REG);
2825
2826         switch (falcon_rev(efx)) {
2827         case FALCON_REV_A0:
2828         case 0xff:
2829                 EFX_ERR(efx, "Falcon rev A0 not supported\n");
2830                 return -ENODEV;
2831
2832         case FALCON_REV_A1:
2833                 if (EFX_OWORD_FIELD(nic_stat, STRAP_PCIE) == 0) {
2834                         EFX_ERR(efx, "Falcon rev A1 PCI-X not supported\n");
2835                         return -ENODEV;
2836                 }
2837                 break;
2838
2839         case FALCON_REV_B0:
2840                 break;
2841
2842         default:
2843                 EFX_ERR(efx, "Unknown Falcon rev %d\n", falcon_rev(efx));
2844                 return -ENODEV;
2845         }
2846
2847         /* Initial assumed speed */
2848         efx->link_speed = EFX_OWORD_FIELD(nic_stat, STRAP_10G) ? 10000 : 1000;
2849
2850         return 0;
2851 }
2852
2853 /* Probe all SPI devices on the NIC */
2854 static void falcon_probe_spi_devices(struct efx_nic *efx)
2855 {
2856         efx_oword_t nic_stat, gpio_ctl, ee_vpd_cfg;
2857         int boot_dev;
2858
2859         falcon_read(efx, &gpio_ctl, GPIO_CTL_REG_KER);
2860         falcon_read(efx, &nic_stat, NIC_STAT_REG);
2861         falcon_read(efx, &ee_vpd_cfg, EE_VPD_CFG_REG_KER);
2862
2863         if (EFX_OWORD_FIELD(gpio_ctl, BOOTED_USING_NVDEVICE)) {
2864                 boot_dev = (EFX_OWORD_FIELD(nic_stat, SF_PRST) ?
2865                             EE_SPI_FLASH : EE_SPI_EEPROM);
2866                 EFX_LOG(efx, "Booted from %s\n",
2867                         boot_dev == EE_SPI_FLASH ? "flash" : "EEPROM");
2868         } else {
2869                 /* Disable VPD and set clock dividers to safe
2870                  * values for initial programming. */
2871                 boot_dev = -1;
2872                 EFX_LOG(efx, "Booted from internal ASIC settings;"
2873                         " setting SPI config\n");
2874                 EFX_POPULATE_OWORD_3(ee_vpd_cfg, EE_VPD_EN, 0,
2875                                      /* 125 MHz / 7 ~= 20 MHz */
2876                                      EE_SF_CLOCK_DIV, 7,
2877                                      /* 125 MHz / 63 ~= 2 MHz */
2878                                      EE_EE_CLOCK_DIV, 63);
2879                 falcon_write(efx, &ee_vpd_cfg, EE_VPD_CFG_REG_KER);
2880         }
2881
2882         if (boot_dev == EE_SPI_FLASH)
2883                 falcon_spi_device_init(efx, &efx->spi_flash, EE_SPI_FLASH,
2884                                        default_flash_type);
2885         if (boot_dev == EE_SPI_EEPROM)
2886                 falcon_spi_device_init(efx, &efx->spi_eeprom, EE_SPI_EEPROM,
2887                                        large_eeprom_type);
2888 }
2889
2890 int falcon_probe_nic(struct efx_nic *efx)
2891 {
2892         struct falcon_nic_data *nic_data;
2893         int rc;
2894
2895         /* Allocate storage for hardware specific data */
2896         nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
2897         if (!nic_data)
2898                 return -ENOMEM;
2899         efx->nic_data = nic_data;
2900
2901         /* Determine number of ports etc. */
2902         rc = falcon_probe_nic_variant(efx);
2903         if (rc)
2904                 goto fail1;
2905
2906         /* Probe secondary function if expected */
2907         if (FALCON_IS_DUAL_FUNC(efx)) {
2908                 struct pci_dev *dev = pci_dev_get(efx->pci_dev);
2909
2910                 while ((dev = pci_get_device(EFX_VENDID_SFC, FALCON_A_S_DEVID,
2911                                              dev))) {
2912                         if (dev->bus == efx->pci_dev->bus &&
2913                             dev->devfn == efx->pci_dev->devfn + 1) {
2914                                 nic_data->pci_dev2 = dev;
2915                                 break;
2916                         }
2917                 }
2918                 if (!nic_data->pci_dev2) {
2919                         EFX_ERR(efx, "failed to find secondary function\n");
2920                         rc = -ENODEV;
2921                         goto fail2;
2922                 }
2923         }
2924
2925         /* Now we can reset the NIC */
2926         rc = falcon_reset_hw(efx, RESET_TYPE_ALL);
2927         if (rc) {
2928                 EFX_ERR(efx, "failed to reset NIC\n");
2929                 goto fail3;
2930         }
2931
2932         /* Allocate memory for INT_KER */
2933         rc = falcon_alloc_buffer(efx, &efx->irq_status, sizeof(efx_oword_t));
2934         if (rc)
2935                 goto fail4;
2936         BUG_ON(efx->irq_status.dma_addr & 0x0f);
2937
2938         EFX_LOG(efx, "INT_KER at %llx (virt %p phys %llx)\n",
2939                 (u64)efx->irq_status.dma_addr,
2940                 efx->irq_status.addr, (u64)virt_to_phys(efx->irq_status.addr));
2941
2942         falcon_probe_spi_devices(efx);
2943
2944         /* Read in the non-volatile configuration */
2945         rc = falcon_probe_nvconfig(efx);
2946         if (rc)
2947                 goto fail5;
2948
2949         /* Initialise I2C adapter */
2950         efx->i2c_adap.owner = THIS_MODULE;
2951         nic_data->i2c_data = falcon_i2c_bit_operations;
2952         nic_data->i2c_data.data = efx;
2953         efx->i2c_adap.algo_data = &nic_data->i2c_data;
2954         efx->i2c_adap.dev.parent = &efx->pci_dev->dev;
2955         strlcpy(efx->i2c_adap.name, "SFC4000 GPIO", sizeof(efx->i2c_adap.name));
2956         rc = i2c_bit_add_bus(&efx->i2c_adap);
2957         if (rc)
2958                 goto fail5;
2959
2960         return 0;
2961
2962  fail5:
2963         falcon_remove_spi_devices(efx);
2964         falcon_free_buffer(efx, &efx->irq_status);
2965  fail4:
2966  fail3:
2967         if (nic_data->pci_dev2) {
2968                 pci_dev_put(nic_data->pci_dev2);
2969                 nic_data->pci_dev2 = NULL;
2970         }
2971  fail2:
2972  fail1:
2973         kfree(efx->nic_data);
2974         return rc;
2975 }
2976
2977 /* This call performs hardware-specific global initialisation, such as
2978  * defining the descriptor cache sizes and number of RSS channels.
2979  * It does not set up any buffers, descriptor rings or event queues.
2980  */
2981 int falcon_init_nic(struct efx_nic *efx)
2982 {
2983         efx_oword_t temp;
2984         unsigned thresh;
2985         int rc;
2986
2987         /* Use on-chip SRAM */
2988         falcon_read(efx, &temp, NIC_STAT_REG);
2989         EFX_SET_OWORD_FIELD(temp, ONCHIP_SRAM, 1);
2990         falcon_write(efx, &temp, NIC_STAT_REG);
2991
2992         /* Set the source of the GMAC clock */
2993         if (falcon_rev(efx) == FALCON_REV_B0) {
2994                 falcon_read(efx, &temp, GPIO_CTL_REG_KER);
2995                 EFX_SET_OWORD_FIELD(temp, GPIO_USE_NIC_CLK, true);
2996                 falcon_write(efx, &temp, GPIO_CTL_REG_KER);
2997         }
2998
2999         /* Set buffer table mode */
3000         EFX_POPULATE_OWORD_1(temp, BUF_TBL_MODE, BUF_TBL_MODE_FULL);
3001         falcon_write(efx, &temp, BUF_TBL_CFG_REG_KER);
3002
3003         rc = falcon_reset_sram(efx);
3004         if (rc)
3005                 return rc;
3006
3007         /* Set positions of descriptor caches in SRAM. */
3008         EFX_POPULATE_OWORD_1(temp, SRM_TX_DC_BASE_ADR, TX_DC_BASE / 8);
3009         falcon_write(efx, &temp, SRM_TX_DC_CFG_REG_KER);
3010         EFX_POPULATE_OWORD_1(temp, SRM_RX_DC_BASE_ADR, RX_DC_BASE / 8);
3011         falcon_write(efx, &temp, SRM_RX_DC_CFG_REG_KER);
3012
3013         /* Set TX descriptor cache size. */
3014         BUILD_BUG_ON(TX_DC_ENTRIES != (16 << TX_DC_ENTRIES_ORDER));
3015         EFX_POPULATE_OWORD_1(temp, TX_DC_SIZE, TX_DC_ENTRIES_ORDER);
3016         falcon_write(efx, &temp, TX_DC_CFG_REG_KER);
3017
3018         /* Set RX descriptor cache size.  Set low watermark to size-8, as
3019          * this allows most efficient prefetching.
3020          */
3021         BUILD_BUG_ON(RX_DC_ENTRIES != (16 << RX_DC_ENTRIES_ORDER));
3022         EFX_POPULATE_OWORD_1(temp, RX_DC_SIZE, RX_DC_ENTRIES_ORDER);
3023         falcon_write(efx, &temp, RX_DC_CFG_REG_KER);
3024         EFX_POPULATE_OWORD_1(temp, RX_DC_PF_LWM, RX_DC_ENTRIES - 8);
3025         falcon_write(efx, &temp, RX_DC_PF_WM_REG_KER);
3026
3027         /* Clear the parity enables on the TX data fifos as
3028          * they produce false parity errors because of timing issues
3029          */
3030         if (EFX_WORKAROUND_5129(efx)) {
3031                 falcon_read(efx, &temp, SPARE_REG_KER);
3032                 EFX_SET_OWORD_FIELD(temp, MEM_PERR_EN_TX_DATA, 0);
3033                 falcon_write(efx, &temp, SPARE_REG_KER);
3034         }
3035
3036         /* Enable all the genuinely fatal interrupts.  (They are still
3037          * masked by the overall interrupt mask, controlled by
3038          * falcon_interrupts()).
3039          *
3040          * Note: All other fatal interrupts are enabled
3041          */
3042         EFX_POPULATE_OWORD_3(temp,
3043                              ILL_ADR_INT_KER_EN, 1,
3044                              RBUF_OWN_INT_KER_EN, 1,
3045                              TBUF_OWN_INT_KER_EN, 1);
3046         EFX_INVERT_OWORD(temp);
3047         falcon_write(efx, &temp, FATAL_INTR_REG_KER);
3048
3049         if (EFX_WORKAROUND_7244(efx)) {
3050                 falcon_read(efx, &temp, RX_FILTER_CTL_REG);
3051                 EFX_SET_OWORD_FIELD(temp, UDP_FULL_SRCH_LIMIT, 8);
3052                 EFX_SET_OWORD_FIELD(temp, UDP_WILD_SRCH_LIMIT, 8);
3053                 EFX_SET_OWORD_FIELD(temp, TCP_FULL_SRCH_LIMIT, 8);
3054                 EFX_SET_OWORD_FIELD(temp, TCP_WILD_SRCH_LIMIT, 8);
3055                 falcon_write(efx, &temp, RX_FILTER_CTL_REG);
3056         }
3057
3058         falcon_setup_rss_indir_table(efx);
3059
3060         /* Setup RX.  Wait for descriptor is broken and must
3061          * be disabled.  RXDP recovery shouldn't be needed, but is.
3062          */
3063         falcon_read(efx, &temp, RX_SELF_RST_REG_KER);
3064         EFX_SET_OWORD_FIELD(temp, RX_NODESC_WAIT_DIS, 1);
3065         EFX_SET_OWORD_FIELD(temp, RX_RECOVERY_EN, 1);
3066         if (EFX_WORKAROUND_5583(efx))
3067                 EFX_SET_OWORD_FIELD(temp, RX_ISCSI_DIS, 1);
3068         falcon_write(efx, &temp, RX_SELF_RST_REG_KER);
3069
3070         /* Disable the ugly timer-based TX DMA backoff and allow TX DMA to be
3071          * controlled by the RX FIFO fill level. Set arbitration to one pkt/Q.
3072          */
3073         falcon_read(efx, &temp, TX_CFG2_REG_KER);
3074         EFX_SET_OWORD_FIELD(temp, TX_RX_SPACER, 0xfe);
3075         EFX_SET_OWORD_FIELD(temp, TX_RX_SPACER_EN, 1);
3076         EFX_SET_OWORD_FIELD(temp, TX_ONE_PKT_PER_Q, 1);
3077         EFX_SET_OWORD_FIELD(temp, TX_CSR_PUSH_EN, 0);
3078         EFX_SET_OWORD_FIELD(temp, TX_DIS_NON_IP_EV, 1);
3079         /* Enable SW_EV to inherit in char driver - assume harmless here */
3080         EFX_SET_OWORD_FIELD(temp, TX_SW_EV_EN, 1);
3081         /* Prefetch threshold 2 => fetch when descriptor cache half empty */
3082         EFX_SET_OWORD_FIELD(temp, TX_PREF_THRESHOLD, 2);
3083         /* Squash TX of packets of 16 bytes or less */
3084         if (falcon_rev(efx) >= FALCON_REV_B0 && EFX_WORKAROUND_9141(efx))
3085                 EFX_SET_OWORD_FIELD(temp, TX_FLUSH_MIN_LEN_EN_B0, 1);
3086         falcon_write(efx, &temp, TX_CFG2_REG_KER);
3087
3088         /* Do not enable TX_NO_EOP_DISC_EN, since it limits packets to 16
3089          * descriptors (which is bad).
3090          */
3091         falcon_read(efx, &temp, TX_CFG_REG_KER);
3092         EFX_SET_OWORD_FIELD(temp, TX_NO_EOP_DISC_EN, 0);
3093         falcon_write(efx, &temp, TX_CFG_REG_KER);
3094
3095         /* RX config */
3096         falcon_read(efx, &temp, RX_CFG_REG_KER);
3097         EFX_SET_OWORD_FIELD_VER(efx, temp, RX_DESC_PUSH_EN, 0);
3098         if (EFX_WORKAROUND_7575(efx))
3099                 EFX_SET_OWORD_FIELD_VER(efx, temp, RX_USR_BUF_SIZE,
3100                                         (3 * 4096) / 32);
3101         if (falcon_rev(efx) >= FALCON_REV_B0)
3102                 EFX_SET_OWORD_FIELD(temp, RX_INGR_EN_B0, 1);
3103
3104         /* RX FIFO flow control thresholds */
3105         thresh = ((rx_xon_thresh_bytes >= 0) ?
3106                   rx_xon_thresh_bytes : efx->type->rx_xon_thresh);
3107         EFX_SET_OWORD_FIELD_VER(efx, temp, RX_XON_MAC_TH, thresh / 256);
3108         thresh = ((rx_xoff_thresh_bytes >= 0) ?
3109                   rx_xoff_thresh_bytes : efx->type->rx_xoff_thresh);
3110         EFX_SET_OWORD_FIELD_VER(efx, temp, RX_XOFF_MAC_TH, thresh / 256);
3111         /* RX control FIFO thresholds [32 entries] */
3112         EFX_SET_OWORD_FIELD_VER(efx, temp, RX_XON_TX_TH, 20);
3113         EFX_SET_OWORD_FIELD_VER(efx, temp, RX_XOFF_TX_TH, 25);
3114         falcon_write(efx, &temp, RX_CFG_REG_KER);
3115
3116         /* Set destination of both TX and RX Flush events */
3117         if (falcon_rev(efx) >= FALCON_REV_B0) {
3118                 EFX_POPULATE_OWORD_1(temp, FLS_EVQ_ID, 0);
3119                 falcon_write(efx, &temp, DP_CTRL_REG);
3120         }
3121
3122         return 0;
3123 }
3124
3125 void falcon_remove_nic(struct efx_nic *efx)
3126 {
3127         struct falcon_nic_data *nic_data = efx->nic_data;
3128         int rc;
3129
3130         /* Remove I2C adapter and clear it in preparation for a retry */
3131         rc = i2c_del_adapter(&efx->i2c_adap);
3132         BUG_ON(rc);
3133         memset(&efx->i2c_adap, 0, sizeof(efx->i2c_adap));
3134
3135         falcon_remove_spi_devices(efx);
3136         falcon_free_buffer(efx, &efx->irq_status);
3137
3138         falcon_reset_hw(efx, RESET_TYPE_ALL);
3139
3140         /* Release the second function after the reset */
3141         if (nic_data->pci_dev2) {
3142                 pci_dev_put(nic_data->pci_dev2);
3143                 nic_data->pci_dev2 = NULL;
3144         }
3145
3146         /* Tear down the private nic state */
3147         kfree(efx->nic_data);
3148         efx->nic_data = NULL;
3149 }
3150
3151 void falcon_update_nic_stats(struct efx_nic *efx)
3152 {
3153         efx_oword_t cnt;
3154
3155         falcon_read(efx, &cnt, RX_NODESC_DROP_REG_KER);
3156         efx->n_rx_nodesc_drop_cnt += EFX_OWORD_FIELD(cnt, RX_NODESC_DROP_CNT);
3157 }
3158
3159 /**************************************************************************
3160  *
3161  * Revision-dependent attributes used by efx.c
3162  *
3163  **************************************************************************
3164  */
3165
3166 struct efx_nic_type falcon_a_nic_type = {
3167         .mem_bar = 2,
3168         .mem_map_size = 0x20000,
3169         .txd_ptr_tbl_base = TX_DESC_PTR_TBL_KER_A1,
3170         .rxd_ptr_tbl_base = RX_DESC_PTR_TBL_KER_A1,
3171         .buf_tbl_base = BUF_TBL_KER_A1,
3172         .evq_ptr_tbl_base = EVQ_PTR_TBL_KER_A1,
3173         .evq_rptr_tbl_base = EVQ_RPTR_REG_KER_A1,
3174         .txd_ring_mask = FALCON_TXD_RING_MASK,
3175         .rxd_ring_mask = FALCON_RXD_RING_MASK,
3176         .evq_size = FALCON_EVQ_SIZE,
3177         .max_dma_mask = FALCON_DMA_MASK,
3178         .tx_dma_mask = FALCON_TX_DMA_MASK,
3179         .bug5391_mask = 0xf,
3180         .rx_xoff_thresh = 2048,
3181         .rx_xon_thresh = 512,
3182         .rx_buffer_padding = 0x24,
3183         .max_interrupt_mode = EFX_INT_MODE_MSI,
3184         .phys_addr_channels = 4,
3185 };
3186
3187 struct efx_nic_type falcon_b_nic_type = {
3188         .mem_bar = 2,
3189         /* Map everything up to and including the RSS indirection
3190          * table.  Don't map MSI-X table, MSI-X PBA since Linux
3191          * requires that they not be mapped.  */
3192         .mem_map_size = RX_RSS_INDIR_TBL_B0 + 0x800,
3193         .txd_ptr_tbl_base = TX_DESC_PTR_TBL_KER_B0,
3194         .rxd_ptr_tbl_base = RX_DESC_PTR_TBL_KER_B0,
3195         .buf_tbl_base = BUF_TBL_KER_B0,
3196         .evq_ptr_tbl_base = EVQ_PTR_TBL_KER_B0,
3197         .evq_rptr_tbl_base = EVQ_RPTR_REG_KER_B0,
3198         .txd_ring_mask = FALCON_TXD_RING_MASK,
3199         .rxd_ring_mask = FALCON_RXD_RING_MASK,
3200         .evq_size = FALCON_EVQ_SIZE,
3201         .max_dma_mask = FALCON_DMA_MASK,
3202         .tx_dma_mask = FALCON_TX_DMA_MASK,
3203         .bug5391_mask = 0,
3204         .rx_xoff_thresh = 54272, /* ~80Kb - 3*max MTU */
3205         .rx_xon_thresh = 27648,  /* ~3*max MTU */
3206         .rx_buffer_padding = 0,
3207         .max_interrupt_mode = EFX_INT_MODE_MSIX,
3208         .phys_addr_channels = 32, /* Hardware limit is 64, but the legacy
3209                                    * interrupt handler only supports 32
3210                                    * channels */
3211 };
3212