1 /*
2  * aQuantia Corporation Network Driver
3  * Copyright (C) 2014-2017 aQuantia Corporation. All rights reserved
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  */
9 
10 /* File aq_ring.c: Definition of functions for Rx/Tx rings. */
11 
12 #include "aq_ring.h"
13 #include "aq_nic.h"
14 #include "aq_hw.h"
15 #include "aq_hw_utils.h"
16 
17 #include <linux/netdevice.h>
18 #include <linux/etherdevice.h>
19 
20 static inline void aq_free_rxpage(struct aq_rxpage *rxpage, struct device *dev)
21 {
22 	unsigned int len = PAGE_SIZE << rxpage->order;
23 
24 	dma_unmap_page(dev, rxpage->daddr, len, DMA_FROM_DEVICE);
25 
26 	/* Drop the ref for being in the ring. */
27 	__free_pages(rxpage->page, rxpage->order);
28 	rxpage->page = NULL;
29 }
30 
31 static int aq_get_rxpage(struct aq_rxpage *rxpage, unsigned int order,
32 			 struct device *dev)
33 {
34 	struct page *page;
35 	dma_addr_t daddr;
36 	int ret = -ENOMEM;
37 
38 	page = dev_alloc_pages(order);
39 	if (unlikely(!page))
40 		goto err_exit;
41 
42 	daddr = dma_map_page(dev, page, 0, PAGE_SIZE << order,
43 			     DMA_FROM_DEVICE);
44 
45 	if (unlikely(dma_mapping_error(dev, daddr)))
46 		goto free_page;
47 
48 	rxpage->page = page;
49 	rxpage->daddr = daddr;
50 	rxpage->order = order;
51 	rxpage->pg_off = 0;
52 
53 	return 0;
54 
55 free_page:
56 	__free_pages(page, order);
57 
58 err_exit:
59 	return ret;
60 }
61 
62 static int aq_get_rxpages(struct aq_ring_s *self, struct aq_ring_buff_s *rxbuf,
63 			  int order)
64 {
65 	int ret;
66 
67 	if (rxbuf->rxdata.page) {
68 		/* One means ring is the only user and can reuse */
69 		if (page_ref_count(rxbuf->rxdata.page) > 1) {
70 			/* Try reuse buffer */
71 			rxbuf->rxdata.pg_off += AQ_CFG_RX_FRAME_MAX;
72 			if (rxbuf->rxdata.pg_off + AQ_CFG_RX_FRAME_MAX <=
73 				(PAGE_SIZE << order)) {
74 				self->stats.rx.pg_flips++;
75 			} else {
76 				/* Buffer exhausted. We have other users and
77 				 * should release this page and realloc
78 				 */
79 				aq_free_rxpage(&rxbuf->rxdata,
80 					       aq_nic_get_dev(self->aq_nic));
81 				self->stats.rx.pg_losts++;
82 			}
83 		} else {
84 			rxbuf->rxdata.pg_off = 0;
85 			self->stats.rx.pg_reuses++;
86 		}
87 	}
88 
89 	if (!rxbuf->rxdata.page) {
90 		ret = aq_get_rxpage(&rxbuf->rxdata, order,
91 				    aq_nic_get_dev(self->aq_nic));
92 		return ret;
93 	}
94 
95 	return 0;
96 }
97 
98 static struct aq_ring_s *aq_ring_alloc(struct aq_ring_s *self,
99 				       struct aq_nic_s *aq_nic)
100 {
101 	int err = 0;
102 
103 	self->buff_ring =
104 		kcalloc(self->size, sizeof(struct aq_ring_buff_s), GFP_KERNEL);
105 
106 	if (!self->buff_ring) {
107 		err = -ENOMEM;
108 		goto err_exit;
109 	}
110 	self->dx_ring = dma_alloc_coherent(aq_nic_get_dev(aq_nic),
111 					   self->size * self->dx_size,
112 					   &self->dx_ring_pa, GFP_KERNEL);
113 	if (!self->dx_ring) {
114 		err = -ENOMEM;
115 		goto err_exit;
116 	}
117 
118 err_exit:
119 	if (err < 0) {
120 		aq_ring_free(self);
121 		self = NULL;
122 	}
123 	return self;
124 }
125 
126 struct aq_ring_s *aq_ring_tx_alloc(struct aq_ring_s *self,
127 				   struct aq_nic_s *aq_nic,
128 				   unsigned int idx,
129 				   struct aq_nic_cfg_s *aq_nic_cfg)
130 {
131 	int err = 0;
132 
133 	self->aq_nic = aq_nic;
134 	self->idx = idx;
135 	self->size = aq_nic_cfg->txds;
136 	self->dx_size = aq_nic_cfg->aq_hw_caps->txd_size;
137 
138 	self = aq_ring_alloc(self, aq_nic);
139 	if (!self) {
140 		err = -ENOMEM;
141 		goto err_exit;
142 	}
143 
144 err_exit:
145 	if (err < 0) {
146 		aq_ring_free(self);
147 		self = NULL;
148 	}
149 	return self;
150 }
151 
152 struct aq_ring_s *aq_ring_rx_alloc(struct aq_ring_s *self,
153 				   struct aq_nic_s *aq_nic,
154 				   unsigned int idx,
155 				   struct aq_nic_cfg_s *aq_nic_cfg)
156 {
157 	int err = 0;
158 
159 	self->aq_nic = aq_nic;
160 	self->idx = idx;
161 	self->size = aq_nic_cfg->rxds;
162 	self->dx_size = aq_nic_cfg->aq_hw_caps->rxd_size;
163 	self->page_order = fls(AQ_CFG_RX_FRAME_MAX / PAGE_SIZE +
164 			       (AQ_CFG_RX_FRAME_MAX % PAGE_SIZE ? 1 : 0)) - 1;
165 
166 	if (aq_nic_cfg->rxpageorder > self->page_order)
167 		self->page_order = aq_nic_cfg->rxpageorder;
168 
169 	self = aq_ring_alloc(self, aq_nic);
170 	if (!self) {
171 		err = -ENOMEM;
172 		goto err_exit;
173 	}
174 
175 err_exit:
176 	if (err < 0) {
177 		aq_ring_free(self);
178 		self = NULL;
179 	}
180 	return self;
181 }
182 
183 int aq_ring_init(struct aq_ring_s *self)
184 {
185 	self->hw_head = 0;
186 	self->sw_head = 0;
187 	self->sw_tail = 0;
188 	return 0;
189 }
190 
191 static inline bool aq_ring_dx_in_range(unsigned int h, unsigned int i,
192 				       unsigned int t)
193 {
194 	return (h < t) ? ((h < i) && (i < t)) : ((h < i) || (i < t));
195 }
196 
197 void aq_ring_update_queue_state(struct aq_ring_s *ring)
198 {
199 	if (aq_ring_avail_dx(ring) <= AQ_CFG_SKB_FRAGS_MAX)
200 		aq_ring_queue_stop(ring);
201 	else if (aq_ring_avail_dx(ring) > AQ_CFG_RESTART_DESC_THRES)
202 		aq_ring_queue_wake(ring);
203 }
204 
205 void aq_ring_queue_wake(struct aq_ring_s *ring)
206 {
207 	struct net_device *ndev = aq_nic_get_ndev(ring->aq_nic);
208 
209 	if (__netif_subqueue_stopped(ndev, ring->idx)) {
210 		netif_wake_subqueue(ndev, ring->idx);
211 		ring->stats.tx.queue_restarts++;
212 	}
213 }
214 
215 void aq_ring_queue_stop(struct aq_ring_s *ring)
216 {
217 	struct net_device *ndev = aq_nic_get_ndev(ring->aq_nic);
218 
219 	if (!__netif_subqueue_stopped(ndev, ring->idx))
220 		netif_stop_subqueue(ndev, ring->idx);
221 }
222 
223 bool aq_ring_tx_clean(struct aq_ring_s *self)
224 {
225 	struct device *dev = aq_nic_get_dev(self->aq_nic);
226 	unsigned int budget = AQ_CFG_TX_CLEAN_BUDGET;
227 
228 	for (; self->sw_head != self->hw_head && budget--;
229 		self->sw_head = aq_ring_next_dx(self, self->sw_head)) {
230 		struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head];
231 
232 		if (likely(buff->is_mapped)) {
233 			if (unlikely(buff->is_sop)) {
234 				if (!buff->is_eop &&
235 				    buff->eop_index != 0xffffU &&
236 				    (!aq_ring_dx_in_range(self->sw_head,
237 						buff->eop_index,
238 						self->hw_head)))
239 					break;
240 
241 				dma_unmap_single(dev, buff->pa, buff->len,
242 						 DMA_TO_DEVICE);
243 			} else {
244 				dma_unmap_page(dev, buff->pa, buff->len,
245 					       DMA_TO_DEVICE);
246 			}
247 		}
248 
249 		if (unlikely(buff->is_eop))
250 			dev_kfree_skb_any(buff->skb);
251 
252 		buff->pa = 0U;
253 		buff->eop_index = 0xffffU;
254 	}
255 
256 	return !!budget;
257 }
258 
259 static void aq_rx_checksum(struct aq_ring_s *self,
260 			   struct aq_ring_buff_s *buff,
261 			   struct sk_buff *skb)
262 {
263 	if (!(self->aq_nic->ndev->features & NETIF_F_RXCSUM))
264 		return;
265 
266 	if (unlikely(buff->is_cso_err)) {
267 		++self->stats.rx.errors;
268 		skb->ip_summed = CHECKSUM_NONE;
269 		return;
270 	}
271 	if (buff->is_ip_cso) {
272 		__skb_incr_checksum_unnecessary(skb);
273 	} else {
274 		skb->ip_summed = CHECKSUM_NONE;
275 	}
276 
277 	if (buff->is_udp_cso || buff->is_tcp_cso)
278 		__skb_incr_checksum_unnecessary(skb);
279 }
280 
281 #define AQ_SKB_ALIGN SKB_DATA_ALIGN(sizeof(struct skb_shared_info))
282 int aq_ring_rx_clean(struct aq_ring_s *self,
283 		     struct napi_struct *napi,
284 		     int *work_done,
285 		     int budget)
286 {
287 	struct net_device *ndev = aq_nic_get_ndev(self->aq_nic);
288 	bool is_rsc_completed = true;
289 	int err = 0;
290 
291 	for (; (self->sw_head != self->hw_head) && budget;
292 		self->sw_head = aq_ring_next_dx(self, self->sw_head),
293 		--budget, ++(*work_done)) {
294 		struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head];
295 		struct aq_ring_buff_s *buff_ = NULL;
296 		struct sk_buff *skb = NULL;
297 		unsigned int next_ = 0U;
298 		unsigned int i = 0U;
299 		u16 hdr_len;
300 
301 		if (buff->is_error)
302 			continue;
303 
304 		if (buff->is_cleaned)
305 			continue;
306 
307 		if (!buff->is_eop) {
308 			for (next_ = buff->next,
309 			     buff_ = &self->buff_ring[next_]; true;
310 			     next_ = buff_->next,
311 			     buff_ = &self->buff_ring[next_]) {
312 				is_rsc_completed =
313 					aq_ring_dx_in_range(self->sw_head,
314 							    next_,
315 							    self->hw_head);
316 
317 				if (unlikely(!is_rsc_completed)) {
318 					is_rsc_completed = false;
319 					break;
320 				}
321 
322 				if (buff_->is_eop)
323 					break;
324 			}
325 
326 			if (!is_rsc_completed) {
327 				err = 0;
328 				goto err_exit;
329 			}
330 		}
331 
332 		dma_sync_single_range_for_cpu(aq_nic_get_dev(self->aq_nic),
333 					      buff->rxdata.daddr,
334 					      buff->rxdata.pg_off,
335 					      buff->len, DMA_FROM_DEVICE);
336 
337 		/* for single fragment packets use build_skb() */
338 		if (buff->is_eop &&
339 		    buff->len <= AQ_CFG_RX_FRAME_MAX - AQ_SKB_ALIGN) {
340 			skb = build_skb(aq_buf_vaddr(&buff->rxdata),
341 					AQ_CFG_RX_FRAME_MAX);
342 			if (unlikely(!skb)) {
343 				err = -ENOMEM;
344 				goto err_exit;
345 			}
346 			skb_put(skb, buff->len);
347 			page_ref_inc(buff->rxdata.page);
348 		} else {
349 			skb = napi_alloc_skb(napi, AQ_CFG_RX_HDR_SIZE);
350 			if (unlikely(!skb)) {
351 				err = -ENOMEM;
352 				goto err_exit;
353 			}
354 
355 			hdr_len = buff->len;
356 			if (hdr_len > AQ_CFG_RX_HDR_SIZE)
357 				hdr_len = eth_get_headlen(skb->dev,
358 							  aq_buf_vaddr(&buff->rxdata),
359 							  AQ_CFG_RX_HDR_SIZE);
360 
361 			memcpy(__skb_put(skb, hdr_len), aq_buf_vaddr(&buff->rxdata),
362 			       ALIGN(hdr_len, sizeof(long)));
363 
364 			if (buff->len - hdr_len > 0) {
365 				skb_add_rx_frag(skb, 0, buff->rxdata.page,
366 						buff->rxdata.pg_off + hdr_len,
367 						buff->len - hdr_len,
368 						AQ_CFG_RX_FRAME_MAX);
369 				page_ref_inc(buff->rxdata.page);
370 			}
371 
372 			if (!buff->is_eop) {
373 				buff_ = buff;
374 				i = 1U;
375 				do {
376 					next_ = buff_->next,
377 					buff_ = &self->buff_ring[next_];
378 
379 					dma_sync_single_range_for_cpu(
380 							aq_nic_get_dev(self->aq_nic),
381 							buff_->rxdata.daddr,
382 							buff_->rxdata.pg_off,
383 							buff_->len,
384 							DMA_FROM_DEVICE);
385 					skb_add_rx_frag(skb, i++,
386 							buff_->rxdata.page,
387 							buff_->rxdata.pg_off,
388 							buff_->len,
389 							AQ_CFG_RX_FRAME_MAX);
390 					page_ref_inc(buff_->rxdata.page);
391 					buff_->is_cleaned = 1;
392 				} while (!buff_->is_eop);
393 			}
394 		}
395 
396 		skb->protocol = eth_type_trans(skb, ndev);
397 
398 		aq_rx_checksum(self, buff, skb);
399 
400 		skb_set_hash(skb, buff->rss_hash,
401 			     buff->is_hash_l4 ? PKT_HASH_TYPE_L4 :
402 			     PKT_HASH_TYPE_NONE);
403 
404 		skb_record_rx_queue(skb, self->idx);
405 
406 		++self->stats.rx.packets;
407 		self->stats.rx.bytes += skb->len;
408 
409 		napi_gro_receive(napi, skb);
410 	}
411 
412 err_exit:
413 	return err;
414 }
415 
416 int aq_ring_rx_fill(struct aq_ring_s *self)
417 {
418 	unsigned int page_order = self->page_order;
419 	struct aq_ring_buff_s *buff = NULL;
420 	int err = 0;
421 	int i = 0;
422 
423 	if (aq_ring_avail_dx(self) < min_t(unsigned int, AQ_CFG_RX_REFILL_THRES,
424 					   self->size / 2))
425 		return err;
426 
427 	for (i = aq_ring_avail_dx(self); i--;
428 		self->sw_tail = aq_ring_next_dx(self, self->sw_tail)) {
429 		buff = &self->buff_ring[self->sw_tail];
430 
431 		buff->flags = 0U;
432 		buff->len = AQ_CFG_RX_FRAME_MAX;
433 
434 		err = aq_get_rxpages(self, buff, page_order);
435 		if (err)
436 			goto err_exit;
437 
438 		buff->pa = aq_buf_daddr(&buff->rxdata);
439 		buff = NULL;
440 	}
441 
442 err_exit:
443 	return err;
444 }
445 
446 void aq_ring_rx_deinit(struct aq_ring_s *self)
447 {
448 	if (!self)
449 		goto err_exit;
450 
451 	for (; self->sw_head != self->sw_tail;
452 		self->sw_head = aq_ring_next_dx(self, self->sw_head)) {
453 		struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head];
454 
455 		aq_free_rxpage(&buff->rxdata, aq_nic_get_dev(self->aq_nic));
456 	}
457 
458 err_exit:;
459 }
460 
461 void aq_ring_free(struct aq_ring_s *self)
462 {
463 	if (!self)
464 		goto err_exit;
465 
466 	kfree(self->buff_ring);
467 
468 	if (self->dx_ring)
469 		dma_free_coherent(aq_nic_get_dev(self->aq_nic),
470 				  self->size * self->dx_size, self->dx_ring,
471 				  self->dx_ring_pa);
472 
473 err_exit:;
474 }
475