xref: /linux/net/core/tso.c (revision 26ba30221c03364d6ed9910be8da4c1fd871b07b)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/export.h>
3 #include <linux/if_vlan.h>
4 #include <net/ip.h>
5 #include <net/tso.h>
6 #include <linux/dma-mapping.h>
7 #include <linux/unaligned.h>
8 
9 void tso_build_hdr(const struct sk_buff *skb, char *hdr, struct tso_t *tso,
10 		   int size, bool is_last)
11 {
12 	int hdr_len = skb_transport_offset(skb) + tso->tlen;
13 	int mac_hdr_len = skb_network_offset(skb);
14 
15 	memcpy(hdr, skb->data, hdr_len);
16 	if (!tso->ipv6) {
17 		struct iphdr *iph = (void *)(hdr + mac_hdr_len);
18 
19 		iph->id = htons(tso->ip_id);
20 		iph->tot_len = htons(size + hdr_len - mac_hdr_len);
21 		tso->ip_id++;
22 	} else {
23 		struct ipv6hdr *iph = (void *)(hdr + mac_hdr_len);
24 
25 		iph->payload_len = htons(size + tso->tlen);
26 	}
27 	hdr += skb_transport_offset(skb);
28 	if (tso->tlen != sizeof(struct udphdr)) {
29 		struct tcphdr *tcph = (struct tcphdr *)hdr;
30 
31 		put_unaligned_be32(tso->tcp_seq, &tcph->seq);
32 
33 		if (!is_last) {
34 			/* Clear all special flags for not last packet */
35 			tcph->psh = 0;
36 			tcph->fin = 0;
37 			tcph->rst = 0;
38 		}
39 	} else {
40 		struct udphdr *uh = (struct udphdr *)hdr;
41 
42 		/* size is after segmentation. */
43 		udp_set_len_short(uh, sizeof(*uh) + size);
44 	}
45 }
46 EXPORT_SYMBOL(tso_build_hdr);
47 
48 void tso_build_data(const struct sk_buff *skb, struct tso_t *tso, int size)
49 {
50 	tso->tcp_seq += size; /* not worth avoiding this operation for UDP */
51 	tso->size -= size;
52 	tso->data += size;
53 
54 	if ((tso->size == 0) &&
55 	    (tso->next_frag_idx < skb_shinfo(skb)->nr_frags)) {
56 		skb_frag_t *frag = &skb_shinfo(skb)->frags[tso->next_frag_idx];
57 
58 		/* Move to next segment */
59 		tso->size = skb_frag_size(frag);
60 		tso->data = skb_frag_address(frag);
61 		tso->next_frag_idx++;
62 	}
63 }
64 EXPORT_SYMBOL(tso_build_data);
65 
66 int tso_start(struct sk_buff *skb, struct tso_t *tso)
67 {
68 	int tlen = skb_is_gso_tcp(skb) ? tcp_hdrlen(skb) : sizeof(struct udphdr);
69 	int hdr_len = skb_transport_offset(skb) + tlen;
70 
71 	tso->tlen = tlen;
72 	tso->ip_id = ntohs(ip_hdr(skb)->id);
73 	tso->tcp_seq = (tlen != sizeof(struct udphdr)) ? ntohl(tcp_hdr(skb)->seq) : 0;
74 	tso->next_frag_idx = 0;
75 	tso->ipv6 = vlan_get_protocol(skb) == htons(ETH_P_IPV6);
76 
77 	/* Build first data */
78 	tso->size = skb_headlen(skb) - hdr_len;
79 	tso->data = skb->data + hdr_len;
80 	if ((tso->size == 0) &&
81 	    (tso->next_frag_idx < skb_shinfo(skb)->nr_frags)) {
82 		skb_frag_t *frag = &skb_shinfo(skb)->frags[tso->next_frag_idx];
83 
84 		/* Move to next segment */
85 		tso->size = skb_frag_size(frag);
86 		tso->data = skb_frag_address(frag);
87 		tso->next_frag_idx++;
88 	}
89 	return hdr_len;
90 }
91 EXPORT_SYMBOL(tso_start);
92 
93 static int tso_dma_iova_try(struct device *dev, struct tso_dma_map *map,
94 			    phys_addr_t phys, size_t linear_len,
95 			    size_t total_len, size_t *offset)
96 {
97 	const struct sk_buff *skb;
98 	unsigned int nr_frags;
99 	int i;
100 
101 	if (!dma_iova_try_alloc(dev, &map->iova_state, phys, total_len))
102 		return 1;
103 
104 	skb = map->skb;
105 	nr_frags = skb_shinfo(skb)->nr_frags;
106 
107 	if (linear_len) {
108 		if (dma_iova_link(dev, &map->iova_state,
109 				  phys, *offset, linear_len,
110 				  DMA_TO_DEVICE, 0))
111 			goto iova_fail;
112 		map->linear_len = linear_len;
113 		*offset += linear_len;
114 	}
115 
116 	for (i = 0; i < nr_frags; i++) {
117 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
118 		unsigned int frag_len = skb_frag_size(frag);
119 
120 		if (dma_iova_link(dev, &map->iova_state,
121 				  skb_frag_phys(frag), *offset,
122 				  frag_len, DMA_TO_DEVICE, 0)) {
123 			map->nr_frags = i;
124 			goto iova_fail;
125 		}
126 		map->frags[i].len = frag_len;
127 		*offset += frag_len;
128 		map->nr_frags = i + 1;
129 	}
130 
131 	if (dma_iova_sync(dev, &map->iova_state, 0, total_len))
132 		goto iova_fail;
133 
134 	return 0;
135 
136 iova_fail:
137 	dma_iova_destroy(dev, &map->iova_state, *offset,
138 			 DMA_TO_DEVICE, 0);
139 	memset(&map->iova_state, 0, sizeof(map->iova_state));
140 
141 	/* reset map state */
142 	map->frag_idx = -1;
143 	map->offset = 0;
144 	map->linear_len = 0;
145 	map->nr_frags = 0;
146 
147 	return 1;
148 }
149 
150 /**
151  * tso_dma_map_init - DMA-map GSO payload regions
152  * @map: map struct to initialize
153  * @dev: device for DMA mapping
154  * @skb: the GSO skb
155  * @hdr_len: per-segment header length in bytes
156  *
157  * DMA-maps the linear payload (after headers) and all frags.
158  * Prefers the DMA IOVA API (one contiguous mapping, one IOTLB sync);
159  * falls back to per-region dma_map_phys() when IOVA is not available.
160  * Positions the iterator at byte 0 of the payload.
161  *
162  * Return: 0 on success, -ENOMEM on DMA mapping failure (partial mappings
163  * are cleaned up internally).
164  */
165 int tso_dma_map_init(struct tso_dma_map *map, struct device *dev,
166 		     const struct sk_buff *skb, unsigned int hdr_len)
167 {
168 	unsigned int linear_len = skb_headlen(skb) - hdr_len;
169 	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
170 	size_t total_len = skb->len - hdr_len;
171 	size_t offset = 0;
172 	phys_addr_t phys;
173 	int i;
174 
175 	map->dev = dev;
176 	map->skb = skb;
177 	map->hdr_len = hdr_len;
178 	map->frag_idx = -1;
179 	map->offset = 0;
180 	map->iova_offset = 0;
181 	map->total_len = total_len;
182 	map->linear_len = 0;
183 	map->nr_frags = 0;
184 	memset(&map->iova_state, 0, sizeof(map->iova_state));
185 
186 	if (!total_len)
187 		return 0;
188 
189 	if (linear_len)
190 		phys = virt_to_phys(skb->data + hdr_len);
191 	else
192 		phys = skb_frag_phys(&skb_shinfo(skb)->frags[0]);
193 
194 	if (tso_dma_iova_try(dev, map, phys, linear_len, total_len, &offset)) {
195 		/* IOVA path failed, map state was reset. Fallback to
196 		 * per-region dma_map_phys()
197 		 */
198 		if (linear_len) {
199 			map->linear_dma = dma_map_phys(dev, phys, linear_len,
200 						       DMA_TO_DEVICE, 0);
201 			if (dma_mapping_error(dev, map->linear_dma))
202 				return -ENOMEM;
203 			map->linear_len = linear_len;
204 		}
205 
206 		for (i = 0; i < nr_frags; i++) {
207 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
208 			unsigned int frag_len = skb_frag_size(frag);
209 
210 			map->frags[i].len = frag_len;
211 			map->frags[i].dma = dma_map_phys(dev, skb_frag_phys(frag),
212 							 frag_len, DMA_TO_DEVICE, 0);
213 			if (dma_mapping_error(dev, map->frags[i].dma)) {
214 				tso_dma_map_cleanup(map);
215 				return -ENOMEM;
216 			}
217 			map->nr_frags = i + 1;
218 		}
219 	}
220 
221 	if (linear_len == 0 && nr_frags > 0)
222 		map->frag_idx = 0;
223 
224 	return 0;
225 }
226 EXPORT_SYMBOL(tso_dma_map_init);
227 
228 /**
229  * tso_dma_map_cleanup - unmap all DMA regions in a tso_dma_map
230  * @map: the map to clean up
231  *
232  * Handles both IOVA and fallback paths. For IOVA, calls
233  * dma_iova_destroy(). For fallback, unmaps each region individually.
234  */
235 void tso_dma_map_cleanup(struct tso_dma_map *map)
236 {
237 	int i;
238 
239 	if (dma_use_iova(&map->iova_state)) {
240 		dma_iova_destroy(map->dev, &map->iova_state, map->total_len,
241 				 DMA_TO_DEVICE, 0);
242 		memset(&map->iova_state, 0, sizeof(map->iova_state));
243 	} else {
244 		if (map->linear_len)
245 			dma_unmap_phys(map->dev, map->linear_dma,
246 				       map->linear_len, DMA_TO_DEVICE, 0);
247 
248 		for (i = 0; i < map->nr_frags; i++)
249 			dma_unmap_phys(map->dev, map->frags[i].dma,
250 				       map->frags[i].len, DMA_TO_DEVICE, 0);
251 	}
252 
253 	map->linear_len = 0;
254 	map->nr_frags = 0;
255 }
256 EXPORT_SYMBOL(tso_dma_map_cleanup);
257 
258 /**
259  * tso_dma_map_count - count descriptors for a payload range
260  * @map: the payload map
261  * @len: number of payload bytes in this segment
262  *
263  * Counts how many contiguous DMA region chunks the next @len bytes
264  * will span, without advancing the iterator. On the IOVA path this
265  * is always 1 (contiguous). On the fallback path, uses region sizes
266  * from the current position.
267  *
268  * Return: the number of descriptors needed for @len bytes of payload.
269  */
270 unsigned int tso_dma_map_count(struct tso_dma_map *map, unsigned int len)
271 {
272 	unsigned int offset = map->offset;
273 	int idx = map->frag_idx;
274 	unsigned int count = 0;
275 
276 	if (!len)
277 		return 0;
278 
279 	if (dma_use_iova(&map->iova_state))
280 		return 1;
281 
282 	while (len > 0) {
283 		unsigned int region_len, chunk;
284 
285 		if (idx == -1)
286 			region_len = map->linear_len;
287 		else
288 			region_len = map->frags[idx].len;
289 
290 		chunk = min(len, region_len - offset);
291 		len -= chunk;
292 		count++;
293 		offset = 0;
294 		idx++;
295 	}
296 
297 	return count;
298 }
299 EXPORT_SYMBOL(tso_dma_map_count);
300 
301 /**
302  * tso_dma_map_next - yield the next DMA address range
303  * @map: the payload map
304  * @addr: output DMA address
305  * @chunk_len: output chunk length
306  * @mapping_len: full DMA mapping length when this chunk starts a new
307  *               mapping region, or 0 when continuing a previous one.
308  *               On the IOVA path this is always 0 (driver must not
309  *               do per-region unmaps; use tso_dma_map_cleanup instead).
310  * @seg_remaining: bytes left in current segment
311  *
312  * Yields the next (dma_addr, chunk_len) pair and advances the iterator.
313  * On the IOVA path, the entire payload is contiguous so each segment
314  * is always a single chunk.
315  *
316  * Return: true if a chunk was yielded, false when @seg_remaining is 0.
317  */
318 bool tso_dma_map_next(struct tso_dma_map *map, dma_addr_t *addr,
319 		      unsigned int *chunk_len, unsigned int *mapping_len,
320 		      unsigned int seg_remaining)
321 {
322 	unsigned int region_len, chunk;
323 
324 	if (!seg_remaining)
325 		return false;
326 
327 	/* IOVA path: contiguous DMA range, no region boundaries */
328 	if (dma_use_iova(&map->iova_state)) {
329 		*addr = map->iova_state.addr + map->iova_offset;
330 		*chunk_len = seg_remaining;
331 		*mapping_len = 0;
332 		map->iova_offset += seg_remaining;
333 		return true;
334 	}
335 
336 	/* Fallback path: per-region iteration */
337 
338 	if (map->frag_idx == -1) {
339 		region_len = map->linear_len;
340 		chunk = min(seg_remaining, region_len - map->offset);
341 		*addr = map->linear_dma + map->offset;
342 	} else {
343 		region_len = map->frags[map->frag_idx].len;
344 		chunk = min(seg_remaining, region_len - map->offset);
345 		*addr = map->frags[map->frag_idx].dma + map->offset;
346 	}
347 
348 	*mapping_len = (map->offset == 0) ? region_len : 0;
349 	*chunk_len = chunk;
350 	map->offset += chunk;
351 
352 	if (map->offset >= region_len) {
353 		map->frag_idx++;
354 		map->offset = 0;
355 	}
356 
357 	return true;
358 }
359 EXPORT_SYMBOL(tso_dma_map_next);
360