1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /**************************************************************************/
4 /* IBM System i and System p Virtual NIC Device Driver */
5 /* Copyright (C) 2014 IBM Corp. */
6 /* Santiago Leon (santi_leon@yahoo.com) */
7 /* Thomas Falcon (tlfalcon@linux.vnet.ibm.com) */
8 /* John Allen (jallen@linux.vnet.ibm.com) */
11 /* This module contains the implementation of a virtual ethernet device */
12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN */
13 /* option of the RS/6000 Platform Architecture to interface with virtual */
14 /* ethernet NICs that are presented to the partition by the hypervisor. */
16 /* Messages are passed between the VNIC driver and the VNIC server using */
17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to */
18 /* issue and receive commands that initiate communication with the server */
19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but */
20 /* are used by the driver to notify the server that a packet is */
21 /* ready for transmission or that a buffer has been added to receive a */
22 /* packet. Subsequently, sCRQs are used by the server to notify the */
23 /* driver that a packet transmission has been completed or that a packet */
24 /* has been received and placed in a waiting buffer. */
26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in */
27 /* which skbs are DMA mapped and immediately unmapped when the transmit */
28 /* or receive has been completed, the VNIC driver is required to use */
29 /* "long term mapping". This entails that large, continuous DMA mapped */
30 /* buffers are allocated on driver initialization and these buffers are */
31 /* then continuously reused to pass skbs to and from the VNIC server. */
33 /**************************************************************************/
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/types.h>
38 #include <linux/errno.h>
39 #include <linux/completion.h>
40 #include <linux/ioport.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kernel.h>
43 #include <linux/netdevice.h>
44 #include <linux/etherdevice.h>
45 #include <linux/skbuff.h>
46 #include <linux/init.h>
47 #include <linux/delay.h>
49 #include <linux/ethtool.h>
50 #include <linux/proc_fs.h>
51 #include <linux/if_arp.h>
54 #include <linux/ipv6.h>
55 #include <linux/irq.h>
56 #include <linux/kthread.h>
57 #include <linux/seq_file.h>
58 #include <linux/interrupt.h>
59 #include <net/net_namespace.h>
60 #include <asm/hvcall.h>
61 #include <linux/atomic.h>
63 #include <asm/iommu.h>
64 #include <linux/uaccess.h>
65 #include <asm/firmware.h>
66 #include <linux/workqueue.h>
67 #include <linux/if_vlan.h>
68 #include <linux/utsname.h>
72 static const char ibmvnic_driver_name[] = "ibmvnic";
73 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
75 MODULE_AUTHOR("Santiago Leon");
76 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
77 MODULE_LICENSE("GPL");
78 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
80 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
81 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
82 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
83 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
84 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
85 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
86 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
87 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
88 static int enable_scrq_irq(struct ibmvnic_adapter *,
89 struct ibmvnic_sub_crq_queue *);
90 static int disable_scrq_irq(struct ibmvnic_adapter *,
91 struct ibmvnic_sub_crq_queue *);
92 static int pending_scrq(struct ibmvnic_adapter *,
93 struct ibmvnic_sub_crq_queue *);
94 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
95 struct ibmvnic_sub_crq_queue *);
96 static int ibmvnic_poll(struct napi_struct *napi, int data);
97 static void send_query_map(struct ibmvnic_adapter *adapter);
98 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
99 static int send_request_unmap(struct ibmvnic_adapter *, u8);
100 static int send_login(struct ibmvnic_adapter *adapter);
101 static void send_query_cap(struct ibmvnic_adapter *adapter);
102 static int init_sub_crqs(struct ibmvnic_adapter *);
103 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
104 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
105 static void release_crq_queue(struct ibmvnic_adapter *);
106 static int __ibmvnic_set_mac(struct net_device *, u8 *);
107 static int init_crq_queue(struct ibmvnic_adapter *adapter);
108 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
109 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
110 struct ibmvnic_sub_crq_queue *tx_scrq);
111 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
112 struct ibmvnic_long_term_buff *ltb);
113 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter);
115 struct ibmvnic_stat {
116 char name[ETH_GSTRING_LEN];
120 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
121 offsetof(struct ibmvnic_statistics, stat))
122 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
124 static const struct ibmvnic_stat ibmvnic_stats[] = {
125 {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
126 {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
127 {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
128 {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
129 {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
130 {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
131 {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
132 {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
133 {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
134 {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
135 {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
136 {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
137 {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
138 {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
139 {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
140 {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
141 {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
142 {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
143 {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
144 {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
145 {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
146 {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
149 static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
151 union ibmvnic_crq crq;
153 memset(&crq, 0, sizeof(crq));
154 crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
155 crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
157 return ibmvnic_send_crq(adapter, &crq);
160 static int send_version_xchg(struct ibmvnic_adapter *adapter)
162 union ibmvnic_crq crq;
164 memset(&crq, 0, sizeof(crq));
165 crq.version_exchange.first = IBMVNIC_CRQ_CMD;
166 crq.version_exchange.cmd = VERSION_EXCHANGE;
167 crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
169 return ibmvnic_send_crq(adapter, &crq);
172 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
173 unsigned long length, unsigned long *number,
176 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
179 rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
187 * ibmvnic_wait_for_completion - Check device state and wait for completion
188 * @adapter: private device data
189 * @comp_done: completion structure to wait for
190 * @timeout: time to wait in milliseconds
192 * Wait for a completion signal or until the timeout limit is reached
193 * while checking that the device is still active.
195 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
196 struct completion *comp_done,
197 unsigned long timeout)
199 struct net_device *netdev;
200 unsigned long div_timeout;
203 netdev = adapter->netdev;
205 div_timeout = msecs_to_jiffies(timeout / retry);
207 if (!adapter->crq.active) {
208 netdev_err(netdev, "Device down!\n");
213 if (wait_for_completion_timeout(comp_done, div_timeout))
216 netdev_err(netdev, "Operation timed out.\n");
221 * reuse_ltb() - Check if a long term buffer can be reused
222 * @ltb: The long term buffer to be checked
223 * @size: The size of the long term buffer.
225 * An LTB can be reused unless its size has changed.
227 * Return: Return true if the LTB can be reused, false otherwise.
229 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size)
231 return (ltb->buff && ltb->size == size);
235 * alloc_long_term_buff() - Allocate a long term buffer (LTB)
237 * @adapter: ibmvnic adapter associated to the LTB
238 * @ltb: container object for the LTB
239 * @size: size of the LTB
241 * Allocate an LTB of the specified size and notify VIOS.
243 * If the given @ltb already has the correct size, reuse it. Otherwise if
244 * its non-NULL, free it. Then allocate a new one of the correct size.
245 * Notify the VIOS either way since we may now be working with a new VIOS.
247 * Allocating larger chunks of memory during resets, specially LPM or under
248 * low memory situations can cause resets to fail/timeout and for LPAR to
249 * lose connectivity. So hold onto the LTB even if we fail to communicate
250 * with the VIOS and reuse it on next open. Free LTB when adapter is closed.
252 * Return: 0 if we were able to allocate the LTB and notify the VIOS and
253 * a negative value otherwise.
255 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
256 struct ibmvnic_long_term_buff *ltb, int size)
258 struct device *dev = &adapter->vdev->dev;
261 if (!reuse_ltb(ltb, size)) {
263 "LTB size changed from 0x%llx to 0x%x, reallocating\n",
265 free_long_term_buff(adapter, ltb);
269 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n",
270 ltb->map_id, ltb->size);
272 ltb->buff = dma_alloc_coherent(dev, size, <b->addr,
275 dev_err(dev, "Couldn't alloc long term buffer\n");
280 ltb->map_id = find_first_zero_bit(adapter->map_ids,
282 bitmap_set(adapter->map_ids, ltb->map_id, 1);
285 "Allocated new LTB [map %d, size 0x%llx]\n",
286 ltb->map_id, ltb->size);
289 /* Ensure ltb is zeroed - specially when reusing it. */
290 memset(ltb->buff, 0, ltb->size);
292 mutex_lock(&adapter->fw_lock);
293 adapter->fw_done_rc = 0;
294 reinit_completion(&adapter->fw_done);
296 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
298 dev_err(dev, "send_request_map failed, rc = %d\n", rc);
302 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
304 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n",
309 if (adapter->fw_done_rc) {
310 dev_err(dev, "Couldn't map LTB, rc = %d\n",
311 adapter->fw_done_rc);
317 /* don't free LTB on communication error - see function header */
318 mutex_unlock(&adapter->fw_lock);
322 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
323 struct ibmvnic_long_term_buff *ltb)
325 struct device *dev = &adapter->vdev->dev;
330 /* VIOS automatically unmaps the long term buffer at remote
331 * end for the following resets:
332 * FAILOVER, MOBILITY, TIMEOUT.
334 if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
335 adapter->reset_reason != VNIC_RESET_MOBILITY &&
336 adapter->reset_reason != VNIC_RESET_TIMEOUT)
337 send_request_unmap(adapter, ltb->map_id);
339 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
342 /* mark this map_id free */
343 bitmap_clear(adapter->map_ids, ltb->map_id, 1);
347 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
351 for (i = 0; i < adapter->num_active_rx_pools; i++)
352 adapter->rx_pool[i].active = 0;
355 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
356 struct ibmvnic_rx_pool *pool)
358 int count = pool->size - atomic_read(&pool->available);
359 u64 handle = adapter->rx_scrq[pool->index]->handle;
360 struct device *dev = &adapter->vdev->dev;
361 struct ibmvnic_ind_xmit_queue *ind_bufp;
362 struct ibmvnic_sub_crq_queue *rx_scrq;
363 union sub_crq *sub_crq;
364 int buffers_added = 0;
365 unsigned long lpar_rc;
377 rx_scrq = adapter->rx_scrq[pool->index];
378 ind_bufp = &rx_scrq->ind_buf;
380 /* netdev_skb_alloc() could have failed after we saved a few skbs
381 * in the indir_buf and we would not have sent them to VIOS yet.
382 * To account for them, start the loop at ind_bufp->index rather
383 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
386 for (i = ind_bufp->index; i < count; ++i) {
387 index = pool->free_map[pool->next_free];
389 /* We maybe reusing the skb from earlier resets. Allocate
390 * only if necessary. But since the LTB may have changed
391 * during reset (see init_rx_pools()), update LTB below
392 * even if reusing skb.
394 skb = pool->rx_buff[index].skb;
396 skb = netdev_alloc_skb(adapter->netdev,
399 dev_err(dev, "Couldn't replenish rx buff\n");
400 adapter->replenish_no_mem++;
405 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
406 pool->next_free = (pool->next_free + 1) % pool->size;
408 /* Copy the skb to the long term mapped DMA buffer */
409 offset = index * pool->buff_size;
410 dst = pool->long_term_buff.buff + offset;
411 memset(dst, 0, pool->buff_size);
412 dma_addr = pool->long_term_buff.addr + offset;
414 /* add the skb to an rx_buff in the pool */
415 pool->rx_buff[index].data = dst;
416 pool->rx_buff[index].dma = dma_addr;
417 pool->rx_buff[index].skb = skb;
418 pool->rx_buff[index].pool_index = pool->index;
419 pool->rx_buff[index].size = pool->buff_size;
421 /* queue the rx_buff for the next send_subcrq_indirect */
422 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
423 memset(sub_crq, 0, sizeof(*sub_crq));
424 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
425 sub_crq->rx_add.correlator =
426 cpu_to_be64((u64)&pool->rx_buff[index]);
427 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
428 sub_crq->rx_add.map_id = pool->long_term_buff.map_id;
430 /* The length field of the sCRQ is defined to be 24 bits so the
431 * buffer size needs to be left shifted by a byte before it is
432 * converted to big endian to prevent the last byte from being
435 #ifdef __LITTLE_ENDIAN__
438 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
440 /* if send_subcrq_indirect queue is full, flush to VIOS */
441 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
444 send_subcrq_indirect(adapter, handle,
445 (u64)ind_bufp->indir_dma,
446 (u64)ind_bufp->index);
447 if (lpar_rc != H_SUCCESS)
449 buffers_added += ind_bufp->index;
450 adapter->replenish_add_buff_success += ind_bufp->index;
454 atomic_add(buffers_added, &pool->available);
458 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
459 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
460 for (i = ind_bufp->index - 1; i >= 0; --i) {
461 struct ibmvnic_rx_buff *rx_buff;
463 pool->next_free = pool->next_free == 0 ?
464 pool->size - 1 : pool->next_free - 1;
465 sub_crq = &ind_bufp->indir_arr[i];
466 rx_buff = (struct ibmvnic_rx_buff *)
467 be64_to_cpu(sub_crq->rx_add.correlator);
468 index = (int)(rx_buff - pool->rx_buff);
469 pool->free_map[pool->next_free] = index;
470 dev_kfree_skb_any(pool->rx_buff[index].skb);
471 pool->rx_buff[index].skb = NULL;
473 adapter->replenish_add_buff_failure += ind_bufp->index;
474 atomic_add(buffers_added, &pool->available);
476 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
477 /* Disable buffer pool replenishment and report carrier off if
478 * queue is closed or pending failover.
479 * Firmware guarantees that a signal will be sent to the
480 * driver, triggering a reset.
482 deactivate_rx_pools(adapter);
483 netif_carrier_off(adapter->netdev);
487 static void replenish_pools(struct ibmvnic_adapter *adapter)
491 adapter->replenish_task_cycles++;
492 for (i = 0; i < adapter->num_active_rx_pools; i++) {
493 if (adapter->rx_pool[i].active)
494 replenish_rx_pool(adapter, &adapter->rx_pool[i]);
497 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
500 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
502 kfree(adapter->tx_stats_buffers);
503 kfree(adapter->rx_stats_buffers);
504 adapter->tx_stats_buffers = NULL;
505 adapter->rx_stats_buffers = NULL;
508 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
510 adapter->tx_stats_buffers =
511 kcalloc(IBMVNIC_MAX_QUEUES,
512 sizeof(struct ibmvnic_tx_queue_stats),
514 if (!adapter->tx_stats_buffers)
517 adapter->rx_stats_buffers =
518 kcalloc(IBMVNIC_MAX_QUEUES,
519 sizeof(struct ibmvnic_rx_queue_stats),
521 if (!adapter->rx_stats_buffers)
527 static void release_stats_token(struct ibmvnic_adapter *adapter)
529 struct device *dev = &adapter->vdev->dev;
531 if (!adapter->stats_token)
534 dma_unmap_single(dev, adapter->stats_token,
535 sizeof(struct ibmvnic_statistics),
537 adapter->stats_token = 0;
540 static int init_stats_token(struct ibmvnic_adapter *adapter)
542 struct device *dev = &adapter->vdev->dev;
546 stok = dma_map_single(dev, &adapter->stats,
547 sizeof(struct ibmvnic_statistics),
549 rc = dma_mapping_error(dev, stok);
551 dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc);
555 adapter->stats_token = stok;
556 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
561 * release_rx_pools() - Release any rx pools attached to @adapter.
562 * @adapter: ibmvnic adapter
564 * Safe to call this multiple times - even if no pools are attached.
566 static void release_rx_pools(struct ibmvnic_adapter *adapter)
568 struct ibmvnic_rx_pool *rx_pool;
571 if (!adapter->rx_pool)
574 for (i = 0; i < adapter->num_active_rx_pools; i++) {
575 rx_pool = &adapter->rx_pool[i];
577 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
579 kfree(rx_pool->free_map);
581 free_long_term_buff(adapter, &rx_pool->long_term_buff);
583 if (!rx_pool->rx_buff)
586 for (j = 0; j < rx_pool->size; j++) {
587 if (rx_pool->rx_buff[j].skb) {
588 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
589 rx_pool->rx_buff[j].skb = NULL;
593 kfree(rx_pool->rx_buff);
596 kfree(adapter->rx_pool);
597 adapter->rx_pool = NULL;
598 adapter->num_active_rx_pools = 0;
599 adapter->prev_rx_pool_size = 0;
603 * reuse_rx_pools() - Check if the existing rx pools can be reused.
604 * @adapter: ibmvnic adapter
606 * Check if the existing rx pools in the adapter can be reused. The
607 * pools can be reused if the pool parameters (number of pools,
608 * number of buffers in the pool and size of each buffer) have not
611 * NOTE: This assumes that all pools have the same number of buffers
612 * which is the case currently. If that changes, we must fix this.
614 * Return: true if the rx pools can be reused, false otherwise.
616 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter)
618 u64 old_num_pools, new_num_pools;
619 u64 old_pool_size, new_pool_size;
620 u64 old_buff_size, new_buff_size;
622 if (!adapter->rx_pool)
625 old_num_pools = adapter->num_active_rx_pools;
626 new_num_pools = adapter->req_rx_queues;
628 old_pool_size = adapter->prev_rx_pool_size;
629 new_pool_size = adapter->req_rx_add_entries_per_subcrq;
631 old_buff_size = adapter->prev_rx_buf_sz;
632 new_buff_size = adapter->cur_rx_buf_sz;
634 if (old_buff_size != new_buff_size ||
635 old_num_pools != new_num_pools ||
636 old_pool_size != new_pool_size)
643 * init_rx_pools(): Initialize the set of receiver pools in the adapter.
644 * @netdev: net device associated with the vnic interface
646 * Initialize the set of receiver pools in the ibmvnic adapter associated
647 * with the net_device @netdev. If possible, reuse the existing rx pools.
648 * Otherwise free any existing pools and allocate a new set of pools
649 * before initializing them.
651 * Return: 0 on success and negative value on error.
653 static int init_rx_pools(struct net_device *netdev)
655 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
656 struct device *dev = &adapter->vdev->dev;
657 struct ibmvnic_rx_pool *rx_pool;
659 u64 pool_size; /* # of buffers in one pool */
663 pool_size = adapter->req_rx_add_entries_per_subcrq;
664 num_pools = adapter->req_rx_queues;
665 buff_size = adapter->cur_rx_buf_sz;
667 if (reuse_rx_pools(adapter)) {
668 dev_dbg(dev, "Reusing rx pools\n");
672 /* Allocate/populate the pools. */
673 release_rx_pools(adapter);
675 adapter->rx_pool = kcalloc(num_pools,
676 sizeof(struct ibmvnic_rx_pool),
678 if (!adapter->rx_pool) {
679 dev_err(dev, "Failed to allocate rx pools\n");
683 /* Set num_active_rx_pools early. If we fail below after partial
684 * allocation, release_rx_pools() will know how many to look for.
686 adapter->num_active_rx_pools = num_pools;
688 for (i = 0; i < num_pools; i++) {
689 rx_pool = &adapter->rx_pool[i];
691 netdev_dbg(adapter->netdev,
692 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
693 i, pool_size, buff_size);
695 rx_pool->size = pool_size;
697 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
699 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
701 if (!rx_pool->free_map) {
702 dev_err(dev, "Couldn't alloc free_map %d\n", i);
707 rx_pool->rx_buff = kcalloc(rx_pool->size,
708 sizeof(struct ibmvnic_rx_buff),
710 if (!rx_pool->rx_buff) {
711 dev_err(dev, "Couldn't alloc rx buffers\n");
717 adapter->prev_rx_pool_size = pool_size;
718 adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz;
721 for (i = 0; i < num_pools; i++) {
722 rx_pool = &adapter->rx_pool[i];
723 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
724 i, rx_pool->size, rx_pool->buff_size);
726 rc = alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
727 rx_pool->size * rx_pool->buff_size);
731 for (j = 0; j < rx_pool->size; ++j) {
732 struct ibmvnic_rx_buff *rx_buff;
734 rx_pool->free_map[j] = j;
736 /* NOTE: Don't clear rx_buff->skb here - will leak
737 * memory! replenish_rx_pool() will reuse skbs or
738 * allocate as necessary.
740 rx_buff = &rx_pool->rx_buff[j];
744 rx_buff->pool_index = 0;
747 /* Mark pool "empty" so replenish_rx_pools() will
748 * update the LTB info for each buffer
750 atomic_set(&rx_pool->available, 0);
751 rx_pool->next_alloc = 0;
752 rx_pool->next_free = 0;
753 /* replenish_rx_pool() may have called deactivate_rx_pools()
754 * on failover. Ensure pool is active now.
760 release_rx_pools(adapter);
762 /* We failed to allocate one or more LTBs or map them on the VIOS.
763 * Hold onto the pools and any LTBs that we did allocate/map.
768 static void release_vpd_data(struct ibmvnic_adapter *adapter)
773 kfree(adapter->vpd->buff);
779 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
780 struct ibmvnic_tx_pool *tx_pool)
782 kfree(tx_pool->tx_buff);
783 kfree(tx_pool->free_map);
784 free_long_term_buff(adapter, &tx_pool->long_term_buff);
788 * release_tx_pools() - Release any tx pools attached to @adapter.
789 * @adapter: ibmvnic adapter
791 * Safe to call this multiple times - even if no pools are attached.
793 static void release_tx_pools(struct ibmvnic_adapter *adapter)
797 /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are
798 * both NULL or both non-NULL. So we only need to check one.
800 if (!adapter->tx_pool)
803 for (i = 0; i < adapter->num_active_tx_pools; i++) {
804 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
805 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
808 kfree(adapter->tx_pool);
809 adapter->tx_pool = NULL;
810 kfree(adapter->tso_pool);
811 adapter->tso_pool = NULL;
812 adapter->num_active_tx_pools = 0;
813 adapter->prev_tx_pool_size = 0;
816 static int init_one_tx_pool(struct net_device *netdev,
817 struct ibmvnic_tx_pool *tx_pool,
818 int pool_size, int buf_size)
822 tx_pool->tx_buff = kcalloc(pool_size,
823 sizeof(struct ibmvnic_tx_buff),
825 if (!tx_pool->tx_buff)
828 tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL);
829 if (!tx_pool->free_map) {
830 kfree(tx_pool->tx_buff);
831 tx_pool->tx_buff = NULL;
835 for (i = 0; i < pool_size; i++)
836 tx_pool->free_map[i] = i;
838 tx_pool->consumer_index = 0;
839 tx_pool->producer_index = 0;
840 tx_pool->num_buffers = pool_size;
841 tx_pool->buf_size = buf_size;
847 * reuse_tx_pools() - Check if the existing tx pools can be reused.
848 * @adapter: ibmvnic adapter
850 * Check if the existing tx pools in the adapter can be reused. The
851 * pools can be reused if the pool parameters (number of pools,
852 * number of buffers in the pool and mtu) have not changed.
854 * NOTE: This assumes that all pools have the same number of buffers
855 * which is the case currently. If that changes, we must fix this.
857 * Return: true if the tx pools can be reused, false otherwise.
859 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter)
861 u64 old_num_pools, new_num_pools;
862 u64 old_pool_size, new_pool_size;
863 u64 old_mtu, new_mtu;
865 if (!adapter->tx_pool)
868 old_num_pools = adapter->num_active_tx_pools;
869 new_num_pools = adapter->num_active_tx_scrqs;
870 old_pool_size = adapter->prev_tx_pool_size;
871 new_pool_size = adapter->req_tx_entries_per_subcrq;
872 old_mtu = adapter->prev_mtu;
873 new_mtu = adapter->req_mtu;
875 if (old_mtu != new_mtu ||
876 old_num_pools != new_num_pools ||
877 old_pool_size != new_pool_size)
884 * init_tx_pools(): Initialize the set of transmit pools in the adapter.
885 * @netdev: net device associated with the vnic interface
887 * Initialize the set of transmit pools in the ibmvnic adapter associated
888 * with the net_device @netdev. If possible, reuse the existing tx pools.
889 * Otherwise free any existing pools and allocate a new set of pools
890 * before initializing them.
892 * Return: 0 on success and negative value on error.
894 static int init_tx_pools(struct net_device *netdev)
896 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
897 struct device *dev = &adapter->vdev->dev;
899 u64 pool_size; /* # of buffers in pool */
903 num_pools = adapter->req_tx_queues;
905 /* We must notify the VIOS about the LTB on all resets - but we only
906 * need to alloc/populate pools if either the number of buffers or
907 * size of each buffer in the pool has changed.
909 if (reuse_tx_pools(adapter)) {
910 netdev_dbg(netdev, "Reusing tx pools\n");
914 /* Allocate/populate the pools. */
915 release_tx_pools(adapter);
917 pool_size = adapter->req_tx_entries_per_subcrq;
918 num_pools = adapter->num_active_tx_scrqs;
920 adapter->tx_pool = kcalloc(num_pools,
921 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
922 if (!adapter->tx_pool)
925 adapter->tso_pool = kcalloc(num_pools,
926 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
927 /* To simplify release_tx_pools() ensure that ->tx_pool and
928 * ->tso_pool are either both NULL or both non-NULL.
930 if (!adapter->tso_pool) {
931 kfree(adapter->tx_pool);
932 adapter->tx_pool = NULL;
936 /* Set num_active_tx_pools early. If we fail below after partial
937 * allocation, release_tx_pools() will know how many to look for.
939 adapter->num_active_tx_pools = num_pools;
941 buff_size = adapter->req_mtu + VLAN_HLEN;
942 buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
944 for (i = 0; i < num_pools; i++) {
945 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n",
946 i, adapter->req_tx_entries_per_subcrq, buff_size);
948 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
949 pool_size, buff_size);
953 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
960 adapter->prev_tx_pool_size = pool_size;
961 adapter->prev_mtu = adapter->req_mtu;
964 /* NOTE: All tx_pools have the same number of buffers (which is
965 * same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS
966 * buffers (see calls init_one_tx_pool() for these).
967 * For consistency, we use tx_pool->num_buffers and
968 * tso_pool->num_buffers below.
971 for (i = 0; i < num_pools; i++) {
972 struct ibmvnic_tx_pool *tso_pool;
973 struct ibmvnic_tx_pool *tx_pool;
976 tx_pool = &adapter->tx_pool[i];
977 ltb_size = tx_pool->num_buffers * tx_pool->buf_size;
978 if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
982 dev_dbg(dev, "Updated LTB for tx pool %d [%p, %d, %d]\n",
983 i, tx_pool->long_term_buff.buff,
984 tx_pool->num_buffers, tx_pool->buf_size);
986 tx_pool->consumer_index = 0;
987 tx_pool->producer_index = 0;
989 for (j = 0; j < tx_pool->num_buffers; j++)
990 tx_pool->free_map[j] = j;
992 tso_pool = &adapter->tso_pool[i];
993 ltb_size = tso_pool->num_buffers * tso_pool->buf_size;
994 if (alloc_long_term_buff(adapter, &tso_pool->long_term_buff,
998 dev_dbg(dev, "Updated LTB for tso pool %d [%p, %d, %d]\n",
999 i, tso_pool->long_term_buff.buff,
1000 tso_pool->num_buffers, tso_pool->buf_size);
1002 tso_pool->consumer_index = 0;
1003 tso_pool->producer_index = 0;
1005 for (j = 0; j < tso_pool->num_buffers; j++)
1006 tso_pool->free_map[j] = j;
1011 release_tx_pools(adapter);
1013 /* We failed to allocate one or more LTBs or map them on the VIOS.
1014 * Hold onto the pools and any LTBs that we did allocate/map.
1019 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
1023 if (adapter->napi_enabled)
1026 for (i = 0; i < adapter->req_rx_queues; i++)
1027 napi_enable(&adapter->napi[i]);
1029 adapter->napi_enabled = true;
1032 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
1036 if (!adapter->napi_enabled)
1039 for (i = 0; i < adapter->req_rx_queues; i++) {
1040 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
1041 napi_disable(&adapter->napi[i]);
1044 adapter->napi_enabled = false;
1047 static int init_napi(struct ibmvnic_adapter *adapter)
1051 adapter->napi = kcalloc(adapter->req_rx_queues,
1052 sizeof(struct napi_struct), GFP_KERNEL);
1056 for (i = 0; i < adapter->req_rx_queues; i++) {
1057 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
1058 netif_napi_add(adapter->netdev, &adapter->napi[i],
1059 ibmvnic_poll, NAPI_POLL_WEIGHT);
1062 adapter->num_active_rx_napi = adapter->req_rx_queues;
1066 static void release_napi(struct ibmvnic_adapter *adapter)
1073 for (i = 0; i < adapter->num_active_rx_napi; i++) {
1074 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
1075 netif_napi_del(&adapter->napi[i]);
1078 kfree(adapter->napi);
1079 adapter->napi = NULL;
1080 adapter->num_active_rx_napi = 0;
1081 adapter->napi_enabled = false;
1084 static const char *adapter_state_to_string(enum vnic_state state)
1109 static int ibmvnic_login(struct net_device *netdev)
1111 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1112 unsigned long timeout = msecs_to_jiffies(20000);
1113 int retry_count = 0;
1120 if (retry_count > retries) {
1121 netdev_warn(netdev, "Login attempts exceeded\n");
1125 adapter->init_done_rc = 0;
1126 reinit_completion(&adapter->init_done);
1127 rc = send_login(adapter);
1131 if (!wait_for_completion_timeout(&adapter->init_done,
1133 netdev_warn(netdev, "Login timed out, retrying...\n");
1135 adapter->init_done_rc = 0;
1140 if (adapter->init_done_rc == ABORTED) {
1141 netdev_warn(netdev, "Login aborted, retrying...\n");
1143 adapter->init_done_rc = 0;
1145 /* FW or device may be busy, so
1146 * wait a bit before retrying login
1149 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
1151 release_sub_crqs(adapter, 1);
1155 "Received partial success, retrying...\n");
1156 adapter->init_done_rc = 0;
1157 reinit_completion(&adapter->init_done);
1158 send_query_cap(adapter);
1159 if (!wait_for_completion_timeout(&adapter->init_done,
1162 "Capabilities query timed out\n");
1166 rc = init_sub_crqs(adapter);
1169 "SCRQ initialization failed\n");
1173 rc = init_sub_crq_irqs(adapter);
1176 "SCRQ irq initialization failed\n");
1179 } else if (adapter->init_done_rc) {
1180 netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n",
1181 adapter->init_done_rc);
1186 __ibmvnic_set_mac(netdev, adapter->mac_addr);
1188 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1192 static void release_login_buffer(struct ibmvnic_adapter *adapter)
1194 kfree(adapter->login_buf);
1195 adapter->login_buf = NULL;
1198 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1200 kfree(adapter->login_rsp_buf);
1201 adapter->login_rsp_buf = NULL;
1204 static void release_resources(struct ibmvnic_adapter *adapter)
1206 release_vpd_data(adapter);
1208 release_napi(adapter);
1209 release_login_buffer(adapter);
1210 release_login_rsp_buffer(adapter);
1213 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1215 struct net_device *netdev = adapter->netdev;
1216 unsigned long timeout = msecs_to_jiffies(20000);
1217 union ibmvnic_crq crq;
1221 netdev_dbg(netdev, "setting link state %d\n", link_state);
1223 memset(&crq, 0, sizeof(crq));
1224 crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1225 crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1226 crq.logical_link_state.link_state = link_state;
1231 reinit_completion(&adapter->init_done);
1232 rc = ibmvnic_send_crq(adapter, &crq);
1234 netdev_err(netdev, "Failed to set link state\n");
1238 if (!wait_for_completion_timeout(&adapter->init_done,
1240 netdev_err(netdev, "timeout setting link state\n");
1244 if (adapter->init_done_rc == PARTIALSUCCESS) {
1245 /* Partuial success, delay and re-send */
1248 } else if (adapter->init_done_rc) {
1249 netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1250 adapter->init_done_rc);
1251 return adapter->init_done_rc;
1258 static int set_real_num_queues(struct net_device *netdev)
1260 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1263 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1264 adapter->req_tx_queues, adapter->req_rx_queues);
1266 rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1268 netdev_err(netdev, "failed to set the number of tx queues\n");
1272 rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1274 netdev_err(netdev, "failed to set the number of rx queues\n");
1279 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1281 struct device *dev = &adapter->vdev->dev;
1282 union ibmvnic_crq crq;
1286 if (adapter->vpd->buff)
1287 len = adapter->vpd->len;
1289 mutex_lock(&adapter->fw_lock);
1290 adapter->fw_done_rc = 0;
1291 reinit_completion(&adapter->fw_done);
1293 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1294 crq.get_vpd_size.cmd = GET_VPD_SIZE;
1295 rc = ibmvnic_send_crq(adapter, &crq);
1297 mutex_unlock(&adapter->fw_lock);
1301 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1303 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1304 mutex_unlock(&adapter->fw_lock);
1307 mutex_unlock(&adapter->fw_lock);
1309 if (!adapter->vpd->len)
1312 if (!adapter->vpd->buff)
1313 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1314 else if (adapter->vpd->len != len)
1315 adapter->vpd->buff =
1316 krealloc(adapter->vpd->buff,
1317 adapter->vpd->len, GFP_KERNEL);
1319 if (!adapter->vpd->buff) {
1320 dev_err(dev, "Could allocate VPD buffer\n");
1324 adapter->vpd->dma_addr =
1325 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1327 if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1328 dev_err(dev, "Could not map VPD buffer\n");
1329 kfree(adapter->vpd->buff);
1330 adapter->vpd->buff = NULL;
1334 mutex_lock(&adapter->fw_lock);
1335 adapter->fw_done_rc = 0;
1336 reinit_completion(&adapter->fw_done);
1338 crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1339 crq.get_vpd.cmd = GET_VPD;
1340 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1341 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1342 rc = ibmvnic_send_crq(adapter, &crq);
1344 kfree(adapter->vpd->buff);
1345 adapter->vpd->buff = NULL;
1346 mutex_unlock(&adapter->fw_lock);
1350 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1352 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1353 kfree(adapter->vpd->buff);
1354 adapter->vpd->buff = NULL;
1355 mutex_unlock(&adapter->fw_lock);
1359 mutex_unlock(&adapter->fw_lock);
1363 static int init_resources(struct ibmvnic_adapter *adapter)
1365 struct net_device *netdev = adapter->netdev;
1368 rc = set_real_num_queues(netdev);
1372 adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1376 /* Vital Product Data (VPD) */
1377 rc = ibmvnic_get_vpd(adapter);
1379 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1383 rc = init_napi(adapter);
1387 send_query_map(adapter);
1389 rc = init_rx_pools(netdev);
1393 rc = init_tx_pools(netdev);
1397 static int __ibmvnic_open(struct net_device *netdev)
1399 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1400 enum vnic_state prev_state = adapter->state;
1403 adapter->state = VNIC_OPENING;
1404 replenish_pools(adapter);
1405 ibmvnic_napi_enable(adapter);
1407 /* We're ready to receive frames, enable the sub-crq interrupts and
1408 * set the logical link state to up
1410 for (i = 0; i < adapter->req_rx_queues; i++) {
1411 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1412 if (prev_state == VNIC_CLOSED)
1413 enable_irq(adapter->rx_scrq[i]->irq);
1414 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1417 for (i = 0; i < adapter->req_tx_queues; i++) {
1418 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1419 if (prev_state == VNIC_CLOSED)
1420 enable_irq(adapter->tx_scrq[i]->irq);
1421 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1422 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1425 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1427 ibmvnic_napi_disable(adapter);
1428 ibmvnic_disable_irqs(adapter);
1432 netif_tx_start_all_queues(netdev);
1434 if (prev_state == VNIC_CLOSED) {
1435 for (i = 0; i < adapter->req_rx_queues; i++)
1436 napi_schedule(&adapter->napi[i]);
1439 adapter->state = VNIC_OPEN;
1443 static int ibmvnic_open(struct net_device *netdev)
1445 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1450 /* If device failover is pending or we are about to reset, just set
1451 * device state and return. Device operation will be handled by reset
1454 * It should be safe to overwrite the adapter->state here. Since
1455 * we hold the rtnl, either the reset has not actually started or
1456 * the rtnl got dropped during the set_link_state() in do_reset().
1457 * In the former case, no one else is changing the state (again we
1458 * have the rtnl) and in the latter case, do_reset() will detect and
1459 * honor our setting below.
1461 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1462 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1463 adapter_state_to_string(adapter->state),
1464 adapter->failover_pending);
1465 adapter->state = VNIC_OPEN;
1470 if (adapter->state != VNIC_CLOSED) {
1471 rc = ibmvnic_login(netdev);
1475 rc = init_resources(adapter);
1477 netdev_err(netdev, "failed to initialize resources\n");
1482 rc = __ibmvnic_open(netdev);
1485 /* If open failed and there is a pending failover or in-progress reset,
1486 * set device state and return. Device operation will be handled by
1487 * reset routine. See also comments above regarding rtnl.
1490 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1491 adapter->state = VNIC_OPEN;
1496 release_resources(adapter);
1497 release_rx_pools(adapter);
1498 release_tx_pools(adapter);
1504 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1506 struct ibmvnic_rx_pool *rx_pool;
1507 struct ibmvnic_rx_buff *rx_buff;
1512 if (!adapter->rx_pool)
1515 rx_scrqs = adapter->num_active_rx_pools;
1516 rx_entries = adapter->req_rx_add_entries_per_subcrq;
1518 /* Free any remaining skbs in the rx buffer pools */
1519 for (i = 0; i < rx_scrqs; i++) {
1520 rx_pool = &adapter->rx_pool[i];
1521 if (!rx_pool || !rx_pool->rx_buff)
1524 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1525 for (j = 0; j < rx_entries; j++) {
1526 rx_buff = &rx_pool->rx_buff[j];
1527 if (rx_buff && rx_buff->skb) {
1528 dev_kfree_skb_any(rx_buff->skb);
1529 rx_buff->skb = NULL;
1535 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1536 struct ibmvnic_tx_pool *tx_pool)
1538 struct ibmvnic_tx_buff *tx_buff;
1542 if (!tx_pool || !tx_pool->tx_buff)
1545 tx_entries = tx_pool->num_buffers;
1547 for (i = 0; i < tx_entries; i++) {
1548 tx_buff = &tx_pool->tx_buff[i];
1549 if (tx_buff && tx_buff->skb) {
1550 dev_kfree_skb_any(tx_buff->skb);
1551 tx_buff->skb = NULL;
1556 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1561 if (!adapter->tx_pool || !adapter->tso_pool)
1564 tx_scrqs = adapter->num_active_tx_pools;
1566 /* Free any remaining skbs in the tx buffer pools */
1567 for (i = 0; i < tx_scrqs; i++) {
1568 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1569 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1570 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1574 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1576 struct net_device *netdev = adapter->netdev;
1579 if (adapter->tx_scrq) {
1580 for (i = 0; i < adapter->req_tx_queues; i++)
1581 if (adapter->tx_scrq[i]->irq) {
1583 "Disabling tx_scrq[%d] irq\n", i);
1584 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1585 disable_irq(adapter->tx_scrq[i]->irq);
1589 if (adapter->rx_scrq) {
1590 for (i = 0; i < adapter->req_rx_queues; i++) {
1591 if (adapter->rx_scrq[i]->irq) {
1593 "Disabling rx_scrq[%d] irq\n", i);
1594 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1595 disable_irq(adapter->rx_scrq[i]->irq);
1601 static void ibmvnic_cleanup(struct net_device *netdev)
1603 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1605 /* ensure that transmissions are stopped if called by do_reset */
1606 if (test_bit(0, &adapter->resetting))
1607 netif_tx_disable(netdev);
1609 netif_tx_stop_all_queues(netdev);
1611 ibmvnic_napi_disable(adapter);
1612 ibmvnic_disable_irqs(adapter);
1615 static int __ibmvnic_close(struct net_device *netdev)
1617 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1620 adapter->state = VNIC_CLOSING;
1621 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1622 adapter->state = VNIC_CLOSED;
1626 static int ibmvnic_close(struct net_device *netdev)
1628 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1631 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
1632 adapter_state_to_string(adapter->state),
1633 adapter->failover_pending,
1634 adapter->force_reset_recovery);
1636 /* If device failover is pending, just set device state and return.
1637 * Device operation will be handled by reset routine.
1639 if (adapter->failover_pending) {
1640 adapter->state = VNIC_CLOSED;
1644 rc = __ibmvnic_close(netdev);
1645 ibmvnic_cleanup(netdev);
1646 clean_rx_pools(adapter);
1647 clean_tx_pools(adapter);
1653 * build_hdr_data - creates L2/L3/L4 header data buffer
1654 * @hdr_field: bitfield determining needed headers
1655 * @skb: socket buffer
1656 * @hdr_len: array of header lengths
1657 * @hdr_data: buffer to write the header to
1659 * Reads hdr_field to determine which headers are needed by firmware.
1660 * Builds a buffer containing these headers. Saves individual header
1661 * lengths and total buffer length to be used to build descriptors.
1663 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1664 int *hdr_len, u8 *hdr_data)
1669 if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1670 hdr_len[0] = sizeof(struct vlan_ethhdr);
1672 hdr_len[0] = sizeof(struct ethhdr);
1674 if (skb->protocol == htons(ETH_P_IP)) {
1675 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1676 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1677 hdr_len[2] = tcp_hdrlen(skb);
1678 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1679 hdr_len[2] = sizeof(struct udphdr);
1680 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1681 hdr_len[1] = sizeof(struct ipv6hdr);
1682 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1683 hdr_len[2] = tcp_hdrlen(skb);
1684 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1685 hdr_len[2] = sizeof(struct udphdr);
1686 } else if (skb->protocol == htons(ETH_P_ARP)) {
1687 hdr_len[1] = arp_hdr_len(skb->dev);
1691 memset(hdr_data, 0, 120);
1692 if ((hdr_field >> 6) & 1) {
1693 hdr = skb_mac_header(skb);
1694 memcpy(hdr_data, hdr, hdr_len[0]);
1698 if ((hdr_field >> 5) & 1) {
1699 hdr = skb_network_header(skb);
1700 memcpy(hdr_data + len, hdr, hdr_len[1]);
1704 if ((hdr_field >> 4) & 1) {
1705 hdr = skb_transport_header(skb);
1706 memcpy(hdr_data + len, hdr, hdr_len[2]);
1713 * create_hdr_descs - create header and header extension descriptors
1714 * @hdr_field: bitfield determining needed headers
1715 * @hdr_data: buffer containing header data
1716 * @len: length of data buffer
1717 * @hdr_len: array of individual header lengths
1718 * @scrq_arr: descriptor array
1720 * Creates header and, if needed, header extension descriptors and
1721 * places them in a descriptor array, scrq_arr
1724 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1725 union sub_crq *scrq_arr)
1727 union sub_crq hdr_desc;
1733 while (tmp_len > 0) {
1734 cur = hdr_data + len - tmp_len;
1736 memset(&hdr_desc, 0, sizeof(hdr_desc));
1737 if (cur != hdr_data) {
1738 data = hdr_desc.hdr_ext.data;
1739 tmp = tmp_len > 29 ? 29 : tmp_len;
1740 hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1741 hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1742 hdr_desc.hdr_ext.len = tmp;
1744 data = hdr_desc.hdr.data;
1745 tmp = tmp_len > 24 ? 24 : tmp_len;
1746 hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1747 hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1748 hdr_desc.hdr.len = tmp;
1749 hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1750 hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1751 hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1752 hdr_desc.hdr.flag = hdr_field << 1;
1754 memcpy(data, cur, tmp);
1756 *scrq_arr = hdr_desc;
1765 * build_hdr_descs_arr - build a header descriptor array
1766 * @skb: tx socket buffer
1767 * @indir_arr: indirect array
1768 * @num_entries: number of descriptors to be sent
1769 * @hdr_field: bit field determining which headers will be sent
1771 * This function will build a TX descriptor array with applicable
1772 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1775 static void build_hdr_descs_arr(struct sk_buff *skb,
1776 union sub_crq *indir_arr,
1777 int *num_entries, u8 hdr_field)
1779 int hdr_len[3] = {0, 0, 0};
1780 u8 hdr_data[140] = {0};
1783 tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1785 *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1789 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1790 struct net_device *netdev)
1792 /* For some backing devices, mishandling of small packets
1793 * can result in a loss of connection or TX stall. Device
1794 * architects recommend that no packet should be smaller
1795 * than the minimum MTU value provided to the driver, so
1796 * pad any packets to that length
1798 if (skb->len < netdev->min_mtu)
1799 return skb_put_padto(skb, netdev->min_mtu);
1804 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1805 struct ibmvnic_sub_crq_queue *tx_scrq)
1807 struct ibmvnic_ind_xmit_queue *ind_bufp;
1808 struct ibmvnic_tx_buff *tx_buff;
1809 struct ibmvnic_tx_pool *tx_pool;
1810 union sub_crq tx_scrq_entry;
1816 ind_bufp = &tx_scrq->ind_buf;
1817 entries = (u64)ind_bufp->index;
1818 queue_num = tx_scrq->pool_index;
1820 for (i = entries - 1; i >= 0; --i) {
1821 tx_scrq_entry = ind_bufp->indir_arr[i];
1822 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1824 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1825 if (index & IBMVNIC_TSO_POOL_MASK) {
1826 tx_pool = &adapter->tso_pool[queue_num];
1827 index &= ~IBMVNIC_TSO_POOL_MASK;
1829 tx_pool = &adapter->tx_pool[queue_num];
1831 tx_pool->free_map[tx_pool->consumer_index] = index;
1832 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1833 tx_pool->num_buffers - 1 :
1834 tx_pool->consumer_index - 1;
1835 tx_buff = &tx_pool->tx_buff[index];
1836 adapter->netdev->stats.tx_packets--;
1837 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1838 adapter->tx_stats_buffers[queue_num].packets--;
1839 adapter->tx_stats_buffers[queue_num].bytes -=
1841 dev_kfree_skb_any(tx_buff->skb);
1842 tx_buff->skb = NULL;
1843 adapter->netdev->stats.tx_dropped++;
1845 ind_bufp->index = 0;
1846 if (atomic_sub_return(entries, &tx_scrq->used) <=
1847 (adapter->req_tx_entries_per_subcrq / 2) &&
1848 __netif_subqueue_stopped(adapter->netdev, queue_num) &&
1849 !test_bit(0, &adapter->resetting)) {
1850 netif_wake_subqueue(adapter->netdev, queue_num);
1851 netdev_dbg(adapter->netdev, "Started queue %d\n",
1856 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1857 struct ibmvnic_sub_crq_queue *tx_scrq)
1859 struct ibmvnic_ind_xmit_queue *ind_bufp;
1865 ind_bufp = &tx_scrq->ind_buf;
1866 dma_addr = (u64)ind_bufp->indir_dma;
1867 entries = (u64)ind_bufp->index;
1868 handle = tx_scrq->handle;
1872 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1874 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1876 ind_bufp->index = 0;
1880 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1882 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1883 int queue_num = skb_get_queue_mapping(skb);
1884 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1885 struct device *dev = &adapter->vdev->dev;
1886 struct ibmvnic_ind_xmit_queue *ind_bufp;
1887 struct ibmvnic_tx_buff *tx_buff = NULL;
1888 struct ibmvnic_sub_crq_queue *tx_scrq;
1889 struct ibmvnic_tx_pool *tx_pool;
1890 unsigned int tx_send_failed = 0;
1891 netdev_tx_t ret = NETDEV_TX_OK;
1892 unsigned int tx_map_failed = 0;
1893 union sub_crq indir_arr[16];
1894 unsigned int tx_dropped = 0;
1895 unsigned int tx_packets = 0;
1896 unsigned int tx_bytes = 0;
1897 dma_addr_t data_dma_addr;
1898 struct netdev_queue *txq;
1899 unsigned long lpar_rc;
1900 union sub_crq tx_crq;
1901 unsigned int offset;
1902 int num_entries = 1;
1907 tx_scrq = adapter->tx_scrq[queue_num];
1908 txq = netdev_get_tx_queue(netdev, queue_num);
1909 ind_bufp = &tx_scrq->ind_buf;
1911 if (test_bit(0, &adapter->resetting)) {
1912 dev_kfree_skb_any(skb);
1920 if (ibmvnic_xmit_workarounds(skb, netdev)) {
1924 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1927 if (skb_is_gso(skb))
1928 tx_pool = &adapter->tso_pool[queue_num];
1930 tx_pool = &adapter->tx_pool[queue_num];
1932 index = tx_pool->free_map[tx_pool->consumer_index];
1934 if (index == IBMVNIC_INVALID_MAP) {
1935 dev_kfree_skb_any(skb);
1938 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1943 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1945 offset = index * tx_pool->buf_size;
1946 dst = tx_pool->long_term_buff.buff + offset;
1947 memset(dst, 0, tx_pool->buf_size);
1948 data_dma_addr = tx_pool->long_term_buff.addr + offset;
1950 if (skb_shinfo(skb)->nr_frags) {
1954 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1955 cur = skb_headlen(skb);
1957 /* Copy the frags */
1958 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1959 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1961 memcpy(dst + cur, skb_frag_address(frag),
1962 skb_frag_size(frag));
1963 cur += skb_frag_size(frag);
1966 skb_copy_from_linear_data(skb, dst, skb->len);
1969 /* post changes to long_term_buff *dst before VIOS accessing it */
1972 tx_pool->consumer_index =
1973 (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1975 tx_buff = &tx_pool->tx_buff[index];
1977 tx_buff->index = index;
1978 tx_buff->pool_index = queue_num;
1980 memset(&tx_crq, 0, sizeof(tx_crq));
1981 tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1982 tx_crq.v1.type = IBMVNIC_TX_DESC;
1983 tx_crq.v1.n_crq_elem = 1;
1984 tx_crq.v1.n_sge = 1;
1985 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1987 if (skb_is_gso(skb))
1988 tx_crq.v1.correlator =
1989 cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1991 tx_crq.v1.correlator = cpu_to_be32(index);
1992 tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1993 tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1994 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1996 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1997 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1998 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
2001 if (skb->protocol == htons(ETH_P_IP)) {
2002 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
2003 proto = ip_hdr(skb)->protocol;
2004 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2005 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2006 proto = ipv6_hdr(skb)->nexthdr;
2009 if (proto == IPPROTO_TCP)
2010 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2011 else if (proto == IPPROTO_UDP)
2012 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2014 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2015 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2018 if (skb_is_gso(skb)) {
2019 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2020 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2024 if ((*hdrs >> 7) & 1)
2025 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
2027 tx_crq.v1.n_crq_elem = num_entries;
2028 tx_buff->num_entries = num_entries;
2029 /* flush buffer if current entry can not fit */
2030 if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2031 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2032 if (lpar_rc != H_SUCCESS)
2036 indir_arr[0] = tx_crq;
2037 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2038 num_entries * sizeof(struct ibmvnic_generic_scrq));
2039 ind_bufp->index += num_entries;
2040 if (__netdev_tx_sent_queue(txq, skb->len,
2041 netdev_xmit_more() &&
2042 ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2043 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2044 if (lpar_rc != H_SUCCESS)
2048 if (atomic_add_return(num_entries, &tx_scrq->used)
2049 >= adapter->req_tx_entries_per_subcrq) {
2050 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2051 netif_stop_subqueue(netdev, queue_num);
2055 tx_bytes += skb->len;
2056 txq_trans_cond_update(txq);
2061 dev_kfree_skb_any(skb);
2062 tx_buff->skb = NULL;
2063 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2064 tx_pool->num_buffers - 1 :
2065 tx_pool->consumer_index - 1;
2068 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2069 dev_err_ratelimited(dev, "tx: send failed\n");
2071 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2072 /* Disable TX and report carrier off if queue is closed
2073 * or pending failover.
2074 * Firmware guarantees that a signal will be sent to the
2075 * driver, triggering a reset or some other action.
2077 netif_tx_stop_all_queues(netdev);
2078 netif_carrier_off(netdev);
2081 netdev->stats.tx_dropped += tx_dropped;
2082 netdev->stats.tx_bytes += tx_bytes;
2083 netdev->stats.tx_packets += tx_packets;
2084 adapter->tx_send_failed += tx_send_failed;
2085 adapter->tx_map_failed += tx_map_failed;
2086 adapter->tx_stats_buffers[queue_num].packets += tx_packets;
2087 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2088 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2093 static void ibmvnic_set_multi(struct net_device *netdev)
2095 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2096 struct netdev_hw_addr *ha;
2097 union ibmvnic_crq crq;
2099 memset(&crq, 0, sizeof(crq));
2100 crq.request_capability.first = IBMVNIC_CRQ_CMD;
2101 crq.request_capability.cmd = REQUEST_CAPABILITY;
2103 if (netdev->flags & IFF_PROMISC) {
2104 if (!adapter->promisc_supported)
2107 if (netdev->flags & IFF_ALLMULTI) {
2108 /* Accept all multicast */
2109 memset(&crq, 0, sizeof(crq));
2110 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2111 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2112 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2113 ibmvnic_send_crq(adapter, &crq);
2114 } else if (netdev_mc_empty(netdev)) {
2115 /* Reject all multicast */
2116 memset(&crq, 0, sizeof(crq));
2117 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2118 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2119 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2120 ibmvnic_send_crq(adapter, &crq);
2122 /* Accept one or more multicast(s) */
2123 netdev_for_each_mc_addr(ha, netdev) {
2124 memset(&crq, 0, sizeof(crq));
2125 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2126 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2127 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2128 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
2130 ibmvnic_send_crq(adapter, &crq);
2136 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2138 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2139 union ibmvnic_crq crq;
2142 if (!is_valid_ether_addr(dev_addr)) {
2143 rc = -EADDRNOTAVAIL;
2147 memset(&crq, 0, sizeof(crq));
2148 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2149 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2150 ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
2152 mutex_lock(&adapter->fw_lock);
2153 adapter->fw_done_rc = 0;
2154 reinit_completion(&adapter->fw_done);
2156 rc = ibmvnic_send_crq(adapter, &crq);
2159 mutex_unlock(&adapter->fw_lock);
2163 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
2164 /* netdev->dev_addr is changed in handle_change_mac_rsp function */
2165 if (rc || adapter->fw_done_rc) {
2167 mutex_unlock(&adapter->fw_lock);
2170 mutex_unlock(&adapter->fw_lock);
2173 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2177 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2179 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2180 struct sockaddr *addr = p;
2184 if (!is_valid_ether_addr(addr->sa_data))
2185 return -EADDRNOTAVAIL;
2187 ether_addr_copy(adapter->mac_addr, addr->sa_data);
2188 if (adapter->state != VNIC_PROBED)
2189 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2194 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2197 case VNIC_RESET_FAILOVER:
2199 case VNIC_RESET_MOBILITY:
2201 case VNIC_RESET_FATAL:
2203 case VNIC_RESET_NON_FATAL:
2205 case VNIC_RESET_TIMEOUT:
2207 case VNIC_RESET_CHANGE_PARAM:
2208 return "CHANGE_PARAM";
2209 case VNIC_RESET_PASSIVE_INIT:
2210 return "PASSIVE_INIT";
2216 * do_reset returns zero if we are able to keep processing reset events, or
2217 * non-zero if we hit a fatal error and must halt.
2219 static int do_reset(struct ibmvnic_adapter *adapter,
2220 struct ibmvnic_rwi *rwi, u32 reset_state)
2222 struct net_device *netdev = adapter->netdev;
2223 u64 old_num_rx_queues, old_num_tx_queues;
2224 u64 old_num_rx_slots, old_num_tx_slots;
2227 netdev_dbg(adapter->netdev,
2228 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2229 adapter_state_to_string(adapter->state),
2230 adapter->failover_pending,
2231 reset_reason_to_string(rwi->reset_reason),
2232 adapter_state_to_string(reset_state));
2234 adapter->reset_reason = rwi->reset_reason;
2235 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2236 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2239 /* Now that we have the rtnl lock, clear any pending failover.
2240 * This will ensure ibmvnic_open() has either completed or will
2241 * block until failover is complete.
2243 if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2244 adapter->failover_pending = false;
2246 /* read the state and check (again) after getting rtnl */
2247 reset_state = adapter->state;
2249 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2254 netif_carrier_off(netdev);
2256 old_num_rx_queues = adapter->req_rx_queues;
2257 old_num_tx_queues = adapter->req_tx_queues;
2258 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2259 old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2261 ibmvnic_cleanup(netdev);
2263 if (reset_state == VNIC_OPEN &&
2264 adapter->reset_reason != VNIC_RESET_MOBILITY &&
2265 adapter->reset_reason != VNIC_RESET_FAILOVER) {
2266 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2267 rc = __ibmvnic_close(netdev);
2271 adapter->state = VNIC_CLOSING;
2273 /* Release the RTNL lock before link state change and
2274 * re-acquire after the link state change to allow
2275 * linkwatch_event to grab the RTNL lock and run during
2279 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2284 if (adapter->state == VNIC_OPEN) {
2285 /* When we dropped rtnl, ibmvnic_open() got
2286 * it and noticed that we are resetting and
2287 * set the adapter state to OPEN. Update our
2288 * new "target" state, and resume the reset
2289 * from VNIC_CLOSING state.
2292 "Open changed state from %s, updating.\n",
2293 adapter_state_to_string(reset_state));
2294 reset_state = VNIC_OPEN;
2295 adapter->state = VNIC_CLOSING;
2298 if (adapter->state != VNIC_CLOSING) {
2299 /* If someone else changed the adapter state
2300 * when we dropped the rtnl, fail the reset
2305 adapter->state = VNIC_CLOSED;
2309 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2310 release_resources(adapter);
2311 release_sub_crqs(adapter, 1);
2312 release_crq_queue(adapter);
2315 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2316 /* remove the closed state so when we call open it appears
2317 * we are coming from the probed state.
2319 adapter->state = VNIC_PROBED;
2321 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2322 rc = init_crq_queue(adapter);
2323 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2324 rc = ibmvnic_reenable_crq_queue(adapter);
2325 release_sub_crqs(adapter, 1);
2327 rc = ibmvnic_reset_crq(adapter);
2328 if (rc == H_CLOSED || rc == H_SUCCESS) {
2329 rc = vio_enable_interrupts(adapter->vdev);
2331 netdev_err(adapter->netdev,
2332 "Reset failed to enable interrupts. rc=%d\n",
2338 netdev_err(adapter->netdev,
2339 "Reset couldn't initialize crq. rc=%d\n", rc);
2343 rc = ibmvnic_reset_init(adapter, true);
2347 /* If the adapter was in PROBE or DOWN state prior to the reset,
2350 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2355 rc = ibmvnic_login(netdev);
2359 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2360 rc = init_resources(adapter);
2363 } else if (adapter->req_rx_queues != old_num_rx_queues ||
2364 adapter->req_tx_queues != old_num_tx_queues ||
2365 adapter->req_rx_add_entries_per_subcrq !=
2367 adapter->req_tx_entries_per_subcrq !=
2369 !adapter->rx_pool ||
2370 !adapter->tso_pool ||
2371 !adapter->tx_pool) {
2372 release_napi(adapter);
2373 release_vpd_data(adapter);
2375 rc = init_resources(adapter);
2380 rc = init_tx_pools(netdev);
2383 "init tx pools failed (%d)\n",
2388 rc = init_rx_pools(netdev);
2391 "init rx pools failed (%d)\n",
2396 ibmvnic_disable_irqs(adapter);
2398 adapter->state = VNIC_CLOSED;
2400 if (reset_state == VNIC_CLOSED) {
2405 rc = __ibmvnic_open(netdev);
2407 rc = IBMVNIC_OPEN_FAILED;
2411 /* refresh device's multicast list */
2412 ibmvnic_set_multi(netdev);
2414 if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2415 adapter->reset_reason == VNIC_RESET_MOBILITY)
2416 __netdev_notify_peers(netdev);
2421 /* restore the adapter state if reset failed */
2423 adapter->state = reset_state;
2424 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2425 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2428 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
2429 adapter_state_to_string(adapter->state),
2430 adapter->failover_pending, rc);
2434 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2435 struct ibmvnic_rwi *rwi, u32 reset_state)
2437 struct net_device *netdev = adapter->netdev;
2440 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
2441 reset_reason_to_string(rwi->reset_reason));
2443 /* read the state and check (again) after getting rtnl */
2444 reset_state = adapter->state;
2446 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2451 netif_carrier_off(netdev);
2452 adapter->reset_reason = rwi->reset_reason;
2454 ibmvnic_cleanup(netdev);
2455 release_resources(adapter);
2456 release_sub_crqs(adapter, 0);
2457 release_crq_queue(adapter);
2459 /* remove the closed state so when we call open it appears
2460 * we are coming from the probed state.
2462 adapter->state = VNIC_PROBED;
2464 reinit_completion(&adapter->init_done);
2465 rc = init_crq_queue(adapter);
2467 netdev_err(adapter->netdev,
2468 "Couldn't initialize crq. rc=%d\n", rc);
2472 rc = ibmvnic_reset_init(adapter, false);
2476 /* If the adapter was in PROBE or DOWN state prior to the reset,
2479 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
2482 rc = ibmvnic_login(netdev);
2486 rc = init_resources(adapter);
2490 ibmvnic_disable_irqs(adapter);
2491 adapter->state = VNIC_CLOSED;
2493 if (reset_state == VNIC_CLOSED)
2496 rc = __ibmvnic_open(netdev);
2498 rc = IBMVNIC_OPEN_FAILED;
2502 __netdev_notify_peers(netdev);
2504 /* restore adapter state if reset failed */
2506 adapter->state = reset_state;
2507 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
2508 adapter_state_to_string(adapter->state),
2509 adapter->failover_pending, rc);
2513 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2515 struct ibmvnic_rwi *rwi;
2516 unsigned long flags;
2518 spin_lock_irqsave(&adapter->rwi_lock, flags);
2520 if (!list_empty(&adapter->rwi_list)) {
2521 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2523 list_del(&rwi->list);
2528 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2533 * do_passive_init - complete probing when partner device is detected.
2534 * @adapter: ibmvnic_adapter struct
2536 * If the ibmvnic device does not have a partner device to communicate with at boot
2537 * and that partner device comes online at a later time, this function is called
2538 * to complete the initialization process of ibmvnic device.
2539 * Caller is expected to hold rtnl_lock().
2541 * Returns non-zero if sub-CRQs are not initialized properly leaving the device
2542 * in the down state.
2543 * Returns 0 upon success and the device is in PROBED state.
2546 static int do_passive_init(struct ibmvnic_adapter *adapter)
2548 unsigned long timeout = msecs_to_jiffies(30000);
2549 struct net_device *netdev = adapter->netdev;
2550 struct device *dev = &adapter->vdev->dev;
2553 netdev_dbg(netdev, "Partner device found, probing.\n");
2555 adapter->state = VNIC_PROBING;
2556 reinit_completion(&adapter->init_done);
2557 adapter->init_done_rc = 0;
2558 adapter->crq.active = true;
2560 rc = send_crq_init_complete(adapter);
2564 rc = send_version_xchg(adapter);
2566 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
2568 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
2569 dev_err(dev, "Initialization sequence timed out\n");
2574 rc = init_sub_crqs(adapter);
2576 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
2580 rc = init_sub_crq_irqs(adapter);
2582 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
2586 netdev->mtu = adapter->req_mtu - ETH_HLEN;
2587 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
2588 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
2590 adapter->state = VNIC_PROBED;
2591 netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
2596 release_sub_crqs(adapter, 1);
2598 adapter->state = VNIC_DOWN;
2602 static void __ibmvnic_reset(struct work_struct *work)
2604 struct ibmvnic_adapter *adapter;
2605 bool saved_state = false;
2606 struct ibmvnic_rwi *tmprwi;
2607 struct ibmvnic_rwi *rwi;
2608 unsigned long flags;
2613 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2615 if (test_and_set_bit_lock(0, &adapter->resetting)) {
2616 queue_delayed_work(system_long_wq,
2617 &adapter->ibmvnic_delayed_reset,
2618 IBMVNIC_RESET_DELAY);
2622 rwi = get_next_rwi(adapter);
2624 spin_lock_irqsave(&adapter->state_lock, flags);
2626 if (adapter->state == VNIC_REMOVING ||
2627 adapter->state == VNIC_REMOVED) {
2628 spin_unlock_irqrestore(&adapter->state_lock, flags);
2635 reset_state = adapter->state;
2638 spin_unlock_irqrestore(&adapter->state_lock, flags);
2640 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
2642 rc = do_passive_init(adapter);
2645 netif_carrier_on(adapter->netdev);
2646 } else if (adapter->force_reset_recovery) {
2647 /* Since we are doing a hard reset now, clear the
2648 * failover_pending flag so we don't ignore any
2649 * future MOBILITY or other resets.
2651 adapter->failover_pending = false;
2653 /* Transport event occurred during previous reset */
2654 if (adapter->wait_for_reset) {
2655 /* Previous was CHANGE_PARAM; caller locked */
2656 adapter->force_reset_recovery = false;
2657 rc = do_hard_reset(adapter, rwi, reset_state);
2660 adapter->force_reset_recovery = false;
2661 rc = do_hard_reset(adapter, rwi, reset_state);
2669 /* If auto-priority-failover is enabled we can get
2670 * back to back failovers during resets, resulting
2671 * in at least two failed resets (from high-priority
2672 * backing device to low-priority one and then back)
2673 * If resets continue to fail beyond that, give the
2674 * adapter some time to settle down before retrying.
2676 if (num_fails >= 3) {
2677 netdev_dbg(adapter->netdev,
2678 "[S:%s] Hard reset failed %d times, waiting 60 secs\n",
2679 adapter_state_to_string(adapter->state),
2681 set_current_state(TASK_UNINTERRUPTIBLE);
2682 schedule_timeout(60 * HZ);
2685 rc = do_reset(adapter, rwi, reset_state);
2688 adapter->last_reset_time = jiffies;
2691 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2693 rwi = get_next_rwi(adapter);
2696 * If there is another reset queued, free the previous rwi
2697 * and process the new reset even if previous reset failed
2698 * (the previous reset could have failed because of a fail
2699 * over for instance, so process the fail over).
2701 * If there are no resets queued and the previous reset failed,
2702 * the adapter would be in an undefined state. So retry the
2703 * previous reset as a hard reset.
2710 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2711 rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
2712 adapter->force_reset_recovery = true;
2715 if (adapter->wait_for_reset) {
2716 adapter->reset_done_rc = rc;
2717 complete(&adapter->reset_done);
2720 clear_bit_unlock(0, &adapter->resetting);
2722 netdev_dbg(adapter->netdev,
2723 "[S:%s FRR:%d WFR:%d] Done processing resets\n",
2724 adapter_state_to_string(adapter->state),
2725 adapter->force_reset_recovery,
2726 adapter->wait_for_reset);
2729 static void __ibmvnic_delayed_reset(struct work_struct *work)
2731 struct ibmvnic_adapter *adapter;
2733 adapter = container_of(work, struct ibmvnic_adapter,
2734 ibmvnic_delayed_reset.work);
2735 __ibmvnic_reset(&adapter->ibmvnic_reset);
2738 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2739 enum ibmvnic_reset_reason reason)
2741 struct list_head *entry, *tmp_entry;
2742 struct ibmvnic_rwi *rwi, *tmp;
2743 struct net_device *netdev = adapter->netdev;
2744 unsigned long flags;
2747 spin_lock_irqsave(&adapter->rwi_lock, flags);
2749 /* If failover is pending don't schedule any other reset.
2750 * Instead let the failover complete. If there is already a
2751 * a failover reset scheduled, we will detect and drop the
2752 * duplicate reset when walking the ->rwi_list below.
2754 if (adapter->state == VNIC_REMOVING ||
2755 adapter->state == VNIC_REMOVED ||
2756 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2758 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2762 if (adapter->state == VNIC_PROBING) {
2763 netdev_warn(netdev, "Adapter reset during probe\n");
2764 adapter->init_done_rc = -EAGAIN;
2769 list_for_each_entry(tmp, &adapter->rwi_list, list) {
2770 if (tmp->reset_reason == reason) {
2771 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
2772 reset_reason_to_string(reason));
2778 rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2783 /* if we just received a transport event,
2784 * flush reset queue and process this reset
2786 if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2787 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2790 rwi->reset_reason = reason;
2791 list_add_tail(&rwi->list, &adapter->rwi_list);
2792 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
2793 reset_reason_to_string(reason));
2794 queue_work(system_long_wq, &adapter->ibmvnic_reset);
2798 /* ibmvnic_close() below can block, so drop the lock first */
2799 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2802 ibmvnic_close(netdev);
2807 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2809 struct ibmvnic_adapter *adapter = netdev_priv(dev);
2811 if (test_bit(0, &adapter->resetting)) {
2812 netdev_err(adapter->netdev,
2813 "Adapter is resetting, skip timeout reset\n");
2816 /* No queuing up reset until at least 5 seconds (default watchdog val)
2819 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2820 netdev_dbg(dev, "Not yet time to tx timeout.\n");
2823 ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2826 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2827 struct ibmvnic_rx_buff *rx_buff)
2829 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2831 rx_buff->skb = NULL;
2833 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2834 pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2836 atomic_dec(&pool->available);
2839 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2841 struct ibmvnic_sub_crq_queue *rx_scrq;
2842 struct ibmvnic_adapter *adapter;
2843 struct net_device *netdev;
2844 int frames_processed;
2848 adapter = netdev_priv(netdev);
2849 scrq_num = (int)(napi - adapter->napi);
2850 frames_processed = 0;
2851 rx_scrq = adapter->rx_scrq[scrq_num];
2854 while (frames_processed < budget) {
2855 struct sk_buff *skb;
2856 struct ibmvnic_rx_buff *rx_buff;
2857 union sub_crq *next;
2862 if (unlikely(test_bit(0, &adapter->resetting) &&
2863 adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2864 enable_scrq_irq(adapter, rx_scrq);
2865 napi_complete_done(napi, frames_processed);
2866 return frames_processed;
2869 if (!pending_scrq(adapter, rx_scrq))
2871 next = ibmvnic_next_scrq(adapter, rx_scrq);
2872 rx_buff = (struct ibmvnic_rx_buff *)
2873 be64_to_cpu(next->rx_comp.correlator);
2874 /* do error checking */
2875 if (next->rx_comp.rc) {
2876 netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2877 be16_to_cpu(next->rx_comp.rc));
2878 /* free the entry */
2879 next->rx_comp.first = 0;
2880 dev_kfree_skb_any(rx_buff->skb);
2881 remove_buff_from_pool(adapter, rx_buff);
2883 } else if (!rx_buff->skb) {
2884 /* free the entry */
2885 next->rx_comp.first = 0;
2886 remove_buff_from_pool(adapter, rx_buff);
2890 length = be32_to_cpu(next->rx_comp.len);
2891 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2892 flags = next->rx_comp.flags;
2894 /* load long_term_buff before copying to skb */
2896 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2899 /* VLAN Header has been stripped by the system firmware and
2900 * needs to be inserted by the driver
2902 if (adapter->rx_vlan_header_insertion &&
2903 (flags & IBMVNIC_VLAN_STRIPPED))
2904 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2905 ntohs(next->rx_comp.vlan_tci));
2907 /* free the entry */
2908 next->rx_comp.first = 0;
2909 remove_buff_from_pool(adapter, rx_buff);
2911 skb_put(skb, length);
2912 skb->protocol = eth_type_trans(skb, netdev);
2913 skb_record_rx_queue(skb, scrq_num);
2915 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2916 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2917 skb->ip_summed = CHECKSUM_UNNECESSARY;
2921 napi_gro_receive(napi, skb); /* send it up */
2922 netdev->stats.rx_packets++;
2923 netdev->stats.rx_bytes += length;
2924 adapter->rx_stats_buffers[scrq_num].packets++;
2925 adapter->rx_stats_buffers[scrq_num].bytes += length;
2929 if (adapter->state != VNIC_CLOSING &&
2930 ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2931 adapter->req_rx_add_entries_per_subcrq / 2) ||
2932 frames_processed < budget))
2933 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2934 if (frames_processed < budget) {
2935 if (napi_complete_done(napi, frames_processed)) {
2936 enable_scrq_irq(adapter, rx_scrq);
2937 if (pending_scrq(adapter, rx_scrq)) {
2938 if (napi_reschedule(napi)) {
2939 disable_scrq_irq(adapter, rx_scrq);
2945 return frames_processed;
2948 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2952 adapter->fallback.mtu = adapter->req_mtu;
2953 adapter->fallback.rx_queues = adapter->req_rx_queues;
2954 adapter->fallback.tx_queues = adapter->req_tx_queues;
2955 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2956 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2958 reinit_completion(&adapter->reset_done);
2959 adapter->wait_for_reset = true;
2960 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2966 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2973 if (adapter->reset_done_rc) {
2975 adapter->desired.mtu = adapter->fallback.mtu;
2976 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2977 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2978 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2979 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2981 reinit_completion(&adapter->reset_done);
2982 adapter->wait_for_reset = true;
2983 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2988 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2996 adapter->wait_for_reset = false;
3001 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
3003 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3005 adapter->desired.mtu = new_mtu + ETH_HLEN;
3007 return wait_for_reset(adapter);
3010 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
3011 struct net_device *dev,
3012 netdev_features_t features)
3014 /* Some backing hardware adapters can not
3015 * handle packets with a MSS less than 224
3016 * or with only one segment.
3018 if (skb_is_gso(skb)) {
3019 if (skb_shinfo(skb)->gso_size < 224 ||
3020 skb_shinfo(skb)->gso_segs == 1)
3021 features &= ~NETIF_F_GSO_MASK;
3027 static const struct net_device_ops ibmvnic_netdev_ops = {
3028 .ndo_open = ibmvnic_open,
3029 .ndo_stop = ibmvnic_close,
3030 .ndo_start_xmit = ibmvnic_xmit,
3031 .ndo_set_rx_mode = ibmvnic_set_multi,
3032 .ndo_set_mac_address = ibmvnic_set_mac,
3033 .ndo_validate_addr = eth_validate_addr,
3034 .ndo_tx_timeout = ibmvnic_tx_timeout,
3035 .ndo_change_mtu = ibmvnic_change_mtu,
3036 .ndo_features_check = ibmvnic_features_check,
3039 /* ethtool functions */
3041 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3042 struct ethtool_link_ksettings *cmd)
3044 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3047 rc = send_query_phys_parms(adapter);
3049 adapter->speed = SPEED_UNKNOWN;
3050 adapter->duplex = DUPLEX_UNKNOWN;
3052 cmd->base.speed = adapter->speed;
3053 cmd->base.duplex = adapter->duplex;
3054 cmd->base.port = PORT_FIBRE;
3055 cmd->base.phy_address = 0;
3056 cmd->base.autoneg = AUTONEG_ENABLE;
3061 static void ibmvnic_get_drvinfo(struct net_device *netdev,
3062 struct ethtool_drvinfo *info)
3064 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3066 strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
3067 strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
3068 strscpy(info->fw_version, adapter->fw_version,
3069 sizeof(info->fw_version));
3072 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3074 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3076 return adapter->msg_enable;
3079 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3081 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3083 adapter->msg_enable = data;
3086 static u32 ibmvnic_get_link(struct net_device *netdev)
3088 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3090 /* Don't need to send a query because we request a logical link up at
3091 * init and then we wait for link state indications
3093 return adapter->logical_link_state;
3096 static void ibmvnic_get_ringparam(struct net_device *netdev,
3097 struct ethtool_ringparam *ring,
3098 struct kernel_ethtool_ringparam *kernel_ring,
3099 struct netlink_ext_ack *extack)
3101 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3103 if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
3104 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3105 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3107 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
3108 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
3110 ring->rx_mini_max_pending = 0;
3111 ring->rx_jumbo_max_pending = 0;
3112 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3113 ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3114 ring->rx_mini_pending = 0;
3115 ring->rx_jumbo_pending = 0;
3118 static int ibmvnic_set_ringparam(struct net_device *netdev,
3119 struct ethtool_ringparam *ring,
3120 struct kernel_ethtool_ringparam *kernel_ring,
3121 struct netlink_ext_ack *extack)
3123 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3127 adapter->desired.rx_entries = ring->rx_pending;
3128 adapter->desired.tx_entries = ring->tx_pending;
3130 ret = wait_for_reset(adapter);
3133 (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
3134 adapter->req_tx_entries_per_subcrq != ring->tx_pending))
3136 "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
3137 ring->rx_pending, ring->tx_pending,
3138 adapter->req_rx_add_entries_per_subcrq,
3139 adapter->req_tx_entries_per_subcrq);
3143 static void ibmvnic_get_channels(struct net_device *netdev,
3144 struct ethtool_channels *channels)
3146 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3148 if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
3149 channels->max_rx = adapter->max_rx_queues;
3150 channels->max_tx = adapter->max_tx_queues;
3152 channels->max_rx = IBMVNIC_MAX_QUEUES;
3153 channels->max_tx = IBMVNIC_MAX_QUEUES;
3156 channels->max_other = 0;
3157 channels->max_combined = 0;
3158 channels->rx_count = adapter->req_rx_queues;
3159 channels->tx_count = adapter->req_tx_queues;
3160 channels->other_count = 0;
3161 channels->combined_count = 0;
3164 static int ibmvnic_set_channels(struct net_device *netdev,
3165 struct ethtool_channels *channels)
3167 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3171 adapter->desired.rx_queues = channels->rx_count;
3172 adapter->desired.tx_queues = channels->tx_count;
3174 ret = wait_for_reset(adapter);
3177 (adapter->req_rx_queues != channels->rx_count ||
3178 adapter->req_tx_queues != channels->tx_count))
3180 "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
3181 channels->rx_count, channels->tx_count,
3182 adapter->req_rx_queues, adapter->req_tx_queues);
3186 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3188 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3191 switch (stringset) {
3193 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
3194 i++, data += ETH_GSTRING_LEN)
3195 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3197 for (i = 0; i < adapter->req_tx_queues; i++) {
3198 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
3199 data += ETH_GSTRING_LEN;
3201 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3202 data += ETH_GSTRING_LEN;
3204 snprintf(data, ETH_GSTRING_LEN,
3205 "tx%d_dropped_packets", i);
3206 data += ETH_GSTRING_LEN;
3209 for (i = 0; i < adapter->req_rx_queues; i++) {
3210 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3211 data += ETH_GSTRING_LEN;
3213 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3214 data += ETH_GSTRING_LEN;
3216 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3217 data += ETH_GSTRING_LEN;
3221 case ETH_SS_PRIV_FLAGS:
3222 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
3223 strcpy(data + i * ETH_GSTRING_LEN,
3224 ibmvnic_priv_flags[i]);
3231 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3233 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3237 return ARRAY_SIZE(ibmvnic_stats) +
3238 adapter->req_tx_queues * NUM_TX_STATS +
3239 adapter->req_rx_queues * NUM_RX_STATS;
3240 case ETH_SS_PRIV_FLAGS:
3241 return ARRAY_SIZE(ibmvnic_priv_flags);
3247 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3248 struct ethtool_stats *stats, u64 *data)
3250 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3251 union ibmvnic_crq crq;
3255 memset(&crq, 0, sizeof(crq));
3256 crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3257 crq.request_statistics.cmd = REQUEST_STATISTICS;
3258 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3259 crq.request_statistics.len =
3260 cpu_to_be32(sizeof(struct ibmvnic_statistics));
3262 /* Wait for data to be written */
3263 reinit_completion(&adapter->stats_done);
3264 rc = ibmvnic_send_crq(adapter, &crq);
3267 rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3271 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3272 data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3273 (adapter, ibmvnic_stats[i].offset));
3275 for (j = 0; j < adapter->req_tx_queues; j++) {
3276 data[i] = adapter->tx_stats_buffers[j].packets;
3278 data[i] = adapter->tx_stats_buffers[j].bytes;
3280 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3284 for (j = 0; j < adapter->req_rx_queues; j++) {
3285 data[i] = adapter->rx_stats_buffers[j].packets;
3287 data[i] = adapter->rx_stats_buffers[j].bytes;
3289 data[i] = adapter->rx_stats_buffers[j].interrupts;
3294 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
3296 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3298 return adapter->priv_flags;
3301 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
3303 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3304 bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
3307 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
3309 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
3314 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3315 .get_drvinfo = ibmvnic_get_drvinfo,
3316 .get_msglevel = ibmvnic_get_msglevel,
3317 .set_msglevel = ibmvnic_set_msglevel,
3318 .get_link = ibmvnic_get_link,
3319 .get_ringparam = ibmvnic_get_ringparam,
3320 .set_ringparam = ibmvnic_set_ringparam,
3321 .get_channels = ibmvnic_get_channels,
3322 .set_channels = ibmvnic_set_channels,
3323 .get_strings = ibmvnic_get_strings,
3324 .get_sset_count = ibmvnic_get_sset_count,
3325 .get_ethtool_stats = ibmvnic_get_ethtool_stats,
3326 .get_link_ksettings = ibmvnic_get_link_ksettings,
3327 .get_priv_flags = ibmvnic_get_priv_flags,
3328 .set_priv_flags = ibmvnic_set_priv_flags,
3331 /* Routines for managing CRQs/sCRQs */
3333 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3334 struct ibmvnic_sub_crq_queue *scrq)
3339 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3344 free_irq(scrq->irq, scrq);
3345 irq_dispose_mapping(scrq->irq);
3350 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3351 atomic_set(&scrq->used, 0);
3353 scrq->ind_buf.index = 0;
3355 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3359 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3360 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3364 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3368 if (!adapter->tx_scrq || !adapter->rx_scrq)
3371 for (i = 0; i < adapter->req_tx_queues; i++) {
3372 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3373 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3378 for (i = 0; i < adapter->req_rx_queues; i++) {
3379 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3380 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3388 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3389 struct ibmvnic_sub_crq_queue *scrq,
3392 struct device *dev = &adapter->vdev->dev;
3395 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3398 /* Close the sub-crqs */
3400 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3401 adapter->vdev->unit_address,
3403 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3406 netdev_err(adapter->netdev,
3407 "Failed to release sub-CRQ %16lx, rc = %ld\n",
3412 dma_free_coherent(dev,
3414 scrq->ind_buf.indir_arr,
3415 scrq->ind_buf.indir_dma);
3417 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3419 free_pages((unsigned long)scrq->msgs, 2);
3423 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3426 struct device *dev = &adapter->vdev->dev;
3427 struct ibmvnic_sub_crq_queue *scrq;
3430 scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3435 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3437 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3438 goto zero_page_failed;
3441 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3443 if (dma_mapping_error(dev, scrq->msg_token)) {
3444 dev_warn(dev, "Couldn't map crq queue messages page\n");
3448 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3449 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3451 if (rc == H_RESOURCE)
3452 rc = ibmvnic_reset_crq(adapter);
3454 if (rc == H_CLOSED) {
3455 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3457 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3461 scrq->adapter = adapter;
3462 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3463 scrq->ind_buf.index = 0;
3465 scrq->ind_buf.indir_arr =
3466 dma_alloc_coherent(dev,
3468 &scrq->ind_buf.indir_dma,
3471 if (!scrq->ind_buf.indir_arr)
3474 spin_lock_init(&scrq->lock);
3476 netdev_dbg(adapter->netdev,
3477 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3478 scrq->crq_num, scrq->hw_irq, scrq->irq);
3484 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3485 adapter->vdev->unit_address,
3487 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3489 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3492 free_pages((unsigned long)scrq->msgs, 2);
3499 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3503 if (adapter->tx_scrq) {
3504 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3505 if (!adapter->tx_scrq[i])
3508 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3510 ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
3511 if (adapter->tx_scrq[i]->irq) {
3512 free_irq(adapter->tx_scrq[i]->irq,
3513 adapter->tx_scrq[i]);
3514 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3515 adapter->tx_scrq[i]->irq = 0;
3518 release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3522 kfree(adapter->tx_scrq);
3523 adapter->tx_scrq = NULL;
3524 adapter->num_active_tx_scrqs = 0;
3527 if (adapter->rx_scrq) {
3528 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3529 if (!adapter->rx_scrq[i])
3532 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3534 if (adapter->rx_scrq[i]->irq) {
3535 free_irq(adapter->rx_scrq[i]->irq,
3536 adapter->rx_scrq[i]);
3537 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3538 adapter->rx_scrq[i]->irq = 0;
3541 release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3545 kfree(adapter->rx_scrq);
3546 adapter->rx_scrq = NULL;
3547 adapter->num_active_rx_scrqs = 0;
3551 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3552 struct ibmvnic_sub_crq_queue *scrq)
3554 struct device *dev = &adapter->vdev->dev;
3557 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3558 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3560 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3565 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3566 struct ibmvnic_sub_crq_queue *scrq)
3568 struct device *dev = &adapter->vdev->dev;
3571 if (scrq->hw_irq > 0x100000000ULL) {
3572 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3576 if (test_bit(0, &adapter->resetting) &&
3577 adapter->reset_reason == VNIC_RESET_MOBILITY) {
3578 u64 val = (0xff000000) | scrq->hw_irq;
3580 rc = plpar_hcall_norets(H_EOI, val);
3581 /* H_EOI would fail with rc = H_FUNCTION when running
3582 * in XIVE mode which is expected, but not an error.
3584 if (rc && (rc != H_FUNCTION))
3585 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3589 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3590 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3592 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3597 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3598 struct ibmvnic_sub_crq_queue *scrq)
3600 struct device *dev = &adapter->vdev->dev;
3601 struct ibmvnic_tx_pool *tx_pool;
3602 struct ibmvnic_tx_buff *txbuff;
3603 struct netdev_queue *txq;
3604 union sub_crq *next;
3609 while (pending_scrq(adapter, scrq)) {
3610 unsigned int pool = scrq->pool_index;
3611 int num_entries = 0;
3612 int total_bytes = 0;
3613 int num_packets = 0;
3615 next = ibmvnic_next_scrq(adapter, scrq);
3616 for (i = 0; i < next->tx_comp.num_comps; i++) {
3617 index = be32_to_cpu(next->tx_comp.correlators[i]);
3618 if (index & IBMVNIC_TSO_POOL_MASK) {
3619 tx_pool = &adapter->tso_pool[pool];
3620 index &= ~IBMVNIC_TSO_POOL_MASK;
3622 tx_pool = &adapter->tx_pool[pool];
3625 txbuff = &tx_pool->tx_buff[index];
3627 num_entries += txbuff->num_entries;
3629 total_bytes += txbuff->skb->len;
3630 if (next->tx_comp.rcs[i]) {
3631 dev_err(dev, "tx error %x\n",
3632 next->tx_comp.rcs[i]);
3633 dev_kfree_skb_irq(txbuff->skb);
3635 dev_consume_skb_irq(txbuff->skb);
3639 netdev_warn(adapter->netdev,
3640 "TX completion received with NULL socket buffer\n");
3642 tx_pool->free_map[tx_pool->producer_index] = index;
3643 tx_pool->producer_index =
3644 (tx_pool->producer_index + 1) %
3645 tx_pool->num_buffers;
3647 /* remove tx_comp scrq*/
3648 next->tx_comp.first = 0;
3650 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3651 netdev_tx_completed_queue(txq, num_packets, total_bytes);
3653 if (atomic_sub_return(num_entries, &scrq->used) <=
3654 (adapter->req_tx_entries_per_subcrq / 2) &&
3655 __netif_subqueue_stopped(adapter->netdev,
3656 scrq->pool_index)) {
3657 netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3658 netdev_dbg(adapter->netdev, "Started queue %d\n",
3663 enable_scrq_irq(adapter, scrq);
3665 if (pending_scrq(adapter, scrq)) {
3666 disable_scrq_irq(adapter, scrq);
3673 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3675 struct ibmvnic_sub_crq_queue *scrq = instance;
3676 struct ibmvnic_adapter *adapter = scrq->adapter;
3678 disable_scrq_irq(adapter, scrq);
3679 ibmvnic_complete_tx(adapter, scrq);
3684 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3686 struct ibmvnic_sub_crq_queue *scrq = instance;
3687 struct ibmvnic_adapter *adapter = scrq->adapter;
3689 /* When booting a kdump kernel we can hit pending interrupts
3690 * prior to completing driver initialization.
3692 if (unlikely(adapter->state != VNIC_OPEN))
3695 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3697 if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3698 disable_scrq_irq(adapter, scrq);
3699 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3705 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3707 struct device *dev = &adapter->vdev->dev;
3708 struct ibmvnic_sub_crq_queue *scrq;
3712 for (i = 0; i < adapter->req_tx_queues; i++) {
3713 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3715 scrq = adapter->tx_scrq[i];
3716 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3720 dev_err(dev, "Error mapping irq\n");
3721 goto req_tx_irq_failed;
3724 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3725 adapter->vdev->unit_address, i);
3726 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3727 0, scrq->name, scrq);
3730 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3732 irq_dispose_mapping(scrq->irq);
3733 goto req_tx_irq_failed;
3737 for (i = 0; i < adapter->req_rx_queues; i++) {
3738 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3740 scrq = adapter->rx_scrq[i];
3741 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3744 dev_err(dev, "Error mapping irq\n");
3745 goto req_rx_irq_failed;
3747 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3748 adapter->vdev->unit_address, i);
3749 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3750 0, scrq->name, scrq);
3752 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3754 irq_dispose_mapping(scrq->irq);
3755 goto req_rx_irq_failed;
3761 for (j = 0; j < i; j++) {
3762 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3763 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3765 i = adapter->req_tx_queues;
3767 for (j = 0; j < i; j++) {
3768 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3769 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3771 release_sub_crqs(adapter, 1);
3775 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3777 struct device *dev = &adapter->vdev->dev;
3778 struct ibmvnic_sub_crq_queue **allqueues;
3779 int registered_queues = 0;
3784 total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3786 allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3790 for (i = 0; i < total_queues; i++) {
3791 allqueues[i] = init_sub_crq_queue(adapter);
3792 if (!allqueues[i]) {
3793 dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3796 registered_queues++;
3799 /* Make sure we were able to register the minimum number of queues */
3800 if (registered_queues <
3801 adapter->min_tx_queues + adapter->min_rx_queues) {
3802 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n");
3806 /* Distribute the failed allocated queues*/
3807 for (i = 0; i < total_queues - registered_queues + more ; i++) {
3808 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3811 if (adapter->req_rx_queues > adapter->min_rx_queues)
3812 adapter->req_rx_queues--;
3817 if (adapter->req_tx_queues > adapter->min_tx_queues)
3818 adapter->req_tx_queues--;
3825 adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3826 sizeof(*adapter->tx_scrq), GFP_KERNEL);
3827 if (!adapter->tx_scrq)
3830 for (i = 0; i < adapter->req_tx_queues; i++) {
3831 adapter->tx_scrq[i] = allqueues[i];
3832 adapter->tx_scrq[i]->pool_index = i;
3833 adapter->num_active_tx_scrqs++;
3836 adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3837 sizeof(*adapter->rx_scrq), GFP_KERNEL);
3838 if (!adapter->rx_scrq)
3841 for (i = 0; i < adapter->req_rx_queues; i++) {
3842 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3843 adapter->rx_scrq[i]->scrq_num = i;
3844 adapter->num_active_rx_scrqs++;
3851 kfree(adapter->tx_scrq);
3852 adapter->tx_scrq = NULL;
3854 for (i = 0; i < registered_queues; i++)
3855 release_sub_crq_queue(adapter, allqueues[i], 1);
3860 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3862 struct device *dev = &adapter->vdev->dev;
3863 union ibmvnic_crq crq;
3867 /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
3868 * the PROMISC flag). Initialize this count upfront. When the tasklet
3869 * receives a response to all of these, it will send the next protocol
3870 * message (QUERY_IP_OFFLOAD).
3872 if (!(adapter->netdev->flags & IFF_PROMISC) ||
3873 adapter->promisc_supported)
3879 /* Sub-CRQ entries are 32 byte long */
3880 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3882 atomic_set(&adapter->running_cap_crqs, cap_reqs);
3884 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3885 adapter->min_rx_add_entries_per_subcrq > entries_page) {
3886 dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3890 if (adapter->desired.mtu)
3891 adapter->req_mtu = adapter->desired.mtu;
3893 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3895 if (!adapter->desired.tx_entries)
3896 adapter->desired.tx_entries =
3897 adapter->max_tx_entries_per_subcrq;
3898 if (!adapter->desired.rx_entries)
3899 adapter->desired.rx_entries =
3900 adapter->max_rx_add_entries_per_subcrq;
3902 max_entries = IBMVNIC_MAX_LTB_SIZE /
3903 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3905 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3906 adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3907 adapter->desired.tx_entries = max_entries;
3910 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3911 adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3912 adapter->desired.rx_entries = max_entries;
3915 if (adapter->desired.tx_entries)
3916 adapter->req_tx_entries_per_subcrq =
3917 adapter->desired.tx_entries;
3919 adapter->req_tx_entries_per_subcrq =
3920 adapter->max_tx_entries_per_subcrq;
3922 if (adapter->desired.rx_entries)
3923 adapter->req_rx_add_entries_per_subcrq =
3924 adapter->desired.rx_entries;
3926 adapter->req_rx_add_entries_per_subcrq =
3927 adapter->max_rx_add_entries_per_subcrq;
3929 if (adapter->desired.tx_queues)
3930 adapter->req_tx_queues =
3931 adapter->desired.tx_queues;
3933 adapter->req_tx_queues =
3934 adapter->opt_tx_comp_sub_queues;
3936 if (adapter->desired.rx_queues)
3937 adapter->req_rx_queues =
3938 adapter->desired.rx_queues;
3940 adapter->req_rx_queues =
3941 adapter->opt_rx_comp_queues;
3943 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3945 atomic_add(cap_reqs, &adapter->running_cap_crqs);
3947 memset(&crq, 0, sizeof(crq));
3948 crq.request_capability.first = IBMVNIC_CRQ_CMD;
3949 crq.request_capability.cmd = REQUEST_CAPABILITY;
3951 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3952 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3954 ibmvnic_send_crq(adapter, &crq);
3956 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3957 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3959 ibmvnic_send_crq(adapter, &crq);
3961 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3962 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3964 ibmvnic_send_crq(adapter, &crq);
3966 crq.request_capability.capability =
3967 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3968 crq.request_capability.number =
3969 cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3971 ibmvnic_send_crq(adapter, &crq);
3973 crq.request_capability.capability =
3974 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3975 crq.request_capability.number =
3976 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3978 ibmvnic_send_crq(adapter, &crq);
3980 crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3981 crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3983 ibmvnic_send_crq(adapter, &crq);
3985 if (adapter->netdev->flags & IFF_PROMISC) {
3986 if (adapter->promisc_supported) {
3987 crq.request_capability.capability =
3988 cpu_to_be16(PROMISC_REQUESTED);
3989 crq.request_capability.number = cpu_to_be64(1);
3991 ibmvnic_send_crq(adapter, &crq);
3994 crq.request_capability.capability =
3995 cpu_to_be16(PROMISC_REQUESTED);
3996 crq.request_capability.number = cpu_to_be64(0);
3998 ibmvnic_send_crq(adapter, &crq);
4001 /* Keep at end to catch any discrepancy between expected and actual
4004 WARN_ON(cap_reqs != 0);
4007 static int pending_scrq(struct ibmvnic_adapter *adapter,
4008 struct ibmvnic_sub_crq_queue *scrq)
4010 union sub_crq *entry = &scrq->msgs[scrq->cur];
4013 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
4015 /* Ensure that the SCRQ valid flag is loaded prior to loading the
4016 * contents of the SCRQ descriptor
4023 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
4024 struct ibmvnic_sub_crq_queue *scrq)
4026 union sub_crq *entry;
4027 unsigned long flags;
4029 spin_lock_irqsave(&scrq->lock, flags);
4030 entry = &scrq->msgs[scrq->cur];
4031 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4032 if (++scrq->cur == scrq->size)
4037 spin_unlock_irqrestore(&scrq->lock, flags);
4039 /* Ensure that the SCRQ valid flag is loaded prior to loading the
4040 * contents of the SCRQ descriptor
4047 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4049 struct ibmvnic_crq_queue *queue = &adapter->crq;
4050 union ibmvnic_crq *crq;
4052 crq = &queue->msgs[queue->cur];
4053 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4054 if (++queue->cur == queue->size)
4063 static void print_subcrq_error(struct device *dev, int rc, const char *func)
4067 dev_warn_ratelimited(dev,
4068 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4072 dev_warn_ratelimited(dev,
4073 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4077 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4082 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4083 u64 remote_handle, u64 ioba, u64 num_entries)
4085 unsigned int ua = adapter->vdev->unit_address;
4086 struct device *dev = &adapter->vdev->dev;
4089 /* Make sure the hypervisor sees the complete request */
4091 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4092 cpu_to_be64(remote_handle),
4096 print_subcrq_error(dev, rc, __func__);
4101 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4102 union ibmvnic_crq *crq)
4104 unsigned int ua = adapter->vdev->unit_address;
4105 struct device *dev = &adapter->vdev->dev;
4106 u64 *u64_crq = (u64 *)crq;
4109 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4110 (unsigned long)cpu_to_be64(u64_crq[0]),
4111 (unsigned long)cpu_to_be64(u64_crq[1]));
4113 if (!adapter->crq.active &&
4114 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4115 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4119 /* Make sure the hypervisor sees the complete request */
4122 rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4123 cpu_to_be64(u64_crq[0]),
4124 cpu_to_be64(u64_crq[1]));
4127 if (rc == H_CLOSED) {
4128 dev_warn(dev, "CRQ Queue closed\n");
4129 /* do not reset, report the fail, wait for passive init from server */
4132 dev_warn(dev, "Send error (rc=%d)\n", rc);
4138 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4140 struct device *dev = &adapter->vdev->dev;
4141 union ibmvnic_crq crq;
4145 memset(&crq, 0, sizeof(crq));
4146 crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4147 crq.generic.cmd = IBMVNIC_CRQ_INIT;
4148 netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4151 rc = ibmvnic_send_crq(adapter, &crq);
4157 } while (retries > 0);
4160 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4167 struct vnic_login_client_data {
4173 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4177 /* Calculate the amount of buffer space needed for the
4178 * vnic client data in the login buffer. There are four entries,
4179 * OS name, LPAR name, device name, and a null last entry.
4181 len = 4 * sizeof(struct vnic_login_client_data);
4182 len += 6; /* "Linux" plus NULL */
4183 len += strlen(utsname()->nodename) + 1;
4184 len += strlen(adapter->netdev->name) + 1;
4189 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4190 struct vnic_login_client_data *vlcd)
4192 const char *os_name = "Linux";
4195 /* Type 1 - LPAR OS */
4197 len = strlen(os_name) + 1;
4198 vlcd->len = cpu_to_be16(len);
4199 strscpy(vlcd->name, os_name, len);
4200 vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4202 /* Type 2 - LPAR name */
4204 len = strlen(utsname()->nodename) + 1;
4205 vlcd->len = cpu_to_be16(len);
4206 strscpy(vlcd->name, utsname()->nodename, len);
4207 vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4209 /* Type 3 - device name */
4211 len = strlen(adapter->netdev->name) + 1;
4212 vlcd->len = cpu_to_be16(len);
4213 strscpy(vlcd->name, adapter->netdev->name, len);
4216 static int send_login(struct ibmvnic_adapter *adapter)
4218 struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4219 struct ibmvnic_login_buffer *login_buffer;
4220 struct device *dev = &adapter->vdev->dev;
4221 struct vnic_login_client_data *vlcd;
4222 dma_addr_t rsp_buffer_token;
4223 dma_addr_t buffer_token;
4224 size_t rsp_buffer_size;
4225 union ibmvnic_crq crq;
4226 int client_data_len;
4233 if (!adapter->tx_scrq || !adapter->rx_scrq) {
4234 netdev_err(adapter->netdev,
4235 "RX or TX queues are not allocated, device login failed\n");
4239 release_login_buffer(adapter);
4240 release_login_rsp_buffer(adapter);
4242 client_data_len = vnic_client_data_len(adapter);
4245 sizeof(struct ibmvnic_login_buffer) +
4246 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4249 login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4251 goto buf_alloc_failed;
4253 buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4255 if (dma_mapping_error(dev, buffer_token)) {
4256 dev_err(dev, "Couldn't map login buffer\n");
4257 goto buf_map_failed;
4260 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4261 sizeof(u64) * adapter->req_tx_queues +
4262 sizeof(u64) * adapter->req_rx_queues +
4263 sizeof(u64) * adapter->req_rx_queues +
4264 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4266 login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4267 if (!login_rsp_buffer)
4268 goto buf_rsp_alloc_failed;
4270 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4271 rsp_buffer_size, DMA_FROM_DEVICE);
4272 if (dma_mapping_error(dev, rsp_buffer_token)) {
4273 dev_err(dev, "Couldn't map login rsp buffer\n");
4274 goto buf_rsp_map_failed;
4277 adapter->login_buf = login_buffer;
4278 adapter->login_buf_token = buffer_token;
4279 adapter->login_buf_sz = buffer_size;
4280 adapter->login_rsp_buf = login_rsp_buffer;
4281 adapter->login_rsp_buf_token = rsp_buffer_token;
4282 adapter->login_rsp_buf_sz = rsp_buffer_size;
4284 login_buffer->len = cpu_to_be32(buffer_size);
4285 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4286 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4287 login_buffer->off_txcomp_subcrqs =
4288 cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4289 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4290 login_buffer->off_rxcomp_subcrqs =
4291 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4292 sizeof(u64) * adapter->req_tx_queues);
4293 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4294 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4296 tx_list_p = (__be64 *)((char *)login_buffer +
4297 sizeof(struct ibmvnic_login_buffer));
4298 rx_list_p = (__be64 *)((char *)login_buffer +
4299 sizeof(struct ibmvnic_login_buffer) +
4300 sizeof(u64) * adapter->req_tx_queues);
4302 for (i = 0; i < adapter->req_tx_queues; i++) {
4303 if (adapter->tx_scrq[i]) {
4305 cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4309 for (i = 0; i < adapter->req_rx_queues; i++) {
4310 if (adapter->rx_scrq[i]) {
4312 cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4316 /* Insert vNIC login client data */
4317 vlcd = (struct vnic_login_client_data *)
4318 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4319 login_buffer->client_data_offset =
4320 cpu_to_be32((char *)vlcd - (char *)login_buffer);
4321 login_buffer->client_data_len = cpu_to_be32(client_data_len);
4323 vnic_add_client_data(adapter, vlcd);
4325 netdev_dbg(adapter->netdev, "Login Buffer:\n");
4326 for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4327 netdev_dbg(adapter->netdev, "%016lx\n",
4328 ((unsigned long *)(adapter->login_buf))[i]);
4331 memset(&crq, 0, sizeof(crq));
4332 crq.login.first = IBMVNIC_CRQ_CMD;
4333 crq.login.cmd = LOGIN;
4334 crq.login.ioba = cpu_to_be32(buffer_token);
4335 crq.login.len = cpu_to_be32(buffer_size);
4337 adapter->login_pending = true;
4338 rc = ibmvnic_send_crq(adapter, &crq);
4340 adapter->login_pending = false;
4341 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4342 goto buf_rsp_map_failed;
4348 kfree(login_rsp_buffer);
4349 adapter->login_rsp_buf = NULL;
4350 buf_rsp_alloc_failed:
4351 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4353 kfree(login_buffer);
4354 adapter->login_buf = NULL;
4359 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4362 union ibmvnic_crq crq;
4364 memset(&crq, 0, sizeof(crq));
4365 crq.request_map.first = IBMVNIC_CRQ_CMD;
4366 crq.request_map.cmd = REQUEST_MAP;
4367 crq.request_map.map_id = map_id;
4368 crq.request_map.ioba = cpu_to_be32(addr);
4369 crq.request_map.len = cpu_to_be32(len);
4370 return ibmvnic_send_crq(adapter, &crq);
4373 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4375 union ibmvnic_crq crq;
4377 memset(&crq, 0, sizeof(crq));
4378 crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4379 crq.request_unmap.cmd = REQUEST_UNMAP;
4380 crq.request_unmap.map_id = map_id;
4381 return ibmvnic_send_crq(adapter, &crq);
4384 static void send_query_map(struct ibmvnic_adapter *adapter)
4386 union ibmvnic_crq crq;
4388 memset(&crq, 0, sizeof(crq));
4389 crq.query_map.first = IBMVNIC_CRQ_CMD;
4390 crq.query_map.cmd = QUERY_MAP;
4391 ibmvnic_send_crq(adapter, &crq);
4394 /* Send a series of CRQs requesting various capabilities of the VNIC server */
4395 static void send_query_cap(struct ibmvnic_adapter *adapter)
4397 union ibmvnic_crq crq;
4400 /* We send out 25 QUERY_CAPABILITY CRQs below. Initialize this count
4401 * upfront. When the tasklet receives a response to all of these, it
4402 * can send out the next protocol messaage (REQUEST_CAPABILITY).
4406 atomic_set(&adapter->running_cap_crqs, cap_reqs);
4408 memset(&crq, 0, sizeof(crq));
4409 crq.query_capability.first = IBMVNIC_CRQ_CMD;
4410 crq.query_capability.cmd = QUERY_CAPABILITY;
4412 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
4413 ibmvnic_send_crq(adapter, &crq);
4416 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
4417 ibmvnic_send_crq(adapter, &crq);
4420 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
4421 ibmvnic_send_crq(adapter, &crq);
4424 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
4425 ibmvnic_send_crq(adapter, &crq);
4428 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4429 ibmvnic_send_crq(adapter, &crq);
4432 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4433 ibmvnic_send_crq(adapter, &crq);
4436 crq.query_capability.capability =
4437 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4438 ibmvnic_send_crq(adapter, &crq);
4441 crq.query_capability.capability =
4442 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4443 ibmvnic_send_crq(adapter, &crq);
4446 crq.query_capability.capability =
4447 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4448 ibmvnic_send_crq(adapter, &crq);
4451 crq.query_capability.capability =
4452 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4453 ibmvnic_send_crq(adapter, &crq);
4456 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4457 ibmvnic_send_crq(adapter, &crq);
4460 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4461 ibmvnic_send_crq(adapter, &crq);
4464 crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4465 ibmvnic_send_crq(adapter, &crq);
4468 crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4469 ibmvnic_send_crq(adapter, &crq);
4472 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4473 ibmvnic_send_crq(adapter, &crq);
4476 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4477 ibmvnic_send_crq(adapter, &crq);
4480 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4481 ibmvnic_send_crq(adapter, &crq);
4484 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4485 ibmvnic_send_crq(adapter, &crq);
4488 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4489 ibmvnic_send_crq(adapter, &crq);
4492 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4493 ibmvnic_send_crq(adapter, &crq);
4496 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4497 ibmvnic_send_crq(adapter, &crq);
4500 crq.query_capability.capability =
4501 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4502 ibmvnic_send_crq(adapter, &crq);
4505 crq.query_capability.capability =
4506 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4507 ibmvnic_send_crq(adapter, &crq);
4510 crq.query_capability.capability =
4511 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4512 ibmvnic_send_crq(adapter, &crq);
4515 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4517 ibmvnic_send_crq(adapter, &crq);
4520 /* Keep at end to catch any discrepancy between expected and actual
4523 WARN_ON(cap_reqs != 0);
4526 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4528 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4529 struct device *dev = &adapter->vdev->dev;
4530 union ibmvnic_crq crq;
4532 adapter->ip_offload_tok =
4534 &adapter->ip_offload_buf,
4538 if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4539 if (!firmware_has_feature(FW_FEATURE_CMO))
4540 dev_err(dev, "Couldn't map offload buffer\n");
4544 memset(&crq, 0, sizeof(crq));
4545 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4546 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4547 crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4548 crq.query_ip_offload.ioba =
4549 cpu_to_be32(adapter->ip_offload_tok);
4551 ibmvnic_send_crq(adapter, &crq);
4554 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4556 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4557 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4558 struct device *dev = &adapter->vdev->dev;
4559 netdev_features_t old_hw_features = 0;
4560 union ibmvnic_crq crq;
4562 adapter->ip_offload_ctrl_tok =
4565 sizeof(adapter->ip_offload_ctrl),
4568 if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4569 dev_err(dev, "Couldn't map ip offload control buffer\n");
4573 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4574 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4575 ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4576 ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4577 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4578 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4579 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4580 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4581 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4582 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4584 /* large_rx disabled for now, additional features needed */
4585 ctrl_buf->large_rx_ipv4 = 0;
4586 ctrl_buf->large_rx_ipv6 = 0;
4588 if (adapter->state != VNIC_PROBING) {
4589 old_hw_features = adapter->netdev->hw_features;
4590 adapter->netdev->hw_features = 0;
4593 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4595 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4596 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4598 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4599 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4601 if ((adapter->netdev->features &
4602 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4603 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4605 if (buf->large_tx_ipv4)
4606 adapter->netdev->hw_features |= NETIF_F_TSO;
4607 if (buf->large_tx_ipv6)
4608 adapter->netdev->hw_features |= NETIF_F_TSO6;
4610 if (adapter->state == VNIC_PROBING) {
4611 adapter->netdev->features |= adapter->netdev->hw_features;
4612 } else if (old_hw_features != adapter->netdev->hw_features) {
4613 netdev_features_t tmp = 0;
4615 /* disable features no longer supported */
4616 adapter->netdev->features &= adapter->netdev->hw_features;
4617 /* turn on features now supported if previously enabled */
4618 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4619 adapter->netdev->hw_features;
4620 adapter->netdev->features |=
4621 tmp & adapter->netdev->wanted_features;
4624 memset(&crq, 0, sizeof(crq));
4625 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4626 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4627 crq.control_ip_offload.len =
4628 cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4629 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4630 ibmvnic_send_crq(adapter, &crq);
4633 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4634 struct ibmvnic_adapter *adapter)
4636 struct device *dev = &adapter->vdev->dev;
4638 if (crq->get_vpd_size_rsp.rc.code) {
4639 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4640 crq->get_vpd_size_rsp.rc.code);
4641 complete(&adapter->fw_done);
4645 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4646 complete(&adapter->fw_done);
4649 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4650 struct ibmvnic_adapter *adapter)
4652 struct device *dev = &adapter->vdev->dev;
4653 unsigned char *substr = NULL;
4654 u8 fw_level_len = 0;
4656 memset(adapter->fw_version, 0, 32);
4658 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4661 if (crq->get_vpd_rsp.rc.code) {
4662 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4663 crq->get_vpd_rsp.rc.code);
4667 /* get the position of the firmware version info
4668 * located after the ASCII 'RM' substring in the buffer
4670 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4672 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4676 /* get length of firmware level ASCII substring */
4677 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4678 fw_level_len = *(substr + 2);
4680 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4684 /* copy firmware version string from vpd into adapter */
4685 if ((substr + 3 + fw_level_len) <
4686 (adapter->vpd->buff + adapter->vpd->len)) {
4687 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4689 dev_info(dev, "FW substr extrapolated VPD buff\n");
4693 if (adapter->fw_version[0] == '\0')
4694 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
4695 complete(&adapter->fw_done);
4698 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4700 struct device *dev = &adapter->vdev->dev;
4701 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4704 dma_unmap_single(dev, adapter->ip_offload_tok,
4705 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4707 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4708 for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4709 netdev_dbg(adapter->netdev, "%016lx\n",
4710 ((unsigned long *)(buf))[i]);
4712 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4713 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4714 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4715 buf->tcp_ipv4_chksum);
4716 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4717 buf->tcp_ipv6_chksum);
4718 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4719 buf->udp_ipv4_chksum);
4720 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4721 buf->udp_ipv6_chksum);
4722 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4723 buf->large_tx_ipv4);
4724 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4725 buf->large_tx_ipv6);
4726 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4727 buf->large_rx_ipv4);
4728 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4729 buf->large_rx_ipv6);
4730 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4731 buf->max_ipv4_header_size);
4732 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4733 buf->max_ipv6_header_size);
4734 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4735 buf->max_tcp_header_size);
4736 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4737 buf->max_udp_header_size);
4738 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4739 buf->max_large_tx_size);
4740 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4741 buf->max_large_rx_size);
4742 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4743 buf->ipv6_extension_header);
4744 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4745 buf->tcp_pseudosum_req);
4746 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4747 buf->num_ipv6_ext_headers);
4748 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4749 buf->off_ipv6_ext_headers);
4751 send_control_ip_offload(adapter);
4754 static const char *ibmvnic_fw_err_cause(u16 cause)
4757 case ADAPTER_PROBLEM:
4758 return "adapter problem";
4760 return "bus problem";
4762 return "firmware problem";
4764 return "device driver problem";
4766 return "EEH recovery";
4768 return "firmware updated";
4770 return "low Memory";
4776 static void handle_error_indication(union ibmvnic_crq *crq,
4777 struct ibmvnic_adapter *adapter)
4779 struct device *dev = &adapter->vdev->dev;
4782 cause = be16_to_cpu(crq->error_indication.error_cause);
4784 dev_warn_ratelimited(dev,
4785 "Firmware reports %serror, cause: %s. Starting recovery...\n",
4786 crq->error_indication.flags
4787 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4788 ibmvnic_fw_err_cause(cause));
4790 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4791 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4793 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4796 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4797 struct ibmvnic_adapter *adapter)
4799 struct net_device *netdev = adapter->netdev;
4800 struct device *dev = &adapter->vdev->dev;
4803 rc = crq->change_mac_addr_rsp.rc.code;
4805 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4808 /* crq->change_mac_addr.mac_addr is the requested one
4809 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4811 eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]);
4812 ether_addr_copy(adapter->mac_addr,
4813 &crq->change_mac_addr_rsp.mac_addr[0]);
4815 complete(&adapter->fw_done);
4819 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4820 struct ibmvnic_adapter *adapter)
4822 struct device *dev = &adapter->vdev->dev;
4826 atomic_dec(&adapter->running_cap_crqs);
4827 netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
4828 atomic_read(&adapter->running_cap_crqs));
4829 switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4831 req_value = &adapter->req_tx_queues;
4835 req_value = &adapter->req_rx_queues;
4838 case REQ_RX_ADD_QUEUES:
4839 req_value = &adapter->req_rx_add_queues;
4842 case REQ_TX_ENTRIES_PER_SUBCRQ:
4843 req_value = &adapter->req_tx_entries_per_subcrq;
4844 name = "tx_entries_per_subcrq";
4846 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4847 req_value = &adapter->req_rx_add_entries_per_subcrq;
4848 name = "rx_add_entries_per_subcrq";
4851 req_value = &adapter->req_mtu;
4854 case PROMISC_REQUESTED:
4855 req_value = &adapter->promisc;
4859 dev_err(dev, "Got invalid cap request rsp %d\n",
4860 crq->request_capability.capability);
4864 switch (crq->request_capability_rsp.rc.code) {
4867 case PARTIALSUCCESS:
4868 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4870 (long)be64_to_cpu(crq->request_capability_rsp.number),
4873 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4875 pr_err("mtu of %llu is not supported. Reverting.\n",
4877 *req_value = adapter->fallback.mtu;
4880 be64_to_cpu(crq->request_capability_rsp.number);
4883 send_request_cap(adapter, 1);
4886 dev_err(dev, "Error %d in request cap rsp\n",
4887 crq->request_capability_rsp.rc.code);
4891 /* Done receiving requested capabilities, query IP offload support */
4892 if (atomic_read(&adapter->running_cap_crqs) == 0)
4893 send_query_ip_offload(adapter);
4896 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4897 struct ibmvnic_adapter *adapter)
4899 struct device *dev = &adapter->vdev->dev;
4900 struct net_device *netdev = adapter->netdev;
4901 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4902 struct ibmvnic_login_buffer *login = adapter->login_buf;
4903 u64 *tx_handle_array;
4904 u64 *rx_handle_array;
4910 /* CHECK: Test/set of login_pending does not need to be atomic
4911 * because only ibmvnic_tasklet tests/clears this.
4913 if (!adapter->login_pending) {
4914 netdev_warn(netdev, "Ignoring unexpected login response\n");
4917 adapter->login_pending = false;
4919 dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4921 dma_unmap_single(dev, adapter->login_rsp_buf_token,
4922 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4924 /* If the number of queues requested can't be allocated by the
4925 * server, the login response will return with code 1. We will need
4926 * to resend the login buffer with fewer queues requested.
4928 if (login_rsp_crq->generic.rc.code) {
4929 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4930 complete(&adapter->init_done);
4934 if (adapter->failover_pending) {
4935 adapter->init_done_rc = -EAGAIN;
4936 netdev_dbg(netdev, "Failover pending, ignoring login response\n");
4937 complete(&adapter->init_done);
4938 /* login response buffer will be released on reset */
4942 netdev->mtu = adapter->req_mtu - ETH_HLEN;
4944 netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4945 for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4946 netdev_dbg(adapter->netdev, "%016lx\n",
4947 ((unsigned long *)(adapter->login_rsp_buf))[i]);
4951 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4952 (be32_to_cpu(login->num_rxcomp_subcrqs) *
4953 adapter->req_rx_add_queues !=
4954 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4955 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4956 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4959 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4960 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4961 /* variable buffer sizes are not supported, so just read the
4964 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4966 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4967 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4969 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4970 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4971 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4972 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4974 for (i = 0; i < num_tx_pools; i++)
4975 adapter->tx_scrq[i]->handle = tx_handle_array[i];
4977 for (i = 0; i < num_rx_pools; i++)
4978 adapter->rx_scrq[i]->handle = rx_handle_array[i];
4980 adapter->num_active_tx_scrqs = num_tx_pools;
4981 adapter->num_active_rx_scrqs = num_rx_pools;
4982 release_login_rsp_buffer(adapter);
4983 release_login_buffer(adapter);
4984 complete(&adapter->init_done);
4989 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4990 struct ibmvnic_adapter *adapter)
4992 struct device *dev = &adapter->vdev->dev;
4995 rc = crq->request_unmap_rsp.rc.code;
4997 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
5000 static void handle_query_map_rsp(union ibmvnic_crq *crq,
5001 struct ibmvnic_adapter *adapter)
5003 struct net_device *netdev = adapter->netdev;
5004 struct device *dev = &adapter->vdev->dev;
5007 rc = crq->query_map_rsp.rc.code;
5009 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
5012 netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
5013 crq->query_map_rsp.page_size,
5014 __be32_to_cpu(crq->query_map_rsp.tot_pages),
5015 __be32_to_cpu(crq->query_map_rsp.free_pages));
5018 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
5019 struct ibmvnic_adapter *adapter)
5021 struct net_device *netdev = adapter->netdev;
5022 struct device *dev = &adapter->vdev->dev;
5025 atomic_dec(&adapter->running_cap_crqs);
5026 netdev_dbg(netdev, "Outstanding queries: %d\n",
5027 atomic_read(&adapter->running_cap_crqs));
5028 rc = crq->query_capability.rc.code;
5030 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
5034 switch (be16_to_cpu(crq->query_capability.capability)) {
5036 adapter->min_tx_queues =
5037 be64_to_cpu(crq->query_capability.number);
5038 netdev_dbg(netdev, "min_tx_queues = %lld\n",
5039 adapter->min_tx_queues);
5042 adapter->min_rx_queues =
5043 be64_to_cpu(crq->query_capability.number);
5044 netdev_dbg(netdev, "min_rx_queues = %lld\n",
5045 adapter->min_rx_queues);
5047 case MIN_RX_ADD_QUEUES:
5048 adapter->min_rx_add_queues =
5049 be64_to_cpu(crq->query_capability.number);
5050 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
5051 adapter->min_rx_add_queues);
5054 adapter->max_tx_queues =
5055 be64_to_cpu(crq->query_capability.number);
5056 netdev_dbg(netdev, "max_tx_queues = %lld\n",
5057 adapter->max_tx_queues);
5060 adapter->max_rx_queues =
5061 be64_to_cpu(crq->query_capability.number);
5062 netdev_dbg(netdev, "max_rx_queues = %lld\n",
5063 adapter->max_rx_queues);
5065 case MAX_RX_ADD_QUEUES:
5066 adapter->max_rx_add_queues =
5067 be64_to_cpu(crq->query_capability.number);
5068 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5069 adapter->max_rx_add_queues);
5071 case MIN_TX_ENTRIES_PER_SUBCRQ:
5072 adapter->min_tx_entries_per_subcrq =
5073 be64_to_cpu(crq->query_capability.number);
5074 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5075 adapter->min_tx_entries_per_subcrq);
5077 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5078 adapter->min_rx_add_entries_per_subcrq =
5079 be64_to_cpu(crq->query_capability.number);
5080 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5081 adapter->min_rx_add_entries_per_subcrq);
5083 case MAX_TX_ENTRIES_PER_SUBCRQ:
5084 adapter->max_tx_entries_per_subcrq =
5085 be64_to_cpu(crq->query_capability.number);
5086 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5087 adapter->max_tx_entries_per_subcrq);
5089 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5090 adapter->max_rx_add_entries_per_subcrq =
5091 be64_to_cpu(crq->query_capability.number);
5092 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5093 adapter->max_rx_add_entries_per_subcrq);
5095 case TCP_IP_OFFLOAD:
5096 adapter->tcp_ip_offload =
5097 be64_to_cpu(crq->query_capability.number);
5098 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5099 adapter->tcp_ip_offload);
5101 case PROMISC_SUPPORTED:
5102 adapter->promisc_supported =
5103 be64_to_cpu(crq->query_capability.number);
5104 netdev_dbg(netdev, "promisc_supported = %lld\n",
5105 adapter->promisc_supported);
5108 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5109 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5110 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5113 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5114 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5115 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5117 case MAX_MULTICAST_FILTERS:
5118 adapter->max_multicast_filters =
5119 be64_to_cpu(crq->query_capability.number);
5120 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5121 adapter->max_multicast_filters);
5123 case VLAN_HEADER_INSERTION:
5124 adapter->vlan_header_insertion =
5125 be64_to_cpu(crq->query_capability.number);
5126 if (adapter->vlan_header_insertion)
5127 netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5128 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5129 adapter->vlan_header_insertion);
5131 case RX_VLAN_HEADER_INSERTION:
5132 adapter->rx_vlan_header_insertion =
5133 be64_to_cpu(crq->query_capability.number);
5134 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5135 adapter->rx_vlan_header_insertion);
5137 case MAX_TX_SG_ENTRIES:
5138 adapter->max_tx_sg_entries =
5139 be64_to_cpu(crq->query_capability.number);
5140 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5141 adapter->max_tx_sg_entries);
5143 case RX_SG_SUPPORTED:
5144 adapter->rx_sg_supported =
5145 be64_to_cpu(crq->query_capability.number);
5146 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5147 adapter->rx_sg_supported);
5149 case OPT_TX_COMP_SUB_QUEUES:
5150 adapter->opt_tx_comp_sub_queues =
5151 be64_to_cpu(crq->query_capability.number);
5152 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5153 adapter->opt_tx_comp_sub_queues);
5155 case OPT_RX_COMP_QUEUES:
5156 adapter->opt_rx_comp_queues =
5157 be64_to_cpu(crq->query_capability.number);
5158 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5159 adapter->opt_rx_comp_queues);
5161 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5162 adapter->opt_rx_bufadd_q_per_rx_comp_q =
5163 be64_to_cpu(crq->query_capability.number);
5164 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5165 adapter->opt_rx_bufadd_q_per_rx_comp_q);
5167 case OPT_TX_ENTRIES_PER_SUBCRQ:
5168 adapter->opt_tx_entries_per_subcrq =
5169 be64_to_cpu(crq->query_capability.number);
5170 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5171 adapter->opt_tx_entries_per_subcrq);
5173 case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5174 adapter->opt_rxba_entries_per_subcrq =
5175 be64_to_cpu(crq->query_capability.number);
5176 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5177 adapter->opt_rxba_entries_per_subcrq);
5179 case TX_RX_DESC_REQ:
5180 adapter->tx_rx_desc_req = crq->query_capability.number;
5181 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5182 adapter->tx_rx_desc_req);
5186 netdev_err(netdev, "Got invalid cap rsp %d\n",
5187 crq->query_capability.capability);
5191 if (atomic_read(&adapter->running_cap_crqs) == 0)
5192 send_request_cap(adapter, 0);
5195 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5197 union ibmvnic_crq crq;
5200 memset(&crq, 0, sizeof(crq));
5201 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5202 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5204 mutex_lock(&adapter->fw_lock);
5205 adapter->fw_done_rc = 0;
5206 reinit_completion(&adapter->fw_done);
5208 rc = ibmvnic_send_crq(adapter, &crq);
5210 mutex_unlock(&adapter->fw_lock);
5214 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
5216 mutex_unlock(&adapter->fw_lock);
5220 mutex_unlock(&adapter->fw_lock);
5221 return adapter->fw_done_rc ? -EIO : 0;
5224 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5225 struct ibmvnic_adapter *adapter)
5227 struct net_device *netdev = adapter->netdev;
5229 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5231 rc = crq->query_phys_parms_rsp.rc.code;
5233 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5237 case IBMVNIC_10MBPS:
5238 adapter->speed = SPEED_10;
5240 case IBMVNIC_100MBPS:
5241 adapter->speed = SPEED_100;
5244 adapter->speed = SPEED_1000;
5246 case IBMVNIC_10GBPS:
5247 adapter->speed = SPEED_10000;
5249 case IBMVNIC_25GBPS:
5250 adapter->speed = SPEED_25000;
5252 case IBMVNIC_40GBPS:
5253 adapter->speed = SPEED_40000;
5255 case IBMVNIC_50GBPS:
5256 adapter->speed = SPEED_50000;
5258 case IBMVNIC_100GBPS:
5259 adapter->speed = SPEED_100000;
5261 case IBMVNIC_200GBPS:
5262 adapter->speed = SPEED_200000;
5265 if (netif_carrier_ok(netdev))
5266 netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5267 adapter->speed = SPEED_UNKNOWN;
5269 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5270 adapter->duplex = DUPLEX_FULL;
5271 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5272 adapter->duplex = DUPLEX_HALF;
5274 adapter->duplex = DUPLEX_UNKNOWN;
5279 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5280 struct ibmvnic_adapter *adapter)
5282 struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5283 struct net_device *netdev = adapter->netdev;
5284 struct device *dev = &adapter->vdev->dev;
5285 u64 *u64_crq = (u64 *)crq;
5288 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5289 (unsigned long)cpu_to_be64(u64_crq[0]),
5290 (unsigned long)cpu_to_be64(u64_crq[1]));
5291 switch (gen_crq->first) {
5292 case IBMVNIC_CRQ_INIT_RSP:
5293 switch (gen_crq->cmd) {
5294 case IBMVNIC_CRQ_INIT:
5295 dev_info(dev, "Partner initialized\n");
5296 adapter->from_passive_init = true;
5297 /* Discard any stale login responses from prev reset.
5298 * CHECK: should we clear even on INIT_COMPLETE?
5300 adapter->login_pending = false;
5302 if (adapter->state == VNIC_DOWN)
5303 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5305 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5307 if (rc && rc != -EBUSY) {
5308 /* We were unable to schedule the failover
5309 * reset either because the adapter was still
5310 * probing (eg: during kexec) or we could not
5311 * allocate memory. Clear the failover_pending
5312 * flag since no one else will. We ignore
5313 * EBUSY because it means either FAILOVER reset
5314 * is already scheduled or the adapter is
5318 "Error %ld scheduling failover reset\n",
5320 adapter->failover_pending = false;
5323 if (!completion_done(&adapter->init_done)) {
5324 complete(&adapter->init_done);
5325 if (!adapter->init_done_rc)
5326 adapter->init_done_rc = -EAGAIN;
5330 case IBMVNIC_CRQ_INIT_COMPLETE:
5331 dev_info(dev, "Partner initialization complete\n");
5332 adapter->crq.active = true;
5333 send_version_xchg(adapter);
5336 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5339 case IBMVNIC_CRQ_XPORT_EVENT:
5340 netif_carrier_off(netdev);
5341 adapter->crq.active = false;
5342 /* terminate any thread waiting for a response
5345 if (!completion_done(&adapter->fw_done)) {
5346 adapter->fw_done_rc = -EIO;
5347 complete(&adapter->fw_done);
5349 if (!completion_done(&adapter->stats_done))
5350 complete(&adapter->stats_done);
5351 if (test_bit(0, &adapter->resetting))
5352 adapter->force_reset_recovery = true;
5353 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
5354 dev_info(dev, "Migrated, re-enabling adapter\n");
5355 ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
5356 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
5357 dev_info(dev, "Backing device failover detected\n");
5358 adapter->failover_pending = true;
5360 /* The adapter lost the connection */
5361 dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
5363 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5366 case IBMVNIC_CRQ_CMD_RSP:
5369 dev_err(dev, "Got an invalid msg type 0x%02x\n",
5374 switch (gen_crq->cmd) {
5375 case VERSION_EXCHANGE_RSP:
5376 rc = crq->version_exchange_rsp.rc.code;
5378 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
5382 be16_to_cpu(crq->version_exchange_rsp.version);
5383 dev_info(dev, "Partner protocol version is %d\n",
5385 send_query_cap(adapter);
5387 case QUERY_CAPABILITY_RSP:
5388 handle_query_cap_rsp(crq, adapter);
5391 handle_query_map_rsp(crq, adapter);
5393 case REQUEST_MAP_RSP:
5394 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
5395 complete(&adapter->fw_done);
5397 case REQUEST_UNMAP_RSP:
5398 handle_request_unmap_rsp(crq, adapter);
5400 case REQUEST_CAPABILITY_RSP:
5401 handle_request_cap_rsp(crq, adapter);
5404 netdev_dbg(netdev, "Got Login Response\n");
5405 handle_login_rsp(crq, adapter);
5407 case LOGICAL_LINK_STATE_RSP:
5409 "Got Logical Link State Response, state: %d rc: %d\n",
5410 crq->logical_link_state_rsp.link_state,
5411 crq->logical_link_state_rsp.rc.code);
5412 adapter->logical_link_state =
5413 crq->logical_link_state_rsp.link_state;
5414 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
5415 complete(&adapter->init_done);
5417 case LINK_STATE_INDICATION:
5418 netdev_dbg(netdev, "Got Logical Link State Indication\n");
5419 adapter->phys_link_state =
5420 crq->link_state_indication.phys_link_state;
5421 adapter->logical_link_state =
5422 crq->link_state_indication.logical_link_state;
5423 if (adapter->phys_link_state && adapter->logical_link_state)
5424 netif_carrier_on(netdev);
5426 netif_carrier_off(netdev);
5428 case CHANGE_MAC_ADDR_RSP:
5429 netdev_dbg(netdev, "Got MAC address change Response\n");
5430 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
5432 case ERROR_INDICATION:
5433 netdev_dbg(netdev, "Got Error Indication\n");
5434 handle_error_indication(crq, adapter);
5436 case REQUEST_STATISTICS_RSP:
5437 netdev_dbg(netdev, "Got Statistics Response\n");
5438 complete(&adapter->stats_done);
5440 case QUERY_IP_OFFLOAD_RSP:
5441 netdev_dbg(netdev, "Got Query IP offload Response\n");
5442 handle_query_ip_offload_rsp(adapter);
5444 case MULTICAST_CTRL_RSP:
5445 netdev_dbg(netdev, "Got multicast control Response\n");
5447 case CONTROL_IP_OFFLOAD_RSP:
5448 netdev_dbg(netdev, "Got Control IP offload Response\n");
5449 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5450 sizeof(adapter->ip_offload_ctrl),
5452 complete(&adapter->init_done);
5454 case COLLECT_FW_TRACE_RSP:
5455 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5456 complete(&adapter->fw_done);
5458 case GET_VPD_SIZE_RSP:
5459 handle_vpd_size_rsp(crq, adapter);
5462 handle_vpd_rsp(crq, adapter);
5464 case QUERY_PHYS_PARMS_RSP:
5465 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5466 complete(&adapter->fw_done);
5469 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5474 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5476 struct ibmvnic_adapter *adapter = instance;
5478 tasklet_schedule(&adapter->tasklet);
5482 static void ibmvnic_tasklet(struct tasklet_struct *t)
5484 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5485 struct ibmvnic_crq_queue *queue = &adapter->crq;
5486 union ibmvnic_crq *crq;
5487 unsigned long flags;
5489 spin_lock_irqsave(&queue->lock, flags);
5491 /* Pull all the valid messages off the CRQ */
5492 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5493 /* This barrier makes sure ibmvnic_next_crq()'s
5494 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5495 * before ibmvnic_handle_crq()'s
5496 * switch(gen_crq->first) and switch(gen_crq->cmd).
5499 ibmvnic_handle_crq(crq, adapter);
5500 crq->generic.first = 0;
5503 spin_unlock_irqrestore(&queue->lock, flags);
5506 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5508 struct vio_dev *vdev = adapter->vdev;
5512 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5513 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5516 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5521 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5523 struct ibmvnic_crq_queue *crq = &adapter->crq;
5524 struct device *dev = &adapter->vdev->dev;
5525 struct vio_dev *vdev = adapter->vdev;
5530 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5531 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5533 /* Clean out the queue */
5537 memset(crq->msgs, 0, PAGE_SIZE);
5539 crq->active = false;
5541 /* And re-open it again */
5542 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5543 crq->msg_token, PAGE_SIZE);
5546 /* Adapter is good, but other end is not ready */
5547 dev_warn(dev, "Partner adapter not ready\n");
5549 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5554 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5556 struct ibmvnic_crq_queue *crq = &adapter->crq;
5557 struct vio_dev *vdev = adapter->vdev;
5563 netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5564 free_irq(vdev->irq, adapter);
5565 tasklet_kill(&adapter->tasklet);
5567 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5568 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5570 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5572 free_page((unsigned long)crq->msgs);
5574 crq->active = false;
5577 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5579 struct ibmvnic_crq_queue *crq = &adapter->crq;
5580 struct device *dev = &adapter->vdev->dev;
5581 struct vio_dev *vdev = adapter->vdev;
5582 int rc, retrc = -ENOMEM;
5587 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5588 /* Should we allocate more than one page? */
5593 crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5594 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5596 if (dma_mapping_error(dev, crq->msg_token))
5599 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5600 crq->msg_token, PAGE_SIZE);
5602 if (rc == H_RESOURCE)
5603 /* maybe kexecing and resource is busy. try a reset */
5604 rc = ibmvnic_reset_crq(adapter);
5607 if (rc == H_CLOSED) {
5608 dev_warn(dev, "Partner adapter not ready\n");
5610 dev_warn(dev, "Error %d opening adapter\n", rc);
5611 goto reg_crq_failed;
5616 tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5618 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5619 snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5620 adapter->vdev->unit_address);
5621 rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5623 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5625 goto req_irq_failed;
5628 rc = vio_enable_interrupts(vdev);
5630 dev_err(dev, "Error %d enabling interrupts\n", rc);
5631 goto req_irq_failed;
5635 spin_lock_init(&crq->lock);
5637 /* process any CRQs that were queued before we enabled interrupts */
5638 tasklet_schedule(&adapter->tasklet);
5643 tasklet_kill(&adapter->tasklet);
5645 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5646 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5648 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5650 free_page((unsigned long)crq->msgs);
5655 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5657 struct device *dev = &adapter->vdev->dev;
5658 unsigned long timeout = msecs_to_jiffies(20000);
5659 u64 old_num_rx_queues = adapter->req_rx_queues;
5660 u64 old_num_tx_queues = adapter->req_tx_queues;
5663 adapter->from_passive_init = false;
5666 reinit_completion(&adapter->init_done);
5668 adapter->init_done_rc = 0;
5669 rc = ibmvnic_send_crq_init(adapter);
5671 dev_err(dev, "Send crq init failed with error %d\n", rc);
5675 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5676 dev_err(dev, "Initialization sequence timed out\n");
5680 if (adapter->init_done_rc) {
5681 release_crq_queue(adapter);
5682 return adapter->init_done_rc;
5685 if (adapter->from_passive_init) {
5686 adapter->state = VNIC_OPEN;
5687 adapter->from_passive_init = false;
5692 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5693 adapter->reset_reason != VNIC_RESET_MOBILITY) {
5694 if (adapter->req_rx_queues != old_num_rx_queues ||
5695 adapter->req_tx_queues != old_num_tx_queues) {
5696 release_sub_crqs(adapter, 0);
5697 rc = init_sub_crqs(adapter);
5699 rc = reset_sub_crq_queues(adapter);
5702 rc = init_sub_crqs(adapter);
5706 dev_err(dev, "Initialization of sub crqs failed\n");
5707 release_crq_queue(adapter);
5711 rc = init_sub_crq_irqs(adapter);
5713 dev_err(dev, "Failed to initialize sub crq irqs\n");
5714 release_crq_queue(adapter);
5720 static struct device_attribute dev_attr_failover;
5722 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5724 struct ibmvnic_adapter *adapter;
5725 struct net_device *netdev;
5726 unsigned char *mac_addr_p;
5730 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5733 mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5734 VETH_MAC_ADDR, NULL);
5737 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5738 __FILE__, __LINE__);
5742 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5743 IBMVNIC_MAX_QUEUES);
5747 adapter = netdev_priv(netdev);
5748 adapter->state = VNIC_PROBING;
5749 dev_set_drvdata(&dev->dev, netdev);
5750 adapter->vdev = dev;
5751 adapter->netdev = netdev;
5752 adapter->login_pending = false;
5753 memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
5754 /* map_ids start at 1, so ensure map_id 0 is always "in-use" */
5755 bitmap_set(adapter->map_ids, 0, 1);
5757 ether_addr_copy(adapter->mac_addr, mac_addr_p);
5758 eth_hw_addr_set(netdev, adapter->mac_addr);
5759 netdev->irq = dev->irq;
5760 netdev->netdev_ops = &ibmvnic_netdev_ops;
5761 netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5762 SET_NETDEV_DEV(netdev, &dev->dev);
5764 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5765 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5766 __ibmvnic_delayed_reset);
5767 INIT_LIST_HEAD(&adapter->rwi_list);
5768 spin_lock_init(&adapter->rwi_lock);
5769 spin_lock_init(&adapter->state_lock);
5770 mutex_init(&adapter->fw_lock);
5771 init_completion(&adapter->init_done);
5772 init_completion(&adapter->fw_done);
5773 init_completion(&adapter->reset_done);
5774 init_completion(&adapter->stats_done);
5775 clear_bit(0, &adapter->resetting);
5776 adapter->prev_rx_buf_sz = 0;
5777 adapter->prev_mtu = 0;
5779 init_success = false;
5781 rc = init_crq_queue(adapter);
5783 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5785 goto ibmvnic_init_fail;
5788 rc = ibmvnic_reset_init(adapter, false);
5789 } while (rc == -EAGAIN);
5791 /* We are ignoring the error from ibmvnic_reset_init() assuming that the
5792 * partner is not ready. CRQ is not active. When the partner becomes
5793 * ready, we will do the passive init reset.
5797 init_success = true;
5799 rc = init_stats_buffers(adapter);
5801 goto ibmvnic_init_fail;
5803 rc = init_stats_token(adapter);
5805 goto ibmvnic_stats_fail;
5807 rc = device_create_file(&dev->dev, &dev_attr_failover);
5809 goto ibmvnic_dev_file_err;
5811 netif_carrier_off(netdev);
5812 rc = register_netdev(netdev);
5814 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5815 goto ibmvnic_register_fail;
5817 dev_info(&dev->dev, "ibmvnic registered\n");
5820 adapter->state = VNIC_PROBED;
5821 netdev->mtu = adapter->req_mtu - ETH_HLEN;
5822 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5823 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5825 adapter->state = VNIC_DOWN;
5828 adapter->wait_for_reset = false;
5829 adapter->last_reset_time = jiffies;
5832 ibmvnic_register_fail:
5833 device_remove_file(&dev->dev, &dev_attr_failover);
5835 ibmvnic_dev_file_err:
5836 release_stats_token(adapter);
5839 release_stats_buffers(adapter);
5842 release_sub_crqs(adapter, 1);
5843 release_crq_queue(adapter);
5844 mutex_destroy(&adapter->fw_lock);
5845 free_netdev(netdev);
5850 static void ibmvnic_remove(struct vio_dev *dev)
5852 struct net_device *netdev = dev_get_drvdata(&dev->dev);
5853 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5854 unsigned long flags;
5856 spin_lock_irqsave(&adapter->state_lock, flags);
5858 /* If ibmvnic_reset() is scheduling a reset, wait for it to
5859 * finish. Then, set the state to REMOVING to prevent it from
5860 * scheduling any more work and to have reset functions ignore
5861 * any resets that have already been scheduled. Drop the lock
5862 * after setting state, so __ibmvnic_reset() which is called
5863 * from the flush_work() below, can make progress.
5865 spin_lock(&adapter->rwi_lock);
5866 adapter->state = VNIC_REMOVING;
5867 spin_unlock(&adapter->rwi_lock);
5869 spin_unlock_irqrestore(&adapter->state_lock, flags);
5871 flush_work(&adapter->ibmvnic_reset);
5872 flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5875 unregister_netdevice(netdev);
5877 release_resources(adapter);
5878 release_rx_pools(adapter);
5879 release_tx_pools(adapter);
5880 release_sub_crqs(adapter, 1);
5881 release_crq_queue(adapter);
5883 release_stats_token(adapter);
5884 release_stats_buffers(adapter);
5886 adapter->state = VNIC_REMOVED;
5889 mutex_destroy(&adapter->fw_lock);
5890 device_remove_file(&dev->dev, &dev_attr_failover);
5891 free_netdev(netdev);
5892 dev_set_drvdata(&dev->dev, NULL);
5895 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5896 const char *buf, size_t count)
5898 struct net_device *netdev = dev_get_drvdata(dev);
5899 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5900 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5901 __be64 session_token;
5904 if (!sysfs_streq(buf, "1"))
5907 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5908 H_GET_SESSION_TOKEN, 0, 0, 0);
5910 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5915 session_token = (__be64)retbuf[0];
5916 netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5917 be64_to_cpu(session_token));
5918 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5919 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5922 "H_VIOCTL initiated failover failed, rc %ld\n",
5926 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
5927 ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5931 static DEVICE_ATTR_WO(failover);
5933 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5935 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5936 struct ibmvnic_adapter *adapter;
5937 struct iommu_table *tbl;
5938 unsigned long ret = 0;
5941 tbl = get_iommu_table_base(&vdev->dev);
5943 /* netdev inits at probe time along with the structures we need below*/
5945 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5947 adapter = netdev_priv(netdev);
5949 ret += PAGE_SIZE; /* the crq message queue */
5950 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5952 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5953 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5955 for (i = 0; i < adapter->num_active_rx_pools; i++)
5956 ret += adapter->rx_pool[i].size *
5957 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5962 static int ibmvnic_resume(struct device *dev)
5964 struct net_device *netdev = dev_get_drvdata(dev);
5965 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5967 if (adapter->state != VNIC_OPEN)
5970 tasklet_schedule(&adapter->tasklet);
5975 static const struct vio_device_id ibmvnic_device_table[] = {
5976 {"network", "IBM,vnic"},
5979 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5981 static const struct dev_pm_ops ibmvnic_pm_ops = {
5982 .resume = ibmvnic_resume
5985 static struct vio_driver ibmvnic_driver = {
5986 .id_table = ibmvnic_device_table,
5987 .probe = ibmvnic_probe,
5988 .remove = ibmvnic_remove,
5989 .get_desired_dma = ibmvnic_get_desired_dma,
5990 .name = ibmvnic_driver_name,
5991 .pm = &ibmvnic_pm_ops,
5994 /* module functions */
5995 static int __init ibmvnic_module_init(void)
5997 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5998 IBMVNIC_DRIVER_VERSION);
6000 return vio_register_driver(&ibmvnic_driver);
6003 static void __exit ibmvnic_module_exit(void)
6005 vio_unregister_driver(&ibmvnic_driver);
6008 module_init(ibmvnic_module_init);
6009 module_exit(ibmvnic_module_exit);