Loading...
Note: File does not exist in v6.8.
1// SPDX-License-Identifier: GPL-2.0-only
2/****************************************************************************
3 * Driver for Solarflare network controllers and boards
4 * Copyright 2010-2012 Solarflare Communications Inc.
5 */
6#include <linux/pci.h>
7#include <linux/module.h>
8#include "net_driver.h"
9#include "efx.h"
10#include "nic.h"
11#include "io.h"
12#include "mcdi.h"
13#include "filter.h"
14#include "mcdi_pcol.h"
15#include "farch_regs.h"
16#include "siena_sriov.h"
17#include "vfdi.h"
18
19/* Number of longs required to track all the VIs in a VF */
20#define VI_MASK_LENGTH BITS_TO_LONGS(1 << EFX_VI_SCALE_MAX)
21
22/* Maximum number of RX queues supported */
23#define VF_MAX_RX_QUEUES 63
24
25/**
26 * enum efx_vf_tx_filter_mode - TX MAC filtering behaviour
27 * @VF_TX_FILTER_OFF: Disabled
28 * @VF_TX_FILTER_AUTO: Enabled if MAC address assigned to VF and only
29 * 2 TX queues allowed per VF.
30 * @VF_TX_FILTER_ON: Enabled
31 */
32enum efx_vf_tx_filter_mode {
33 VF_TX_FILTER_OFF,
34 VF_TX_FILTER_AUTO,
35 VF_TX_FILTER_ON,
36};
37
38/**
39 * struct siena_vf - Back-end resource and protocol state for a PCI VF
40 * @efx: The Efx NIC owning this VF
41 * @pci_rid: The PCI requester ID for this VF
42 * @pci_name: The PCI name (formatted address) of this VF
43 * @index: Index of VF within its port and PF.
44 * @req: VFDI incoming request work item. Incoming USR_EV events are received
45 * by the NAPI handler, but must be handled by executing MCDI requests
46 * inside a work item.
47 * @req_addr: VFDI incoming request DMA address (in VF's PCI address space).
48 * @req_type: Expected next incoming (from VF) %VFDI_EV_TYPE member.
49 * @req_seqno: Expected next incoming (from VF) %VFDI_EV_SEQ member.
50 * @msg_seqno: Next %VFDI_EV_SEQ member to reply to VF. Protected by
51 * @status_lock
52 * @busy: VFDI request queued to be processed or being processed. Receiving
53 * a VFDI request when @busy is set is an error condition.
54 * @buf: Incoming VFDI requests are DMA from the VF into this buffer.
55 * @buftbl_base: Buffer table entries for this VF start at this index.
56 * @rx_filtering: Receive filtering has been requested by the VF driver.
57 * @rx_filter_flags: The flags sent in the %VFDI_OP_INSERT_FILTER request.
58 * @rx_filter_qid: VF relative qid for RX filter requested by VF.
59 * @rx_filter_id: Receive MAC filter ID. Only one filter per VF is supported.
60 * @tx_filter_mode: Transmit MAC filtering mode.
61 * @tx_filter_id: Transmit MAC filter ID.
62 * @addr: The MAC address and outer vlan tag of the VF.
63 * @status_addr: VF DMA address of page for &struct vfdi_status updates.
64 * @status_lock: Mutex protecting @msg_seqno, @status_addr, @addr,
65 * @peer_page_addrs and @peer_page_count from simultaneous
66 * updates by the VM and consumption by
67 * efx_siena_sriov_update_vf_addr()
68 * @peer_page_addrs: Pointer to an array of guest pages for local addresses.
69 * @peer_page_count: Number of entries in @peer_page_count.
70 * @evq0_addrs: Array of guest pages backing evq0.
71 * @evq0_count: Number of entries in @evq0_addrs.
72 * @flush_waitq: wait queue used by %VFDI_OP_FINI_ALL_QUEUES handler
73 * to wait for flush completions.
74 * @txq_lock: Mutex for TX queue allocation.
75 * @txq_mask: Mask of initialized transmit queues.
76 * @txq_count: Number of initialized transmit queues.
77 * @rxq_mask: Mask of initialized receive queues.
78 * @rxq_count: Number of initialized receive queues.
79 * @rxq_retry_mask: Mask or receive queues that need to be flushed again
80 * due to flush failure.
81 * @rxq_retry_count: Number of receive queues in @rxq_retry_mask.
82 * @reset_work: Work item to schedule a VF reset.
83 */
84struct siena_vf {
85 struct efx_nic *efx;
86 unsigned int pci_rid;
87 char pci_name[13]; /* dddd:bb:dd.f */
88 unsigned int index;
89 struct work_struct req;
90 u64 req_addr;
91 int req_type;
92 unsigned req_seqno;
93 unsigned msg_seqno;
94 bool busy;
95 struct efx_buffer buf;
96 unsigned buftbl_base;
97 bool rx_filtering;
98 enum efx_filter_flags rx_filter_flags;
99 unsigned rx_filter_qid;
100 int rx_filter_id;
101 enum efx_vf_tx_filter_mode tx_filter_mode;
102 int tx_filter_id;
103 struct vfdi_endpoint addr;
104 u64 status_addr;
105 struct mutex status_lock;
106 u64 *peer_page_addrs;
107 unsigned peer_page_count;
108 u64 evq0_addrs[EFX_MAX_VF_EVQ_SIZE * sizeof(efx_qword_t) /
109 EFX_BUF_SIZE];
110 unsigned evq0_count;
111 wait_queue_head_t flush_waitq;
112 struct mutex txq_lock;
113 unsigned long txq_mask[VI_MASK_LENGTH];
114 unsigned txq_count;
115 unsigned long rxq_mask[VI_MASK_LENGTH];
116 unsigned rxq_count;
117 unsigned long rxq_retry_mask[VI_MASK_LENGTH];
118 atomic_t rxq_retry_count;
119 struct work_struct reset_work;
120};
121
122struct efx_memcpy_req {
123 unsigned int from_rid;
124 void *from_buf;
125 u64 from_addr;
126 unsigned int to_rid;
127 u64 to_addr;
128 unsigned length;
129};
130
131/**
132 * struct efx_local_addr - A MAC address on the vswitch without a VF.
133 *
134 * Siena does not have a switch, so VFs can't transmit data to each
135 * other. Instead the VFs must be made aware of the local addresses
136 * on the vswitch, so that they can arrange for an alternative
137 * software datapath to be used.
138 *
139 * @link: List head for insertion into efx->local_addr_list.
140 * @addr: Ethernet address
141 */
142struct efx_local_addr {
143 struct list_head link;
144 u8 addr[ETH_ALEN];
145};
146
147/**
148 * struct efx_endpoint_page - Page of vfdi_endpoint structures
149 *
150 * @link: List head for insertion into efx->local_page_list.
151 * @ptr: Pointer to page.
152 * @addr: DMA address of page.
153 */
154struct efx_endpoint_page {
155 struct list_head link;
156 void *ptr;
157 dma_addr_t addr;
158};
159
160/* Buffer table entries are reserved txq0,rxq0,evq0,txq1,rxq1,evq1 */
161#define EFX_BUFTBL_TXQ_BASE(_vf, _qid) \
162 ((_vf)->buftbl_base + EFX_VF_BUFTBL_PER_VI * (_qid))
163#define EFX_BUFTBL_RXQ_BASE(_vf, _qid) \
164 (EFX_BUFTBL_TXQ_BASE(_vf, _qid) + \
165 (EFX_MAX_DMAQ_SIZE * sizeof(efx_qword_t) / EFX_BUF_SIZE))
166#define EFX_BUFTBL_EVQ_BASE(_vf, _qid) \
167 (EFX_BUFTBL_TXQ_BASE(_vf, _qid) + \
168 (2 * EFX_MAX_DMAQ_SIZE * sizeof(efx_qword_t) / EFX_BUF_SIZE))
169
170#define EFX_FIELD_MASK(_field) \
171 ((1 << _field ## _WIDTH) - 1)
172
173/* VFs can only use this many transmit channels */
174static unsigned int vf_max_tx_channels = 2;
175module_param(vf_max_tx_channels, uint, 0444);
176MODULE_PARM_DESC(vf_max_tx_channels,
177 "Limit the number of TX channels VFs can use");
178
179static int max_vfs = -1;
180module_param(max_vfs, int, 0444);
181MODULE_PARM_DESC(max_vfs,
182 "Reduce the number of VFs initialized by the driver");
183
184/* Workqueue used by VFDI communication. We can't use the global
185 * workqueue because it may be running the VF driver's probe()
186 * routine, which will be blocked there waiting for a VFDI response.
187 */
188static struct workqueue_struct *vfdi_workqueue;
189
190static unsigned abs_index(struct siena_vf *vf, unsigned index)
191{
192 return EFX_VI_BASE + vf->index * efx_vf_size(vf->efx) + index;
193}
194
195static int efx_siena_sriov_cmd(struct efx_nic *efx, bool enable,
196 unsigned *vi_scale_out, unsigned *vf_total_out)
197{
198 MCDI_DECLARE_BUF(inbuf, MC_CMD_SRIOV_IN_LEN);
199 MCDI_DECLARE_BUF(outbuf, MC_CMD_SRIOV_OUT_LEN);
200 unsigned vi_scale, vf_total;
201 size_t outlen;
202 int rc;
203
204 MCDI_SET_DWORD(inbuf, SRIOV_IN_ENABLE, enable ? 1 : 0);
205 MCDI_SET_DWORD(inbuf, SRIOV_IN_VI_BASE, EFX_VI_BASE);
206 MCDI_SET_DWORD(inbuf, SRIOV_IN_VF_COUNT, efx->vf_count);
207
208 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_SRIOV, inbuf, MC_CMD_SRIOV_IN_LEN,
209 outbuf, MC_CMD_SRIOV_OUT_LEN, &outlen);
210 if (rc)
211 return rc;
212 if (outlen < MC_CMD_SRIOV_OUT_LEN)
213 return -EIO;
214
215 vf_total = MCDI_DWORD(outbuf, SRIOV_OUT_VF_TOTAL);
216 vi_scale = MCDI_DWORD(outbuf, SRIOV_OUT_VI_SCALE);
217 if (vi_scale > EFX_VI_SCALE_MAX)
218 return -EOPNOTSUPP;
219
220 if (vi_scale_out)
221 *vi_scale_out = vi_scale;
222 if (vf_total_out)
223 *vf_total_out = vf_total;
224
225 return 0;
226}
227
228static void efx_siena_sriov_usrev(struct efx_nic *efx, bool enabled)
229{
230 struct siena_nic_data *nic_data = efx->nic_data;
231 efx_oword_t reg;
232
233 EFX_POPULATE_OWORD_2(reg,
234 FRF_CZ_USREV_DIS, enabled ? 0 : 1,
235 FRF_CZ_DFLT_EVQ, nic_data->vfdi_channel->channel);
236 efx_writeo(efx, ®, FR_CZ_USR_EV_CFG);
237}
238
239static int efx_siena_sriov_memcpy(struct efx_nic *efx,
240 struct efx_memcpy_req *req,
241 unsigned int count)
242{
243 MCDI_DECLARE_BUF(inbuf, MCDI_CTL_SDU_LEN_MAX_V1);
244 MCDI_DECLARE_STRUCT_PTR(record);
245 unsigned int index, used;
246 u64 from_addr;
247 u32 from_rid;
248 int rc;
249
250 mb(); /* Finish writing source/reading dest before DMA starts */
251
252 if (WARN_ON(count > MC_CMD_MEMCPY_IN_RECORD_MAXNUM))
253 return -ENOBUFS;
254 used = MC_CMD_MEMCPY_IN_LEN(count);
255
256 for (index = 0; index < count; index++) {
257 record = MCDI_ARRAY_STRUCT_PTR(inbuf, MEMCPY_IN_RECORD, index);
258 MCDI_SET_DWORD(record, MEMCPY_RECORD_TYPEDEF_NUM_RECORDS,
259 count);
260 MCDI_SET_DWORD(record, MEMCPY_RECORD_TYPEDEF_TO_RID,
261 req->to_rid);
262 MCDI_SET_QWORD(record, MEMCPY_RECORD_TYPEDEF_TO_ADDR,
263 req->to_addr);
264 if (req->from_buf == NULL) {
265 from_rid = req->from_rid;
266 from_addr = req->from_addr;
267 } else {
268 if (WARN_ON(used + req->length >
269 MCDI_CTL_SDU_LEN_MAX_V1)) {
270 rc = -ENOBUFS;
271 goto out;
272 }
273
274 from_rid = MC_CMD_MEMCPY_RECORD_TYPEDEF_RID_INLINE;
275 from_addr = used;
276 memcpy(_MCDI_PTR(inbuf, used), req->from_buf,
277 req->length);
278 used += req->length;
279 }
280
281 MCDI_SET_DWORD(record, MEMCPY_RECORD_TYPEDEF_FROM_RID, from_rid);
282 MCDI_SET_QWORD(record, MEMCPY_RECORD_TYPEDEF_FROM_ADDR,
283 from_addr);
284 MCDI_SET_DWORD(record, MEMCPY_RECORD_TYPEDEF_LENGTH,
285 req->length);
286
287 ++req;
288 }
289
290 rc = efx_mcdi_rpc(efx, MC_CMD_MEMCPY, inbuf, used, NULL, 0, NULL);
291out:
292 mb(); /* Don't write source/read dest before DMA is complete */
293
294 return rc;
295}
296
297/* The TX filter is entirely controlled by this driver, and is modified
298 * underneath the feet of the VF
299 */
300static void efx_siena_sriov_reset_tx_filter(struct siena_vf *vf)
301{
302 struct efx_nic *efx = vf->efx;
303 struct efx_filter_spec filter;
304 u16 vlan;
305 int rc;
306
307 if (vf->tx_filter_id != -1) {
308 efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
309 vf->tx_filter_id);
310 netif_dbg(efx, hw, efx->net_dev, "Removed vf %s tx filter %d\n",
311 vf->pci_name, vf->tx_filter_id);
312 vf->tx_filter_id = -1;
313 }
314
315 if (is_zero_ether_addr(vf->addr.mac_addr))
316 return;
317
318 /* Turn on TX filtering automatically if not explicitly
319 * enabled or disabled.
320 */
321 if (vf->tx_filter_mode == VF_TX_FILTER_AUTO && vf_max_tx_channels <= 2)
322 vf->tx_filter_mode = VF_TX_FILTER_ON;
323
324 vlan = ntohs(vf->addr.tci) & VLAN_VID_MASK;
325 efx_filter_init_tx(&filter, abs_index(vf, 0));
326 rc = efx_filter_set_eth_local(&filter,
327 vlan ? vlan : EFX_FILTER_VID_UNSPEC,
328 vf->addr.mac_addr);
329 BUG_ON(rc);
330
331 rc = efx_filter_insert_filter(efx, &filter, true);
332 if (rc < 0) {
333 netif_warn(efx, hw, efx->net_dev,
334 "Unable to migrate tx filter for vf %s\n",
335 vf->pci_name);
336 } else {
337 netif_dbg(efx, hw, efx->net_dev, "Inserted vf %s tx filter %d\n",
338 vf->pci_name, rc);
339 vf->tx_filter_id = rc;
340 }
341}
342
343/* The RX filter is managed here on behalf of the VF driver */
344static void efx_siena_sriov_reset_rx_filter(struct siena_vf *vf)
345{
346 struct efx_nic *efx = vf->efx;
347 struct efx_filter_spec filter;
348 u16 vlan;
349 int rc;
350
351 if (vf->rx_filter_id != -1) {
352 efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
353 vf->rx_filter_id);
354 netif_dbg(efx, hw, efx->net_dev, "Removed vf %s rx filter %d\n",
355 vf->pci_name, vf->rx_filter_id);
356 vf->rx_filter_id = -1;
357 }
358
359 if (!vf->rx_filtering || is_zero_ether_addr(vf->addr.mac_addr))
360 return;
361
362 vlan = ntohs(vf->addr.tci) & VLAN_VID_MASK;
363 efx_filter_init_rx(&filter, EFX_FILTER_PRI_REQUIRED,
364 vf->rx_filter_flags,
365 abs_index(vf, vf->rx_filter_qid));
366 rc = efx_filter_set_eth_local(&filter,
367 vlan ? vlan : EFX_FILTER_VID_UNSPEC,
368 vf->addr.mac_addr);
369 BUG_ON(rc);
370
371 rc = efx_filter_insert_filter(efx, &filter, true);
372 if (rc < 0) {
373 netif_warn(efx, hw, efx->net_dev,
374 "Unable to insert rx filter for vf %s\n",
375 vf->pci_name);
376 } else {
377 netif_dbg(efx, hw, efx->net_dev, "Inserted vf %s rx filter %d\n",
378 vf->pci_name, rc);
379 vf->rx_filter_id = rc;
380 }
381}
382
383static void __efx_siena_sriov_update_vf_addr(struct siena_vf *vf)
384{
385 struct efx_nic *efx = vf->efx;
386 struct siena_nic_data *nic_data = efx->nic_data;
387
388 efx_siena_sriov_reset_tx_filter(vf);
389 efx_siena_sriov_reset_rx_filter(vf);
390 queue_work(vfdi_workqueue, &nic_data->peer_work);
391}
392
393/* Push the peer list to this VF. The caller must hold status_lock to interlock
394 * with VFDI requests, and they must be serialised against manipulation of
395 * local_page_list, either by acquiring local_lock or by running from
396 * efx_siena_sriov_peer_work()
397 */
398static void __efx_siena_sriov_push_vf_status(struct siena_vf *vf)
399{
400 struct efx_nic *efx = vf->efx;
401 struct siena_nic_data *nic_data = efx->nic_data;
402 struct vfdi_status *status = nic_data->vfdi_status.addr;
403 struct efx_memcpy_req copy[4];
404 struct efx_endpoint_page *epp;
405 unsigned int pos, count;
406 unsigned data_offset;
407 efx_qword_t event;
408
409 WARN_ON(!mutex_is_locked(&vf->status_lock));
410 WARN_ON(!vf->status_addr);
411
412 status->local = vf->addr;
413 status->generation_end = ++status->generation_start;
414
415 memset(copy, '\0', sizeof(copy));
416 /* Write generation_start */
417 copy[0].from_buf = &status->generation_start;
418 copy[0].to_rid = vf->pci_rid;
419 copy[0].to_addr = vf->status_addr + offsetof(struct vfdi_status,
420 generation_start);
421 copy[0].length = sizeof(status->generation_start);
422 /* DMA the rest of the structure (excluding the generations). This
423 * assumes that the non-generation portion of vfdi_status is in
424 * one chunk starting at the version member.
425 */
426 data_offset = offsetof(struct vfdi_status, version);
427 copy[1].from_rid = efx->pci_dev->devfn;
428 copy[1].from_addr = nic_data->vfdi_status.dma_addr + data_offset;
429 copy[1].to_rid = vf->pci_rid;
430 copy[1].to_addr = vf->status_addr + data_offset;
431 copy[1].length = status->length - data_offset;
432
433 /* Copy the peer pages */
434 pos = 2;
435 count = 0;
436 list_for_each_entry(epp, &nic_data->local_page_list, link) {
437 if (count == vf->peer_page_count) {
438 /* The VF driver will know they need to provide more
439 * pages because peer_addr_count is too large.
440 */
441 break;
442 }
443 copy[pos].from_buf = NULL;
444 copy[pos].from_rid = efx->pci_dev->devfn;
445 copy[pos].from_addr = epp->addr;
446 copy[pos].to_rid = vf->pci_rid;
447 copy[pos].to_addr = vf->peer_page_addrs[count];
448 copy[pos].length = EFX_PAGE_SIZE;
449
450 if (++pos == ARRAY_SIZE(copy)) {
451 efx_siena_sriov_memcpy(efx, copy, ARRAY_SIZE(copy));
452 pos = 0;
453 }
454 ++count;
455 }
456
457 /* Write generation_end */
458 copy[pos].from_buf = &status->generation_end;
459 copy[pos].to_rid = vf->pci_rid;
460 copy[pos].to_addr = vf->status_addr + offsetof(struct vfdi_status,
461 generation_end);
462 copy[pos].length = sizeof(status->generation_end);
463 efx_siena_sriov_memcpy(efx, copy, pos + 1);
464
465 /* Notify the guest */
466 EFX_POPULATE_QWORD_3(event,
467 FSF_AZ_EV_CODE, FSE_CZ_EV_CODE_USER_EV,
468 VFDI_EV_SEQ, (vf->msg_seqno & 0xff),
469 VFDI_EV_TYPE, VFDI_EV_TYPE_STATUS);
470 ++vf->msg_seqno;
471 efx_farch_generate_event(efx,
472 EFX_VI_BASE + vf->index * efx_vf_size(efx),
473 &event);
474}
475
476static void efx_siena_sriov_bufs(struct efx_nic *efx, unsigned offset,
477 u64 *addr, unsigned count)
478{
479 efx_qword_t buf;
480 unsigned pos;
481
482 for (pos = 0; pos < count; ++pos) {
483 EFX_POPULATE_QWORD_3(buf,
484 FRF_AZ_BUF_ADR_REGION, 0,
485 FRF_AZ_BUF_ADR_FBUF,
486 addr ? addr[pos] >> 12 : 0,
487 FRF_AZ_BUF_OWNER_ID_FBUF, 0);
488 efx_sram_writeq(efx, efx->membase + FR_BZ_BUF_FULL_TBL,
489 &buf, offset + pos);
490 }
491}
492
493static bool bad_vf_index(struct efx_nic *efx, unsigned index)
494{
495 return index >= efx_vf_size(efx);
496}
497
498static bool bad_buf_count(unsigned buf_count, unsigned max_entry_count)
499{
500 unsigned max_buf_count = max_entry_count *
501 sizeof(efx_qword_t) / EFX_BUF_SIZE;
502
503 return ((buf_count & (buf_count - 1)) || buf_count > max_buf_count);
504}
505
506/* Check that VI specified by per-port index belongs to a VF.
507 * Optionally set VF index and VI index within the VF.
508 */
509static bool map_vi_index(struct efx_nic *efx, unsigned abs_index,
510 struct siena_vf **vf_out, unsigned *rel_index_out)
511{
512 struct siena_nic_data *nic_data = efx->nic_data;
513 unsigned vf_i;
514
515 if (abs_index < EFX_VI_BASE)
516 return true;
517 vf_i = (abs_index - EFX_VI_BASE) / efx_vf_size(efx);
518 if (vf_i >= efx->vf_init_count)
519 return true;
520
521 if (vf_out)
522 *vf_out = nic_data->vf + vf_i;
523 if (rel_index_out)
524 *rel_index_out = abs_index % efx_vf_size(efx);
525 return false;
526}
527
528static int efx_vfdi_init_evq(struct siena_vf *vf)
529{
530 struct efx_nic *efx = vf->efx;
531 struct vfdi_req *req = vf->buf.addr;
532 unsigned vf_evq = req->u.init_evq.index;
533 unsigned buf_count = req->u.init_evq.buf_count;
534 unsigned abs_evq = abs_index(vf, vf_evq);
535 unsigned buftbl = EFX_BUFTBL_EVQ_BASE(vf, vf_evq);
536 efx_oword_t reg;
537
538 if (bad_vf_index(efx, vf_evq) ||
539 bad_buf_count(buf_count, EFX_MAX_VF_EVQ_SIZE)) {
540 if (net_ratelimit())
541 netif_err(efx, hw, efx->net_dev,
542 "ERROR: Invalid INIT_EVQ from %s: evq %d bufs %d\n",
543 vf->pci_name, vf_evq, buf_count);
544 return VFDI_RC_EINVAL;
545 }
546
547 efx_siena_sriov_bufs(efx, buftbl, req->u.init_evq.addr, buf_count);
548
549 EFX_POPULATE_OWORD_3(reg,
550 FRF_CZ_TIMER_Q_EN, 1,
551 FRF_CZ_HOST_NOTIFY_MODE, 0,
552 FRF_CZ_TIMER_MODE, FFE_CZ_TIMER_MODE_DIS);
553 efx_writeo_table(efx, ®, FR_BZ_TIMER_TBL, abs_evq);
554 EFX_POPULATE_OWORD_3(reg,
555 FRF_AZ_EVQ_EN, 1,
556 FRF_AZ_EVQ_SIZE, __ffs(buf_count),
557 FRF_AZ_EVQ_BUF_BASE_ID, buftbl);
558 efx_writeo_table(efx, ®, FR_BZ_EVQ_PTR_TBL, abs_evq);
559
560 if (vf_evq == 0) {
561 memcpy(vf->evq0_addrs, req->u.init_evq.addr,
562 buf_count * sizeof(u64));
563 vf->evq0_count = buf_count;
564 }
565
566 return VFDI_RC_SUCCESS;
567}
568
569static int efx_vfdi_init_rxq(struct siena_vf *vf)
570{
571 struct efx_nic *efx = vf->efx;
572 struct vfdi_req *req = vf->buf.addr;
573 unsigned vf_rxq = req->u.init_rxq.index;
574 unsigned vf_evq = req->u.init_rxq.evq;
575 unsigned buf_count = req->u.init_rxq.buf_count;
576 unsigned buftbl = EFX_BUFTBL_RXQ_BASE(vf, vf_rxq);
577 unsigned label;
578 efx_oword_t reg;
579
580 if (bad_vf_index(efx, vf_evq) || bad_vf_index(efx, vf_rxq) ||
581 vf_rxq >= VF_MAX_RX_QUEUES ||
582 bad_buf_count(buf_count, EFX_MAX_DMAQ_SIZE)) {
583 if (net_ratelimit())
584 netif_err(efx, hw, efx->net_dev,
585 "ERROR: Invalid INIT_RXQ from %s: rxq %d evq %d "
586 "buf_count %d\n", vf->pci_name, vf_rxq,
587 vf_evq, buf_count);
588 return VFDI_RC_EINVAL;
589 }
590 if (__test_and_set_bit(req->u.init_rxq.index, vf->rxq_mask))
591 ++vf->rxq_count;
592 efx_siena_sriov_bufs(efx, buftbl, req->u.init_rxq.addr, buf_count);
593
594 label = req->u.init_rxq.label & EFX_FIELD_MASK(FRF_AZ_RX_DESCQ_LABEL);
595 EFX_POPULATE_OWORD_6(reg,
596 FRF_AZ_RX_DESCQ_BUF_BASE_ID, buftbl,
597 FRF_AZ_RX_DESCQ_EVQ_ID, abs_index(vf, vf_evq),
598 FRF_AZ_RX_DESCQ_LABEL, label,
599 FRF_AZ_RX_DESCQ_SIZE, __ffs(buf_count),
600 FRF_AZ_RX_DESCQ_JUMBO,
601 !!(req->u.init_rxq.flags &
602 VFDI_RXQ_FLAG_SCATTER_EN),
603 FRF_AZ_RX_DESCQ_EN, 1);
604 efx_writeo_table(efx, ®, FR_BZ_RX_DESC_PTR_TBL,
605 abs_index(vf, vf_rxq));
606
607 return VFDI_RC_SUCCESS;
608}
609
610static int efx_vfdi_init_txq(struct siena_vf *vf)
611{
612 struct efx_nic *efx = vf->efx;
613 struct vfdi_req *req = vf->buf.addr;
614 unsigned vf_txq = req->u.init_txq.index;
615 unsigned vf_evq = req->u.init_txq.evq;
616 unsigned buf_count = req->u.init_txq.buf_count;
617 unsigned buftbl = EFX_BUFTBL_TXQ_BASE(vf, vf_txq);
618 unsigned label, eth_filt_en;
619 efx_oword_t reg;
620
621 if (bad_vf_index(efx, vf_evq) || bad_vf_index(efx, vf_txq) ||
622 vf_txq >= vf_max_tx_channels ||
623 bad_buf_count(buf_count, EFX_MAX_DMAQ_SIZE)) {
624 if (net_ratelimit())
625 netif_err(efx, hw, efx->net_dev,
626 "ERROR: Invalid INIT_TXQ from %s: txq %d evq %d "
627 "buf_count %d\n", vf->pci_name, vf_txq,
628 vf_evq, buf_count);
629 return VFDI_RC_EINVAL;
630 }
631
632 mutex_lock(&vf->txq_lock);
633 if (__test_and_set_bit(req->u.init_txq.index, vf->txq_mask))
634 ++vf->txq_count;
635 mutex_unlock(&vf->txq_lock);
636 efx_siena_sriov_bufs(efx, buftbl, req->u.init_txq.addr, buf_count);
637
638 eth_filt_en = vf->tx_filter_mode == VF_TX_FILTER_ON;
639
640 label = req->u.init_txq.label & EFX_FIELD_MASK(FRF_AZ_TX_DESCQ_LABEL);
641 EFX_POPULATE_OWORD_8(reg,
642 FRF_CZ_TX_DPT_Q_MASK_WIDTH, min(efx->vi_scale, 1U),
643 FRF_CZ_TX_DPT_ETH_FILT_EN, eth_filt_en,
644 FRF_AZ_TX_DESCQ_EN, 1,
645 FRF_AZ_TX_DESCQ_BUF_BASE_ID, buftbl,
646 FRF_AZ_TX_DESCQ_EVQ_ID, abs_index(vf, vf_evq),
647 FRF_AZ_TX_DESCQ_LABEL, label,
648 FRF_AZ_TX_DESCQ_SIZE, __ffs(buf_count),
649 FRF_BZ_TX_NON_IP_DROP_DIS, 1);
650 efx_writeo_table(efx, ®, FR_BZ_TX_DESC_PTR_TBL,
651 abs_index(vf, vf_txq));
652
653 return VFDI_RC_SUCCESS;
654}
655
656/* Returns true when efx_vfdi_fini_all_queues should wake */
657static bool efx_vfdi_flush_wake(struct siena_vf *vf)
658{
659 /* Ensure that all updates are visible to efx_vfdi_fini_all_queues() */
660 smp_mb();
661
662 return (!vf->txq_count && !vf->rxq_count) ||
663 atomic_read(&vf->rxq_retry_count);
664}
665
666static void efx_vfdi_flush_clear(struct siena_vf *vf)
667{
668 memset(vf->txq_mask, 0, sizeof(vf->txq_mask));
669 vf->txq_count = 0;
670 memset(vf->rxq_mask, 0, sizeof(vf->rxq_mask));
671 vf->rxq_count = 0;
672 memset(vf->rxq_retry_mask, 0, sizeof(vf->rxq_retry_mask));
673 atomic_set(&vf->rxq_retry_count, 0);
674}
675
676static int efx_vfdi_fini_all_queues(struct siena_vf *vf)
677{
678 struct efx_nic *efx = vf->efx;
679 efx_oword_t reg;
680 unsigned count = efx_vf_size(efx);
681 unsigned vf_offset = EFX_VI_BASE + vf->index * efx_vf_size(efx);
682 unsigned timeout = HZ;
683 unsigned index, rxqs_count;
684 MCDI_DECLARE_BUF(inbuf, MC_CMD_FLUSH_RX_QUEUES_IN_LENMAX);
685 int rc;
686
687 BUILD_BUG_ON(VF_MAX_RX_QUEUES >
688 MC_CMD_FLUSH_RX_QUEUES_IN_QID_OFST_MAXNUM);
689
690 rtnl_lock();
691 siena_prepare_flush(efx);
692 rtnl_unlock();
693
694 /* Flush all the initialized queues */
695 rxqs_count = 0;
696 for (index = 0; index < count; ++index) {
697 if (test_bit(index, vf->txq_mask)) {
698 EFX_POPULATE_OWORD_2(reg,
699 FRF_AZ_TX_FLUSH_DESCQ_CMD, 1,
700 FRF_AZ_TX_FLUSH_DESCQ,
701 vf_offset + index);
702 efx_writeo(efx, ®, FR_AZ_TX_FLUSH_DESCQ);
703 }
704 if (test_bit(index, vf->rxq_mask)) {
705 MCDI_SET_ARRAY_DWORD(
706 inbuf, FLUSH_RX_QUEUES_IN_QID_OFST,
707 rxqs_count, vf_offset + index);
708 rxqs_count++;
709 }
710 }
711
712 atomic_set(&vf->rxq_retry_count, 0);
713 while (timeout && (vf->rxq_count || vf->txq_count)) {
714 rc = efx_mcdi_rpc(efx, MC_CMD_FLUSH_RX_QUEUES, inbuf,
715 MC_CMD_FLUSH_RX_QUEUES_IN_LEN(rxqs_count),
716 NULL, 0, NULL);
717 WARN_ON(rc < 0);
718
719 timeout = wait_event_timeout(vf->flush_waitq,
720 efx_vfdi_flush_wake(vf),
721 timeout);
722 rxqs_count = 0;
723 for (index = 0; index < count; ++index) {
724 if (test_and_clear_bit(index, vf->rxq_retry_mask)) {
725 atomic_dec(&vf->rxq_retry_count);
726 MCDI_SET_ARRAY_DWORD(
727 inbuf, FLUSH_RX_QUEUES_IN_QID_OFST,
728 rxqs_count, vf_offset + index);
729 rxqs_count++;
730 }
731 }
732 }
733
734 rtnl_lock();
735 siena_finish_flush(efx);
736 rtnl_unlock();
737
738 /* Irrespective of success/failure, fini the queues */
739 EFX_ZERO_OWORD(reg);
740 for (index = 0; index < count; ++index) {
741 efx_writeo_table(efx, ®, FR_BZ_RX_DESC_PTR_TBL,
742 vf_offset + index);
743 efx_writeo_table(efx, ®, FR_BZ_TX_DESC_PTR_TBL,
744 vf_offset + index);
745 efx_writeo_table(efx, ®, FR_BZ_EVQ_PTR_TBL,
746 vf_offset + index);
747 efx_writeo_table(efx, ®, FR_BZ_TIMER_TBL,
748 vf_offset + index);
749 }
750 efx_siena_sriov_bufs(efx, vf->buftbl_base, NULL,
751 EFX_VF_BUFTBL_PER_VI * efx_vf_size(efx));
752 efx_vfdi_flush_clear(vf);
753
754 vf->evq0_count = 0;
755
756 return timeout ? 0 : VFDI_RC_ETIMEDOUT;
757}
758
759static int efx_vfdi_insert_filter(struct siena_vf *vf)
760{
761 struct efx_nic *efx = vf->efx;
762 struct siena_nic_data *nic_data = efx->nic_data;
763 struct vfdi_req *req = vf->buf.addr;
764 unsigned vf_rxq = req->u.mac_filter.rxq;
765 unsigned flags;
766
767 if (bad_vf_index(efx, vf_rxq) || vf->rx_filtering) {
768 if (net_ratelimit())
769 netif_err(efx, hw, efx->net_dev,
770 "ERROR: Invalid INSERT_FILTER from %s: rxq %d "
771 "flags 0x%x\n", vf->pci_name, vf_rxq,
772 req->u.mac_filter.flags);
773 return VFDI_RC_EINVAL;
774 }
775
776 flags = 0;
777 if (req->u.mac_filter.flags & VFDI_MAC_FILTER_FLAG_RSS)
778 flags |= EFX_FILTER_FLAG_RX_RSS;
779 if (req->u.mac_filter.flags & VFDI_MAC_FILTER_FLAG_SCATTER)
780 flags |= EFX_FILTER_FLAG_RX_SCATTER;
781 vf->rx_filter_flags = flags;
782 vf->rx_filter_qid = vf_rxq;
783 vf->rx_filtering = true;
784
785 efx_siena_sriov_reset_rx_filter(vf);
786 queue_work(vfdi_workqueue, &nic_data->peer_work);
787
788 return VFDI_RC_SUCCESS;
789}
790
791static int efx_vfdi_remove_all_filters(struct siena_vf *vf)
792{
793 struct efx_nic *efx = vf->efx;
794 struct siena_nic_data *nic_data = efx->nic_data;
795
796 vf->rx_filtering = false;
797 efx_siena_sriov_reset_rx_filter(vf);
798 queue_work(vfdi_workqueue, &nic_data->peer_work);
799
800 return VFDI_RC_SUCCESS;
801}
802
803static int efx_vfdi_set_status_page(struct siena_vf *vf)
804{
805 struct efx_nic *efx = vf->efx;
806 struct siena_nic_data *nic_data = efx->nic_data;
807 struct vfdi_req *req = vf->buf.addr;
808 u64 page_count = req->u.set_status_page.peer_page_count;
809 u64 max_page_count =
810 (EFX_PAGE_SIZE -
811 offsetof(struct vfdi_req, u.set_status_page.peer_page_addr[0]))
812 / sizeof(req->u.set_status_page.peer_page_addr[0]);
813
814 if (!req->u.set_status_page.dma_addr || page_count > max_page_count) {
815 if (net_ratelimit())
816 netif_err(efx, hw, efx->net_dev,
817 "ERROR: Invalid SET_STATUS_PAGE from %s\n",
818 vf->pci_name);
819 return VFDI_RC_EINVAL;
820 }
821
822 mutex_lock(&nic_data->local_lock);
823 mutex_lock(&vf->status_lock);
824 vf->status_addr = req->u.set_status_page.dma_addr;
825
826 kfree(vf->peer_page_addrs);
827 vf->peer_page_addrs = NULL;
828 vf->peer_page_count = 0;
829
830 if (page_count) {
831 vf->peer_page_addrs = kcalloc(page_count, sizeof(u64),
832 GFP_KERNEL);
833 if (vf->peer_page_addrs) {
834 memcpy(vf->peer_page_addrs,
835 req->u.set_status_page.peer_page_addr,
836 page_count * sizeof(u64));
837 vf->peer_page_count = page_count;
838 }
839 }
840
841 __efx_siena_sriov_push_vf_status(vf);
842 mutex_unlock(&vf->status_lock);
843 mutex_unlock(&nic_data->local_lock);
844
845 return VFDI_RC_SUCCESS;
846}
847
848static int efx_vfdi_clear_status_page(struct siena_vf *vf)
849{
850 mutex_lock(&vf->status_lock);
851 vf->status_addr = 0;
852 mutex_unlock(&vf->status_lock);
853
854 return VFDI_RC_SUCCESS;
855}
856
857typedef int (*efx_vfdi_op_t)(struct siena_vf *vf);
858
859static const efx_vfdi_op_t vfdi_ops[VFDI_OP_LIMIT] = {
860 [VFDI_OP_INIT_EVQ] = efx_vfdi_init_evq,
861 [VFDI_OP_INIT_TXQ] = efx_vfdi_init_txq,
862 [VFDI_OP_INIT_RXQ] = efx_vfdi_init_rxq,
863 [VFDI_OP_FINI_ALL_QUEUES] = efx_vfdi_fini_all_queues,
864 [VFDI_OP_INSERT_FILTER] = efx_vfdi_insert_filter,
865 [VFDI_OP_REMOVE_ALL_FILTERS] = efx_vfdi_remove_all_filters,
866 [VFDI_OP_SET_STATUS_PAGE] = efx_vfdi_set_status_page,
867 [VFDI_OP_CLEAR_STATUS_PAGE] = efx_vfdi_clear_status_page,
868};
869
870static void efx_siena_sriov_vfdi(struct work_struct *work)
871{
872 struct siena_vf *vf = container_of(work, struct siena_vf, req);
873 struct efx_nic *efx = vf->efx;
874 struct vfdi_req *req = vf->buf.addr;
875 struct efx_memcpy_req copy[2];
876 int rc;
877
878 /* Copy this page into the local address space */
879 memset(copy, '\0', sizeof(copy));
880 copy[0].from_rid = vf->pci_rid;
881 copy[0].from_addr = vf->req_addr;
882 copy[0].to_rid = efx->pci_dev->devfn;
883 copy[0].to_addr = vf->buf.dma_addr;
884 copy[0].length = EFX_PAGE_SIZE;
885 rc = efx_siena_sriov_memcpy(efx, copy, 1);
886 if (rc) {
887 /* If we can't get the request, we can't reply to the caller */
888 if (net_ratelimit())
889 netif_err(efx, hw, efx->net_dev,
890 "ERROR: Unable to fetch VFDI request from %s rc %d\n",
891 vf->pci_name, -rc);
892 vf->busy = false;
893 return;
894 }
895
896 if (req->op < VFDI_OP_LIMIT && vfdi_ops[req->op] != NULL) {
897 rc = vfdi_ops[req->op](vf);
898 if (rc == 0) {
899 netif_dbg(efx, hw, efx->net_dev,
900 "vfdi request %d from %s ok\n",
901 req->op, vf->pci_name);
902 }
903 } else {
904 netif_dbg(efx, hw, efx->net_dev,
905 "ERROR: Unrecognised request %d from VF %s addr "
906 "%llx\n", req->op, vf->pci_name,
907 (unsigned long long)vf->req_addr);
908 rc = VFDI_RC_EOPNOTSUPP;
909 }
910
911 /* Allow subsequent VF requests */
912 vf->busy = false;
913 smp_wmb();
914
915 /* Respond to the request */
916 req->rc = rc;
917 req->op = VFDI_OP_RESPONSE;
918
919 memset(copy, '\0', sizeof(copy));
920 copy[0].from_buf = &req->rc;
921 copy[0].to_rid = vf->pci_rid;
922 copy[0].to_addr = vf->req_addr + offsetof(struct vfdi_req, rc);
923 copy[0].length = sizeof(req->rc);
924 copy[1].from_buf = &req->op;
925 copy[1].to_rid = vf->pci_rid;
926 copy[1].to_addr = vf->req_addr + offsetof(struct vfdi_req, op);
927 copy[1].length = sizeof(req->op);
928
929 (void)efx_siena_sriov_memcpy(efx, copy, ARRAY_SIZE(copy));
930}
931
932
933
934/* After a reset the event queues inside the guests no longer exist. Fill the
935 * event ring in guest memory with VFDI reset events, then (re-initialise) the
936 * event queue to raise an interrupt. The guest driver will then recover.
937 */
938
939static void efx_siena_sriov_reset_vf(struct siena_vf *vf,
940 struct efx_buffer *buffer)
941{
942 struct efx_nic *efx = vf->efx;
943 struct efx_memcpy_req copy_req[4];
944 efx_qword_t event;
945 unsigned int pos, count, k, buftbl, abs_evq;
946 efx_oword_t reg;
947 efx_dword_t ptr;
948 int rc;
949
950 BUG_ON(buffer->len != EFX_PAGE_SIZE);
951
952 if (!vf->evq0_count)
953 return;
954 BUG_ON(vf->evq0_count & (vf->evq0_count - 1));
955
956 mutex_lock(&vf->status_lock);
957 EFX_POPULATE_QWORD_3(event,
958 FSF_AZ_EV_CODE, FSE_CZ_EV_CODE_USER_EV,
959 VFDI_EV_SEQ, vf->msg_seqno,
960 VFDI_EV_TYPE, VFDI_EV_TYPE_RESET);
961 vf->msg_seqno++;
962 for (pos = 0; pos < EFX_PAGE_SIZE; pos += sizeof(event))
963 memcpy(buffer->addr + pos, &event, sizeof(event));
964
965 for (pos = 0; pos < vf->evq0_count; pos += count) {
966 count = min_t(unsigned, vf->evq0_count - pos,
967 ARRAY_SIZE(copy_req));
968 for (k = 0; k < count; k++) {
969 copy_req[k].from_buf = NULL;
970 copy_req[k].from_rid = efx->pci_dev->devfn;
971 copy_req[k].from_addr = buffer->dma_addr;
972 copy_req[k].to_rid = vf->pci_rid;
973 copy_req[k].to_addr = vf->evq0_addrs[pos + k];
974 copy_req[k].length = EFX_PAGE_SIZE;
975 }
976 rc = efx_siena_sriov_memcpy(efx, copy_req, count);
977 if (rc) {
978 if (net_ratelimit())
979 netif_err(efx, hw, efx->net_dev,
980 "ERROR: Unable to notify %s of reset"
981 ": %d\n", vf->pci_name, -rc);
982 break;
983 }
984 }
985
986 /* Reinitialise, arm and trigger evq0 */
987 abs_evq = abs_index(vf, 0);
988 buftbl = EFX_BUFTBL_EVQ_BASE(vf, 0);
989 efx_siena_sriov_bufs(efx, buftbl, vf->evq0_addrs, vf->evq0_count);
990
991 EFX_POPULATE_OWORD_3(reg,
992 FRF_CZ_TIMER_Q_EN, 1,
993 FRF_CZ_HOST_NOTIFY_MODE, 0,
994 FRF_CZ_TIMER_MODE, FFE_CZ_TIMER_MODE_DIS);
995 efx_writeo_table(efx, ®, FR_BZ_TIMER_TBL, abs_evq);
996 EFX_POPULATE_OWORD_3(reg,
997 FRF_AZ_EVQ_EN, 1,
998 FRF_AZ_EVQ_SIZE, __ffs(vf->evq0_count),
999 FRF_AZ_EVQ_BUF_BASE_ID, buftbl);
1000 efx_writeo_table(efx, ®, FR_BZ_EVQ_PTR_TBL, abs_evq);
1001 EFX_POPULATE_DWORD_1(ptr, FRF_AZ_EVQ_RPTR, 0);
1002 efx_writed(efx, &ptr, FR_BZ_EVQ_RPTR + FR_BZ_EVQ_RPTR_STEP * abs_evq);
1003
1004 mutex_unlock(&vf->status_lock);
1005}
1006
1007static void efx_siena_sriov_reset_vf_work(struct work_struct *work)
1008{
1009 struct siena_vf *vf = container_of(work, struct siena_vf, req);
1010 struct efx_nic *efx = vf->efx;
1011 struct efx_buffer buf;
1012
1013 if (!efx_nic_alloc_buffer(efx, &buf, EFX_PAGE_SIZE, GFP_NOIO)) {
1014 efx_siena_sriov_reset_vf(vf, &buf);
1015 efx_nic_free_buffer(efx, &buf);
1016 }
1017}
1018
1019static void efx_siena_sriov_handle_no_channel(struct efx_nic *efx)
1020{
1021 netif_err(efx, drv, efx->net_dev,
1022 "ERROR: IOV requires MSI-X and 1 additional interrupt"
1023 "vector. IOV disabled\n");
1024 efx->vf_count = 0;
1025}
1026
1027static int efx_siena_sriov_probe_channel(struct efx_channel *channel)
1028{
1029 struct siena_nic_data *nic_data = channel->efx->nic_data;
1030 nic_data->vfdi_channel = channel;
1031
1032 return 0;
1033}
1034
1035static void
1036efx_siena_sriov_get_channel_name(struct efx_channel *channel,
1037 char *buf, size_t len)
1038{
1039 snprintf(buf, len, "%s-iov", channel->efx->name);
1040}
1041
1042static const struct efx_channel_type efx_siena_sriov_channel_type = {
1043 .handle_no_channel = efx_siena_sriov_handle_no_channel,
1044 .pre_probe = efx_siena_sriov_probe_channel,
1045 .post_remove = efx_channel_dummy_op_void,
1046 .get_name = efx_siena_sriov_get_channel_name,
1047 /* no copy operation; channel must not be reallocated */
1048 .keep_eventq = true,
1049};
1050
1051void efx_siena_sriov_probe(struct efx_nic *efx)
1052{
1053 unsigned count;
1054
1055 if (!max_vfs)
1056 return;
1057
1058 if (efx_siena_sriov_cmd(efx, false, &efx->vi_scale, &count)) {
1059 netif_info(efx, probe, efx->net_dev, "no SR-IOV VFs probed\n");
1060 return;
1061 }
1062 if (count > 0 && count > max_vfs)
1063 count = max_vfs;
1064
1065 /* efx_nic_dimension_resources() will reduce vf_count as appopriate */
1066 efx->vf_count = count;
1067
1068 efx->extra_channel_type[EFX_EXTRA_CHANNEL_IOV] = &efx_siena_sriov_channel_type;
1069}
1070
1071/* Copy the list of individual addresses into the vfdi_status.peers
1072 * array and auxiliary pages, protected by %local_lock. Drop that lock
1073 * and then broadcast the address list to every VF.
1074 */
1075static void efx_siena_sriov_peer_work(struct work_struct *data)
1076{
1077 struct siena_nic_data *nic_data = container_of(data,
1078 struct siena_nic_data,
1079 peer_work);
1080 struct efx_nic *efx = nic_data->efx;
1081 struct vfdi_status *vfdi_status = nic_data->vfdi_status.addr;
1082 struct siena_vf *vf;
1083 struct efx_local_addr *local_addr;
1084 struct vfdi_endpoint *peer;
1085 struct efx_endpoint_page *epp;
1086 struct list_head pages;
1087 unsigned int peer_space;
1088 unsigned int peer_count;
1089 unsigned int pos;
1090
1091 mutex_lock(&nic_data->local_lock);
1092
1093 /* Move the existing peer pages off %local_page_list */
1094 INIT_LIST_HEAD(&pages);
1095 list_splice_tail_init(&nic_data->local_page_list, &pages);
1096
1097 /* Populate the VF addresses starting from entry 1 (entry 0 is
1098 * the PF address)
1099 */
1100 peer = vfdi_status->peers + 1;
1101 peer_space = ARRAY_SIZE(vfdi_status->peers) - 1;
1102 peer_count = 1;
1103 for (pos = 0; pos < efx->vf_count; ++pos) {
1104 vf = nic_data->vf + pos;
1105
1106 mutex_lock(&vf->status_lock);
1107 if (vf->rx_filtering && !is_zero_ether_addr(vf->addr.mac_addr)) {
1108 *peer++ = vf->addr;
1109 ++peer_count;
1110 --peer_space;
1111 BUG_ON(peer_space == 0);
1112 }
1113 mutex_unlock(&vf->status_lock);
1114 }
1115
1116 /* Fill the remaining addresses */
1117 list_for_each_entry(local_addr, &nic_data->local_addr_list, link) {
1118 ether_addr_copy(peer->mac_addr, local_addr->addr);
1119 peer->tci = 0;
1120 ++peer;
1121 ++peer_count;
1122 if (--peer_space == 0) {
1123 if (list_empty(&pages)) {
1124 epp = kmalloc(sizeof(*epp), GFP_KERNEL);
1125 if (!epp)
1126 break;
1127 epp->ptr = dma_alloc_coherent(
1128 &efx->pci_dev->dev, EFX_PAGE_SIZE,
1129 &epp->addr, GFP_KERNEL);
1130 if (!epp->ptr) {
1131 kfree(epp);
1132 break;
1133 }
1134 } else {
1135 epp = list_first_entry(
1136 &pages, struct efx_endpoint_page, link);
1137 list_del(&epp->link);
1138 }
1139
1140 list_add_tail(&epp->link, &nic_data->local_page_list);
1141 peer = (struct vfdi_endpoint *)epp->ptr;
1142 peer_space = EFX_PAGE_SIZE / sizeof(struct vfdi_endpoint);
1143 }
1144 }
1145 vfdi_status->peer_count = peer_count;
1146 mutex_unlock(&nic_data->local_lock);
1147
1148 /* Free any now unused endpoint pages */
1149 while (!list_empty(&pages)) {
1150 epp = list_first_entry(
1151 &pages, struct efx_endpoint_page, link);
1152 list_del(&epp->link);
1153 dma_free_coherent(&efx->pci_dev->dev, EFX_PAGE_SIZE,
1154 epp->ptr, epp->addr);
1155 kfree(epp);
1156 }
1157
1158 /* Finally, push the pages */
1159 for (pos = 0; pos < efx->vf_count; ++pos) {
1160 vf = nic_data->vf + pos;
1161
1162 mutex_lock(&vf->status_lock);
1163 if (vf->status_addr)
1164 __efx_siena_sriov_push_vf_status(vf);
1165 mutex_unlock(&vf->status_lock);
1166 }
1167}
1168
1169static void efx_siena_sriov_free_local(struct efx_nic *efx)
1170{
1171 struct siena_nic_data *nic_data = efx->nic_data;
1172 struct efx_local_addr *local_addr;
1173 struct efx_endpoint_page *epp;
1174
1175 while (!list_empty(&nic_data->local_addr_list)) {
1176 local_addr = list_first_entry(&nic_data->local_addr_list,
1177 struct efx_local_addr, link);
1178 list_del(&local_addr->link);
1179 kfree(local_addr);
1180 }
1181
1182 while (!list_empty(&nic_data->local_page_list)) {
1183 epp = list_first_entry(&nic_data->local_page_list,
1184 struct efx_endpoint_page, link);
1185 list_del(&epp->link);
1186 dma_free_coherent(&efx->pci_dev->dev, EFX_PAGE_SIZE,
1187 epp->ptr, epp->addr);
1188 kfree(epp);
1189 }
1190}
1191
1192static int efx_siena_sriov_vf_alloc(struct efx_nic *efx)
1193{
1194 unsigned index;
1195 struct siena_vf *vf;
1196 struct siena_nic_data *nic_data = efx->nic_data;
1197
1198 nic_data->vf = kcalloc(efx->vf_count, sizeof(*nic_data->vf),
1199 GFP_KERNEL);
1200 if (!nic_data->vf)
1201 return -ENOMEM;
1202
1203 for (index = 0; index < efx->vf_count; ++index) {
1204 vf = nic_data->vf + index;
1205
1206 vf->efx = efx;
1207 vf->index = index;
1208 vf->rx_filter_id = -1;
1209 vf->tx_filter_mode = VF_TX_FILTER_AUTO;
1210 vf->tx_filter_id = -1;
1211 INIT_WORK(&vf->req, efx_siena_sriov_vfdi);
1212 INIT_WORK(&vf->reset_work, efx_siena_sriov_reset_vf_work);
1213 init_waitqueue_head(&vf->flush_waitq);
1214 mutex_init(&vf->status_lock);
1215 mutex_init(&vf->txq_lock);
1216 }
1217
1218 return 0;
1219}
1220
1221static void efx_siena_sriov_vfs_fini(struct efx_nic *efx)
1222{
1223 struct siena_nic_data *nic_data = efx->nic_data;
1224 struct siena_vf *vf;
1225 unsigned int pos;
1226
1227 for (pos = 0; pos < efx->vf_count; ++pos) {
1228 vf = nic_data->vf + pos;
1229
1230 efx_nic_free_buffer(efx, &vf->buf);
1231 kfree(vf->peer_page_addrs);
1232 vf->peer_page_addrs = NULL;
1233 vf->peer_page_count = 0;
1234
1235 vf->evq0_count = 0;
1236 }
1237}
1238
1239static int efx_siena_sriov_vfs_init(struct efx_nic *efx)
1240{
1241 struct pci_dev *pci_dev = efx->pci_dev;
1242 struct siena_nic_data *nic_data = efx->nic_data;
1243 unsigned index, devfn, sriov, buftbl_base;
1244 u16 offset, stride;
1245 struct siena_vf *vf;
1246 int rc;
1247
1248 sriov = pci_find_ext_capability(pci_dev, PCI_EXT_CAP_ID_SRIOV);
1249 if (!sriov)
1250 return -ENOENT;
1251
1252 pci_read_config_word(pci_dev, sriov + PCI_SRIOV_VF_OFFSET, &offset);
1253 pci_read_config_word(pci_dev, sriov + PCI_SRIOV_VF_STRIDE, &stride);
1254
1255 buftbl_base = nic_data->vf_buftbl_base;
1256 devfn = pci_dev->devfn + offset;
1257 for (index = 0; index < efx->vf_count; ++index) {
1258 vf = nic_data->vf + index;
1259
1260 /* Reserve buffer entries */
1261 vf->buftbl_base = buftbl_base;
1262 buftbl_base += EFX_VF_BUFTBL_PER_VI * efx_vf_size(efx);
1263
1264 vf->pci_rid = devfn;
1265 snprintf(vf->pci_name, sizeof(vf->pci_name),
1266 "%04x:%02x:%02x.%d",
1267 pci_domain_nr(pci_dev->bus), pci_dev->bus->number,
1268 PCI_SLOT(devfn), PCI_FUNC(devfn));
1269
1270 rc = efx_nic_alloc_buffer(efx, &vf->buf, EFX_PAGE_SIZE,
1271 GFP_KERNEL);
1272 if (rc)
1273 goto fail;
1274
1275 devfn += stride;
1276 }
1277
1278 return 0;
1279
1280fail:
1281 efx_siena_sriov_vfs_fini(efx);
1282 return rc;
1283}
1284
1285int efx_siena_sriov_init(struct efx_nic *efx)
1286{
1287 struct net_device *net_dev = efx->net_dev;
1288 struct siena_nic_data *nic_data = efx->nic_data;
1289 struct vfdi_status *vfdi_status;
1290 int rc;
1291
1292 /* Ensure there's room for vf_channel */
1293 BUILD_BUG_ON(EFX_MAX_CHANNELS + 1 >= EFX_VI_BASE);
1294 /* Ensure that VI_BASE is aligned on VI_SCALE */
1295 BUILD_BUG_ON(EFX_VI_BASE & ((1 << EFX_VI_SCALE_MAX) - 1));
1296
1297 if (efx->vf_count == 0)
1298 return 0;
1299
1300 rc = efx_siena_sriov_cmd(efx, true, NULL, NULL);
1301 if (rc)
1302 goto fail_cmd;
1303
1304 rc = efx_nic_alloc_buffer(efx, &nic_data->vfdi_status,
1305 sizeof(*vfdi_status), GFP_KERNEL);
1306 if (rc)
1307 goto fail_status;
1308 vfdi_status = nic_data->vfdi_status.addr;
1309 memset(vfdi_status, 0, sizeof(*vfdi_status));
1310 vfdi_status->version = 1;
1311 vfdi_status->length = sizeof(*vfdi_status);
1312 vfdi_status->max_tx_channels = vf_max_tx_channels;
1313 vfdi_status->vi_scale = efx->vi_scale;
1314 vfdi_status->rss_rxq_count = efx->rss_spread;
1315 vfdi_status->peer_count = 1 + efx->vf_count;
1316 vfdi_status->timer_quantum_ns = efx->timer_quantum_ns;
1317
1318 rc = efx_siena_sriov_vf_alloc(efx);
1319 if (rc)
1320 goto fail_alloc;
1321
1322 mutex_init(&nic_data->local_lock);
1323 INIT_WORK(&nic_data->peer_work, efx_siena_sriov_peer_work);
1324 INIT_LIST_HEAD(&nic_data->local_addr_list);
1325 INIT_LIST_HEAD(&nic_data->local_page_list);
1326
1327 rc = efx_siena_sriov_vfs_init(efx);
1328 if (rc)
1329 goto fail_vfs;
1330
1331 rtnl_lock();
1332 ether_addr_copy(vfdi_status->peers[0].mac_addr, net_dev->dev_addr);
1333 efx->vf_init_count = efx->vf_count;
1334 rtnl_unlock();
1335
1336 efx_siena_sriov_usrev(efx, true);
1337
1338 /* At this point we must be ready to accept VFDI requests */
1339
1340 rc = pci_enable_sriov(efx->pci_dev, efx->vf_count);
1341 if (rc)
1342 goto fail_pci;
1343
1344 netif_info(efx, probe, net_dev,
1345 "enabled SR-IOV for %d VFs, %d VI per VF\n",
1346 efx->vf_count, efx_vf_size(efx));
1347 return 0;
1348
1349fail_pci:
1350 efx_siena_sriov_usrev(efx, false);
1351 rtnl_lock();
1352 efx->vf_init_count = 0;
1353 rtnl_unlock();
1354 efx_siena_sriov_vfs_fini(efx);
1355fail_vfs:
1356 cancel_work_sync(&nic_data->peer_work);
1357 efx_siena_sriov_free_local(efx);
1358 kfree(nic_data->vf);
1359fail_alloc:
1360 efx_nic_free_buffer(efx, &nic_data->vfdi_status);
1361fail_status:
1362 efx_siena_sriov_cmd(efx, false, NULL, NULL);
1363fail_cmd:
1364 return rc;
1365}
1366
1367void efx_siena_sriov_fini(struct efx_nic *efx)
1368{
1369 struct siena_vf *vf;
1370 unsigned int pos;
1371 struct siena_nic_data *nic_data = efx->nic_data;
1372
1373 if (efx->vf_init_count == 0)
1374 return;
1375
1376 /* Disable all interfaces to reconfiguration */
1377 BUG_ON(nic_data->vfdi_channel->enabled);
1378 efx_siena_sriov_usrev(efx, false);
1379 rtnl_lock();
1380 efx->vf_init_count = 0;
1381 rtnl_unlock();
1382
1383 /* Flush all reconfiguration work */
1384 for (pos = 0; pos < efx->vf_count; ++pos) {
1385 vf = nic_data->vf + pos;
1386 cancel_work_sync(&vf->req);
1387 cancel_work_sync(&vf->reset_work);
1388 }
1389 cancel_work_sync(&nic_data->peer_work);
1390
1391 pci_disable_sriov(efx->pci_dev);
1392
1393 /* Tear down back-end state */
1394 efx_siena_sriov_vfs_fini(efx);
1395 efx_siena_sriov_free_local(efx);
1396 kfree(nic_data->vf);
1397 efx_nic_free_buffer(efx, &nic_data->vfdi_status);
1398 efx_siena_sriov_cmd(efx, false, NULL, NULL);
1399}
1400
1401void efx_siena_sriov_event(struct efx_channel *channel, efx_qword_t *event)
1402{
1403 struct efx_nic *efx = channel->efx;
1404 struct siena_vf *vf;
1405 unsigned qid, seq, type, data;
1406
1407 qid = EFX_QWORD_FIELD(*event, FSF_CZ_USER_QID);
1408
1409 /* USR_EV_REG_VALUE is dword0, so access the VFDI_EV fields directly */
1410 BUILD_BUG_ON(FSF_CZ_USER_EV_REG_VALUE_LBN != 0);
1411 seq = EFX_QWORD_FIELD(*event, VFDI_EV_SEQ);
1412 type = EFX_QWORD_FIELD(*event, VFDI_EV_TYPE);
1413 data = EFX_QWORD_FIELD(*event, VFDI_EV_DATA);
1414
1415 netif_vdbg(efx, hw, efx->net_dev,
1416 "USR_EV event from qid %d seq 0x%x type %d data 0x%x\n",
1417 qid, seq, type, data);
1418
1419 if (map_vi_index(efx, qid, &vf, NULL))
1420 return;
1421 if (vf->busy)
1422 goto error;
1423
1424 if (type == VFDI_EV_TYPE_REQ_WORD0) {
1425 /* Resynchronise */
1426 vf->req_type = VFDI_EV_TYPE_REQ_WORD0;
1427 vf->req_seqno = seq + 1;
1428 vf->req_addr = 0;
1429 } else if (seq != (vf->req_seqno++ & 0xff) || type != vf->req_type)
1430 goto error;
1431
1432 switch (vf->req_type) {
1433 case VFDI_EV_TYPE_REQ_WORD0:
1434 case VFDI_EV_TYPE_REQ_WORD1:
1435 case VFDI_EV_TYPE_REQ_WORD2:
1436 vf->req_addr |= (u64)data << (vf->req_type << 4);
1437 ++vf->req_type;
1438 return;
1439
1440 case VFDI_EV_TYPE_REQ_WORD3:
1441 vf->req_addr |= (u64)data << 48;
1442 vf->req_type = VFDI_EV_TYPE_REQ_WORD0;
1443 vf->busy = true;
1444 queue_work(vfdi_workqueue, &vf->req);
1445 return;
1446 }
1447
1448error:
1449 if (net_ratelimit())
1450 netif_err(efx, hw, efx->net_dev,
1451 "ERROR: Screaming VFDI request from %s\n",
1452 vf->pci_name);
1453 /* Reset the request and sequence number */
1454 vf->req_type = VFDI_EV_TYPE_REQ_WORD0;
1455 vf->req_seqno = seq + 1;
1456}
1457
1458void efx_siena_sriov_flr(struct efx_nic *efx, unsigned vf_i)
1459{
1460 struct siena_nic_data *nic_data = efx->nic_data;
1461 struct siena_vf *vf;
1462
1463 if (vf_i > efx->vf_init_count)
1464 return;
1465 vf = nic_data->vf + vf_i;
1466 netif_info(efx, hw, efx->net_dev,
1467 "FLR on VF %s\n", vf->pci_name);
1468
1469 vf->status_addr = 0;
1470 efx_vfdi_remove_all_filters(vf);
1471 efx_vfdi_flush_clear(vf);
1472
1473 vf->evq0_count = 0;
1474}
1475
1476int efx_siena_sriov_mac_address_changed(struct efx_nic *efx)
1477{
1478 struct siena_nic_data *nic_data = efx->nic_data;
1479 struct vfdi_status *vfdi_status = nic_data->vfdi_status.addr;
1480
1481 if (!efx->vf_init_count)
1482 return 0;
1483 ether_addr_copy(vfdi_status->peers[0].mac_addr,
1484 efx->net_dev->dev_addr);
1485 queue_work(vfdi_workqueue, &nic_data->peer_work);
1486
1487 return 0;
1488}
1489
1490void efx_siena_sriov_tx_flush_done(struct efx_nic *efx, efx_qword_t *event)
1491{
1492 struct siena_vf *vf;
1493 unsigned queue, qid;
1494
1495 queue = EFX_QWORD_FIELD(*event, FSF_AZ_DRIVER_EV_SUBDATA);
1496 if (map_vi_index(efx, queue, &vf, &qid))
1497 return;
1498 /* Ignore flush completions triggered by an FLR */
1499 if (!test_bit(qid, vf->txq_mask))
1500 return;
1501
1502 __clear_bit(qid, vf->txq_mask);
1503 --vf->txq_count;
1504
1505 if (efx_vfdi_flush_wake(vf))
1506 wake_up(&vf->flush_waitq);
1507}
1508
1509void efx_siena_sriov_rx_flush_done(struct efx_nic *efx, efx_qword_t *event)
1510{
1511 struct siena_vf *vf;
1512 unsigned ev_failed, queue, qid;
1513
1514 queue = EFX_QWORD_FIELD(*event, FSF_AZ_DRIVER_EV_RX_DESCQ_ID);
1515 ev_failed = EFX_QWORD_FIELD(*event,
1516 FSF_AZ_DRIVER_EV_RX_FLUSH_FAIL);
1517 if (map_vi_index(efx, queue, &vf, &qid))
1518 return;
1519 if (!test_bit(qid, vf->rxq_mask))
1520 return;
1521
1522 if (ev_failed) {
1523 set_bit(qid, vf->rxq_retry_mask);
1524 atomic_inc(&vf->rxq_retry_count);
1525 } else {
1526 __clear_bit(qid, vf->rxq_mask);
1527 --vf->rxq_count;
1528 }
1529 if (efx_vfdi_flush_wake(vf))
1530 wake_up(&vf->flush_waitq);
1531}
1532
1533/* Called from napi. Schedule the reset work item */
1534void efx_siena_sriov_desc_fetch_err(struct efx_nic *efx, unsigned dmaq)
1535{
1536 struct siena_vf *vf;
1537 unsigned int rel;
1538
1539 if (map_vi_index(efx, dmaq, &vf, &rel))
1540 return;
1541
1542 if (net_ratelimit())
1543 netif_err(efx, hw, efx->net_dev,
1544 "VF %d DMA Q %d reports descriptor fetch error.\n",
1545 vf->index, rel);
1546 queue_work(vfdi_workqueue, &vf->reset_work);
1547}
1548
1549/* Reset all VFs */
1550void efx_siena_sriov_reset(struct efx_nic *efx)
1551{
1552 struct siena_nic_data *nic_data = efx->nic_data;
1553 unsigned int vf_i;
1554 struct efx_buffer buf;
1555 struct siena_vf *vf;
1556
1557 ASSERT_RTNL();
1558
1559 if (efx->vf_init_count == 0)
1560 return;
1561
1562 efx_siena_sriov_usrev(efx, true);
1563 (void)efx_siena_sriov_cmd(efx, true, NULL, NULL);
1564
1565 if (efx_nic_alloc_buffer(efx, &buf, EFX_PAGE_SIZE, GFP_NOIO))
1566 return;
1567
1568 for (vf_i = 0; vf_i < efx->vf_init_count; ++vf_i) {
1569 vf = nic_data->vf + vf_i;
1570 efx_siena_sriov_reset_vf(vf, &buf);
1571 }
1572
1573 efx_nic_free_buffer(efx, &buf);
1574}
1575
1576int efx_init_sriov(void)
1577{
1578 /* A single threaded workqueue is sufficient. efx_siena_sriov_vfdi() and
1579 * efx_siena_sriov_peer_work() spend almost all their time sleeping for
1580 * MCDI to complete anyway
1581 */
1582 vfdi_workqueue = create_singlethread_workqueue("sfc_vfdi");
1583 if (!vfdi_workqueue)
1584 return -ENOMEM;
1585 return 0;
1586}
1587
1588void efx_fini_sriov(void)
1589{
1590 destroy_workqueue(vfdi_workqueue);
1591}
1592
1593int efx_siena_sriov_set_vf_mac(struct efx_nic *efx, int vf_i, u8 *mac)
1594{
1595 struct siena_nic_data *nic_data = efx->nic_data;
1596 struct siena_vf *vf;
1597
1598 if (vf_i >= efx->vf_init_count)
1599 return -EINVAL;
1600 vf = nic_data->vf + vf_i;
1601
1602 mutex_lock(&vf->status_lock);
1603 ether_addr_copy(vf->addr.mac_addr, mac);
1604 __efx_siena_sriov_update_vf_addr(vf);
1605 mutex_unlock(&vf->status_lock);
1606
1607 return 0;
1608}
1609
1610int efx_siena_sriov_set_vf_vlan(struct efx_nic *efx, int vf_i,
1611 u16 vlan, u8 qos)
1612{
1613 struct siena_nic_data *nic_data = efx->nic_data;
1614 struct siena_vf *vf;
1615 u16 tci;
1616
1617 if (vf_i >= efx->vf_init_count)
1618 return -EINVAL;
1619 vf = nic_data->vf + vf_i;
1620
1621 mutex_lock(&vf->status_lock);
1622 tci = (vlan & VLAN_VID_MASK) | ((qos & 0x7) << VLAN_PRIO_SHIFT);
1623 vf->addr.tci = htons(tci);
1624 __efx_siena_sriov_update_vf_addr(vf);
1625 mutex_unlock(&vf->status_lock);
1626
1627 return 0;
1628}
1629
1630int efx_siena_sriov_set_vf_spoofchk(struct efx_nic *efx, int vf_i,
1631 bool spoofchk)
1632{
1633 struct siena_nic_data *nic_data = efx->nic_data;
1634 struct siena_vf *vf;
1635 int rc;
1636
1637 if (vf_i >= efx->vf_init_count)
1638 return -EINVAL;
1639 vf = nic_data->vf + vf_i;
1640
1641 mutex_lock(&vf->txq_lock);
1642 if (vf->txq_count == 0) {
1643 vf->tx_filter_mode =
1644 spoofchk ? VF_TX_FILTER_ON : VF_TX_FILTER_OFF;
1645 rc = 0;
1646 } else {
1647 /* This cannot be changed while TX queues are running */
1648 rc = -EBUSY;
1649 }
1650 mutex_unlock(&vf->txq_lock);
1651 return rc;
1652}
1653
1654int efx_siena_sriov_get_vf_config(struct efx_nic *efx, int vf_i,
1655 struct ifla_vf_info *ivi)
1656{
1657 struct siena_nic_data *nic_data = efx->nic_data;
1658 struct siena_vf *vf;
1659 u16 tci;
1660
1661 if (vf_i >= efx->vf_init_count)
1662 return -EINVAL;
1663 vf = nic_data->vf + vf_i;
1664
1665 ivi->vf = vf_i;
1666 ether_addr_copy(ivi->mac, vf->addr.mac_addr);
1667 ivi->max_tx_rate = 0;
1668 ivi->min_tx_rate = 0;
1669 tci = ntohs(vf->addr.tci);
1670 ivi->vlan = tci & VLAN_VID_MASK;
1671 ivi->qos = (tci >> VLAN_PRIO_SHIFT) & 0x7;
1672 ivi->spoofchk = vf->tx_filter_mode == VF_TX_FILTER_ON;
1673
1674 return 0;
1675}
1676
1677bool efx_siena_sriov_wanted(struct efx_nic *efx)
1678{
1679 return efx->vf_count != 0;
1680}
1681
1682int efx_siena_sriov_configure(struct efx_nic *efx, int num_vfs)
1683{
1684 return 0;
1685}