xref: /linux/drivers/net/ethernet/tehuti/tn40.c (revision a6a6a98094116b60e5523a571d9443c53325f5b1)
1 // SPDX-License-Identifier: GPL-2.0+
2 /* Copyright (c) Tehuti Networks Ltd. */
3 
4 #include <linux/bitfield.h>
5 #include <linux/ethtool.h>
6 #include <linux/firmware.h>
7 #include <linux/if_vlan.h>
8 #include <linux/iopoll.h>
9 #include <linux/netdevice.h>
10 #include <linux/pci.h>
11 #include <linux/phylink.h>
12 #include <linux/vmalloc.h>
13 #include <net/page_pool/helpers.h>
14 
15 #include "tn40.h"
16 
17 #define TN40_SHORT_PACKET_SIZE 60
18 #define TN40_FIRMWARE_NAME "tehuti/bdx.bin"
19 
20 static void tn40_enable_interrupts(struct tn40_priv *priv)
21 {
22 	tn40_write_reg(priv, TN40_REG_IMR, priv->isr_mask);
23 }
24 
25 static void tn40_disable_interrupts(struct tn40_priv *priv)
26 {
27 	tn40_write_reg(priv, TN40_REG_IMR, 0);
28 }
29 
30 static int tn40_fifo_alloc(struct tn40_priv *priv, struct tn40_fifo *f,
31 			   int fsz_type,
32 			   u16 reg_cfg0, u16 reg_cfg1,
33 			   u16 reg_rptr, u16 reg_wptr)
34 {
35 	u16 memsz = TN40_FIFO_SIZE * (1 << fsz_type);
36 	u64 cfg_base;
37 
38 	memset(f, 0, sizeof(struct tn40_fifo));
39 	/* 1K extra space is allocated at the end of the fifo to simplify
40 	 * processing of descriptors that wraps around fifo's end.
41 	 */
42 	f->va = dma_alloc_coherent(&priv->pdev->dev,
43 				   memsz + TN40_FIFO_EXTRA_SPACE, &f->da,
44 				   GFP_KERNEL);
45 	if (!f->va)
46 		return -ENOMEM;
47 
48 	f->reg_cfg0 = reg_cfg0;
49 	f->reg_cfg1 = reg_cfg1;
50 	f->reg_rptr = reg_rptr;
51 	f->reg_wptr = reg_wptr;
52 	f->rptr = 0;
53 	f->wptr = 0;
54 	f->memsz = memsz;
55 	f->size_mask = memsz - 1;
56 	cfg_base = lower_32_bits((f->da & TN40_TX_RX_CFG0_BASE) | fsz_type);
57 	tn40_write_reg(priv, reg_cfg0, cfg_base);
58 	tn40_write_reg(priv, reg_cfg1, upper_32_bits(f->da));
59 	return 0;
60 }
61 
62 static void tn40_fifo_free(struct tn40_priv *priv, struct tn40_fifo *f)
63 {
64 	dma_free_coherent(&priv->pdev->dev,
65 			  f->memsz + TN40_FIFO_EXTRA_SPACE, f->va, f->da);
66 }
67 
68 static struct tn40_rxdb *tn40_rxdb_alloc(int nelem)
69 {
70 	size_t size = sizeof(struct tn40_rxdb) + (nelem * sizeof(int)) +
71 	    (nelem * sizeof(struct tn40_rx_map));
72 	struct tn40_rxdb *db;
73 	int i;
74 
75 	db = vzalloc(size);
76 	if (db) {
77 		db->stack = (int *)(db + 1);
78 		db->elems = (void *)(db->stack + nelem);
79 		db->nelem = nelem;
80 		db->top = nelem;
81 		/* make the first alloc close to db struct */
82 		for (i = 0; i < nelem; i++)
83 			db->stack[i] = nelem - i - 1;
84 	}
85 	return db;
86 }
87 
88 static void tn40_rxdb_free(struct tn40_rxdb *db)
89 {
90 	vfree(db);
91 }
92 
93 static int tn40_rxdb_alloc_elem(struct tn40_rxdb *db)
94 {
95 	return db->stack[--db->top];
96 }
97 
98 static void *tn40_rxdb_addr_elem(struct tn40_rxdb *db, unsigned int n)
99 {
100 	return db->elems + n;
101 }
102 
103 static int tn40_rxdb_available(struct tn40_rxdb *db)
104 {
105 	return db->top;
106 }
107 
108 static void tn40_rxdb_free_elem(struct tn40_rxdb *db, unsigned int n)
109 {
110 	db->stack[db->top++] = n;
111 }
112 
113 /**
114  * tn40_create_rx_ring - Initialize RX all related HW and SW resources
115  * @priv: NIC private structure
116  *
117  * create_rx_ring creates rxf and rxd fifos, updates the relevant HW registers,
118  * preallocates skbs for rx. It assumes that Rx is disabled in HW funcs are
119  * grouped for better cache usage
120  *
121  * RxD fifo is smaller then RxF fifo by design. Upon high load, RxD will be
122  * filled and packets will be dropped by the NIC without getting into the host
123  * or generating interrupts. In this situation the host has no chance of
124  * processing all the packets. Dropping packets by the NIC is cheaper, since it
125  * takes 0 CPU cycles.
126  *
127  * Return: 0 on success and negative value on error.
128  */
129 static int tn40_create_rx_ring(struct tn40_priv *priv)
130 {
131 	struct page_pool_params pp = {
132 		.dev = &priv->pdev->dev,
133 		.napi = &priv->napi,
134 		.dma_dir = DMA_FROM_DEVICE,
135 		.netdev = priv->ndev,
136 		.flags = PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV,
137 		.max_len = PAGE_SIZE,
138 	};
139 	int ret, pkt_size, nr;
140 
141 	priv->page_pool = page_pool_create(&pp);
142 	if (IS_ERR(priv->page_pool))
143 		return PTR_ERR(priv->page_pool);
144 
145 	ret = tn40_fifo_alloc(priv, &priv->rxd_fifo0.m, priv->rxd_size,
146 			      TN40_REG_RXD_CFG0_0, TN40_REG_RXD_CFG1_0,
147 			      TN40_REG_RXD_RPTR_0, TN40_REG_RXD_WPTR_0);
148 	if (ret)
149 		goto err_destroy_page_pool;
150 
151 	ret = tn40_fifo_alloc(priv, &priv->rxf_fifo0.m, priv->rxf_size,
152 			      TN40_REG_RXF_CFG0_0, TN40_REG_RXF_CFG1_0,
153 			      TN40_REG_RXF_RPTR_0, TN40_REG_RXF_WPTR_0);
154 	if (ret)
155 		goto err_free_rxd;
156 
157 	pkt_size = priv->ndev->mtu + VLAN_ETH_HLEN;
158 	priv->rxf_fifo0.m.pktsz = pkt_size;
159 	nr = priv->rxf_fifo0.m.memsz / sizeof(struct tn40_rxf_desc);
160 	priv->rxdb0 = tn40_rxdb_alloc(nr);
161 	if (!priv->rxdb0) {
162 		ret = -ENOMEM;
163 		goto err_free_rxf;
164 	}
165 	return 0;
166 err_free_rxf:
167 	tn40_fifo_free(priv, &priv->rxf_fifo0.m);
168 err_free_rxd:
169 	tn40_fifo_free(priv, &priv->rxd_fifo0.m);
170 err_destroy_page_pool:
171 	page_pool_destroy(priv->page_pool);
172 	return ret;
173 }
174 
175 static void tn40_rx_free_buffers(struct tn40_priv *priv)
176 {
177 	struct tn40_rxdb *db = priv->rxdb0;
178 	struct tn40_rx_map *dm;
179 	u16 i;
180 
181 	netdev_dbg(priv->ndev, "total =%d free =%d busy =%d\n", db->nelem,
182 		   tn40_rxdb_available(db),
183 		   db->nelem - tn40_rxdb_available(db));
184 
185 	for (i = 0; i < db->nelem; i++) {
186 		dm = tn40_rxdb_addr_elem(db, i);
187 		if (dm->page)
188 			page_pool_put_full_page(priv->page_pool, dm->page,
189 						false);
190 	}
191 }
192 
193 static void tn40_destroy_rx_ring(struct tn40_priv *priv)
194 {
195 	if (priv->rxdb0) {
196 		tn40_rx_free_buffers(priv);
197 		tn40_rxdb_free(priv->rxdb0);
198 		priv->rxdb0 = NULL;
199 	}
200 	tn40_fifo_free(priv, &priv->rxf_fifo0.m);
201 	tn40_fifo_free(priv, &priv->rxd_fifo0.m);
202 	page_pool_destroy(priv->page_pool);
203 }
204 
205 static void tn40_set_rx_desc(struct tn40_priv *priv, int idx, u64 dma)
206 {
207 	struct tn40_rxf_fifo *f = &priv->rxf_fifo0;
208 	struct tn40_rxf_desc *rxfd;
209 	int delta;
210 
211 	rxfd = (struct tn40_rxf_desc *)(f->m.va + f->m.wptr);
212 	rxfd->info = cpu_to_le32(0x10003);	/* INFO =1 BC =3 */
213 	rxfd->va_lo = cpu_to_le32(idx);
214 	rxfd->pa_lo = cpu_to_le32(lower_32_bits(dma));
215 	rxfd->pa_hi = cpu_to_le32(upper_32_bits(dma));
216 	rxfd->len = cpu_to_le32(f->m.pktsz);
217 	f->m.wptr += sizeof(struct tn40_rxf_desc);
218 	delta = f->m.wptr - f->m.memsz;
219 	if (unlikely(delta >= 0)) {
220 		f->m.wptr = delta;
221 		if (delta > 0) {
222 			memcpy(f->m.va, f->m.va + f->m.memsz, delta);
223 			netdev_dbg(priv->ndev,
224 				   "wrapped rxd descriptor\n");
225 		}
226 	}
227 }
228 
229 /**
230  * tn40_rx_alloc_buffers - Fill rxf fifo with buffers.
231  *
232  * @priv: NIC's private structure
233  *
234  * rx_alloc_buffers allocates buffers via the page pool API, builds rxf descs
235  * and pushes them (rxf descr) into the rxf fifo. The pages are stored in rxdb.
236  * To calculate the free space, we uses the cached values of RPTR and WPTR
237  * when needed. This function also updates RPTR and WPTR.
238  */
239 static void tn40_rx_alloc_buffers(struct tn40_priv *priv)
240 {
241 	struct tn40_rxf_fifo *f = &priv->rxf_fifo0;
242 	struct tn40_rxdb *db = priv->rxdb0;
243 	struct tn40_rx_map *dm;
244 	struct page *page;
245 	int dno, i, idx;
246 
247 	dno = tn40_rxdb_available(db) - 1;
248 	for (i = dno; i > 0; i--) {
249 		page = page_pool_dev_alloc_pages(priv->page_pool);
250 		if (!page)
251 			break;
252 
253 		idx = tn40_rxdb_alloc_elem(db);
254 		tn40_set_rx_desc(priv, idx, page_pool_get_dma_addr(page));
255 		dm = tn40_rxdb_addr_elem(db, idx);
256 		dm->page = page;
257 	}
258 	if (i != dno)
259 		tn40_write_reg(priv, f->m.reg_wptr,
260 			       f->m.wptr & TN40_TXF_WPTR_WR_PTR);
261 	netdev_dbg(priv->ndev, "write_reg 0x%04x f->m.reg_wptr 0x%x\n",
262 		   f->m.reg_wptr, f->m.wptr & TN40_TXF_WPTR_WR_PTR);
263 	netdev_dbg(priv->ndev, "read_reg  0x%04x f->m.reg_rptr=0x%x\n",
264 		   f->m.reg_rptr, tn40_read_reg(priv, f->m.reg_rptr));
265 	netdev_dbg(priv->ndev, "write_reg 0x%04x f->m.reg_wptr=0x%x\n",
266 		   f->m.reg_wptr, tn40_read_reg(priv, f->m.reg_wptr));
267 }
268 
269 static void tn40_recycle_rx_buffer(struct tn40_priv *priv,
270 				   struct tn40_rxd_desc *rxdd)
271 {
272 	struct tn40_rxf_fifo *f = &priv->rxf_fifo0;
273 	struct tn40_rx_map *dm;
274 	int idx;
275 
276 	idx = le32_to_cpu(rxdd->va_lo);
277 	dm = tn40_rxdb_addr_elem(priv->rxdb0, idx);
278 	tn40_set_rx_desc(priv, idx, page_pool_get_dma_addr(dm->page));
279 
280 	tn40_write_reg(priv, f->m.reg_wptr, f->m.wptr & TN40_TXF_WPTR_WR_PTR);
281 }
282 
283 static int tn40_rx_receive(struct tn40_priv *priv, int budget)
284 {
285 	struct tn40_rxd_fifo *f = &priv->rxd_fifo0;
286 	u32 rxd_val1, rxd_err, pkt_id;
287 	int tmp_len, size, done = 0;
288 	struct tn40_rxdb *db = NULL;
289 	struct tn40_rxd_desc *rxdd;
290 	struct tn40_rx_map *dm;
291 	struct sk_buff *skb;
292 	u16 len, rxd_vlan;
293 	int idx;
294 
295 	f->m.wptr = tn40_read_reg(priv, f->m.reg_wptr) & TN40_TXF_WPTR_WR_PTR;
296 	size = f->m.wptr - f->m.rptr;
297 	if (size < 0)
298 		size += f->m.memsz;	/* Size is negative :-) */
299 
300 	while (size > 0) {
301 		rxdd = (struct tn40_rxd_desc *)(f->m.va + f->m.rptr);
302 		db = priv->rxdb0;
303 
304 		/* We have a chicken and egg problem here. If the
305 		 * descriptor is wrapped we first need to copy the tail
306 		 * of the descriptor to the end of the buffer before
307 		 * extracting values from the descriptor. However in
308 		 * order to know if the descriptor is wrapped we need to
309 		 * obtain the length of the descriptor from (the
310 		 * wrapped) descriptor. Luckily the length is the first
311 		 * word of the descriptor. Descriptor lengths are
312 		 * multiples of 8 bytes so in case of a wrapped
313 		 * descriptor the first 8 bytes guaranteed to appear
314 		 * before the end of the buffer. We first obtain the
315 		 * length, we then copy the rest of the descriptor if
316 		 * needed and then extract the rest of the values from
317 		 * the descriptor.
318 		 *
319 		 * Do not change the order of operations as it will
320 		 * break the code!!!
321 		 */
322 		rxd_val1 = le32_to_cpu(rxdd->rxd_val1);
323 		tmp_len = TN40_GET_RXD_BC(rxd_val1) << 3;
324 		pkt_id = TN40_GET_RXD_PKT_ID(rxd_val1);
325 		size -= tmp_len;
326 		/* CHECK FOR A PARTIALLY ARRIVED DESCRIPTOR */
327 		if (size < 0) {
328 			netdev_dbg(priv->ndev,
329 				   "%s partially arrived desc tmp_len %d\n",
330 				   __func__, tmp_len);
331 			break;
332 		}
333 		/* make sure that the descriptor fully is arrived
334 		 * before reading the rest of the descriptor.
335 		 */
336 		rmb();
337 
338 		/* A special treatment is given to non-contiguous
339 		 * descriptors that start near the end, wraps around
340 		 * and continue at the beginning. The second part is
341 		 * copied right after the first, and then descriptor
342 		 * is interpreted as normal. The fifo has an extra
343 		 * space to allow such operations.
344 		 */
345 
346 		/* HAVE WE REACHED THE END OF THE QUEUE? */
347 		f->m.rptr += tmp_len;
348 		tmp_len = f->m.rptr - f->m.memsz;
349 		if (unlikely(tmp_len >= 0)) {
350 			f->m.rptr = tmp_len;
351 			if (tmp_len > 0) {
352 				/* COPY PARTIAL DESCRIPTOR
353 				 * TO THE END OF THE QUEUE
354 				 */
355 				netdev_dbg(priv->ndev,
356 					   "wrapped desc rptr=%d tmp_len=%d\n",
357 					   f->m.rptr, tmp_len);
358 				memcpy(f->m.va + f->m.memsz, f->m.va, tmp_len);
359 			}
360 		}
361 		idx = le32_to_cpu(rxdd->va_lo);
362 		dm = tn40_rxdb_addr_elem(db, idx);
363 		prefetch(dm);
364 
365 		len = le16_to_cpu(rxdd->len);
366 		rxd_vlan = le16_to_cpu(rxdd->rxd_vlan);
367 		/* CHECK FOR ERRORS */
368 		rxd_err = TN40_GET_RXD_ERR(rxd_val1);
369 		if (unlikely(rxd_err)) {
370 			u64_stats_update_begin(&priv->syncp);
371 			priv->stats.rx_errors++;
372 			u64_stats_update_end(&priv->syncp);
373 			tn40_recycle_rx_buffer(priv, rxdd);
374 			continue;
375 		}
376 
377 		skb = napi_build_skb(page_address(dm->page), PAGE_SIZE);
378 		if (!skb) {
379 			u64_stats_update_begin(&priv->syncp);
380 			priv->stats.rx_dropped++;
381 			u64_stats_update_end(&priv->syncp);
382 			tn40_recycle_rx_buffer(priv, rxdd);
383 			break;
384 		}
385 		skb_mark_for_recycle(skb);
386 		skb_put(skb, len);
387 		skb->protocol = eth_type_trans(skb, priv->ndev);
388 		skb->ip_summed =
389 		    (pkt_id == 0) ? CHECKSUM_NONE : CHECKSUM_UNNECESSARY;
390 		if (TN40_GET_RXD_VTAG(rxd_val1))
391 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
392 					       TN40_GET_RXD_VLAN_TCI(rxd_vlan));
393 
394 		dm->page = NULL;
395 		tn40_rxdb_free_elem(db, idx);
396 
397 		napi_gro_receive(&priv->napi, skb);
398 
399 		u64_stats_update_begin(&priv->syncp);
400 		priv->stats.rx_bytes += len;
401 		u64_stats_update_end(&priv->syncp);
402 
403 		if (unlikely(++done >= budget))
404 			break;
405 	}
406 	u64_stats_update_begin(&priv->syncp);
407 	priv->stats.rx_packets += done;
408 	u64_stats_update_end(&priv->syncp);
409 	/* FIXME: Do something to minimize pci accesses */
410 	tn40_write_reg(priv, f->m.reg_rptr, f->m.rptr & TN40_TXF_WPTR_WR_PTR);
411 	tn40_rx_alloc_buffers(priv);
412 	return done;
413 }
414 
415 /* TX HW/SW interaction overview
416  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
417  * There are 2 types of TX communication channels between driver and NIC.
418  * 1) TX Free Fifo - TXF - Holds ack descriptors for sent packets.
419  * 2) TX Data Fifo - TXD - Holds descriptors of full buffers.
420  *
421  * Currently the NIC supports TSO, checksumming and gather DMA
422  * UFO and IP fragmentation is on the way.
423  *
424  * RX SW Data Structures
425  * ~~~~~~~~~~~~~~~~~~~~~
426  * TXDB is used to keep track of all skbs owned by SW and their DMA addresses.
427  * For TX case, ownership lasts from getting the packet via hard_xmit and
428  * until the HW acknowledges sending the packet by TXF descriptors.
429  * TXDB is implemented as a cyclic buffer.
430  *
431  * FIFO objects keep info about the fifo's size and location, relevant HW
432  * registers, usage and skb db. Each RXD and RXF fifo has their own fifo
433  * structure. Implemented as simple struct.
434  *
435  * TX SW Execution Flow
436  * ~~~~~~~~~~~~~~~~~~~~
437  * OS calls the driver's hard_xmit method with a packet to send. The driver
438  * creates DMA mappings, builds TXD descriptors and kicks the HW by updating
439  * TXD WPTR.
440  *
441  * When a packet is sent, The HW write a TXF descriptor and the SW
442  * frees the original skb. To prevent TXD fifo overflow without
443  * reading HW registers every time, the SW deploys "tx level"
444  * technique. Upon startup, the tx level is initialized to TXD fifo
445  * length. For every sent packet, the SW gets its TXD descriptor size
446  * (from a pre-calculated array) and subtracts it from tx level.  The
447  * size is also stored in txdb. When a TXF ack arrives, the SW fetched
448  * the size of the original TXD descriptor from the txdb and adds it
449  * to the tx level. When the Tx level drops below some predefined
450  * threshold, the driver stops the TX queue. When the TX level rises
451  * above that level, the tx queue is enabled again.
452  *
453  * This technique avoids excessive reading of RPTR and WPTR registers.
454  * As our benchmarks shows, it adds 1.5 Gbit/sec to NIC's throughput.
455  */
456 static void tn40_do_tx_db_ptr_next(struct tn40_txdb *db,
457 				   struct tn40_tx_map **pptr)
458 {
459 	++*pptr;
460 	if (unlikely(*pptr == db->end))
461 		*pptr = db->start;
462 }
463 
464 static void tn40_tx_db_inc_rptr(struct tn40_txdb *db)
465 {
466 	tn40_do_tx_db_ptr_next(db, &db->rptr);
467 }
468 
469 static void tn40_tx_db_inc_wptr(struct tn40_txdb *db)
470 {
471 	tn40_do_tx_db_ptr_next(db, &db->wptr);
472 }
473 
474 static int tn40_tx_db_init(struct tn40_txdb *d, int sz_type)
475 {
476 	int memsz = TN40_FIFO_SIZE * (1 << (sz_type + 1));
477 
478 	d->start = vzalloc(memsz);
479 	if (!d->start)
480 		return -ENOMEM;
481 	/* In order to differentiate between an empty db state and a full db
482 	 * state at least one element should always be empty in order to
483 	 * avoid rptr == wptr, which means that the db is empty.
484 	 */
485 	d->size = memsz / sizeof(struct tn40_tx_map) - 1;
486 	d->end = d->start + d->size + 1;	/* just after last element */
487 
488 	/* All dbs are created empty */
489 	d->rptr = d->start;
490 	d->wptr = d->start;
491 	return 0;
492 }
493 
494 static void tn40_tx_db_close(struct tn40_txdb *d)
495 {
496 	if (d->start) {
497 		vfree(d->start);
498 		d->start = NULL;
499 	}
500 }
501 
502 /* Sizes of tx desc (including padding if needed) as function of the SKB's
503  * frag number
504  * 7 - is number of lwords in txd with one phys buffer
505  * 3 - is number of lwords used for every additional phys buffer
506  * for (i = 0; i < TN40_MAX_PBL; i++) {
507  *	lwords = 7 + (i * 3);
508  *	if (lwords & 1)
509  *		lwords++;	pad it with 1 lword
510  *	tn40_txd_sizes[i].bytes = lwords << 2;
511  *	tn40_txd_sizes[i].qwords = lwords >> 1;
512  * }
513  */
514 static struct {
515 	u16 bytes;
516 	u16 qwords;	/* qword = 64 bit */
517 } tn40_txd_sizes[] = {
518 	{0x20, 0x04},
519 	{0x28, 0x05},
520 	{0x38, 0x07},
521 	{0x40, 0x08},
522 	{0x50, 0x0a},
523 	{0x58, 0x0b},
524 	{0x68, 0x0d},
525 	{0x70, 0x0e},
526 	{0x80, 0x10},
527 	{0x88, 0x11},
528 	{0x98, 0x13},
529 	{0xa0, 0x14},
530 	{0xb0, 0x16},
531 	{0xb8, 0x17},
532 	{0xc8, 0x19},
533 	{0xd0, 0x1a},
534 	{0xe0, 0x1c},
535 	{0xe8, 0x1d},
536 	{0xf8, 0x1f},
537 };
538 
539 static void tn40_pbl_set(struct tn40_pbl *pbl, dma_addr_t dma, int len)
540 {
541 	pbl->len = cpu_to_le32(len);
542 	pbl->pa_lo = cpu_to_le32(lower_32_bits(dma));
543 	pbl->pa_hi = cpu_to_le32(upper_32_bits(dma));
544 }
545 
546 static void tn40_txdb_set(struct tn40_txdb *db, dma_addr_t dma, int len)
547 {
548 	db->wptr->len = len;
549 	db->wptr->addr.dma = dma;
550 }
551 
552 struct tn40_mapping_info {
553 	dma_addr_t dma;
554 	size_t size;
555 };
556 
557 /**
558  * tn40_tx_map_skb - create and store DMA mappings for skb's data blocks
559  * @priv: NIC private structure
560  * @skb: socket buffer to map
561  * @txdd: pointer to tx descriptor to be updated
562  * @pkt_len: pointer to unsigned long value
563  *
564  * This function creates DMA mappings for skb's data blocks and writes them to
565  * PBL of a new tx descriptor. It also stores them in the tx db, so they could
566  * be unmapped after the data has been sent. It is the responsibility of the
567  * caller to make sure that there is enough space in the txdb. The last
568  * element holds a pointer to skb itself and is marked with a zero length.
569  *
570  * Return: 0 on success and negative value on error.
571  */
572 static int tn40_tx_map_skb(struct tn40_priv *priv, struct sk_buff *skb,
573 			   struct tn40_txd_desc *txdd, unsigned int *pkt_len)
574 {
575 	struct tn40_mapping_info info[TN40_MAX_PBL];
576 	int nr_frags = skb_shinfo(skb)->nr_frags;
577 	struct tn40_pbl *pbl = &txdd->pbl[0];
578 	struct tn40_txdb *db = &priv->txdb;
579 	unsigned int size;
580 	int i, len, ret;
581 	dma_addr_t dma;
582 
583 	netdev_dbg(priv->ndev, "TX skb %p skbLen %d dataLen %d frags %d\n", skb,
584 		   skb->len, skb->data_len, nr_frags);
585 	if (nr_frags > TN40_MAX_PBL - 1) {
586 		ret = skb_linearize(skb);
587 		if (ret)
588 			return ret;
589 		nr_frags = skb_shinfo(skb)->nr_frags;
590 	}
591 	/* initial skb */
592 	len = skb->len - skb->data_len;
593 	dma = dma_map_single(&priv->pdev->dev, skb->data, len,
594 			     DMA_TO_DEVICE);
595 	ret = dma_mapping_error(&priv->pdev->dev, dma);
596 	if (ret)
597 		return ret;
598 
599 	tn40_txdb_set(db, dma, len);
600 	tn40_pbl_set(pbl++, db->wptr->addr.dma, db->wptr->len);
601 	*pkt_len = db->wptr->len;
602 
603 	for (i = 0; i < nr_frags; i++) {
604 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
605 
606 		size = skb_frag_size(frag);
607 		dma = skb_frag_dma_map(&priv->pdev->dev, frag, 0,
608 				       size, DMA_TO_DEVICE);
609 
610 		ret = dma_mapping_error(&priv->pdev->dev, dma);
611 		if (ret)
612 			goto mapping_error;
613 		info[i].dma = dma;
614 		info[i].size = size;
615 	}
616 
617 	for (i = 0; i < nr_frags; i++) {
618 		tn40_tx_db_inc_wptr(db);
619 		tn40_txdb_set(db, info[i].dma, info[i].size);
620 		tn40_pbl_set(pbl++, db->wptr->addr.dma, db->wptr->len);
621 		*pkt_len += db->wptr->len;
622 	}
623 
624 	/* SHORT_PKT_FIX */
625 	if (skb->len < TN40_SHORT_PACKET_SIZE)
626 		++nr_frags;
627 
628 	/* Add skb clean up info. */
629 	tn40_tx_db_inc_wptr(db);
630 	db->wptr->len = -tn40_txd_sizes[nr_frags].bytes;
631 	db->wptr->addr.skb = skb;
632 	tn40_tx_db_inc_wptr(db);
633 
634 	return 0;
635  mapping_error:
636 	dma_unmap_page(&priv->pdev->dev, db->wptr->addr.dma, db->wptr->len,
637 		       DMA_TO_DEVICE);
638 	for (; i > 0; i--)
639 		dma_unmap_page(&priv->pdev->dev, info[i - 1].dma,
640 			       info[i - 1].size, DMA_TO_DEVICE);
641 	return -ENOMEM;
642 }
643 
644 static int tn40_create_tx_ring(struct tn40_priv *priv)
645 {
646 	int ret;
647 
648 	ret = tn40_fifo_alloc(priv, &priv->txd_fifo0.m, priv->txd_size,
649 			      TN40_REG_TXD_CFG0_0, TN40_REG_TXD_CFG1_0,
650 			      TN40_REG_TXD_RPTR_0, TN40_REG_TXD_WPTR_0);
651 	if (ret)
652 		return ret;
653 
654 	ret = tn40_fifo_alloc(priv, &priv->txf_fifo0.m, priv->txf_size,
655 			      TN40_REG_TXF_CFG0_0, TN40_REG_TXF_CFG1_0,
656 			      TN40_REG_TXF_RPTR_0, TN40_REG_TXF_WPTR_0);
657 	if (ret)
658 		goto err_free_txd;
659 
660 	/* The TX db has to keep mappings for all packets sent (on
661 	 * TxD) and not yet reclaimed (on TxF).
662 	 */
663 	ret = tn40_tx_db_init(&priv->txdb, max(priv->txd_size, priv->txf_size));
664 	if (ret)
665 		goto err_free_txf;
666 
667 	/* SHORT_PKT_FIX */
668 	priv->b0_len = 64;
669 	priv->b0_va = dma_alloc_coherent(&priv->pdev->dev, priv->b0_len,
670 					 &priv->b0_dma, GFP_KERNEL);
671 	if (!priv->b0_va)
672 		goto err_free_db;
673 
674 	priv->tx_level = TN40_MAX_TX_LEVEL;
675 	priv->tx_update_mark = priv->tx_level - 1024;
676 	return 0;
677 err_free_db:
678 	tn40_tx_db_close(&priv->txdb);
679 err_free_txf:
680 	tn40_fifo_free(priv, &priv->txf_fifo0.m);
681 err_free_txd:
682 	tn40_fifo_free(priv, &priv->txd_fifo0.m);
683 	return -ENOMEM;
684 }
685 
686 /**
687  * tn40_tx_space - Calculate the available space in the TX fifo.
688  * @priv: NIC private structure
689  *
690  * Return: available space in TX fifo in bytes
691  */
692 static int tn40_tx_space(struct tn40_priv *priv)
693 {
694 	struct tn40_txd_fifo *f = &priv->txd_fifo0;
695 	int fsize;
696 
697 	f->m.rptr = tn40_read_reg(priv, f->m.reg_rptr) & TN40_TXF_WPTR_WR_PTR;
698 	fsize = f->m.rptr - f->m.wptr;
699 	if (fsize <= 0)
700 		fsize = f->m.memsz + fsize;
701 	return fsize;
702 }
703 
704 #define TN40_TXD_FULL_CHECKSUM 7
705 
706 static netdev_tx_t tn40_start_xmit(struct sk_buff *skb, struct net_device *ndev)
707 {
708 	struct tn40_priv *priv = netdev_priv(ndev);
709 	struct tn40_txd_fifo *f = &priv->txd_fifo0;
710 	int txd_checksum = TN40_TXD_FULL_CHECKSUM;
711 	struct tn40_txd_desc *txdd;
712 	int nr_frags, len, err;
713 	unsigned int pkt_len;
714 	int txd_vlan_id = 0;
715 	int txd_lgsnd = 0;
716 	int txd_vtag = 0;
717 	int txd_mss = 0;
718 
719 	/* Build tx descriptor */
720 	txdd = (struct tn40_txd_desc *)(f->m.va + f->m.wptr);
721 	err = tn40_tx_map_skb(priv, skb, txdd, &pkt_len);
722 	if (err) {
723 		u64_stats_update_begin(&priv->syncp);
724 		priv->stats.tx_dropped++;
725 		u64_stats_update_end(&priv->syncp);
726 		dev_kfree_skb(skb);
727 		return NETDEV_TX_OK;
728 	}
729 	nr_frags = skb_shinfo(skb)->nr_frags;
730 	if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL))
731 		txd_checksum = 0;
732 
733 	if (skb_shinfo(skb)->gso_size) {
734 		txd_mss = skb_shinfo(skb)->gso_size;
735 		txd_lgsnd = 1;
736 		netdev_dbg(priv->ndev, "skb %p pkt len %d gso size = %d\n", skb,
737 			   pkt_len, txd_mss);
738 	}
739 	if (skb_vlan_tag_present(skb)) {
740 		/* Don't cut VLAN ID to 12 bits */
741 		txd_vlan_id = skb_vlan_tag_get(skb);
742 		txd_vtag = 1;
743 	}
744 	txdd->va_hi = 0;
745 	txdd->va_lo = 0;
746 	txdd->length = cpu_to_le16(pkt_len);
747 	txdd->mss = cpu_to_le16(txd_mss);
748 	txdd->txd_val1 =
749 		cpu_to_le32(TN40_TXD_W1_VAL
750 			    (tn40_txd_sizes[nr_frags].qwords, txd_checksum,
751 			     txd_vtag, txd_lgsnd, txd_vlan_id));
752 	netdev_dbg(priv->ndev, "=== w1 qwords[%d] %d =====\n", nr_frags,
753 		   tn40_txd_sizes[nr_frags].qwords);
754 	netdev_dbg(priv->ndev, "=== TxD desc =====================\n");
755 	netdev_dbg(priv->ndev, "=== w1: 0x%x ================\n",
756 		   txdd->txd_val1);
757 	netdev_dbg(priv->ndev, "=== w2: mss 0x%x len 0x%x\n", txdd->mss,
758 		   txdd->length);
759 	/* SHORT_PKT_FIX */
760 	if (pkt_len < TN40_SHORT_PACKET_SIZE) {
761 		struct tn40_pbl *pbl = &txdd->pbl[++nr_frags];
762 
763 		txdd->length = cpu_to_le16(TN40_SHORT_PACKET_SIZE);
764 		txdd->txd_val1 =
765 			cpu_to_le32(TN40_TXD_W1_VAL
766 				    (tn40_txd_sizes[nr_frags].qwords,
767 				     txd_checksum, txd_vtag, txd_lgsnd,
768 				     txd_vlan_id));
769 		pbl->len = cpu_to_le32(TN40_SHORT_PACKET_SIZE - pkt_len);
770 		pbl->pa_lo = cpu_to_le32(lower_32_bits(priv->b0_dma));
771 		pbl->pa_hi = cpu_to_le32(upper_32_bits(priv->b0_dma));
772 		netdev_dbg(priv->ndev, "=== SHORT_PKT_FIX   ==============\n");
773 		netdev_dbg(priv->ndev, "=== nr_frags : %d   ==============\n",
774 			   nr_frags);
775 	}
776 
777 	/* Increment TXD write pointer. In case of fifo wrapping copy
778 	 * reminder of the descriptor to the beginning.
779 	 */
780 	f->m.wptr += tn40_txd_sizes[nr_frags].bytes;
781 	len = f->m.wptr - f->m.memsz;
782 	if (unlikely(len >= 0)) {
783 		f->m.wptr = len;
784 		if (len > 0)
785 			memcpy(f->m.va, f->m.va + f->m.memsz, len);
786 	}
787 	/* Force memory writes to complete before letting the HW know
788 	 * there are new descriptors to fetch.
789 	 */
790 	wmb();
791 
792 	priv->tx_level -= tn40_txd_sizes[nr_frags].bytes;
793 	if (priv->tx_level > priv->tx_update_mark) {
794 		tn40_write_reg(priv, f->m.reg_wptr,
795 			       f->m.wptr & TN40_TXF_WPTR_WR_PTR);
796 	} else {
797 		if (priv->tx_noupd++ > TN40_NO_UPD_PACKETS) {
798 			priv->tx_noupd = 0;
799 			tn40_write_reg(priv, f->m.reg_wptr,
800 				       f->m.wptr & TN40_TXF_WPTR_WR_PTR);
801 		}
802 	}
803 
804 	u64_stats_update_begin(&priv->syncp);
805 	priv->stats.tx_packets++;
806 	priv->stats.tx_bytes += pkt_len;
807 	u64_stats_update_end(&priv->syncp);
808 	if (priv->tx_level < TN40_MIN_TX_LEVEL) {
809 		netdev_dbg(priv->ndev, "TX Q STOP level %d\n", priv->tx_level);
810 		netif_stop_queue(ndev);
811 	}
812 
813 	return NETDEV_TX_OK;
814 }
815 
816 static void tn40_tx_cleanup(struct tn40_priv *priv)
817 {
818 	struct tn40_txf_fifo *f = &priv->txf_fifo0;
819 	struct tn40_txdb *db = &priv->txdb;
820 	int tx_level = 0;
821 
822 	f->m.wptr = tn40_read_reg(priv, f->m.reg_wptr) & TN40_TXF_WPTR_MASK;
823 
824 	netif_tx_lock(priv->ndev);
825 	while (f->m.wptr != f->m.rptr) {
826 		f->m.rptr += TN40_TXF_DESC_SZ;
827 		f->m.rptr &= f->m.size_mask;
828 		/* Unmap all fragments */
829 		/* First has to come tx_maps containing DMA */
830 		do {
831 			dma_addr_t addr = db->rptr->addr.dma;
832 			size_t size =  db->rptr->len;
833 
834 			netif_tx_unlock(priv->ndev);
835 			dma_unmap_page(&priv->pdev->dev, addr,
836 				       size, DMA_TO_DEVICE);
837 			netif_tx_lock(priv->ndev);
838 			tn40_tx_db_inc_rptr(db);
839 		} while (db->rptr->len > 0);
840 		tx_level -= db->rptr->len; /* '-' Because the len is negative */
841 
842 		/* Now should come skb pointer - free it */
843 		dev_kfree_skb_any(db->rptr->addr.skb);
844 		netdev_dbg(priv->ndev, "dev_kfree_skb_any %p %d\n",
845 			   db->rptr->addr.skb, -db->rptr->len);
846 		tn40_tx_db_inc_rptr(db);
847 	}
848 
849 	/* Let the HW know which TXF descriptors were cleaned */
850 	tn40_write_reg(priv, f->m.reg_rptr, f->m.rptr & TN40_TXF_WPTR_WR_PTR);
851 
852 	/* We reclaimed resources, so in case the Q is stopped by xmit
853 	 * callback, we resume the transmission and use tx_lock to
854 	 * synchronize with xmit.
855 	 */
856 	priv->tx_level += tx_level;
857 	if (priv->tx_noupd) {
858 		priv->tx_noupd = 0;
859 		tn40_write_reg(priv, priv->txd_fifo0.m.reg_wptr,
860 			       priv->txd_fifo0.m.wptr & TN40_TXF_WPTR_WR_PTR);
861 	}
862 	if (unlikely(netif_queue_stopped(priv->ndev) &&
863 		     netif_carrier_ok(priv->ndev) &&
864 		     (priv->tx_level >= TN40_MAX_TX_LEVEL / 2))) {
865 		netdev_dbg(priv->ndev, "TX Q WAKE level %d\n", priv->tx_level);
866 		netif_wake_queue(priv->ndev);
867 	}
868 	netif_tx_unlock(priv->ndev);
869 }
870 
871 static void tn40_tx_free_skbs(struct tn40_priv *priv)
872 {
873 	struct tn40_txdb *db = &priv->txdb;
874 
875 	while (db->rptr != db->wptr) {
876 		if (likely(db->rptr->len))
877 			dma_unmap_page(&priv->pdev->dev, db->rptr->addr.dma,
878 				       db->rptr->len, DMA_TO_DEVICE);
879 		else
880 			dev_kfree_skb(db->rptr->addr.skb);
881 		tn40_tx_db_inc_rptr(db);
882 	}
883 }
884 
885 static void tn40_destroy_tx_ring(struct tn40_priv *priv)
886 {
887 	tn40_tx_free_skbs(priv);
888 	tn40_fifo_free(priv, &priv->txd_fifo0.m);
889 	tn40_fifo_free(priv, &priv->txf_fifo0.m);
890 	tn40_tx_db_close(&priv->txdb);
891 	/* SHORT_PKT_FIX */
892 	if (priv->b0_len) {
893 		dma_free_coherent(&priv->pdev->dev, priv->b0_len, priv->b0_va,
894 				  priv->b0_dma);
895 		priv->b0_len = 0;
896 	}
897 }
898 
899 /**
900  * tn40_tx_push_desc - Push a descriptor to TxD fifo.
901  *
902  * @priv: NIC private structure
903  * @data: desc's data
904  * @size: desc's size
905  *
906  * This function pushes desc to TxD fifo and overlaps it if needed.
907  *
908  * This function does not check for available space, nor does it check
909  * that the data size is smaller than the fifo size. Checking for
910  * space is the responsibility of the caller.
911  */
912 static void tn40_tx_push_desc(struct tn40_priv *priv, void *data, int size)
913 {
914 	struct tn40_txd_fifo *f = &priv->txd_fifo0;
915 	int i = f->m.memsz - f->m.wptr;
916 
917 	if (size == 0)
918 		return;
919 
920 	if (i > size) {
921 		memcpy(f->m.va + f->m.wptr, data, size);
922 		f->m.wptr += size;
923 	} else {
924 		memcpy(f->m.va + f->m.wptr, data, i);
925 		f->m.wptr = size - i;
926 		memcpy(f->m.va, data + i, f->m.wptr);
927 	}
928 	tn40_write_reg(priv, f->m.reg_wptr, f->m.wptr & TN40_TXF_WPTR_WR_PTR);
929 }
930 
931 /**
932  * tn40_tx_push_desc_safe - push descriptor to TxD fifo in a safe way.
933  *
934  * @priv: NIC private structure
935  * @data: descriptor data
936  * @size: descriptor size
937  *
938  * This function does check for available space and, if necessary,
939  * waits for the NIC to read existing data before writing new data.
940  */
941 static void tn40_tx_push_desc_safe(struct tn40_priv *priv, void *data, int size)
942 {
943 	int timer = 0;
944 
945 	while (size > 0) {
946 		/* We subtract 8 because when the fifo is full rptr ==
947 		 * wptr, which also means that fifo is empty, we can
948 		 * understand the difference, but could the HW do the
949 		 * same ???
950 		 */
951 		int avail = tn40_tx_space(priv) - 8;
952 
953 		if (avail <= 0) {
954 			if (timer++ > 300) /* Prevent endless loop */
955 				break;
956 			/* Give the HW a chance to clean the fifo */
957 			usleep_range(50, 60);
958 			continue;
959 		}
960 		avail = min(avail, size);
961 		netdev_dbg(priv->ndev,
962 			   "about to push  %d bytes starting %p size %d\n",
963 			   avail, data, size);
964 		tn40_tx_push_desc(priv, data, avail);
965 		size -= avail;
966 		data += avail;
967 	}
968 }
969 
970 int tn40_set_link_speed(struct tn40_priv *priv, u32 speed)
971 {
972 	u32 val;
973 	int i;
974 
975 	netdev_dbg(priv->ndev, "speed %d\n", speed);
976 	switch (speed) {
977 	case SPEED_10000:
978 	case SPEED_5000:
979 	case SPEED_2500:
980 		netdev_dbg(priv->ndev, "link_speed %d\n", speed);
981 
982 		tn40_write_reg(priv, 0x1010, 0x217);	/*ETHSD.REFCLK_CONF  */
983 		tn40_write_reg(priv, 0x104c, 0x4c);	/*ETHSD.L0_RX_PCNT  */
984 		tn40_write_reg(priv, 0x1050, 0x4c);	/*ETHSD.L1_RX_PCNT  */
985 		tn40_write_reg(priv, 0x1054, 0x4c);	/*ETHSD.L2_RX_PCNT  */
986 		tn40_write_reg(priv, 0x1058, 0x4c);	/*ETHSD.L3_RX_PCNT  */
987 		tn40_write_reg(priv, 0x102c, 0x434);	/*ETHSD.L0_TX_PCNT  */
988 		tn40_write_reg(priv, 0x1030, 0x434);	/*ETHSD.L1_TX_PCNT  */
989 		tn40_write_reg(priv, 0x1034, 0x434);	/*ETHSD.L2_TX_PCNT  */
990 		tn40_write_reg(priv, 0x1038, 0x434);	/*ETHSD.L3_TX_PCNT  */
991 		tn40_write_reg(priv, 0x6300, 0x0400);	/*MAC.PCS_CTRL */
992 
993 		tn40_write_reg(priv, 0x1018, 0x00);	/*Mike2 */
994 		udelay(5);
995 		tn40_write_reg(priv, 0x1018, 0x04);	/*Mike2 */
996 		udelay(5);
997 		tn40_write_reg(priv, 0x1018, 0x06);	/*Mike2 */
998 		udelay(5);
999 		/*MikeFix1 */
1000 		/*L0: 0x103c , L1: 0x1040 , L2: 0x1044 , L3: 0x1048 =0x81644 */
1001 		tn40_write_reg(priv, 0x103c, 0x81644);	/*ETHSD.L0_TX_DCNT  */
1002 		tn40_write_reg(priv, 0x1040, 0x81644);	/*ETHSD.L1_TX_DCNT  */
1003 		tn40_write_reg(priv, 0x1044, 0x81644);	/*ETHSD.L2_TX_DCNT  */
1004 		tn40_write_reg(priv, 0x1048, 0x81644);	/*ETHSD.L3_TX_DCNT  */
1005 		tn40_write_reg(priv, 0x1014, 0x043);	/*ETHSD.INIT_STAT */
1006 		for (i = 1000; i; i--) {
1007 			usleep_range(50, 60);
1008 			/*ETHSD.INIT_STAT */
1009 			val = tn40_read_reg(priv, 0x1014);
1010 			if (val & (1 << 9)) {
1011 				/*ETHSD.INIT_STAT */
1012 				tn40_write_reg(priv, 0x1014, 0x3);
1013 				/*ETHSD.INIT_STAT */
1014 				val = tn40_read_reg(priv, 0x1014);
1015 
1016 				break;
1017 			}
1018 		}
1019 		if (!i)
1020 			netdev_err(priv->ndev, "MAC init timeout!\n");
1021 
1022 		tn40_write_reg(priv, 0x6350, 0x0);	/*MAC.PCS_IF_MODE */
1023 		tn40_write_reg(priv, TN40_REG_CTRLST, 0xC13);	/*0x93//0x13 */
1024 		tn40_write_reg(priv, 0x111c, 0x7ff);	/*MAC.MAC_RST_CNT */
1025 		usleep_range(2000, 2100);
1026 
1027 		tn40_write_reg(priv, 0x111c, 0x0);	/*MAC.MAC_RST_CNT */
1028 		break;
1029 
1030 	case SPEED_1000:
1031 	case SPEED_100:
1032 		tn40_write_reg(priv, 0x1010, 0x613);	/*ETHSD.REFCLK_CONF */
1033 		tn40_write_reg(priv, 0x104c, 0x4d);	/*ETHSD.L0_RX_PCNT  */
1034 		tn40_write_reg(priv, 0x1050, 0x0);	/*ETHSD.L1_RX_PCNT  */
1035 		tn40_write_reg(priv, 0x1054, 0x0);	/*ETHSD.L2_RX_PCNT  */
1036 		tn40_write_reg(priv, 0x1058, 0x0);	/*ETHSD.L3_RX_PCNT  */
1037 		tn40_write_reg(priv, 0x102c, 0x35);	/*ETHSD.L0_TX_PCNT  */
1038 		tn40_write_reg(priv, 0x1030, 0x0);	/*ETHSD.L1_TX_PCNT  */
1039 		tn40_write_reg(priv, 0x1034, 0x0);	/*ETHSD.L2_TX_PCNT  */
1040 		tn40_write_reg(priv, 0x1038, 0x0);	/*ETHSD.L3_TX_PCNT  */
1041 		tn40_write_reg(priv, 0x6300, 0x01140);	/*MAC.PCS_CTRL */
1042 
1043 		tn40_write_reg(priv, 0x1014, 0x043);	/*ETHSD.INIT_STAT */
1044 		for (i = 1000; i; i--) {
1045 			usleep_range(50, 60);
1046 			val = tn40_read_reg(priv, 0x1014); /*ETHSD.INIT_STAT */
1047 			if (val & (1 << 9)) {
1048 				/*ETHSD.INIT_STAT */
1049 				tn40_write_reg(priv, 0x1014, 0x3);
1050 				/*ETHSD.INIT_STAT */
1051 				val = tn40_read_reg(priv, 0x1014);
1052 
1053 				break;
1054 			}
1055 		}
1056 		if (!i)
1057 			netdev_err(priv->ndev, "MAC init timeout!\n");
1058 
1059 		tn40_write_reg(priv, 0x6350, 0x2b);	/*MAC.PCS_IF_MODE 1g */
1060 		tn40_write_reg(priv, 0x6310, 0x9801);	/*MAC.PCS_DEV_AB */
1061 
1062 		tn40_write_reg(priv, 0x6314, 0x1);	/*MAC.PCS_PART_AB */
1063 		tn40_write_reg(priv, 0x6348, 0xc8);	/*MAC.PCS_LINK_LO */
1064 		tn40_write_reg(priv, 0x634c, 0xc8);	/*MAC.PCS_LINK_HI */
1065 		usleep_range(50, 60);
1066 		tn40_write_reg(priv, TN40_REG_CTRLST, 0xC13);	/*0x93//0x13 */
1067 		tn40_write_reg(priv, 0x111c, 0x7ff);	/*MAC.MAC_RST_CNT */
1068 		usleep_range(2000, 2100);
1069 
1070 		tn40_write_reg(priv, 0x111c, 0x0);	/*MAC.MAC_RST_CNT */
1071 		tn40_write_reg(priv, 0x6300, 0x1140);	/*MAC.PCS_CTRL */
1072 		break;
1073 
1074 	case 0:		/* Link down */
1075 		tn40_write_reg(priv, 0x104c, 0x0);	/*ETHSD.L0_RX_PCNT  */
1076 		tn40_write_reg(priv, 0x1050, 0x0);	/*ETHSD.L1_RX_PCNT  */
1077 		tn40_write_reg(priv, 0x1054, 0x0);	/*ETHSD.L2_RX_PCNT  */
1078 		tn40_write_reg(priv, 0x1058, 0x0);	/*ETHSD.L3_RX_PCNT  */
1079 		tn40_write_reg(priv, 0x102c, 0x0);	/*ETHSD.L0_TX_PCNT  */
1080 		tn40_write_reg(priv, 0x1030, 0x0);	/*ETHSD.L1_TX_PCNT  */
1081 		tn40_write_reg(priv, 0x1034, 0x0);	/*ETHSD.L2_TX_PCNT  */
1082 		tn40_write_reg(priv, 0x1038, 0x0);	/*ETHSD.L3_TX_PCNT  */
1083 
1084 		tn40_write_reg(priv, TN40_REG_CTRLST, 0x800);
1085 		tn40_write_reg(priv, 0x111c, 0x7ff);	/*MAC.MAC_RST_CNT */
1086 		usleep_range(2000, 2100);
1087 
1088 		tn40_write_reg(priv, 0x111c, 0x0);	/*MAC.MAC_RST_CNT */
1089 		break;
1090 
1091 	default:
1092 		netdev_err(priv->ndev,
1093 			   "Link speed was not identified yet (%d)\n", speed);
1094 		speed = 0;
1095 		break;
1096 	}
1097 	return speed;
1098 }
1099 
1100 static void tn40_link_changed(struct tn40_priv *priv)
1101 {
1102 	u32 link = tn40_read_reg(priv,
1103 				 TN40_REG_MAC_LNK_STAT) & TN40_MAC_LINK_STAT;
1104 
1105 	netdev_dbg(priv->ndev, "link changed %u\n", link);
1106 }
1107 
1108 static void tn40_isr_extra(struct tn40_priv *priv, u32 isr)
1109 {
1110 	if (isr & (TN40_IR_LNKCHG0 | TN40_IR_LNKCHG1 | TN40_IR_TMR0)) {
1111 		netdev_dbg(priv->ndev, "isr = 0x%x\n", isr);
1112 		tn40_link_changed(priv);
1113 	}
1114 }
1115 
1116 static irqreturn_t tn40_isr_napi(int irq, void *dev)
1117 {
1118 	struct tn40_priv *priv = netdev_priv((struct net_device *)dev);
1119 	u32 isr;
1120 
1121 	isr = tn40_read_reg(priv, TN40_REG_ISR_MSK0);
1122 
1123 	if (unlikely(!isr)) {
1124 		tn40_enable_interrupts(priv);
1125 		return IRQ_NONE;	/* Not our interrupt */
1126 	}
1127 
1128 	if (isr & TN40_IR_EXTRA)
1129 		tn40_isr_extra(priv, isr);
1130 
1131 	if (isr & (TN40_IR_RX_DESC_0 | TN40_IR_TX_FREE_0 | TN40_IR_TMR1)) {
1132 		if (likely(napi_schedule_prep(&priv->napi))) {
1133 			__napi_schedule(&priv->napi);
1134 			return IRQ_HANDLED;
1135 		}
1136 		/* We get here if an interrupt has slept into the
1137 		 * small time window between these lines in
1138 		 * tn40_poll: tn40_enable_interrupts(priv); return 0;
1139 		 *
1140 		 * Currently interrupts are disabled (since we read
1141 		 * the ISR register) and we have failed to register
1142 		 * the next poll. So we read the regs to trigger the
1143 		 * chip and allow further interrupts.
1144 		 */
1145 		tn40_read_reg(priv, TN40_REG_TXF_WPTR_0);
1146 		tn40_read_reg(priv, TN40_REG_RXD_WPTR_0);
1147 	}
1148 
1149 	tn40_enable_interrupts(priv);
1150 	return IRQ_HANDLED;
1151 }
1152 
1153 static int tn40_poll(struct napi_struct *napi, int budget)
1154 {
1155 	struct tn40_priv *priv = container_of(napi, struct tn40_priv, napi);
1156 	int work_done;
1157 
1158 	tn40_tx_cleanup(priv);
1159 
1160 	if (!budget)
1161 		return 0;
1162 
1163 	work_done = tn40_rx_receive(priv, budget);
1164 	if (work_done == budget)
1165 		return budget;
1166 
1167 	if (napi_complete_done(napi, work_done))
1168 		tn40_enable_interrupts(priv);
1169 	return work_done;
1170 }
1171 
1172 static int tn40_fw_load(struct tn40_priv *priv)
1173 {
1174 	const struct firmware *fw = NULL;
1175 	int master, ret;
1176 	u32 val;
1177 
1178 	ret = request_firmware(&fw, TN40_FIRMWARE_NAME, &priv->pdev->dev);
1179 	if (ret)
1180 		return ret;
1181 
1182 	master = tn40_read_reg(priv, TN40_REG_INIT_SEMAPHORE);
1183 	if (!tn40_read_reg(priv, TN40_REG_INIT_STATUS) && master) {
1184 		netdev_dbg(priv->ndev, "Loading FW...\n");
1185 		tn40_tx_push_desc_safe(priv, (void *)fw->data, fw->size);
1186 		msleep(100);
1187 	}
1188 	ret = read_poll_timeout(tn40_read_reg, val, val, 2000, 400000, false,
1189 				priv, TN40_REG_INIT_STATUS);
1190 	if (master)
1191 		tn40_write_reg(priv, TN40_REG_INIT_SEMAPHORE, 1);
1192 
1193 	if (ret) {
1194 		netdev_err(priv->ndev, "firmware loading failed\n");
1195 		netdev_dbg(priv->ndev, "VPC: 0x%x VIC: 0x%x STATUS: 0x%xd\n",
1196 			   tn40_read_reg(priv, TN40_REG_VPC),
1197 			   tn40_read_reg(priv, TN40_REG_VIC),
1198 			   tn40_read_reg(priv, TN40_REG_INIT_STATUS));
1199 		ret = -EIO;
1200 	} else {
1201 		netdev_dbg(priv->ndev, "firmware loading success\n");
1202 	}
1203 	release_firmware(fw);
1204 	return ret;
1205 }
1206 
1207 static void tn40_restore_mac(struct net_device *ndev, struct tn40_priv *priv)
1208 {
1209 	u32 val;
1210 
1211 	netdev_dbg(priv->ndev, "mac0 =%x mac1 =%x mac2 =%x\n",
1212 		   tn40_read_reg(priv, TN40_REG_UNC_MAC0_A),
1213 		   tn40_read_reg(priv, TN40_REG_UNC_MAC1_A),
1214 		   tn40_read_reg(priv, TN40_REG_UNC_MAC2_A));
1215 
1216 	val = (ndev->dev_addr[0] << 8) | (ndev->dev_addr[1]);
1217 	tn40_write_reg(priv, TN40_REG_UNC_MAC2_A, val);
1218 	val = (ndev->dev_addr[2] << 8) | (ndev->dev_addr[3]);
1219 	tn40_write_reg(priv, TN40_REG_UNC_MAC1_A, val);
1220 	val = (ndev->dev_addr[4] << 8) | (ndev->dev_addr[5]);
1221 	tn40_write_reg(priv, TN40_REG_UNC_MAC0_A, val);
1222 
1223 	/* More then IP MAC address */
1224 	tn40_write_reg(priv, TN40_REG_MAC_ADDR_0,
1225 		       (ndev->dev_addr[3] << 24) | (ndev->dev_addr[2] << 16) |
1226 		       (ndev->dev_addr[1] << 8) | (ndev->dev_addr[0]));
1227 	tn40_write_reg(priv, TN40_REG_MAC_ADDR_1,
1228 		       (ndev->dev_addr[5] << 8) | (ndev->dev_addr[4]));
1229 
1230 	netdev_dbg(priv->ndev, "mac0 =%x mac1 =%x mac2 =%x\n",
1231 		   tn40_read_reg(priv, TN40_REG_UNC_MAC0_A),
1232 		   tn40_read_reg(priv, TN40_REG_UNC_MAC1_A),
1233 		   tn40_read_reg(priv, TN40_REG_UNC_MAC2_A));
1234 }
1235 
1236 static void tn40_hw_start(struct tn40_priv *priv)
1237 {
1238 	tn40_write_reg(priv, TN40_REG_FRM_LENGTH, 0X3FE0);
1239 	tn40_write_reg(priv, TN40_REG_GMAC_RXF_A, 0X10fd);
1240 	/*MikeFix1 */
1241 	/*L0: 0x103c , L1: 0x1040 , L2: 0x1044 , L3: 0x1048 =0x81644 */
1242 	tn40_write_reg(priv, 0x103c, 0x81644);	/*ETHSD.L0_TX_DCNT  */
1243 	tn40_write_reg(priv, 0x1040, 0x81644);	/*ETHSD.L1_TX_DCNT  */
1244 	tn40_write_reg(priv, 0x1044, 0x81644);	/*ETHSD.L2_TX_DCNT  */
1245 	tn40_write_reg(priv, 0x1048, 0x81644);	/*ETHSD.L3_TX_DCNT  */
1246 	tn40_write_reg(priv, TN40_REG_RX_FIFO_SECTION, 0x10);
1247 	tn40_write_reg(priv, TN40_REG_TX_FIFO_SECTION, 0xE00010);
1248 	tn40_write_reg(priv, TN40_REG_RX_FULLNESS, 0);
1249 	tn40_write_reg(priv, TN40_REG_TX_FULLNESS, 0);
1250 
1251 	tn40_write_reg(priv, TN40_REG_VGLB, 0);
1252 	tn40_write_reg(priv, TN40_REG_MAX_FRAME_A,
1253 		       priv->rxf_fifo0.m.pktsz & TN40_MAX_FRAME_AB_VAL);
1254 	tn40_write_reg(priv, TN40_REG_RDINTCM0, priv->rdintcm);
1255 	tn40_write_reg(priv, TN40_REG_RDINTCM2, 0);
1256 
1257 	/* old val = 0x300064 */
1258 	tn40_write_reg(priv, TN40_REG_TDINTCM0, priv->tdintcm);
1259 
1260 	/* Enable timer interrupt once in 2 secs. */
1261 	tn40_restore_mac(priv->ndev, priv);
1262 
1263 	/* Pause frame */
1264 	tn40_write_reg(priv, 0x12E0, 0x28);
1265 	tn40_write_reg(priv, TN40_REG_PAUSE_QUANT, 0xFFFF);
1266 	tn40_write_reg(priv, 0x6064, 0xF);
1267 
1268 	tn40_write_reg(priv, TN40_REG_GMAC_RXF_A,
1269 		       TN40_GMAC_RX_FILTER_OSEN | TN40_GMAC_RX_FILTER_TXFC |
1270 		       TN40_GMAC_RX_FILTER_AM | TN40_GMAC_RX_FILTER_AB);
1271 
1272 	tn40_enable_interrupts(priv);
1273 }
1274 
1275 static int tn40_hw_reset(struct tn40_priv *priv)
1276 {
1277 	u32 val;
1278 
1279 	/* Reset sequences: read, write 1, read, write 0 */
1280 	val = tn40_read_reg(priv, TN40_REG_CLKPLL);
1281 	tn40_write_reg(priv, TN40_REG_CLKPLL, (val | TN40_CLKPLL_SFTRST) + 0x8);
1282 	usleep_range(50, 60);
1283 	val = tn40_read_reg(priv, TN40_REG_CLKPLL);
1284 	tn40_write_reg(priv, TN40_REG_CLKPLL, val & ~TN40_CLKPLL_SFTRST);
1285 
1286 	/* Check that the PLLs are locked and reset ended */
1287 	val = read_poll_timeout(tn40_read_reg, val,
1288 				(val & TN40_CLKPLL_LKD) == TN40_CLKPLL_LKD,
1289 				10000, 700000, false, priv, TN40_REG_CLKPLL);
1290 	if (val)
1291 		return -EIO;
1292 
1293 	usleep_range(50, 60);
1294 	/* Do any PCI-E read transaction */
1295 	tn40_read_reg(priv, TN40_REG_RXD_CFG0_0);
1296 	return 0;
1297 }
1298 
1299 static void tn40_sw_reset(struct tn40_priv *priv)
1300 {
1301 	int i, ret;
1302 	u32 val;
1303 
1304 	/* 1. load MAC (obsolete) */
1305 	/* 2. disable Rx (and Tx) */
1306 	tn40_write_reg(priv, TN40_REG_GMAC_RXF_A, 0);
1307 	msleep(100);
1308 	/* 3. Disable port */
1309 	tn40_write_reg(priv, TN40_REG_DIS_PORT, 1);
1310 	/* 4. Disable queue */
1311 	tn40_write_reg(priv, TN40_REG_DIS_QU, 1);
1312 	/* 5. Wait until hw is disabled */
1313 	ret = read_poll_timeout(tn40_read_reg, val, val & 1, 10000, 500000,
1314 				false, priv, TN40_REG_RST_PORT);
1315 	if (ret)
1316 		netdev_err(priv->ndev, "SW reset timeout. continuing anyway\n");
1317 
1318 	/* 6. Disable interrupts */
1319 	tn40_write_reg(priv, TN40_REG_RDINTCM0, 0);
1320 	tn40_write_reg(priv, TN40_REG_TDINTCM0, 0);
1321 	tn40_write_reg(priv, TN40_REG_IMR, 0);
1322 	tn40_read_reg(priv, TN40_REG_ISR);
1323 
1324 	/* 7. Reset queue */
1325 	tn40_write_reg(priv, TN40_REG_RST_QU, 1);
1326 	/* 8. Reset port */
1327 	tn40_write_reg(priv, TN40_REG_RST_PORT, 1);
1328 	/* 9. Zero all read and write pointers */
1329 	for (i = TN40_REG_TXD_WPTR_0; i <= TN40_REG_TXF_RPTR_3; i += 0x10)
1330 		tn40_write_reg(priv, i, 0);
1331 	/* 10. Unset port disable */
1332 	tn40_write_reg(priv, TN40_REG_DIS_PORT, 0);
1333 	/* 11. Unset queue disable */
1334 	tn40_write_reg(priv, TN40_REG_DIS_QU, 0);
1335 	/* 12. Unset queue reset */
1336 	tn40_write_reg(priv, TN40_REG_RST_QU, 0);
1337 	/* 13. Unset port reset */
1338 	tn40_write_reg(priv, TN40_REG_RST_PORT, 0);
1339 	/* 14. Enable Rx */
1340 	/* Skipped. will be done later */
1341 }
1342 
1343 static int tn40_start(struct tn40_priv *priv)
1344 {
1345 	int ret;
1346 
1347 	ret = tn40_create_tx_ring(priv);
1348 	if (ret) {
1349 		netdev_err(priv->ndev, "failed to tx init %d\n", ret);
1350 		return ret;
1351 	}
1352 
1353 	ret = tn40_create_rx_ring(priv);
1354 	if (ret) {
1355 		netdev_err(priv->ndev, "failed to rx init %d\n", ret);
1356 		goto err_tx_ring;
1357 	}
1358 
1359 	tn40_rx_alloc_buffers(priv);
1360 	if (tn40_rxdb_available(priv->rxdb0) != 1) {
1361 		ret = -ENOMEM;
1362 		netdev_err(priv->ndev, "failed to allocate rx buffers\n");
1363 		goto err_rx_ring;
1364 	}
1365 
1366 	ret = request_irq(priv->pdev->irq, &tn40_isr_napi, IRQF_SHARED,
1367 			  priv->ndev->name, priv->ndev);
1368 	if (ret) {
1369 		netdev_err(priv->ndev, "failed to request irq %d\n", ret);
1370 		goto err_rx_ring;
1371 	}
1372 
1373 	tn40_hw_start(priv);
1374 	return 0;
1375 err_rx_ring:
1376 	tn40_destroy_rx_ring(priv);
1377 err_tx_ring:
1378 	tn40_destroy_tx_ring(priv);
1379 	return ret;
1380 }
1381 
1382 static void tn40_stop(struct tn40_priv *priv)
1383 {
1384 	tn40_disable_interrupts(priv);
1385 	free_irq(priv->pdev->irq, priv->ndev);
1386 	tn40_sw_reset(priv);
1387 	tn40_destroy_tx_ring(priv);
1388 	tn40_destroy_rx_ring(priv);
1389 }
1390 
1391 static int tn40_close(struct net_device *ndev)
1392 {
1393 	struct tn40_priv *priv = netdev_priv(ndev);
1394 
1395 	phylink_stop(priv->phylink);
1396 	phylink_disconnect_phy(priv->phylink);
1397 
1398 	napi_disable(&priv->napi);
1399 	netif_napi_del(&priv->napi);
1400 	tn40_stop(priv);
1401 	return 0;
1402 }
1403 
1404 static int tn40_open(struct net_device *dev)
1405 {
1406 	struct tn40_priv *priv = netdev_priv(dev);
1407 	int ret;
1408 
1409 	ret = phylink_connect_phy(priv->phylink, priv->phydev);
1410 	if (ret) {
1411 		netdev_err(dev, "failed to connect to phy %d\n", ret);
1412 		return ret;
1413 	}
1414 	tn40_sw_reset(priv);
1415 	ret = tn40_start(priv);
1416 	if (ret) {
1417 		phylink_disconnect_phy(priv->phylink);
1418 		netdev_err(dev, "failed to start %d\n", ret);
1419 		return ret;
1420 	}
1421 	napi_enable(&priv->napi);
1422 	phylink_start(priv->phylink);
1423 	netif_start_queue(priv->ndev);
1424 	return 0;
1425 }
1426 
1427 static void __tn40_vlan_rx_vid(struct net_device *ndev, uint16_t vid,
1428 			       int enable)
1429 {
1430 	struct tn40_priv *priv = netdev_priv(ndev);
1431 	u32 reg, bit, val;
1432 
1433 	netdev_dbg(priv->ndev, "vid =%d value =%d\n", (int)vid, enable);
1434 	reg = TN40_REG_VLAN_0 + (vid / 32) * 4;
1435 	bit = 1 << vid % 32;
1436 	val = tn40_read_reg(priv, reg);
1437 	netdev_dbg(priv->ndev, "reg =%x, val =%x, bit =%d\n", reg, val, bit);
1438 	if (enable)
1439 		val |= bit;
1440 	else
1441 		val &= ~bit;
1442 	netdev_dbg(priv->ndev, "new val %x\n", val);
1443 	tn40_write_reg(priv, reg, val);
1444 }
1445 
1446 static int tn40_vlan_rx_add_vid(struct net_device *ndev,
1447 				__always_unused __be16 proto, u16 vid)
1448 {
1449 	__tn40_vlan_rx_vid(ndev, vid, 1);
1450 	return 0;
1451 }
1452 
1453 static int tn40_vlan_rx_kill_vid(struct net_device *ndev,
1454 				 __always_unused __be16 proto, u16 vid)
1455 {
1456 	__tn40_vlan_rx_vid(ndev, vid, 0);
1457 	return 0;
1458 }
1459 
1460 static void tn40_setmulti(struct net_device *ndev)
1461 {
1462 	u32 rxf_val = TN40_GMAC_RX_FILTER_AM | TN40_GMAC_RX_FILTER_AB |
1463 		TN40_GMAC_RX_FILTER_OSEN | TN40_GMAC_RX_FILTER_TXFC;
1464 	struct tn40_priv *priv = netdev_priv(ndev);
1465 	int i;
1466 
1467 	/* IMF - imperfect (hash) rx multicast filter */
1468 	/* PMF - perfect rx multicast filter */
1469 
1470 	/* FIXME: RXE(OFF) */
1471 	if (ndev->flags & IFF_PROMISC) {
1472 		rxf_val |= TN40_GMAC_RX_FILTER_PRM;
1473 	} else if (ndev->flags & IFF_ALLMULTI) {
1474 		/* set IMF to accept all multicast frames */
1475 		for (i = 0; i < TN40_MAC_MCST_HASH_NUM; i++)
1476 			tn40_write_reg(priv,
1477 				       TN40_REG_RX_MCST_HASH0 + i * 4, ~0);
1478 	} else if (netdev_mc_count(ndev)) {
1479 		struct netdev_hw_addr *mclist;
1480 		u32 reg, val;
1481 		u8 hash;
1482 
1483 		/* Set IMF to deny all multicast frames */
1484 		for (i = 0; i < TN40_MAC_MCST_HASH_NUM; i++)
1485 			tn40_write_reg(priv,
1486 				       TN40_REG_RX_MCST_HASH0 + i * 4, 0);
1487 
1488 		/* Set PMF to deny all multicast frames */
1489 		for (i = 0; i < TN40_MAC_MCST_NUM; i++) {
1490 			tn40_write_reg(priv,
1491 				       TN40_REG_RX_MAC_MCST0 + i * 8, 0);
1492 			tn40_write_reg(priv,
1493 				       TN40_REG_RX_MAC_MCST1 + i * 8, 0);
1494 		}
1495 		/* Use PMF to accept first MAC_MCST_NUM (15) addresses */
1496 
1497 		/* TBD: Sort the addresses and write them in ascending
1498 		 * order into RX_MAC_MCST regs. we skip this phase now
1499 		 * and accept ALL multicast frames through IMF. Accept
1500 		 * the rest of addresses throw IMF.
1501 		 */
1502 		netdev_for_each_mc_addr(mclist, ndev) {
1503 			hash = 0;
1504 			for (i = 0; i < ETH_ALEN; i++)
1505 				hash ^= mclist->addr[i];
1506 
1507 			reg = TN40_REG_RX_MCST_HASH0 + ((hash >> 5) << 2);
1508 			val = tn40_read_reg(priv, reg);
1509 			val |= (1 << (hash % 32));
1510 			tn40_write_reg(priv, reg, val);
1511 		}
1512 	} else {
1513 		rxf_val |= TN40_GMAC_RX_FILTER_AB;
1514 	}
1515 	tn40_write_reg(priv, TN40_REG_GMAC_RXF_A, rxf_val);
1516 	/* Enable RX */
1517 	/* FIXME: RXE(ON) */
1518 }
1519 
1520 static int tn40_set_mac(struct net_device *ndev, void *p)
1521 {
1522 	struct tn40_priv *priv = netdev_priv(ndev);
1523 	struct sockaddr *addr = p;
1524 
1525 	eth_hw_addr_set(ndev, addr->sa_data);
1526 	tn40_restore_mac(ndev, priv);
1527 	return 0;
1528 }
1529 
1530 static void tn40_mac_init(struct tn40_priv *priv)
1531 {
1532 	u8 addr[ETH_ALEN];
1533 	u64 val;
1534 
1535 	val = (u64)tn40_read_reg(priv, TN40_REG_UNC_MAC0_A);
1536 	val |= (u64)tn40_read_reg(priv, TN40_REG_UNC_MAC1_A) << 16;
1537 	val |= (u64)tn40_read_reg(priv, TN40_REG_UNC_MAC2_A) << 32;
1538 
1539 	u64_to_ether_addr(val, addr);
1540 	eth_hw_addr_set(priv->ndev, addr);
1541 }
1542 
1543 static void tn40_get_stats(struct net_device *ndev,
1544 			   struct rtnl_link_stats64 *stats)
1545 {
1546 	struct tn40_priv *priv = netdev_priv(ndev);
1547 	unsigned int start;
1548 
1549 	do {
1550 		start = u64_stats_fetch_begin(&priv->syncp);
1551 		stats->tx_packets = priv->stats.tx_packets;
1552 		stats->tx_bytes = priv->stats.tx_bytes;
1553 		stats->tx_dropped = priv->stats.tx_dropped;
1554 
1555 		stats->rx_packets = priv->stats.rx_packets;
1556 		stats->rx_bytes = priv->stats.rx_bytes;
1557 		stats->rx_dropped = priv->stats.rx_dropped;
1558 		stats->rx_errors = priv->stats.rx_errors;
1559 	} while (u64_stats_fetch_retry(&priv->syncp, start));
1560 }
1561 
1562 static const struct net_device_ops tn40_netdev_ops = {
1563 	.ndo_open = tn40_open,
1564 	.ndo_stop = tn40_close,
1565 	.ndo_start_xmit = tn40_start_xmit,
1566 	.ndo_validate_addr = eth_validate_addr,
1567 	.ndo_set_rx_mode = tn40_setmulti,
1568 	.ndo_get_stats64 = tn40_get_stats,
1569 	.ndo_set_mac_address = tn40_set_mac,
1570 	.ndo_vlan_rx_add_vid = tn40_vlan_rx_add_vid,
1571 	.ndo_vlan_rx_kill_vid = tn40_vlan_rx_kill_vid,
1572 };
1573 
1574 static int tn40_priv_init(struct tn40_priv *priv)
1575 {
1576 	int ret;
1577 
1578 	tn40_set_link_speed(priv, 0);
1579 
1580 	/* Set GPIO[9:0] to output 0 */
1581 	tn40_write_reg(priv, 0x51E0, 0x30010006);	/* GPIO_OE_ WR CMD */
1582 	tn40_write_reg(priv, 0x51F0, 0x0);	/* GPIO_OE_ DATA */
1583 	tn40_write_reg(priv, TN40_REG_MDIO_CMD_STAT, 0x3ec8);
1584 
1585 	/* we use tx descriptors to load a firmware. */
1586 	ret = tn40_create_tx_ring(priv);
1587 	if (ret)
1588 		return ret;
1589 	ret = tn40_fw_load(priv);
1590 	tn40_destroy_tx_ring(priv);
1591 	return ret;
1592 }
1593 
1594 static struct net_device *tn40_netdev_alloc(struct pci_dev *pdev)
1595 {
1596 	struct net_device *ndev;
1597 
1598 	ndev = devm_alloc_etherdev(&pdev->dev, sizeof(struct tn40_priv));
1599 	if (!ndev)
1600 		return NULL;
1601 	ndev->netdev_ops = &tn40_netdev_ops;
1602 	ndev->tx_queue_len = TN40_NDEV_TXQ_LEN;
1603 	ndev->mem_start = pci_resource_start(pdev, 0);
1604 	ndev->mem_end = pci_resource_end(pdev, 0);
1605 	ndev->min_mtu = ETH_ZLEN;
1606 	ndev->max_mtu = TN40_MAX_MTU;
1607 
1608 	ndev->features = NETIF_F_IP_CSUM |
1609 		NETIF_F_SG |
1610 		NETIF_F_FRAGLIST |
1611 		NETIF_F_TSO | NETIF_F_GRO |
1612 		NETIF_F_RXCSUM |
1613 		NETIF_F_RXHASH |
1614 		NETIF_F_HW_VLAN_CTAG_TX |
1615 		NETIF_F_HW_VLAN_CTAG_RX |
1616 		NETIF_F_HW_VLAN_CTAG_FILTER;
1617 	ndev->vlan_features = NETIF_F_IP_CSUM |
1618 			       NETIF_F_SG |
1619 			       NETIF_F_TSO | NETIF_F_GRO | NETIF_F_RXHASH;
1620 
1621 	if (dma_get_mask(&pdev->dev) == DMA_BIT_MASK(64)) {
1622 		ndev->features |= NETIF_F_HIGHDMA;
1623 		ndev->vlan_features |= NETIF_F_HIGHDMA;
1624 	}
1625 	ndev->hw_features |= ndev->features;
1626 
1627 	SET_NETDEV_DEV(ndev, &pdev->dev);
1628 	netif_stop_queue(ndev);
1629 	return ndev;
1630 }
1631 
1632 static int tn40_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1633 {
1634 	struct net_device *ndev;
1635 	struct tn40_priv *priv;
1636 	unsigned int nvec = 1;
1637 	void __iomem *regs;
1638 	int ret;
1639 
1640 	ret = pci_enable_device(pdev);
1641 	if (ret)
1642 		return ret;
1643 	ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
1644 	if (ret) {
1645 		dev_err(&pdev->dev, "failed to set DMA mask.\n");
1646 		goto err_disable_device;
1647 	}
1648 
1649 	ret = pci_request_regions(pdev, TN40_DRV_NAME);
1650 	if (ret) {
1651 		dev_err(&pdev->dev, "failed to request PCI regions.\n");
1652 		goto err_disable_device;
1653 	}
1654 
1655 	pci_set_master(pdev);
1656 
1657 	regs = pci_iomap(pdev, 0, TN40_REGS_SIZE);
1658 	if (!regs) {
1659 		ret = -EIO;
1660 		dev_err(&pdev->dev, "failed to map PCI bar.\n");
1661 		goto err_free_regions;
1662 	}
1663 
1664 	ndev = tn40_netdev_alloc(pdev);
1665 	if (!ndev) {
1666 		ret = -ENOMEM;
1667 		dev_err(&pdev->dev, "failed to allocate netdev.\n");
1668 		goto err_iounmap;
1669 	}
1670 
1671 	priv = netdev_priv(ndev);
1672 	pci_set_drvdata(pdev, priv);
1673 	netif_napi_add(ndev, &priv->napi, tn40_poll);
1674 
1675 	priv->regs = regs;
1676 	priv->pdev = pdev;
1677 	priv->ndev = ndev;
1678 	/* Initialize fifo sizes. */
1679 	priv->txd_size = 3;
1680 	priv->txf_size = 3;
1681 	priv->rxd_size = 3;
1682 	priv->rxf_size = 3;
1683 	/* Initialize the initial coalescing registers. */
1684 	priv->rdintcm = TN40_INT_REG_VAL(0x20, 1, 4, 12);
1685 	priv->tdintcm = TN40_INT_REG_VAL(0x20, 1, 0, 12);
1686 
1687 	ret = tn40_hw_reset(priv);
1688 	if (ret) {
1689 		dev_err(&pdev->dev, "failed to reset HW.\n");
1690 		goto err_unset_drvdata;
1691 	}
1692 
1693 	ret = pci_alloc_irq_vectors(pdev, 1, nvec, PCI_IRQ_MSI);
1694 	if (ret < 0) {
1695 		dev_err(&pdev->dev, "failed to allocate irq.\n");
1696 		goto err_unset_drvdata;
1697 	}
1698 
1699 	ret = tn40_mdiobus_init(priv);
1700 	if (ret) {
1701 		dev_err(&pdev->dev, "failed to initialize mdio bus.\n");
1702 		goto err_free_irq;
1703 	}
1704 
1705 	priv->stats_flag =
1706 		((tn40_read_reg(priv, TN40_FPGA_VER) & 0xFFF) != 308);
1707 	u64_stats_init(&priv->syncp);
1708 
1709 	priv->isr_mask = TN40_IR_RX_FREE_0 | TN40_IR_LNKCHG0 | TN40_IR_PSE |
1710 		TN40_IR_TMR0 | TN40_IR_RX_DESC_0 | TN40_IR_TX_FREE_0 |
1711 		TN40_IR_TMR1;
1712 
1713 	tn40_mac_init(priv);
1714 	ret = tn40_phy_register(priv);
1715 	if (ret) {
1716 		dev_err(&pdev->dev, "failed to set up PHY.\n");
1717 		goto err_free_irq;
1718 	}
1719 
1720 	ret = tn40_priv_init(priv);
1721 	if (ret) {
1722 		dev_err(&pdev->dev, "failed to initialize tn40_priv.\n");
1723 		goto err_unregister_phydev;
1724 	}
1725 
1726 	ret = register_netdev(ndev);
1727 	if (ret) {
1728 		dev_err(&pdev->dev, "failed to register netdev.\n");
1729 		goto err_unregister_phydev;
1730 	}
1731 	return 0;
1732 err_unregister_phydev:
1733 	tn40_phy_unregister(priv);
1734 err_free_irq:
1735 	pci_free_irq_vectors(pdev);
1736 err_unset_drvdata:
1737 	pci_set_drvdata(pdev, NULL);
1738 err_iounmap:
1739 	iounmap(regs);
1740 err_free_regions:
1741 	pci_release_regions(pdev);
1742 err_disable_device:
1743 	pci_disable_device(pdev);
1744 	return ret;
1745 }
1746 
1747 static void tn40_remove(struct pci_dev *pdev)
1748 {
1749 	struct tn40_priv *priv = pci_get_drvdata(pdev);
1750 	struct net_device *ndev = priv->ndev;
1751 
1752 	unregister_netdev(ndev);
1753 
1754 	tn40_phy_unregister(priv);
1755 	pci_free_irq_vectors(priv->pdev);
1756 	pci_set_drvdata(pdev, NULL);
1757 	iounmap(priv->regs);
1758 	pci_release_regions(pdev);
1759 	pci_disable_device(pdev);
1760 }
1761 
1762 static const struct pci_device_id tn40_id_table[] = {
1763 	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_TEHUTI, 0x4022,
1764 			 PCI_VENDOR_ID_TEHUTI, 0x3015) },
1765 	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_TEHUTI, 0x4022,
1766 			 PCI_VENDOR_ID_DLINK, 0x4d00) },
1767 	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_TEHUTI, 0x4022,
1768 			 PCI_VENDOR_ID_ASUSTEK, 0x8709) },
1769 	{ PCI_DEVICE_SUB(PCI_VENDOR_ID_TEHUTI, 0x4022,
1770 			 PCI_VENDOR_ID_EDIMAX, 0x8103) },
1771 	{ }
1772 };
1773 
1774 static struct pci_driver tn40_driver = {
1775 	.name = TN40_DRV_NAME,
1776 	.id_table = tn40_id_table,
1777 	.probe = tn40_probe,
1778 	.remove = tn40_remove,
1779 };
1780 
1781 module_pci_driver(tn40_driver);
1782 
1783 MODULE_DEVICE_TABLE(pci, tn40_id_table);
1784 MODULE_LICENSE("GPL");
1785 MODULE_FIRMWARE(TN40_FIRMWARE_NAME);
1786 MODULE_DESCRIPTION("Tehuti Network TN40xx Driver");
1787