Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * IBM Power Virtual Ethernet Device Driver
4 *
5 * Copyright (C) IBM Corporation, 2003, 2010
6 *
7 * Authors: Dave Larson <larson1@us.ibm.com>
8 * Santiago Leon <santil@linux.vnet.ibm.com>
9 * Brian King <brking@linux.vnet.ibm.com>
10 * Robert Jennings <rcj@linux.vnet.ibm.com>
11 * Anton Blanchard <anton@au.ibm.com>
12 */
13
14#include <linux/module.h>
15#include <linux/types.h>
16#include <linux/errno.h>
17#include <linux/dma-mapping.h>
18#include <linux/kernel.h>
19#include <linux/netdevice.h>
20#include <linux/etherdevice.h>
21#include <linux/skbuff.h>
22#include <linux/init.h>
23#include <linux/interrupt.h>
24#include <linux/mm.h>
25#include <linux/pm.h>
26#include <linux/ethtool.h>
27#include <linux/in.h>
28#include <linux/ip.h>
29#include <linux/ipv6.h>
30#include <linux/slab.h>
31#include <asm/hvcall.h>
32#include <linux/atomic.h>
33#include <asm/vio.h>
34#include <asm/iommu.h>
35#include <asm/firmware.h>
36#include <net/tcp.h>
37#include <net/ip6_checksum.h>
38
39#include "ibmveth.h"
40
41static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
42static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
43static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
44
45static struct kobj_type ktype_veth_pool;
46
47
48static const char ibmveth_driver_name[] = "ibmveth";
49static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
50#define ibmveth_driver_version "1.06"
51
52MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
53MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
54MODULE_LICENSE("GPL");
55MODULE_VERSION(ibmveth_driver_version);
56
57static unsigned int tx_copybreak __read_mostly = 128;
58module_param(tx_copybreak, uint, 0644);
59MODULE_PARM_DESC(tx_copybreak,
60 "Maximum size of packet that is copied to a new buffer on transmit");
61
62static unsigned int rx_copybreak __read_mostly = 128;
63module_param(rx_copybreak, uint, 0644);
64MODULE_PARM_DESC(rx_copybreak,
65 "Maximum size of packet that is copied to a new buffer on receive");
66
67static unsigned int rx_flush __read_mostly = 0;
68module_param(rx_flush, uint, 0644);
69MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
70
71static bool old_large_send __read_mostly;
72module_param(old_large_send, bool, 0444);
73MODULE_PARM_DESC(old_large_send,
74 "Use old large send method on firmware that supports the new method");
75
76struct ibmveth_stat {
77 char name[ETH_GSTRING_LEN];
78 int offset;
79};
80
81#define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
82#define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
83
84static struct ibmveth_stat ibmveth_stats[] = {
85 { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
86 { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
87 { "replenish_add_buff_failure",
88 IBMVETH_STAT_OFF(replenish_add_buff_failure) },
89 { "replenish_add_buff_success",
90 IBMVETH_STAT_OFF(replenish_add_buff_success) },
91 { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
92 { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
93 { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
94 { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
95 { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
96 { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
97 { "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },
98 { "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) },
99 { "fw_enabled_large_send", IBMVETH_STAT_OFF(fw_large_send_support) }
100};
101
102/* simple methods of getting data from the current rxq entry */
103static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
104{
105 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
106}
107
108static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
109{
110 return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
111 IBMVETH_RXQ_TOGGLE_SHIFT;
112}
113
114static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
115{
116 return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
117}
118
119static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
120{
121 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
122}
123
124static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
125{
126 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
127}
128
129static inline int ibmveth_rxq_large_packet(struct ibmveth_adapter *adapter)
130{
131 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_LRG_PKT;
132}
133
134static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
135{
136 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
137}
138
139static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
140{
141 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
142}
143
144/* setup the initial settings for a buffer pool */
145static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
146 u32 pool_index, u32 pool_size,
147 u32 buff_size, u32 pool_active)
148{
149 pool->size = pool_size;
150 pool->index = pool_index;
151 pool->buff_size = buff_size;
152 pool->threshold = pool_size * 7 / 8;
153 pool->active = pool_active;
154}
155
156/* allocate and setup an buffer pool - called during open */
157static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
158{
159 int i;
160
161 pool->free_map = kmalloc_array(pool->size, sizeof(u16), GFP_KERNEL);
162
163 if (!pool->free_map)
164 return -1;
165
166 pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), GFP_KERNEL);
167 if (!pool->dma_addr) {
168 kfree(pool->free_map);
169 pool->free_map = NULL;
170 return -1;
171 }
172
173 pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
174
175 if (!pool->skbuff) {
176 kfree(pool->dma_addr);
177 pool->dma_addr = NULL;
178
179 kfree(pool->free_map);
180 pool->free_map = NULL;
181 return -1;
182 }
183
184 for (i = 0; i < pool->size; ++i)
185 pool->free_map[i] = i;
186
187 atomic_set(&pool->available, 0);
188 pool->producer_index = 0;
189 pool->consumer_index = 0;
190
191 return 0;
192}
193
194static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
195{
196 unsigned long offset;
197
198 for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
199 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
200}
201
202/* replenish the buffers for a pool. note that we don't need to
203 * skb_reserve these since they are used for incoming...
204 */
205static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
206 struct ibmveth_buff_pool *pool)
207{
208 u32 i;
209 u32 count = pool->size - atomic_read(&pool->available);
210 u32 buffers_added = 0;
211 struct sk_buff *skb;
212 unsigned int free_index, index;
213 u64 correlator;
214 unsigned long lpar_rc;
215 dma_addr_t dma_addr;
216
217 mb();
218
219 for (i = 0; i < count; ++i) {
220 union ibmveth_buf_desc desc;
221
222 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
223
224 if (!skb) {
225 netdev_dbg(adapter->netdev,
226 "replenish: unable to allocate skb\n");
227 adapter->replenish_no_mem++;
228 break;
229 }
230
231 free_index = pool->consumer_index;
232 pool->consumer_index++;
233 if (pool->consumer_index >= pool->size)
234 pool->consumer_index = 0;
235 index = pool->free_map[free_index];
236
237 BUG_ON(index == IBM_VETH_INVALID_MAP);
238 BUG_ON(pool->skbuff[index] != NULL);
239
240 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
241 pool->buff_size, DMA_FROM_DEVICE);
242
243 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
244 goto failure;
245
246 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
247 pool->dma_addr[index] = dma_addr;
248 pool->skbuff[index] = skb;
249
250 correlator = ((u64)pool->index << 32) | index;
251 *(u64 *)skb->data = correlator;
252
253 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
254 desc.fields.address = dma_addr;
255
256 if (rx_flush) {
257 unsigned int len = min(pool->buff_size,
258 adapter->netdev->mtu +
259 IBMVETH_BUFF_OH);
260 ibmveth_flush_buffer(skb->data, len);
261 }
262 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
263 desc.desc);
264
265 if (lpar_rc != H_SUCCESS) {
266 goto failure;
267 } else {
268 buffers_added++;
269 adapter->replenish_add_buff_success++;
270 }
271 }
272
273 mb();
274 atomic_add(buffers_added, &(pool->available));
275 return;
276
277failure:
278 pool->free_map[free_index] = index;
279 pool->skbuff[index] = NULL;
280 if (pool->consumer_index == 0)
281 pool->consumer_index = pool->size - 1;
282 else
283 pool->consumer_index--;
284 if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
285 dma_unmap_single(&adapter->vdev->dev,
286 pool->dma_addr[index], pool->buff_size,
287 DMA_FROM_DEVICE);
288 dev_kfree_skb_any(skb);
289 adapter->replenish_add_buff_failure++;
290
291 mb();
292 atomic_add(buffers_added, &(pool->available));
293}
294
295/*
296 * The final 8 bytes of the buffer list is a counter of frames dropped
297 * because there was not a buffer in the buffer list capable of holding
298 * the frame.
299 */
300static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
301{
302 __be64 *p = adapter->buffer_list_addr + 4096 - 8;
303
304 adapter->rx_no_buffer = be64_to_cpup(p);
305}
306
307/* replenish routine */
308static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
309{
310 int i;
311
312 adapter->replenish_task_cycles++;
313
314 for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
315 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
316
317 if (pool->active &&
318 (atomic_read(&pool->available) < pool->threshold))
319 ibmveth_replenish_buffer_pool(adapter, pool);
320 }
321
322 ibmveth_update_rx_no_buffer(adapter);
323}
324
325/* empty and free ana buffer pool - also used to do cleanup in error paths */
326static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
327 struct ibmveth_buff_pool *pool)
328{
329 int i;
330
331 kfree(pool->free_map);
332 pool->free_map = NULL;
333
334 if (pool->skbuff && pool->dma_addr) {
335 for (i = 0; i < pool->size; ++i) {
336 struct sk_buff *skb = pool->skbuff[i];
337 if (skb) {
338 dma_unmap_single(&adapter->vdev->dev,
339 pool->dma_addr[i],
340 pool->buff_size,
341 DMA_FROM_DEVICE);
342 dev_kfree_skb_any(skb);
343 pool->skbuff[i] = NULL;
344 }
345 }
346 }
347
348 if (pool->dma_addr) {
349 kfree(pool->dma_addr);
350 pool->dma_addr = NULL;
351 }
352
353 if (pool->skbuff) {
354 kfree(pool->skbuff);
355 pool->skbuff = NULL;
356 }
357}
358
359/* remove a buffer from a pool */
360static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
361 u64 correlator)
362{
363 unsigned int pool = correlator >> 32;
364 unsigned int index = correlator & 0xffffffffUL;
365 unsigned int free_index;
366 struct sk_buff *skb;
367
368 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
369 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
370
371 skb = adapter->rx_buff_pool[pool].skbuff[index];
372
373 BUG_ON(skb == NULL);
374
375 adapter->rx_buff_pool[pool].skbuff[index] = NULL;
376
377 dma_unmap_single(&adapter->vdev->dev,
378 adapter->rx_buff_pool[pool].dma_addr[index],
379 adapter->rx_buff_pool[pool].buff_size,
380 DMA_FROM_DEVICE);
381
382 free_index = adapter->rx_buff_pool[pool].producer_index;
383 adapter->rx_buff_pool[pool].producer_index++;
384 if (adapter->rx_buff_pool[pool].producer_index >=
385 adapter->rx_buff_pool[pool].size)
386 adapter->rx_buff_pool[pool].producer_index = 0;
387 adapter->rx_buff_pool[pool].free_map[free_index] = index;
388
389 mb();
390
391 atomic_dec(&(adapter->rx_buff_pool[pool].available));
392}
393
394/* get the current buffer on the rx queue */
395static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
396{
397 u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
398 unsigned int pool = correlator >> 32;
399 unsigned int index = correlator & 0xffffffffUL;
400
401 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
402 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
403
404 return adapter->rx_buff_pool[pool].skbuff[index];
405}
406
407/* recycle the current buffer on the rx queue */
408static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
409{
410 u32 q_index = adapter->rx_queue.index;
411 u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
412 unsigned int pool = correlator >> 32;
413 unsigned int index = correlator & 0xffffffffUL;
414 union ibmveth_buf_desc desc;
415 unsigned long lpar_rc;
416 int ret = 1;
417
418 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
419 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
420
421 if (!adapter->rx_buff_pool[pool].active) {
422 ibmveth_rxq_harvest_buffer(adapter);
423 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
424 goto out;
425 }
426
427 desc.fields.flags_len = IBMVETH_BUF_VALID |
428 adapter->rx_buff_pool[pool].buff_size;
429 desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
430
431 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
432
433 if (lpar_rc != H_SUCCESS) {
434 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
435 "during recycle rc=%ld", lpar_rc);
436 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
437 ret = 0;
438 }
439
440 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
441 adapter->rx_queue.index = 0;
442 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
443 }
444
445out:
446 return ret;
447}
448
449static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
450{
451 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
452
453 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
454 adapter->rx_queue.index = 0;
455 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
456 }
457}
458
459static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
460 union ibmveth_buf_desc rxq_desc, u64 mac_address)
461{
462 int rc, try_again = 1;
463
464 /*
465 * After a kexec the adapter will still be open, so our attempt to
466 * open it will fail. So if we get a failure we free the adapter and
467 * try again, but only once.
468 */
469retry:
470 rc = h_register_logical_lan(adapter->vdev->unit_address,
471 adapter->buffer_list_dma, rxq_desc.desc,
472 adapter->filter_list_dma, mac_address);
473
474 if (rc != H_SUCCESS && try_again) {
475 do {
476 rc = h_free_logical_lan(adapter->vdev->unit_address);
477 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
478
479 try_again = 0;
480 goto retry;
481 }
482
483 return rc;
484}
485
486static u64 ibmveth_encode_mac_addr(u8 *mac)
487{
488 int i;
489 u64 encoded = 0;
490
491 for (i = 0; i < ETH_ALEN; i++)
492 encoded = (encoded << 8) | mac[i];
493
494 return encoded;
495}
496
497static int ibmveth_open(struct net_device *netdev)
498{
499 struct ibmveth_adapter *adapter = netdev_priv(netdev);
500 u64 mac_address;
501 int rxq_entries = 1;
502 unsigned long lpar_rc;
503 int rc;
504 union ibmveth_buf_desc rxq_desc;
505 int i;
506 struct device *dev;
507
508 netdev_dbg(netdev, "open starting\n");
509
510 napi_enable(&adapter->napi);
511
512 for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
513 rxq_entries += adapter->rx_buff_pool[i].size;
514
515 rc = -ENOMEM;
516 adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
517 if (!adapter->buffer_list_addr) {
518 netdev_err(netdev, "unable to allocate list pages\n");
519 goto out;
520 }
521
522 adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
523 if (!adapter->filter_list_addr) {
524 netdev_err(netdev, "unable to allocate filter pages\n");
525 goto out_free_buffer_list;
526 }
527
528 dev = &adapter->vdev->dev;
529
530 adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
531 rxq_entries;
532 adapter->rx_queue.queue_addr =
533 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
534 &adapter->rx_queue.queue_dma, GFP_KERNEL);
535 if (!adapter->rx_queue.queue_addr)
536 goto out_free_filter_list;
537
538 adapter->buffer_list_dma = dma_map_single(dev,
539 adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
540 if (dma_mapping_error(dev, adapter->buffer_list_dma)) {
541 netdev_err(netdev, "unable to map buffer list pages\n");
542 goto out_free_queue_mem;
543 }
544
545 adapter->filter_list_dma = dma_map_single(dev,
546 adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
547 if (dma_mapping_error(dev, adapter->filter_list_dma)) {
548 netdev_err(netdev, "unable to map filter list pages\n");
549 goto out_unmap_buffer_list;
550 }
551
552 adapter->rx_queue.index = 0;
553 adapter->rx_queue.num_slots = rxq_entries;
554 adapter->rx_queue.toggle = 1;
555
556 mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
557
558 rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
559 adapter->rx_queue.queue_len;
560 rxq_desc.fields.address = adapter->rx_queue.queue_dma;
561
562 netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
563 netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
564 netdev_dbg(netdev, "receive q @ 0x%p\n", adapter->rx_queue.queue_addr);
565
566 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
567
568 lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
569
570 if (lpar_rc != H_SUCCESS) {
571 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
572 lpar_rc);
573 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
574 "desc:0x%llx MAC:0x%llx\n",
575 adapter->buffer_list_dma,
576 adapter->filter_list_dma,
577 rxq_desc.desc,
578 mac_address);
579 rc = -ENONET;
580 goto out_unmap_filter_list;
581 }
582
583 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
584 if (!adapter->rx_buff_pool[i].active)
585 continue;
586 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
587 netdev_err(netdev, "unable to alloc pool\n");
588 adapter->rx_buff_pool[i].active = 0;
589 rc = -ENOMEM;
590 goto out_free_buffer_pools;
591 }
592 }
593
594 netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
595 rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
596 netdev);
597 if (rc != 0) {
598 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
599 netdev->irq, rc);
600 do {
601 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
602 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
603
604 goto out_free_buffer_pools;
605 }
606
607 rc = -ENOMEM;
608 adapter->bounce_buffer =
609 kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
610 if (!adapter->bounce_buffer)
611 goto out_free_irq;
612
613 adapter->bounce_buffer_dma =
614 dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
615 netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
616 if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
617 netdev_err(netdev, "unable to map bounce buffer\n");
618 goto out_free_bounce_buffer;
619 }
620
621 netdev_dbg(netdev, "initial replenish cycle\n");
622 ibmveth_interrupt(netdev->irq, netdev);
623
624 netif_start_queue(netdev);
625
626 netdev_dbg(netdev, "open complete\n");
627
628 return 0;
629
630out_free_bounce_buffer:
631 kfree(adapter->bounce_buffer);
632out_free_irq:
633 free_irq(netdev->irq, netdev);
634out_free_buffer_pools:
635 while (--i >= 0) {
636 if (adapter->rx_buff_pool[i].active)
637 ibmveth_free_buffer_pool(adapter,
638 &adapter->rx_buff_pool[i]);
639 }
640out_unmap_filter_list:
641 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
642 DMA_BIDIRECTIONAL);
643out_unmap_buffer_list:
644 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
645 DMA_BIDIRECTIONAL);
646out_free_queue_mem:
647 dma_free_coherent(dev, adapter->rx_queue.queue_len,
648 adapter->rx_queue.queue_addr,
649 adapter->rx_queue.queue_dma);
650out_free_filter_list:
651 free_page((unsigned long)adapter->filter_list_addr);
652out_free_buffer_list:
653 free_page((unsigned long)adapter->buffer_list_addr);
654out:
655 napi_disable(&adapter->napi);
656 return rc;
657}
658
659static int ibmveth_close(struct net_device *netdev)
660{
661 struct ibmveth_adapter *adapter = netdev_priv(netdev);
662 struct device *dev = &adapter->vdev->dev;
663 long lpar_rc;
664 int i;
665
666 netdev_dbg(netdev, "close starting\n");
667
668 napi_disable(&adapter->napi);
669
670 if (!adapter->pool_config)
671 netif_stop_queue(netdev);
672
673 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
674
675 do {
676 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
677 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
678
679 if (lpar_rc != H_SUCCESS) {
680 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
681 "continuing with close\n", lpar_rc);
682 }
683
684 free_irq(netdev->irq, netdev);
685
686 ibmveth_update_rx_no_buffer(adapter);
687
688 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
689 DMA_BIDIRECTIONAL);
690 free_page((unsigned long)adapter->buffer_list_addr);
691
692 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
693 DMA_BIDIRECTIONAL);
694 free_page((unsigned long)adapter->filter_list_addr);
695
696 dma_free_coherent(dev, adapter->rx_queue.queue_len,
697 adapter->rx_queue.queue_addr,
698 adapter->rx_queue.queue_dma);
699
700 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
701 if (adapter->rx_buff_pool[i].active)
702 ibmveth_free_buffer_pool(adapter,
703 &adapter->rx_buff_pool[i]);
704
705 dma_unmap_single(&adapter->vdev->dev, adapter->bounce_buffer_dma,
706 adapter->netdev->mtu + IBMVETH_BUFF_OH,
707 DMA_BIDIRECTIONAL);
708 kfree(adapter->bounce_buffer);
709
710 netdev_dbg(netdev, "close complete\n");
711
712 return 0;
713}
714
715static int netdev_get_link_ksettings(struct net_device *dev,
716 struct ethtool_link_ksettings *cmd)
717{
718 u32 supported, advertising;
719
720 supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
721 SUPPORTED_FIBRE);
722 advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
723 ADVERTISED_FIBRE);
724 cmd->base.speed = SPEED_1000;
725 cmd->base.duplex = DUPLEX_FULL;
726 cmd->base.port = PORT_FIBRE;
727 cmd->base.phy_address = 0;
728 cmd->base.autoneg = AUTONEG_ENABLE;
729
730 ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
731 supported);
732 ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
733 advertising);
734
735 return 0;
736}
737
738static void netdev_get_drvinfo(struct net_device *dev,
739 struct ethtool_drvinfo *info)
740{
741 strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
742 strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
743}
744
745static netdev_features_t ibmveth_fix_features(struct net_device *dev,
746 netdev_features_t features)
747{
748 /*
749 * Since the ibmveth firmware interface does not have the
750 * concept of separate tx/rx checksum offload enable, if rx
751 * checksum is disabled we also have to disable tx checksum
752 * offload. Once we disable rx checksum offload, we are no
753 * longer allowed to send tx buffers that are not properly
754 * checksummed.
755 */
756
757 if (!(features & NETIF_F_RXCSUM))
758 features &= ~NETIF_F_CSUM_MASK;
759
760 return features;
761}
762
763static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
764{
765 struct ibmveth_adapter *adapter = netdev_priv(dev);
766 unsigned long set_attr, clr_attr, ret_attr;
767 unsigned long set_attr6, clr_attr6;
768 long ret, ret4, ret6;
769 int rc1 = 0, rc2 = 0;
770 int restart = 0;
771
772 if (netif_running(dev)) {
773 restart = 1;
774 adapter->pool_config = 1;
775 ibmveth_close(dev);
776 adapter->pool_config = 0;
777 }
778
779 set_attr = 0;
780 clr_attr = 0;
781 set_attr6 = 0;
782 clr_attr6 = 0;
783
784 if (data) {
785 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
786 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
787 } else {
788 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
789 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
790 }
791
792 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
793
794 if (ret == H_SUCCESS &&
795 (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
796 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
797 set_attr, &ret_attr);
798
799 if (ret4 != H_SUCCESS) {
800 netdev_err(dev, "unable to change IPv4 checksum "
801 "offload settings. %d rc=%ld\n",
802 data, ret4);
803
804 h_illan_attributes(adapter->vdev->unit_address,
805 set_attr, clr_attr, &ret_attr);
806
807 if (data == 1)
808 dev->features &= ~NETIF_F_IP_CSUM;
809
810 } else {
811 adapter->fw_ipv4_csum_support = data;
812 }
813
814 ret6 = h_illan_attributes(adapter->vdev->unit_address,
815 clr_attr6, set_attr6, &ret_attr);
816
817 if (ret6 != H_SUCCESS) {
818 netdev_err(dev, "unable to change IPv6 checksum "
819 "offload settings. %d rc=%ld\n",
820 data, ret6);
821
822 h_illan_attributes(adapter->vdev->unit_address,
823 set_attr6, clr_attr6, &ret_attr);
824
825 if (data == 1)
826 dev->features &= ~NETIF_F_IPV6_CSUM;
827
828 } else
829 adapter->fw_ipv6_csum_support = data;
830
831 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
832 adapter->rx_csum = data;
833 else
834 rc1 = -EIO;
835 } else {
836 rc1 = -EIO;
837 netdev_err(dev, "unable to change checksum offload settings."
838 " %d rc=%ld ret_attr=%lx\n", data, ret,
839 ret_attr);
840 }
841
842 if (restart)
843 rc2 = ibmveth_open(dev);
844
845 return rc1 ? rc1 : rc2;
846}
847
848static int ibmveth_set_tso(struct net_device *dev, u32 data)
849{
850 struct ibmveth_adapter *adapter = netdev_priv(dev);
851 unsigned long set_attr, clr_attr, ret_attr;
852 long ret1, ret2;
853 int rc1 = 0, rc2 = 0;
854 int restart = 0;
855
856 if (netif_running(dev)) {
857 restart = 1;
858 adapter->pool_config = 1;
859 ibmveth_close(dev);
860 adapter->pool_config = 0;
861 }
862
863 set_attr = 0;
864 clr_attr = 0;
865
866 if (data)
867 set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
868 else
869 clr_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
870
871 ret1 = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
872
873 if (ret1 == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
874 !old_large_send) {
875 ret2 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
876 set_attr, &ret_attr);
877
878 if (ret2 != H_SUCCESS) {
879 netdev_err(dev, "unable to change tso settings. %d rc=%ld\n",
880 data, ret2);
881
882 h_illan_attributes(adapter->vdev->unit_address,
883 set_attr, clr_attr, &ret_attr);
884
885 if (data == 1)
886 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
887 rc1 = -EIO;
888
889 } else {
890 adapter->fw_large_send_support = data;
891 adapter->large_send = data;
892 }
893 } else {
894 /* Older firmware version of large send offload does not
895 * support tcp6/ipv6
896 */
897 if (data == 1) {
898 dev->features &= ~NETIF_F_TSO6;
899 netdev_info(dev, "TSO feature requires all partitions to have updated driver");
900 }
901 adapter->large_send = data;
902 }
903
904 if (restart)
905 rc2 = ibmveth_open(dev);
906
907 return rc1 ? rc1 : rc2;
908}
909
910static int ibmveth_set_features(struct net_device *dev,
911 netdev_features_t features)
912{
913 struct ibmveth_adapter *adapter = netdev_priv(dev);
914 int rx_csum = !!(features & NETIF_F_RXCSUM);
915 int large_send = !!(features & (NETIF_F_TSO | NETIF_F_TSO6));
916 int rc1 = 0, rc2 = 0;
917
918 if (rx_csum != adapter->rx_csum) {
919 rc1 = ibmveth_set_csum_offload(dev, rx_csum);
920 if (rc1 && !adapter->rx_csum)
921 dev->features =
922 features & ~(NETIF_F_CSUM_MASK |
923 NETIF_F_RXCSUM);
924 }
925
926 if (large_send != adapter->large_send) {
927 rc2 = ibmveth_set_tso(dev, large_send);
928 if (rc2 && !adapter->large_send)
929 dev->features =
930 features & ~(NETIF_F_TSO | NETIF_F_TSO6);
931 }
932
933 return rc1 ? rc1 : rc2;
934}
935
936static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
937{
938 int i;
939
940 if (stringset != ETH_SS_STATS)
941 return;
942
943 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
944 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
945}
946
947static int ibmveth_get_sset_count(struct net_device *dev, int sset)
948{
949 switch (sset) {
950 case ETH_SS_STATS:
951 return ARRAY_SIZE(ibmveth_stats);
952 default:
953 return -EOPNOTSUPP;
954 }
955}
956
957static void ibmveth_get_ethtool_stats(struct net_device *dev,
958 struct ethtool_stats *stats, u64 *data)
959{
960 int i;
961 struct ibmveth_adapter *adapter = netdev_priv(dev);
962
963 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
964 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
965}
966
967static const struct ethtool_ops netdev_ethtool_ops = {
968 .get_drvinfo = netdev_get_drvinfo,
969 .get_link = ethtool_op_get_link,
970 .get_strings = ibmveth_get_strings,
971 .get_sset_count = ibmveth_get_sset_count,
972 .get_ethtool_stats = ibmveth_get_ethtool_stats,
973 .get_link_ksettings = netdev_get_link_ksettings,
974};
975
976static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
977{
978 return -EOPNOTSUPP;
979}
980
981#define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
982
983static int ibmveth_send(struct ibmveth_adapter *adapter,
984 union ibmveth_buf_desc *descs, unsigned long mss)
985{
986 unsigned long correlator;
987 unsigned int retry_count;
988 unsigned long ret;
989
990 /*
991 * The retry count sets a maximum for the number of broadcast and
992 * multicast destinations within the system.
993 */
994 retry_count = 1024;
995 correlator = 0;
996 do {
997 ret = h_send_logical_lan(adapter->vdev->unit_address,
998 descs[0].desc, descs[1].desc,
999 descs[2].desc, descs[3].desc,
1000 descs[4].desc, descs[5].desc,
1001 correlator, &correlator, mss,
1002 adapter->fw_large_send_support);
1003 } while ((ret == H_BUSY) && (retry_count--));
1004
1005 if (ret != H_SUCCESS && ret != H_DROPPED) {
1006 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
1007 "with rc=%ld\n", ret);
1008 return 1;
1009 }
1010
1011 return 0;
1012}
1013
1014static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
1015 struct net_device *netdev)
1016{
1017 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1018 unsigned int desc_flags;
1019 union ibmveth_buf_desc descs[6];
1020 int last, i;
1021 int force_bounce = 0;
1022 dma_addr_t dma_addr;
1023 unsigned long mss = 0;
1024
1025 /* veth doesn't handle frag_list, so linearize the skb.
1026 * When GRO is enabled SKB's can have frag_list.
1027 */
1028 if (adapter->is_active_trunk &&
1029 skb_has_frag_list(skb) && __skb_linearize(skb)) {
1030 netdev->stats.tx_dropped++;
1031 goto out;
1032 }
1033
1034 /*
1035 * veth handles a maximum of 6 segments including the header, so
1036 * we have to linearize the skb if there are more than this.
1037 */
1038 if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
1039 netdev->stats.tx_dropped++;
1040 goto out;
1041 }
1042
1043 /* veth can't checksum offload UDP */
1044 if (skb->ip_summed == CHECKSUM_PARTIAL &&
1045 ((skb->protocol == htons(ETH_P_IP) &&
1046 ip_hdr(skb)->protocol != IPPROTO_TCP) ||
1047 (skb->protocol == htons(ETH_P_IPV6) &&
1048 ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
1049 skb_checksum_help(skb)) {
1050
1051 netdev_err(netdev, "tx: failed to checksum packet\n");
1052 netdev->stats.tx_dropped++;
1053 goto out;
1054 }
1055
1056 desc_flags = IBMVETH_BUF_VALID;
1057
1058 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1059 unsigned char *buf = skb_transport_header(skb) +
1060 skb->csum_offset;
1061
1062 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1063
1064 /* Need to zero out the checksum */
1065 buf[0] = 0;
1066 buf[1] = 0;
1067
1068 if (skb_is_gso(skb) && adapter->fw_large_send_support)
1069 desc_flags |= IBMVETH_BUF_LRG_SND;
1070 }
1071
1072retry_bounce:
1073 memset(descs, 0, sizeof(descs));
1074
1075 /*
1076 * If a linear packet is below the rx threshold then
1077 * copy it into the static bounce buffer. This avoids the
1078 * cost of a TCE insert and remove.
1079 */
1080 if (force_bounce || (!skb_is_nonlinear(skb) &&
1081 (skb->len < tx_copybreak))) {
1082 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1083 skb->len);
1084
1085 descs[0].fields.flags_len = desc_flags | skb->len;
1086 descs[0].fields.address = adapter->bounce_buffer_dma;
1087
1088 if (ibmveth_send(adapter, descs, 0)) {
1089 adapter->tx_send_failed++;
1090 netdev->stats.tx_dropped++;
1091 } else {
1092 netdev->stats.tx_packets++;
1093 netdev->stats.tx_bytes += skb->len;
1094 }
1095
1096 goto out;
1097 }
1098
1099 /* Map the header */
1100 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1101 skb_headlen(skb), DMA_TO_DEVICE);
1102 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1103 goto map_failed;
1104
1105 descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1106 descs[0].fields.address = dma_addr;
1107
1108 /* Map the frags */
1109 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1110 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1111
1112 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1113 skb_frag_size(frag), DMA_TO_DEVICE);
1114
1115 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1116 goto map_failed_frags;
1117
1118 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1119 descs[i+1].fields.address = dma_addr;
1120 }
1121
1122 if (skb->ip_summed == CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1123 if (adapter->fw_large_send_support) {
1124 mss = (unsigned long)skb_shinfo(skb)->gso_size;
1125 adapter->tx_large_packets++;
1126 } else if (!skb_is_gso_v6(skb)) {
1127 /* Put -1 in the IP checksum to tell phyp it
1128 * is a largesend packet. Put the mss in
1129 * the TCP checksum.
1130 */
1131 ip_hdr(skb)->check = 0xffff;
1132 tcp_hdr(skb)->check =
1133 cpu_to_be16(skb_shinfo(skb)->gso_size);
1134 adapter->tx_large_packets++;
1135 }
1136 }
1137
1138 if (ibmveth_send(adapter, descs, mss)) {
1139 adapter->tx_send_failed++;
1140 netdev->stats.tx_dropped++;
1141 } else {
1142 netdev->stats.tx_packets++;
1143 netdev->stats.tx_bytes += skb->len;
1144 }
1145
1146 dma_unmap_single(&adapter->vdev->dev,
1147 descs[0].fields.address,
1148 descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1149 DMA_TO_DEVICE);
1150
1151 for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1152 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1153 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1154 DMA_TO_DEVICE);
1155
1156out:
1157 dev_consume_skb_any(skb);
1158 return NETDEV_TX_OK;
1159
1160map_failed_frags:
1161 last = i+1;
1162 for (i = 1; i < last; i++)
1163 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1164 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1165 DMA_TO_DEVICE);
1166
1167 dma_unmap_single(&adapter->vdev->dev,
1168 descs[0].fields.address,
1169 descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1170 DMA_TO_DEVICE);
1171map_failed:
1172 if (!firmware_has_feature(FW_FEATURE_CMO))
1173 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1174 adapter->tx_map_failed++;
1175 if (skb_linearize(skb)) {
1176 netdev->stats.tx_dropped++;
1177 goto out;
1178 }
1179 force_bounce = 1;
1180 goto retry_bounce;
1181}
1182
1183static void ibmveth_rx_mss_helper(struct sk_buff *skb, u16 mss, int lrg_pkt)
1184{
1185 struct tcphdr *tcph;
1186 int offset = 0;
1187 int hdr_len;
1188
1189 /* only TCP packets will be aggregated */
1190 if (skb->protocol == htons(ETH_P_IP)) {
1191 struct iphdr *iph = (struct iphdr *)skb->data;
1192
1193 if (iph->protocol == IPPROTO_TCP) {
1194 offset = iph->ihl * 4;
1195 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1196 } else {
1197 return;
1198 }
1199 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1200 struct ipv6hdr *iph6 = (struct ipv6hdr *)skb->data;
1201
1202 if (iph6->nexthdr == IPPROTO_TCP) {
1203 offset = sizeof(struct ipv6hdr);
1204 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1205 } else {
1206 return;
1207 }
1208 } else {
1209 return;
1210 }
1211 /* if mss is not set through Large Packet bit/mss in rx buffer,
1212 * expect that the mss will be written to the tcp header checksum.
1213 */
1214 tcph = (struct tcphdr *)(skb->data + offset);
1215 if (lrg_pkt) {
1216 skb_shinfo(skb)->gso_size = mss;
1217 } else if (offset) {
1218 skb_shinfo(skb)->gso_size = ntohs(tcph->check);
1219 tcph->check = 0;
1220 }
1221
1222 if (skb_shinfo(skb)->gso_size) {
1223 hdr_len = offset + tcph->doff * 4;
1224 skb_shinfo(skb)->gso_segs =
1225 DIV_ROUND_UP(skb->len - hdr_len,
1226 skb_shinfo(skb)->gso_size);
1227 }
1228}
1229
1230static void ibmveth_rx_csum_helper(struct sk_buff *skb,
1231 struct ibmveth_adapter *adapter)
1232{
1233 struct iphdr *iph = NULL;
1234 struct ipv6hdr *iph6 = NULL;
1235 __be16 skb_proto = 0;
1236 u16 iphlen = 0;
1237 u16 iph_proto = 0;
1238 u16 tcphdrlen = 0;
1239
1240 skb_proto = be16_to_cpu(skb->protocol);
1241
1242 if (skb_proto == ETH_P_IP) {
1243 iph = (struct iphdr *)skb->data;
1244
1245 /* If the IP checksum is not offloaded and if the packet
1246 * is large send, the checksum must be rebuilt.
1247 */
1248 if (iph->check == 0xffff) {
1249 iph->check = 0;
1250 iph->check = ip_fast_csum((unsigned char *)iph,
1251 iph->ihl);
1252 }
1253
1254 iphlen = iph->ihl * 4;
1255 iph_proto = iph->protocol;
1256 } else if (skb_proto == ETH_P_IPV6) {
1257 iph6 = (struct ipv6hdr *)skb->data;
1258 iphlen = sizeof(struct ipv6hdr);
1259 iph_proto = iph6->nexthdr;
1260 }
1261
1262 /* In OVS environment, when a flow is not cached, specifically for a
1263 * new TCP connection, the first packet information is passed up
1264 * the user space for finding a flow. During this process, OVS computes
1265 * checksum on the first packet when CHECKSUM_PARTIAL flag is set.
1266 *
1267 * Given that we zeroed out TCP checksum field in transmit path
1268 * (refer ibmveth_start_xmit routine) as we set "no checksum bit",
1269 * OVS computed checksum will be incorrect w/o TCP pseudo checksum
1270 * in the packet. This leads to OVS dropping the packet and hence
1271 * TCP retransmissions are seen.
1272 *
1273 * So, re-compute TCP pseudo header checksum.
1274 */
1275 if (iph_proto == IPPROTO_TCP && adapter->is_active_trunk) {
1276 struct tcphdr *tcph = (struct tcphdr *)(skb->data + iphlen);
1277
1278 tcphdrlen = skb->len - iphlen;
1279
1280 /* Recompute TCP pseudo header checksum */
1281 if (skb_proto == ETH_P_IP)
1282 tcph->check = ~csum_tcpudp_magic(iph->saddr,
1283 iph->daddr, tcphdrlen, iph_proto, 0);
1284 else if (skb_proto == ETH_P_IPV6)
1285 tcph->check = ~csum_ipv6_magic(&iph6->saddr,
1286 &iph6->daddr, tcphdrlen, iph_proto, 0);
1287
1288 /* Setup SKB fields for checksum offload */
1289 skb_partial_csum_set(skb, iphlen,
1290 offsetof(struct tcphdr, check));
1291 skb_reset_network_header(skb);
1292 }
1293}
1294
1295static int ibmveth_poll(struct napi_struct *napi, int budget)
1296{
1297 struct ibmveth_adapter *adapter =
1298 container_of(napi, struct ibmveth_adapter, napi);
1299 struct net_device *netdev = adapter->netdev;
1300 int frames_processed = 0;
1301 unsigned long lpar_rc;
1302 u16 mss = 0;
1303
1304 while (frames_processed < budget) {
1305 if (!ibmveth_rxq_pending_buffer(adapter))
1306 break;
1307
1308 smp_rmb();
1309 if (!ibmveth_rxq_buffer_valid(adapter)) {
1310 wmb(); /* suggested by larson1 */
1311 adapter->rx_invalid_buffer++;
1312 netdev_dbg(netdev, "recycling invalid buffer\n");
1313 ibmveth_rxq_recycle_buffer(adapter);
1314 } else {
1315 struct sk_buff *skb, *new_skb;
1316 int length = ibmveth_rxq_frame_length(adapter);
1317 int offset = ibmveth_rxq_frame_offset(adapter);
1318 int csum_good = ibmveth_rxq_csum_good(adapter);
1319 int lrg_pkt = ibmveth_rxq_large_packet(adapter);
1320
1321 skb = ibmveth_rxq_get_buffer(adapter);
1322
1323 /* if the large packet bit is set in the rx queue
1324 * descriptor, the mss will be written by PHYP eight
1325 * bytes from the start of the rx buffer, which is
1326 * skb->data at this stage
1327 */
1328 if (lrg_pkt) {
1329 __be64 *rxmss = (__be64 *)(skb->data + 8);
1330
1331 mss = (u16)be64_to_cpu(*rxmss);
1332 }
1333
1334 new_skb = NULL;
1335 if (length < rx_copybreak)
1336 new_skb = netdev_alloc_skb(netdev, length);
1337
1338 if (new_skb) {
1339 skb_copy_to_linear_data(new_skb,
1340 skb->data + offset,
1341 length);
1342 if (rx_flush)
1343 ibmveth_flush_buffer(skb->data,
1344 length + offset);
1345 if (!ibmveth_rxq_recycle_buffer(adapter))
1346 kfree_skb(skb);
1347 skb = new_skb;
1348 } else {
1349 ibmveth_rxq_harvest_buffer(adapter);
1350 skb_reserve(skb, offset);
1351 }
1352
1353 skb_put(skb, length);
1354 skb->protocol = eth_type_trans(skb, netdev);
1355
1356 if (csum_good) {
1357 skb->ip_summed = CHECKSUM_UNNECESSARY;
1358 ibmveth_rx_csum_helper(skb, adapter);
1359 }
1360
1361 if (length > netdev->mtu + ETH_HLEN) {
1362 ibmveth_rx_mss_helper(skb, mss, lrg_pkt);
1363 adapter->rx_large_packets++;
1364 }
1365
1366 napi_gro_receive(napi, skb); /* send it up */
1367
1368 netdev->stats.rx_packets++;
1369 netdev->stats.rx_bytes += length;
1370 frames_processed++;
1371 }
1372 }
1373
1374 ibmveth_replenish_task(adapter);
1375
1376 if (frames_processed < budget) {
1377 napi_complete_done(napi, frames_processed);
1378
1379 /* We think we are done - reenable interrupts,
1380 * then check once more to make sure we are done.
1381 */
1382 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1383 VIO_IRQ_ENABLE);
1384
1385 BUG_ON(lpar_rc != H_SUCCESS);
1386
1387 if (ibmveth_rxq_pending_buffer(adapter) &&
1388 napi_reschedule(napi)) {
1389 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1390 VIO_IRQ_DISABLE);
1391 }
1392 }
1393
1394 return frames_processed;
1395}
1396
1397static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1398{
1399 struct net_device *netdev = dev_instance;
1400 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1401 unsigned long lpar_rc;
1402
1403 if (napi_schedule_prep(&adapter->napi)) {
1404 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1405 VIO_IRQ_DISABLE);
1406 BUG_ON(lpar_rc != H_SUCCESS);
1407 __napi_schedule(&adapter->napi);
1408 }
1409 return IRQ_HANDLED;
1410}
1411
1412static void ibmveth_set_multicast_list(struct net_device *netdev)
1413{
1414 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1415 unsigned long lpar_rc;
1416
1417 if ((netdev->flags & IFF_PROMISC) ||
1418 (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1419 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1420 IbmVethMcastEnableRecv |
1421 IbmVethMcastDisableFiltering,
1422 0);
1423 if (lpar_rc != H_SUCCESS) {
1424 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1425 "entering promisc mode\n", lpar_rc);
1426 }
1427 } else {
1428 struct netdev_hw_addr *ha;
1429 /* clear the filter table & disable filtering */
1430 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1431 IbmVethMcastEnableRecv |
1432 IbmVethMcastDisableFiltering |
1433 IbmVethMcastClearFilterTable,
1434 0);
1435 if (lpar_rc != H_SUCCESS) {
1436 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1437 "attempting to clear filter table\n",
1438 lpar_rc);
1439 }
1440 /* add the addresses to the filter table */
1441 netdev_for_each_mc_addr(ha, netdev) {
1442 /* add the multicast address to the filter table */
1443 u64 mcast_addr;
1444 mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1445 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1446 IbmVethMcastAddFilter,
1447 mcast_addr);
1448 if (lpar_rc != H_SUCCESS) {
1449 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1450 "when adding an entry to the filter "
1451 "table\n", lpar_rc);
1452 }
1453 }
1454
1455 /* re-enable filtering */
1456 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1457 IbmVethMcastEnableFiltering,
1458 0);
1459 if (lpar_rc != H_SUCCESS) {
1460 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1461 "enabling filtering\n", lpar_rc);
1462 }
1463 }
1464}
1465
1466static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1467{
1468 struct ibmveth_adapter *adapter = netdev_priv(dev);
1469 struct vio_dev *viodev = adapter->vdev;
1470 int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1471 int i, rc;
1472 int need_restart = 0;
1473
1474 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1475 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1476 break;
1477
1478 if (i == IBMVETH_NUM_BUFF_POOLS)
1479 return -EINVAL;
1480
1481 /* Deactivate all the buffer pools so that the next loop can activate
1482 only the buffer pools necessary to hold the new MTU */
1483 if (netif_running(adapter->netdev)) {
1484 need_restart = 1;
1485 adapter->pool_config = 1;
1486 ibmveth_close(adapter->netdev);
1487 adapter->pool_config = 0;
1488 }
1489
1490 /* Look for an active buffer pool that can hold the new MTU */
1491 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1492 adapter->rx_buff_pool[i].active = 1;
1493
1494 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1495 dev->mtu = new_mtu;
1496 vio_cmo_set_dev_desired(viodev,
1497 ibmveth_get_desired_dma
1498 (viodev));
1499 if (need_restart) {
1500 return ibmveth_open(adapter->netdev);
1501 }
1502 return 0;
1503 }
1504 }
1505
1506 if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1507 return rc;
1508
1509 return -EINVAL;
1510}
1511
1512#ifdef CONFIG_NET_POLL_CONTROLLER
1513static void ibmveth_poll_controller(struct net_device *dev)
1514{
1515 ibmveth_replenish_task(netdev_priv(dev));
1516 ibmveth_interrupt(dev->irq, dev);
1517}
1518#endif
1519
1520/**
1521 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1522 *
1523 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1524 *
1525 * Return value:
1526 * Number of bytes of IO data the driver will need to perform well.
1527 */
1528static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1529{
1530 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1531 struct ibmveth_adapter *adapter;
1532 struct iommu_table *tbl;
1533 unsigned long ret;
1534 int i;
1535 int rxqentries = 1;
1536
1537 tbl = get_iommu_table_base(&vdev->dev);
1538
1539 /* netdev inits at probe time along with the structures we need below*/
1540 if (netdev == NULL)
1541 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1542
1543 adapter = netdev_priv(netdev);
1544
1545 ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1546 ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1547
1548 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1549 /* add the size of the active receive buffers */
1550 if (adapter->rx_buff_pool[i].active)
1551 ret +=
1552 adapter->rx_buff_pool[i].size *
1553 IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1554 buff_size, tbl);
1555 rxqentries += adapter->rx_buff_pool[i].size;
1556 }
1557 /* add the size of the receive queue entries */
1558 ret += IOMMU_PAGE_ALIGN(
1559 rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1560
1561 return ret;
1562}
1563
1564static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1565{
1566 struct ibmveth_adapter *adapter = netdev_priv(dev);
1567 struct sockaddr *addr = p;
1568 u64 mac_address;
1569 int rc;
1570
1571 if (!is_valid_ether_addr(addr->sa_data))
1572 return -EADDRNOTAVAIL;
1573
1574 mac_address = ibmveth_encode_mac_addr(addr->sa_data);
1575 rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1576 if (rc) {
1577 netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1578 return rc;
1579 }
1580
1581 ether_addr_copy(dev->dev_addr, addr->sa_data);
1582
1583 return 0;
1584}
1585
1586static const struct net_device_ops ibmveth_netdev_ops = {
1587 .ndo_open = ibmveth_open,
1588 .ndo_stop = ibmveth_close,
1589 .ndo_start_xmit = ibmveth_start_xmit,
1590 .ndo_set_rx_mode = ibmveth_set_multicast_list,
1591 .ndo_do_ioctl = ibmveth_ioctl,
1592 .ndo_change_mtu = ibmveth_change_mtu,
1593 .ndo_fix_features = ibmveth_fix_features,
1594 .ndo_set_features = ibmveth_set_features,
1595 .ndo_validate_addr = eth_validate_addr,
1596 .ndo_set_mac_address = ibmveth_set_mac_addr,
1597#ifdef CONFIG_NET_POLL_CONTROLLER
1598 .ndo_poll_controller = ibmveth_poll_controller,
1599#endif
1600};
1601
1602static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1603{
1604 int rc, i, mac_len;
1605 struct net_device *netdev;
1606 struct ibmveth_adapter *adapter;
1607 unsigned char *mac_addr_p;
1608 __be32 *mcastFilterSize_p;
1609 long ret;
1610 unsigned long ret_attr;
1611
1612 dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1613 dev->unit_address);
1614
1615 mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1616 &mac_len);
1617 if (!mac_addr_p) {
1618 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1619 return -EINVAL;
1620 }
1621 /* Workaround for old/broken pHyp */
1622 if (mac_len == 8)
1623 mac_addr_p += 2;
1624 else if (mac_len != 6) {
1625 dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1626 mac_len);
1627 return -EINVAL;
1628 }
1629
1630 mcastFilterSize_p = (__be32 *)vio_get_attribute(dev,
1631 VETH_MCAST_FILTER_SIZE,
1632 NULL);
1633 if (!mcastFilterSize_p) {
1634 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1635 "attribute\n");
1636 return -EINVAL;
1637 }
1638
1639 netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1640
1641 if (!netdev)
1642 return -ENOMEM;
1643
1644 adapter = netdev_priv(netdev);
1645 dev_set_drvdata(&dev->dev, netdev);
1646
1647 adapter->vdev = dev;
1648 adapter->netdev = netdev;
1649 adapter->mcastFilterSize = be32_to_cpu(*mcastFilterSize_p);
1650 adapter->pool_config = 0;
1651
1652 netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1653
1654 netdev->irq = dev->irq;
1655 netdev->netdev_ops = &ibmveth_netdev_ops;
1656 netdev->ethtool_ops = &netdev_ethtool_ops;
1657 SET_NETDEV_DEV(netdev, &dev->dev);
1658 netdev->hw_features = NETIF_F_SG;
1659 if (vio_get_attribute(dev, "ibm,illan-options", NULL) != NULL) {
1660 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1661 NETIF_F_RXCSUM;
1662 }
1663
1664 netdev->features |= netdev->hw_features;
1665
1666 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
1667
1668 /* If running older firmware, TSO should not be enabled by default */
1669 if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
1670 !old_large_send) {
1671 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1672 netdev->features |= netdev->hw_features;
1673 } else {
1674 netdev->hw_features |= NETIF_F_TSO;
1675 }
1676
1677 adapter->is_active_trunk = false;
1678 if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK)) {
1679 adapter->is_active_trunk = true;
1680 netdev->hw_features |= NETIF_F_FRAGLIST;
1681 netdev->features |= NETIF_F_FRAGLIST;
1682 }
1683
1684 netdev->min_mtu = IBMVETH_MIN_MTU;
1685 netdev->max_mtu = ETH_MAX_MTU;
1686
1687 memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1688
1689 if (firmware_has_feature(FW_FEATURE_CMO))
1690 memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1691
1692 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1693 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1694 int error;
1695
1696 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1697 pool_count[i], pool_size[i],
1698 pool_active[i]);
1699 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1700 &dev->dev.kobj, "pool%d", i);
1701 if (!error)
1702 kobject_uevent(kobj, KOBJ_ADD);
1703 }
1704
1705 netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1706 netdev_dbg(netdev, "registering netdev...\n");
1707
1708 ibmveth_set_features(netdev, netdev->features);
1709
1710 rc = register_netdev(netdev);
1711
1712 if (rc) {
1713 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1714 free_netdev(netdev);
1715 return rc;
1716 }
1717
1718 netdev_dbg(netdev, "registered\n");
1719
1720 return 0;
1721}
1722
1723static int ibmveth_remove(struct vio_dev *dev)
1724{
1725 struct net_device *netdev = dev_get_drvdata(&dev->dev);
1726 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1727 int i;
1728
1729 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1730 kobject_put(&adapter->rx_buff_pool[i].kobj);
1731
1732 unregister_netdev(netdev);
1733
1734 free_netdev(netdev);
1735 dev_set_drvdata(&dev->dev, NULL);
1736
1737 return 0;
1738}
1739
1740static struct attribute veth_active_attr;
1741static struct attribute veth_num_attr;
1742static struct attribute veth_size_attr;
1743
1744static ssize_t veth_pool_show(struct kobject *kobj,
1745 struct attribute *attr, char *buf)
1746{
1747 struct ibmveth_buff_pool *pool = container_of(kobj,
1748 struct ibmveth_buff_pool,
1749 kobj);
1750
1751 if (attr == &veth_active_attr)
1752 return sprintf(buf, "%d\n", pool->active);
1753 else if (attr == &veth_num_attr)
1754 return sprintf(buf, "%d\n", pool->size);
1755 else if (attr == &veth_size_attr)
1756 return sprintf(buf, "%d\n", pool->buff_size);
1757 return 0;
1758}
1759
1760static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1761 const char *buf, size_t count)
1762{
1763 struct ibmveth_buff_pool *pool = container_of(kobj,
1764 struct ibmveth_buff_pool,
1765 kobj);
1766 struct net_device *netdev = dev_get_drvdata(
1767 container_of(kobj->parent, struct device, kobj));
1768 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1769 long value = simple_strtol(buf, NULL, 10);
1770 long rc;
1771
1772 if (attr == &veth_active_attr) {
1773 if (value && !pool->active) {
1774 if (netif_running(netdev)) {
1775 if (ibmveth_alloc_buffer_pool(pool)) {
1776 netdev_err(netdev,
1777 "unable to alloc pool\n");
1778 return -ENOMEM;
1779 }
1780 pool->active = 1;
1781 adapter->pool_config = 1;
1782 ibmveth_close(netdev);
1783 adapter->pool_config = 0;
1784 if ((rc = ibmveth_open(netdev)))
1785 return rc;
1786 } else {
1787 pool->active = 1;
1788 }
1789 } else if (!value && pool->active) {
1790 int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1791 int i;
1792 /* Make sure there is a buffer pool with buffers that
1793 can hold a packet of the size of the MTU */
1794 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1795 if (pool == &adapter->rx_buff_pool[i])
1796 continue;
1797 if (!adapter->rx_buff_pool[i].active)
1798 continue;
1799 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1800 break;
1801 }
1802
1803 if (i == IBMVETH_NUM_BUFF_POOLS) {
1804 netdev_err(netdev, "no active pool >= MTU\n");
1805 return -EPERM;
1806 }
1807
1808 if (netif_running(netdev)) {
1809 adapter->pool_config = 1;
1810 ibmveth_close(netdev);
1811 pool->active = 0;
1812 adapter->pool_config = 0;
1813 if ((rc = ibmveth_open(netdev)))
1814 return rc;
1815 }
1816 pool->active = 0;
1817 }
1818 } else if (attr == &veth_num_attr) {
1819 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1820 return -EINVAL;
1821 } else {
1822 if (netif_running(netdev)) {
1823 adapter->pool_config = 1;
1824 ibmveth_close(netdev);
1825 adapter->pool_config = 0;
1826 pool->size = value;
1827 if ((rc = ibmveth_open(netdev)))
1828 return rc;
1829 } else {
1830 pool->size = value;
1831 }
1832 }
1833 } else if (attr == &veth_size_attr) {
1834 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1835 return -EINVAL;
1836 } else {
1837 if (netif_running(netdev)) {
1838 adapter->pool_config = 1;
1839 ibmveth_close(netdev);
1840 adapter->pool_config = 0;
1841 pool->buff_size = value;
1842 if ((rc = ibmveth_open(netdev)))
1843 return rc;
1844 } else {
1845 pool->buff_size = value;
1846 }
1847 }
1848 }
1849
1850 /* kick the interrupt handler to allocate/deallocate pools */
1851 ibmveth_interrupt(netdev->irq, netdev);
1852 return count;
1853}
1854
1855
1856#define ATTR(_name, _mode) \
1857 struct attribute veth_##_name##_attr = { \
1858 .name = __stringify(_name), .mode = _mode, \
1859 };
1860
1861static ATTR(active, 0644);
1862static ATTR(num, 0644);
1863static ATTR(size, 0644);
1864
1865static struct attribute *veth_pool_attrs[] = {
1866 &veth_active_attr,
1867 &veth_num_attr,
1868 &veth_size_attr,
1869 NULL,
1870};
1871
1872static const struct sysfs_ops veth_pool_ops = {
1873 .show = veth_pool_show,
1874 .store = veth_pool_store,
1875};
1876
1877static struct kobj_type ktype_veth_pool = {
1878 .release = NULL,
1879 .sysfs_ops = &veth_pool_ops,
1880 .default_attrs = veth_pool_attrs,
1881};
1882
1883static int ibmveth_resume(struct device *dev)
1884{
1885 struct net_device *netdev = dev_get_drvdata(dev);
1886 ibmveth_interrupt(netdev->irq, netdev);
1887 return 0;
1888}
1889
1890static const struct vio_device_id ibmveth_device_table[] = {
1891 { "network", "IBM,l-lan"},
1892 { "", "" }
1893};
1894MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1895
1896static const struct dev_pm_ops ibmveth_pm_ops = {
1897 .resume = ibmveth_resume
1898};
1899
1900static struct vio_driver ibmveth_driver = {
1901 .id_table = ibmveth_device_table,
1902 .probe = ibmveth_probe,
1903 .remove = ibmveth_remove,
1904 .get_desired_dma = ibmveth_get_desired_dma,
1905 .name = ibmveth_driver_name,
1906 .pm = &ibmveth_pm_ops,
1907};
1908
1909static int __init ibmveth_module_init(void)
1910{
1911 printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1912 ibmveth_driver_string, ibmveth_driver_version);
1913
1914 return vio_register_driver(&ibmveth_driver);
1915}
1916
1917static void __exit ibmveth_module_exit(void)
1918{
1919 vio_unregister_driver(&ibmveth_driver);
1920}
1921
1922module_init(ibmveth_module_init);
1923module_exit(ibmveth_module_exit);
1/*
2 * IBM Power Virtual Ethernet Device Driver
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, see <http://www.gnu.org/licenses/>.
16 *
17 * Copyright (C) IBM Corporation, 2003, 2010
18 *
19 * Authors: Dave Larson <larson1@us.ibm.com>
20 * Santiago Leon <santil@linux.vnet.ibm.com>
21 * Brian King <brking@linux.vnet.ibm.com>
22 * Robert Jennings <rcj@linux.vnet.ibm.com>
23 * Anton Blanchard <anton@au.ibm.com>
24 */
25
26#include <linux/module.h>
27#include <linux/moduleparam.h>
28#include <linux/types.h>
29#include <linux/errno.h>
30#include <linux/dma-mapping.h>
31#include <linux/kernel.h>
32#include <linux/netdevice.h>
33#include <linux/etherdevice.h>
34#include <linux/skbuff.h>
35#include <linux/init.h>
36#include <linux/interrupt.h>
37#include <linux/mm.h>
38#include <linux/pm.h>
39#include <linux/ethtool.h>
40#include <linux/in.h>
41#include <linux/ip.h>
42#include <linux/ipv6.h>
43#include <linux/slab.h>
44#include <asm/hvcall.h>
45#include <linux/atomic.h>
46#include <asm/vio.h>
47#include <asm/iommu.h>
48#include <asm/firmware.h>
49
50#include "ibmveth.h"
51
52static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
53static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
54static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
55
56static struct kobj_type ktype_veth_pool;
57
58
59static const char ibmveth_driver_name[] = "ibmveth";
60static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
61#define ibmveth_driver_version "1.06"
62
63MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
64MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
65MODULE_LICENSE("GPL");
66MODULE_VERSION(ibmveth_driver_version);
67
68static unsigned int tx_copybreak __read_mostly = 128;
69module_param(tx_copybreak, uint, 0644);
70MODULE_PARM_DESC(tx_copybreak,
71 "Maximum size of packet that is copied to a new buffer on transmit");
72
73static unsigned int rx_copybreak __read_mostly = 128;
74module_param(rx_copybreak, uint, 0644);
75MODULE_PARM_DESC(rx_copybreak,
76 "Maximum size of packet that is copied to a new buffer on receive");
77
78static unsigned int rx_flush __read_mostly = 0;
79module_param(rx_flush, uint, 0644);
80MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
81
82static bool old_large_send __read_mostly;
83module_param(old_large_send, bool, S_IRUGO);
84MODULE_PARM_DESC(old_large_send,
85 "Use old large send method on firmware that supports the new method");
86
87struct ibmveth_stat {
88 char name[ETH_GSTRING_LEN];
89 int offset;
90};
91
92#define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
93#define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
94
95struct ibmveth_stat ibmveth_stats[] = {
96 { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
97 { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
98 { "replenish_add_buff_failure",
99 IBMVETH_STAT_OFF(replenish_add_buff_failure) },
100 { "replenish_add_buff_success",
101 IBMVETH_STAT_OFF(replenish_add_buff_success) },
102 { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
103 { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
104 { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
105 { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
106 { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
107 { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
108 { "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },
109 { "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) },
110 { "fw_enabled_large_send", IBMVETH_STAT_OFF(fw_large_send_support) }
111};
112
113/* simple methods of getting data from the current rxq entry */
114static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
115{
116 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
117}
118
119static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
120{
121 return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
122 IBMVETH_RXQ_TOGGLE_SHIFT;
123}
124
125static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
126{
127 return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
128}
129
130static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
131{
132 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
133}
134
135static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
136{
137 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
138}
139
140static inline int ibmveth_rxq_large_packet(struct ibmveth_adapter *adapter)
141{
142 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_LRG_PKT;
143}
144
145static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
146{
147 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
148}
149
150static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
151{
152 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
153}
154
155/* setup the initial settings for a buffer pool */
156static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
157 u32 pool_index, u32 pool_size,
158 u32 buff_size, u32 pool_active)
159{
160 pool->size = pool_size;
161 pool->index = pool_index;
162 pool->buff_size = buff_size;
163 pool->threshold = pool_size * 7 / 8;
164 pool->active = pool_active;
165}
166
167/* allocate and setup an buffer pool - called during open */
168static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
169{
170 int i;
171
172 pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
173
174 if (!pool->free_map)
175 return -1;
176
177 pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), GFP_KERNEL);
178 if (!pool->dma_addr) {
179 kfree(pool->free_map);
180 pool->free_map = NULL;
181 return -1;
182 }
183
184 pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
185
186 if (!pool->skbuff) {
187 kfree(pool->dma_addr);
188 pool->dma_addr = NULL;
189
190 kfree(pool->free_map);
191 pool->free_map = NULL;
192 return -1;
193 }
194
195 for (i = 0; i < pool->size; ++i)
196 pool->free_map[i] = i;
197
198 atomic_set(&pool->available, 0);
199 pool->producer_index = 0;
200 pool->consumer_index = 0;
201
202 return 0;
203}
204
205static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
206{
207 unsigned long offset;
208
209 for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
210 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
211}
212
213/* replenish the buffers for a pool. note that we don't need to
214 * skb_reserve these since they are used for incoming...
215 */
216static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
217 struct ibmveth_buff_pool *pool)
218{
219 u32 i;
220 u32 count = pool->size - atomic_read(&pool->available);
221 u32 buffers_added = 0;
222 struct sk_buff *skb;
223 unsigned int free_index, index;
224 u64 correlator;
225 unsigned long lpar_rc;
226 dma_addr_t dma_addr;
227
228 mb();
229
230 for (i = 0; i < count; ++i) {
231 union ibmveth_buf_desc desc;
232
233 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
234
235 if (!skb) {
236 netdev_dbg(adapter->netdev,
237 "replenish: unable to allocate skb\n");
238 adapter->replenish_no_mem++;
239 break;
240 }
241
242 free_index = pool->consumer_index;
243 pool->consumer_index++;
244 if (pool->consumer_index >= pool->size)
245 pool->consumer_index = 0;
246 index = pool->free_map[free_index];
247
248 BUG_ON(index == IBM_VETH_INVALID_MAP);
249 BUG_ON(pool->skbuff[index] != NULL);
250
251 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
252 pool->buff_size, DMA_FROM_DEVICE);
253
254 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
255 goto failure;
256
257 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
258 pool->dma_addr[index] = dma_addr;
259 pool->skbuff[index] = skb;
260
261 correlator = ((u64)pool->index << 32) | index;
262 *(u64 *)skb->data = correlator;
263
264 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
265 desc.fields.address = dma_addr;
266
267 if (rx_flush) {
268 unsigned int len = min(pool->buff_size,
269 adapter->netdev->mtu +
270 IBMVETH_BUFF_OH);
271 ibmveth_flush_buffer(skb->data, len);
272 }
273 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
274 desc.desc);
275
276 if (lpar_rc != H_SUCCESS) {
277 goto failure;
278 } else {
279 buffers_added++;
280 adapter->replenish_add_buff_success++;
281 }
282 }
283
284 mb();
285 atomic_add(buffers_added, &(pool->available));
286 return;
287
288failure:
289 pool->free_map[free_index] = index;
290 pool->skbuff[index] = NULL;
291 if (pool->consumer_index == 0)
292 pool->consumer_index = pool->size - 1;
293 else
294 pool->consumer_index--;
295 if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
296 dma_unmap_single(&adapter->vdev->dev,
297 pool->dma_addr[index], pool->buff_size,
298 DMA_FROM_DEVICE);
299 dev_kfree_skb_any(skb);
300 adapter->replenish_add_buff_failure++;
301
302 mb();
303 atomic_add(buffers_added, &(pool->available));
304}
305
306/*
307 * The final 8 bytes of the buffer list is a counter of frames dropped
308 * because there was not a buffer in the buffer list capable of holding
309 * the frame.
310 */
311static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
312{
313 __be64 *p = adapter->buffer_list_addr + 4096 - 8;
314
315 adapter->rx_no_buffer = be64_to_cpup(p);
316}
317
318/* replenish routine */
319static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
320{
321 int i;
322
323 adapter->replenish_task_cycles++;
324
325 for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
326 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
327
328 if (pool->active &&
329 (atomic_read(&pool->available) < pool->threshold))
330 ibmveth_replenish_buffer_pool(adapter, pool);
331 }
332
333 ibmveth_update_rx_no_buffer(adapter);
334}
335
336/* empty and free ana buffer pool - also used to do cleanup in error paths */
337static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
338 struct ibmveth_buff_pool *pool)
339{
340 int i;
341
342 kfree(pool->free_map);
343 pool->free_map = NULL;
344
345 if (pool->skbuff && pool->dma_addr) {
346 for (i = 0; i < pool->size; ++i) {
347 struct sk_buff *skb = pool->skbuff[i];
348 if (skb) {
349 dma_unmap_single(&adapter->vdev->dev,
350 pool->dma_addr[i],
351 pool->buff_size,
352 DMA_FROM_DEVICE);
353 dev_kfree_skb_any(skb);
354 pool->skbuff[i] = NULL;
355 }
356 }
357 }
358
359 if (pool->dma_addr) {
360 kfree(pool->dma_addr);
361 pool->dma_addr = NULL;
362 }
363
364 if (pool->skbuff) {
365 kfree(pool->skbuff);
366 pool->skbuff = NULL;
367 }
368}
369
370/* remove a buffer from a pool */
371static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
372 u64 correlator)
373{
374 unsigned int pool = correlator >> 32;
375 unsigned int index = correlator & 0xffffffffUL;
376 unsigned int free_index;
377 struct sk_buff *skb;
378
379 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
380 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
381
382 skb = adapter->rx_buff_pool[pool].skbuff[index];
383
384 BUG_ON(skb == NULL);
385
386 adapter->rx_buff_pool[pool].skbuff[index] = NULL;
387
388 dma_unmap_single(&adapter->vdev->dev,
389 adapter->rx_buff_pool[pool].dma_addr[index],
390 adapter->rx_buff_pool[pool].buff_size,
391 DMA_FROM_DEVICE);
392
393 free_index = adapter->rx_buff_pool[pool].producer_index;
394 adapter->rx_buff_pool[pool].producer_index++;
395 if (adapter->rx_buff_pool[pool].producer_index >=
396 adapter->rx_buff_pool[pool].size)
397 adapter->rx_buff_pool[pool].producer_index = 0;
398 adapter->rx_buff_pool[pool].free_map[free_index] = index;
399
400 mb();
401
402 atomic_dec(&(adapter->rx_buff_pool[pool].available));
403}
404
405/* get the current buffer on the rx queue */
406static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
407{
408 u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
409 unsigned int pool = correlator >> 32;
410 unsigned int index = correlator & 0xffffffffUL;
411
412 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
413 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
414
415 return adapter->rx_buff_pool[pool].skbuff[index];
416}
417
418/* recycle the current buffer on the rx queue */
419static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
420{
421 u32 q_index = adapter->rx_queue.index;
422 u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
423 unsigned int pool = correlator >> 32;
424 unsigned int index = correlator & 0xffffffffUL;
425 union ibmveth_buf_desc desc;
426 unsigned long lpar_rc;
427 int ret = 1;
428
429 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
430 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
431
432 if (!adapter->rx_buff_pool[pool].active) {
433 ibmveth_rxq_harvest_buffer(adapter);
434 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
435 goto out;
436 }
437
438 desc.fields.flags_len = IBMVETH_BUF_VALID |
439 adapter->rx_buff_pool[pool].buff_size;
440 desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
441
442 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
443
444 if (lpar_rc != H_SUCCESS) {
445 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
446 "during recycle rc=%ld", lpar_rc);
447 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
448 ret = 0;
449 }
450
451 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
452 adapter->rx_queue.index = 0;
453 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
454 }
455
456out:
457 return ret;
458}
459
460static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
461{
462 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
463
464 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
465 adapter->rx_queue.index = 0;
466 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
467 }
468}
469
470static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
471{
472 int i;
473 struct device *dev = &adapter->vdev->dev;
474
475 if (adapter->buffer_list_addr != NULL) {
476 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
477 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
478 DMA_BIDIRECTIONAL);
479 adapter->buffer_list_dma = DMA_ERROR_CODE;
480 }
481 free_page((unsigned long)adapter->buffer_list_addr);
482 adapter->buffer_list_addr = NULL;
483 }
484
485 if (adapter->filter_list_addr != NULL) {
486 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
487 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
488 DMA_BIDIRECTIONAL);
489 adapter->filter_list_dma = DMA_ERROR_CODE;
490 }
491 free_page((unsigned long)adapter->filter_list_addr);
492 adapter->filter_list_addr = NULL;
493 }
494
495 if (adapter->rx_queue.queue_addr != NULL) {
496 dma_free_coherent(dev, adapter->rx_queue.queue_len,
497 adapter->rx_queue.queue_addr,
498 adapter->rx_queue.queue_dma);
499 adapter->rx_queue.queue_addr = NULL;
500 }
501
502 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
503 if (adapter->rx_buff_pool[i].active)
504 ibmveth_free_buffer_pool(adapter,
505 &adapter->rx_buff_pool[i]);
506
507 if (adapter->bounce_buffer != NULL) {
508 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
509 dma_unmap_single(&adapter->vdev->dev,
510 adapter->bounce_buffer_dma,
511 adapter->netdev->mtu + IBMVETH_BUFF_OH,
512 DMA_BIDIRECTIONAL);
513 adapter->bounce_buffer_dma = DMA_ERROR_CODE;
514 }
515 kfree(adapter->bounce_buffer);
516 adapter->bounce_buffer = NULL;
517 }
518}
519
520static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
521 union ibmveth_buf_desc rxq_desc, u64 mac_address)
522{
523 int rc, try_again = 1;
524
525 /*
526 * After a kexec the adapter will still be open, so our attempt to
527 * open it will fail. So if we get a failure we free the adapter and
528 * try again, but only once.
529 */
530retry:
531 rc = h_register_logical_lan(adapter->vdev->unit_address,
532 adapter->buffer_list_dma, rxq_desc.desc,
533 adapter->filter_list_dma, mac_address);
534
535 if (rc != H_SUCCESS && try_again) {
536 do {
537 rc = h_free_logical_lan(adapter->vdev->unit_address);
538 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
539
540 try_again = 0;
541 goto retry;
542 }
543
544 return rc;
545}
546
547static u64 ibmveth_encode_mac_addr(u8 *mac)
548{
549 int i;
550 u64 encoded = 0;
551
552 for (i = 0; i < ETH_ALEN; i++)
553 encoded = (encoded << 8) | mac[i];
554
555 return encoded;
556}
557
558static int ibmveth_open(struct net_device *netdev)
559{
560 struct ibmveth_adapter *adapter = netdev_priv(netdev);
561 u64 mac_address;
562 int rxq_entries = 1;
563 unsigned long lpar_rc;
564 int rc;
565 union ibmveth_buf_desc rxq_desc;
566 int i;
567 struct device *dev;
568
569 netdev_dbg(netdev, "open starting\n");
570
571 napi_enable(&adapter->napi);
572
573 for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
574 rxq_entries += adapter->rx_buff_pool[i].size;
575
576 adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
577 adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
578
579 if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
580 netdev_err(netdev, "unable to allocate filter or buffer list "
581 "pages\n");
582 rc = -ENOMEM;
583 goto err_out;
584 }
585
586 dev = &adapter->vdev->dev;
587
588 adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
589 rxq_entries;
590 adapter->rx_queue.queue_addr =
591 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
592 &adapter->rx_queue.queue_dma, GFP_KERNEL);
593 if (!adapter->rx_queue.queue_addr) {
594 rc = -ENOMEM;
595 goto err_out;
596 }
597
598 adapter->buffer_list_dma = dma_map_single(dev,
599 adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
600 adapter->filter_list_dma = dma_map_single(dev,
601 adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
602
603 if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
604 (dma_mapping_error(dev, adapter->filter_list_dma))) {
605 netdev_err(netdev, "unable to map filter or buffer list "
606 "pages\n");
607 rc = -ENOMEM;
608 goto err_out;
609 }
610
611 adapter->rx_queue.index = 0;
612 adapter->rx_queue.num_slots = rxq_entries;
613 adapter->rx_queue.toggle = 1;
614
615 mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
616
617 rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
618 adapter->rx_queue.queue_len;
619 rxq_desc.fields.address = adapter->rx_queue.queue_dma;
620
621 netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
622 netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
623 netdev_dbg(netdev, "receive q @ 0x%p\n", adapter->rx_queue.queue_addr);
624
625 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
626
627 lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
628
629 if (lpar_rc != H_SUCCESS) {
630 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
631 lpar_rc);
632 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
633 "desc:0x%llx MAC:0x%llx\n",
634 adapter->buffer_list_dma,
635 adapter->filter_list_dma,
636 rxq_desc.desc,
637 mac_address);
638 rc = -ENONET;
639 goto err_out;
640 }
641
642 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
643 if (!adapter->rx_buff_pool[i].active)
644 continue;
645 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
646 netdev_err(netdev, "unable to alloc pool\n");
647 adapter->rx_buff_pool[i].active = 0;
648 rc = -ENOMEM;
649 goto err_out;
650 }
651 }
652
653 netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
654 rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
655 netdev);
656 if (rc != 0) {
657 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
658 netdev->irq, rc);
659 do {
660 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
661 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
662
663 goto err_out;
664 }
665
666 adapter->bounce_buffer =
667 kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
668 if (!adapter->bounce_buffer) {
669 rc = -ENOMEM;
670 goto err_out_free_irq;
671 }
672 adapter->bounce_buffer_dma =
673 dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
674 netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
675 if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
676 netdev_err(netdev, "unable to map bounce buffer\n");
677 rc = -ENOMEM;
678 goto err_out_free_irq;
679 }
680
681 netdev_dbg(netdev, "initial replenish cycle\n");
682 ibmveth_interrupt(netdev->irq, netdev);
683
684 netif_start_queue(netdev);
685
686 netdev_dbg(netdev, "open complete\n");
687
688 return 0;
689
690err_out_free_irq:
691 free_irq(netdev->irq, netdev);
692err_out:
693 ibmveth_cleanup(adapter);
694 napi_disable(&adapter->napi);
695 return rc;
696}
697
698static int ibmveth_close(struct net_device *netdev)
699{
700 struct ibmveth_adapter *adapter = netdev_priv(netdev);
701 long lpar_rc;
702
703 netdev_dbg(netdev, "close starting\n");
704
705 napi_disable(&adapter->napi);
706
707 if (!adapter->pool_config)
708 netif_stop_queue(netdev);
709
710 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
711
712 do {
713 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
714 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
715
716 if (lpar_rc != H_SUCCESS) {
717 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
718 "continuing with close\n", lpar_rc);
719 }
720
721 free_irq(netdev->irq, netdev);
722
723 ibmveth_update_rx_no_buffer(adapter);
724
725 ibmveth_cleanup(adapter);
726
727 netdev_dbg(netdev, "close complete\n");
728
729 return 0;
730}
731
732static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
733{
734 cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
735 SUPPORTED_FIBRE);
736 cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
737 ADVERTISED_FIBRE);
738 ethtool_cmd_speed_set(cmd, SPEED_1000);
739 cmd->duplex = DUPLEX_FULL;
740 cmd->port = PORT_FIBRE;
741 cmd->phy_address = 0;
742 cmd->transceiver = XCVR_INTERNAL;
743 cmd->autoneg = AUTONEG_ENABLE;
744 cmd->maxtxpkt = 0;
745 cmd->maxrxpkt = 1;
746 return 0;
747}
748
749static void netdev_get_drvinfo(struct net_device *dev,
750 struct ethtool_drvinfo *info)
751{
752 strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
753 strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
754}
755
756static netdev_features_t ibmveth_fix_features(struct net_device *dev,
757 netdev_features_t features)
758{
759 /*
760 * Since the ibmveth firmware interface does not have the
761 * concept of separate tx/rx checksum offload enable, if rx
762 * checksum is disabled we also have to disable tx checksum
763 * offload. Once we disable rx checksum offload, we are no
764 * longer allowed to send tx buffers that are not properly
765 * checksummed.
766 */
767
768 if (!(features & NETIF_F_RXCSUM))
769 features &= ~NETIF_F_CSUM_MASK;
770
771 return features;
772}
773
774static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
775{
776 struct ibmveth_adapter *adapter = netdev_priv(dev);
777 unsigned long set_attr, clr_attr, ret_attr;
778 unsigned long set_attr6, clr_attr6;
779 long ret, ret4, ret6;
780 int rc1 = 0, rc2 = 0;
781 int restart = 0;
782
783 if (netif_running(dev)) {
784 restart = 1;
785 adapter->pool_config = 1;
786 ibmveth_close(dev);
787 adapter->pool_config = 0;
788 }
789
790 set_attr = 0;
791 clr_attr = 0;
792 set_attr6 = 0;
793 clr_attr6 = 0;
794
795 if (data) {
796 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
797 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
798 } else {
799 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
800 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
801 }
802
803 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
804
805 if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
806 !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
807 (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
808 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
809 set_attr, &ret_attr);
810
811 if (ret4 != H_SUCCESS) {
812 netdev_err(dev, "unable to change IPv4 checksum "
813 "offload settings. %d rc=%ld\n",
814 data, ret4);
815
816 h_illan_attributes(adapter->vdev->unit_address,
817 set_attr, clr_attr, &ret_attr);
818
819 if (data == 1)
820 dev->features &= ~NETIF_F_IP_CSUM;
821
822 } else {
823 adapter->fw_ipv4_csum_support = data;
824 }
825
826 ret6 = h_illan_attributes(adapter->vdev->unit_address,
827 clr_attr6, set_attr6, &ret_attr);
828
829 if (ret6 != H_SUCCESS) {
830 netdev_err(dev, "unable to change IPv6 checksum "
831 "offload settings. %d rc=%ld\n",
832 data, ret6);
833
834 h_illan_attributes(adapter->vdev->unit_address,
835 set_attr6, clr_attr6, &ret_attr);
836
837 if (data == 1)
838 dev->features &= ~NETIF_F_IPV6_CSUM;
839
840 } else
841 adapter->fw_ipv6_csum_support = data;
842
843 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
844 adapter->rx_csum = data;
845 else
846 rc1 = -EIO;
847 } else {
848 rc1 = -EIO;
849 netdev_err(dev, "unable to change checksum offload settings."
850 " %d rc=%ld ret_attr=%lx\n", data, ret,
851 ret_attr);
852 }
853
854 if (restart)
855 rc2 = ibmveth_open(dev);
856
857 return rc1 ? rc1 : rc2;
858}
859
860static int ibmveth_set_tso(struct net_device *dev, u32 data)
861{
862 struct ibmveth_adapter *adapter = netdev_priv(dev);
863 unsigned long set_attr, clr_attr, ret_attr;
864 long ret1, ret2;
865 int rc1 = 0, rc2 = 0;
866 int restart = 0;
867
868 if (netif_running(dev)) {
869 restart = 1;
870 adapter->pool_config = 1;
871 ibmveth_close(dev);
872 adapter->pool_config = 0;
873 }
874
875 set_attr = 0;
876 clr_attr = 0;
877
878 if (data)
879 set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
880 else
881 clr_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
882
883 ret1 = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
884
885 if (ret1 == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
886 !old_large_send) {
887 ret2 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
888 set_attr, &ret_attr);
889
890 if (ret2 != H_SUCCESS) {
891 netdev_err(dev, "unable to change tso settings. %d rc=%ld\n",
892 data, ret2);
893
894 h_illan_attributes(adapter->vdev->unit_address,
895 set_attr, clr_attr, &ret_attr);
896
897 if (data == 1)
898 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
899 rc1 = -EIO;
900
901 } else {
902 adapter->fw_large_send_support = data;
903 adapter->large_send = data;
904 }
905 } else {
906 /* Older firmware version of large send offload does not
907 * support tcp6/ipv6
908 */
909 if (data == 1) {
910 dev->features &= ~NETIF_F_TSO6;
911 netdev_info(dev, "TSO feature requires all partitions to have updated driver");
912 }
913 adapter->large_send = data;
914 }
915
916 if (restart)
917 rc2 = ibmveth_open(dev);
918
919 return rc1 ? rc1 : rc2;
920}
921
922static int ibmveth_set_features(struct net_device *dev,
923 netdev_features_t features)
924{
925 struct ibmveth_adapter *adapter = netdev_priv(dev);
926 int rx_csum = !!(features & NETIF_F_RXCSUM);
927 int large_send = !!(features & (NETIF_F_TSO | NETIF_F_TSO6));
928 int rc1 = 0, rc2 = 0;
929
930 if (rx_csum != adapter->rx_csum) {
931 rc1 = ibmveth_set_csum_offload(dev, rx_csum);
932 if (rc1 && !adapter->rx_csum)
933 dev->features =
934 features & ~(NETIF_F_CSUM_MASK |
935 NETIF_F_RXCSUM);
936 }
937
938 if (large_send != adapter->large_send) {
939 rc2 = ibmveth_set_tso(dev, large_send);
940 if (rc2 && !adapter->large_send)
941 dev->features =
942 features & ~(NETIF_F_TSO | NETIF_F_TSO6);
943 }
944
945 return rc1 ? rc1 : rc2;
946}
947
948static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
949{
950 int i;
951
952 if (stringset != ETH_SS_STATS)
953 return;
954
955 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
956 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
957}
958
959static int ibmveth_get_sset_count(struct net_device *dev, int sset)
960{
961 switch (sset) {
962 case ETH_SS_STATS:
963 return ARRAY_SIZE(ibmveth_stats);
964 default:
965 return -EOPNOTSUPP;
966 }
967}
968
969static void ibmveth_get_ethtool_stats(struct net_device *dev,
970 struct ethtool_stats *stats, u64 *data)
971{
972 int i;
973 struct ibmveth_adapter *adapter = netdev_priv(dev);
974
975 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
976 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
977}
978
979static const struct ethtool_ops netdev_ethtool_ops = {
980 .get_drvinfo = netdev_get_drvinfo,
981 .get_settings = netdev_get_settings,
982 .get_link = ethtool_op_get_link,
983 .get_strings = ibmveth_get_strings,
984 .get_sset_count = ibmveth_get_sset_count,
985 .get_ethtool_stats = ibmveth_get_ethtool_stats,
986};
987
988static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
989{
990 return -EOPNOTSUPP;
991}
992
993#define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
994
995static int ibmveth_send(struct ibmveth_adapter *adapter,
996 union ibmveth_buf_desc *descs, unsigned long mss)
997{
998 unsigned long correlator;
999 unsigned int retry_count;
1000 unsigned long ret;
1001
1002 /*
1003 * The retry count sets a maximum for the number of broadcast and
1004 * multicast destinations within the system.
1005 */
1006 retry_count = 1024;
1007 correlator = 0;
1008 do {
1009 ret = h_send_logical_lan(adapter->vdev->unit_address,
1010 descs[0].desc, descs[1].desc,
1011 descs[2].desc, descs[3].desc,
1012 descs[4].desc, descs[5].desc,
1013 correlator, &correlator, mss,
1014 adapter->fw_large_send_support);
1015 } while ((ret == H_BUSY) && (retry_count--));
1016
1017 if (ret != H_SUCCESS && ret != H_DROPPED) {
1018 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
1019 "with rc=%ld\n", ret);
1020 return 1;
1021 }
1022
1023 return 0;
1024}
1025
1026static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
1027 struct net_device *netdev)
1028{
1029 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1030 unsigned int desc_flags;
1031 union ibmveth_buf_desc descs[6];
1032 int last, i;
1033 int force_bounce = 0;
1034 dma_addr_t dma_addr;
1035 unsigned long mss = 0;
1036
1037 /*
1038 * veth handles a maximum of 6 segments including the header, so
1039 * we have to linearize the skb if there are more than this.
1040 */
1041 if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
1042 netdev->stats.tx_dropped++;
1043 goto out;
1044 }
1045
1046 /* veth can't checksum offload UDP */
1047 if (skb->ip_summed == CHECKSUM_PARTIAL &&
1048 ((skb->protocol == htons(ETH_P_IP) &&
1049 ip_hdr(skb)->protocol != IPPROTO_TCP) ||
1050 (skb->protocol == htons(ETH_P_IPV6) &&
1051 ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
1052 skb_checksum_help(skb)) {
1053
1054 netdev_err(netdev, "tx: failed to checksum packet\n");
1055 netdev->stats.tx_dropped++;
1056 goto out;
1057 }
1058
1059 desc_flags = IBMVETH_BUF_VALID;
1060
1061 if (skb_is_gso(skb) && adapter->fw_large_send_support)
1062 desc_flags |= IBMVETH_BUF_LRG_SND;
1063
1064 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1065 unsigned char *buf = skb_transport_header(skb) +
1066 skb->csum_offset;
1067
1068 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1069
1070 /* Need to zero out the checksum */
1071 buf[0] = 0;
1072 buf[1] = 0;
1073 }
1074
1075retry_bounce:
1076 memset(descs, 0, sizeof(descs));
1077
1078 /*
1079 * If a linear packet is below the rx threshold then
1080 * copy it into the static bounce buffer. This avoids the
1081 * cost of a TCE insert and remove.
1082 */
1083 if (force_bounce || (!skb_is_nonlinear(skb) &&
1084 (skb->len < tx_copybreak))) {
1085 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1086 skb->len);
1087
1088 descs[0].fields.flags_len = desc_flags | skb->len;
1089 descs[0].fields.address = adapter->bounce_buffer_dma;
1090
1091 if (ibmveth_send(adapter, descs, 0)) {
1092 adapter->tx_send_failed++;
1093 netdev->stats.tx_dropped++;
1094 } else {
1095 netdev->stats.tx_packets++;
1096 netdev->stats.tx_bytes += skb->len;
1097 }
1098
1099 goto out;
1100 }
1101
1102 /* Map the header */
1103 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1104 skb_headlen(skb), DMA_TO_DEVICE);
1105 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1106 goto map_failed;
1107
1108 descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1109 descs[0].fields.address = dma_addr;
1110
1111 /* Map the frags */
1112 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1113 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1114
1115 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1116 skb_frag_size(frag), DMA_TO_DEVICE);
1117
1118 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1119 goto map_failed_frags;
1120
1121 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1122 descs[i+1].fields.address = dma_addr;
1123 }
1124
1125 if (skb_is_gso(skb)) {
1126 if (adapter->fw_large_send_support) {
1127 mss = (unsigned long)skb_shinfo(skb)->gso_size;
1128 adapter->tx_large_packets++;
1129 } else if (!skb_is_gso_v6(skb)) {
1130 /* Put -1 in the IP checksum to tell phyp it
1131 * is a largesend packet. Put the mss in
1132 * the TCP checksum.
1133 */
1134 ip_hdr(skb)->check = 0xffff;
1135 tcp_hdr(skb)->check =
1136 cpu_to_be16(skb_shinfo(skb)->gso_size);
1137 adapter->tx_large_packets++;
1138 }
1139 }
1140
1141 if (ibmveth_send(adapter, descs, mss)) {
1142 adapter->tx_send_failed++;
1143 netdev->stats.tx_dropped++;
1144 } else {
1145 netdev->stats.tx_packets++;
1146 netdev->stats.tx_bytes += skb->len;
1147 }
1148
1149 dma_unmap_single(&adapter->vdev->dev,
1150 descs[0].fields.address,
1151 descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1152 DMA_TO_DEVICE);
1153
1154 for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1155 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1156 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1157 DMA_TO_DEVICE);
1158
1159out:
1160 dev_consume_skb_any(skb);
1161 return NETDEV_TX_OK;
1162
1163map_failed_frags:
1164 last = i+1;
1165 for (i = 0; i < last; i++)
1166 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1167 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1168 DMA_TO_DEVICE);
1169
1170map_failed:
1171 if (!firmware_has_feature(FW_FEATURE_CMO))
1172 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1173 adapter->tx_map_failed++;
1174 if (skb_linearize(skb)) {
1175 netdev->stats.tx_dropped++;
1176 goto out;
1177 }
1178 force_bounce = 1;
1179 goto retry_bounce;
1180}
1181
1182static void ibmveth_rx_mss_helper(struct sk_buff *skb, u16 mss, int lrg_pkt)
1183{
1184 struct tcphdr *tcph;
1185 int offset = 0;
1186 int hdr_len;
1187
1188 /* only TCP packets will be aggregated */
1189 if (skb->protocol == htons(ETH_P_IP)) {
1190 struct iphdr *iph = (struct iphdr *)skb->data;
1191
1192 if (iph->protocol == IPPROTO_TCP) {
1193 offset = iph->ihl * 4;
1194 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1195 } else {
1196 return;
1197 }
1198 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1199 struct ipv6hdr *iph6 = (struct ipv6hdr *)skb->data;
1200
1201 if (iph6->nexthdr == IPPROTO_TCP) {
1202 offset = sizeof(struct ipv6hdr);
1203 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1204 } else {
1205 return;
1206 }
1207 } else {
1208 return;
1209 }
1210 /* if mss is not set through Large Packet bit/mss in rx buffer,
1211 * expect that the mss will be written to the tcp header checksum.
1212 */
1213 tcph = (struct tcphdr *)(skb->data + offset);
1214 if (lrg_pkt) {
1215 skb_shinfo(skb)->gso_size = mss;
1216 } else if (offset) {
1217 skb_shinfo(skb)->gso_size = ntohs(tcph->check);
1218 tcph->check = 0;
1219 }
1220
1221 if (skb_shinfo(skb)->gso_size) {
1222 hdr_len = offset + tcph->doff * 4;
1223 skb_shinfo(skb)->gso_segs =
1224 DIV_ROUND_UP(skb->len - hdr_len,
1225 skb_shinfo(skb)->gso_size);
1226 }
1227}
1228
1229static int ibmveth_poll(struct napi_struct *napi, int budget)
1230{
1231 struct ibmveth_adapter *adapter =
1232 container_of(napi, struct ibmveth_adapter, napi);
1233 struct net_device *netdev = adapter->netdev;
1234 int frames_processed = 0;
1235 unsigned long lpar_rc;
1236 struct iphdr *iph;
1237 u16 mss = 0;
1238
1239restart_poll:
1240 while (frames_processed < budget) {
1241 if (!ibmveth_rxq_pending_buffer(adapter))
1242 break;
1243
1244 smp_rmb();
1245 if (!ibmveth_rxq_buffer_valid(adapter)) {
1246 wmb(); /* suggested by larson1 */
1247 adapter->rx_invalid_buffer++;
1248 netdev_dbg(netdev, "recycling invalid buffer\n");
1249 ibmveth_rxq_recycle_buffer(adapter);
1250 } else {
1251 struct sk_buff *skb, *new_skb;
1252 int length = ibmveth_rxq_frame_length(adapter);
1253 int offset = ibmveth_rxq_frame_offset(adapter);
1254 int csum_good = ibmveth_rxq_csum_good(adapter);
1255 int lrg_pkt = ibmveth_rxq_large_packet(adapter);
1256
1257 skb = ibmveth_rxq_get_buffer(adapter);
1258
1259 /* if the large packet bit is set in the rx queue
1260 * descriptor, the mss will be written by PHYP eight
1261 * bytes from the start of the rx buffer, which is
1262 * skb->data at this stage
1263 */
1264 if (lrg_pkt) {
1265 __be64 *rxmss = (__be64 *)(skb->data + 8);
1266
1267 mss = (u16)be64_to_cpu(*rxmss);
1268 }
1269
1270 new_skb = NULL;
1271 if (length < rx_copybreak)
1272 new_skb = netdev_alloc_skb(netdev, length);
1273
1274 if (new_skb) {
1275 skb_copy_to_linear_data(new_skb,
1276 skb->data + offset,
1277 length);
1278 if (rx_flush)
1279 ibmveth_flush_buffer(skb->data,
1280 length + offset);
1281 if (!ibmveth_rxq_recycle_buffer(adapter))
1282 kfree_skb(skb);
1283 skb = new_skb;
1284 } else {
1285 ibmveth_rxq_harvest_buffer(adapter);
1286 skb_reserve(skb, offset);
1287 }
1288
1289 skb_put(skb, length);
1290 skb->protocol = eth_type_trans(skb, netdev);
1291
1292 if (csum_good) {
1293 skb->ip_summed = CHECKSUM_UNNECESSARY;
1294 if (be16_to_cpu(skb->protocol) == ETH_P_IP) {
1295 iph = (struct iphdr *)skb->data;
1296
1297 /* If the IP checksum is not offloaded and if the packet
1298 * is large send, the checksum must be rebuilt.
1299 */
1300 if (iph->check == 0xffff) {
1301 iph->check = 0;
1302 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
1303 }
1304 }
1305 }
1306
1307 if (length > netdev->mtu + ETH_HLEN) {
1308 ibmveth_rx_mss_helper(skb, mss, lrg_pkt);
1309 adapter->rx_large_packets++;
1310 }
1311
1312 napi_gro_receive(napi, skb); /* send it up */
1313
1314 netdev->stats.rx_packets++;
1315 netdev->stats.rx_bytes += length;
1316 frames_processed++;
1317 }
1318 }
1319
1320 ibmveth_replenish_task(adapter);
1321
1322 if (frames_processed < budget) {
1323 napi_complete(napi);
1324
1325 /* We think we are done - reenable interrupts,
1326 * then check once more to make sure we are done.
1327 */
1328 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1329 VIO_IRQ_ENABLE);
1330
1331 BUG_ON(lpar_rc != H_SUCCESS);
1332
1333 if (ibmveth_rxq_pending_buffer(adapter) &&
1334 napi_reschedule(napi)) {
1335 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1336 VIO_IRQ_DISABLE);
1337 goto restart_poll;
1338 }
1339 }
1340
1341 return frames_processed;
1342}
1343
1344static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1345{
1346 struct net_device *netdev = dev_instance;
1347 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1348 unsigned long lpar_rc;
1349
1350 if (napi_schedule_prep(&adapter->napi)) {
1351 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1352 VIO_IRQ_DISABLE);
1353 BUG_ON(lpar_rc != H_SUCCESS);
1354 __napi_schedule(&adapter->napi);
1355 }
1356 return IRQ_HANDLED;
1357}
1358
1359static void ibmveth_set_multicast_list(struct net_device *netdev)
1360{
1361 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1362 unsigned long lpar_rc;
1363
1364 if ((netdev->flags & IFF_PROMISC) ||
1365 (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1366 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1367 IbmVethMcastEnableRecv |
1368 IbmVethMcastDisableFiltering,
1369 0);
1370 if (lpar_rc != H_SUCCESS) {
1371 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1372 "entering promisc mode\n", lpar_rc);
1373 }
1374 } else {
1375 struct netdev_hw_addr *ha;
1376 /* clear the filter table & disable filtering */
1377 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1378 IbmVethMcastEnableRecv |
1379 IbmVethMcastDisableFiltering |
1380 IbmVethMcastClearFilterTable,
1381 0);
1382 if (lpar_rc != H_SUCCESS) {
1383 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1384 "attempting to clear filter table\n",
1385 lpar_rc);
1386 }
1387 /* add the addresses to the filter table */
1388 netdev_for_each_mc_addr(ha, netdev) {
1389 /* add the multicast address to the filter table */
1390 u64 mcast_addr;
1391 mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1392 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1393 IbmVethMcastAddFilter,
1394 mcast_addr);
1395 if (lpar_rc != H_SUCCESS) {
1396 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1397 "when adding an entry to the filter "
1398 "table\n", lpar_rc);
1399 }
1400 }
1401
1402 /* re-enable filtering */
1403 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1404 IbmVethMcastEnableFiltering,
1405 0);
1406 if (lpar_rc != H_SUCCESS) {
1407 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1408 "enabling filtering\n", lpar_rc);
1409 }
1410 }
1411}
1412
1413static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1414{
1415 struct ibmveth_adapter *adapter = netdev_priv(dev);
1416 struct vio_dev *viodev = adapter->vdev;
1417 int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1418 int i, rc;
1419 int need_restart = 0;
1420
1421 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1422 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1423 break;
1424
1425 if (i == IBMVETH_NUM_BUFF_POOLS)
1426 return -EINVAL;
1427
1428 /* Deactivate all the buffer pools so that the next loop can activate
1429 only the buffer pools necessary to hold the new MTU */
1430 if (netif_running(adapter->netdev)) {
1431 need_restart = 1;
1432 adapter->pool_config = 1;
1433 ibmveth_close(adapter->netdev);
1434 adapter->pool_config = 0;
1435 }
1436
1437 /* Look for an active buffer pool that can hold the new MTU */
1438 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1439 adapter->rx_buff_pool[i].active = 1;
1440
1441 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1442 dev->mtu = new_mtu;
1443 vio_cmo_set_dev_desired(viodev,
1444 ibmveth_get_desired_dma
1445 (viodev));
1446 if (need_restart) {
1447 return ibmveth_open(adapter->netdev);
1448 }
1449 return 0;
1450 }
1451 }
1452
1453 if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1454 return rc;
1455
1456 return -EINVAL;
1457}
1458
1459#ifdef CONFIG_NET_POLL_CONTROLLER
1460static void ibmveth_poll_controller(struct net_device *dev)
1461{
1462 ibmveth_replenish_task(netdev_priv(dev));
1463 ibmveth_interrupt(dev->irq, dev);
1464}
1465#endif
1466
1467/**
1468 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1469 *
1470 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1471 *
1472 * Return value:
1473 * Number of bytes of IO data the driver will need to perform well.
1474 */
1475static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1476{
1477 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1478 struct ibmveth_adapter *adapter;
1479 struct iommu_table *tbl;
1480 unsigned long ret;
1481 int i;
1482 int rxqentries = 1;
1483
1484 tbl = get_iommu_table_base(&vdev->dev);
1485
1486 /* netdev inits at probe time along with the structures we need below*/
1487 if (netdev == NULL)
1488 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1489
1490 adapter = netdev_priv(netdev);
1491
1492 ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1493 ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1494
1495 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1496 /* add the size of the active receive buffers */
1497 if (adapter->rx_buff_pool[i].active)
1498 ret +=
1499 adapter->rx_buff_pool[i].size *
1500 IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1501 buff_size, tbl);
1502 rxqentries += adapter->rx_buff_pool[i].size;
1503 }
1504 /* add the size of the receive queue entries */
1505 ret += IOMMU_PAGE_ALIGN(
1506 rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1507
1508 return ret;
1509}
1510
1511static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1512{
1513 struct ibmveth_adapter *adapter = netdev_priv(dev);
1514 struct sockaddr *addr = p;
1515 u64 mac_address;
1516 int rc;
1517
1518 if (!is_valid_ether_addr(addr->sa_data))
1519 return -EADDRNOTAVAIL;
1520
1521 mac_address = ibmveth_encode_mac_addr(addr->sa_data);
1522 rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1523 if (rc) {
1524 netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1525 return rc;
1526 }
1527
1528 ether_addr_copy(dev->dev_addr, addr->sa_data);
1529
1530 return 0;
1531}
1532
1533static const struct net_device_ops ibmveth_netdev_ops = {
1534 .ndo_open = ibmveth_open,
1535 .ndo_stop = ibmveth_close,
1536 .ndo_start_xmit = ibmveth_start_xmit,
1537 .ndo_set_rx_mode = ibmveth_set_multicast_list,
1538 .ndo_do_ioctl = ibmveth_ioctl,
1539 .ndo_change_mtu = ibmveth_change_mtu,
1540 .ndo_fix_features = ibmveth_fix_features,
1541 .ndo_set_features = ibmveth_set_features,
1542 .ndo_validate_addr = eth_validate_addr,
1543 .ndo_set_mac_address = ibmveth_set_mac_addr,
1544#ifdef CONFIG_NET_POLL_CONTROLLER
1545 .ndo_poll_controller = ibmveth_poll_controller,
1546#endif
1547};
1548
1549static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1550{
1551 int rc, i, mac_len;
1552 struct net_device *netdev;
1553 struct ibmveth_adapter *adapter;
1554 unsigned char *mac_addr_p;
1555 unsigned int *mcastFilterSize_p;
1556 long ret;
1557 unsigned long ret_attr;
1558
1559 dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1560 dev->unit_address);
1561
1562 mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1563 &mac_len);
1564 if (!mac_addr_p) {
1565 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1566 return -EINVAL;
1567 }
1568 /* Workaround for old/broken pHyp */
1569 if (mac_len == 8)
1570 mac_addr_p += 2;
1571 else if (mac_len != 6) {
1572 dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1573 mac_len);
1574 return -EINVAL;
1575 }
1576
1577 mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1578 VETH_MCAST_FILTER_SIZE, NULL);
1579 if (!mcastFilterSize_p) {
1580 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1581 "attribute\n");
1582 return -EINVAL;
1583 }
1584
1585 netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1586
1587 if (!netdev)
1588 return -ENOMEM;
1589
1590 adapter = netdev_priv(netdev);
1591 dev_set_drvdata(&dev->dev, netdev);
1592
1593 adapter->vdev = dev;
1594 adapter->netdev = netdev;
1595 adapter->mcastFilterSize = *mcastFilterSize_p;
1596 adapter->pool_config = 0;
1597
1598 netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1599
1600 netdev->irq = dev->irq;
1601 netdev->netdev_ops = &ibmveth_netdev_ops;
1602 netdev->ethtool_ops = &netdev_ethtool_ops;
1603 SET_NETDEV_DEV(netdev, &dev->dev);
1604 netdev->hw_features = NETIF_F_SG;
1605 if (vio_get_attribute(dev, "ibm,illan-options", NULL) != NULL) {
1606 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1607 NETIF_F_RXCSUM;
1608 }
1609
1610 netdev->features |= netdev->hw_features;
1611
1612 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
1613
1614 /* If running older firmware, TSO should not be enabled by default */
1615 if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
1616 !old_large_send) {
1617 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1618 netdev->features |= netdev->hw_features;
1619 } else {
1620 netdev->hw_features |= NETIF_F_TSO;
1621 }
1622
1623 netdev->min_mtu = IBMVETH_MIN_MTU;
1624 netdev->max_mtu = ETH_MAX_MTU;
1625
1626 memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1627
1628 if (firmware_has_feature(FW_FEATURE_CMO))
1629 memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1630
1631 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1632 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1633 int error;
1634
1635 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1636 pool_count[i], pool_size[i],
1637 pool_active[i]);
1638 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1639 &dev->dev.kobj, "pool%d", i);
1640 if (!error)
1641 kobject_uevent(kobj, KOBJ_ADD);
1642 }
1643
1644 netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1645
1646 adapter->buffer_list_dma = DMA_ERROR_CODE;
1647 adapter->filter_list_dma = DMA_ERROR_CODE;
1648 adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1649
1650 netdev_dbg(netdev, "registering netdev...\n");
1651
1652 ibmveth_set_features(netdev, netdev->features);
1653
1654 rc = register_netdev(netdev);
1655
1656 if (rc) {
1657 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1658 free_netdev(netdev);
1659 return rc;
1660 }
1661
1662 netdev_dbg(netdev, "registered\n");
1663
1664 return 0;
1665}
1666
1667static int ibmveth_remove(struct vio_dev *dev)
1668{
1669 struct net_device *netdev = dev_get_drvdata(&dev->dev);
1670 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1671 int i;
1672
1673 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1674 kobject_put(&adapter->rx_buff_pool[i].kobj);
1675
1676 unregister_netdev(netdev);
1677
1678 free_netdev(netdev);
1679 dev_set_drvdata(&dev->dev, NULL);
1680
1681 return 0;
1682}
1683
1684static struct attribute veth_active_attr;
1685static struct attribute veth_num_attr;
1686static struct attribute veth_size_attr;
1687
1688static ssize_t veth_pool_show(struct kobject *kobj,
1689 struct attribute *attr, char *buf)
1690{
1691 struct ibmveth_buff_pool *pool = container_of(kobj,
1692 struct ibmveth_buff_pool,
1693 kobj);
1694
1695 if (attr == &veth_active_attr)
1696 return sprintf(buf, "%d\n", pool->active);
1697 else if (attr == &veth_num_attr)
1698 return sprintf(buf, "%d\n", pool->size);
1699 else if (attr == &veth_size_attr)
1700 return sprintf(buf, "%d\n", pool->buff_size);
1701 return 0;
1702}
1703
1704static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1705 const char *buf, size_t count)
1706{
1707 struct ibmveth_buff_pool *pool = container_of(kobj,
1708 struct ibmveth_buff_pool,
1709 kobj);
1710 struct net_device *netdev = dev_get_drvdata(
1711 container_of(kobj->parent, struct device, kobj));
1712 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1713 long value = simple_strtol(buf, NULL, 10);
1714 long rc;
1715
1716 if (attr == &veth_active_attr) {
1717 if (value && !pool->active) {
1718 if (netif_running(netdev)) {
1719 if (ibmveth_alloc_buffer_pool(pool)) {
1720 netdev_err(netdev,
1721 "unable to alloc pool\n");
1722 return -ENOMEM;
1723 }
1724 pool->active = 1;
1725 adapter->pool_config = 1;
1726 ibmveth_close(netdev);
1727 adapter->pool_config = 0;
1728 if ((rc = ibmveth_open(netdev)))
1729 return rc;
1730 } else {
1731 pool->active = 1;
1732 }
1733 } else if (!value && pool->active) {
1734 int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1735 int i;
1736 /* Make sure there is a buffer pool with buffers that
1737 can hold a packet of the size of the MTU */
1738 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1739 if (pool == &adapter->rx_buff_pool[i])
1740 continue;
1741 if (!adapter->rx_buff_pool[i].active)
1742 continue;
1743 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1744 break;
1745 }
1746
1747 if (i == IBMVETH_NUM_BUFF_POOLS) {
1748 netdev_err(netdev, "no active pool >= MTU\n");
1749 return -EPERM;
1750 }
1751
1752 if (netif_running(netdev)) {
1753 adapter->pool_config = 1;
1754 ibmveth_close(netdev);
1755 pool->active = 0;
1756 adapter->pool_config = 0;
1757 if ((rc = ibmveth_open(netdev)))
1758 return rc;
1759 }
1760 pool->active = 0;
1761 }
1762 } else if (attr == &veth_num_attr) {
1763 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1764 return -EINVAL;
1765 } else {
1766 if (netif_running(netdev)) {
1767 adapter->pool_config = 1;
1768 ibmveth_close(netdev);
1769 adapter->pool_config = 0;
1770 pool->size = value;
1771 if ((rc = ibmveth_open(netdev)))
1772 return rc;
1773 } else {
1774 pool->size = value;
1775 }
1776 }
1777 } else if (attr == &veth_size_attr) {
1778 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1779 return -EINVAL;
1780 } else {
1781 if (netif_running(netdev)) {
1782 adapter->pool_config = 1;
1783 ibmveth_close(netdev);
1784 adapter->pool_config = 0;
1785 pool->buff_size = value;
1786 if ((rc = ibmveth_open(netdev)))
1787 return rc;
1788 } else {
1789 pool->buff_size = value;
1790 }
1791 }
1792 }
1793
1794 /* kick the interrupt handler to allocate/deallocate pools */
1795 ibmveth_interrupt(netdev->irq, netdev);
1796 return count;
1797}
1798
1799
1800#define ATTR(_name, _mode) \
1801 struct attribute veth_##_name##_attr = { \
1802 .name = __stringify(_name), .mode = _mode, \
1803 };
1804
1805static ATTR(active, 0644);
1806static ATTR(num, 0644);
1807static ATTR(size, 0644);
1808
1809static struct attribute *veth_pool_attrs[] = {
1810 &veth_active_attr,
1811 &veth_num_attr,
1812 &veth_size_attr,
1813 NULL,
1814};
1815
1816static const struct sysfs_ops veth_pool_ops = {
1817 .show = veth_pool_show,
1818 .store = veth_pool_store,
1819};
1820
1821static struct kobj_type ktype_veth_pool = {
1822 .release = NULL,
1823 .sysfs_ops = &veth_pool_ops,
1824 .default_attrs = veth_pool_attrs,
1825};
1826
1827static int ibmveth_resume(struct device *dev)
1828{
1829 struct net_device *netdev = dev_get_drvdata(dev);
1830 ibmveth_interrupt(netdev->irq, netdev);
1831 return 0;
1832}
1833
1834static struct vio_device_id ibmveth_device_table[] = {
1835 { "network", "IBM,l-lan"},
1836 { "", "" }
1837};
1838MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1839
1840static struct dev_pm_ops ibmveth_pm_ops = {
1841 .resume = ibmveth_resume
1842};
1843
1844static struct vio_driver ibmveth_driver = {
1845 .id_table = ibmveth_device_table,
1846 .probe = ibmveth_probe,
1847 .remove = ibmveth_remove,
1848 .get_desired_dma = ibmveth_get_desired_dma,
1849 .name = ibmveth_driver_name,
1850 .pm = &ibmveth_pm_ops,
1851};
1852
1853static int __init ibmveth_module_init(void)
1854{
1855 printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1856 ibmveth_driver_string, ibmveth_driver_version);
1857
1858 return vio_register_driver(&ibmveth_driver);
1859}
1860
1861static void __exit ibmveth_module_exit(void)
1862{
1863 vio_unregister_driver(&ibmveth_driver);
1864}
1865
1866module_init(ibmveth_module_init);
1867module_exit(ibmveth_module_exit);